2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* linux/mm/vmscan.c
|
|
|
|
*
|
|
|
|
* Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
|
|
|
|
*
|
|
|
|
* Swap reorganised 29.12.95, Stephen Tweedie.
|
|
|
|
* kswapd added: 7.1.96 sct
|
|
|
|
* Removed kswapd_ctl limits, and swap out as many pages as needed
|
|
|
|
* to bring the system back to freepages.high: 2.4.97, Rik van Riel.
|
|
|
|
* Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com).
|
|
|
|
* Multiqueue VM started 5.8.00, Rik van Riel.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/module.h>
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 15:04:11 +07:00
|
|
|
#include <linux/gfp.h>
|
2005-04-17 05:20:36 +07:00
|
|
|
#include <linux/kernel_stat.h>
|
|
|
|
#include <linux/swap.h>
|
|
|
|
#include <linux/pagemap.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/highmem.h>
|
2006-09-27 15:50:00 +07:00
|
|
|
#include <linux/vmstat.h>
|
2005-04-17 05:20:36 +07:00
|
|
|
#include <linux/file.h>
|
|
|
|
#include <linux/writeback.h>
|
|
|
|
#include <linux/blkdev.h>
|
|
|
|
#include <linux/buffer_head.h> /* for try_to_release_page(),
|
|
|
|
buffer_heads_over_limit */
|
|
|
|
#include <linux/mm_inline.h>
|
|
|
|
#include <linux/backing-dev.h>
|
|
|
|
#include <linux/rmap.h>
|
|
|
|
#include <linux/topology.h>
|
|
|
|
#include <linux/cpu.h>
|
|
|
|
#include <linux/cpuset.h>
|
2011-01-14 06:45:56 +07:00
|
|
|
#include <linux/compaction.h>
|
2005-04-17 05:20:36 +07:00
|
|
|
#include <linux/notifier.h>
|
|
|
|
#include <linux/rwsem.h>
|
2006-03-22 15:09:04 +07:00
|
|
|
#include <linux/delay.h>
|
2006-06-27 16:53:33 +07:00
|
|
|
#include <linux/kthread.h>
|
2006-12-07 11:34:23 +07:00
|
|
|
#include <linux/freezer.h>
|
2008-02-07 15:13:56 +07:00
|
|
|
#include <linux/memcontrol.h>
|
2008-07-25 15:48:52 +07:00
|
|
|
#include <linux/delayacct.h>
|
2008-10-19 10:26:53 +07:00
|
|
|
#include <linux/sysctl.h>
|
vmscan: all_unreclaimable() use zone->all_unreclaimable as a name
all_unreclaimable check in direct reclaim has been introduced at 2.6.19
by following commit.
2006 Sep 25; commit 408d8544; oom: use unreclaimable info
And it went through strange history. firstly, following commit broke
the logic unintentionally.
2008 Apr 29; commit a41f24ea; page allocator: smarter retry of
costly-order allocations
Two years later, I've found obvious meaningless code fragment and
restored original intention by following commit.
2010 Jun 04; commit bb21c7ce; vmscan: fix do_try_to_free_pages()
return value when priority==0
But, the logic didn't works when 32bit highmem system goes hibernation
and Minchan slightly changed the algorithm and fixed it .
2010 Sep 22: commit d1908362: vmscan: check all_unreclaimable
in direct reclaim path
But, recently, Andrey Vagin found the new corner case. Look,
struct zone {
..
int all_unreclaimable;
..
unsigned long pages_scanned;
..
}
zone->all_unreclaimable and zone->pages_scanned are neigher atomic
variables nor protected by lock. Therefore zones can become a state of
zone->page_scanned=0 and zone->all_unreclaimable=1. In this case, current
all_unreclaimable() return false even though zone->all_unreclaimabe=1.
This resulted in the kernel hanging up when executing a loop of the form
1. fork
2. mmap
3. touch memory
4. read memory
5. munmmap
as described in
http://www.gossamer-threads.com/lists/linux/kernel/1348725#1348725
Is this ignorable minor issue? No. Unfortunately, x86 has very small dma
zone and it become zone->all_unreclamble=1 easily. and if it become
all_unreclaimable=1, it never restore all_unreclaimable=0. Why? if
all_unreclaimable=1, vmscan only try DEF_PRIORITY reclaim and
a-few-lru-pages>>DEF_PRIORITY always makes 0. that mean no page scan at
all!
Eventually, oom-killer never works on such systems. That said, we can't
use zone->pages_scanned for this purpose. This patch restore
all_unreclaimable() use zone->all_unreclaimable as old. and in addition,
to add oom_killer_disabled check to avoid reintroduce the issue of commit
d1908362 ("vmscan: check all_unreclaimable in direct reclaim path").
Reported-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-15 05:22:12 +07:00
|
|
|
#include <linux/oom.h>
|
2011-05-21 02:50:29 +07:00
|
|
|
#include <linux/prefetch.h>
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
#include <asm/tlbflush.h>
|
|
|
|
#include <asm/div64.h>
|
|
|
|
|
|
|
|
#include <linux/swapops.h>
|
|
|
|
|
2006-03-22 15:08:33 +07:00
|
|
|
#include "internal.h"
|
|
|
|
|
2010-08-10 07:19:16 +07:00
|
|
|
#define CREATE_TRACE_POINTS
|
|
|
|
#include <trace/events/vmscan.h>
|
|
|
|
|
2011-01-14 06:45:55 +07:00
|
|
|
/*
|
2011-01-14 06:46:00 +07:00
|
|
|
* reclaim_mode determines how the inactive list is shrunk
|
|
|
|
* RECLAIM_MODE_SINGLE: Reclaim only order-0 pages
|
|
|
|
* RECLAIM_MODE_ASYNC: Do not block
|
|
|
|
* RECLAIM_MODE_SYNC: Allow blocking e.g. call wait_on_page_writeback
|
|
|
|
* RECLAIM_MODE_LUMPYRECLAIM: For high-order allocations, take a reference
|
2011-01-14 06:45:55 +07:00
|
|
|
* page from the LRU and reclaim all pages within a
|
|
|
|
* naturally aligned range
|
2011-01-14 06:46:00 +07:00
|
|
|
* RECLAIM_MODE_COMPACTION: For high-order allocations, reclaim a number of
|
2011-01-14 06:45:56 +07:00
|
|
|
* order-0 pages and then compact the zone
|
2011-01-14 06:45:55 +07:00
|
|
|
*/
|
2011-01-14 06:46:00 +07:00
|
|
|
typedef unsigned __bitwise__ reclaim_mode_t;
|
|
|
|
#define RECLAIM_MODE_SINGLE ((__force reclaim_mode_t)0x01u)
|
|
|
|
#define RECLAIM_MODE_ASYNC ((__force reclaim_mode_t)0x02u)
|
|
|
|
#define RECLAIM_MODE_SYNC ((__force reclaim_mode_t)0x04u)
|
|
|
|
#define RECLAIM_MODE_LUMPYRECLAIM ((__force reclaim_mode_t)0x08u)
|
|
|
|
#define RECLAIM_MODE_COMPACTION ((__force reclaim_mode_t)0x10u)
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
struct scan_control {
|
|
|
|
/* Incremented by the number of inactive pages that were scanned */
|
|
|
|
unsigned long nr_scanned;
|
|
|
|
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
/* Number of pages freed so far during a call to shrink_zones() */
|
|
|
|
unsigned long nr_reclaimed;
|
|
|
|
|
2009-12-15 08:59:10 +07:00
|
|
|
/* How many pages shrink_list() should reclaim */
|
|
|
|
unsigned long nr_to_reclaim;
|
|
|
|
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
unsigned long hibernation_mode;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This context's GFP mask */
|
2005-10-21 14:18:50 +07:00
|
|
|
gfp_t gfp_mask;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
int may_writepage;
|
|
|
|
|
2009-04-01 05:19:30 +07:00
|
|
|
/* Can mapped pages be reclaimed? */
|
|
|
|
int may_unmap;
|
2006-01-19 08:42:30 +07:00
|
|
|
|
2009-04-22 02:24:57 +07:00
|
|
|
/* Can pages be swapped as part of reclaim? */
|
|
|
|
int may_swap;
|
|
|
|
|
2007-07-17 18:03:16 +07:00
|
|
|
int order;
|
2008-02-07 15:13:56 +07:00
|
|
|
|
2010-05-25 04:32:37 +07:00
|
|
|
/*
|
2010-08-17 17:09:09 +07:00
|
|
|
* Intend to reclaim enough continuous memory rather than reclaim
|
|
|
|
* enough amount of memory. i.e, mode for high order allocation.
|
2010-05-25 04:32:37 +07:00
|
|
|
*/
|
2011-01-14 06:46:00 +07:00
|
|
|
reclaim_mode_t reclaim_mode;
|
2010-05-25 04:32:37 +07:00
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
/*
|
|
|
|
* The memory cgroup that hit its limit and as a result is the
|
|
|
|
* primary target of this reclaim invocation.
|
|
|
|
*/
|
|
|
|
struct mem_cgroup *target_mem_cgroup;
|
2008-02-07 15:13:56 +07:00
|
|
|
|
2009-04-01 05:23:31 +07:00
|
|
|
/*
|
|
|
|
* Nodemask of nodes allowed by the caller. If NULL, all nodes
|
|
|
|
* are scanned.
|
|
|
|
*/
|
|
|
|
nodemask_t *nodemask;
|
2005-04-17 05:20:36 +07:00
|
|
|
};
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
struct mem_cgroup_zone {
|
|
|
|
struct mem_cgroup *mem_cgroup;
|
|
|
|
struct zone *zone;
|
|
|
|
};
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
|
|
|
|
|
|
|
|
#ifdef ARCH_HAS_PREFETCH
|
|
|
|
#define prefetch_prev_lru_page(_page, _base, _field) \
|
|
|
|
do { \
|
|
|
|
if ((_page)->lru.prev != _base) { \
|
|
|
|
struct page *prev; \
|
|
|
|
\
|
|
|
|
prev = lru_to_page(&(_page->lru)); \
|
|
|
|
prefetch(&prev->_field); \
|
|
|
|
} \
|
|
|
|
} while (0)
|
|
|
|
#else
|
|
|
|
#define prefetch_prev_lru_page(_page, _base, _field) do { } while (0)
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef ARCH_HAS_PREFETCHW
|
|
|
|
#define prefetchw_prev_lru_page(_page, _base, _field) \
|
|
|
|
do { \
|
|
|
|
if ((_page)->lru.prev != _base) { \
|
|
|
|
struct page *prev; \
|
|
|
|
\
|
|
|
|
prev = lru_to_page(&(_page->lru)); \
|
|
|
|
prefetchw(&prev->_field); \
|
|
|
|
} \
|
|
|
|
} while (0)
|
|
|
|
#else
|
|
|
|
#define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0)
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/*
|
|
|
|
* From 0 .. 100. Higher means more swappy.
|
|
|
|
*/
|
|
|
|
int vm_swappiness = 60;
|
2006-06-23 16:03:47 +07:00
|
|
|
long vm_total_pages; /* The total number of pages which the VM controls */
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
static LIST_HEAD(shrinker_list);
|
|
|
|
static DECLARE_RWSEM(shrinker_rwsem);
|
|
|
|
|
2008-03-05 05:28:39 +07:00
|
|
|
#ifdef CONFIG_CGROUP_MEM_RES_CTLR
|
2012-01-13 08:17:50 +07:00
|
|
|
static bool global_reclaim(struct scan_control *sc)
|
|
|
|
{
|
2012-01-13 08:17:52 +07:00
|
|
|
return !sc->target_mem_cgroup;
|
2012-01-13 08:17:50 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static bool scanning_global_lru(struct mem_cgroup_zone *mz)
|
2012-01-13 08:17:50 +07:00
|
|
|
{
|
2012-01-13 08:17:52 +07:00
|
|
|
return !mz->mem_cgroup;
|
2012-01-13 08:17:50 +07:00
|
|
|
}
|
2008-02-07 15:14:29 +07:00
|
|
|
#else
|
2012-01-13 08:17:50 +07:00
|
|
|
static bool global_reclaim(struct scan_control *sc)
|
|
|
|
{
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static bool scanning_global_lru(struct mem_cgroup_zone *mz)
|
2012-01-13 08:17:50 +07:00
|
|
|
{
|
|
|
|
return true;
|
|
|
|
}
|
2008-02-07 15:14:29 +07:00
|
|
|
#endif
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static struct zone_reclaim_stat *get_reclaim_stat(struct mem_cgroup_zone *mz)
|
2009-01-08 09:08:15 +07:00
|
|
|
{
|
2012-01-13 08:17:52 +07:00
|
|
|
if (!scanning_global_lru(mz))
|
|
|
|
return mem_cgroup_get_reclaim_stat(mz->mem_cgroup, mz->zone);
|
2009-01-08 09:08:20 +07:00
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
return &mz->zone->reclaim_stat;
|
2009-01-08 09:08:15 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static unsigned long zone_nr_lru_pages(struct mem_cgroup_zone *mz,
|
|
|
|
enum lru_list lru)
|
2009-01-08 09:08:16 +07:00
|
|
|
{
|
2012-01-13 08:17:52 +07:00
|
|
|
if (!scanning_global_lru(mz))
|
|
|
|
return mem_cgroup_zone_nr_lru_pages(mz->mem_cgroup,
|
|
|
|
zone_to_nid(mz->zone),
|
|
|
|
zone_idx(mz->zone),
|
|
|
|
BIT(lru));
|
2009-01-08 09:08:19 +07:00
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
return zone_page_state(mz->zone, NR_LRU_BASE + lru);
|
2009-01-08 09:08:16 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* Add a shrinker callback to be called from the vm
|
|
|
|
*/
|
2007-07-17 18:03:17 +07:00
|
|
|
void register_shrinker(struct shrinker *shrinker)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2011-12-09 05:33:54 +07:00
|
|
|
atomic_long_set(&shrinker->nr_in_batch, 0);
|
2007-07-17 18:03:17 +07:00
|
|
|
down_write(&shrinker_rwsem);
|
|
|
|
list_add_tail(&shrinker->list, &shrinker_list);
|
|
|
|
up_write(&shrinker_rwsem);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2007-07-17 18:03:17 +07:00
|
|
|
EXPORT_SYMBOL(register_shrinker);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Remove one
|
|
|
|
*/
|
2007-07-17 18:03:17 +07:00
|
|
|
void unregister_shrinker(struct shrinker *shrinker)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
down_write(&shrinker_rwsem);
|
|
|
|
list_del(&shrinker->list);
|
|
|
|
up_write(&shrinker_rwsem);
|
|
|
|
}
|
2007-07-17 18:03:17 +07:00
|
|
|
EXPORT_SYMBOL(unregister_shrinker);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-05-25 07:12:27 +07:00
|
|
|
static inline int do_shrinker_shrink(struct shrinker *shrinker,
|
|
|
|
struct shrink_control *sc,
|
|
|
|
unsigned long nr_to_scan)
|
|
|
|
{
|
|
|
|
sc->nr_to_scan = nr_to_scan;
|
|
|
|
return (*shrinker->shrink)(shrinker, sc);
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
#define SHRINK_BATCH 128
|
|
|
|
/*
|
|
|
|
* Call the shrink functions to age shrinkable caches
|
|
|
|
*
|
|
|
|
* Here we assume it costs one seek to replace a lru page and that it also
|
|
|
|
* takes a seek to recreate a cache object. With this in mind we age equal
|
|
|
|
* percentages of the lru and ageable caches. This should balance the seeks
|
|
|
|
* generated by these structures.
|
|
|
|
*
|
2007-10-20 06:27:18 +07:00
|
|
|
* If the vm encountered mapped pages on the LRU it increase the pressure on
|
2005-04-17 05:20:36 +07:00
|
|
|
* slab to avoid swapping.
|
|
|
|
*
|
|
|
|
* We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits.
|
|
|
|
*
|
|
|
|
* `lru_pages' represents the number of on-LRU pages in all the zones which
|
|
|
|
* are eligible for the caller's allocation attempt. It is used for balancing
|
|
|
|
* slab reclaim versus page reclaim.
|
2005-06-22 07:14:35 +07:00
|
|
|
*
|
|
|
|
* Returns the number of slab objects which we shrunk.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2011-05-25 07:12:26 +07:00
|
|
|
unsigned long shrink_slab(struct shrink_control *shrink,
|
2011-05-25 07:12:27 +07:00
|
|
|
unsigned long nr_pages_scanned,
|
2011-05-25 07:12:26 +07:00
|
|
|
unsigned long lru_pages)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct shrinker *shrinker;
|
2006-03-22 15:08:19 +07:00
|
|
|
unsigned long ret = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-05-25 07:12:27 +07:00
|
|
|
if (nr_pages_scanned == 0)
|
|
|
|
nr_pages_scanned = SWAP_CLUSTER_MAX;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-05-25 07:11:11 +07:00
|
|
|
if (!down_read_trylock(&shrinker_rwsem)) {
|
|
|
|
/* Assume we'll be able to shrink next time */
|
|
|
|
ret = 1;
|
|
|
|
goto out;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
list_for_each_entry(shrinker, &shrinker_list, list) {
|
|
|
|
unsigned long long delta;
|
2011-12-09 05:33:51 +07:00
|
|
|
long total_scan;
|
|
|
|
long max_pass;
|
2011-07-08 11:14:34 +07:00
|
|
|
int shrink_ret = 0;
|
2011-07-08 11:14:35 +07:00
|
|
|
long nr;
|
|
|
|
long new_nr;
|
2011-07-08 11:14:37 +07:00
|
|
|
long batch_size = shrinker->batch ? shrinker->batch
|
|
|
|
: SHRINK_BATCH;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-12-09 05:33:51 +07:00
|
|
|
max_pass = do_shrinker_shrink(shrinker, shrink, 0);
|
|
|
|
if (max_pass <= 0)
|
|
|
|
continue;
|
|
|
|
|
2011-07-08 11:14:35 +07:00
|
|
|
/*
|
|
|
|
* copy the current shrinker scan count into a local variable
|
|
|
|
* and zero it so that other concurrent shrinker invocations
|
|
|
|
* don't also do this scanning work.
|
|
|
|
*/
|
2011-12-09 05:33:54 +07:00
|
|
|
nr = atomic_long_xchg(&shrinker->nr_in_batch, 0);
|
2011-07-08 11:14:35 +07:00
|
|
|
|
|
|
|
total_scan = nr;
|
2011-05-25 07:12:27 +07:00
|
|
|
delta = (4 * nr_pages_scanned) / shrinker->seeks;
|
2005-11-29 04:44:15 +07:00
|
|
|
delta *= max_pass;
|
2005-04-17 05:20:36 +07:00
|
|
|
do_div(delta, lru_pages + 1);
|
2011-07-08 11:14:35 +07:00
|
|
|
total_scan += delta;
|
|
|
|
if (total_scan < 0) {
|
2009-04-01 05:23:29 +07:00
|
|
|
printk(KERN_ERR "shrink_slab: %pF negative objects to "
|
|
|
|
"delete nr=%ld\n",
|
2011-07-08 11:14:35 +07:00
|
|
|
shrinker->shrink, total_scan);
|
|
|
|
total_scan = max_pass;
|
2005-11-29 04:44:15 +07:00
|
|
|
}
|
|
|
|
|
vmscan: reduce wind up shrinker->nr when shrinker can't do work
When a shrinker returns -1 to shrink_slab() to indicate it cannot do
any work given the current memory reclaim requirements, it adds the
entire total_scan count to shrinker->nr. The idea ehind this is that
whenteh shrinker is next called and can do work, it will do the work
of the previously aborted shrinker call as well.
However, if a filesystem is doing lots of allocation with GFP_NOFS
set, then we get many, many more aborts from the shrinkers than we
do successful calls. The result is that shrinker->nr winds up to
it's maximum permissible value (twice the current cache size) and
then when the next shrinker call that can do work is issued, it
has enough scan count built up to free the entire cache twice over.
This manifests itself in the cache going from full to empty in a
matter of seconds, even when only a small part of the cache is
needed to be emptied to free sufficient memory.
Under metadata intensive workloads on ext4 and XFS, I'm seeing the
VFS caches increase memory consumption up to 75% of memory (no page
cache pressure) over a period of 30-60s, and then the shrinker
empties them down to zero in the space of 2-3s. This cycle repeats
over and over again, with the shrinker completely trashing the inode
and dentry caches every minute or so the workload continues.
This behaviour was made obvious by the shrink_slab tracepoints added
earlier in the series, and made worse by the patch that corrected
the concurrent accounting of shrinker->nr.
To avoid this problem, stop repeated small increments of the total
scan value from winding shrinker->nr up to a value that can cause
the entire cache to be freed. We still need to allow it to wind up,
so use the delta as the "large scan" threshold check - if the delta
is more than a quarter of the entire cache size, then it is a large
scan and allowed to cause lots of windup because we are clearly
needing to free lots of memory.
If it isn't a large scan then limit the total scan to half the size
of the cache so that windup never increases to consume the whole
cache. Reducing the total scan limit further does not allow enough
wind-up to maintain the current levels of performance, whilst a
higher threshold does not prevent the windup from freeing the entire
cache under sustained workloads.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-07-08 11:14:36 +07:00
|
|
|
/*
|
|
|
|
* We need to avoid excessive windup on filesystem shrinkers
|
|
|
|
* due to large numbers of GFP_NOFS allocations causing the
|
|
|
|
* shrinkers to return -1 all the time. This results in a large
|
|
|
|
* nr being built up so when a shrink that can do some work
|
|
|
|
* comes along it empties the entire cache due to nr >>>
|
|
|
|
* max_pass. This is bad for sustaining a working set in
|
|
|
|
* memory.
|
|
|
|
*
|
|
|
|
* Hence only allow the shrinker to scan the entire cache when
|
|
|
|
* a large delta change is calculated directly.
|
|
|
|
*/
|
|
|
|
if (delta < max_pass / 4)
|
|
|
|
total_scan = min(total_scan, max_pass / 2);
|
|
|
|
|
2005-11-29 04:44:15 +07:00
|
|
|
/*
|
|
|
|
* Avoid risking looping forever due to too large nr value:
|
|
|
|
* never try to free more than twice the estimate number of
|
|
|
|
* freeable entries.
|
|
|
|
*/
|
2011-07-08 11:14:35 +07:00
|
|
|
if (total_scan > max_pass * 2)
|
|
|
|
total_scan = max_pass * 2;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-07-08 11:14:35 +07:00
|
|
|
trace_mm_shrink_slab_start(shrinker, shrink, nr,
|
2011-07-08 11:14:34 +07:00
|
|
|
nr_pages_scanned, lru_pages,
|
|
|
|
max_pass, delta, total_scan);
|
|
|
|
|
2011-07-08 11:14:37 +07:00
|
|
|
while (total_scan >= batch_size) {
|
2005-06-22 07:14:35 +07:00
|
|
|
int nr_before;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-05-25 07:12:27 +07:00
|
|
|
nr_before = do_shrinker_shrink(shrinker, shrink, 0);
|
|
|
|
shrink_ret = do_shrinker_shrink(shrinker, shrink,
|
2011-07-08 11:14:37 +07:00
|
|
|
batch_size);
|
2005-04-17 05:20:36 +07:00
|
|
|
if (shrink_ret == -1)
|
|
|
|
break;
|
2005-06-22 07:14:35 +07:00
|
|
|
if (shrink_ret < nr_before)
|
|
|
|
ret += nr_before - shrink_ret;
|
2011-07-08 11:14:37 +07:00
|
|
|
count_vm_events(SLABS_SCANNED, batch_size);
|
|
|
|
total_scan -= batch_size;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
cond_resched();
|
|
|
|
}
|
|
|
|
|
2011-07-08 11:14:35 +07:00
|
|
|
/*
|
|
|
|
* move the unused scan count back into the shrinker in a
|
|
|
|
* manner that handles concurrent updates. If we exhausted the
|
|
|
|
* scan, there is no need to do an update.
|
|
|
|
*/
|
2011-12-09 05:33:54 +07:00
|
|
|
if (total_scan > 0)
|
|
|
|
new_nr = atomic_long_add_return(total_scan,
|
|
|
|
&shrinker->nr_in_batch);
|
|
|
|
else
|
|
|
|
new_nr = atomic_long_read(&shrinker->nr_in_batch);
|
2011-07-08 11:14:35 +07:00
|
|
|
|
|
|
|
trace_mm_shrink_slab_end(shrinker, shrink_ret, nr, new_nr);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
up_read(&shrinker_rwsem);
|
2011-05-25 07:11:11 +07:00
|
|
|
out:
|
|
|
|
cond_resched();
|
2005-06-22 07:14:35 +07:00
|
|
|
return ret;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2011-01-14 06:46:00 +07:00
|
|
|
static void set_reclaim_mode(int priority, struct scan_control *sc,
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
bool sync)
|
|
|
|
{
|
2011-01-14 06:46:00 +07:00
|
|
|
reclaim_mode_t syncmode = sync ? RECLAIM_MODE_SYNC : RECLAIM_MODE_ASYNC;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
|
|
|
|
/*
|
2011-01-14 06:45:56 +07:00
|
|
|
* Initially assume we are entering either lumpy reclaim or
|
|
|
|
* reclaim/compaction.Depending on the order, we will either set the
|
|
|
|
* sync mode or just reclaim order-0 pages later.
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
*/
|
2011-01-14 06:45:56 +07:00
|
|
|
if (COMPACTION_BUILD)
|
2011-01-14 06:46:00 +07:00
|
|
|
sc->reclaim_mode = RECLAIM_MODE_COMPACTION;
|
2011-01-14 06:45:56 +07:00
|
|
|
else
|
2011-01-14 06:46:00 +07:00
|
|
|
sc->reclaim_mode = RECLAIM_MODE_LUMPYRECLAIM;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
|
|
|
|
/*
|
2011-01-14 06:45:56 +07:00
|
|
|
* Avoid using lumpy reclaim or reclaim/compaction if possible by
|
|
|
|
* restricting when its set to either costly allocations or when
|
|
|
|
* under memory pressure
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
*/
|
|
|
|
if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
|
2011-01-14 06:46:00 +07:00
|
|
|
sc->reclaim_mode |= syncmode;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
else if (sc->order && priority < DEF_PRIORITY - 2)
|
2011-01-14 06:46:00 +07:00
|
|
|
sc->reclaim_mode |= syncmode;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
else
|
2011-01-14 06:46:00 +07:00
|
|
|
sc->reclaim_mode = RECLAIM_MODE_SINGLE | RECLAIM_MODE_ASYNC;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
}
|
|
|
|
|
2011-01-14 06:46:00 +07:00
|
|
|
static void reset_reclaim_mode(struct scan_control *sc)
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
{
|
2011-01-14 06:46:00 +07:00
|
|
|
sc->reclaim_mode = RECLAIM_MODE_SINGLE | RECLAIM_MODE_ASYNC;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
static inline int is_page_cache_freeable(struct page *page)
|
|
|
|
{
|
2009-09-22 07:03:00 +07:00
|
|
|
/*
|
|
|
|
* A freeable page cache page is referenced only by the caller
|
|
|
|
* that isolated the page, the page cache radix tree and
|
|
|
|
* optional buffer heads at page->private.
|
|
|
|
*/
|
2009-09-22 07:02:59 +07:00
|
|
|
return page_count(page) - page_has_private(page) == 2;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
static int may_write_to_queue(struct backing_dev_info *bdi,
|
|
|
|
struct scan_control *sc)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2006-01-08 16:00:47 +07:00
|
|
|
if (current->flags & PF_SWAPWRITE)
|
2005-04-17 05:20:36 +07:00
|
|
|
return 1;
|
|
|
|
if (!bdi_write_congested(bdi))
|
|
|
|
return 1;
|
|
|
|
if (bdi == current->backing_dev_info)
|
|
|
|
return 1;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
|
|
|
|
/* lumpy reclaim for hugepage often need a lot of write */
|
|
|
|
if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
|
|
|
|
return 1;
|
2005-04-17 05:20:36 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We detected a synchronous write error writing a page out. Probably
|
|
|
|
* -ENOSPC. We need to propagate that into the address_space for a subsequent
|
|
|
|
* fsync(), msync() or close().
|
|
|
|
*
|
|
|
|
* The tricky part is that after writepage we cannot touch the mapping: nothing
|
|
|
|
* prevents it from being freed up. But we have a ref on the page and once
|
|
|
|
* that page is locked, the mapping is pinned.
|
|
|
|
*
|
|
|
|
* We're allowed to run sleeping lock_page() here because we know the caller has
|
|
|
|
* __GFP_FS.
|
|
|
|
*/
|
|
|
|
static void handle_write_error(struct address_space *mapping,
|
|
|
|
struct page *page, int error)
|
|
|
|
{
|
2011-03-10 14:52:07 +07:00
|
|
|
lock_page(page);
|
2007-05-08 14:23:25 +07:00
|
|
|
if (page_mapping(page) == mapping)
|
|
|
|
mapping_set_error(mapping, error);
|
2005-04-17 05:20:36 +07:00
|
|
|
unlock_page(page);
|
|
|
|
}
|
|
|
|
|
2006-06-23 16:03:38 +07:00
|
|
|
/* possible outcome of pageout() */
|
|
|
|
typedef enum {
|
|
|
|
/* failed to write page out, page is locked */
|
|
|
|
PAGE_KEEP,
|
|
|
|
/* move page to the active list, page is locked */
|
|
|
|
PAGE_ACTIVATE,
|
|
|
|
/* page has been sent to the disk successfully, page is unlocked */
|
|
|
|
PAGE_SUCCESS,
|
|
|
|
/* page is clean and locked */
|
|
|
|
PAGE_CLEAN,
|
|
|
|
} pageout_t;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
[PATCH] vmscan: rename functions
We have:
try_to_free_pages
->shrink_caches(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_cache(struct zone *, ...)
->shrink_list(struct list_head *, ...)
->refill_inactive_list((struct zone *, ...)
which is fairly irrational.
Rename things so that we have
try_to_free_pages
->shrink_zones(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_inactive_list(struct zone *, ...)
->shrink_page_list(struct list_head *, ...)
->shrink_active_list(struct zone *, ...)
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-22 15:08:21 +07:00
|
|
|
* pageout is called by shrink_page_list() for each dirty page.
|
|
|
|
* Calls ->writepage().
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2007-08-23 04:01:26 +07:00
|
|
|
static pageout_t pageout(struct page *page, struct address_space *mapping,
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
struct scan_control *sc)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
/*
|
|
|
|
* If the page is dirty, only perform writeback if that write
|
|
|
|
* will be non-blocking. To prevent this allocation from being
|
|
|
|
* stalled by pagecache activity. But note that there may be
|
|
|
|
* stalls if we need to run get_block(). We could test
|
|
|
|
* PagePrivate for that.
|
|
|
|
*
|
2009-12-15 08:58:49 +07:00
|
|
|
* If this process is currently in __generic_file_aio_write() against
|
2005-04-17 05:20:36 +07:00
|
|
|
* this page's queue, we can perform writeback even if that
|
|
|
|
* will block.
|
|
|
|
*
|
|
|
|
* If the page is swapcache, write it back even if that would
|
|
|
|
* block, for some throttling. This happens by accident, because
|
|
|
|
* swap_backing_dev_info is bust: it doesn't reflect the
|
|
|
|
* congestion state of the swapdevs. Easy to fix, if needed.
|
|
|
|
*/
|
|
|
|
if (!is_page_cache_freeable(page))
|
|
|
|
return PAGE_KEEP;
|
|
|
|
if (!mapping) {
|
|
|
|
/*
|
|
|
|
* Some data journaling orphaned pages can have
|
|
|
|
* page->mapping == NULL while being dirty with clean buffers.
|
|
|
|
*/
|
2009-04-03 22:42:36 +07:00
|
|
|
if (page_has_private(page)) {
|
2005-04-17 05:20:36 +07:00
|
|
|
if (try_to_free_buffers(page)) {
|
|
|
|
ClearPageDirty(page);
|
2008-04-30 14:55:07 +07:00
|
|
|
printk("%s: orphaned page\n", __func__);
|
2005-04-17 05:20:36 +07:00
|
|
|
return PAGE_CLEAN;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return PAGE_KEEP;
|
|
|
|
}
|
|
|
|
if (mapping->a_ops->writepage == NULL)
|
|
|
|
return PAGE_ACTIVATE;
|
2010-10-27 04:21:45 +07:00
|
|
|
if (!may_write_to_queue(mapping->backing_dev_info, sc))
|
2005-04-17 05:20:36 +07:00
|
|
|
return PAGE_KEEP;
|
|
|
|
|
|
|
|
if (clear_page_dirty_for_io(page)) {
|
|
|
|
int res;
|
|
|
|
struct writeback_control wbc = {
|
|
|
|
.sync_mode = WB_SYNC_NONE,
|
|
|
|
.nr_to_write = SWAP_CLUSTER_MAX,
|
[PATCH] writeback: fix range handling
When a writeback_control's `start' and `end' fields are used to
indicate a one-byte-range starting at file offset zero, the required
values of .start=0,.end=0 mean that the ->writepages() implementation
has no way of telling that it is being asked to perform a range
request. Because we're currently overloading (start == 0 && end == 0)
to mean "this is not a write-a-range request".
To make all this sane, the patch changes range of writeback_control.
So caller does: If it is calling ->writepages() to write pages, it
sets range (range_start/end or range_cyclic) always.
And if range_cyclic is true, ->writepages() thinks the range is
cyclic, otherwise it just uses range_start and range_end.
This patch does,
- Add LLONG_MAX, LLONG_MIN, ULLONG_MAX to include/linux/kernel.h
-1 is usually ok for range_end (type is long long). But, if someone did,
range_end += val; range_end is "val - 1"
u64val = range_end >> bits; u64val is "~(0ULL)"
or something, they are wrong. So, this adds LLONG_MAX to avoid nasty
things, and uses LLONG_MAX for range_end.
- All callers of ->writepages() sets range_start/end or range_cyclic.
- Fix updates of ->writeback_index. It seems already bit strange.
If it starts at 0 and ended by check of nr_to_write, this last
index may reduce chance to scan end of file. So, this updates
->writeback_index only if range_cyclic is true or whole-file is
scanned.
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Nathan Scott <nathans@sgi.com>
Cc: Anton Altaparmakov <aia21@cantab.net>
Cc: Steven French <sfrench@us.ibm.com>
Cc: "Vladimir V. Saveliev" <vs@namesys.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-23 16:03:26 +07:00
|
|
|
.range_start = 0,
|
|
|
|
.range_end = LLONG_MAX,
|
2005-04-17 05:20:36 +07:00
|
|
|
.for_reclaim = 1,
|
|
|
|
};
|
|
|
|
|
|
|
|
SetPageReclaim(page);
|
|
|
|
res = mapping->a_ops->writepage(page, &wbc);
|
|
|
|
if (res < 0)
|
|
|
|
handle_write_error(mapping, page, res);
|
2005-12-16 05:28:17 +07:00
|
|
|
if (res == AOP_WRITEPAGE_ACTIVATE) {
|
2005-04-17 05:20:36 +07:00
|
|
|
ClearPageReclaim(page);
|
|
|
|
return PAGE_ACTIVATE;
|
|
|
|
}
|
2007-08-23 04:01:26 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
if (!PageWriteback(page)) {
|
|
|
|
/* synchronous write or broken a_ops? */
|
|
|
|
ClearPageReclaim(page);
|
|
|
|
}
|
2010-08-10 07:19:18 +07:00
|
|
|
trace_mm_vmscan_writepage(page,
|
2011-01-14 06:46:00 +07:00
|
|
|
trace_reclaim_flags(page, sc->reclaim_mode));
|
2006-09-27 15:50:00 +07:00
|
|
|
inc_zone_page_state(page, NR_VMSCAN_WRITE);
|
2005-04-17 05:20:36 +07:00
|
|
|
return PAGE_SUCCESS;
|
|
|
|
}
|
|
|
|
|
|
|
|
return PAGE_CLEAN;
|
|
|
|
}
|
|
|
|
|
2006-10-17 14:09:36 +07:00
|
|
|
/*
|
2008-07-26 09:45:30 +07:00
|
|
|
* Same as remove_mapping, but if the page is removed from the mapping, it
|
|
|
|
* gets returned with a refcount of 0.
|
2006-10-17 14:09:36 +07:00
|
|
|
*/
|
2008-07-26 09:45:30 +07:00
|
|
|
static int __remove_mapping(struct address_space *mapping, struct page *page)
|
2006-01-08 16:00:48 +07:00
|
|
|
{
|
2006-09-26 13:31:23 +07:00
|
|
|
BUG_ON(!PageLocked(page));
|
|
|
|
BUG_ON(mapping != page_mapping(page));
|
2006-01-08 16:00:48 +07:00
|
|
|
|
2008-07-26 09:45:32 +07:00
|
|
|
spin_lock_irq(&mapping->tree_lock);
|
2006-01-08 16:00:48 +07:00
|
|
|
/*
|
2006-09-27 15:50:02 +07:00
|
|
|
* The non racy check for a busy page.
|
|
|
|
*
|
|
|
|
* Must be careful with the order of the tests. When someone has
|
|
|
|
* a ref to the page, it may be possible that they dirty it then
|
|
|
|
* drop the reference. So if PageDirty is tested before page_count
|
|
|
|
* here, then the following race may occur:
|
|
|
|
*
|
|
|
|
* get_user_pages(&page);
|
|
|
|
* [user mapping goes away]
|
|
|
|
* write_to(page);
|
|
|
|
* !PageDirty(page) [good]
|
|
|
|
* SetPageDirty(page);
|
|
|
|
* put_page(page);
|
|
|
|
* !page_count(page) [good, discard it]
|
|
|
|
*
|
|
|
|
* [oops, our write_to data is lost]
|
|
|
|
*
|
|
|
|
* Reversing the order of the tests ensures such a situation cannot
|
|
|
|
* escape unnoticed. The smp_rmb is needed to ensure the page->flags
|
|
|
|
* load is not satisfied before that of page->_count.
|
|
|
|
*
|
|
|
|
* Note that if SetPageDirty is always performed via set_page_dirty,
|
|
|
|
* and thus under tree_lock, then this ordering is not required.
|
2006-01-08 16:00:48 +07:00
|
|
|
*/
|
2008-07-26 09:45:30 +07:00
|
|
|
if (!page_freeze_refs(page, 2))
|
2006-01-08 16:00:48 +07:00
|
|
|
goto cannot_free;
|
2008-07-26 09:45:30 +07:00
|
|
|
/* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */
|
|
|
|
if (unlikely(PageDirty(page))) {
|
|
|
|
page_unfreeze_refs(page, 2);
|
2006-01-08 16:00:48 +07:00
|
|
|
goto cannot_free;
|
2008-07-26 09:45:30 +07:00
|
|
|
}
|
2006-01-08 16:00:48 +07:00
|
|
|
|
|
|
|
if (PageSwapCache(page)) {
|
|
|
|
swp_entry_t swap = { .val = page_private(page) };
|
|
|
|
__delete_from_swap_cache(page);
|
2008-07-26 09:45:32 +07:00
|
|
|
spin_unlock_irq(&mapping->tree_lock);
|
2009-06-17 05:32:52 +07:00
|
|
|
swapcache_free(swap, page);
|
2008-07-26 09:45:30 +07:00
|
|
|
} else {
|
2010-12-02 01:35:19 +07:00
|
|
|
void (*freepage)(struct page *);
|
|
|
|
|
|
|
|
freepage = mapping->a_ops->freepage;
|
|
|
|
|
2011-03-23 06:32:44 +07:00
|
|
|
__delete_from_page_cache(page);
|
2008-07-26 09:45:32 +07:00
|
|
|
spin_unlock_irq(&mapping->tree_lock);
|
2009-05-29 04:34:28 +07:00
|
|
|
mem_cgroup_uncharge_cache_page(page);
|
2010-12-02 01:35:19 +07:00
|
|
|
|
|
|
|
if (freepage != NULL)
|
|
|
|
freepage(page);
|
2006-01-08 16:00:48 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
cannot_free:
|
2008-07-26 09:45:32 +07:00
|
|
|
spin_unlock_irq(&mapping->tree_lock);
|
2006-01-08 16:00:48 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-07-26 09:45:30 +07:00
|
|
|
/*
|
|
|
|
* Attempt to detach a locked page from its ->mapping. If it is dirty or if
|
|
|
|
* someone else has a ref on the page, abort and return 0. If it was
|
|
|
|
* successfully detached, return 1. Assumes the caller has a single ref on
|
|
|
|
* this page.
|
|
|
|
*/
|
|
|
|
int remove_mapping(struct address_space *mapping, struct page *page)
|
|
|
|
{
|
|
|
|
if (__remove_mapping(mapping, page)) {
|
|
|
|
/*
|
|
|
|
* Unfreezing the refcount with 1 rather than 2 effectively
|
|
|
|
* drops the pagecache ref for us without requiring another
|
|
|
|
* atomic operation.
|
|
|
|
*/
|
|
|
|
page_unfreeze_refs(page, 1);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
/**
|
|
|
|
* putback_lru_page - put previously isolated page onto appropriate LRU list
|
|
|
|
* @page: page to be put back to appropriate lru list
|
|
|
|
*
|
|
|
|
* Add previously isolated @page to appropriate LRU list.
|
|
|
|
* Page may still be unevictable for other reasons.
|
|
|
|
*
|
|
|
|
* lru_lock must not be held, interrupts must be enabled.
|
|
|
|
*/
|
|
|
|
void putback_lru_page(struct page *page)
|
|
|
|
{
|
|
|
|
int lru;
|
|
|
|
int active = !!TestClearPageActive(page);
|
2008-10-19 10:26:40 +07:00
|
|
|
int was_unevictable = PageUnevictable(page);
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
|
|
|
|
VM_BUG_ON(PageLRU(page));
|
|
|
|
|
|
|
|
redo:
|
|
|
|
ClearPageUnevictable(page);
|
|
|
|
|
|
|
|
if (page_evictable(page, NULL)) {
|
|
|
|
/*
|
|
|
|
* For evictable pages, we can use the cache.
|
|
|
|
* In event of a race, worst case is we end up with an
|
|
|
|
* unevictable page on [in]active list.
|
|
|
|
* We know how to handle that.
|
|
|
|
*/
|
2009-09-22 07:02:58 +07:00
|
|
|
lru = active + page_lru_base_type(page);
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
lru_cache_add_lru(page, lru);
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Put unevictable pages directly on zone's unevictable
|
|
|
|
* list.
|
|
|
|
*/
|
|
|
|
lru = LRU_UNEVICTABLE;
|
|
|
|
add_page_to_unevictable_list(page);
|
2009-10-27 06:50:00 +07:00
|
|
|
/*
|
2011-11-01 07:09:28 +07:00
|
|
|
* When racing with an mlock or AS_UNEVICTABLE clearing
|
|
|
|
* (page is unlocked) make sure that if the other thread
|
|
|
|
* does not observe our setting of PG_lru and fails
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
* isolation/check_move_unevictable_pages,
|
2011-11-01 07:09:28 +07:00
|
|
|
* we see PG_mlocked/AS_UNEVICTABLE cleared below and move
|
2009-10-27 06:50:00 +07:00
|
|
|
* the page back to the evictable list.
|
|
|
|
*
|
2011-11-01 07:09:28 +07:00
|
|
|
* The other side is TestClearPageMlocked() or shmem_lock().
|
2009-10-27 06:50:00 +07:00
|
|
|
*/
|
|
|
|
smp_mb();
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* page's status can change while we move it among lru. If an evictable
|
|
|
|
* page is on unevictable list, it never be freed. To avoid that,
|
|
|
|
* check after we added it to the list, again.
|
|
|
|
*/
|
|
|
|
if (lru == LRU_UNEVICTABLE && page_evictable(page, NULL)) {
|
|
|
|
if (!isolate_lru_page(page)) {
|
|
|
|
put_page(page);
|
|
|
|
goto redo;
|
|
|
|
}
|
|
|
|
/* This means someone else dropped this page from LRU
|
|
|
|
* So, it will be freed or putback to LRU again. There is
|
|
|
|
* nothing to do here.
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
|
2008-10-19 10:26:40 +07:00
|
|
|
if (was_unevictable && lru != LRU_UNEVICTABLE)
|
|
|
|
count_vm_event(UNEVICTABLE_PGRESCUED);
|
|
|
|
else if (!was_unevictable && lru == LRU_UNEVICTABLE)
|
|
|
|
count_vm_event(UNEVICTABLE_PGCULLED);
|
|
|
|
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
put_page(page); /* drop ref from isolate */
|
|
|
|
}
|
|
|
|
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
enum page_references {
|
|
|
|
PAGEREF_RECLAIM,
|
|
|
|
PAGEREF_RECLAIM_CLEAN,
|
2010-03-06 04:42:22 +07:00
|
|
|
PAGEREF_KEEP,
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
PAGEREF_ACTIVATE,
|
|
|
|
};
|
|
|
|
|
|
|
|
static enum page_references page_check_references(struct page *page,
|
2012-01-13 08:17:52 +07:00
|
|
|
struct mem_cgroup_zone *mz,
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
struct scan_control *sc)
|
|
|
|
{
|
2010-03-06 04:42:22 +07:00
|
|
|
int referenced_ptes, referenced_page;
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
unsigned long vm_flags;
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
referenced_ptes = page_referenced(page, 1, mz->mem_cgroup, &vm_flags);
|
2010-03-06 04:42:22 +07:00
|
|
|
referenced_page = TestClearPageReferenced(page);
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
|
|
|
|
/* Lumpy reclaim - ignore references */
|
2011-01-14 06:46:00 +07:00
|
|
|
if (sc->reclaim_mode & RECLAIM_MODE_LUMPYRECLAIM)
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
return PAGEREF_RECLAIM;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Mlock lost the isolation race with us. Let try_to_unmap()
|
|
|
|
* move the page to the unevictable list.
|
|
|
|
*/
|
|
|
|
if (vm_flags & VM_LOCKED)
|
|
|
|
return PAGEREF_RECLAIM;
|
|
|
|
|
2010-03-06 04:42:22 +07:00
|
|
|
if (referenced_ptes) {
|
|
|
|
if (PageAnon(page))
|
|
|
|
return PAGEREF_ACTIVATE;
|
|
|
|
/*
|
|
|
|
* All mapped pages start out with page table
|
|
|
|
* references from the instantiating fault, so we need
|
|
|
|
* to look twice if a mapped file page is used more
|
|
|
|
* than once.
|
|
|
|
*
|
|
|
|
* Mark it and spare it for another trip around the
|
|
|
|
* inactive list. Another page table reference will
|
|
|
|
* lead to its activation.
|
|
|
|
*
|
|
|
|
* Note: the mark is set for activated pages as well
|
|
|
|
* so that recently deactivated but used pages are
|
|
|
|
* quickly recovered.
|
|
|
|
*/
|
|
|
|
SetPageReferenced(page);
|
|
|
|
|
2012-01-11 06:06:59 +07:00
|
|
|
if (referenced_page || referenced_ptes > 1)
|
2010-03-06 04:42:22 +07:00
|
|
|
return PAGEREF_ACTIVATE;
|
|
|
|
|
2012-01-11 06:07:03 +07:00
|
|
|
/*
|
|
|
|
* Activate file-backed executable pages after first usage.
|
|
|
|
*/
|
|
|
|
if (vm_flags & VM_EXEC)
|
|
|
|
return PAGEREF_ACTIVATE;
|
|
|
|
|
2010-03-06 04:42:22 +07:00
|
|
|
return PAGEREF_KEEP;
|
|
|
|
}
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
|
|
|
|
/* Reclaim if clean, defer dirty pages to writeback */
|
2010-10-27 04:21:46 +07:00
|
|
|
if (referenced_page && !PageSwapBacked(page))
|
2010-03-06 04:42:22 +07:00
|
|
|
return PAGEREF_RECLAIM_CLEAN;
|
|
|
|
|
|
|
|
return PAGEREF_RECLAIM;
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
[PATCH] vmscan: rename functions
We have:
try_to_free_pages
->shrink_caches(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_cache(struct zone *, ...)
->shrink_list(struct list_head *, ...)
->refill_inactive_list((struct zone *, ...)
which is fairly irrational.
Rename things so that we have
try_to_free_pages
->shrink_zones(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_inactive_list(struct zone *, ...)
->shrink_page_list(struct list_head *, ...)
->shrink_active_list(struct zone *, ...)
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-22 15:08:21 +07:00
|
|
|
* shrink_page_list() returns the number of reclaimed pages
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
[PATCH] vmscan: rename functions
We have:
try_to_free_pages
->shrink_caches(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_cache(struct zone *, ...)
->shrink_list(struct list_head *, ...)
->refill_inactive_list((struct zone *, ...)
which is fairly irrational.
Rename things so that we have
try_to_free_pages
->shrink_zones(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_inactive_list(struct zone *, ...)
->shrink_page_list(struct list_head *, ...)
->shrink_active_list(struct zone *, ...)
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-22 15:08:21 +07:00
|
|
|
static unsigned long shrink_page_list(struct list_head *page_list,
|
2012-01-13 08:17:52 +07:00
|
|
|
struct mem_cgroup_zone *mz,
|
2011-11-01 07:07:51 +07:00
|
|
|
struct scan_control *sc,
|
2011-11-01 07:07:56 +07:00
|
|
|
int priority,
|
|
|
|
unsigned long *ret_nr_dirty,
|
|
|
|
unsigned long *ret_nr_writeback)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
LIST_HEAD(ret_pages);
|
2010-08-10 07:19:31 +07:00
|
|
|
LIST_HEAD(free_pages);
|
2005-04-17 05:20:36 +07:00
|
|
|
int pgactivate = 0;
|
2010-10-27 04:21:45 +07:00
|
|
|
unsigned long nr_dirty = 0;
|
|
|
|
unsigned long nr_congested = 0;
|
2006-03-22 15:08:20 +07:00
|
|
|
unsigned long nr_reclaimed = 0;
|
2011-11-01 07:07:56 +07:00
|
|
|
unsigned long nr_writeback = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
cond_resched();
|
|
|
|
|
|
|
|
while (!list_empty(page_list)) {
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
enum page_references references;
|
2005-04-17 05:20:36 +07:00
|
|
|
struct address_space *mapping;
|
|
|
|
struct page *page;
|
|
|
|
int may_enter_fs;
|
|
|
|
|
|
|
|
cond_resched();
|
|
|
|
|
|
|
|
page = lru_to_page(page_list);
|
|
|
|
list_del(&page->lru);
|
|
|
|
|
2008-08-02 17:01:03 +07:00
|
|
|
if (!trylock_page(page))
|
2005-04-17 05:20:36 +07:00
|
|
|
goto keep;
|
|
|
|
|
2006-09-26 13:30:55 +07:00
|
|
|
VM_BUG_ON(PageActive(page));
|
2012-01-13 08:17:52 +07:00
|
|
|
VM_BUG_ON(page_zone(page) != mz->zone);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
sc->nr_scanned++;
|
2006-02-12 08:55:53 +07:00
|
|
|
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
if (unlikely(!page_evictable(page, NULL)))
|
|
|
|
goto cull_mlocked;
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
|
2009-04-01 05:19:30 +07:00
|
|
|
if (!sc->may_unmap && page_mapped(page))
|
2006-02-12 08:55:53 +07:00
|
|
|
goto keep_locked;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Double the slab pressure for mapped and swapcache pages */
|
|
|
|
if (page_mapped(page) || PageSwapCache(page))
|
|
|
|
sc->nr_scanned++;
|
|
|
|
|
2007-08-23 04:01:26 +07:00
|
|
|
may_enter_fs = (sc->gfp_mask & __GFP_FS) ||
|
|
|
|
(PageSwapCache(page) && (sc->gfp_mask & __GFP_IO));
|
|
|
|
|
|
|
|
if (PageWriteback(page)) {
|
2011-11-01 07:07:56 +07:00
|
|
|
nr_writeback++;
|
2007-08-23 04:01:26 +07:00
|
|
|
/*
|
2011-11-01 07:07:42 +07:00
|
|
|
* Synchronous reclaim cannot queue pages for
|
|
|
|
* writeback due to the possibility of stack overflow
|
|
|
|
* but if it encounters a page under writeback, wait
|
|
|
|
* for the IO to complete.
|
2007-08-23 04:01:26 +07:00
|
|
|
*/
|
2011-01-14 06:46:00 +07:00
|
|
|
if ((sc->reclaim_mode & RECLAIM_MODE_SYNC) &&
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
may_enter_fs)
|
2007-08-23 04:01:26 +07:00
|
|
|
wait_on_page_writeback(page);
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
else {
|
|
|
|
unlock_page(page);
|
|
|
|
goto keep_lumpy;
|
|
|
|
}
|
2007-08-23 04:01:26 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
references = page_check_references(page, mz, sc);
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
switch (references) {
|
|
|
|
case PAGEREF_ACTIVATE:
|
2005-04-17 05:20:36 +07:00
|
|
|
goto activate_locked;
|
2010-03-06 04:42:22 +07:00
|
|
|
case PAGEREF_KEEP:
|
|
|
|
goto keep_locked;
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
case PAGEREF_RECLAIM:
|
|
|
|
case PAGEREF_RECLAIM_CLEAN:
|
|
|
|
; /* try to reclaim the page below */
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Anonymous process memory has backing store?
|
|
|
|
* Try to allocate it some swap space here.
|
|
|
|
*/
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
if (PageAnon(page) && !PageSwapCache(page)) {
|
2008-11-20 06:36:37 +07:00
|
|
|
if (!(sc->gfp_mask & __GFP_IO))
|
|
|
|
goto keep_locked;
|
2009-01-07 05:39:39 +07:00
|
|
|
if (!add_to_swap(page))
|
2005-04-17 05:20:36 +07:00
|
|
|
goto activate_locked;
|
2008-11-20 06:36:37 +07:00
|
|
|
may_enter_fs = 1;
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
mapping = page_mapping(page);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The page is mapped into the page tables of one or more
|
|
|
|
* processes. Try to unmap it here.
|
|
|
|
*/
|
|
|
|
if (page_mapped(page) && mapping) {
|
2009-09-16 16:50:10 +07:00
|
|
|
switch (try_to_unmap(page, TTU_UNMAP)) {
|
2005-04-17 05:20:36 +07:00
|
|
|
case SWAP_FAIL:
|
|
|
|
goto activate_locked;
|
|
|
|
case SWAP_AGAIN:
|
|
|
|
goto keep_locked;
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
case SWAP_MLOCK:
|
|
|
|
goto cull_mlocked;
|
2005-04-17 05:20:36 +07:00
|
|
|
case SWAP_SUCCESS:
|
|
|
|
; /* try to free the page below */
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (PageDirty(page)) {
|
2010-10-27 04:21:45 +07:00
|
|
|
nr_dirty++;
|
|
|
|
|
mm: vmscan: do not writeback filesystem pages in direct reclaim
Testing from the XFS folk revealed that there is still too much I/O from
the end of the LRU in kswapd. Previously it was considered acceptable by
VM people for a small number of pages to be written back from reclaim with
testing generally showing about 0.3% of pages reclaimed were written back
(higher if memory was low). That writing back a small number of pages is
ok has been heavily disputed for quite some time and Dave Chinner
explained it well;
It doesn't have to be a very high number to be a problem. IO
is orders of magnitude slower than the CPU time it takes to
flush a page, so the cost of making a bad flush decision is
very high. And single page writeback from the LRU is almost
always a bad flush decision.
To complicate matters, filesystems respond very differently to requests
from reclaim according to Christoph Hellwig;
xfs tries to write it back if the requester is kswapd
ext4 ignores the request if it's a delayed allocation
btrfs ignores the request
As a result, each filesystem has different performance characteristics
when under memory pressure and there are many pages being dirtied. In
some cases, the request is ignored entirely so the VM cannot depend on the
IO being dispatched.
The objective of this series is to reduce writing of filesystem-backed
pages from reclaim, play nicely with writeback that is already in progress
and throttle reclaim appropriately when writeback pages are encountered.
The assumption is that the flushers will always write pages faster than if
reclaim issues the IO.
A secondary goal is to avoid the problem whereby direct reclaim splices
two potentially deep call stacks together.
There is a potential new problem as reclaim has less control over how long
before a page in a particularly zone or container is cleaned and direct
reclaimers depend on kswapd or flusher threads to do the necessary work.
However, as filesystems sometimes ignore direct reclaim requests already,
it is not expected to be a serious issue.
Patch 1 disables writeback of filesystem pages from direct reclaim
entirely. Anonymous pages are still written.
Patch 2 removes dead code in lumpy reclaim as it is no longer able
to synchronously write pages. This hurts lumpy reclaim but
there is an expectation that compaction is used for hugepage
allocations these days and lumpy reclaim's days are numbered.
Patches 3-4 add warnings to XFS and ext4 if called from
direct reclaim. With patch 1, this "never happens" and is
intended to catch regressions in this logic in the future.
Patch 5 disables writeback of filesystem pages from kswapd unless
the priority is raised to the point where kswapd is considered
to be in trouble.
Patch 6 throttles reclaimers if too many dirty pages are being
encountered and the zones or backing devices are congested.
Patch 7 invalidates dirty pages found at the end of the LRU so they
are reclaimed quickly after being written back rather than
waiting for a reclaimer to find them
I consider this series to be orthogonal to the writeback work but it is
worth noting that the writeback work affects the viability of patch 8 in
particular.
I tested this on ext4 and xfs using fs_mark, a simple writeback test based
on dd and a micro benchmark that does a streaming write to a large mapping
(exercises use-once LRU logic) followed by streaming writes to a mix of
anonymous and file-backed mappings. The command line for fs_mark when
botted with 512M looked something like
./fs_mark -d /tmp/fsmark-2676 -D 100 -N 150 -n 150 -L 25 -t 1 -S0 -s 10485760
The number of files was adjusted depending on the amount of available
memory so that the files created was about 3xRAM. For multiple threads,
the -d switch is specified multiple times.
The test machine is x86-64 with an older generation of AMD processor with
4 cores. The underlying storage was 4 disks configured as RAID-0 as this
was the best configuration of storage I had available. Swap is on a
separate disk. Dirty ratio was tuned to 40% instead of the default of
20%.
Testing was run with and without monitors to both verify that the patches
were operating as expected and that any performance gain was real and not
due to interference from monitors.
Here is a summary of results based on testing XFS.
512M1P-xfs Files/s mean 32.69 ( 0.00%) 34.44 ( 5.08%)
512M1P-xfs Elapsed Time fsmark 51.41 48.29
512M1P-xfs Elapsed Time simple-wb 114.09 108.61
512M1P-xfs Elapsed Time mmap-strm 113.46 109.34
512M1P-xfs Kswapd efficiency fsmark 62% 63%
512M1P-xfs Kswapd efficiency simple-wb 56% 61%
512M1P-xfs Kswapd efficiency mmap-strm 44% 42%
512M-xfs Files/s mean 30.78 ( 0.00%) 35.94 (14.36%)
512M-xfs Elapsed Time fsmark 56.08 48.90
512M-xfs Elapsed Time simple-wb 112.22 98.13
512M-xfs Elapsed Time mmap-strm 219.15 196.67
512M-xfs Kswapd efficiency fsmark 54% 56%
512M-xfs Kswapd efficiency simple-wb 54% 55%
512M-xfs Kswapd efficiency mmap-strm 45% 44%
512M-4X-xfs Files/s mean 30.31 ( 0.00%) 33.33 ( 9.06%)
512M-4X-xfs Elapsed Time fsmark 63.26 55.88
512M-4X-xfs Elapsed Time simple-wb 100.90 90.25
512M-4X-xfs Elapsed Time mmap-strm 261.73 255.38
512M-4X-xfs Kswapd efficiency fsmark 49% 50%
512M-4X-xfs Kswapd efficiency simple-wb 54% 56%
512M-4X-xfs Kswapd efficiency mmap-strm 37% 36%
512M-16X-xfs Files/s mean 60.89 ( 0.00%) 65.22 ( 6.64%)
512M-16X-xfs Elapsed Time fsmark 67.47 58.25
512M-16X-xfs Elapsed Time simple-wb 103.22 90.89
512M-16X-xfs Elapsed Time mmap-strm 237.09 198.82
512M-16X-xfs Kswapd efficiency fsmark 45% 46%
512M-16X-xfs Kswapd efficiency simple-wb 53% 55%
512M-16X-xfs Kswapd efficiency mmap-strm 33% 33%
Up until 512-4X, the FSmark improvements were statistically significant.
For the 4X and 16X tests the results were within standard deviations but
just barely. The time to completion for all tests is improved which is an
important result. In general, kswapd efficiency is not affected by
skipping dirty pages.
1024M1P-xfs Files/s mean 39.09 ( 0.00%) 41.15 ( 5.01%)
1024M1P-xfs Elapsed Time fsmark 84.14 80.41
1024M1P-xfs Elapsed Time simple-wb 210.77 184.78
1024M1P-xfs Elapsed Time mmap-strm 162.00 160.34
1024M1P-xfs Kswapd efficiency fsmark 69% 75%
1024M1P-xfs Kswapd efficiency simple-wb 71% 77%
1024M1P-xfs Kswapd efficiency mmap-strm 43% 44%
1024M-xfs Files/s mean 35.45 ( 0.00%) 37.00 ( 4.19%)
1024M-xfs Elapsed Time fsmark 94.59 91.00
1024M-xfs Elapsed Time simple-wb 229.84 195.08
1024M-xfs Elapsed Time mmap-strm 405.38 440.29
1024M-xfs Kswapd efficiency fsmark 79% 71%
1024M-xfs Kswapd efficiency simple-wb 74% 74%
1024M-xfs Kswapd efficiency mmap-strm 39% 42%
1024M-4X-xfs Files/s mean 32.63 ( 0.00%) 35.05 ( 6.90%)
1024M-4X-xfs Elapsed Time fsmark 103.33 97.74
1024M-4X-xfs Elapsed Time simple-wb 204.48 178.57
1024M-4X-xfs Elapsed Time mmap-strm 528.38 511.88
1024M-4X-xfs Kswapd efficiency fsmark 81% 70%
1024M-4X-xfs Kswapd efficiency simple-wb 73% 72%
1024M-4X-xfs Kswapd efficiency mmap-strm 39% 38%
1024M-16X-xfs Files/s mean 42.65 ( 0.00%) 42.97 ( 0.74%)
1024M-16X-xfs Elapsed Time fsmark 103.11 99.11
1024M-16X-xfs Elapsed Time simple-wb 200.83 178.24
1024M-16X-xfs Elapsed Time mmap-strm 397.35 459.82
1024M-16X-xfs Kswapd efficiency fsmark 84% 69%
1024M-16X-xfs Kswapd efficiency simple-wb 74% 73%
1024M-16X-xfs Kswapd efficiency mmap-strm 39% 40%
All FSMark tests up to 16X had statistically significant improvements.
For the most part, tests are completing faster with the exception of the
streaming writes to a mixture of anonymous and file-backed mappings which
were slower in two cases
In the cases where the mmap-strm tests were slower, there was more
swapping due to dirty pages being skipped. The number of additional pages
swapped is almost identical to the fewer number of pages written from
reclaim. In other words, roughly the same number of pages were reclaimed
but swapping was slower. As the test is a bit unrealistic and stresses
memory heavily, the small shift is acceptable.
4608M1P-xfs Files/s mean 29.75 ( 0.00%) 30.96 ( 3.91%)
4608M1P-xfs Elapsed Time fsmark 512.01 492.15
4608M1P-xfs Elapsed Time simple-wb 618.18 566.24
4608M1P-xfs Elapsed Time mmap-strm 488.05 465.07
4608M1P-xfs Kswapd efficiency fsmark 93% 86%
4608M1P-xfs Kswapd efficiency simple-wb 88% 84%
4608M1P-xfs Kswapd efficiency mmap-strm 46% 45%
4608M-xfs Files/s mean 27.60 ( 0.00%) 28.85 ( 4.33%)
4608M-xfs Elapsed Time fsmark 555.96 532.34
4608M-xfs Elapsed Time simple-wb 659.72 571.85
4608M-xfs Elapsed Time mmap-strm 1082.57 1146.38
4608M-xfs Kswapd efficiency fsmark 89% 91%
4608M-xfs Kswapd efficiency simple-wb 88% 82%
4608M-xfs Kswapd efficiency mmap-strm 48% 46%
4608M-4X-xfs Files/s mean 26.00 ( 0.00%) 27.47 ( 5.35%)
4608M-4X-xfs Elapsed Time fsmark 592.91 564.00
4608M-4X-xfs Elapsed Time simple-wb 616.65 575.07
4608M-4X-xfs Elapsed Time mmap-strm 1773.02 1631.53
4608M-4X-xfs Kswapd efficiency fsmark 90% 94%
4608M-4X-xfs Kswapd efficiency simple-wb 87% 82%
4608M-4X-xfs Kswapd efficiency mmap-strm 43% 43%
4608M-16X-xfs Files/s mean 26.07 ( 0.00%) 26.42 ( 1.32%)
4608M-16X-xfs Elapsed Time fsmark 602.69 585.78
4608M-16X-xfs Elapsed Time simple-wb 606.60 573.81
4608M-16X-xfs Elapsed Time mmap-strm 1549.75 1441.86
4608M-16X-xfs Kswapd efficiency fsmark 98% 98%
4608M-16X-xfs Kswapd efficiency simple-wb 88% 82%
4608M-16X-xfs Kswapd efficiency mmap-strm 44% 42%
Unlike the other tests, the fsmark results are not statistically
significant but the min and max times are both improved and for the most
part, tests completed faster.
There are other indications that this is an improvement as well. For
example, in the vast majority of cases, there were fewer pages scanned by
direct reclaim implying in many cases that stalls due to direct reclaim
are reduced. KSwapd is scanning more due to skipping dirty pages which is
unfortunate but the CPU usage is still acceptable
In an earlier set of tests, I used blktrace and in almost all cases
throughput throughout the entire test was higher. However, I ended up
discarding those results as recording blktrace data was too heavy for my
liking.
On a laptop, I plugged in a USB stick and ran a similar tests of tests
using it as backing storage. A desktop environment was running and for
the entire duration of the tests, firefox and gnome terminal were
launching and exiting to vaguely simulate a user.
1024M-xfs Files/s mean 0.41 ( 0.00%) 0.44 ( 6.82%)
1024M-xfs Elapsed Time fsmark 2053.52 1641.03
1024M-xfs Elapsed Time simple-wb 1229.53 768.05
1024M-xfs Elapsed Time mmap-strm 4126.44 4597.03
1024M-xfs Kswapd efficiency fsmark 84% 85%
1024M-xfs Kswapd efficiency simple-wb 92% 81%
1024M-xfs Kswapd efficiency mmap-strm 60% 51%
1024M-xfs Avg wait ms fsmark 5404.53 4473.87
1024M-xfs Avg wait ms simple-wb 2541.35 1453.54
1024M-xfs Avg wait ms mmap-strm 3400.25 3852.53
The mmap-strm results were hurt because firefox launching had a tendency
to push the test out of memory. On the postive side, firefox launched
marginally faster with the patches applied. Time to completion for many
tests was faster but more importantly - the "Avg wait" time as measured by
iostat was far lower implying the system would be more responsive. It was
also the case that "Avg wait ms" on the root filesystem was lower. I
tested it manually and while the system felt slightly more responsive
while copying data to a USB stick, it was marginal enough that it could be
my imagination.
This patch: do not writeback filesystem pages in direct reclaim.
When kswapd is failing to keep zones above the min watermark, a process
will enter direct reclaim in the same manner kswapd does. If a dirty page
is encountered during the scan, this page is written to backing storage
using mapping->writepage.
This causes two problems. First, it can result in very deep call stacks,
particularly if the target storage or filesystem are complex. Some
filesystems ignore write requests from direct reclaim as a result. The
second is that a single-page flush is inefficient in terms of IO. While
there is an expectation that the elevator will merge requests, this does
not always happen. Quoting Christoph Hellwig;
The elevator has a relatively small window it can operate on,
and can never fix up a bad large scale writeback pattern.
This patch prevents direct reclaim writing back filesystem pages by
checking if current is kswapd. Anonymous pages are still written to swap
as there is not the equivalent of a flusher thread for anonymous pages.
If the dirty pages cannot be written back, they are placed back on the LRU
lists. There is now a direct dependency on dirty page balancing to
prevent too many pages in the system being dirtied which would prevent
reclaim making forward progress.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:07:38 +07:00
|
|
|
/*
|
|
|
|
* Only kswapd can writeback filesystem pages to
|
2011-11-01 07:07:51 +07:00
|
|
|
* avoid risk of stack overflow but do not writeback
|
|
|
|
* unless under significant pressure.
|
mm: vmscan: do not writeback filesystem pages in direct reclaim
Testing from the XFS folk revealed that there is still too much I/O from
the end of the LRU in kswapd. Previously it was considered acceptable by
VM people for a small number of pages to be written back from reclaim with
testing generally showing about 0.3% of pages reclaimed were written back
(higher if memory was low). That writing back a small number of pages is
ok has been heavily disputed for quite some time and Dave Chinner
explained it well;
It doesn't have to be a very high number to be a problem. IO
is orders of magnitude slower than the CPU time it takes to
flush a page, so the cost of making a bad flush decision is
very high. And single page writeback from the LRU is almost
always a bad flush decision.
To complicate matters, filesystems respond very differently to requests
from reclaim according to Christoph Hellwig;
xfs tries to write it back if the requester is kswapd
ext4 ignores the request if it's a delayed allocation
btrfs ignores the request
As a result, each filesystem has different performance characteristics
when under memory pressure and there are many pages being dirtied. In
some cases, the request is ignored entirely so the VM cannot depend on the
IO being dispatched.
The objective of this series is to reduce writing of filesystem-backed
pages from reclaim, play nicely with writeback that is already in progress
and throttle reclaim appropriately when writeback pages are encountered.
The assumption is that the flushers will always write pages faster than if
reclaim issues the IO.
A secondary goal is to avoid the problem whereby direct reclaim splices
two potentially deep call stacks together.
There is a potential new problem as reclaim has less control over how long
before a page in a particularly zone or container is cleaned and direct
reclaimers depend on kswapd or flusher threads to do the necessary work.
However, as filesystems sometimes ignore direct reclaim requests already,
it is not expected to be a serious issue.
Patch 1 disables writeback of filesystem pages from direct reclaim
entirely. Anonymous pages are still written.
Patch 2 removes dead code in lumpy reclaim as it is no longer able
to synchronously write pages. This hurts lumpy reclaim but
there is an expectation that compaction is used for hugepage
allocations these days and lumpy reclaim's days are numbered.
Patches 3-4 add warnings to XFS and ext4 if called from
direct reclaim. With patch 1, this "never happens" and is
intended to catch regressions in this logic in the future.
Patch 5 disables writeback of filesystem pages from kswapd unless
the priority is raised to the point where kswapd is considered
to be in trouble.
Patch 6 throttles reclaimers if too many dirty pages are being
encountered and the zones or backing devices are congested.
Patch 7 invalidates dirty pages found at the end of the LRU so they
are reclaimed quickly after being written back rather than
waiting for a reclaimer to find them
I consider this series to be orthogonal to the writeback work but it is
worth noting that the writeback work affects the viability of patch 8 in
particular.
I tested this on ext4 and xfs using fs_mark, a simple writeback test based
on dd and a micro benchmark that does a streaming write to a large mapping
(exercises use-once LRU logic) followed by streaming writes to a mix of
anonymous and file-backed mappings. The command line for fs_mark when
botted with 512M looked something like
./fs_mark -d /tmp/fsmark-2676 -D 100 -N 150 -n 150 -L 25 -t 1 -S0 -s 10485760
The number of files was adjusted depending on the amount of available
memory so that the files created was about 3xRAM. For multiple threads,
the -d switch is specified multiple times.
The test machine is x86-64 with an older generation of AMD processor with
4 cores. The underlying storage was 4 disks configured as RAID-0 as this
was the best configuration of storage I had available. Swap is on a
separate disk. Dirty ratio was tuned to 40% instead of the default of
20%.
Testing was run with and without monitors to both verify that the patches
were operating as expected and that any performance gain was real and not
due to interference from monitors.
Here is a summary of results based on testing XFS.
512M1P-xfs Files/s mean 32.69 ( 0.00%) 34.44 ( 5.08%)
512M1P-xfs Elapsed Time fsmark 51.41 48.29
512M1P-xfs Elapsed Time simple-wb 114.09 108.61
512M1P-xfs Elapsed Time mmap-strm 113.46 109.34
512M1P-xfs Kswapd efficiency fsmark 62% 63%
512M1P-xfs Kswapd efficiency simple-wb 56% 61%
512M1P-xfs Kswapd efficiency mmap-strm 44% 42%
512M-xfs Files/s mean 30.78 ( 0.00%) 35.94 (14.36%)
512M-xfs Elapsed Time fsmark 56.08 48.90
512M-xfs Elapsed Time simple-wb 112.22 98.13
512M-xfs Elapsed Time mmap-strm 219.15 196.67
512M-xfs Kswapd efficiency fsmark 54% 56%
512M-xfs Kswapd efficiency simple-wb 54% 55%
512M-xfs Kswapd efficiency mmap-strm 45% 44%
512M-4X-xfs Files/s mean 30.31 ( 0.00%) 33.33 ( 9.06%)
512M-4X-xfs Elapsed Time fsmark 63.26 55.88
512M-4X-xfs Elapsed Time simple-wb 100.90 90.25
512M-4X-xfs Elapsed Time mmap-strm 261.73 255.38
512M-4X-xfs Kswapd efficiency fsmark 49% 50%
512M-4X-xfs Kswapd efficiency simple-wb 54% 56%
512M-4X-xfs Kswapd efficiency mmap-strm 37% 36%
512M-16X-xfs Files/s mean 60.89 ( 0.00%) 65.22 ( 6.64%)
512M-16X-xfs Elapsed Time fsmark 67.47 58.25
512M-16X-xfs Elapsed Time simple-wb 103.22 90.89
512M-16X-xfs Elapsed Time mmap-strm 237.09 198.82
512M-16X-xfs Kswapd efficiency fsmark 45% 46%
512M-16X-xfs Kswapd efficiency simple-wb 53% 55%
512M-16X-xfs Kswapd efficiency mmap-strm 33% 33%
Up until 512-4X, the FSmark improvements were statistically significant.
For the 4X and 16X tests the results were within standard deviations but
just barely. The time to completion for all tests is improved which is an
important result. In general, kswapd efficiency is not affected by
skipping dirty pages.
1024M1P-xfs Files/s mean 39.09 ( 0.00%) 41.15 ( 5.01%)
1024M1P-xfs Elapsed Time fsmark 84.14 80.41
1024M1P-xfs Elapsed Time simple-wb 210.77 184.78
1024M1P-xfs Elapsed Time mmap-strm 162.00 160.34
1024M1P-xfs Kswapd efficiency fsmark 69% 75%
1024M1P-xfs Kswapd efficiency simple-wb 71% 77%
1024M1P-xfs Kswapd efficiency mmap-strm 43% 44%
1024M-xfs Files/s mean 35.45 ( 0.00%) 37.00 ( 4.19%)
1024M-xfs Elapsed Time fsmark 94.59 91.00
1024M-xfs Elapsed Time simple-wb 229.84 195.08
1024M-xfs Elapsed Time mmap-strm 405.38 440.29
1024M-xfs Kswapd efficiency fsmark 79% 71%
1024M-xfs Kswapd efficiency simple-wb 74% 74%
1024M-xfs Kswapd efficiency mmap-strm 39% 42%
1024M-4X-xfs Files/s mean 32.63 ( 0.00%) 35.05 ( 6.90%)
1024M-4X-xfs Elapsed Time fsmark 103.33 97.74
1024M-4X-xfs Elapsed Time simple-wb 204.48 178.57
1024M-4X-xfs Elapsed Time mmap-strm 528.38 511.88
1024M-4X-xfs Kswapd efficiency fsmark 81% 70%
1024M-4X-xfs Kswapd efficiency simple-wb 73% 72%
1024M-4X-xfs Kswapd efficiency mmap-strm 39% 38%
1024M-16X-xfs Files/s mean 42.65 ( 0.00%) 42.97 ( 0.74%)
1024M-16X-xfs Elapsed Time fsmark 103.11 99.11
1024M-16X-xfs Elapsed Time simple-wb 200.83 178.24
1024M-16X-xfs Elapsed Time mmap-strm 397.35 459.82
1024M-16X-xfs Kswapd efficiency fsmark 84% 69%
1024M-16X-xfs Kswapd efficiency simple-wb 74% 73%
1024M-16X-xfs Kswapd efficiency mmap-strm 39% 40%
All FSMark tests up to 16X had statistically significant improvements.
For the most part, tests are completing faster with the exception of the
streaming writes to a mixture of anonymous and file-backed mappings which
were slower in two cases
In the cases where the mmap-strm tests were slower, there was more
swapping due to dirty pages being skipped. The number of additional pages
swapped is almost identical to the fewer number of pages written from
reclaim. In other words, roughly the same number of pages were reclaimed
but swapping was slower. As the test is a bit unrealistic and stresses
memory heavily, the small shift is acceptable.
4608M1P-xfs Files/s mean 29.75 ( 0.00%) 30.96 ( 3.91%)
4608M1P-xfs Elapsed Time fsmark 512.01 492.15
4608M1P-xfs Elapsed Time simple-wb 618.18 566.24
4608M1P-xfs Elapsed Time mmap-strm 488.05 465.07
4608M1P-xfs Kswapd efficiency fsmark 93% 86%
4608M1P-xfs Kswapd efficiency simple-wb 88% 84%
4608M1P-xfs Kswapd efficiency mmap-strm 46% 45%
4608M-xfs Files/s mean 27.60 ( 0.00%) 28.85 ( 4.33%)
4608M-xfs Elapsed Time fsmark 555.96 532.34
4608M-xfs Elapsed Time simple-wb 659.72 571.85
4608M-xfs Elapsed Time mmap-strm 1082.57 1146.38
4608M-xfs Kswapd efficiency fsmark 89% 91%
4608M-xfs Kswapd efficiency simple-wb 88% 82%
4608M-xfs Kswapd efficiency mmap-strm 48% 46%
4608M-4X-xfs Files/s mean 26.00 ( 0.00%) 27.47 ( 5.35%)
4608M-4X-xfs Elapsed Time fsmark 592.91 564.00
4608M-4X-xfs Elapsed Time simple-wb 616.65 575.07
4608M-4X-xfs Elapsed Time mmap-strm 1773.02 1631.53
4608M-4X-xfs Kswapd efficiency fsmark 90% 94%
4608M-4X-xfs Kswapd efficiency simple-wb 87% 82%
4608M-4X-xfs Kswapd efficiency mmap-strm 43% 43%
4608M-16X-xfs Files/s mean 26.07 ( 0.00%) 26.42 ( 1.32%)
4608M-16X-xfs Elapsed Time fsmark 602.69 585.78
4608M-16X-xfs Elapsed Time simple-wb 606.60 573.81
4608M-16X-xfs Elapsed Time mmap-strm 1549.75 1441.86
4608M-16X-xfs Kswapd efficiency fsmark 98% 98%
4608M-16X-xfs Kswapd efficiency simple-wb 88% 82%
4608M-16X-xfs Kswapd efficiency mmap-strm 44% 42%
Unlike the other tests, the fsmark results are not statistically
significant but the min and max times are both improved and for the most
part, tests completed faster.
There are other indications that this is an improvement as well. For
example, in the vast majority of cases, there were fewer pages scanned by
direct reclaim implying in many cases that stalls due to direct reclaim
are reduced. KSwapd is scanning more due to skipping dirty pages which is
unfortunate but the CPU usage is still acceptable
In an earlier set of tests, I used blktrace and in almost all cases
throughput throughout the entire test was higher. However, I ended up
discarding those results as recording blktrace data was too heavy for my
liking.
On a laptop, I plugged in a USB stick and ran a similar tests of tests
using it as backing storage. A desktop environment was running and for
the entire duration of the tests, firefox and gnome terminal were
launching and exiting to vaguely simulate a user.
1024M-xfs Files/s mean 0.41 ( 0.00%) 0.44 ( 6.82%)
1024M-xfs Elapsed Time fsmark 2053.52 1641.03
1024M-xfs Elapsed Time simple-wb 1229.53 768.05
1024M-xfs Elapsed Time mmap-strm 4126.44 4597.03
1024M-xfs Kswapd efficiency fsmark 84% 85%
1024M-xfs Kswapd efficiency simple-wb 92% 81%
1024M-xfs Kswapd efficiency mmap-strm 60% 51%
1024M-xfs Avg wait ms fsmark 5404.53 4473.87
1024M-xfs Avg wait ms simple-wb 2541.35 1453.54
1024M-xfs Avg wait ms mmap-strm 3400.25 3852.53
The mmap-strm results were hurt because firefox launching had a tendency
to push the test out of memory. On the postive side, firefox launched
marginally faster with the patches applied. Time to completion for many
tests was faster but more importantly - the "Avg wait" time as measured by
iostat was far lower implying the system would be more responsive. It was
also the case that "Avg wait ms" on the root filesystem was lower. I
tested it manually and while the system felt slightly more responsive
while copying data to a USB stick, it was marginal enough that it could be
my imagination.
This patch: do not writeback filesystem pages in direct reclaim.
When kswapd is failing to keep zones above the min watermark, a process
will enter direct reclaim in the same manner kswapd does. If a dirty page
is encountered during the scan, this page is written to backing storage
using mapping->writepage.
This causes two problems. First, it can result in very deep call stacks,
particularly if the target storage or filesystem are complex. Some
filesystems ignore write requests from direct reclaim as a result. The
second is that a single-page flush is inefficient in terms of IO. While
there is an expectation that the elevator will merge requests, this does
not always happen. Quoting Christoph Hellwig;
The elevator has a relatively small window it can operate on,
and can never fix up a bad large scale writeback pattern.
This patch prevents direct reclaim writing back filesystem pages by
checking if current is kswapd. Anonymous pages are still written to swap
as there is not the equivalent of a flusher thread for anonymous pages.
If the dirty pages cannot be written back, they are placed back on the LRU
lists. There is now a direct dependency on dirty page balancing to
prevent too many pages in the system being dirtied which would prevent
reclaim making forward progress.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:07:38 +07:00
|
|
|
*/
|
2011-11-01 07:07:51 +07:00
|
|
|
if (page_is_file_cache(page) &&
|
|
|
|
(!current_is_kswapd() || priority >= DEF_PRIORITY - 2)) {
|
2011-11-01 07:07:59 +07:00
|
|
|
/*
|
|
|
|
* Immediately reclaim when written back.
|
|
|
|
* Similar in principal to deactivate_page()
|
|
|
|
* except we already have the page isolated
|
|
|
|
* and know it's dirty
|
|
|
|
*/
|
|
|
|
inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE);
|
|
|
|
SetPageReclaim(page);
|
|
|
|
|
mm: vmscan: do not writeback filesystem pages in direct reclaim
Testing from the XFS folk revealed that there is still too much I/O from
the end of the LRU in kswapd. Previously it was considered acceptable by
VM people for a small number of pages to be written back from reclaim with
testing generally showing about 0.3% of pages reclaimed were written back
(higher if memory was low). That writing back a small number of pages is
ok has been heavily disputed for quite some time and Dave Chinner
explained it well;
It doesn't have to be a very high number to be a problem. IO
is orders of magnitude slower than the CPU time it takes to
flush a page, so the cost of making a bad flush decision is
very high. And single page writeback from the LRU is almost
always a bad flush decision.
To complicate matters, filesystems respond very differently to requests
from reclaim according to Christoph Hellwig;
xfs tries to write it back if the requester is kswapd
ext4 ignores the request if it's a delayed allocation
btrfs ignores the request
As a result, each filesystem has different performance characteristics
when under memory pressure and there are many pages being dirtied. In
some cases, the request is ignored entirely so the VM cannot depend on the
IO being dispatched.
The objective of this series is to reduce writing of filesystem-backed
pages from reclaim, play nicely with writeback that is already in progress
and throttle reclaim appropriately when writeback pages are encountered.
The assumption is that the flushers will always write pages faster than if
reclaim issues the IO.
A secondary goal is to avoid the problem whereby direct reclaim splices
two potentially deep call stacks together.
There is a potential new problem as reclaim has less control over how long
before a page in a particularly zone or container is cleaned and direct
reclaimers depend on kswapd or flusher threads to do the necessary work.
However, as filesystems sometimes ignore direct reclaim requests already,
it is not expected to be a serious issue.
Patch 1 disables writeback of filesystem pages from direct reclaim
entirely. Anonymous pages are still written.
Patch 2 removes dead code in lumpy reclaim as it is no longer able
to synchronously write pages. This hurts lumpy reclaim but
there is an expectation that compaction is used for hugepage
allocations these days and lumpy reclaim's days are numbered.
Patches 3-4 add warnings to XFS and ext4 if called from
direct reclaim. With patch 1, this "never happens" and is
intended to catch regressions in this logic in the future.
Patch 5 disables writeback of filesystem pages from kswapd unless
the priority is raised to the point where kswapd is considered
to be in trouble.
Patch 6 throttles reclaimers if too many dirty pages are being
encountered and the zones or backing devices are congested.
Patch 7 invalidates dirty pages found at the end of the LRU so they
are reclaimed quickly after being written back rather than
waiting for a reclaimer to find them
I consider this series to be orthogonal to the writeback work but it is
worth noting that the writeback work affects the viability of patch 8 in
particular.
I tested this on ext4 and xfs using fs_mark, a simple writeback test based
on dd and a micro benchmark that does a streaming write to a large mapping
(exercises use-once LRU logic) followed by streaming writes to a mix of
anonymous and file-backed mappings. The command line for fs_mark when
botted with 512M looked something like
./fs_mark -d /tmp/fsmark-2676 -D 100 -N 150 -n 150 -L 25 -t 1 -S0 -s 10485760
The number of files was adjusted depending on the amount of available
memory so that the files created was about 3xRAM. For multiple threads,
the -d switch is specified multiple times.
The test machine is x86-64 with an older generation of AMD processor with
4 cores. The underlying storage was 4 disks configured as RAID-0 as this
was the best configuration of storage I had available. Swap is on a
separate disk. Dirty ratio was tuned to 40% instead of the default of
20%.
Testing was run with and without monitors to both verify that the patches
were operating as expected and that any performance gain was real and not
due to interference from monitors.
Here is a summary of results based on testing XFS.
512M1P-xfs Files/s mean 32.69 ( 0.00%) 34.44 ( 5.08%)
512M1P-xfs Elapsed Time fsmark 51.41 48.29
512M1P-xfs Elapsed Time simple-wb 114.09 108.61
512M1P-xfs Elapsed Time mmap-strm 113.46 109.34
512M1P-xfs Kswapd efficiency fsmark 62% 63%
512M1P-xfs Kswapd efficiency simple-wb 56% 61%
512M1P-xfs Kswapd efficiency mmap-strm 44% 42%
512M-xfs Files/s mean 30.78 ( 0.00%) 35.94 (14.36%)
512M-xfs Elapsed Time fsmark 56.08 48.90
512M-xfs Elapsed Time simple-wb 112.22 98.13
512M-xfs Elapsed Time mmap-strm 219.15 196.67
512M-xfs Kswapd efficiency fsmark 54% 56%
512M-xfs Kswapd efficiency simple-wb 54% 55%
512M-xfs Kswapd efficiency mmap-strm 45% 44%
512M-4X-xfs Files/s mean 30.31 ( 0.00%) 33.33 ( 9.06%)
512M-4X-xfs Elapsed Time fsmark 63.26 55.88
512M-4X-xfs Elapsed Time simple-wb 100.90 90.25
512M-4X-xfs Elapsed Time mmap-strm 261.73 255.38
512M-4X-xfs Kswapd efficiency fsmark 49% 50%
512M-4X-xfs Kswapd efficiency simple-wb 54% 56%
512M-4X-xfs Kswapd efficiency mmap-strm 37% 36%
512M-16X-xfs Files/s mean 60.89 ( 0.00%) 65.22 ( 6.64%)
512M-16X-xfs Elapsed Time fsmark 67.47 58.25
512M-16X-xfs Elapsed Time simple-wb 103.22 90.89
512M-16X-xfs Elapsed Time mmap-strm 237.09 198.82
512M-16X-xfs Kswapd efficiency fsmark 45% 46%
512M-16X-xfs Kswapd efficiency simple-wb 53% 55%
512M-16X-xfs Kswapd efficiency mmap-strm 33% 33%
Up until 512-4X, the FSmark improvements were statistically significant.
For the 4X and 16X tests the results were within standard deviations but
just barely. The time to completion for all tests is improved which is an
important result. In general, kswapd efficiency is not affected by
skipping dirty pages.
1024M1P-xfs Files/s mean 39.09 ( 0.00%) 41.15 ( 5.01%)
1024M1P-xfs Elapsed Time fsmark 84.14 80.41
1024M1P-xfs Elapsed Time simple-wb 210.77 184.78
1024M1P-xfs Elapsed Time mmap-strm 162.00 160.34
1024M1P-xfs Kswapd efficiency fsmark 69% 75%
1024M1P-xfs Kswapd efficiency simple-wb 71% 77%
1024M1P-xfs Kswapd efficiency mmap-strm 43% 44%
1024M-xfs Files/s mean 35.45 ( 0.00%) 37.00 ( 4.19%)
1024M-xfs Elapsed Time fsmark 94.59 91.00
1024M-xfs Elapsed Time simple-wb 229.84 195.08
1024M-xfs Elapsed Time mmap-strm 405.38 440.29
1024M-xfs Kswapd efficiency fsmark 79% 71%
1024M-xfs Kswapd efficiency simple-wb 74% 74%
1024M-xfs Kswapd efficiency mmap-strm 39% 42%
1024M-4X-xfs Files/s mean 32.63 ( 0.00%) 35.05 ( 6.90%)
1024M-4X-xfs Elapsed Time fsmark 103.33 97.74
1024M-4X-xfs Elapsed Time simple-wb 204.48 178.57
1024M-4X-xfs Elapsed Time mmap-strm 528.38 511.88
1024M-4X-xfs Kswapd efficiency fsmark 81% 70%
1024M-4X-xfs Kswapd efficiency simple-wb 73% 72%
1024M-4X-xfs Kswapd efficiency mmap-strm 39% 38%
1024M-16X-xfs Files/s mean 42.65 ( 0.00%) 42.97 ( 0.74%)
1024M-16X-xfs Elapsed Time fsmark 103.11 99.11
1024M-16X-xfs Elapsed Time simple-wb 200.83 178.24
1024M-16X-xfs Elapsed Time mmap-strm 397.35 459.82
1024M-16X-xfs Kswapd efficiency fsmark 84% 69%
1024M-16X-xfs Kswapd efficiency simple-wb 74% 73%
1024M-16X-xfs Kswapd efficiency mmap-strm 39% 40%
All FSMark tests up to 16X had statistically significant improvements.
For the most part, tests are completing faster with the exception of the
streaming writes to a mixture of anonymous and file-backed mappings which
were slower in two cases
In the cases where the mmap-strm tests were slower, there was more
swapping due to dirty pages being skipped. The number of additional pages
swapped is almost identical to the fewer number of pages written from
reclaim. In other words, roughly the same number of pages were reclaimed
but swapping was slower. As the test is a bit unrealistic and stresses
memory heavily, the small shift is acceptable.
4608M1P-xfs Files/s mean 29.75 ( 0.00%) 30.96 ( 3.91%)
4608M1P-xfs Elapsed Time fsmark 512.01 492.15
4608M1P-xfs Elapsed Time simple-wb 618.18 566.24
4608M1P-xfs Elapsed Time mmap-strm 488.05 465.07
4608M1P-xfs Kswapd efficiency fsmark 93% 86%
4608M1P-xfs Kswapd efficiency simple-wb 88% 84%
4608M1P-xfs Kswapd efficiency mmap-strm 46% 45%
4608M-xfs Files/s mean 27.60 ( 0.00%) 28.85 ( 4.33%)
4608M-xfs Elapsed Time fsmark 555.96 532.34
4608M-xfs Elapsed Time simple-wb 659.72 571.85
4608M-xfs Elapsed Time mmap-strm 1082.57 1146.38
4608M-xfs Kswapd efficiency fsmark 89% 91%
4608M-xfs Kswapd efficiency simple-wb 88% 82%
4608M-xfs Kswapd efficiency mmap-strm 48% 46%
4608M-4X-xfs Files/s mean 26.00 ( 0.00%) 27.47 ( 5.35%)
4608M-4X-xfs Elapsed Time fsmark 592.91 564.00
4608M-4X-xfs Elapsed Time simple-wb 616.65 575.07
4608M-4X-xfs Elapsed Time mmap-strm 1773.02 1631.53
4608M-4X-xfs Kswapd efficiency fsmark 90% 94%
4608M-4X-xfs Kswapd efficiency simple-wb 87% 82%
4608M-4X-xfs Kswapd efficiency mmap-strm 43% 43%
4608M-16X-xfs Files/s mean 26.07 ( 0.00%) 26.42 ( 1.32%)
4608M-16X-xfs Elapsed Time fsmark 602.69 585.78
4608M-16X-xfs Elapsed Time simple-wb 606.60 573.81
4608M-16X-xfs Elapsed Time mmap-strm 1549.75 1441.86
4608M-16X-xfs Kswapd efficiency fsmark 98% 98%
4608M-16X-xfs Kswapd efficiency simple-wb 88% 82%
4608M-16X-xfs Kswapd efficiency mmap-strm 44% 42%
Unlike the other tests, the fsmark results are not statistically
significant but the min and max times are both improved and for the most
part, tests completed faster.
There are other indications that this is an improvement as well. For
example, in the vast majority of cases, there were fewer pages scanned by
direct reclaim implying in many cases that stalls due to direct reclaim
are reduced. KSwapd is scanning more due to skipping dirty pages which is
unfortunate but the CPU usage is still acceptable
In an earlier set of tests, I used blktrace and in almost all cases
throughput throughout the entire test was higher. However, I ended up
discarding those results as recording blktrace data was too heavy for my
liking.
On a laptop, I plugged in a USB stick and ran a similar tests of tests
using it as backing storage. A desktop environment was running and for
the entire duration of the tests, firefox and gnome terminal were
launching and exiting to vaguely simulate a user.
1024M-xfs Files/s mean 0.41 ( 0.00%) 0.44 ( 6.82%)
1024M-xfs Elapsed Time fsmark 2053.52 1641.03
1024M-xfs Elapsed Time simple-wb 1229.53 768.05
1024M-xfs Elapsed Time mmap-strm 4126.44 4597.03
1024M-xfs Kswapd efficiency fsmark 84% 85%
1024M-xfs Kswapd efficiency simple-wb 92% 81%
1024M-xfs Kswapd efficiency mmap-strm 60% 51%
1024M-xfs Avg wait ms fsmark 5404.53 4473.87
1024M-xfs Avg wait ms simple-wb 2541.35 1453.54
1024M-xfs Avg wait ms mmap-strm 3400.25 3852.53
The mmap-strm results were hurt because firefox launching had a tendency
to push the test out of memory. On the postive side, firefox launched
marginally faster with the patches applied. Time to completion for many
tests was faster but more importantly - the "Avg wait" time as measured by
iostat was far lower implying the system would be more responsive. It was
also the case that "Avg wait ms" on the root filesystem was lower. I
tested it manually and while the system felt slightly more responsive
while copying data to a USB stick, it was marginal enough that it could be
my imagination.
This patch: do not writeback filesystem pages in direct reclaim.
When kswapd is failing to keep zones above the min watermark, a process
will enter direct reclaim in the same manner kswapd does. If a dirty page
is encountered during the scan, this page is written to backing storage
using mapping->writepage.
This causes two problems. First, it can result in very deep call stacks,
particularly if the target storage or filesystem are complex. Some
filesystems ignore write requests from direct reclaim as a result. The
second is that a single-page flush is inefficient in terms of IO. While
there is an expectation that the elevator will merge requests, this does
not always happen. Quoting Christoph Hellwig;
The elevator has a relatively small window it can operate on,
and can never fix up a bad large scale writeback pattern.
This patch prevents direct reclaim writing back filesystem pages by
checking if current is kswapd. Anonymous pages are still written to swap
as there is not the equivalent of a flusher thread for anonymous pages.
If the dirty pages cannot be written back, they are placed back on the LRU
lists. There is now a direct dependency on dirty page balancing to
prevent too many pages in the system being dirtied which would prevent
reclaim making forward progress.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Alex Elder <aelder@sgi.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:07:38 +07:00
|
|
|
goto keep_locked;
|
|
|
|
}
|
|
|
|
|
vmscan: factor out page reference checks
The used-once mapped file page detection patchset.
It is meant to help workloads with large amounts of shortly used file
mappings, like rtorrent hashing a file or git when dealing with loose
objects (git gc on a bigger site?).
Right now, the VM activates referenced mapped file pages on first
encounter on the inactive list and it takes a full memory cycle to
reclaim them again. When those pages dominate memory, the system
no longer has a meaningful notion of 'working set' and is required
to give up the active list to make reclaim progress. Obviously,
this results in rather bad scanning latencies and the wrong pages
being reclaimed.
This patch makes the VM be more careful about activating mapped file
pages in the first place. The minimum granted lifetime without
another memory access becomes an inactive list cycle instead of the
full memory cycle, which is more natural given the mentioned loads.
This test resembles a hashing rtorrent process. Sequentially, 32MB
chunks of a file are mapped into memory, hashed (sha1) and unmapped
again. While this happens, every 5 seconds a process is launched and
its execution time taken:
python2.4 -c 'import pydoc'
old: max=2.31s mean=1.26s (0.34)
new: max=1.25s mean=0.32s (0.32)
find /etc -type f
old: max=2.52s mean=1.44s (0.43)
new: max=1.92s mean=0.12s (0.17)
vim -c ':quit'
old: max=6.14s mean=4.03s (0.49)
new: max=3.48s mean=2.41s (0.25)
mplayer --help
old: max=8.08s mean=5.74s (1.02)
new: max=3.79s mean=1.32s (0.81)
overall hash time (stdev):
old: time=1192.30 (12.85) thruput=25.78mb/s (0.27)
new: time=1060.27 (32.58) thruput=29.02mb/s (0.88) (-11%)
I also tested kernbench with regular IO streaming in the background to
see whether the delayed activation of frequently used mapped file
pages had a negative impact on performance in the presence of pressure
on the inactive list. The patch made no significant difference in
timing, neither for kernbench nor for the streaming IO throughput.
The first patch submission raised concerns about the cost of the extra
faults for actually activated pages on machines that have no hardware
support for young page table entries.
I created an artificial worst case scenario on an ARM machine with
around 300MHz and 64MB of memory to figure out the dimensions
involved. The test would mmap a file of 20MB, then
1. touch all its pages to fault them in
2. force one full scan cycle on the inactive file LRU
-- old: mapping pages activated
-- new: mapping pages inactive
3. touch the mapping pages again
-- old and new: fault exceptions to set the young bits
4. force another full scan cycle on the inactive file LRU
5. touch the mapping pages one last time
-- new: fault exceptions to set the young bits
The test showed an overall increase of 6% in time over 100 iterations
of the above (old: ~212sec, new: ~225sec). 13 secs total overhead /
(100 * 5k pages), ignoring the execution time of the test itself,
makes for about 25us overhead for every page that gets actually
activated. Note:
1. File mapping the size of one third of main memory, _completely_
in active use across memory pressure - i.e., most pages referenced
within one LRU cycle. This should be rare to non-existant,
especially on such embedded setups.
2. Many huge activation batches. Those batches only occur when the
working set fluctuates. If it changes completely between every full
LRU cycle, you have problematic reclaim overhead anyway.
3. Access of activated pages at maximum speed: sequential loads from
every single page without doing anything in between. In reality,
the extra faults will get distributed between actual operations on
the data.
So even if a workload manages to get the VM into the situation of
activating a third of memory in one go on such a setup, it will take
2.2 seconds instead 2.1 without the patch.
Comparing the numbers (and my user-experience over several months),
I think this change is an overall improvement to the VM.
Patch 1 is only refactoring to break up that ugly compound conditional
in shrink_page_list() and make it easy to document and add new checks
in a readable fashion.
Patch 2 gets rid of the obsolete page_mapping_inuse(). It's not
strictly related to #3, but it was in the original submission and is a
net simplification, so I kept it.
Patch 3 implements used-once detection of mapped file pages.
This patch:
Moving the big conditional into its own predicate function makes the code
a bit easier to read and allows for better commenting on the checks
one-by-one.
This is just cleaning up, no semantics should have been changed.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: OSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-06 04:42:19 +07:00
|
|
|
if (references == PAGEREF_RECLAIM_CLEAN)
|
2005-04-17 05:20:36 +07:00
|
|
|
goto keep_locked;
|
2008-03-25 02:29:52 +07:00
|
|
|
if (!may_enter_fs)
|
2005-04-17 05:20:36 +07:00
|
|
|
goto keep_locked;
|
2006-02-01 18:05:28 +07:00
|
|
|
if (!sc->may_writepage)
|
2005-04-17 05:20:36 +07:00
|
|
|
goto keep_locked;
|
|
|
|
|
|
|
|
/* Page is dirty, try to write it out here */
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
switch (pageout(page, mapping, sc)) {
|
2005-04-17 05:20:36 +07:00
|
|
|
case PAGE_KEEP:
|
2010-10-27 04:21:45 +07:00
|
|
|
nr_congested++;
|
2005-04-17 05:20:36 +07:00
|
|
|
goto keep_locked;
|
|
|
|
case PAGE_ACTIVATE:
|
|
|
|
goto activate_locked;
|
|
|
|
case PAGE_SUCCESS:
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
if (PageWriteback(page))
|
|
|
|
goto keep_lumpy;
|
|
|
|
if (PageDirty(page))
|
2005-04-17 05:20:36 +07:00
|
|
|
goto keep;
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* A synchronous write - probably a ramdisk. Go
|
|
|
|
* ahead and try to reclaim the page.
|
|
|
|
*/
|
2008-08-02 17:01:03 +07:00
|
|
|
if (!trylock_page(page))
|
2005-04-17 05:20:36 +07:00
|
|
|
goto keep;
|
|
|
|
if (PageDirty(page) || PageWriteback(page))
|
|
|
|
goto keep_locked;
|
|
|
|
mapping = page_mapping(page);
|
|
|
|
case PAGE_CLEAN:
|
|
|
|
; /* try to free the page below */
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the page has buffers, try to free the buffer mappings
|
|
|
|
* associated with this page. If we succeed we try to free
|
|
|
|
* the page as well.
|
|
|
|
*
|
|
|
|
* We do this even if the page is PageDirty().
|
|
|
|
* try_to_release_page() does not perform I/O, but it is
|
|
|
|
* possible for a page to have PageDirty set, but it is actually
|
|
|
|
* clean (all its buffers are clean). This happens if the
|
|
|
|
* buffers were written out directly, with submit_bh(). ext3
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
* will do this, as well as the blockdev mapping.
|
2005-04-17 05:20:36 +07:00
|
|
|
* try_to_release_page() will discover that cleanness and will
|
|
|
|
* drop the buffers and mark the page clean - it can be freed.
|
|
|
|
*
|
|
|
|
* Rarely, pages can have buffers and no ->mapping. These are
|
|
|
|
* the pages which were not successfully invalidated in
|
|
|
|
* truncate_complete_page(). We try to drop those buffers here
|
|
|
|
* and if that worked, and the page is no longer mapped into
|
|
|
|
* process address space (page_count == 1) it can be freed.
|
|
|
|
* Otherwise, leave the page on the LRU so it is swappable.
|
|
|
|
*/
|
2009-04-03 22:42:36 +07:00
|
|
|
if (page_has_private(page)) {
|
2005-04-17 05:20:36 +07:00
|
|
|
if (!try_to_release_page(page, sc->gfp_mask))
|
|
|
|
goto activate_locked;
|
2008-07-26 09:45:30 +07:00
|
|
|
if (!mapping && page_count(page) == 1) {
|
|
|
|
unlock_page(page);
|
|
|
|
if (put_page_testzero(page))
|
|
|
|
goto free_it;
|
|
|
|
else {
|
|
|
|
/*
|
|
|
|
* rare race with speculative reference.
|
|
|
|
* the speculative reference will free
|
|
|
|
* this page shortly, so we may
|
|
|
|
* increment nr_reclaimed here (and
|
|
|
|
* leave it off the LRU).
|
|
|
|
*/
|
|
|
|
nr_reclaimed++;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2008-07-26 09:45:30 +07:00
|
|
|
if (!mapping || !__remove_mapping(mapping, page))
|
2006-01-08 16:00:48 +07:00
|
|
|
goto keep_locked;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-10-19 10:26:58 +07:00
|
|
|
/*
|
|
|
|
* At this point, we have no other references and there is
|
|
|
|
* no way to pick any more up (removed from LRU, removed
|
|
|
|
* from pagecache). Can use non-atomic bitops now (and
|
|
|
|
* we obviously don't have to worry about waking up a process
|
|
|
|
* waiting on the page lock, because there are no references.
|
|
|
|
*/
|
|
|
|
__clear_page_locked(page);
|
2008-07-26 09:45:30 +07:00
|
|
|
free_it:
|
2006-03-22 15:08:20 +07:00
|
|
|
nr_reclaimed++;
|
2010-08-10 07:19:31 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Is there need to periodically free_page_list? It would
|
|
|
|
* appear not as the counts should be low
|
|
|
|
*/
|
|
|
|
list_add(&page->lru, &free_pages);
|
2005-04-17 05:20:36 +07:00
|
|
|
continue;
|
|
|
|
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
cull_mlocked:
|
2009-01-07 05:39:38 +07:00
|
|
|
if (PageSwapCache(page))
|
|
|
|
try_to_free_swap(page);
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
unlock_page(page);
|
|
|
|
putback_lru_page(page);
|
2011-01-14 06:46:00 +07:00
|
|
|
reset_reclaim_mode(sc);
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
continue;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
activate_locked:
|
2008-10-19 10:26:23 +07:00
|
|
|
/* Not a candidate for swapping, so reclaim swap space. */
|
|
|
|
if (PageSwapCache(page) && vm_swap_full())
|
2009-01-07 05:39:36 +07:00
|
|
|
try_to_free_swap(page);
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
VM_BUG_ON(PageActive(page));
|
2005-04-17 05:20:36 +07:00
|
|
|
SetPageActive(page);
|
|
|
|
pgactivate++;
|
|
|
|
keep_locked:
|
|
|
|
unlock_page(page);
|
|
|
|
keep:
|
2011-01-14 06:46:00 +07:00
|
|
|
reset_reclaim_mode(sc);
|
vmscan: narrow the scenarios in whcih lumpy reclaim uses synchrounous reclaim
shrink_page_list() can decide to give up reclaiming a page under a
number of conditions such as
1. trylock_page() failure
2. page is unevictable
3. zone reclaim and page is mapped
4. PageWriteback() is true
5. page is swapbacked and swap is full
6. add_to_swap() failure
7. page is dirty and gfpmask don't have GFP_IO, GFP_FS
8. page is pinned
9. IO queue is congested
10. pageout() start IO, but not finished
With lumpy reclaim, failures result in entering synchronous lumpy reclaim
but this can be unnecessary. In cases (2), (3), (5), (6), (7) and (8),
there is no point retrying. This patch causes lumpy reclaim to abort when
it is known it will fail.
Case (9) is more interesting. current behavior is,
1. start shrink_page_list(async)
2. found queue_congested()
3. skip pageout write
4. still start shrink_page_list(sync)
5. wait on a lot of pages
6. again, found queue_congested()
7. give up pageout write again
So, it's useless time wasting. However, just skipping page reclaim is
also notgood as x86 allocating a huge page needs 512 pages for example.
It can have more dirty pages than queue congestion threshold (~=128).
After this patch, pageout() behaves as follows;
- If order > PAGE_ALLOC_COSTLY_ORDER
Ignore queue congestion always.
- If order <= PAGE_ALLOC_COSTLY_ORDER
skip write page and disable lumpy reclaim.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:42 +07:00
|
|
|
keep_lumpy:
|
2005-04-17 05:20:36 +07:00
|
|
|
list_add(&page->lru, &ret_pages);
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
VM_BUG_ON(PageLRU(page) || PageUnevictable(page));
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2010-08-10 07:19:31 +07:00
|
|
|
|
2010-10-27 04:21:45 +07:00
|
|
|
/*
|
|
|
|
* Tag a zone as congested if all the dirty pages encountered were
|
|
|
|
* backed by a congested BDI. In this case, reclaimers should just
|
|
|
|
* back off and wait for congestion to clear because further reclaim
|
|
|
|
* will encounter the same problem
|
|
|
|
*/
|
2012-01-13 08:17:50 +07:00
|
|
|
if (nr_dirty && nr_dirty == nr_congested && global_reclaim(sc))
|
2012-01-13 08:17:52 +07:00
|
|
|
zone_set_flag(mz->zone, ZONE_CONGESTED);
|
2010-10-27 04:21:45 +07:00
|
|
|
|
2012-01-11 06:07:04 +07:00
|
|
|
free_hot_cold_page_list(&free_pages, 1);
|
2010-08-10 07:19:31 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
list_splice(&ret_pages, page_list);
|
2006-06-30 15:55:45 +07:00
|
|
|
count_vm_events(PGACTIVATE, pgactivate);
|
2011-11-01 07:07:56 +07:00
|
|
|
*ret_nr_dirty += nr_dirty;
|
|
|
|
*ret_nr_writeback += nr_writeback;
|
2006-03-22 15:08:20 +07:00
|
|
|
return nr_reclaimed;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2007-07-17 18:03:16 +07:00
|
|
|
/*
|
|
|
|
* Attempt to remove the specified page from its LRU. Only take this page
|
|
|
|
* if it is of the appropriate PageActive status. Pages which are being
|
|
|
|
* freed elsewhere are also ignored.
|
|
|
|
*
|
|
|
|
* page: page to consider
|
|
|
|
* mode: one of the LRU isolation modes defined above
|
|
|
|
*
|
|
|
|
* returns 0 on success, -ve errno on failure.
|
|
|
|
*/
|
2011-11-01 07:06:47 +07:00
|
|
|
int __isolate_lru_page(struct page *page, isolate_mode_t mode, int file)
|
2007-07-17 18:03:16 +07:00
|
|
|
{
|
2011-11-01 07:06:47 +07:00
|
|
|
bool all_lru_mode;
|
2007-07-17 18:03:16 +07:00
|
|
|
int ret = -EINVAL;
|
|
|
|
|
|
|
|
/* Only take pages on the LRU. */
|
|
|
|
if (!PageLRU(page))
|
|
|
|
return ret;
|
|
|
|
|
2011-11-01 07:06:47 +07:00
|
|
|
all_lru_mode = (mode & (ISOLATE_ACTIVE|ISOLATE_INACTIVE)) ==
|
|
|
|
(ISOLATE_ACTIVE|ISOLATE_INACTIVE);
|
|
|
|
|
2007-07-17 18:03:16 +07:00
|
|
|
/*
|
|
|
|
* When checking the active state, we need to be sure we are
|
|
|
|
* dealing with comparible boolean values. Take the logical not
|
|
|
|
* of each.
|
|
|
|
*/
|
2011-11-01 07:06:47 +07:00
|
|
|
if (!all_lru_mode && !PageActive(page) != !(mode & ISOLATE_ACTIVE))
|
2007-07-17 18:03:16 +07:00
|
|
|
return ret;
|
|
|
|
|
2011-11-01 07:06:47 +07:00
|
|
|
if (!all_lru_mode && !!page_is_file_cache(page) != file)
|
2008-10-19 10:26:32 +07:00
|
|
|
return ret;
|
|
|
|
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
/*
|
|
|
|
* When this function is being called for lumpy reclaim, we
|
|
|
|
* initially look into all LRU pages, active, inactive and
|
|
|
|
* unevictable; only give shrink_page_list evictable pages.
|
|
|
|
*/
|
|
|
|
if (PageUnevictable(page))
|
|
|
|
return ret;
|
|
|
|
|
2007-07-17 18:03:16 +07:00
|
|
|
ret = -EBUSY;
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 09:08:01 +07:00
|
|
|
|
2012-01-13 08:19:38 +07:00
|
|
|
/*
|
|
|
|
* To minimise LRU disruption, the caller can indicate that it only
|
|
|
|
* wants to isolate pages it will be able to operate on without
|
|
|
|
* blocking - clean pages for the most part.
|
|
|
|
*
|
|
|
|
* ISOLATE_CLEAN means that only clean pages should be isolated. This
|
|
|
|
* is used by reclaim when it is cannot write to backing storage
|
|
|
|
*
|
|
|
|
* ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages
|
|
|
|
* that it is possible to migrate without blocking
|
|
|
|
*/
|
|
|
|
if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) {
|
|
|
|
/* All the caller can do on PageWriteback is block */
|
|
|
|
if (PageWriteback(page))
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (PageDirty(page)) {
|
|
|
|
struct address_space *mapping;
|
|
|
|
|
|
|
|
/* ISOLATE_CLEAN means only clean pages */
|
|
|
|
if (mode & ISOLATE_CLEAN)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Only pages without mappings or that have a
|
|
|
|
* ->migratepage callback are possible to migrate
|
|
|
|
* without blocking
|
|
|
|
*/
|
|
|
|
mapping = page_mapping(page);
|
|
|
|
if (mapping && !mapping->a_ops->migratepage)
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
2011-11-01 07:06:51 +07:00
|
|
|
|
2011-11-01 07:06:55 +07:00
|
|
|
if ((mode & ISOLATE_UNMAPPED) && page_mapped(page))
|
|
|
|
return ret;
|
|
|
|
|
2007-07-17 18:03:16 +07:00
|
|
|
if (likely(get_page_unless_zero(page))) {
|
|
|
|
/*
|
|
|
|
* Be careful not to clear PageLRU until after we're
|
|
|
|
* sure the page is not being freed elsewhere -- the
|
|
|
|
* page release code relies on it.
|
|
|
|
*/
|
|
|
|
ClearPageLRU(page);
|
|
|
|
ret = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* zone->lru_lock is heavily contended. Some of the functions that
|
|
|
|
* shrink the lists perform better by taking out a batch of pages
|
|
|
|
* and working on them outside the LRU lock.
|
|
|
|
*
|
|
|
|
* For pagecache intensive workloads, this function is the hottest
|
|
|
|
* spot in the kernel (apart from copy_*_user functions).
|
|
|
|
*
|
|
|
|
* Appropriate locks must be held before calling this function.
|
|
|
|
*
|
|
|
|
* @nr_to_scan: The number of pages to look through on the list.
|
2012-01-13 08:20:06 +07:00
|
|
|
* @mz: The mem_cgroup_zone to pull pages from.
|
2005-04-17 05:20:36 +07:00
|
|
|
* @dst: The temp list to put pages on to.
|
2012-01-13 08:20:06 +07:00
|
|
|
* @nr_scanned: The number of pages that were scanned.
|
2012-03-22 06:33:51 +07:00
|
|
|
* @sc: The scan_control struct for this reclaim session
|
2007-07-17 18:03:16 +07:00
|
|
|
* @mode: One of the LRU isolation modes
|
2012-01-13 08:20:06 +07:00
|
|
|
* @active: True [1] if isolating active pages
|
2008-10-19 10:26:32 +07:00
|
|
|
* @file: True [1] if isolating file [!anon] pages
|
2005-04-17 05:20:36 +07:00
|
|
|
*
|
|
|
|
* returns how many pages were moved onto *@dst.
|
|
|
|
*/
|
2006-03-22 15:08:19 +07:00
|
|
|
static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
|
2012-01-13 08:20:06 +07:00
|
|
|
struct mem_cgroup_zone *mz, struct list_head *dst,
|
2012-03-22 06:33:51 +07:00
|
|
|
unsigned long *nr_scanned, struct scan_control *sc,
|
|
|
|
isolate_mode_t mode, int active, int file)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2012-01-13 08:20:06 +07:00
|
|
|
struct lruvec *lruvec;
|
|
|
|
struct list_head *src;
|
2006-03-22 15:08:19 +07:00
|
|
|
unsigned long nr_taken = 0;
|
2010-08-10 07:19:17 +07:00
|
|
|
unsigned long nr_lumpy_taken = 0;
|
|
|
|
unsigned long nr_lumpy_dirty = 0;
|
|
|
|
unsigned long nr_lumpy_failed = 0;
|
2006-03-22 15:08:23 +07:00
|
|
|
unsigned long scan;
|
2012-01-13 08:20:06 +07:00
|
|
|
int lru = LRU_BASE;
|
|
|
|
|
|
|
|
lruvec = mem_cgroup_zone_lruvec(mz->zone, mz->mem_cgroup);
|
|
|
|
if (active)
|
|
|
|
lru += LRU_ACTIVE;
|
|
|
|
if (file)
|
|
|
|
lru += LRU_FILE;
|
|
|
|
src = &lruvec->lists[lru];
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-03-22 15:08:23 +07:00
|
|
|
for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
|
2007-07-17 18:03:16 +07:00
|
|
|
struct page *page;
|
|
|
|
unsigned long pfn;
|
|
|
|
unsigned long end_pfn;
|
|
|
|
unsigned long page_pfn;
|
|
|
|
int zone_id;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
page = lru_to_page(src);
|
|
|
|
prefetchw_prev_lru_page(page, src, flags);
|
|
|
|
|
2006-09-26 13:30:55 +07:00
|
|
|
VM_BUG_ON(!PageLRU(page));
|
2006-03-22 15:07:59 +07:00
|
|
|
|
2008-10-19 10:26:32 +07:00
|
|
|
switch (__isolate_lru_page(page, mode, file)) {
|
2007-07-17 18:03:16 +07:00
|
|
|
case 0:
|
2012-01-13 08:18:15 +07:00
|
|
|
mem_cgroup_lru_del(page);
|
2007-07-17 18:03:16 +07:00
|
|
|
list_move(&page->lru, dst);
|
2011-01-14 06:47:13 +07:00
|
|
|
nr_taken += hpage_nr_pages(page);
|
2007-07-17 18:03:16 +07:00
|
|
|
break;
|
|
|
|
|
|
|
|
case -EBUSY:
|
|
|
|
/* else it is being freed elsewhere */
|
|
|
|
list_move(&page->lru, src);
|
|
|
|
continue;
|
2006-03-22 15:07:58 +07:00
|
|
|
|
2007-07-17 18:03:16 +07:00
|
|
|
default:
|
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
|
2012-03-22 06:33:51 +07:00
|
|
|
if (!sc->order || !(sc->reclaim_mode & RECLAIM_MODE_LUMPYRECLAIM))
|
2007-07-17 18:03:16 +07:00
|
|
|
continue;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Attempt to take all pages in the order aligned region
|
|
|
|
* surrounding the tag page. Only take those pages of
|
|
|
|
* the same active state as that tag page. We may safely
|
|
|
|
* round the target page pfn down to the requested order
|
2011-03-31 08:57:33 +07:00
|
|
|
* as the mem_map is guaranteed valid out to MAX_ORDER,
|
2007-07-17 18:03:16 +07:00
|
|
|
* where that page is in a different zone we will detect
|
|
|
|
* it from its zone id and abort this block scan.
|
|
|
|
*/
|
|
|
|
zone_id = page_zone_id(page);
|
|
|
|
page_pfn = page_to_pfn(page);
|
2012-03-22 06:33:51 +07:00
|
|
|
pfn = page_pfn & ~((1 << sc->order) - 1);
|
|
|
|
end_pfn = pfn + (1 << sc->order);
|
2007-07-17 18:03:16 +07:00
|
|
|
for (; pfn < end_pfn; pfn++) {
|
|
|
|
struct page *cursor_page;
|
|
|
|
|
|
|
|
/* The target page is in the block, ignore it. */
|
|
|
|
if (unlikely(pfn == page_pfn))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* Avoid holes within the zone. */
|
|
|
|
if (unlikely(!pfn_valid_within(pfn)))
|
|
|
|
break;
|
|
|
|
|
|
|
|
cursor_page = pfn_to_page(pfn);
|
2008-10-19 10:26:32 +07:00
|
|
|
|
2007-07-17 18:03:16 +07:00
|
|
|
/* Check that we have not crossed a zone boundary. */
|
|
|
|
if (unlikely(page_zone_id(cursor_page) != zone_id))
|
2010-10-27 04:21:44 +07:00
|
|
|
break;
|
2009-09-22 07:01:43 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If we don't have enough swap space, reclaiming of
|
|
|
|
* anon page which don't already have a swap slot is
|
|
|
|
* pointless.
|
|
|
|
*/
|
2012-01-11 06:08:33 +07:00
|
|
|
if (nr_swap_pages <= 0 && PageSwapBacked(cursor_page) &&
|
2010-10-27 04:21:44 +07:00
|
|
|
!PageSwapCache(cursor_page))
|
|
|
|
break;
|
2009-09-22 07:01:43 +07:00
|
|
|
|
2009-06-17 05:33:24 +07:00
|
|
|
if (__isolate_lru_page(cursor_page, mode, file) == 0) {
|
2012-01-13 08:19:29 +07:00
|
|
|
unsigned int isolated_pages;
|
|
|
|
|
2012-01-13 08:18:15 +07:00
|
|
|
mem_cgroup_lru_del(cursor_page);
|
2007-07-17 18:03:16 +07:00
|
|
|
list_move(&cursor_page->lru, dst);
|
2012-01-13 08:19:29 +07:00
|
|
|
isolated_pages = hpage_nr_pages(cursor_page);
|
|
|
|
nr_taken += isolated_pages;
|
|
|
|
nr_lumpy_taken += isolated_pages;
|
2010-08-10 07:19:17 +07:00
|
|
|
if (PageDirty(cursor_page))
|
2012-01-13 08:19:29 +07:00
|
|
|
nr_lumpy_dirty += isolated_pages;
|
2007-07-17 18:03:16 +07:00
|
|
|
scan++;
|
2012-01-13 08:19:29 +07:00
|
|
|
pfn += isolated_pages - 1;
|
2010-08-10 07:19:17 +07:00
|
|
|
} else {
|
2011-06-16 05:08:51 +07:00
|
|
|
/*
|
|
|
|
* Check if the page is freed already.
|
|
|
|
*
|
|
|
|
* We can't use page_count() as that
|
|
|
|
* requires compound_head and we don't
|
|
|
|
* have a pin on the page here. If a
|
|
|
|
* page is tail, we may or may not
|
|
|
|
* have isolated the head, so assume
|
|
|
|
* it's not free, it'd be tricky to
|
|
|
|
* track the head status without a
|
|
|
|
* page pin.
|
|
|
|
*/
|
|
|
|
if (!PageTail(cursor_page) &&
|
|
|
|
!atomic_read(&cursor_page->_count))
|
2010-10-27 04:21:44 +07:00
|
|
|
continue;
|
|
|
|
break;
|
2007-07-17 18:03:16 +07:00
|
|
|
}
|
|
|
|
}
|
2010-10-27 04:21:44 +07:00
|
|
|
|
|
|
|
/* If we break out of the loop above, lumpy reclaim failed */
|
|
|
|
if (pfn < end_pfn)
|
|
|
|
nr_lumpy_failed++;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:20:06 +07:00
|
|
|
*nr_scanned = scan;
|
2010-08-10 07:19:17 +07:00
|
|
|
|
2012-03-22 06:33:51 +07:00
|
|
|
trace_mm_vmscan_lru_isolate(sc->order,
|
2010-08-10 07:19:17 +07:00
|
|
|
nr_to_scan, scan,
|
|
|
|
nr_taken,
|
|
|
|
nr_lumpy_taken, nr_lumpy_dirty, nr_lumpy_failed,
|
2012-01-13 08:19:20 +07:00
|
|
|
mode, file);
|
2005-04-17 05:20:36 +07:00
|
|
|
return nr_taken;
|
|
|
|
}
|
|
|
|
|
vmscan: move isolate_lru_page() to vmscan.c
On large memory systems, the VM can spend way too much time scanning
through pages that it cannot (or should not) evict from memory. Not only
does it use up CPU time, but it also provokes lock contention and can
leave large systems under memory presure in a catatonic state.
This patch series improves VM scalability by:
1) putting filesystem backed, swap backed and unevictable pages
onto their own LRUs, so the system only scans the pages that it
can/should evict from memory
2) switching to two handed clock replacement for the anonymous LRUs,
so the number of pages that need to be scanned when the system
starts swapping is bound to a reasonable number
3) keeping unevictable pages off the LRU completely, so the
VM does not waste CPU time scanning them. ramfs, ramdisk,
SHM_LOCKED shared memory segments and mlock()ed VMA pages
are keept on the unevictable list.
This patch:
isolate_lru_page logically belongs to be in vmscan.c than migrate.c.
It is tough, because we don't need that function without memory migration
so there is a valid argument to have it in migrate.c. However a
subsequent patch needs to make use of it in the core mm, so we can happily
move it to vmscan.c.
Also, make the function a little more generic by not requiring that it
adds an isolated page to a given list. Callers can do that.
Note that we now have '__isolate_lru_page()', that does
something quite different, visible outside of vmscan.c
for use with memory controller. Methinks we need to
rationalize these names/purposes. --lts
[akpm@linux-foundation.org: fix mm/memory_hotplug.c build]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:09 +07:00
|
|
|
/**
|
|
|
|
* isolate_lru_page - tries to isolate a page from its LRU list
|
|
|
|
* @page: page to isolate from its LRU list
|
|
|
|
*
|
|
|
|
* Isolates a @page from an LRU list, clears PageLRU and adjusts the
|
|
|
|
* vmstat statistic corresponding to whatever LRU list the page was on.
|
|
|
|
*
|
|
|
|
* Returns 0 if the page was removed from an LRU list.
|
|
|
|
* Returns -EBUSY if the page was not on an LRU list.
|
|
|
|
*
|
|
|
|
* The returned page will have PageLRU() cleared. If it was found on
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
* the active list, it will have PageActive set. If it was found on
|
|
|
|
* the unevictable list, it will have the PageUnevictable bit set. That flag
|
|
|
|
* may need to be cleared by the caller before letting the page go.
|
vmscan: move isolate_lru_page() to vmscan.c
On large memory systems, the VM can spend way too much time scanning
through pages that it cannot (or should not) evict from memory. Not only
does it use up CPU time, but it also provokes lock contention and can
leave large systems under memory presure in a catatonic state.
This patch series improves VM scalability by:
1) putting filesystem backed, swap backed and unevictable pages
onto their own LRUs, so the system only scans the pages that it
can/should evict from memory
2) switching to two handed clock replacement for the anonymous LRUs,
so the number of pages that need to be scanned when the system
starts swapping is bound to a reasonable number
3) keeping unevictable pages off the LRU completely, so the
VM does not waste CPU time scanning them. ramfs, ramdisk,
SHM_LOCKED shared memory segments and mlock()ed VMA pages
are keept on the unevictable list.
This patch:
isolate_lru_page logically belongs to be in vmscan.c than migrate.c.
It is tough, because we don't need that function without memory migration
so there is a valid argument to have it in migrate.c. However a
subsequent patch needs to make use of it in the core mm, so we can happily
move it to vmscan.c.
Also, make the function a little more generic by not requiring that it
adds an isolated page to a given list. Callers can do that.
Note that we now have '__isolate_lru_page()', that does
something quite different, visible outside of vmscan.c
for use with memory controller. Methinks we need to
rationalize these names/purposes. --lts
[akpm@linux-foundation.org: fix mm/memory_hotplug.c build]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:09 +07:00
|
|
|
*
|
|
|
|
* The vmstat statistic corresponding to the list on which the page was
|
|
|
|
* found will be decremented.
|
|
|
|
*
|
|
|
|
* Restrictions:
|
|
|
|
* (1) Must be called with an elevated refcount on the page. This is a
|
|
|
|
* fundamentnal difference from isolate_lru_pages (which is called
|
|
|
|
* without a stable reference).
|
|
|
|
* (2) the lru_lock must not be held.
|
|
|
|
* (3) interrupts must be enabled.
|
|
|
|
*/
|
|
|
|
int isolate_lru_page(struct page *page)
|
|
|
|
{
|
|
|
|
int ret = -EBUSY;
|
|
|
|
|
2011-05-25 07:12:21 +07:00
|
|
|
VM_BUG_ON(!page_count(page));
|
|
|
|
|
vmscan: move isolate_lru_page() to vmscan.c
On large memory systems, the VM can spend way too much time scanning
through pages that it cannot (or should not) evict from memory. Not only
does it use up CPU time, but it also provokes lock contention and can
leave large systems under memory presure in a catatonic state.
This patch series improves VM scalability by:
1) putting filesystem backed, swap backed and unevictable pages
onto their own LRUs, so the system only scans the pages that it
can/should evict from memory
2) switching to two handed clock replacement for the anonymous LRUs,
so the number of pages that need to be scanned when the system
starts swapping is bound to a reasonable number
3) keeping unevictable pages off the LRU completely, so the
VM does not waste CPU time scanning them. ramfs, ramdisk,
SHM_LOCKED shared memory segments and mlock()ed VMA pages
are keept on the unevictable list.
This patch:
isolate_lru_page logically belongs to be in vmscan.c than migrate.c.
It is tough, because we don't need that function without memory migration
so there is a valid argument to have it in migrate.c. However a
subsequent patch needs to make use of it in the core mm, so we can happily
move it to vmscan.c.
Also, make the function a little more generic by not requiring that it
adds an isolated page to a given list. Callers can do that.
Note that we now have '__isolate_lru_page()', that does
something quite different, visible outside of vmscan.c
for use with memory controller. Methinks we need to
rationalize these names/purposes. --lts
[akpm@linux-foundation.org: fix mm/memory_hotplug.c build]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:09 +07:00
|
|
|
if (PageLRU(page)) {
|
|
|
|
struct zone *zone = page_zone(page);
|
|
|
|
|
|
|
|
spin_lock_irq(&zone->lru_lock);
|
2011-05-25 07:12:21 +07:00
|
|
|
if (PageLRU(page)) {
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
int lru = page_lru(page);
|
vmscan: move isolate_lru_page() to vmscan.c
On large memory systems, the VM can spend way too much time scanning
through pages that it cannot (or should not) evict from memory. Not only
does it use up CPU time, but it also provokes lock contention and can
leave large systems under memory presure in a catatonic state.
This patch series improves VM scalability by:
1) putting filesystem backed, swap backed and unevictable pages
onto their own LRUs, so the system only scans the pages that it
can/should evict from memory
2) switching to two handed clock replacement for the anonymous LRUs,
so the number of pages that need to be scanned when the system
starts swapping is bound to a reasonable number
3) keeping unevictable pages off the LRU completely, so the
VM does not waste CPU time scanning them. ramfs, ramdisk,
SHM_LOCKED shared memory segments and mlock()ed VMA pages
are keept on the unevictable list.
This patch:
isolate_lru_page logically belongs to be in vmscan.c than migrate.c.
It is tough, because we don't need that function without memory migration
so there is a valid argument to have it in migrate.c. However a
subsequent patch needs to make use of it in the core mm, so we can happily
move it to vmscan.c.
Also, make the function a little more generic by not requiring that it
adds an isolated page to a given list. Callers can do that.
Note that we now have '__isolate_lru_page()', that does
something quite different, visible outside of vmscan.c
for use with memory controller. Methinks we need to
rationalize these names/purposes. --lts
[akpm@linux-foundation.org: fix mm/memory_hotplug.c build]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:09 +07:00
|
|
|
ret = 0;
|
2011-05-25 07:12:21 +07:00
|
|
|
get_page(page);
|
vmscan: move isolate_lru_page() to vmscan.c
On large memory systems, the VM can spend way too much time scanning
through pages that it cannot (or should not) evict from memory. Not only
does it use up CPU time, but it also provokes lock contention and can
leave large systems under memory presure in a catatonic state.
This patch series improves VM scalability by:
1) putting filesystem backed, swap backed and unevictable pages
onto their own LRUs, so the system only scans the pages that it
can/should evict from memory
2) switching to two handed clock replacement for the anonymous LRUs,
so the number of pages that need to be scanned when the system
starts swapping is bound to a reasonable number
3) keeping unevictable pages off the LRU completely, so the
VM does not waste CPU time scanning them. ramfs, ramdisk,
SHM_LOCKED shared memory segments and mlock()ed VMA pages
are keept on the unevictable list.
This patch:
isolate_lru_page logically belongs to be in vmscan.c than migrate.c.
It is tough, because we don't need that function without memory migration
so there is a valid argument to have it in migrate.c. However a
subsequent patch needs to make use of it in the core mm, so we can happily
move it to vmscan.c.
Also, make the function a little more generic by not requiring that it
adds an isolated page to a given list. Callers can do that.
Note that we now have '__isolate_lru_page()', that does
something quite different, visible outside of vmscan.c
for use with memory controller. Methinks we need to
rationalize these names/purposes. --lts
[akpm@linux-foundation.org: fix mm/memory_hotplug.c build]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:09 +07:00
|
|
|
ClearPageLRU(page);
|
2008-10-19 10:26:32 +07:00
|
|
|
|
|
|
|
del_page_from_lru_list(zone, page, lru);
|
vmscan: move isolate_lru_page() to vmscan.c
On large memory systems, the VM can spend way too much time scanning
through pages that it cannot (or should not) evict from memory. Not only
does it use up CPU time, but it also provokes lock contention and can
leave large systems under memory presure in a catatonic state.
This patch series improves VM scalability by:
1) putting filesystem backed, swap backed and unevictable pages
onto their own LRUs, so the system only scans the pages that it
can/should evict from memory
2) switching to two handed clock replacement for the anonymous LRUs,
so the number of pages that need to be scanned when the system
starts swapping is bound to a reasonable number
3) keeping unevictable pages off the LRU completely, so the
VM does not waste CPU time scanning them. ramfs, ramdisk,
SHM_LOCKED shared memory segments and mlock()ed VMA pages
are keept on the unevictable list.
This patch:
isolate_lru_page logically belongs to be in vmscan.c than migrate.c.
It is tough, because we don't need that function without memory migration
so there is a valid argument to have it in migrate.c. However a
subsequent patch needs to make use of it in the core mm, so we can happily
move it to vmscan.c.
Also, make the function a little more generic by not requiring that it
adds an isolated page to a given list. Callers can do that.
Note that we now have '__isolate_lru_page()', that does
something quite different, visible outside of vmscan.c
for use with memory controller. Methinks we need to
rationalize these names/purposes. --lts
[akpm@linux-foundation.org: fix mm/memory_hotplug.c build]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:09 +07:00
|
|
|
}
|
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-09-22 07:01:38 +07:00
|
|
|
/*
|
|
|
|
* Are there way too many processes in the direct reclaim path already?
|
|
|
|
*/
|
|
|
|
static int too_many_isolated(struct zone *zone, int file,
|
|
|
|
struct scan_control *sc)
|
|
|
|
{
|
|
|
|
unsigned long inactive, isolated;
|
|
|
|
|
|
|
|
if (current_is_kswapd())
|
|
|
|
return 0;
|
|
|
|
|
2012-01-13 08:17:50 +07:00
|
|
|
if (!global_reclaim(sc))
|
2009-09-22 07:01:38 +07:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (file) {
|
|
|
|
inactive = zone_page_state(zone, NR_INACTIVE_FILE);
|
|
|
|
isolated = zone_page_state(zone, NR_ISOLATED_FILE);
|
|
|
|
} else {
|
|
|
|
inactive = zone_page_state(zone, NR_INACTIVE_ANON);
|
|
|
|
isolated = zone_page_state(zone, NR_ISOLATED_ANON);
|
|
|
|
}
|
|
|
|
|
|
|
|
return isolated > inactive;
|
|
|
|
}
|
|
|
|
|
2010-08-10 07:19:30 +07:00
|
|
|
static noinline_for_stack void
|
2012-01-13 08:20:07 +07:00
|
|
|
putback_inactive_pages(struct mem_cgroup_zone *mz,
|
|
|
|
struct list_head *page_list)
|
2010-08-10 07:19:30 +07:00
|
|
|
{
|
2012-01-13 08:17:52 +07:00
|
|
|
struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz);
|
2012-01-13 08:20:07 +07:00
|
|
|
struct zone *zone = mz->zone;
|
|
|
|
LIST_HEAD(pages_to_free);
|
2010-08-10 07:19:30 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Put back any unfreeable pages.
|
|
|
|
*/
|
|
|
|
while (!list_empty(page_list)) {
|
2012-01-13 08:20:07 +07:00
|
|
|
struct page *page = lru_to_page(page_list);
|
2010-08-10 07:19:30 +07:00
|
|
|
int lru;
|
2012-01-13 08:20:07 +07:00
|
|
|
|
2010-08-10 07:19:30 +07:00
|
|
|
VM_BUG_ON(PageLRU(page));
|
|
|
|
list_del(&page->lru);
|
|
|
|
if (unlikely(!page_evictable(page, NULL))) {
|
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
|
|
|
putback_lru_page(page);
|
|
|
|
spin_lock_irq(&zone->lru_lock);
|
|
|
|
continue;
|
|
|
|
}
|
2011-01-18 05:42:19 +07:00
|
|
|
SetPageLRU(page);
|
2010-08-10 07:19:30 +07:00
|
|
|
lru = page_lru(page);
|
2011-01-18 05:42:19 +07:00
|
|
|
add_page_to_lru_list(zone, page, lru);
|
2010-08-10 07:19:30 +07:00
|
|
|
if (is_active_lru(lru)) {
|
|
|
|
int file = is_file_lru(lru);
|
2011-01-14 06:47:13 +07:00
|
|
|
int numpages = hpage_nr_pages(page);
|
|
|
|
reclaim_stat->recent_rotated[file] += numpages;
|
2010-08-10 07:19:30 +07:00
|
|
|
}
|
2012-01-13 08:19:56 +07:00
|
|
|
if (put_page_testzero(page)) {
|
|
|
|
__ClearPageLRU(page);
|
|
|
|
__ClearPageActive(page);
|
|
|
|
del_page_from_lru_list(zone, page, lru);
|
|
|
|
|
|
|
|
if (unlikely(PageCompound(page))) {
|
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
|
|
|
(*get_compound_page_dtor(page))(page);
|
|
|
|
spin_lock_irq(&zone->lru_lock);
|
|
|
|
} else
|
|
|
|
list_add(&page->lru, &pages_to_free);
|
2010-08-10 07:19:30 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-01-13 08:20:07 +07:00
|
|
|
/*
|
|
|
|
* To save our caller's stack, now use input list for pages to free.
|
|
|
|
*/
|
|
|
|
list_splice(&pages_to_free, page_list);
|
2010-08-10 07:19:30 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static noinline_for_stack void
|
|
|
|
update_isolated_counts(struct mem_cgroup_zone *mz,
|
2012-01-13 08:20:07 +07:00
|
|
|
struct list_head *page_list,
|
2012-01-13 08:17:52 +07:00
|
|
|
unsigned long *nr_anon,
|
2012-01-13 08:20:07 +07:00
|
|
|
unsigned long *nr_file)
|
2010-08-10 07:19:33 +07:00
|
|
|
{
|
2012-01-13 08:17:52 +07:00
|
|
|
struct zone *zone = mz->zone;
|
2010-08-10 07:19:33 +07:00
|
|
|
unsigned int count[NR_LRU_LISTS] = { 0, };
|
2012-01-13 08:20:07 +07:00
|
|
|
unsigned long nr_active = 0;
|
|
|
|
struct page *page;
|
|
|
|
int lru;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Count pages and clear active flags
|
|
|
|
*/
|
|
|
|
list_for_each_entry(page, page_list, lru) {
|
|
|
|
int numpages = hpage_nr_pages(page);
|
|
|
|
lru = page_lru_base_type(page);
|
|
|
|
if (PageActive(page)) {
|
|
|
|
lru += LRU_ACTIVE;
|
|
|
|
ClearPageActive(page);
|
|
|
|
nr_active += numpages;
|
|
|
|
}
|
|
|
|
count[lru] += numpages;
|
|
|
|
}
|
2010-08-10 07:19:33 +07:00
|
|
|
|
2012-03-22 06:34:02 +07:00
|
|
|
preempt_disable();
|
2010-08-10 07:19:33 +07:00
|
|
|
__count_vm_events(PGDEACTIVATE, nr_active);
|
|
|
|
|
|
|
|
__mod_zone_page_state(zone, NR_ACTIVE_FILE,
|
|
|
|
-count[LRU_ACTIVE_FILE]);
|
|
|
|
__mod_zone_page_state(zone, NR_INACTIVE_FILE,
|
|
|
|
-count[LRU_INACTIVE_FILE]);
|
|
|
|
__mod_zone_page_state(zone, NR_ACTIVE_ANON,
|
|
|
|
-count[LRU_ACTIVE_ANON]);
|
|
|
|
__mod_zone_page_state(zone, NR_INACTIVE_ANON,
|
|
|
|
-count[LRU_INACTIVE_ANON]);
|
|
|
|
|
|
|
|
*nr_anon = count[LRU_ACTIVE_ANON] + count[LRU_INACTIVE_ANON];
|
|
|
|
*nr_file = count[LRU_ACTIVE_FILE] + count[LRU_INACTIVE_FILE];
|
|
|
|
|
2012-03-22 06:34:02 +07:00
|
|
|
__mod_zone_page_state(zone, NR_ISOLATED_ANON, *nr_anon);
|
|
|
|
__mod_zone_page_state(zone, NR_ISOLATED_FILE, *nr_file);
|
|
|
|
preempt_enable();
|
2010-08-10 07:19:33 +07:00
|
|
|
}
|
|
|
|
|
vmscan: raise the bar to PAGEOUT_IO_SYNC stalls
Fix "system goes unresponsive under memory pressure and lots of
dirty/writeback pages" bug.
http://lkml.org/lkml/2010/4/4/86
In the above thread, Andreas Mohr described that
Invoking any command locked up for minutes (note that I'm
talking about attempted additional I/O to the _other_,
_unaffected_ main system HDD - such as loading some shell
binaries -, NOT the external SSD18M!!).
This happens when the two conditions are both meet:
- under memory pressure
- writing heavily to a slow device
OOM also happens in Andreas' system. The OOM trace shows that 3 processes
are stuck in wait_on_page_writeback() in the direct reclaim path. One in
do_fork() and the other two in unix_stream_sendmsg(). They are blocked on
this condition:
(sc->order && priority < DEF_PRIORITY - 2)
which was introduced in commit 78dc583d (vmscan: low order lumpy reclaim
also should use PAGEOUT_IO_SYNC) one year ago. That condition may be too
permissive. In Andreas' case, 512MB/1024 = 512KB. If the direct reclaim
for the order-1 fork() allocation runs into a range of 512KB
hard-to-reclaim LRU pages, it will be stalled.
It's a severe problem in three ways.
Firstly, it can easily happen in daily desktop usage. vmscan priority can
easily go below (DEF_PRIORITY - 2) on _local_ memory pressure. Even if
the system has 50% globally reclaimable pages, it still has good
opportunity to have 0.1% sized hard-to-reclaim ranges. For example, a
simple dd can easily create a big range (up to 20%) of dirty pages in the
LRU lists. And order-1 to order-3 allocations are more than common with
SLUB. Try "grep -v '1 :' /proc/slabinfo" to get the list of high order
slab caches. For example, the order-1 radix_tree_node slab cache may
stall applications at swap-in time; the order-3 inode cache on most
filesystems may stall applications when trying to read some file; the
order-2 proc_inode_cache may stall applications when trying to open a
/proc file.
Secondly, once triggered, it will stall unrelated processes (not doing IO
at all) in the system. This "one slow USB device stalls the whole system"
avalanching effect is very bad.
Thirdly, once stalled, the stall time could be intolerable long for the
users. When there are 20MB queued writeback pages and USB 1.1 is writing
them in 1MB/s, wait_on_page_writeback() will stuck for up to 20 seconds.
Not to mention it may be called multiple times.
So raise the bar to only enable PAGEOUT_IO_SYNC when priority goes below
DEF_PRIORITY/3, or 6.25% LRU size. As the default dirty throttle ratio is
20%, it will hardly be triggered by pure dirty pages. We'd better treat
PAGEOUT_IO_SYNC as some last resort workaround -- its stall time is so
uncomfortably long (easily goes beyond 1s).
The bar is only raised for (order < PAGE_ALLOC_COSTLY_ORDER) allocations,
which are easy to satisfy in 1TB memory boxes. So, although 6.25% of
memory could be an awful lot of pages to scan on a system with 1TB of
memory, it won't really have to busy scan that much.
Andreas tested an older version of this patch and reported that it mostly
fixed his problem. Mel Gorman helped improve it and KOSAKI Motohiro will
fix it further in the next patch.
Reported-by: Andreas Mohr <andi@lisas.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-10 07:20:01 +07:00
|
|
|
/*
|
2011-11-01 07:07:42 +07:00
|
|
|
* Returns true if a direct reclaim should wait on pages under writeback.
|
vmscan: raise the bar to PAGEOUT_IO_SYNC stalls
Fix "system goes unresponsive under memory pressure and lots of
dirty/writeback pages" bug.
http://lkml.org/lkml/2010/4/4/86
In the above thread, Andreas Mohr described that
Invoking any command locked up for minutes (note that I'm
talking about attempted additional I/O to the _other_,
_unaffected_ main system HDD - such as loading some shell
binaries -, NOT the external SSD18M!!).
This happens when the two conditions are both meet:
- under memory pressure
- writing heavily to a slow device
OOM also happens in Andreas' system. The OOM trace shows that 3 processes
are stuck in wait_on_page_writeback() in the direct reclaim path. One in
do_fork() and the other two in unix_stream_sendmsg(). They are blocked on
this condition:
(sc->order && priority < DEF_PRIORITY - 2)
which was introduced in commit 78dc583d (vmscan: low order lumpy reclaim
also should use PAGEOUT_IO_SYNC) one year ago. That condition may be too
permissive. In Andreas' case, 512MB/1024 = 512KB. If the direct reclaim
for the order-1 fork() allocation runs into a range of 512KB
hard-to-reclaim LRU pages, it will be stalled.
It's a severe problem in three ways.
Firstly, it can easily happen in daily desktop usage. vmscan priority can
easily go below (DEF_PRIORITY - 2) on _local_ memory pressure. Even if
the system has 50% globally reclaimable pages, it still has good
opportunity to have 0.1% sized hard-to-reclaim ranges. For example, a
simple dd can easily create a big range (up to 20%) of dirty pages in the
LRU lists. And order-1 to order-3 allocations are more than common with
SLUB. Try "grep -v '1 :' /proc/slabinfo" to get the list of high order
slab caches. For example, the order-1 radix_tree_node slab cache may
stall applications at swap-in time; the order-3 inode cache on most
filesystems may stall applications when trying to read some file; the
order-2 proc_inode_cache may stall applications when trying to open a
/proc file.
Secondly, once triggered, it will stall unrelated processes (not doing IO
at all) in the system. This "one slow USB device stalls the whole system"
avalanching effect is very bad.
Thirdly, once stalled, the stall time could be intolerable long for the
users. When there are 20MB queued writeback pages and USB 1.1 is writing
them in 1MB/s, wait_on_page_writeback() will stuck for up to 20 seconds.
Not to mention it may be called multiple times.
So raise the bar to only enable PAGEOUT_IO_SYNC when priority goes below
DEF_PRIORITY/3, or 6.25% LRU size. As the default dirty throttle ratio is
20%, it will hardly be triggered by pure dirty pages. We'd better treat
PAGEOUT_IO_SYNC as some last resort workaround -- its stall time is so
uncomfortably long (easily goes beyond 1s).
The bar is only raised for (order < PAGE_ALLOC_COSTLY_ORDER) allocations,
which are easy to satisfy in 1TB memory boxes. So, although 6.25% of
memory could be an awful lot of pages to scan on a system with 1TB of
memory, it won't really have to busy scan that much.
Andreas tested an older version of this patch and reported that it mostly
fixed his problem. Mel Gorman helped improve it and KOSAKI Motohiro will
fix it further in the next patch.
Reported-by: Andreas Mohr <andi@lisas.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-10 07:20:01 +07:00
|
|
|
*
|
|
|
|
* If we are direct reclaiming for contiguous pages and we do not reclaim
|
|
|
|
* everything in the list, try again and wait for writeback IO to complete.
|
|
|
|
* This will stall high-order allocations noticeably. Only do that when really
|
|
|
|
* need to free the pages under high memory pressure.
|
|
|
|
*/
|
|
|
|
static inline bool should_reclaim_stall(unsigned long nr_taken,
|
|
|
|
unsigned long nr_freed,
|
|
|
|
int priority,
|
|
|
|
struct scan_control *sc)
|
|
|
|
{
|
|
|
|
int lumpy_stall_priority;
|
|
|
|
|
|
|
|
/* kswapd should not stall on sync IO */
|
|
|
|
if (current_is_kswapd())
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* Only stall on lumpy reclaim */
|
2011-01-14 06:46:00 +07:00
|
|
|
if (sc->reclaim_mode & RECLAIM_MODE_SINGLE)
|
vmscan: raise the bar to PAGEOUT_IO_SYNC stalls
Fix "system goes unresponsive under memory pressure and lots of
dirty/writeback pages" bug.
http://lkml.org/lkml/2010/4/4/86
In the above thread, Andreas Mohr described that
Invoking any command locked up for minutes (note that I'm
talking about attempted additional I/O to the _other_,
_unaffected_ main system HDD - such as loading some shell
binaries -, NOT the external SSD18M!!).
This happens when the two conditions are both meet:
- under memory pressure
- writing heavily to a slow device
OOM also happens in Andreas' system. The OOM trace shows that 3 processes
are stuck in wait_on_page_writeback() in the direct reclaim path. One in
do_fork() and the other two in unix_stream_sendmsg(). They are blocked on
this condition:
(sc->order && priority < DEF_PRIORITY - 2)
which was introduced in commit 78dc583d (vmscan: low order lumpy reclaim
also should use PAGEOUT_IO_SYNC) one year ago. That condition may be too
permissive. In Andreas' case, 512MB/1024 = 512KB. If the direct reclaim
for the order-1 fork() allocation runs into a range of 512KB
hard-to-reclaim LRU pages, it will be stalled.
It's a severe problem in three ways.
Firstly, it can easily happen in daily desktop usage. vmscan priority can
easily go below (DEF_PRIORITY - 2) on _local_ memory pressure. Even if
the system has 50% globally reclaimable pages, it still has good
opportunity to have 0.1% sized hard-to-reclaim ranges. For example, a
simple dd can easily create a big range (up to 20%) of dirty pages in the
LRU lists. And order-1 to order-3 allocations are more than common with
SLUB. Try "grep -v '1 :' /proc/slabinfo" to get the list of high order
slab caches. For example, the order-1 radix_tree_node slab cache may
stall applications at swap-in time; the order-3 inode cache on most
filesystems may stall applications when trying to read some file; the
order-2 proc_inode_cache may stall applications when trying to open a
/proc file.
Secondly, once triggered, it will stall unrelated processes (not doing IO
at all) in the system. This "one slow USB device stalls the whole system"
avalanching effect is very bad.
Thirdly, once stalled, the stall time could be intolerable long for the
users. When there are 20MB queued writeback pages and USB 1.1 is writing
them in 1MB/s, wait_on_page_writeback() will stuck for up to 20 seconds.
Not to mention it may be called multiple times.
So raise the bar to only enable PAGEOUT_IO_SYNC when priority goes below
DEF_PRIORITY/3, or 6.25% LRU size. As the default dirty throttle ratio is
20%, it will hardly be triggered by pure dirty pages. We'd better treat
PAGEOUT_IO_SYNC as some last resort workaround -- its stall time is so
uncomfortably long (easily goes beyond 1s).
The bar is only raised for (order < PAGE_ALLOC_COSTLY_ORDER) allocations,
which are easy to satisfy in 1TB memory boxes. So, although 6.25% of
memory could be an awful lot of pages to scan on a system with 1TB of
memory, it won't really have to busy scan that much.
Andreas tested an older version of this patch and reported that it mostly
fixed his problem. Mel Gorman helped improve it and KOSAKI Motohiro will
fix it further in the next patch.
Reported-by: Andreas Mohr <andi@lisas.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-10 07:20:01 +07:00
|
|
|
return false;
|
|
|
|
|
2011-08-23 23:28:02 +07:00
|
|
|
/* If we have reclaimed everything on the isolated list, no stall */
|
vmscan: raise the bar to PAGEOUT_IO_SYNC stalls
Fix "system goes unresponsive under memory pressure and lots of
dirty/writeback pages" bug.
http://lkml.org/lkml/2010/4/4/86
In the above thread, Andreas Mohr described that
Invoking any command locked up for minutes (note that I'm
talking about attempted additional I/O to the _other_,
_unaffected_ main system HDD - such as loading some shell
binaries -, NOT the external SSD18M!!).
This happens when the two conditions are both meet:
- under memory pressure
- writing heavily to a slow device
OOM also happens in Andreas' system. The OOM trace shows that 3 processes
are stuck in wait_on_page_writeback() in the direct reclaim path. One in
do_fork() and the other two in unix_stream_sendmsg(). They are blocked on
this condition:
(sc->order && priority < DEF_PRIORITY - 2)
which was introduced in commit 78dc583d (vmscan: low order lumpy reclaim
also should use PAGEOUT_IO_SYNC) one year ago. That condition may be too
permissive. In Andreas' case, 512MB/1024 = 512KB. If the direct reclaim
for the order-1 fork() allocation runs into a range of 512KB
hard-to-reclaim LRU pages, it will be stalled.
It's a severe problem in three ways.
Firstly, it can easily happen in daily desktop usage. vmscan priority can
easily go below (DEF_PRIORITY - 2) on _local_ memory pressure. Even if
the system has 50% globally reclaimable pages, it still has good
opportunity to have 0.1% sized hard-to-reclaim ranges. For example, a
simple dd can easily create a big range (up to 20%) of dirty pages in the
LRU lists. And order-1 to order-3 allocations are more than common with
SLUB. Try "grep -v '1 :' /proc/slabinfo" to get the list of high order
slab caches. For example, the order-1 radix_tree_node slab cache may
stall applications at swap-in time; the order-3 inode cache on most
filesystems may stall applications when trying to read some file; the
order-2 proc_inode_cache may stall applications when trying to open a
/proc file.
Secondly, once triggered, it will stall unrelated processes (not doing IO
at all) in the system. This "one slow USB device stalls the whole system"
avalanching effect is very bad.
Thirdly, once stalled, the stall time could be intolerable long for the
users. When there are 20MB queued writeback pages and USB 1.1 is writing
them in 1MB/s, wait_on_page_writeback() will stuck for up to 20 seconds.
Not to mention it may be called multiple times.
So raise the bar to only enable PAGEOUT_IO_SYNC when priority goes below
DEF_PRIORITY/3, or 6.25% LRU size. As the default dirty throttle ratio is
20%, it will hardly be triggered by pure dirty pages. We'd better treat
PAGEOUT_IO_SYNC as some last resort workaround -- its stall time is so
uncomfortably long (easily goes beyond 1s).
The bar is only raised for (order < PAGE_ALLOC_COSTLY_ORDER) allocations,
which are easy to satisfy in 1TB memory boxes. So, although 6.25% of
memory could be an awful lot of pages to scan on a system with 1TB of
memory, it won't really have to busy scan that much.
Andreas tested an older version of this patch and reported that it mostly
fixed his problem. Mel Gorman helped improve it and KOSAKI Motohiro will
fix it further in the next patch.
Reported-by: Andreas Mohr <andi@lisas.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-10 07:20:01 +07:00
|
|
|
if (nr_freed == nr_taken)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* For high-order allocations, there are two stall thresholds.
|
|
|
|
* High-cost allocations stall immediately where as lower
|
|
|
|
* order allocations such as stacks require the scanning
|
|
|
|
* priority to be much higher before stalling.
|
|
|
|
*/
|
|
|
|
if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
|
|
|
|
lumpy_stall_priority = DEF_PRIORITY;
|
|
|
|
else
|
|
|
|
lumpy_stall_priority = DEF_PRIORITY / 3;
|
|
|
|
|
|
|
|
return priority <= lumpy_stall_priority;
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
[PATCH] vmscan: rename functions
We have:
try_to_free_pages
->shrink_caches(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_cache(struct zone *, ...)
->shrink_list(struct list_head *, ...)
->refill_inactive_list((struct zone *, ...)
which is fairly irrational.
Rename things so that we have
try_to_free_pages
->shrink_zones(struct zone **zones, ..)
->shrink_zone(struct zone *, ...)
->shrink_inactive_list(struct zone *, ...)
->shrink_page_list(struct list_head *, ...)
->shrink_active_list(struct zone *, ...)
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-22 15:08:21 +07:00
|
|
|
* shrink_inactive_list() is a helper for shrink_zone(). It returns the number
|
|
|
|
* of reclaimed pages
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2010-08-10 07:19:30 +07:00
|
|
|
static noinline_for_stack unsigned long
|
2012-01-13 08:17:52 +07:00
|
|
|
shrink_inactive_list(unsigned long nr_to_scan, struct mem_cgroup_zone *mz,
|
|
|
|
struct scan_control *sc, int priority, int file)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
LIST_HEAD(page_list);
|
2010-08-10 07:19:28 +07:00
|
|
|
unsigned long nr_scanned;
|
2006-03-22 15:08:20 +07:00
|
|
|
unsigned long nr_reclaimed = 0;
|
2010-08-10 07:19:28 +07:00
|
|
|
unsigned long nr_taken;
|
|
|
|
unsigned long nr_anon;
|
|
|
|
unsigned long nr_file;
|
2011-11-01 07:07:56 +07:00
|
|
|
unsigned long nr_dirty = 0;
|
|
|
|
unsigned long nr_writeback = 0;
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode_t isolate_mode = ISOLATE_INACTIVE;
|
2012-01-13 08:17:52 +07:00
|
|
|
struct zone *zone = mz->zone;
|
2012-03-22 06:34:02 +07:00
|
|
|
struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz);
|
2009-06-17 05:31:40 +07:00
|
|
|
|
2009-09-22 07:01:38 +07:00
|
|
|
while (unlikely(too_many_isolated(zone, file, sc))) {
|
2009-10-27 06:49:35 +07:00
|
|
|
congestion_wait(BLK_RW_ASYNC, HZ/10);
|
2009-09-22 07:01:38 +07:00
|
|
|
|
|
|
|
/* We are about to die and free our memory. Return now. */
|
|
|
|
if (fatal_signal_pending(current))
|
|
|
|
return SWAP_CLUSTER_MAX;
|
|
|
|
}
|
|
|
|
|
2011-01-14 06:46:00 +07:00
|
|
|
set_reclaim_mode(priority, sc, false);
|
2011-11-01 07:06:47 +07:00
|
|
|
if (sc->reclaim_mode & RECLAIM_MODE_LUMPYRECLAIM)
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode |= ISOLATE_ACTIVE;
|
2011-11-01 07:06:47 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
lru_add_drain();
|
2011-11-01 07:06:55 +07:00
|
|
|
|
|
|
|
if (!sc->may_unmap)
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode |= ISOLATE_UNMAPPED;
|
2011-11-01 07:06:55 +07:00
|
|
|
if (!sc->may_writepage)
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode |= ISOLATE_CLEAN;
|
2011-11-01 07:06:55 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
spin_lock_irq(&zone->lru_lock);
|
2009-09-22 07:01:36 +07:00
|
|
|
|
2012-03-22 06:33:51 +07:00
|
|
|
nr_taken = isolate_lru_pages(nr_to_scan, mz, &page_list, &nr_scanned,
|
|
|
|
sc, isolate_mode, 0, file);
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc)) {
|
2010-08-10 07:19:28 +07:00
|
|
|
zone->pages_scanned += nr_scanned;
|
|
|
|
if (current_is_kswapd())
|
|
|
|
__count_zone_vm_events(PGSCAN_KSWAPD, zone,
|
|
|
|
nr_scanned);
|
|
|
|
else
|
|
|
|
__count_zone_vm_events(PGSCAN_DIRECT, zone,
|
|
|
|
nr_scanned);
|
|
|
|
}
|
2012-03-22 06:34:02 +07:00
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
2009-09-22 07:01:36 +07:00
|
|
|
|
2012-03-22 06:34:02 +07:00
|
|
|
if (nr_taken == 0)
|
2010-08-10 07:19:30 +07:00
|
|
|
return 0;
|
2007-07-17 18:03:16 +07:00
|
|
|
|
2012-01-13 08:20:07 +07:00
|
|
|
update_isolated_counts(mz, &page_list, &nr_anon, &nr_file);
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
nr_reclaimed = shrink_page_list(&page_list, mz, sc, priority,
|
2011-11-01 07:07:56 +07:00
|
|
|
&nr_dirty, &nr_writeback);
|
2007-08-23 04:01:26 +07:00
|
|
|
|
vmscan: raise the bar to PAGEOUT_IO_SYNC stalls
Fix "system goes unresponsive under memory pressure and lots of
dirty/writeback pages" bug.
http://lkml.org/lkml/2010/4/4/86
In the above thread, Andreas Mohr described that
Invoking any command locked up for minutes (note that I'm
talking about attempted additional I/O to the _other_,
_unaffected_ main system HDD - such as loading some shell
binaries -, NOT the external SSD18M!!).
This happens when the two conditions are both meet:
- under memory pressure
- writing heavily to a slow device
OOM also happens in Andreas' system. The OOM trace shows that 3 processes
are stuck in wait_on_page_writeback() in the direct reclaim path. One in
do_fork() and the other two in unix_stream_sendmsg(). They are blocked on
this condition:
(sc->order && priority < DEF_PRIORITY - 2)
which was introduced in commit 78dc583d (vmscan: low order lumpy reclaim
also should use PAGEOUT_IO_SYNC) one year ago. That condition may be too
permissive. In Andreas' case, 512MB/1024 = 512KB. If the direct reclaim
for the order-1 fork() allocation runs into a range of 512KB
hard-to-reclaim LRU pages, it will be stalled.
It's a severe problem in three ways.
Firstly, it can easily happen in daily desktop usage. vmscan priority can
easily go below (DEF_PRIORITY - 2) on _local_ memory pressure. Even if
the system has 50% globally reclaimable pages, it still has good
opportunity to have 0.1% sized hard-to-reclaim ranges. For example, a
simple dd can easily create a big range (up to 20%) of dirty pages in the
LRU lists. And order-1 to order-3 allocations are more than common with
SLUB. Try "grep -v '1 :' /proc/slabinfo" to get the list of high order
slab caches. For example, the order-1 radix_tree_node slab cache may
stall applications at swap-in time; the order-3 inode cache on most
filesystems may stall applications when trying to read some file; the
order-2 proc_inode_cache may stall applications when trying to open a
/proc file.
Secondly, once triggered, it will stall unrelated processes (not doing IO
at all) in the system. This "one slow USB device stalls the whole system"
avalanching effect is very bad.
Thirdly, once stalled, the stall time could be intolerable long for the
users. When there are 20MB queued writeback pages and USB 1.1 is writing
them in 1MB/s, wait_on_page_writeback() will stuck for up to 20 seconds.
Not to mention it may be called multiple times.
So raise the bar to only enable PAGEOUT_IO_SYNC when priority goes below
DEF_PRIORITY/3, or 6.25% LRU size. As the default dirty throttle ratio is
20%, it will hardly be triggered by pure dirty pages. We'd better treat
PAGEOUT_IO_SYNC as some last resort workaround -- its stall time is so
uncomfortably long (easily goes beyond 1s).
The bar is only raised for (order < PAGE_ALLOC_COSTLY_ORDER) allocations,
which are easy to satisfy in 1TB memory boxes. So, although 6.25% of
memory could be an awful lot of pages to scan on a system with 1TB of
memory, it won't really have to busy scan that much.
Andreas tested an older version of this patch and reported that it mostly
fixed his problem. Mel Gorman helped improve it and KOSAKI Motohiro will
fix it further in the next patch.
Reported-by: Andreas Mohr <andi@lisas.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-10 07:20:01 +07:00
|
|
|
/* Check if we should syncronously wait for writeback */
|
|
|
|
if (should_reclaim_stall(nr_taken, nr_reclaimed, priority, sc)) {
|
2011-01-14 06:46:00 +07:00
|
|
|
set_reclaim_mode(priority, sc, true);
|
2012-01-13 08:17:52 +07:00
|
|
|
nr_reclaimed += shrink_page_list(&page_list, mz, sc,
|
2011-11-01 07:07:56 +07:00
|
|
|
priority, &nr_dirty, &nr_writeback);
|
2010-08-10 07:19:28 +07:00
|
|
|
}
|
2009-09-22 07:01:36 +07:00
|
|
|
|
2012-01-13 08:20:07 +07:00
|
|
|
spin_lock_irq(&zone->lru_lock);
|
|
|
|
|
2012-03-22 06:34:02 +07:00
|
|
|
reclaim_stat->recent_scanned[0] += nr_anon;
|
|
|
|
reclaim_stat->recent_scanned[1] += nr_file;
|
|
|
|
|
2012-04-26 06:01:48 +07:00
|
|
|
if (global_reclaim(sc)) {
|
|
|
|
if (current_is_kswapd())
|
|
|
|
__count_zone_vm_events(PGSTEAL_KSWAPD, zone,
|
|
|
|
nr_reclaimed);
|
|
|
|
else
|
|
|
|
__count_zone_vm_events(PGSTEAL_DIRECT, zone,
|
|
|
|
nr_reclaimed);
|
|
|
|
}
|
2006-01-06 15:11:20 +07:00
|
|
|
|
2012-01-13 08:20:07 +07:00
|
|
|
putback_inactive_pages(mz, &page_list);
|
|
|
|
|
|
|
|
__mod_zone_page_state(zone, NR_ISOLATED_ANON, -nr_anon);
|
|
|
|
__mod_zone_page_state(zone, NR_ISOLATED_FILE, -nr_file);
|
|
|
|
|
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
|
|
|
|
|
|
|
free_hot_cold_page_list(&page_list, 1);
|
tracing, vmscan: add trace events for LRU list shrinking
There have been numerous reports of stalls that pointed at the problem
being somewhere in the VM. There are multiple roots to the problems which
means dealing with any of the root problems in isolation is tricky to
justify on their own and they would still need integration testing. This
patch series puts together two different patch sets which in combination
should tackle some of the root causes of latency problems being reported.
Patch 1 adds a tracepoint for shrink_inactive_list. For this series, the
most important results is being able to calculate the scanning/reclaim
ratio as a measure of the amount of work being done by page reclaim.
Patch 2 accounts for time spent in congestion_wait.
Patches 3-6 were originally developed by Kosaki Motohiro but reworked for
this series. It has been noted that lumpy reclaim is far too aggressive
and trashes the system somewhat. As SLUB uses high-order allocations, a
large cost incurred by lumpy reclaim will be noticeable. It was also
reported during transparent hugepage support testing that lumpy reclaim
was trashing the system and these patches should mitigate that problem
without disabling lumpy reclaim.
Patch 7 adds wait_iff_congested() and replaces some callers of
congestion_wait(). wait_iff_congested() only sleeps if there is a BDI
that is currently congested. Patch 8 notes that any BDI being congested
is not necessarily a problem because there could be multiple BDIs of
varying speeds and numberous zones. It attempts to track when a zone
being reclaimed contains many pages backed by a congested BDI and if so,
reclaimers wait on the congestion queue.
I ran a number of tests with monitoring on X86, X86-64 and PPC64. Each
machine had 3G of RAM and the CPUs were
X86: Intel P4 2-core
X86-64: AMD Phenom 4-core
PPC64: PPC970MP
Each used a single disk and the onboard IO controller. Dirty ratio was
left at 20. I'm just going to report for X86-64 and PPC64 in a vague
attempt to keep this report short. Four kernels were tested each based on
v2.6.36-rc4
traceonly-v2r2: Patches 1 and 2 to instrument vmscan reclaims and congestion_wait
lowlumpy-v2r3: Patches 1-6 to test if lumpy reclaim is better
waitcongest-v2r3: Patches 1-7 to only wait on congestion
waitwriteback-v2r4: Patches 1-8 to detect when a zone is congested
nocongest-v1r5: Patches 1-3 for testing wait_iff_congestion
nodirect-v1r5: Patches 1-10 to disable filesystem writeback for better IO
The tests run were as follows
kernbench
compile-based benchmark. Smoke test performance
sysbench
OLTP read-only benchmark. Will be re-run in the future as read-write
micro-mapped-file-stream
This is a micro-benchmark from Johannes Weiner that accesses a
large sparse-file through mmap(). It was configured to run in only
single-CPU mode but can be indicative of how well page reclaim
identifies suitable pages.
stress-highalloc
Tries to allocate huge pages under heavy load.
kernbench, iozone and sysbench did not report any performance regression
on any machine. sysbench did pressure the system lightly and there was
reclaim activity but there were no difference of major interest between
the kernels.
X86-64 micro-mapped-file-stream
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3 waitwriteback-v2r4
pgalloc_dma 1639.00 ( 0.00%) 667.00 (-145.73%) 1167.00 ( -40.45%) 578.00 (-183.56%)
pgalloc_dma32 2842410.00 ( 0.00%) 2842626.00 ( 0.01%) 2843043.00 ( 0.02%) 2843014.00 ( 0.02%)
pgalloc_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgsteal_dma 729.00 ( 0.00%) 85.00 (-757.65%) 609.00 ( -19.70%) 125.00 (-483.20%)
pgsteal_dma32 2338721.00 ( 0.00%) 2447354.00 ( 4.44%) 2429536.00 ( 3.74%) 2436772.00 ( 4.02%)
pgsteal_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_kswapd_dma 1469.00 ( 0.00%) 532.00 (-176.13%) 1078.00 ( -36.27%) 220.00 (-567.73%)
pgscan_kswapd_dma32 4597713.00 ( 0.00%) 4503597.00 ( -2.09%) 4295673.00 ( -7.03%) 3891686.00 ( -18.14%)
pgscan_kswapd_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_direct_dma 71.00 ( 0.00%) 134.00 ( 47.01%) 243.00 ( 70.78%) 352.00 ( 79.83%)
pgscan_direct_dma32 305820.00 ( 0.00%) 280204.00 ( -9.14%) 600518.00 ( 49.07%) 957485.00 ( 68.06%)
pgscan_direct_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pageoutrun 16296.00 ( 0.00%) 21254.00 ( 23.33%) 18447.00 ( 11.66%) 20067.00 ( 18.79%)
allocstall 443.00 ( 0.00%) 273.00 ( -62.27%) 513.00 ( 13.65%) 1568.00 ( 71.75%)
These are based on the raw figures taken from /proc/vmstat. It's a rough
measure of reclaim activity. Note that allocstall counts are higher
because we are entering direct reclaim more often as a result of not
sleeping in congestion. In itself, it's not necessarily a bad thing.
It's easier to get a view of what happened from the vmscan tracepoint
report.
FTrace Reclaim Statistics: vmscan
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3 waitwriteback-v2r4
Direct reclaims 443 273 513 1568
Direct reclaim pages scanned 305968 280402 600825 957933
Direct reclaim pages reclaimed 43503 19005 30327 117191
Direct reclaim write file async I/O 0 0 0 0
Direct reclaim write anon async I/O 0 3 4 12
Direct reclaim write file sync I/O 0 0 0 0
Direct reclaim write anon sync I/O 0 0 0 0
Wake kswapd requests 187649 132338 191695 267701
Kswapd wakeups 3 1 4 1
Kswapd pages scanned 4599269 4454162 4296815 3891906
Kswapd pages reclaimed 2295947 2428434 2399818 2319706
Kswapd reclaim write file async I/O 1 0 1 1
Kswapd reclaim write anon async I/O 59 187 41 222
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 4.34 2.52 6.63 2.96
Time kswapd awake (seconds) 11.15 10.25 11.01 10.19
Total pages scanned 4905237 4734564 4897640 4849839
Total pages reclaimed 2339450 2447439 2430145 2436897
%age total pages scanned/reclaimed 47.69% 51.69% 49.62% 50.25%
%age total pages scanned/written 0.00% 0.00% 0.00% 0.00%
%age file pages scanned/written 0.00% 0.00% 0.00% 0.00%
Percentage Time Spent Direct Reclaim 29.23% 19.02% 38.48% 20.25%
Percentage Time kswapd Awake 78.58% 78.85% 76.83% 79.86%
What is interesting here for nocongest in particular is that while direct
reclaim scans more pages, the overall number of pages scanned remains the
same and the ratio of pages scanned to pages reclaimed is more or less the
same. In other words, while we are sleeping less, reclaim is not doing
more work and as direct reclaim and kswapd is awake for less time, it
would appear to be doing less work.
FTrace Reclaim Statistics: congestion_wait
Direct number congest waited 87 196 64 0
Direct time congest waited 4604ms 4732ms 5420ms 0ms
Direct full congest waited 72 145 53 0
Direct number conditional waited 0 0 324 1315
Direct time conditional waited 0ms 0ms 0ms 0ms
Direct full conditional waited 0 0 0 0
KSwapd number congest waited 20 10 15 7
KSwapd time congest waited 1264ms 536ms 884ms 284ms
KSwapd full congest waited 10 4 6 2
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
The vanilla kernel spent 8 seconds asleep in direct reclaim and no time at
all asleep with the patches.
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 10.51 10.73 10.6 11.66
Total Elapsed Time (seconds) 14.19 13.00 14.33 12.76
Overall, the tests completed faster. It is interesting to note that backing off further
when a zone is congested and not just a BDI was more efficient overall.
PPC64 micro-mapped-file-stream
pgalloc_dma 3024660.00 ( 0.00%) 3027185.00 ( 0.08%) 3025845.00 ( 0.04%) 3026281.00 ( 0.05%)
pgalloc_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgsteal_dma 2508073.00 ( 0.00%) 2565351.00 ( 2.23%) 2463577.00 ( -1.81%) 2532263.00 ( 0.96%)
pgsteal_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_kswapd_dma 4601307.00 ( 0.00%) 4128076.00 ( -11.46%) 3912317.00 ( -17.61%) 3377165.00 ( -36.25%)
pgscan_kswapd_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_direct_dma 629825.00 ( 0.00%) 971622.00 ( 35.18%) 1063938.00 ( 40.80%) 1711935.00 ( 63.21%)
pgscan_direct_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pageoutrun 27776.00 ( 0.00%) 20458.00 ( -35.77%) 18763.00 ( -48.04%) 18157.00 ( -52.98%)
allocstall 977.00 ( 0.00%) 2751.00 ( 64.49%) 2098.00 ( 53.43%) 5136.00 ( 80.98%)
Similar trends to x86-64. allocstalls are up but it's not necessarily bad.
FTrace Reclaim Statistics: vmscan
Direct reclaims 977 2709 2098 5136
Direct reclaim pages scanned 629825 963814 1063938 1711935
Direct reclaim pages reclaimed 75550 242538 150904 387647
Direct reclaim write file async I/O 0 0 0 2
Direct reclaim write anon async I/O 0 10 0 4
Direct reclaim write file sync I/O 0 0 0 0
Direct reclaim write anon sync I/O 0 0 0 0
Wake kswapd requests 392119 1201712 571935 571921
Kswapd wakeups 3 2 3 3
Kswapd pages scanned 4601307 4128076 3912317 3377165
Kswapd pages reclaimed 2432523 2318797 2312673 2144616
Kswapd reclaim write file async I/O 20 1 1 1
Kswapd reclaim write anon async I/O 57 132 11 121
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 6.19 7.30 13.04 10.88
Time kswapd awake (seconds) 21.73 26.51 25.55 23.90
Total pages scanned 5231132 5091890 4976255 5089100
Total pages reclaimed 2508073 2561335 2463577 2532263
%age total pages scanned/reclaimed 47.95% 50.30% 49.51% 49.76%
%age total pages scanned/written 0.00% 0.00% 0.00% 0.00%
%age file pages scanned/written 0.00% 0.00% 0.00% 0.00%
Percentage Time Spent Direct Reclaim 18.89% 20.65% 32.65% 27.65%
Percentage Time kswapd Awake 72.39% 80.68% 78.21% 77.40%
Again, a similar trend that the congestion_wait changes mean that direct
reclaim scans more pages but the overall number of pages scanned while
slightly reduced, are very similar. The ratio of scanning/reclaimed
remains roughly similar. The downside is that kswapd and direct reclaim
was awake longer and for a larger percentage of the overall workload.
It's possible there were big differences in the amount of time spent
reclaiming slab pages between the different kernels which is plausible
considering that the micro tests runs after fsmark and sysbench.
Trace Reclaim Statistics: congestion_wait
Direct number congest waited 845 1312 104 0
Direct time congest waited 19416ms 26560ms 7544ms 0ms
Direct full congest waited 745 1105 72 0
Direct number conditional waited 0 0 1322 2935
Direct time conditional waited 0ms 0ms 12ms 312ms
Direct full conditional waited 0 0 0 3
KSwapd number congest waited 39 102 75 63
KSwapd time congest waited 2484ms 6760ms 5756ms 3716ms
KSwapd full congest waited 20 48 46 25
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
The vanilla kernel spent 20 seconds asleep in direct reclaim and only
312ms asleep with the patches. The time kswapd spent congest waited was
also reduced by a large factor.
MMTests Statistics: duration
ser/Sys Time Running Test (seconds) 26.58 28.05 26.9 28.47
Total Elapsed Time (seconds) 30.02 32.86 32.67 30.88
With all patches applies, the completion times are very similar.
X86-64 STRESS-HIGHALLOC
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Pass 1 82.00 ( 0.00%) 84.00 ( 2.00%) 85.00 ( 3.00%) 85.00 ( 3.00%)
Pass 2 90.00 ( 0.00%) 87.00 (-3.00%) 88.00 (-2.00%) 89.00 (-1.00%)
At Rest 92.00 ( 0.00%) 90.00 (-2.00%) 90.00 (-2.00%) 91.00 (-1.00%)
Success figures across the board are broadly similar.
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Direct reclaims 1045 944 886 887
Direct reclaim pages scanned 135091 119604 109382 101019
Direct reclaim pages reclaimed 88599 47535 47863 46671
Direct reclaim write file async I/O 494 283 465 280
Direct reclaim write anon async I/O 29357 13710 16656 13462
Direct reclaim write file sync I/O 154 2 2 3
Direct reclaim write anon sync I/O 14594 571 509 561
Wake kswapd requests 7491 933 872 892
Kswapd wakeups 814 778 731 780
Kswapd pages scanned 7290822 15341158 11916436 13703442
Kswapd pages reclaimed 3587336 3142496 3094392 3187151
Kswapd reclaim write file async I/O 91975 32317 28022 29628
Kswapd reclaim write anon async I/O 1992022 789307 829745 849769
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 4588.93 2467.16 2495.41 2547.07
Time kswapd awake (seconds) 2497.66 1020.16 1098.06 1176.82
Total pages scanned 7425913 15460762 12025818 13804461
Total pages reclaimed 3675935 3190031 3142255 3233822
%age total pages scanned/reclaimed 49.50% 20.63% 26.13% 23.43%
%age total pages scanned/written 28.66% 5.41% 7.28% 6.47%
%age file pages scanned/written 1.25% 0.21% 0.24% 0.22%
Percentage Time Spent Direct Reclaim 57.33% 42.15% 42.41% 42.99%
Percentage Time kswapd Awake 43.56% 27.87% 29.76% 31.25%
Scanned/reclaimed ratios again look good with big improvements in
efficiency. The Scanned/written ratios also look much improved. With a
better scanned/written ration, there is an expectation that IO would be
more efficient and indeed, the time spent in direct reclaim is much
reduced by the full series and kswapd spends a little less time awake.
Overall, indications here are that allocations were happening much faster
and this can be seen with a graph of the latency figures as the
allocations were taking place
http://www.csn.ul.ie/~mel/postings/vmscanreduce-20101509/highalloc-interlatency-hydra-mean.ps
FTrace Reclaim Statistics: congestion_wait
Direct number congest waited 1333 204 169 4
Direct time congest waited 78896ms 8288ms 7260ms 200ms
Direct full congest waited 756 92 69 2
Direct number conditional waited 0 0 26 186
Direct time conditional waited 0ms 0ms 0ms 2504ms
Direct full conditional waited 0 0 0 25
KSwapd number congest waited 4 395 227 282
KSwapd time congest waited 384ms 25136ms 10508ms 18380ms
KSwapd full congest waited 3 232 98 176
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
KSwapd full conditional waited 318 0 312 9
Overall, the time spent speeping is reduced. kswapd is still hitting
congestion_wait() but that is because there are callers remaining where it
wasn't clear in advance if they should be changed to wait_iff_congested()
or not. Overall the sleep imes are reduced though - from 79ish seconds to
about 19.
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 3415.43 3386.65 3388.39 3377.5
Total Elapsed Time (seconds) 5733.48 3660.33 3689.41 3765.39
With the full series, the time to complete the tests are reduced by 30%
PPC64 STRESS-HIGHALLOC
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Pass 1 17.00 ( 0.00%) 34.00 (17.00%) 38.00 (21.00%) 43.00 (26.00%)
Pass 2 25.00 ( 0.00%) 37.00 (12.00%) 42.00 (17.00%) 46.00 (21.00%)
At Rest 49.00 ( 0.00%) 43.00 (-6.00%) 45.00 (-4.00%) 51.00 ( 2.00%)
Success rates there are *way* up particularly considering that the 16MB
huge pages on PPC64 mean that it's always much harder to allocate them.
FTrace Reclaim Statistics: vmscan
stress-highalloc stress-highalloc stress-highalloc stress-highalloc
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Direct reclaims 499 505 564 509
Direct reclaim pages scanned 223478 41898 51818 45605
Direct reclaim pages reclaimed 137730 21148 27161 23455
Direct reclaim write file async I/O 399 136 162 136
Direct reclaim write anon async I/O 46977 2865 4686 3998
Direct reclaim write file sync I/O 29 0 1 3
Direct reclaim write anon sync I/O 31023 159 237 239
Wake kswapd requests 420 351 360 326
Kswapd wakeups 185 294 249 277
Kswapd pages scanned 15703488 16392500 17821724 17598737
Kswapd pages reclaimed 5808466 2908858 3139386 3145435
Kswapd reclaim write file async I/O 159938 18400 18717 13473
Kswapd reclaim write anon async I/O 3467554 228957 322799 234278
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 9665.35 1707.81 2374.32 1871.23
Time kswapd awake (seconds) 9401.21 1367.86 1951.75 1328.88
Total pages scanned 15926966 16434398 17873542 17644342
Total pages reclaimed 5946196 2930006 3166547 3168890
%age total pages scanned/reclaimed 37.33% 17.83% 17.72% 17.96%
%age total pages scanned/written 23.27% 1.52% 1.94% 1.43%
%age file pages scanned/written 1.01% 0.11% 0.11% 0.08%
Percentage Time Spent Direct Reclaim 44.55% 35.10% 41.42% 36.91%
Percentage Time kswapd Awake 86.71% 43.58% 52.67% 41.14%
While the scanning rates are slightly up, the scanned/reclaimed and
scanned/written figures are much improved. The time spent in direct
reclaim and with kswapd are massively reduced, mostly by the lowlumpy
patches.
FTrace Reclaim Statistics: congestion_wait
Direct number congest waited 725 303 126 3
Direct time congest waited 45524ms 9180ms 5936ms 300ms
Direct full congest waited 487 190 52 3
Direct number conditional waited 0 0 200 301
Direct time conditional waited 0ms 0ms 0ms 1904ms
Direct full conditional waited 0 0 0 19
KSwapd number congest waited 0 2 23 4
KSwapd time congest waited 0ms 200ms 420ms 404ms
KSwapd full congest waited 0 2 2 4
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
Not as dramatic a story here but the time spent asleep is reduced and we
can still see what wait_iff_congested is going to sleep when necessary.
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 12028.09 3157.17 3357.79 3199.16
Total Elapsed Time (seconds) 10842.07 3138.72 3705.54 3229.85
The time to complete this test goes way down. With the full series, we
are allocating over twice the number of huge pages in 30% of the time and
there is a corresponding impact on the allocation latency graph available
at.
http://www.csn.ul.ie/~mel/postings/vmscanreduce-20101509/highalloc-interlatency-powyah-mean.ps
This patch:
Add a trace event for shrink_inactive_list() and updates the sample
postprocessing script appropriately. It can be used to determine how many
pages were reclaimed and for non-lumpy reclaim where exactly the pages
were reclaimed from.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:40 +07:00
|
|
|
|
2011-11-01 07:07:56 +07:00
|
|
|
/*
|
|
|
|
* If reclaim is isolating dirty pages under writeback, it implies
|
|
|
|
* that the long-lived page allocation rate is exceeding the page
|
|
|
|
* laundering rate. Either the global limits are not being effective
|
|
|
|
* at throttling processes due to the page distribution throughout
|
|
|
|
* zones or there is heavy usage of a slow backing device. The
|
|
|
|
* only option is to throttle from reclaim context which is not ideal
|
|
|
|
* as there is no guarantee the dirtying process is throttled in the
|
|
|
|
* same way balance_dirty_pages() manages.
|
|
|
|
*
|
|
|
|
* This scales the number of dirty pages that must be under writeback
|
|
|
|
* before throttling depending on priority. It is a simple backoff
|
|
|
|
* function that has the most effect in the range DEF_PRIORITY to
|
|
|
|
* DEF_PRIORITY-2 which is the priority reclaim is considered to be
|
|
|
|
* in trouble and reclaim is considered to be in trouble.
|
|
|
|
*
|
|
|
|
* DEF_PRIORITY 100% isolated pages must be PageWriteback to throttle
|
|
|
|
* DEF_PRIORITY-1 50% must be PageWriteback
|
|
|
|
* DEF_PRIORITY-2 25% must be PageWriteback, kswapd in trouble
|
|
|
|
* ...
|
|
|
|
* DEF_PRIORITY-6 For SWAP_CLUSTER_MAX isolated pages, throttle if any
|
|
|
|
* isolated page is PageWriteback
|
|
|
|
*/
|
|
|
|
if (nr_writeback && nr_writeback >= (nr_taken >> (DEF_PRIORITY-priority)))
|
|
|
|
wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10);
|
|
|
|
|
tracing, vmscan: add trace events for LRU list shrinking
There have been numerous reports of stalls that pointed at the problem
being somewhere in the VM. There are multiple roots to the problems which
means dealing with any of the root problems in isolation is tricky to
justify on their own and they would still need integration testing. This
patch series puts together two different patch sets which in combination
should tackle some of the root causes of latency problems being reported.
Patch 1 adds a tracepoint for shrink_inactive_list. For this series, the
most important results is being able to calculate the scanning/reclaim
ratio as a measure of the amount of work being done by page reclaim.
Patch 2 accounts for time spent in congestion_wait.
Patches 3-6 were originally developed by Kosaki Motohiro but reworked for
this series. It has been noted that lumpy reclaim is far too aggressive
and trashes the system somewhat. As SLUB uses high-order allocations, a
large cost incurred by lumpy reclaim will be noticeable. It was also
reported during transparent hugepage support testing that lumpy reclaim
was trashing the system and these patches should mitigate that problem
without disabling lumpy reclaim.
Patch 7 adds wait_iff_congested() and replaces some callers of
congestion_wait(). wait_iff_congested() only sleeps if there is a BDI
that is currently congested. Patch 8 notes that any BDI being congested
is not necessarily a problem because there could be multiple BDIs of
varying speeds and numberous zones. It attempts to track when a zone
being reclaimed contains many pages backed by a congested BDI and if so,
reclaimers wait on the congestion queue.
I ran a number of tests with monitoring on X86, X86-64 and PPC64. Each
machine had 3G of RAM and the CPUs were
X86: Intel P4 2-core
X86-64: AMD Phenom 4-core
PPC64: PPC970MP
Each used a single disk and the onboard IO controller. Dirty ratio was
left at 20. I'm just going to report for X86-64 and PPC64 in a vague
attempt to keep this report short. Four kernels were tested each based on
v2.6.36-rc4
traceonly-v2r2: Patches 1 and 2 to instrument vmscan reclaims and congestion_wait
lowlumpy-v2r3: Patches 1-6 to test if lumpy reclaim is better
waitcongest-v2r3: Patches 1-7 to only wait on congestion
waitwriteback-v2r4: Patches 1-8 to detect when a zone is congested
nocongest-v1r5: Patches 1-3 for testing wait_iff_congestion
nodirect-v1r5: Patches 1-10 to disable filesystem writeback for better IO
The tests run were as follows
kernbench
compile-based benchmark. Smoke test performance
sysbench
OLTP read-only benchmark. Will be re-run in the future as read-write
micro-mapped-file-stream
This is a micro-benchmark from Johannes Weiner that accesses a
large sparse-file through mmap(). It was configured to run in only
single-CPU mode but can be indicative of how well page reclaim
identifies suitable pages.
stress-highalloc
Tries to allocate huge pages under heavy load.
kernbench, iozone and sysbench did not report any performance regression
on any machine. sysbench did pressure the system lightly and there was
reclaim activity but there were no difference of major interest between
the kernels.
X86-64 micro-mapped-file-stream
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3 waitwriteback-v2r4
pgalloc_dma 1639.00 ( 0.00%) 667.00 (-145.73%) 1167.00 ( -40.45%) 578.00 (-183.56%)
pgalloc_dma32 2842410.00 ( 0.00%) 2842626.00 ( 0.01%) 2843043.00 ( 0.02%) 2843014.00 ( 0.02%)
pgalloc_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgsteal_dma 729.00 ( 0.00%) 85.00 (-757.65%) 609.00 ( -19.70%) 125.00 (-483.20%)
pgsteal_dma32 2338721.00 ( 0.00%) 2447354.00 ( 4.44%) 2429536.00 ( 3.74%) 2436772.00 ( 4.02%)
pgsteal_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_kswapd_dma 1469.00 ( 0.00%) 532.00 (-176.13%) 1078.00 ( -36.27%) 220.00 (-567.73%)
pgscan_kswapd_dma32 4597713.00 ( 0.00%) 4503597.00 ( -2.09%) 4295673.00 ( -7.03%) 3891686.00 ( -18.14%)
pgscan_kswapd_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_direct_dma 71.00 ( 0.00%) 134.00 ( 47.01%) 243.00 ( 70.78%) 352.00 ( 79.83%)
pgscan_direct_dma32 305820.00 ( 0.00%) 280204.00 ( -9.14%) 600518.00 ( 49.07%) 957485.00 ( 68.06%)
pgscan_direct_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pageoutrun 16296.00 ( 0.00%) 21254.00 ( 23.33%) 18447.00 ( 11.66%) 20067.00 ( 18.79%)
allocstall 443.00 ( 0.00%) 273.00 ( -62.27%) 513.00 ( 13.65%) 1568.00 ( 71.75%)
These are based on the raw figures taken from /proc/vmstat. It's a rough
measure of reclaim activity. Note that allocstall counts are higher
because we are entering direct reclaim more often as a result of not
sleeping in congestion. In itself, it's not necessarily a bad thing.
It's easier to get a view of what happened from the vmscan tracepoint
report.
FTrace Reclaim Statistics: vmscan
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3 waitwriteback-v2r4
Direct reclaims 443 273 513 1568
Direct reclaim pages scanned 305968 280402 600825 957933
Direct reclaim pages reclaimed 43503 19005 30327 117191
Direct reclaim write file async I/O 0 0 0 0
Direct reclaim write anon async I/O 0 3 4 12
Direct reclaim write file sync I/O 0 0 0 0
Direct reclaim write anon sync I/O 0 0 0 0
Wake kswapd requests 187649 132338 191695 267701
Kswapd wakeups 3 1 4 1
Kswapd pages scanned 4599269 4454162 4296815 3891906
Kswapd pages reclaimed 2295947 2428434 2399818 2319706
Kswapd reclaim write file async I/O 1 0 1 1
Kswapd reclaim write anon async I/O 59 187 41 222
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 4.34 2.52 6.63 2.96
Time kswapd awake (seconds) 11.15 10.25 11.01 10.19
Total pages scanned 4905237 4734564 4897640 4849839
Total pages reclaimed 2339450 2447439 2430145 2436897
%age total pages scanned/reclaimed 47.69% 51.69% 49.62% 50.25%
%age total pages scanned/written 0.00% 0.00% 0.00% 0.00%
%age file pages scanned/written 0.00% 0.00% 0.00% 0.00%
Percentage Time Spent Direct Reclaim 29.23% 19.02% 38.48% 20.25%
Percentage Time kswapd Awake 78.58% 78.85% 76.83% 79.86%
What is interesting here for nocongest in particular is that while direct
reclaim scans more pages, the overall number of pages scanned remains the
same and the ratio of pages scanned to pages reclaimed is more or less the
same. In other words, while we are sleeping less, reclaim is not doing
more work and as direct reclaim and kswapd is awake for less time, it
would appear to be doing less work.
FTrace Reclaim Statistics: congestion_wait
Direct number congest waited 87 196 64 0
Direct time congest waited 4604ms 4732ms 5420ms 0ms
Direct full congest waited 72 145 53 0
Direct number conditional waited 0 0 324 1315
Direct time conditional waited 0ms 0ms 0ms 0ms
Direct full conditional waited 0 0 0 0
KSwapd number congest waited 20 10 15 7
KSwapd time congest waited 1264ms 536ms 884ms 284ms
KSwapd full congest waited 10 4 6 2
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
The vanilla kernel spent 8 seconds asleep in direct reclaim and no time at
all asleep with the patches.
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 10.51 10.73 10.6 11.66
Total Elapsed Time (seconds) 14.19 13.00 14.33 12.76
Overall, the tests completed faster. It is interesting to note that backing off further
when a zone is congested and not just a BDI was more efficient overall.
PPC64 micro-mapped-file-stream
pgalloc_dma 3024660.00 ( 0.00%) 3027185.00 ( 0.08%) 3025845.00 ( 0.04%) 3026281.00 ( 0.05%)
pgalloc_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgsteal_dma 2508073.00 ( 0.00%) 2565351.00 ( 2.23%) 2463577.00 ( -1.81%) 2532263.00 ( 0.96%)
pgsteal_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_kswapd_dma 4601307.00 ( 0.00%) 4128076.00 ( -11.46%) 3912317.00 ( -17.61%) 3377165.00 ( -36.25%)
pgscan_kswapd_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pgscan_direct_dma 629825.00 ( 0.00%) 971622.00 ( 35.18%) 1063938.00 ( 40.80%) 1711935.00 ( 63.21%)
pgscan_direct_normal 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%)
pageoutrun 27776.00 ( 0.00%) 20458.00 ( -35.77%) 18763.00 ( -48.04%) 18157.00 ( -52.98%)
allocstall 977.00 ( 0.00%) 2751.00 ( 64.49%) 2098.00 ( 53.43%) 5136.00 ( 80.98%)
Similar trends to x86-64. allocstalls are up but it's not necessarily bad.
FTrace Reclaim Statistics: vmscan
Direct reclaims 977 2709 2098 5136
Direct reclaim pages scanned 629825 963814 1063938 1711935
Direct reclaim pages reclaimed 75550 242538 150904 387647
Direct reclaim write file async I/O 0 0 0 2
Direct reclaim write anon async I/O 0 10 0 4
Direct reclaim write file sync I/O 0 0 0 0
Direct reclaim write anon sync I/O 0 0 0 0
Wake kswapd requests 392119 1201712 571935 571921
Kswapd wakeups 3 2 3 3
Kswapd pages scanned 4601307 4128076 3912317 3377165
Kswapd pages reclaimed 2432523 2318797 2312673 2144616
Kswapd reclaim write file async I/O 20 1 1 1
Kswapd reclaim write anon async I/O 57 132 11 121
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 6.19 7.30 13.04 10.88
Time kswapd awake (seconds) 21.73 26.51 25.55 23.90
Total pages scanned 5231132 5091890 4976255 5089100
Total pages reclaimed 2508073 2561335 2463577 2532263
%age total pages scanned/reclaimed 47.95% 50.30% 49.51% 49.76%
%age total pages scanned/written 0.00% 0.00% 0.00% 0.00%
%age file pages scanned/written 0.00% 0.00% 0.00% 0.00%
Percentage Time Spent Direct Reclaim 18.89% 20.65% 32.65% 27.65%
Percentage Time kswapd Awake 72.39% 80.68% 78.21% 77.40%
Again, a similar trend that the congestion_wait changes mean that direct
reclaim scans more pages but the overall number of pages scanned while
slightly reduced, are very similar. The ratio of scanning/reclaimed
remains roughly similar. The downside is that kswapd and direct reclaim
was awake longer and for a larger percentage of the overall workload.
It's possible there were big differences in the amount of time spent
reclaiming slab pages between the different kernels which is plausible
considering that the micro tests runs after fsmark and sysbench.
Trace Reclaim Statistics: congestion_wait
Direct number congest waited 845 1312 104 0
Direct time congest waited 19416ms 26560ms 7544ms 0ms
Direct full congest waited 745 1105 72 0
Direct number conditional waited 0 0 1322 2935
Direct time conditional waited 0ms 0ms 12ms 312ms
Direct full conditional waited 0 0 0 3
KSwapd number congest waited 39 102 75 63
KSwapd time congest waited 2484ms 6760ms 5756ms 3716ms
KSwapd full congest waited 20 48 46 25
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
The vanilla kernel spent 20 seconds asleep in direct reclaim and only
312ms asleep with the patches. The time kswapd spent congest waited was
also reduced by a large factor.
MMTests Statistics: duration
ser/Sys Time Running Test (seconds) 26.58 28.05 26.9 28.47
Total Elapsed Time (seconds) 30.02 32.86 32.67 30.88
With all patches applies, the completion times are very similar.
X86-64 STRESS-HIGHALLOC
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Pass 1 82.00 ( 0.00%) 84.00 ( 2.00%) 85.00 ( 3.00%) 85.00 ( 3.00%)
Pass 2 90.00 ( 0.00%) 87.00 (-3.00%) 88.00 (-2.00%) 89.00 (-1.00%)
At Rest 92.00 ( 0.00%) 90.00 (-2.00%) 90.00 (-2.00%) 91.00 (-1.00%)
Success figures across the board are broadly similar.
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Direct reclaims 1045 944 886 887
Direct reclaim pages scanned 135091 119604 109382 101019
Direct reclaim pages reclaimed 88599 47535 47863 46671
Direct reclaim write file async I/O 494 283 465 280
Direct reclaim write anon async I/O 29357 13710 16656 13462
Direct reclaim write file sync I/O 154 2 2 3
Direct reclaim write anon sync I/O 14594 571 509 561
Wake kswapd requests 7491 933 872 892
Kswapd wakeups 814 778 731 780
Kswapd pages scanned 7290822 15341158 11916436 13703442
Kswapd pages reclaimed 3587336 3142496 3094392 3187151
Kswapd reclaim write file async I/O 91975 32317 28022 29628
Kswapd reclaim write anon async I/O 1992022 789307 829745 849769
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 4588.93 2467.16 2495.41 2547.07
Time kswapd awake (seconds) 2497.66 1020.16 1098.06 1176.82
Total pages scanned 7425913 15460762 12025818 13804461
Total pages reclaimed 3675935 3190031 3142255 3233822
%age total pages scanned/reclaimed 49.50% 20.63% 26.13% 23.43%
%age total pages scanned/written 28.66% 5.41% 7.28% 6.47%
%age file pages scanned/written 1.25% 0.21% 0.24% 0.22%
Percentage Time Spent Direct Reclaim 57.33% 42.15% 42.41% 42.99%
Percentage Time kswapd Awake 43.56% 27.87% 29.76% 31.25%
Scanned/reclaimed ratios again look good with big improvements in
efficiency. The Scanned/written ratios also look much improved. With a
better scanned/written ration, there is an expectation that IO would be
more efficient and indeed, the time spent in direct reclaim is much
reduced by the full series and kswapd spends a little less time awake.
Overall, indications here are that allocations were happening much faster
and this can be seen with a graph of the latency figures as the
allocations were taking place
http://www.csn.ul.ie/~mel/postings/vmscanreduce-20101509/highalloc-interlatency-hydra-mean.ps
FTrace Reclaim Statistics: congestion_wait
Direct number congest waited 1333 204 169 4
Direct time congest waited 78896ms 8288ms 7260ms 200ms
Direct full congest waited 756 92 69 2
Direct number conditional waited 0 0 26 186
Direct time conditional waited 0ms 0ms 0ms 2504ms
Direct full conditional waited 0 0 0 25
KSwapd number congest waited 4 395 227 282
KSwapd time congest waited 384ms 25136ms 10508ms 18380ms
KSwapd full congest waited 3 232 98 176
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
KSwapd full conditional waited 318 0 312 9
Overall, the time spent speeping is reduced. kswapd is still hitting
congestion_wait() but that is because there are callers remaining where it
wasn't clear in advance if they should be changed to wait_iff_congested()
or not. Overall the sleep imes are reduced though - from 79ish seconds to
about 19.
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 3415.43 3386.65 3388.39 3377.5
Total Elapsed Time (seconds) 5733.48 3660.33 3689.41 3765.39
With the full series, the time to complete the tests are reduced by 30%
PPC64 STRESS-HIGHALLOC
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Pass 1 17.00 ( 0.00%) 34.00 (17.00%) 38.00 (21.00%) 43.00 (26.00%)
Pass 2 25.00 ( 0.00%) 37.00 (12.00%) 42.00 (17.00%) 46.00 (21.00%)
At Rest 49.00 ( 0.00%) 43.00 (-6.00%) 45.00 (-4.00%) 51.00 ( 2.00%)
Success rates there are *way* up particularly considering that the 16MB
huge pages on PPC64 mean that it's always much harder to allocate them.
FTrace Reclaim Statistics: vmscan
stress-highalloc stress-highalloc stress-highalloc stress-highalloc
traceonly-v2r2 lowlumpy-v2r3 waitcongest-v2r3waitwriteback-v2r4
Direct reclaims 499 505 564 509
Direct reclaim pages scanned 223478 41898 51818 45605
Direct reclaim pages reclaimed 137730 21148 27161 23455
Direct reclaim write file async I/O 399 136 162 136
Direct reclaim write anon async I/O 46977 2865 4686 3998
Direct reclaim write file sync I/O 29 0 1 3
Direct reclaim write anon sync I/O 31023 159 237 239
Wake kswapd requests 420 351 360 326
Kswapd wakeups 185 294 249 277
Kswapd pages scanned 15703488 16392500 17821724 17598737
Kswapd pages reclaimed 5808466 2908858 3139386 3145435
Kswapd reclaim write file async I/O 159938 18400 18717 13473
Kswapd reclaim write anon async I/O 3467554 228957 322799 234278
Kswapd reclaim write file sync I/O 0 0 0 0
Kswapd reclaim write anon sync I/O 0 0 0 0
Time stalled direct reclaim (seconds) 9665.35 1707.81 2374.32 1871.23
Time kswapd awake (seconds) 9401.21 1367.86 1951.75 1328.88
Total pages scanned 15926966 16434398 17873542 17644342
Total pages reclaimed 5946196 2930006 3166547 3168890
%age total pages scanned/reclaimed 37.33% 17.83% 17.72% 17.96%
%age total pages scanned/written 23.27% 1.52% 1.94% 1.43%
%age file pages scanned/written 1.01% 0.11% 0.11% 0.08%
Percentage Time Spent Direct Reclaim 44.55% 35.10% 41.42% 36.91%
Percentage Time kswapd Awake 86.71% 43.58% 52.67% 41.14%
While the scanning rates are slightly up, the scanned/reclaimed and
scanned/written figures are much improved. The time spent in direct
reclaim and with kswapd are massively reduced, mostly by the lowlumpy
patches.
FTrace Reclaim Statistics: congestion_wait
Direct number congest waited 725 303 126 3
Direct time congest waited 45524ms 9180ms 5936ms 300ms
Direct full congest waited 487 190 52 3
Direct number conditional waited 0 0 200 301
Direct time conditional waited 0ms 0ms 0ms 1904ms
Direct full conditional waited 0 0 0 19
KSwapd number congest waited 0 2 23 4
KSwapd time congest waited 0ms 200ms 420ms 404ms
KSwapd full congest waited 0 2 2 4
KSwapd number conditional waited 0 0 0 0
KSwapd time conditional waited 0ms 0ms 0ms 0ms
KSwapd full conditional waited 0 0 0 0
Not as dramatic a story here but the time spent asleep is reduced and we
can still see what wait_iff_congested is going to sleep when necessary.
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 12028.09 3157.17 3357.79 3199.16
Total Elapsed Time (seconds) 10842.07 3138.72 3705.54 3229.85
The time to complete this test goes way down. With the full series, we
are allocating over twice the number of huge pages in 30% of the time and
there is a corresponding impact on the allocation latency graph available
at.
http://www.csn.ul.ie/~mel/postings/vmscanreduce-20101509/highalloc-interlatency-powyah-mean.ps
This patch:
Add a trace event for shrink_inactive_list() and updates the sample
postprocessing script appropriately. It can be used to determine how many
pages were reclaimed and for non-lumpy reclaim where exactly the pages
were reclaimed from.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-10-27 04:21:40 +07:00
|
|
|
trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id,
|
|
|
|
zone_idx(zone),
|
|
|
|
nr_scanned, nr_reclaimed,
|
|
|
|
priority,
|
2011-01-14 06:46:00 +07:00
|
|
|
trace_shrink_flags(file, sc->reclaim_mode));
|
2006-03-22 15:08:20 +07:00
|
|
|
return nr_reclaimed;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This moves pages from the active list to the inactive list.
|
|
|
|
*
|
|
|
|
* We move them the other way if the page is referenced by one or more
|
|
|
|
* processes, from rmap.
|
|
|
|
*
|
|
|
|
* If the pages are mostly unmapped, the processing is fast and it is
|
|
|
|
* appropriate to hold zone->lru_lock across the whole operation. But if
|
|
|
|
* the pages are mapped, the processing is slow (page_referenced()) so we
|
|
|
|
* should drop zone->lru_lock around each page. It's impossible to balance
|
|
|
|
* this, so instead we remove the pages from the LRU while processing them.
|
|
|
|
* It is safe to rely on PG_active against the non-LRU pages in here because
|
|
|
|
* nobody will play with that bit on a non-LRU page.
|
|
|
|
*
|
|
|
|
* The downside is that we have to touch page->_count against each page.
|
|
|
|
* But we had to alter page->flags anyway.
|
|
|
|
*/
|
2008-02-07 15:14:37 +07:00
|
|
|
|
2009-06-17 05:33:13 +07:00
|
|
|
static void move_active_pages_to_lru(struct zone *zone,
|
|
|
|
struct list_head *list,
|
2012-01-13 08:19:56 +07:00
|
|
|
struct list_head *pages_to_free,
|
2009-06-17 05:33:13 +07:00
|
|
|
enum lru_list lru)
|
|
|
|
{
|
|
|
|
unsigned long pgmoved = 0;
|
|
|
|
struct page *page;
|
|
|
|
|
|
|
|
while (!list_empty(list)) {
|
2012-01-13 08:18:15 +07:00
|
|
|
struct lruvec *lruvec;
|
|
|
|
|
2009-06-17 05:33:13 +07:00
|
|
|
page = lru_to_page(list);
|
|
|
|
|
|
|
|
VM_BUG_ON(PageLRU(page));
|
|
|
|
SetPageLRU(page);
|
|
|
|
|
2012-01-13 08:18:15 +07:00
|
|
|
lruvec = mem_cgroup_lru_add_list(zone, page, lru);
|
|
|
|
list_move(&page->lru, &lruvec->lists[lru]);
|
2011-01-14 06:47:13 +07:00
|
|
|
pgmoved += hpage_nr_pages(page);
|
2009-06-17 05:33:13 +07:00
|
|
|
|
2012-01-13 08:19:56 +07:00
|
|
|
if (put_page_testzero(page)) {
|
|
|
|
__ClearPageLRU(page);
|
|
|
|
__ClearPageActive(page);
|
|
|
|
del_page_from_lru_list(zone, page, lru);
|
|
|
|
|
|
|
|
if (unlikely(PageCompound(page))) {
|
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
|
|
|
(*get_compound_page_dtor(page))(page);
|
|
|
|
spin_lock_irq(&zone->lru_lock);
|
|
|
|
} else
|
|
|
|
list_add(&page->lru, pages_to_free);
|
2009-06-17 05:33:13 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
__mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved);
|
|
|
|
if (!is_active_lru(lru))
|
|
|
|
__count_vm_events(PGDEACTIVATE, pgmoved);
|
|
|
|
}
|
2008-02-07 15:14:37 +07:00
|
|
|
|
2012-01-13 08:20:06 +07:00
|
|
|
static void shrink_active_list(unsigned long nr_to_scan,
|
2012-01-13 08:17:52 +07:00
|
|
|
struct mem_cgroup_zone *mz,
|
|
|
|
struct scan_control *sc,
|
|
|
|
int priority, int file)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2009-09-22 07:01:35 +07:00
|
|
|
unsigned long nr_taken;
|
2012-01-13 08:20:06 +07:00
|
|
|
unsigned long nr_scanned;
|
2009-06-17 05:33:05 +07:00
|
|
|
unsigned long vm_flags;
|
2005-04-17 05:20:36 +07:00
|
|
|
LIST_HEAD(l_hold); /* The pages which were snipped off */
|
vmscan: make mapped executable pages the first class citizen
Protect referenced PROT_EXEC mapped pages from being deactivated.
PROT_EXEC(or its internal presentation VM_EXEC) pages normally belong to some
currently running executables and their linked libraries, they shall really be
cached aggressively to provide good user experiences.
Thanks to Johannes Weiner for the advice to reuse the VMA walk in
page_referenced() to get the PROT_EXEC bit.
[more details]
( The consequences of this patch will have to be discussed together with
Rik van Riel's recent patch "vmscan: evict use-once pages first". )
( Some of the good points and insights are taken into this changelog.
Thanks to all the involved people for the great LKML discussions. )
the problem
===========
For a typical desktop, the most precious working set is composed of
*actively accessed*
(1) memory mapped executables
(2) and their anonymous pages
(3) and other files
(4) and the dcache/icache/.. slabs
while the least important data are
(5) infrequently used or use-once files
For a typical desktop, one major problem is busty and large amount of (5)
use-once files flushing out the working set.
Inside the working set, (4) dcache/icache have already been too sticky ;-)
So we only have to care (2) anonymous and (1)(3) file pages.
anonymous pages
===============
Anonymous pages are effectively immune to the streaming IO attack, because we
now have separate file/anon LRU lists. When the use-once files crowd into the
file LRU, the list's "quality" is significantly lowered. Therefore the scan
balance policy in get_scan_ratio() will choose to scan the (low quality) file
LRU much more frequently than the anon LRU.
file pages
==========
Rik proposed to *not* scan the active file LRU when the inactive list grows
larger than active list. This guarantees that when there are use-once streaming
IO, and the working set is not too large(so that active_size < inactive_size),
the active file LRU will *not* be scanned at all. So the not-too-large working
set can be well protected.
But there are also situations where the file working set is a bit large so that
(active_size >= inactive_size), or the streaming IOs are not purely use-once.
In these cases, the active list will be scanned slowly. Because the current
shrink_active_list() policy is to deactivate active pages regardless of their
referenced bits. The deactivated pages become susceptible to the streaming IO
attack: the inactive list could be scanned fast (500MB / 50MBps = 10s) so that
the deactivated pages don't have enough time to get re-referenced. Because a
user tend to switch between windows in intervals from seconds to minutes.
This patch holds mapped executable pages in the active list as long as they
are referenced during each full scan of the active list. Because the active
list is normally scanned much slower, they get longer grace time (eg. 100s)
for further references, which better matches the pace of user operations.
Therefore this patch greatly prolongs the in-cache time of executable code,
when there are moderate memory pressures.
before patch: guaranteed to be cached if reference intervals < I
after patch: guaranteed to be cached if reference intervals < I+A
(except when randomly reclaimed by the lumpy reclaim)
where
A = time to fully scan the active file LRU
I = time to fully scan the inactive file LRU
Note that normally A >> I.
side effects
============
This patch is safe in general, it restores the pre-2.6.28 mmap() behavior
but in a much smaller and well targeted scope.
One may worry about some one to abuse the PROT_EXEC heuristic. But as
Andrew Morton stated, there are other tricks to getting that sort of boost.
Another concern is the PROT_EXEC mapped pages growing large in rare cases,
and therefore hurting reclaim efficiency. But a sane application targeted for
large audience will never use PROT_EXEC for data mappings. If some home made
application tries to abuse that bit, it shall be aware of the consequences.
If it is abused to scale of 2/3 total memory, it gains nothing but overheads.
benchmarks
==========
1) memory tight desktop
1.1) brief summary
- clock time and major faults are reduced by 50%;
- pswpin numbers are reduced to ~1/3.
That means X desktop responsiveness is doubled under high memory/swap pressure.
1.2) test scenario
- nfsroot gnome desktop with 512M physical memory
- run some programs, and switch between the existing windows
after starting each new program.
1.3) progress timing (seconds)
before after programs
0.02 0.02 N xeyes
0.75 0.76 N firefox
2.02 1.88 N nautilus
3.36 3.17 N nautilus --browser
5.26 4.89 N gthumb
7.12 6.47 N gedit
9.22 8.16 N xpdf /usr/share/doc/shared-mime-info/shared-mime-info-spec.pdf
13.58 12.55 N xterm
15.87 14.57 N mlterm
18.63 17.06 N gnome-terminal
21.16 18.90 N urxvt
26.24 23.48 N gnome-system-monitor
28.72 26.52 N gnome-help
32.15 29.65 N gnome-dictionary
39.66 36.12 N /usr/games/sol
43.16 39.27 N /usr/games/gnometris
48.65 42.56 N /usr/games/gnect
53.31 47.03 N /usr/games/gtali
58.60 52.05 N /usr/games/iagno
65.77 55.42 N /usr/games/gnotravex
70.76 61.47 N /usr/games/mahjongg
76.15 67.11 N /usr/games/gnome-sudoku
86.32 75.15 N /usr/games/glines
92.21 79.70 N /usr/games/glchess
103.79 88.48 N /usr/games/gnomine
113.84 96.51 N /usr/games/gnotski
124.40 102.19 N /usr/games/gnibbles
137.41 114.93 N /usr/games/gnobots2
155.53 125.02 N /usr/games/blackjack
179.85 135.11 N /usr/games/same-gnome
224.49 154.50 N /usr/bin/gnome-window-properties
248.44 162.09 N /usr/bin/gnome-default-applications-properties
282.62 173.29 N /usr/bin/gnome-at-properties
323.72 188.21 N /usr/bin/gnome-typing-monitor
363.99 199.93 N /usr/bin/gnome-at-visual
394.21 206.95 N /usr/bin/gnome-sound-properties
435.14 224.49 N /usr/bin/gnome-at-mobility
463.05 234.11 N /usr/bin/gnome-keybinding-properties
503.75 248.59 N /usr/bin/gnome-about-me
554.00 276.27 N /usr/bin/gnome-display-properties
615.48 304.39 N /usr/bin/gnome-network-preferences
693.03 342.01 N /usr/bin/gnome-mouse-properties
759.90 388.58 N /usr/bin/gnome-appearance-properties
937.90 508.47 N /usr/bin/gnome-control-center
1109.75 587.57 N /usr/bin/gnome-keyboard-properties
1399.05 758.16 N : oocalc
1524.64 830.03 N : oodraw
1684.31 900.03 N : ooimpress
1874.04 993.91 N : oomath
2115.12 1081.89 N : ooweb
2369.02 1161.99 N : oowriter
Note that the last ": oo*" commands are actually commented out.
1.4) vmstat numbers (some relevant ones are marked with *)
before after
nr_free_pages 1293 3898
nr_inactive_anon 59956 53460
nr_active_anon 26815 30026
nr_inactive_file 2657 3218
nr_active_file 2019 2806
nr_unevictable 4 4
nr_mlock 4 4
nr_anon_pages 26706 27859
*nr_mapped 3542 4469
nr_file_pages 72232 67681
nr_dirty 1 0
nr_writeback 123 19
nr_slab_reclaimable 3375 3534
nr_slab_unreclaimable 11405 10665
nr_page_table_pages 8106 7864
nr_unstable 0 0
nr_bounce 0 0
*nr_vmscan_write 394776 230839
nr_writeback_temp 0 0
numa_hit 6843353 3318676
numa_miss 0 0
numa_foreign 0 0
numa_interleave 1719 1719
numa_local 6843353 3318676
numa_other 0 0
*pgpgin 5954683 2057175
*pgpgout 1578276 922744
*pswpin 1486615 512238
*pswpout 394568 230685
pgalloc_dma 277432 56602
pgalloc_dma32 6769477 3310348
pgalloc_normal 0 0
pgalloc_movable 0 0
pgfree 7048396 3371118
pgactivate 2036343 1471492
pgdeactivate 2189691 1612829
pgfault 3702176 3100702
*pgmajfault 452116 201343
pgrefill_dma 12185 7127
pgrefill_dma32 334384 653703
pgrefill_normal 0 0
pgrefill_movable 0 0
pgsteal_dma 74214 22179
pgsteal_dma32 3334164 1638029
pgsteal_normal 0 0
pgsteal_movable 0 0
pgscan_kswapd_dma 1081421 1216199
pgscan_kswapd_dma32 58979118 46002810
pgscan_kswapd_normal 0 0
pgscan_kswapd_movable 0 0
pgscan_direct_dma 2015438 1086109
pgscan_direct_dma32 55787823 36101597
pgscan_direct_normal 0 0
pgscan_direct_movable 0 0
pginodesteal 3461 7281
slabs_scanned 564864 527616
kswapd_steal 2889797 1448082
kswapd_inodesteal 14827 14835
pageoutrun 43459 21562
allocstall 9653 4032
pgrotated 384216 228631
1.5) free numbers at the end of the tests
before patch:
total used free shared buffers cached
Mem: 474 467 7 0 0 236
-/+ buffers/cache: 230 243
Swap: 1023 418 605
after patch:
total used free shared buffers cached
Mem: 474 457 16 0 0 236
-/+ buffers/cache: 221 253
Swap: 1023 404 619
2) memory flushing in a file server
2.1) brief summary
The number of major faults from 50 to 3 during 10% cache hot reads.
That means this patch successfully stops major faults when the active file
list is slowly scanned when there are partially cache hot streaming IO.
2.2) test scenario
Do 100000 pread(size=110 pages, offset=(i*100) pages), where 10% of the
pages will be activated:
for i in `seq 0 100 10000000`; do echo $i 110; done > pattern-hot-10
iotrace.rb --load pattern-hot-10 --play /b/sparse
vmmon nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
and monitor /proc/vmstat during the time. The test box has 2G memory.
I carried out tests on fresh booted console as well as X desktop, and
fetched the vmstat numbers on
(1) begin: shortly after the big read IO starts;
(2) end: just before the big read IO stops;
(3) restore: the big read IO stops and the zsh working set restored
(4) restore X: after IO, switch back and forth between the urxvt and firefox
windows to restore their working set.
2.3) console mode results
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.29 VM_EXEC protection ON:
begin: 2481 2237 8694 630 0 574299
end: 275 231976 233914 633 776271 20933042
restore: 370 232154 234524 691 777183 20958453
2.6.29 VM_EXEC protection ON (second run):
begin: 2434 2237 8493 629 0 574195
end: 284 231970 233536 632 771918 20896129
restore: 399 232218 234789 690 774526 20957909
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 2479 2344 9659 210 0 579643
end: 284 232010 234142 260 772776 20917184
restore: 379 232159 234371 301 774888 20967849
The above console numbers show that
- The startup pgmajfault of 2.6.30-rc4-mm is merely 1/3 that of 2.6.29.
I'd attribute that improvement to the mmap readahead improvements :-)
- The pgmajfault increment during the file copy is 633-630=3 vs 260-210=50.
That's a huge improvement - which means with the VM_EXEC protection logic,
active mmap pages is pretty safe even under partially cache hot streaming IO.
- when active:inactive file lru size reaches 1:1, their scan rates is 1:20.8
under 10% cache hot IO. (computed with formula Dpgdeactivate:Dpgfree)
That roughly means the active mmap pages get 20.8 more chances to get
re-referenced to stay in memory.
- The absolute nr_mapped drops considerably to 1/9 during the big IO, and the
dropped pages are mostly inactive ones. The patch has almost no impact in
this aspect, that means it won't unnecessarily increase memory pressure.
(In contrast, your 20% mmap protection ratio will keep them all, and
therefore eliminate the extra 41 major faults to restore working set
of zsh etc.)
The iotrace.rb read throughput is
151.194384MB/s 284.198252s 100001x 450560b --load pattern-hot-10 --play /b/sparse
which means the inactive list is rotated at the speed of 250MB/s,
so a full scan of which takes about 3.5 seconds, while a full scan
of active file list takes about 77 seconds.
2.4) X mode results
We can reach roughly the same conclusions for X desktop:
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.30-rc4-mm VM_EXEC protection ON:
begin: 9740 8920 64075 561 0 678360
end: 768 218254 220029 565 798953 21057006
restore: 857 218543 220987 606 799462 21075710
restore X: 2414 218560 225344 797 799462 21080795
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 9368 5035 26389 554 0 633391
end: 770 218449 221230 661 646472 17832500
restore: 1113 218466 220978 710 649881 17905235
restore X: 2687 218650 225484 947 802700 21083584
- the absolute nr_mapped drops considerably (to 1/13 of the original size)
during the streaming IO.
- the delta of pgmajfault is 3 vs 107 during IO, or 236 vs 393
during the whole process.
Cc: Elladan <elladan@eskimo.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:12 +07:00
|
|
|
LIST_HEAD(l_active);
|
2008-10-19 10:26:14 +07:00
|
|
|
LIST_HEAD(l_inactive);
|
2005-04-17 05:20:36 +07:00
|
|
|
struct page *page;
|
2012-01-13 08:17:52 +07:00
|
|
|
struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz);
|
2009-09-22 07:01:35 +07:00
|
|
|
unsigned long nr_rotated = 0;
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode_t isolate_mode = ISOLATE_ACTIVE;
|
2012-01-13 08:17:52 +07:00
|
|
|
struct zone *zone = mz->zone;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
lru_add_drain();
|
2011-11-01 07:06:55 +07:00
|
|
|
|
2012-03-22 06:34:17 +07:00
|
|
|
reset_reclaim_mode(sc);
|
|
|
|
|
2011-11-01 07:06:55 +07:00
|
|
|
if (!sc->may_unmap)
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode |= ISOLATE_UNMAPPED;
|
2011-11-01 07:06:55 +07:00
|
|
|
if (!sc->may_writepage)
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode |= ISOLATE_CLEAN;
|
2011-11-01 07:06:55 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
spin_lock_irq(&zone->lru_lock);
|
2012-01-13 08:18:15 +07:00
|
|
|
|
2012-03-22 06:33:51 +07:00
|
|
|
nr_taken = isolate_lru_pages(nr_to_scan, mz, &l_hold, &nr_scanned, sc,
|
2012-03-22 06:33:48 +07:00
|
|
|
isolate_mode, 1, file);
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc))
|
2012-01-13 08:20:06 +07:00
|
|
|
zone->pages_scanned += nr_scanned;
|
2012-01-13 08:17:50 +07:00
|
|
|
|
2009-09-22 07:02:56 +07:00
|
|
|
reclaim_stat->recent_scanned[file] += nr_taken;
|
2008-02-07 15:14:37 +07:00
|
|
|
|
2012-01-13 08:20:06 +07:00
|
|
|
__count_zone_vm_events(PGREFILL, zone, nr_scanned);
|
2008-10-19 10:26:32 +07:00
|
|
|
if (file)
|
2009-09-22 07:01:35 +07:00
|
|
|
__mod_zone_page_state(zone, NR_ACTIVE_FILE, -nr_taken);
|
2008-10-19 10:26:32 +07:00
|
|
|
else
|
2009-09-22 07:01:35 +07:00
|
|
|
__mod_zone_page_state(zone, NR_ACTIVE_ANON, -nr_taken);
|
2009-09-22 07:01:37 +07:00
|
|
|
__mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken);
|
2005-04-17 05:20:36 +07:00
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
|
|
|
|
|
|
|
while (!list_empty(&l_hold)) {
|
|
|
|
cond_resched();
|
|
|
|
page = lru_to_page(&l_hold);
|
|
|
|
list_del(&page->lru);
|
2008-10-19 10:26:35 +07:00
|
|
|
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
if (unlikely(!page_evictable(page, NULL))) {
|
|
|
|
putback_lru_page(page);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2012-03-22 06:34:00 +07:00
|
|
|
if (unlikely(buffer_heads_over_limit)) {
|
|
|
|
if (page_has_private(page) && trylock_page(page)) {
|
|
|
|
if (page_has_private(page))
|
|
|
|
try_to_release_page(page, 0);
|
|
|
|
unlock_page(page);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
if (page_referenced(page, 0, mz->mem_cgroup, &vm_flags)) {
|
2011-01-14 06:47:13 +07:00
|
|
|
nr_rotated += hpage_nr_pages(page);
|
vmscan: make mapped executable pages the first class citizen
Protect referenced PROT_EXEC mapped pages from being deactivated.
PROT_EXEC(or its internal presentation VM_EXEC) pages normally belong to some
currently running executables and their linked libraries, they shall really be
cached aggressively to provide good user experiences.
Thanks to Johannes Weiner for the advice to reuse the VMA walk in
page_referenced() to get the PROT_EXEC bit.
[more details]
( The consequences of this patch will have to be discussed together with
Rik van Riel's recent patch "vmscan: evict use-once pages first". )
( Some of the good points and insights are taken into this changelog.
Thanks to all the involved people for the great LKML discussions. )
the problem
===========
For a typical desktop, the most precious working set is composed of
*actively accessed*
(1) memory mapped executables
(2) and their anonymous pages
(3) and other files
(4) and the dcache/icache/.. slabs
while the least important data are
(5) infrequently used or use-once files
For a typical desktop, one major problem is busty and large amount of (5)
use-once files flushing out the working set.
Inside the working set, (4) dcache/icache have already been too sticky ;-)
So we only have to care (2) anonymous and (1)(3) file pages.
anonymous pages
===============
Anonymous pages are effectively immune to the streaming IO attack, because we
now have separate file/anon LRU lists. When the use-once files crowd into the
file LRU, the list's "quality" is significantly lowered. Therefore the scan
balance policy in get_scan_ratio() will choose to scan the (low quality) file
LRU much more frequently than the anon LRU.
file pages
==========
Rik proposed to *not* scan the active file LRU when the inactive list grows
larger than active list. This guarantees that when there are use-once streaming
IO, and the working set is not too large(so that active_size < inactive_size),
the active file LRU will *not* be scanned at all. So the not-too-large working
set can be well protected.
But there are also situations where the file working set is a bit large so that
(active_size >= inactive_size), or the streaming IOs are not purely use-once.
In these cases, the active list will be scanned slowly. Because the current
shrink_active_list() policy is to deactivate active pages regardless of their
referenced bits. The deactivated pages become susceptible to the streaming IO
attack: the inactive list could be scanned fast (500MB / 50MBps = 10s) so that
the deactivated pages don't have enough time to get re-referenced. Because a
user tend to switch between windows in intervals from seconds to minutes.
This patch holds mapped executable pages in the active list as long as they
are referenced during each full scan of the active list. Because the active
list is normally scanned much slower, they get longer grace time (eg. 100s)
for further references, which better matches the pace of user operations.
Therefore this patch greatly prolongs the in-cache time of executable code,
when there are moderate memory pressures.
before patch: guaranteed to be cached if reference intervals < I
after patch: guaranteed to be cached if reference intervals < I+A
(except when randomly reclaimed by the lumpy reclaim)
where
A = time to fully scan the active file LRU
I = time to fully scan the inactive file LRU
Note that normally A >> I.
side effects
============
This patch is safe in general, it restores the pre-2.6.28 mmap() behavior
but in a much smaller and well targeted scope.
One may worry about some one to abuse the PROT_EXEC heuristic. But as
Andrew Morton stated, there are other tricks to getting that sort of boost.
Another concern is the PROT_EXEC mapped pages growing large in rare cases,
and therefore hurting reclaim efficiency. But a sane application targeted for
large audience will never use PROT_EXEC for data mappings. If some home made
application tries to abuse that bit, it shall be aware of the consequences.
If it is abused to scale of 2/3 total memory, it gains nothing but overheads.
benchmarks
==========
1) memory tight desktop
1.1) brief summary
- clock time and major faults are reduced by 50%;
- pswpin numbers are reduced to ~1/3.
That means X desktop responsiveness is doubled under high memory/swap pressure.
1.2) test scenario
- nfsroot gnome desktop with 512M physical memory
- run some programs, and switch between the existing windows
after starting each new program.
1.3) progress timing (seconds)
before after programs
0.02 0.02 N xeyes
0.75 0.76 N firefox
2.02 1.88 N nautilus
3.36 3.17 N nautilus --browser
5.26 4.89 N gthumb
7.12 6.47 N gedit
9.22 8.16 N xpdf /usr/share/doc/shared-mime-info/shared-mime-info-spec.pdf
13.58 12.55 N xterm
15.87 14.57 N mlterm
18.63 17.06 N gnome-terminal
21.16 18.90 N urxvt
26.24 23.48 N gnome-system-monitor
28.72 26.52 N gnome-help
32.15 29.65 N gnome-dictionary
39.66 36.12 N /usr/games/sol
43.16 39.27 N /usr/games/gnometris
48.65 42.56 N /usr/games/gnect
53.31 47.03 N /usr/games/gtali
58.60 52.05 N /usr/games/iagno
65.77 55.42 N /usr/games/gnotravex
70.76 61.47 N /usr/games/mahjongg
76.15 67.11 N /usr/games/gnome-sudoku
86.32 75.15 N /usr/games/glines
92.21 79.70 N /usr/games/glchess
103.79 88.48 N /usr/games/gnomine
113.84 96.51 N /usr/games/gnotski
124.40 102.19 N /usr/games/gnibbles
137.41 114.93 N /usr/games/gnobots2
155.53 125.02 N /usr/games/blackjack
179.85 135.11 N /usr/games/same-gnome
224.49 154.50 N /usr/bin/gnome-window-properties
248.44 162.09 N /usr/bin/gnome-default-applications-properties
282.62 173.29 N /usr/bin/gnome-at-properties
323.72 188.21 N /usr/bin/gnome-typing-monitor
363.99 199.93 N /usr/bin/gnome-at-visual
394.21 206.95 N /usr/bin/gnome-sound-properties
435.14 224.49 N /usr/bin/gnome-at-mobility
463.05 234.11 N /usr/bin/gnome-keybinding-properties
503.75 248.59 N /usr/bin/gnome-about-me
554.00 276.27 N /usr/bin/gnome-display-properties
615.48 304.39 N /usr/bin/gnome-network-preferences
693.03 342.01 N /usr/bin/gnome-mouse-properties
759.90 388.58 N /usr/bin/gnome-appearance-properties
937.90 508.47 N /usr/bin/gnome-control-center
1109.75 587.57 N /usr/bin/gnome-keyboard-properties
1399.05 758.16 N : oocalc
1524.64 830.03 N : oodraw
1684.31 900.03 N : ooimpress
1874.04 993.91 N : oomath
2115.12 1081.89 N : ooweb
2369.02 1161.99 N : oowriter
Note that the last ": oo*" commands are actually commented out.
1.4) vmstat numbers (some relevant ones are marked with *)
before after
nr_free_pages 1293 3898
nr_inactive_anon 59956 53460
nr_active_anon 26815 30026
nr_inactive_file 2657 3218
nr_active_file 2019 2806
nr_unevictable 4 4
nr_mlock 4 4
nr_anon_pages 26706 27859
*nr_mapped 3542 4469
nr_file_pages 72232 67681
nr_dirty 1 0
nr_writeback 123 19
nr_slab_reclaimable 3375 3534
nr_slab_unreclaimable 11405 10665
nr_page_table_pages 8106 7864
nr_unstable 0 0
nr_bounce 0 0
*nr_vmscan_write 394776 230839
nr_writeback_temp 0 0
numa_hit 6843353 3318676
numa_miss 0 0
numa_foreign 0 0
numa_interleave 1719 1719
numa_local 6843353 3318676
numa_other 0 0
*pgpgin 5954683 2057175
*pgpgout 1578276 922744
*pswpin 1486615 512238
*pswpout 394568 230685
pgalloc_dma 277432 56602
pgalloc_dma32 6769477 3310348
pgalloc_normal 0 0
pgalloc_movable 0 0
pgfree 7048396 3371118
pgactivate 2036343 1471492
pgdeactivate 2189691 1612829
pgfault 3702176 3100702
*pgmajfault 452116 201343
pgrefill_dma 12185 7127
pgrefill_dma32 334384 653703
pgrefill_normal 0 0
pgrefill_movable 0 0
pgsteal_dma 74214 22179
pgsteal_dma32 3334164 1638029
pgsteal_normal 0 0
pgsteal_movable 0 0
pgscan_kswapd_dma 1081421 1216199
pgscan_kswapd_dma32 58979118 46002810
pgscan_kswapd_normal 0 0
pgscan_kswapd_movable 0 0
pgscan_direct_dma 2015438 1086109
pgscan_direct_dma32 55787823 36101597
pgscan_direct_normal 0 0
pgscan_direct_movable 0 0
pginodesteal 3461 7281
slabs_scanned 564864 527616
kswapd_steal 2889797 1448082
kswapd_inodesteal 14827 14835
pageoutrun 43459 21562
allocstall 9653 4032
pgrotated 384216 228631
1.5) free numbers at the end of the tests
before patch:
total used free shared buffers cached
Mem: 474 467 7 0 0 236
-/+ buffers/cache: 230 243
Swap: 1023 418 605
after patch:
total used free shared buffers cached
Mem: 474 457 16 0 0 236
-/+ buffers/cache: 221 253
Swap: 1023 404 619
2) memory flushing in a file server
2.1) brief summary
The number of major faults from 50 to 3 during 10% cache hot reads.
That means this patch successfully stops major faults when the active file
list is slowly scanned when there are partially cache hot streaming IO.
2.2) test scenario
Do 100000 pread(size=110 pages, offset=(i*100) pages), where 10% of the
pages will be activated:
for i in `seq 0 100 10000000`; do echo $i 110; done > pattern-hot-10
iotrace.rb --load pattern-hot-10 --play /b/sparse
vmmon nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
and monitor /proc/vmstat during the time. The test box has 2G memory.
I carried out tests on fresh booted console as well as X desktop, and
fetched the vmstat numbers on
(1) begin: shortly after the big read IO starts;
(2) end: just before the big read IO stops;
(3) restore: the big read IO stops and the zsh working set restored
(4) restore X: after IO, switch back and forth between the urxvt and firefox
windows to restore their working set.
2.3) console mode results
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.29 VM_EXEC protection ON:
begin: 2481 2237 8694 630 0 574299
end: 275 231976 233914 633 776271 20933042
restore: 370 232154 234524 691 777183 20958453
2.6.29 VM_EXEC protection ON (second run):
begin: 2434 2237 8493 629 0 574195
end: 284 231970 233536 632 771918 20896129
restore: 399 232218 234789 690 774526 20957909
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 2479 2344 9659 210 0 579643
end: 284 232010 234142 260 772776 20917184
restore: 379 232159 234371 301 774888 20967849
The above console numbers show that
- The startup pgmajfault of 2.6.30-rc4-mm is merely 1/3 that of 2.6.29.
I'd attribute that improvement to the mmap readahead improvements :-)
- The pgmajfault increment during the file copy is 633-630=3 vs 260-210=50.
That's a huge improvement - which means with the VM_EXEC protection logic,
active mmap pages is pretty safe even under partially cache hot streaming IO.
- when active:inactive file lru size reaches 1:1, their scan rates is 1:20.8
under 10% cache hot IO. (computed with formula Dpgdeactivate:Dpgfree)
That roughly means the active mmap pages get 20.8 more chances to get
re-referenced to stay in memory.
- The absolute nr_mapped drops considerably to 1/9 during the big IO, and the
dropped pages are mostly inactive ones. The patch has almost no impact in
this aspect, that means it won't unnecessarily increase memory pressure.
(In contrast, your 20% mmap protection ratio will keep them all, and
therefore eliminate the extra 41 major faults to restore working set
of zsh etc.)
The iotrace.rb read throughput is
151.194384MB/s 284.198252s 100001x 450560b --load pattern-hot-10 --play /b/sparse
which means the inactive list is rotated at the speed of 250MB/s,
so a full scan of which takes about 3.5 seconds, while a full scan
of active file list takes about 77 seconds.
2.4) X mode results
We can reach roughly the same conclusions for X desktop:
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.30-rc4-mm VM_EXEC protection ON:
begin: 9740 8920 64075 561 0 678360
end: 768 218254 220029 565 798953 21057006
restore: 857 218543 220987 606 799462 21075710
restore X: 2414 218560 225344 797 799462 21080795
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 9368 5035 26389 554 0 633391
end: 770 218449 221230 661 646472 17832500
restore: 1113 218466 220978 710 649881 17905235
restore X: 2687 218650 225484 947 802700 21083584
- the absolute nr_mapped drops considerably (to 1/13 of the original size)
during the streaming IO.
- the delta of pgmajfault is 3 vs 107 during IO, or 236 vs 393
during the whole process.
Cc: Elladan <elladan@eskimo.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:12 +07:00
|
|
|
/*
|
|
|
|
* Identify referenced, file-backed active pages and
|
|
|
|
* give them one more trip around the active list. So
|
|
|
|
* that executable code get better chances to stay in
|
|
|
|
* memory under moderate memory pressure. Anon pages
|
|
|
|
* are not likely to be evicted by use-once streaming
|
|
|
|
* IO, plus JVM can create lots of anon VM_EXEC pages,
|
|
|
|
* so we ignore them here.
|
|
|
|
*/
|
2009-10-27 06:49:53 +07:00
|
|
|
if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) {
|
vmscan: make mapped executable pages the first class citizen
Protect referenced PROT_EXEC mapped pages from being deactivated.
PROT_EXEC(or its internal presentation VM_EXEC) pages normally belong to some
currently running executables and their linked libraries, they shall really be
cached aggressively to provide good user experiences.
Thanks to Johannes Weiner for the advice to reuse the VMA walk in
page_referenced() to get the PROT_EXEC bit.
[more details]
( The consequences of this patch will have to be discussed together with
Rik van Riel's recent patch "vmscan: evict use-once pages first". )
( Some of the good points and insights are taken into this changelog.
Thanks to all the involved people for the great LKML discussions. )
the problem
===========
For a typical desktop, the most precious working set is composed of
*actively accessed*
(1) memory mapped executables
(2) and their anonymous pages
(3) and other files
(4) and the dcache/icache/.. slabs
while the least important data are
(5) infrequently used or use-once files
For a typical desktop, one major problem is busty and large amount of (5)
use-once files flushing out the working set.
Inside the working set, (4) dcache/icache have already been too sticky ;-)
So we only have to care (2) anonymous and (1)(3) file pages.
anonymous pages
===============
Anonymous pages are effectively immune to the streaming IO attack, because we
now have separate file/anon LRU lists. When the use-once files crowd into the
file LRU, the list's "quality" is significantly lowered. Therefore the scan
balance policy in get_scan_ratio() will choose to scan the (low quality) file
LRU much more frequently than the anon LRU.
file pages
==========
Rik proposed to *not* scan the active file LRU when the inactive list grows
larger than active list. This guarantees that when there are use-once streaming
IO, and the working set is not too large(so that active_size < inactive_size),
the active file LRU will *not* be scanned at all. So the not-too-large working
set can be well protected.
But there are also situations where the file working set is a bit large so that
(active_size >= inactive_size), or the streaming IOs are not purely use-once.
In these cases, the active list will be scanned slowly. Because the current
shrink_active_list() policy is to deactivate active pages regardless of their
referenced bits. The deactivated pages become susceptible to the streaming IO
attack: the inactive list could be scanned fast (500MB / 50MBps = 10s) so that
the deactivated pages don't have enough time to get re-referenced. Because a
user tend to switch between windows in intervals from seconds to minutes.
This patch holds mapped executable pages in the active list as long as they
are referenced during each full scan of the active list. Because the active
list is normally scanned much slower, they get longer grace time (eg. 100s)
for further references, which better matches the pace of user operations.
Therefore this patch greatly prolongs the in-cache time of executable code,
when there are moderate memory pressures.
before patch: guaranteed to be cached if reference intervals < I
after patch: guaranteed to be cached if reference intervals < I+A
(except when randomly reclaimed by the lumpy reclaim)
where
A = time to fully scan the active file LRU
I = time to fully scan the inactive file LRU
Note that normally A >> I.
side effects
============
This patch is safe in general, it restores the pre-2.6.28 mmap() behavior
but in a much smaller and well targeted scope.
One may worry about some one to abuse the PROT_EXEC heuristic. But as
Andrew Morton stated, there are other tricks to getting that sort of boost.
Another concern is the PROT_EXEC mapped pages growing large in rare cases,
and therefore hurting reclaim efficiency. But a sane application targeted for
large audience will never use PROT_EXEC for data mappings. If some home made
application tries to abuse that bit, it shall be aware of the consequences.
If it is abused to scale of 2/3 total memory, it gains nothing but overheads.
benchmarks
==========
1) memory tight desktop
1.1) brief summary
- clock time and major faults are reduced by 50%;
- pswpin numbers are reduced to ~1/3.
That means X desktop responsiveness is doubled under high memory/swap pressure.
1.2) test scenario
- nfsroot gnome desktop with 512M physical memory
- run some programs, and switch between the existing windows
after starting each new program.
1.3) progress timing (seconds)
before after programs
0.02 0.02 N xeyes
0.75 0.76 N firefox
2.02 1.88 N nautilus
3.36 3.17 N nautilus --browser
5.26 4.89 N gthumb
7.12 6.47 N gedit
9.22 8.16 N xpdf /usr/share/doc/shared-mime-info/shared-mime-info-spec.pdf
13.58 12.55 N xterm
15.87 14.57 N mlterm
18.63 17.06 N gnome-terminal
21.16 18.90 N urxvt
26.24 23.48 N gnome-system-monitor
28.72 26.52 N gnome-help
32.15 29.65 N gnome-dictionary
39.66 36.12 N /usr/games/sol
43.16 39.27 N /usr/games/gnometris
48.65 42.56 N /usr/games/gnect
53.31 47.03 N /usr/games/gtali
58.60 52.05 N /usr/games/iagno
65.77 55.42 N /usr/games/gnotravex
70.76 61.47 N /usr/games/mahjongg
76.15 67.11 N /usr/games/gnome-sudoku
86.32 75.15 N /usr/games/glines
92.21 79.70 N /usr/games/glchess
103.79 88.48 N /usr/games/gnomine
113.84 96.51 N /usr/games/gnotski
124.40 102.19 N /usr/games/gnibbles
137.41 114.93 N /usr/games/gnobots2
155.53 125.02 N /usr/games/blackjack
179.85 135.11 N /usr/games/same-gnome
224.49 154.50 N /usr/bin/gnome-window-properties
248.44 162.09 N /usr/bin/gnome-default-applications-properties
282.62 173.29 N /usr/bin/gnome-at-properties
323.72 188.21 N /usr/bin/gnome-typing-monitor
363.99 199.93 N /usr/bin/gnome-at-visual
394.21 206.95 N /usr/bin/gnome-sound-properties
435.14 224.49 N /usr/bin/gnome-at-mobility
463.05 234.11 N /usr/bin/gnome-keybinding-properties
503.75 248.59 N /usr/bin/gnome-about-me
554.00 276.27 N /usr/bin/gnome-display-properties
615.48 304.39 N /usr/bin/gnome-network-preferences
693.03 342.01 N /usr/bin/gnome-mouse-properties
759.90 388.58 N /usr/bin/gnome-appearance-properties
937.90 508.47 N /usr/bin/gnome-control-center
1109.75 587.57 N /usr/bin/gnome-keyboard-properties
1399.05 758.16 N : oocalc
1524.64 830.03 N : oodraw
1684.31 900.03 N : ooimpress
1874.04 993.91 N : oomath
2115.12 1081.89 N : ooweb
2369.02 1161.99 N : oowriter
Note that the last ": oo*" commands are actually commented out.
1.4) vmstat numbers (some relevant ones are marked with *)
before after
nr_free_pages 1293 3898
nr_inactive_anon 59956 53460
nr_active_anon 26815 30026
nr_inactive_file 2657 3218
nr_active_file 2019 2806
nr_unevictable 4 4
nr_mlock 4 4
nr_anon_pages 26706 27859
*nr_mapped 3542 4469
nr_file_pages 72232 67681
nr_dirty 1 0
nr_writeback 123 19
nr_slab_reclaimable 3375 3534
nr_slab_unreclaimable 11405 10665
nr_page_table_pages 8106 7864
nr_unstable 0 0
nr_bounce 0 0
*nr_vmscan_write 394776 230839
nr_writeback_temp 0 0
numa_hit 6843353 3318676
numa_miss 0 0
numa_foreign 0 0
numa_interleave 1719 1719
numa_local 6843353 3318676
numa_other 0 0
*pgpgin 5954683 2057175
*pgpgout 1578276 922744
*pswpin 1486615 512238
*pswpout 394568 230685
pgalloc_dma 277432 56602
pgalloc_dma32 6769477 3310348
pgalloc_normal 0 0
pgalloc_movable 0 0
pgfree 7048396 3371118
pgactivate 2036343 1471492
pgdeactivate 2189691 1612829
pgfault 3702176 3100702
*pgmajfault 452116 201343
pgrefill_dma 12185 7127
pgrefill_dma32 334384 653703
pgrefill_normal 0 0
pgrefill_movable 0 0
pgsteal_dma 74214 22179
pgsteal_dma32 3334164 1638029
pgsteal_normal 0 0
pgsteal_movable 0 0
pgscan_kswapd_dma 1081421 1216199
pgscan_kswapd_dma32 58979118 46002810
pgscan_kswapd_normal 0 0
pgscan_kswapd_movable 0 0
pgscan_direct_dma 2015438 1086109
pgscan_direct_dma32 55787823 36101597
pgscan_direct_normal 0 0
pgscan_direct_movable 0 0
pginodesteal 3461 7281
slabs_scanned 564864 527616
kswapd_steal 2889797 1448082
kswapd_inodesteal 14827 14835
pageoutrun 43459 21562
allocstall 9653 4032
pgrotated 384216 228631
1.5) free numbers at the end of the tests
before patch:
total used free shared buffers cached
Mem: 474 467 7 0 0 236
-/+ buffers/cache: 230 243
Swap: 1023 418 605
after patch:
total used free shared buffers cached
Mem: 474 457 16 0 0 236
-/+ buffers/cache: 221 253
Swap: 1023 404 619
2) memory flushing in a file server
2.1) brief summary
The number of major faults from 50 to 3 during 10% cache hot reads.
That means this patch successfully stops major faults when the active file
list is slowly scanned when there are partially cache hot streaming IO.
2.2) test scenario
Do 100000 pread(size=110 pages, offset=(i*100) pages), where 10% of the
pages will be activated:
for i in `seq 0 100 10000000`; do echo $i 110; done > pattern-hot-10
iotrace.rb --load pattern-hot-10 --play /b/sparse
vmmon nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
and monitor /proc/vmstat during the time. The test box has 2G memory.
I carried out tests on fresh booted console as well as X desktop, and
fetched the vmstat numbers on
(1) begin: shortly after the big read IO starts;
(2) end: just before the big read IO stops;
(3) restore: the big read IO stops and the zsh working set restored
(4) restore X: after IO, switch back and forth between the urxvt and firefox
windows to restore their working set.
2.3) console mode results
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.29 VM_EXEC protection ON:
begin: 2481 2237 8694 630 0 574299
end: 275 231976 233914 633 776271 20933042
restore: 370 232154 234524 691 777183 20958453
2.6.29 VM_EXEC protection ON (second run):
begin: 2434 2237 8493 629 0 574195
end: 284 231970 233536 632 771918 20896129
restore: 399 232218 234789 690 774526 20957909
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 2479 2344 9659 210 0 579643
end: 284 232010 234142 260 772776 20917184
restore: 379 232159 234371 301 774888 20967849
The above console numbers show that
- The startup pgmajfault of 2.6.30-rc4-mm is merely 1/3 that of 2.6.29.
I'd attribute that improvement to the mmap readahead improvements :-)
- The pgmajfault increment during the file copy is 633-630=3 vs 260-210=50.
That's a huge improvement - which means with the VM_EXEC protection logic,
active mmap pages is pretty safe even under partially cache hot streaming IO.
- when active:inactive file lru size reaches 1:1, their scan rates is 1:20.8
under 10% cache hot IO. (computed with formula Dpgdeactivate:Dpgfree)
That roughly means the active mmap pages get 20.8 more chances to get
re-referenced to stay in memory.
- The absolute nr_mapped drops considerably to 1/9 during the big IO, and the
dropped pages are mostly inactive ones. The patch has almost no impact in
this aspect, that means it won't unnecessarily increase memory pressure.
(In contrast, your 20% mmap protection ratio will keep them all, and
therefore eliminate the extra 41 major faults to restore working set
of zsh etc.)
The iotrace.rb read throughput is
151.194384MB/s 284.198252s 100001x 450560b --load pattern-hot-10 --play /b/sparse
which means the inactive list is rotated at the speed of 250MB/s,
so a full scan of which takes about 3.5 seconds, while a full scan
of active file list takes about 77 seconds.
2.4) X mode results
We can reach roughly the same conclusions for X desktop:
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.30-rc4-mm VM_EXEC protection ON:
begin: 9740 8920 64075 561 0 678360
end: 768 218254 220029 565 798953 21057006
restore: 857 218543 220987 606 799462 21075710
restore X: 2414 218560 225344 797 799462 21080795
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 9368 5035 26389 554 0 633391
end: 770 218449 221230 661 646472 17832500
restore: 1113 218466 220978 710 649881 17905235
restore X: 2687 218650 225484 947 802700 21083584
- the absolute nr_mapped drops considerably (to 1/13 of the original size)
during the streaming IO.
- the delta of pgmajfault is 3 vs 107 during IO, or 236 vs 393
during the whole process.
Cc: Elladan <elladan@eskimo.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:12 +07:00
|
|
|
list_add(&page->lru, &l_active);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
2008-10-19 10:26:35 +07:00
|
|
|
|
2009-09-22 07:01:44 +07:00
|
|
|
ClearPageActive(page); /* we are de-activating */
|
2005-04-17 05:20:36 +07:00
|
|
|
list_add(&page->lru, &l_inactive);
|
|
|
|
}
|
|
|
|
|
2009-01-07 05:40:13 +07:00
|
|
|
/*
|
vmscan: make mapped executable pages the first class citizen
Protect referenced PROT_EXEC mapped pages from being deactivated.
PROT_EXEC(or its internal presentation VM_EXEC) pages normally belong to some
currently running executables and their linked libraries, they shall really be
cached aggressively to provide good user experiences.
Thanks to Johannes Weiner for the advice to reuse the VMA walk in
page_referenced() to get the PROT_EXEC bit.
[more details]
( The consequences of this patch will have to be discussed together with
Rik van Riel's recent patch "vmscan: evict use-once pages first". )
( Some of the good points and insights are taken into this changelog.
Thanks to all the involved people for the great LKML discussions. )
the problem
===========
For a typical desktop, the most precious working set is composed of
*actively accessed*
(1) memory mapped executables
(2) and their anonymous pages
(3) and other files
(4) and the dcache/icache/.. slabs
while the least important data are
(5) infrequently used or use-once files
For a typical desktop, one major problem is busty and large amount of (5)
use-once files flushing out the working set.
Inside the working set, (4) dcache/icache have already been too sticky ;-)
So we only have to care (2) anonymous and (1)(3) file pages.
anonymous pages
===============
Anonymous pages are effectively immune to the streaming IO attack, because we
now have separate file/anon LRU lists. When the use-once files crowd into the
file LRU, the list's "quality" is significantly lowered. Therefore the scan
balance policy in get_scan_ratio() will choose to scan the (low quality) file
LRU much more frequently than the anon LRU.
file pages
==========
Rik proposed to *not* scan the active file LRU when the inactive list grows
larger than active list. This guarantees that when there are use-once streaming
IO, and the working set is not too large(so that active_size < inactive_size),
the active file LRU will *not* be scanned at all. So the not-too-large working
set can be well protected.
But there are also situations where the file working set is a bit large so that
(active_size >= inactive_size), or the streaming IOs are not purely use-once.
In these cases, the active list will be scanned slowly. Because the current
shrink_active_list() policy is to deactivate active pages regardless of their
referenced bits. The deactivated pages become susceptible to the streaming IO
attack: the inactive list could be scanned fast (500MB / 50MBps = 10s) so that
the deactivated pages don't have enough time to get re-referenced. Because a
user tend to switch between windows in intervals from seconds to minutes.
This patch holds mapped executable pages in the active list as long as they
are referenced during each full scan of the active list. Because the active
list is normally scanned much slower, they get longer grace time (eg. 100s)
for further references, which better matches the pace of user operations.
Therefore this patch greatly prolongs the in-cache time of executable code,
when there are moderate memory pressures.
before patch: guaranteed to be cached if reference intervals < I
after patch: guaranteed to be cached if reference intervals < I+A
(except when randomly reclaimed by the lumpy reclaim)
where
A = time to fully scan the active file LRU
I = time to fully scan the inactive file LRU
Note that normally A >> I.
side effects
============
This patch is safe in general, it restores the pre-2.6.28 mmap() behavior
but in a much smaller and well targeted scope.
One may worry about some one to abuse the PROT_EXEC heuristic. But as
Andrew Morton stated, there are other tricks to getting that sort of boost.
Another concern is the PROT_EXEC mapped pages growing large in rare cases,
and therefore hurting reclaim efficiency. But a sane application targeted for
large audience will never use PROT_EXEC for data mappings. If some home made
application tries to abuse that bit, it shall be aware of the consequences.
If it is abused to scale of 2/3 total memory, it gains nothing but overheads.
benchmarks
==========
1) memory tight desktop
1.1) brief summary
- clock time and major faults are reduced by 50%;
- pswpin numbers are reduced to ~1/3.
That means X desktop responsiveness is doubled under high memory/swap pressure.
1.2) test scenario
- nfsroot gnome desktop with 512M physical memory
- run some programs, and switch between the existing windows
after starting each new program.
1.3) progress timing (seconds)
before after programs
0.02 0.02 N xeyes
0.75 0.76 N firefox
2.02 1.88 N nautilus
3.36 3.17 N nautilus --browser
5.26 4.89 N gthumb
7.12 6.47 N gedit
9.22 8.16 N xpdf /usr/share/doc/shared-mime-info/shared-mime-info-spec.pdf
13.58 12.55 N xterm
15.87 14.57 N mlterm
18.63 17.06 N gnome-terminal
21.16 18.90 N urxvt
26.24 23.48 N gnome-system-monitor
28.72 26.52 N gnome-help
32.15 29.65 N gnome-dictionary
39.66 36.12 N /usr/games/sol
43.16 39.27 N /usr/games/gnometris
48.65 42.56 N /usr/games/gnect
53.31 47.03 N /usr/games/gtali
58.60 52.05 N /usr/games/iagno
65.77 55.42 N /usr/games/gnotravex
70.76 61.47 N /usr/games/mahjongg
76.15 67.11 N /usr/games/gnome-sudoku
86.32 75.15 N /usr/games/glines
92.21 79.70 N /usr/games/glchess
103.79 88.48 N /usr/games/gnomine
113.84 96.51 N /usr/games/gnotski
124.40 102.19 N /usr/games/gnibbles
137.41 114.93 N /usr/games/gnobots2
155.53 125.02 N /usr/games/blackjack
179.85 135.11 N /usr/games/same-gnome
224.49 154.50 N /usr/bin/gnome-window-properties
248.44 162.09 N /usr/bin/gnome-default-applications-properties
282.62 173.29 N /usr/bin/gnome-at-properties
323.72 188.21 N /usr/bin/gnome-typing-monitor
363.99 199.93 N /usr/bin/gnome-at-visual
394.21 206.95 N /usr/bin/gnome-sound-properties
435.14 224.49 N /usr/bin/gnome-at-mobility
463.05 234.11 N /usr/bin/gnome-keybinding-properties
503.75 248.59 N /usr/bin/gnome-about-me
554.00 276.27 N /usr/bin/gnome-display-properties
615.48 304.39 N /usr/bin/gnome-network-preferences
693.03 342.01 N /usr/bin/gnome-mouse-properties
759.90 388.58 N /usr/bin/gnome-appearance-properties
937.90 508.47 N /usr/bin/gnome-control-center
1109.75 587.57 N /usr/bin/gnome-keyboard-properties
1399.05 758.16 N : oocalc
1524.64 830.03 N : oodraw
1684.31 900.03 N : ooimpress
1874.04 993.91 N : oomath
2115.12 1081.89 N : ooweb
2369.02 1161.99 N : oowriter
Note that the last ": oo*" commands are actually commented out.
1.4) vmstat numbers (some relevant ones are marked with *)
before after
nr_free_pages 1293 3898
nr_inactive_anon 59956 53460
nr_active_anon 26815 30026
nr_inactive_file 2657 3218
nr_active_file 2019 2806
nr_unevictable 4 4
nr_mlock 4 4
nr_anon_pages 26706 27859
*nr_mapped 3542 4469
nr_file_pages 72232 67681
nr_dirty 1 0
nr_writeback 123 19
nr_slab_reclaimable 3375 3534
nr_slab_unreclaimable 11405 10665
nr_page_table_pages 8106 7864
nr_unstable 0 0
nr_bounce 0 0
*nr_vmscan_write 394776 230839
nr_writeback_temp 0 0
numa_hit 6843353 3318676
numa_miss 0 0
numa_foreign 0 0
numa_interleave 1719 1719
numa_local 6843353 3318676
numa_other 0 0
*pgpgin 5954683 2057175
*pgpgout 1578276 922744
*pswpin 1486615 512238
*pswpout 394568 230685
pgalloc_dma 277432 56602
pgalloc_dma32 6769477 3310348
pgalloc_normal 0 0
pgalloc_movable 0 0
pgfree 7048396 3371118
pgactivate 2036343 1471492
pgdeactivate 2189691 1612829
pgfault 3702176 3100702
*pgmajfault 452116 201343
pgrefill_dma 12185 7127
pgrefill_dma32 334384 653703
pgrefill_normal 0 0
pgrefill_movable 0 0
pgsteal_dma 74214 22179
pgsteal_dma32 3334164 1638029
pgsteal_normal 0 0
pgsteal_movable 0 0
pgscan_kswapd_dma 1081421 1216199
pgscan_kswapd_dma32 58979118 46002810
pgscan_kswapd_normal 0 0
pgscan_kswapd_movable 0 0
pgscan_direct_dma 2015438 1086109
pgscan_direct_dma32 55787823 36101597
pgscan_direct_normal 0 0
pgscan_direct_movable 0 0
pginodesteal 3461 7281
slabs_scanned 564864 527616
kswapd_steal 2889797 1448082
kswapd_inodesteal 14827 14835
pageoutrun 43459 21562
allocstall 9653 4032
pgrotated 384216 228631
1.5) free numbers at the end of the tests
before patch:
total used free shared buffers cached
Mem: 474 467 7 0 0 236
-/+ buffers/cache: 230 243
Swap: 1023 418 605
after patch:
total used free shared buffers cached
Mem: 474 457 16 0 0 236
-/+ buffers/cache: 221 253
Swap: 1023 404 619
2) memory flushing in a file server
2.1) brief summary
The number of major faults from 50 to 3 during 10% cache hot reads.
That means this patch successfully stops major faults when the active file
list is slowly scanned when there are partially cache hot streaming IO.
2.2) test scenario
Do 100000 pread(size=110 pages, offset=(i*100) pages), where 10% of the
pages will be activated:
for i in `seq 0 100 10000000`; do echo $i 110; done > pattern-hot-10
iotrace.rb --load pattern-hot-10 --play /b/sparse
vmmon nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
and monitor /proc/vmstat during the time. The test box has 2G memory.
I carried out tests on fresh booted console as well as X desktop, and
fetched the vmstat numbers on
(1) begin: shortly after the big read IO starts;
(2) end: just before the big read IO stops;
(3) restore: the big read IO stops and the zsh working set restored
(4) restore X: after IO, switch back and forth between the urxvt and firefox
windows to restore their working set.
2.3) console mode results
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.29 VM_EXEC protection ON:
begin: 2481 2237 8694 630 0 574299
end: 275 231976 233914 633 776271 20933042
restore: 370 232154 234524 691 777183 20958453
2.6.29 VM_EXEC protection ON (second run):
begin: 2434 2237 8493 629 0 574195
end: 284 231970 233536 632 771918 20896129
restore: 399 232218 234789 690 774526 20957909
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 2479 2344 9659 210 0 579643
end: 284 232010 234142 260 772776 20917184
restore: 379 232159 234371 301 774888 20967849
The above console numbers show that
- The startup pgmajfault of 2.6.30-rc4-mm is merely 1/3 that of 2.6.29.
I'd attribute that improvement to the mmap readahead improvements :-)
- The pgmajfault increment during the file copy is 633-630=3 vs 260-210=50.
That's a huge improvement - which means with the VM_EXEC protection logic,
active mmap pages is pretty safe even under partially cache hot streaming IO.
- when active:inactive file lru size reaches 1:1, their scan rates is 1:20.8
under 10% cache hot IO. (computed with formula Dpgdeactivate:Dpgfree)
That roughly means the active mmap pages get 20.8 more chances to get
re-referenced to stay in memory.
- The absolute nr_mapped drops considerably to 1/9 during the big IO, and the
dropped pages are mostly inactive ones. The patch has almost no impact in
this aspect, that means it won't unnecessarily increase memory pressure.
(In contrast, your 20% mmap protection ratio will keep them all, and
therefore eliminate the extra 41 major faults to restore working set
of zsh etc.)
The iotrace.rb read throughput is
151.194384MB/s 284.198252s 100001x 450560b --load pattern-hot-10 --play /b/sparse
which means the inactive list is rotated at the speed of 250MB/s,
so a full scan of which takes about 3.5 seconds, while a full scan
of active file list takes about 77 seconds.
2.4) X mode results
We can reach roughly the same conclusions for X desktop:
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.30-rc4-mm VM_EXEC protection ON:
begin: 9740 8920 64075 561 0 678360
end: 768 218254 220029 565 798953 21057006
restore: 857 218543 220987 606 799462 21075710
restore X: 2414 218560 225344 797 799462 21080795
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 9368 5035 26389 554 0 633391
end: 770 218449 221230 661 646472 17832500
restore: 1113 218466 220978 710 649881 17905235
restore X: 2687 218650 225484 947 802700 21083584
- the absolute nr_mapped drops considerably (to 1/13 of the original size)
during the streaming IO.
- the delta of pgmajfault is 3 vs 107 during IO, or 236 vs 393
during the whole process.
Cc: Elladan <elladan@eskimo.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:12 +07:00
|
|
|
* Move pages back to the lru list.
|
2009-01-07 05:40:13 +07:00
|
|
|
*/
|
2008-12-01 09:00:35 +07:00
|
|
|
spin_lock_irq(&zone->lru_lock);
|
2008-10-19 10:26:34 +07:00
|
|
|
/*
|
vmscan: make mapped executable pages the first class citizen
Protect referenced PROT_EXEC mapped pages from being deactivated.
PROT_EXEC(or its internal presentation VM_EXEC) pages normally belong to some
currently running executables and their linked libraries, they shall really be
cached aggressively to provide good user experiences.
Thanks to Johannes Weiner for the advice to reuse the VMA walk in
page_referenced() to get the PROT_EXEC bit.
[more details]
( The consequences of this patch will have to be discussed together with
Rik van Riel's recent patch "vmscan: evict use-once pages first". )
( Some of the good points and insights are taken into this changelog.
Thanks to all the involved people for the great LKML discussions. )
the problem
===========
For a typical desktop, the most precious working set is composed of
*actively accessed*
(1) memory mapped executables
(2) and their anonymous pages
(3) and other files
(4) and the dcache/icache/.. slabs
while the least important data are
(5) infrequently used or use-once files
For a typical desktop, one major problem is busty and large amount of (5)
use-once files flushing out the working set.
Inside the working set, (4) dcache/icache have already been too sticky ;-)
So we only have to care (2) anonymous and (1)(3) file pages.
anonymous pages
===============
Anonymous pages are effectively immune to the streaming IO attack, because we
now have separate file/anon LRU lists. When the use-once files crowd into the
file LRU, the list's "quality" is significantly lowered. Therefore the scan
balance policy in get_scan_ratio() will choose to scan the (low quality) file
LRU much more frequently than the anon LRU.
file pages
==========
Rik proposed to *not* scan the active file LRU when the inactive list grows
larger than active list. This guarantees that when there are use-once streaming
IO, and the working set is not too large(so that active_size < inactive_size),
the active file LRU will *not* be scanned at all. So the not-too-large working
set can be well protected.
But there are also situations where the file working set is a bit large so that
(active_size >= inactive_size), or the streaming IOs are not purely use-once.
In these cases, the active list will be scanned slowly. Because the current
shrink_active_list() policy is to deactivate active pages regardless of their
referenced bits. The deactivated pages become susceptible to the streaming IO
attack: the inactive list could be scanned fast (500MB / 50MBps = 10s) so that
the deactivated pages don't have enough time to get re-referenced. Because a
user tend to switch between windows in intervals from seconds to minutes.
This patch holds mapped executable pages in the active list as long as they
are referenced during each full scan of the active list. Because the active
list is normally scanned much slower, they get longer grace time (eg. 100s)
for further references, which better matches the pace of user operations.
Therefore this patch greatly prolongs the in-cache time of executable code,
when there are moderate memory pressures.
before patch: guaranteed to be cached if reference intervals < I
after patch: guaranteed to be cached if reference intervals < I+A
(except when randomly reclaimed by the lumpy reclaim)
where
A = time to fully scan the active file LRU
I = time to fully scan the inactive file LRU
Note that normally A >> I.
side effects
============
This patch is safe in general, it restores the pre-2.6.28 mmap() behavior
but in a much smaller and well targeted scope.
One may worry about some one to abuse the PROT_EXEC heuristic. But as
Andrew Morton stated, there are other tricks to getting that sort of boost.
Another concern is the PROT_EXEC mapped pages growing large in rare cases,
and therefore hurting reclaim efficiency. But a sane application targeted for
large audience will never use PROT_EXEC for data mappings. If some home made
application tries to abuse that bit, it shall be aware of the consequences.
If it is abused to scale of 2/3 total memory, it gains nothing but overheads.
benchmarks
==========
1) memory tight desktop
1.1) brief summary
- clock time and major faults are reduced by 50%;
- pswpin numbers are reduced to ~1/3.
That means X desktop responsiveness is doubled under high memory/swap pressure.
1.2) test scenario
- nfsroot gnome desktop with 512M physical memory
- run some programs, and switch between the existing windows
after starting each new program.
1.3) progress timing (seconds)
before after programs
0.02 0.02 N xeyes
0.75 0.76 N firefox
2.02 1.88 N nautilus
3.36 3.17 N nautilus --browser
5.26 4.89 N gthumb
7.12 6.47 N gedit
9.22 8.16 N xpdf /usr/share/doc/shared-mime-info/shared-mime-info-spec.pdf
13.58 12.55 N xterm
15.87 14.57 N mlterm
18.63 17.06 N gnome-terminal
21.16 18.90 N urxvt
26.24 23.48 N gnome-system-monitor
28.72 26.52 N gnome-help
32.15 29.65 N gnome-dictionary
39.66 36.12 N /usr/games/sol
43.16 39.27 N /usr/games/gnometris
48.65 42.56 N /usr/games/gnect
53.31 47.03 N /usr/games/gtali
58.60 52.05 N /usr/games/iagno
65.77 55.42 N /usr/games/gnotravex
70.76 61.47 N /usr/games/mahjongg
76.15 67.11 N /usr/games/gnome-sudoku
86.32 75.15 N /usr/games/glines
92.21 79.70 N /usr/games/glchess
103.79 88.48 N /usr/games/gnomine
113.84 96.51 N /usr/games/gnotski
124.40 102.19 N /usr/games/gnibbles
137.41 114.93 N /usr/games/gnobots2
155.53 125.02 N /usr/games/blackjack
179.85 135.11 N /usr/games/same-gnome
224.49 154.50 N /usr/bin/gnome-window-properties
248.44 162.09 N /usr/bin/gnome-default-applications-properties
282.62 173.29 N /usr/bin/gnome-at-properties
323.72 188.21 N /usr/bin/gnome-typing-monitor
363.99 199.93 N /usr/bin/gnome-at-visual
394.21 206.95 N /usr/bin/gnome-sound-properties
435.14 224.49 N /usr/bin/gnome-at-mobility
463.05 234.11 N /usr/bin/gnome-keybinding-properties
503.75 248.59 N /usr/bin/gnome-about-me
554.00 276.27 N /usr/bin/gnome-display-properties
615.48 304.39 N /usr/bin/gnome-network-preferences
693.03 342.01 N /usr/bin/gnome-mouse-properties
759.90 388.58 N /usr/bin/gnome-appearance-properties
937.90 508.47 N /usr/bin/gnome-control-center
1109.75 587.57 N /usr/bin/gnome-keyboard-properties
1399.05 758.16 N : oocalc
1524.64 830.03 N : oodraw
1684.31 900.03 N : ooimpress
1874.04 993.91 N : oomath
2115.12 1081.89 N : ooweb
2369.02 1161.99 N : oowriter
Note that the last ": oo*" commands are actually commented out.
1.4) vmstat numbers (some relevant ones are marked with *)
before after
nr_free_pages 1293 3898
nr_inactive_anon 59956 53460
nr_active_anon 26815 30026
nr_inactive_file 2657 3218
nr_active_file 2019 2806
nr_unevictable 4 4
nr_mlock 4 4
nr_anon_pages 26706 27859
*nr_mapped 3542 4469
nr_file_pages 72232 67681
nr_dirty 1 0
nr_writeback 123 19
nr_slab_reclaimable 3375 3534
nr_slab_unreclaimable 11405 10665
nr_page_table_pages 8106 7864
nr_unstable 0 0
nr_bounce 0 0
*nr_vmscan_write 394776 230839
nr_writeback_temp 0 0
numa_hit 6843353 3318676
numa_miss 0 0
numa_foreign 0 0
numa_interleave 1719 1719
numa_local 6843353 3318676
numa_other 0 0
*pgpgin 5954683 2057175
*pgpgout 1578276 922744
*pswpin 1486615 512238
*pswpout 394568 230685
pgalloc_dma 277432 56602
pgalloc_dma32 6769477 3310348
pgalloc_normal 0 0
pgalloc_movable 0 0
pgfree 7048396 3371118
pgactivate 2036343 1471492
pgdeactivate 2189691 1612829
pgfault 3702176 3100702
*pgmajfault 452116 201343
pgrefill_dma 12185 7127
pgrefill_dma32 334384 653703
pgrefill_normal 0 0
pgrefill_movable 0 0
pgsteal_dma 74214 22179
pgsteal_dma32 3334164 1638029
pgsteal_normal 0 0
pgsteal_movable 0 0
pgscan_kswapd_dma 1081421 1216199
pgscan_kswapd_dma32 58979118 46002810
pgscan_kswapd_normal 0 0
pgscan_kswapd_movable 0 0
pgscan_direct_dma 2015438 1086109
pgscan_direct_dma32 55787823 36101597
pgscan_direct_normal 0 0
pgscan_direct_movable 0 0
pginodesteal 3461 7281
slabs_scanned 564864 527616
kswapd_steal 2889797 1448082
kswapd_inodesteal 14827 14835
pageoutrun 43459 21562
allocstall 9653 4032
pgrotated 384216 228631
1.5) free numbers at the end of the tests
before patch:
total used free shared buffers cached
Mem: 474 467 7 0 0 236
-/+ buffers/cache: 230 243
Swap: 1023 418 605
after patch:
total used free shared buffers cached
Mem: 474 457 16 0 0 236
-/+ buffers/cache: 221 253
Swap: 1023 404 619
2) memory flushing in a file server
2.1) brief summary
The number of major faults from 50 to 3 during 10% cache hot reads.
That means this patch successfully stops major faults when the active file
list is slowly scanned when there are partially cache hot streaming IO.
2.2) test scenario
Do 100000 pread(size=110 pages, offset=(i*100) pages), where 10% of the
pages will be activated:
for i in `seq 0 100 10000000`; do echo $i 110; done > pattern-hot-10
iotrace.rb --load pattern-hot-10 --play /b/sparse
vmmon nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
and monitor /proc/vmstat during the time. The test box has 2G memory.
I carried out tests on fresh booted console as well as X desktop, and
fetched the vmstat numbers on
(1) begin: shortly after the big read IO starts;
(2) end: just before the big read IO stops;
(3) restore: the big read IO stops and the zsh working set restored
(4) restore X: after IO, switch back and forth between the urxvt and firefox
windows to restore their working set.
2.3) console mode results
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.29 VM_EXEC protection ON:
begin: 2481 2237 8694 630 0 574299
end: 275 231976 233914 633 776271 20933042
restore: 370 232154 234524 691 777183 20958453
2.6.29 VM_EXEC protection ON (second run):
begin: 2434 2237 8493 629 0 574195
end: 284 231970 233536 632 771918 20896129
restore: 399 232218 234789 690 774526 20957909
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 2479 2344 9659 210 0 579643
end: 284 232010 234142 260 772776 20917184
restore: 379 232159 234371 301 774888 20967849
The above console numbers show that
- The startup pgmajfault of 2.6.30-rc4-mm is merely 1/3 that of 2.6.29.
I'd attribute that improvement to the mmap readahead improvements :-)
- The pgmajfault increment during the file copy is 633-630=3 vs 260-210=50.
That's a huge improvement - which means with the VM_EXEC protection logic,
active mmap pages is pretty safe even under partially cache hot streaming IO.
- when active:inactive file lru size reaches 1:1, their scan rates is 1:20.8
under 10% cache hot IO. (computed with formula Dpgdeactivate:Dpgfree)
That roughly means the active mmap pages get 20.8 more chances to get
re-referenced to stay in memory.
- The absolute nr_mapped drops considerably to 1/9 during the big IO, and the
dropped pages are mostly inactive ones. The patch has almost no impact in
this aspect, that means it won't unnecessarily increase memory pressure.
(In contrast, your 20% mmap protection ratio will keep them all, and
therefore eliminate the extra 41 major faults to restore working set
of zsh etc.)
The iotrace.rb read throughput is
151.194384MB/s 284.198252s 100001x 450560b --load pattern-hot-10 --play /b/sparse
which means the inactive list is rotated at the speed of 250MB/s,
so a full scan of which takes about 3.5 seconds, while a full scan
of active file list takes about 77 seconds.
2.4) X mode results
We can reach roughly the same conclusions for X desktop:
nr_mapped nr_active_file nr_inactive_file pgmajfault pgdeactivate pgfree
2.6.30-rc4-mm VM_EXEC protection ON:
begin: 9740 8920 64075 561 0 678360
end: 768 218254 220029 565 798953 21057006
restore: 857 218543 220987 606 799462 21075710
restore X: 2414 218560 225344 797 799462 21080795
2.6.30-rc4-mm VM_EXEC protection OFF:
begin: 9368 5035 26389 554 0 633391
end: 770 218449 221230 661 646472 17832500
restore: 1113 218466 220978 710 649881 17905235
restore X: 2687 218650 225484 947 802700 21083584
- the absolute nr_mapped drops considerably (to 1/13 of the original size)
during the streaming IO.
- the delta of pgmajfault is 3 vs 107 during IO, or 236 vs 393
during the whole process.
Cc: Elladan <elladan@eskimo.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:12 +07:00
|
|
|
* Count referenced pages from currently used mappings as rotated,
|
|
|
|
* even though only some of them are actually re-activated. This
|
|
|
|
* helps balance scan pressure between file and anonymous pages in
|
|
|
|
* get_scan_ratio.
|
2008-10-19 10:26:35 +07:00
|
|
|
*/
|
2009-09-22 07:02:56 +07:00
|
|
|
reclaim_stat->recent_rotated[file] += nr_rotated;
|
2008-10-19 10:26:34 +07:00
|
|
|
|
2012-01-13 08:19:56 +07:00
|
|
|
move_active_pages_to_lru(zone, &l_active, &l_hold,
|
2009-06-17 05:33:13 +07:00
|
|
|
LRU_ACTIVE + file * LRU_FILE);
|
2012-01-13 08:19:56 +07:00
|
|
|
move_active_pages_to_lru(zone, &l_inactive, &l_hold,
|
2009-06-17 05:33:13 +07:00
|
|
|
LRU_BASE + file * LRU_FILE);
|
2009-09-22 07:01:37 +07:00
|
|
|
__mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken);
|
2006-06-30 15:55:45 +07:00
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
2012-01-13 08:19:56 +07:00
|
|
|
|
|
|
|
free_hot_cold_page_list(&l_hold, 1);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2010-10-27 04:21:31 +07:00
|
|
|
#ifdef CONFIG_SWAP
|
2009-01-08 09:08:18 +07:00
|
|
|
static int inactive_anon_is_low_global(struct zone *zone)
|
2009-01-08 09:08:14 +07:00
|
|
|
{
|
|
|
|
unsigned long active, inactive;
|
|
|
|
|
|
|
|
active = zone_page_state(zone, NR_ACTIVE_ANON);
|
|
|
|
inactive = zone_page_state(zone, NR_INACTIVE_ANON);
|
|
|
|
|
|
|
|
if (inactive * zone->inactive_ratio < active)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-01-08 09:08:18 +07:00
|
|
|
/**
|
|
|
|
* inactive_anon_is_low - check if anonymous pages need to be deactivated
|
|
|
|
* @zone: zone to check
|
|
|
|
* @sc: scan control of this context
|
|
|
|
*
|
|
|
|
* Returns true if the zone does not have enough inactive anon pages,
|
|
|
|
* meaning some active anon pages need to be deactivated.
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
static int inactive_anon_is_low(struct mem_cgroup_zone *mz)
|
2009-01-08 09:08:18 +07:00
|
|
|
{
|
2010-10-27 04:21:31 +07:00
|
|
|
/*
|
|
|
|
* If we don't have swap space, anonymous page deactivation
|
|
|
|
* is pointless.
|
|
|
|
*/
|
|
|
|
if (!total_swap_pages)
|
|
|
|
return 0;
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
if (!scanning_global_lru(mz))
|
|
|
|
return mem_cgroup_inactive_anon_is_low(mz->mem_cgroup,
|
|
|
|
mz->zone);
|
|
|
|
|
|
|
|
return inactive_anon_is_low_global(mz->zone);
|
2009-01-08 09:08:18 +07:00
|
|
|
}
|
2010-10-27 04:21:31 +07:00
|
|
|
#else
|
2012-01-13 08:17:52 +07:00
|
|
|
static inline int inactive_anon_is_low(struct mem_cgroup_zone *mz)
|
2010-10-27 04:21:31 +07:00
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
2009-01-08 09:08:18 +07:00
|
|
|
|
2009-06-17 05:32:28 +07:00
|
|
|
static int inactive_file_is_low_global(struct zone *zone)
|
|
|
|
{
|
|
|
|
unsigned long active, inactive;
|
|
|
|
|
|
|
|
active = zone_page_state(zone, NR_ACTIVE_FILE);
|
|
|
|
inactive = zone_page_state(zone, NR_INACTIVE_FILE);
|
|
|
|
|
|
|
|
return (active > inactive);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* inactive_file_is_low - check if file pages need to be deactivated
|
2012-01-13 08:17:52 +07:00
|
|
|
* @mz: memory cgroup and zone to check
|
2009-06-17 05:32:28 +07:00
|
|
|
*
|
|
|
|
* When the system is doing streaming IO, memory pressure here
|
|
|
|
* ensures that active file pages get deactivated, until more
|
|
|
|
* than half of the file pages are on the inactive list.
|
|
|
|
*
|
|
|
|
* Once we get to that situation, protect the system's working
|
|
|
|
* set from being evicted by disabling active file page aging.
|
|
|
|
*
|
|
|
|
* This uses a different ratio than the anonymous pages, because
|
|
|
|
* the page cache uses a use-once replacement algorithm.
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
static int inactive_file_is_low(struct mem_cgroup_zone *mz)
|
2009-06-17 05:32:28 +07:00
|
|
|
{
|
2012-01-13 08:17:52 +07:00
|
|
|
if (!scanning_global_lru(mz))
|
|
|
|
return mem_cgroup_inactive_file_is_low(mz->mem_cgroup,
|
|
|
|
mz->zone);
|
2009-06-17 05:32:28 +07:00
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
return inactive_file_is_low_global(mz->zone);
|
2009-06-17 05:32:28 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static int inactive_list_is_low(struct mem_cgroup_zone *mz, int file)
|
2009-12-15 08:59:48 +07:00
|
|
|
{
|
|
|
|
if (file)
|
2012-01-13 08:17:52 +07:00
|
|
|
return inactive_file_is_low(mz);
|
2009-12-15 08:59:48 +07:00
|
|
|
else
|
2012-01-13 08:17:52 +07:00
|
|
|
return inactive_anon_is_low(mz);
|
2009-12-15 08:59:48 +07:00
|
|
|
}
|
|
|
|
|
2008-10-19 10:26:32 +07:00
|
|
|
static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
|
2012-01-13 08:17:52 +07:00
|
|
|
struct mem_cgroup_zone *mz,
|
|
|
|
struct scan_control *sc, int priority)
|
2008-10-19 10:26:14 +07:00
|
|
|
{
|
2008-10-19 10:26:32 +07:00
|
|
|
int file = is_file_lru(lru);
|
|
|
|
|
2009-12-15 08:59:48 +07:00
|
|
|
if (is_active_lru(lru)) {
|
2012-01-13 08:17:52 +07:00
|
|
|
if (inactive_list_is_low(mz, file))
|
|
|
|
shrink_active_list(nr_to_scan, mz, sc, priority, file);
|
2008-10-19 10:26:34 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
return shrink_inactive_list(nr_to_scan, mz, sc, priority, file);
|
2008-10-19 10:26:32 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static int vmscan_swappiness(struct mem_cgroup_zone *mz,
|
|
|
|
struct scan_control *sc)
|
2011-07-27 06:08:21 +07:00
|
|
|
{
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc))
|
2011-07-27 06:08:21 +07:00
|
|
|
return vm_swappiness;
|
2012-01-13 08:17:52 +07:00
|
|
|
return mem_cgroup_swappiness(mz->mem_cgroup);
|
2011-07-27 06:08:21 +07:00
|
|
|
}
|
|
|
|
|
2008-10-19 10:26:32 +07:00
|
|
|
/*
|
|
|
|
* Determine how aggressively the anon and file LRU lists should be
|
|
|
|
* scanned. The relative value of each set of LRU lists is determined
|
|
|
|
* by looking at the fraction of the pages scanned we did rotate back
|
|
|
|
* onto the active list instead of evict.
|
|
|
|
*
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
* nr[0] = anon pages to scan; nr[1] = file pages to scan
|
2008-10-19 10:26:32 +07:00
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
static void get_scan_count(struct mem_cgroup_zone *mz, struct scan_control *sc,
|
|
|
|
unsigned long *nr, int priority)
|
2008-10-19 10:26:32 +07:00
|
|
|
{
|
|
|
|
unsigned long anon, file, free;
|
|
|
|
unsigned long anon_prio, file_prio;
|
|
|
|
unsigned long ap, fp;
|
2012-01-13 08:17:52 +07:00
|
|
|
struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(mz);
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
u64 fraction[2], denominator;
|
2012-01-13 08:20:01 +07:00
|
|
|
enum lru_list lru;
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
int noswap = 0;
|
2011-09-15 06:21:52 +07:00
|
|
|
bool force_scan = false;
|
memcg: fix get_scan_count() for small targets
During memory reclaim we determine the number of pages to be scanned per
zone as
(anon + file) >> priority.
Assume
scan = (anon + file) >> priority.
If scan < SWAP_CLUSTER_MAX, the scan will be skipped for this time and
priority gets higher. This has some problems.
1. This increases priority as 1 without any scan.
To do scan in this priority, amount of pages should be larger than 512M.
If pages>>priority < SWAP_CLUSTER_MAX, it's recorded and scan will be
batched, later. (But we lose 1 priority.)
If memory size is below 16M, pages >> priority is 0 and no scan in
DEF_PRIORITY forever.
2. If zone->all_unreclaimabe==true, it's scanned only when priority==0.
So, x86's ZONE_DMA will never be recoverred until the user of pages
frees memory by itself.
3. With memcg, the limit of memory can be small. When using small memcg,
it gets priority < DEF_PRIORITY-2 very easily and need to call
wait_iff_congested().
For doing scan before priorty=9, 64MB of memory should be used.
Then, this patch tries to scan SWAP_CLUSTER_MAX of pages in force...when
1. the target is enough small.
2. it's kswapd or memcg reclaim.
Then we can avoid rapid priority drop and may be able to recover
all_unreclaimable in a small zones. And this patch removes nr_saved_scan.
This will allow scanning in this priority even when pages >> priority is
very small.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Ying Han <yinghan@google.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-05-27 06:25:34 +07:00
|
|
|
|
2011-11-01 07:07:27 +07:00
|
|
|
/*
|
|
|
|
* If the zone or memcg is small, nr[l] can be 0. This
|
|
|
|
* results in no scanning on this priority and a potential
|
|
|
|
* priority drop. Global direct reclaim can go to the next
|
|
|
|
* zone and tends to have no problems. Global kswapd is for
|
|
|
|
* zone balancing and it needs to scan a minimum amount. When
|
|
|
|
* reclaiming for a memcg, a priority drop can cause high
|
|
|
|
* latencies, so it's better to scan a minimum amount there as
|
|
|
|
* well.
|
|
|
|
*/
|
2012-01-13 08:18:06 +07:00
|
|
|
if (current_is_kswapd() && mz->zone->all_unreclaimable)
|
2011-09-15 06:21:52 +07:00
|
|
|
force_scan = true;
|
2012-01-13 08:17:50 +07:00
|
|
|
if (!global_reclaim(sc))
|
2011-09-15 06:21:52 +07:00
|
|
|
force_scan = true;
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
|
|
|
|
/* If we have no swap space, do not bother scanning anon pages. */
|
|
|
|
if (!sc->may_swap || (nr_swap_pages <= 0)) {
|
|
|
|
noswap = 1;
|
|
|
|
fraction[0] = 0;
|
|
|
|
fraction[1] = 1;
|
|
|
|
denominator = 1;
|
|
|
|
goto out;
|
|
|
|
}
|
2008-10-19 10:26:32 +07:00
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
anon = zone_nr_lru_pages(mz, LRU_ACTIVE_ANON) +
|
|
|
|
zone_nr_lru_pages(mz, LRU_INACTIVE_ANON);
|
|
|
|
file = zone_nr_lru_pages(mz, LRU_ACTIVE_FILE) +
|
|
|
|
zone_nr_lru_pages(mz, LRU_INACTIVE_FILE);
|
2011-09-15 06:21:52 +07:00
|
|
|
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc)) {
|
2012-01-13 08:17:52 +07:00
|
|
|
free = zone_page_state(mz->zone, NR_FREE_PAGES);
|
2009-01-08 09:08:17 +07:00
|
|
|
/* If we have very few page cache pages,
|
|
|
|
force-scan anon pages. */
|
2012-01-13 08:17:52 +07:00
|
|
|
if (unlikely(file + free <= high_wmark_pages(mz->zone))) {
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
fraction[0] = 1;
|
|
|
|
fraction[1] = 0;
|
|
|
|
denominator = 1;
|
|
|
|
goto out;
|
2009-01-08 09:08:17 +07:00
|
|
|
}
|
2008-10-19 10:26:32 +07:00
|
|
|
}
|
|
|
|
|
2010-08-10 07:19:51 +07:00
|
|
|
/*
|
|
|
|
* With swappiness at 100, anonymous and file have the same priority.
|
|
|
|
* This scanning priority is essentially the inverse of IO cost.
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
anon_prio = vmscan_swappiness(mz, sc);
|
|
|
|
file_prio = 200 - vmscan_swappiness(mz, sc);
|
2010-08-10 07:19:51 +07:00
|
|
|
|
2008-10-19 10:26:32 +07:00
|
|
|
/*
|
|
|
|
* OK, so we have swap space and a fair amount of page cache
|
|
|
|
* pages. We use the recently rotated / recently scanned
|
|
|
|
* ratios to determine how valuable each cache is.
|
|
|
|
*
|
|
|
|
* Because workloads change over time (and to avoid overflow)
|
|
|
|
* we keep these statistics as a floating average, which ends
|
|
|
|
* up weighing recent references more than old ones.
|
|
|
|
*
|
|
|
|
* anon in [0], file in [1]
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
spin_lock_irq(&mz->zone->lru_lock);
|
2009-01-08 09:08:15 +07:00
|
|
|
if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) {
|
|
|
|
reclaim_stat->recent_scanned[0] /= 2;
|
|
|
|
reclaim_stat->recent_rotated[0] /= 2;
|
2008-10-19 10:26:32 +07:00
|
|
|
}
|
|
|
|
|
2009-01-08 09:08:15 +07:00
|
|
|
if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) {
|
|
|
|
reclaim_stat->recent_scanned[1] /= 2;
|
|
|
|
reclaim_stat->recent_rotated[1] /= 2;
|
2008-10-19 10:26:32 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2008-11-20 06:36:44 +07:00
|
|
|
* The amount of pressure on anon vs file pages is inversely
|
|
|
|
* proportional to the fraction of recently scanned pages on
|
|
|
|
* each list that were recently referenced and in active use.
|
2008-10-19 10:26:32 +07:00
|
|
|
*/
|
2009-01-08 09:08:15 +07:00
|
|
|
ap = (anon_prio + 1) * (reclaim_stat->recent_scanned[0] + 1);
|
|
|
|
ap /= reclaim_stat->recent_rotated[0] + 1;
|
2008-10-19 10:26:32 +07:00
|
|
|
|
2009-01-08 09:08:15 +07:00
|
|
|
fp = (file_prio + 1) * (reclaim_stat->recent_scanned[1] + 1);
|
|
|
|
fp /= reclaim_stat->recent_rotated[1] + 1;
|
2012-01-13 08:17:52 +07:00
|
|
|
spin_unlock_irq(&mz->zone->lru_lock);
|
2008-10-19 10:26:32 +07:00
|
|
|
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
fraction[0] = ap;
|
|
|
|
fraction[1] = fp;
|
|
|
|
denominator = ap + fp + 1;
|
|
|
|
out:
|
2012-01-13 08:20:01 +07:00
|
|
|
for_each_evictable_lru(lru) {
|
|
|
|
int file = is_file_lru(lru);
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
unsigned long scan;
|
2009-06-17 05:32:29 +07:00
|
|
|
|
2012-01-13 08:20:01 +07:00
|
|
|
scan = zone_nr_lru_pages(mz, lru);
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
if (priority || noswap) {
|
|
|
|
scan >>= priority;
|
2011-11-01 07:07:27 +07:00
|
|
|
if (!scan && force_scan)
|
|
|
|
scan = SWAP_CLUSTER_MAX;
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
scan = div64_u64(scan * fraction[file], denominator);
|
|
|
|
}
|
2012-01-13 08:20:01 +07:00
|
|
|
nr[lru] = scan;
|
vmscan: prevent get_scan_ratio() rounding errors
get_scan_ratio() calculates percentage and if the percentage is < 1%, it
will round percentage down to 0% and cause we completely ignore scanning
anon/file pages to reclaim memory even the total anon/file pages are very
big.
To avoid underflow, we don't use percentage, instead we directly calculate
how many pages should be scaned. In this way, we should get several
scanned pages for < 1% percent.
This has some benefits:
1. increase our calculation precision
2. making our scan more smoothly. Without this, if percent[x] is
underflow, shrink_zone() doesn't scan any pages and suddenly it scans
all pages when priority is zero. With this, even priority isn't zero,
shrink_zone() gets chance to scan some pages.
Note, this patch doesn't really change logics, but just increase
precision. For system with a lot of memory, this might slightly changes
behavior. For example, in a sequential file read workload, without the
patch, we don't swap any anon pages. With it, if anon memory size is
bigger than 16G, we will see one anon page swapped. The 16G is calculated
as PAGE_SIZE * priority(4096) * (fp/ap). fp/ap is assumed to be 1024
which is common in this workload. So the impact sounds not a big deal.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-25 04:32:36 +07:00
|
|
|
}
|
2009-06-17 05:32:29 +07:00
|
|
|
}
|
2008-10-19 10:26:32 +07:00
|
|
|
|
2011-01-14 06:45:56 +07:00
|
|
|
/*
|
|
|
|
* Reclaim/compaction depends on a number of pages being freed. To avoid
|
|
|
|
* disruption to the system, a small number of order-0 pages continue to be
|
|
|
|
* rotated and reclaimed in the normal fashion. However, by the time we get
|
|
|
|
* back to the allocator and call try_to_compact_zone(), we ensure that
|
|
|
|
* there are enough free pages for it to be likely successful
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
static inline bool should_continue_reclaim(struct mem_cgroup_zone *mz,
|
2011-01-14 06:45:56 +07:00
|
|
|
unsigned long nr_reclaimed,
|
|
|
|
unsigned long nr_scanned,
|
|
|
|
struct scan_control *sc)
|
|
|
|
{
|
|
|
|
unsigned long pages_for_compaction;
|
|
|
|
unsigned long inactive_lru_pages;
|
|
|
|
|
|
|
|
/* If not in reclaim/compaction mode, stop */
|
2011-01-14 06:46:00 +07:00
|
|
|
if (!(sc->reclaim_mode & RECLAIM_MODE_COMPACTION))
|
2011-01-14 06:45:56 +07:00
|
|
|
return false;
|
|
|
|
|
mm: vmscan: stop reclaim/compaction earlier due to insufficient progress if !__GFP_REPEAT
should_continue_reclaim() for reclaim/compaction allows scanning to
continue even if pages are not being reclaimed until the full list is
scanned. In terms of allocation success, this makes sense but potentially
it introduces unwanted latency for high-order allocations such as
transparent hugepages and network jumbo frames that would prefer to fail
the allocation attempt and fallback to order-0 pages. Worse, there is a
potential that the full LRU scan will clear all the young bits, distort
page aging information and potentially push pages into swap that would
have otherwise remained resident.
This patch will stop reclaim/compaction if no pages were reclaimed in the
last SWAP_CLUSTER_MAX pages that were considered. For allocations such as
hugetlbfs that use __GFP_REPEAT and have fewer fallback options, the full
LRU list may still be scanned.
Order-0 allocation should not be affected because RECLAIM_MODE_COMPACTION
is not set so the following avoids the gfp_mask being examined:
if (!(sc->reclaim_mode & RECLAIM_MODE_COMPACTION))
return false;
A tool was developed based on ftrace that tracked the latency of
high-order allocations while transparent hugepage support was enabled and
three benchmarks were run. The "fix-infinite" figures are 2.6.38-rc4 with
Johannes's patch "vmscan: fix zone shrinking exit when scan work is done"
applied.
STREAM Highorder Allocation Latency Statistics
fix-infinite break-early
1 :: Count 10298 10229
1 :: Min 0.4560 0.4640
1 :: Mean 1.0589 1.0183
1 :: Max 14.5990 11.7510
1 :: Stddev 0.5208 0.4719
2 :: Count 2 1
2 :: Min 1.8610 3.7240
2 :: Mean 3.4325 3.7240
2 :: Max 5.0040 3.7240
2 :: Stddev 1.5715 0.0000
9 :: Count 111696 111694
9 :: Min 0.5230 0.4110
9 :: Mean 10.5831 10.5718
9 :: Max 38.4480 43.2900
9 :: Stddev 1.1147 1.1325
Mean time for order-1 allocations is reduced. order-2 looks increased but
with so few allocations, it's not particularly significant. THP mean
allocation latency is also reduced. That said, allocation time varies so
significantly that the reductions are within noise.
Max allocation time is reduced by a significant amount for low-order
allocations but reduced for THP allocations which presumably are now
breaking before reclaim has done enough work.
SysBench Highorder Allocation Latency Statistics
fix-infinite break-early
1 :: Count 15745 15677
1 :: Min 0.4250 0.4550
1 :: Mean 1.1023 1.0810
1 :: Max 14.4590 10.8220
1 :: Stddev 0.5117 0.5100
2 :: Count 1 1
2 :: Min 3.0040 2.1530
2 :: Mean 3.0040 2.1530
2 :: Max 3.0040 2.1530
2 :: Stddev 0.0000 0.0000
9 :: Count 2017 1931
9 :: Min 0.4980 0.7480
9 :: Mean 10.4717 10.3840
9 :: Max 24.9460 26.2500
9 :: Stddev 1.1726 1.1966
Again, mean time for order-1 allocations is reduced while order-2
allocations are too few to draw conclusions from. The mean time for THP
allocations is also slightly reduced albeit the reductions are within
varianes.
Once again, our maximum allocation time is significantly reduced for
low-order allocations and slightly increased for THP allocations.
Anon stream mmap reference Highorder Allocation Latency Statistics
1 :: Count 1376 1790
1 :: Min 0.4940 0.5010
1 :: Mean 1.0289 0.9732
1 :: Max 6.2670 4.2540
1 :: Stddev 0.4142 0.2785
2 :: Count 1 -
2 :: Min 1.9060 -
2 :: Mean 1.9060 -
2 :: Max 1.9060 -
2 :: Stddev 0.0000 -
9 :: Count 11266 11257
9 :: Min 0.4990 0.4940
9 :: Mean 27250.4669 24256.1919
9 :: Max 11439211.0000 6008885.0000
9 :: Stddev 226427.4624 186298.1430
This benchmark creates one thread per CPU which references an amount of
anonymous memory 1.5 times the size of physical RAM. This pounds swap
quite heavily and is intended to exercise THP a bit.
Mean allocation time for order-1 is reduced as before. It's also reduced
for THP allocations but the variations here are pretty massive due to
swap. As before, maximum allocation times are significantly reduced.
Overall, the patch reduces the mean and maximum allocation latencies for
the smaller high-order allocations. This was with Slab configured so it
would be expected to be more significant with Slub which uses these size
allocations more aggressively.
The mean allocation times for THP allocations are also slightly reduced.
The maximum latency was slightly increased as predicted by the comments
due to reclaim/compaction breaking early. However, workloads care more
about the latency of lower-order allocations than THP so it's an
acceptable trade-off.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-02-26 05:44:20 +07:00
|
|
|
/* Consider stopping depending on scan and reclaim activity */
|
|
|
|
if (sc->gfp_mask & __GFP_REPEAT) {
|
|
|
|
/*
|
|
|
|
* For __GFP_REPEAT allocations, stop reclaiming if the
|
|
|
|
* full LRU list has been scanned and we are still failing
|
|
|
|
* to reclaim pages. This full LRU scan is potentially
|
|
|
|
* expensive but a __GFP_REPEAT caller really wants to succeed
|
|
|
|
*/
|
|
|
|
if (!nr_reclaimed && !nr_scanned)
|
|
|
|
return false;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* For non-__GFP_REPEAT allocations which can presumably
|
|
|
|
* fail without consequence, stop if we failed to reclaim
|
|
|
|
* any pages from the last SWAP_CLUSTER_MAX number of
|
|
|
|
* pages that were scanned. This will return to the
|
|
|
|
* caller faster at the risk reclaim/compaction and
|
|
|
|
* the resulting allocation attempt fails
|
|
|
|
*/
|
|
|
|
if (!nr_reclaimed)
|
|
|
|
return false;
|
|
|
|
}
|
2011-01-14 06:45:56 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If we have not reclaimed enough pages for compaction and the
|
|
|
|
* inactive lists are large enough, continue reclaiming
|
|
|
|
*/
|
|
|
|
pages_for_compaction = (2UL << sc->order);
|
2012-01-13 08:17:52 +07:00
|
|
|
inactive_lru_pages = zone_nr_lru_pages(mz, LRU_INACTIVE_FILE);
|
2012-01-11 06:08:18 +07:00
|
|
|
if (nr_swap_pages > 0)
|
2012-01-13 08:17:52 +07:00
|
|
|
inactive_lru_pages += zone_nr_lru_pages(mz, LRU_INACTIVE_ANON);
|
2011-01-14 06:45:56 +07:00
|
|
|
if (sc->nr_reclaimed < pages_for_compaction &&
|
|
|
|
inactive_lru_pages > pages_for_compaction)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
/* If compaction would go ahead or the allocation would succeed, stop */
|
2012-01-13 08:17:52 +07:00
|
|
|
switch (compaction_suitable(mz->zone, sc->order)) {
|
2011-01-14 06:45:56 +07:00
|
|
|
case COMPACT_PARTIAL:
|
|
|
|
case COMPACT_CONTINUE:
|
|
|
|
return false;
|
|
|
|
default:
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* This is a basic per-zone page freer. Used by both kswapd and direct reclaim.
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
static void shrink_mem_cgroup_zone(int priority, struct mem_cgroup_zone *mz,
|
|
|
|
struct scan_control *sc)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2008-10-19 10:26:14 +07:00
|
|
|
unsigned long nr[NR_LRU_LISTS];
|
2006-03-22 15:08:18 +07:00
|
|
|
unsigned long nr_to_scan;
|
2012-01-13 08:20:01 +07:00
|
|
|
enum lru_list lru;
|
2011-02-11 06:01:34 +07:00
|
|
|
unsigned long nr_reclaimed, nr_scanned;
|
2009-12-15 08:59:10 +07:00
|
|
|
unsigned long nr_to_reclaim = sc->nr_to_reclaim;
|
2011-11-01 07:07:03 +07:00
|
|
|
struct blk_plug plug;
|
2008-10-19 10:26:55 +07:00
|
|
|
|
2011-01-14 06:45:56 +07:00
|
|
|
restart:
|
|
|
|
nr_reclaimed = 0;
|
2011-02-11 06:01:34 +07:00
|
|
|
nr_scanned = sc->nr_scanned;
|
2012-01-13 08:17:52 +07:00
|
|
|
get_scan_count(mz, sc, nr, priority);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-11-01 07:07:03 +07:00
|
|
|
blk_start_plug(&plug);
|
2008-10-19 10:26:34 +07:00
|
|
|
while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] ||
|
|
|
|
nr[LRU_INACTIVE_FILE]) {
|
2012-01-13 08:20:01 +07:00
|
|
|
for_each_evictable_lru(lru) {
|
|
|
|
if (nr[lru]) {
|
2009-12-15 08:59:14 +07:00
|
|
|
nr_to_scan = min_t(unsigned long,
|
2012-01-13 08:20:01 +07:00
|
|
|
nr[lru], SWAP_CLUSTER_MAX);
|
|
|
|
nr[lru] -= nr_to_scan;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2012-01-13 08:20:01 +07:00
|
|
|
nr_reclaimed += shrink_list(lru, nr_to_scan,
|
2012-01-13 08:17:52 +07:00
|
|
|
mz, sc, priority);
|
2008-10-19 10:26:14 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
/*
|
|
|
|
* On large memory systems, scan >> priority can become
|
|
|
|
* really large. This is fine for the starting priority;
|
|
|
|
* we want to put equal scanning pressure on each zone.
|
|
|
|
* However, if the VM has a harder time of freeing pages,
|
|
|
|
* with multiple processes reclaiming pages, the total
|
|
|
|
* freeing target can get unreasonably large.
|
|
|
|
*/
|
2012-04-13 02:49:16 +07:00
|
|
|
if (nr_reclaimed >= nr_to_reclaim && priority < DEF_PRIORITY)
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
break;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2011-11-01 07:07:03 +07:00
|
|
|
blk_finish_plug(&plug);
|
2011-01-14 06:45:56 +07:00
|
|
|
sc->nr_reclaimed += nr_reclaimed;
|
2009-01-07 05:40:02 +07:00
|
|
|
|
2008-10-19 10:26:34 +07:00
|
|
|
/*
|
|
|
|
* Even if we did not try to evict anon pages at all, we want to
|
|
|
|
* rebalance the anon lru active/inactive ratio.
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
if (inactive_anon_is_low(mz))
|
|
|
|
shrink_active_list(SWAP_CLUSTER_MAX, mz, sc, priority, 0);
|
2008-10-19 10:26:34 +07:00
|
|
|
|
2011-01-14 06:45:56 +07:00
|
|
|
/* reclaim/compaction might need reclaim to continue */
|
2012-01-13 08:17:52 +07:00
|
|
|
if (should_continue_reclaim(mz, nr_reclaimed,
|
2011-01-14 06:45:56 +07:00
|
|
|
sc->nr_scanned - nr_scanned, sc))
|
|
|
|
goto restart;
|
|
|
|
|
2007-03-01 11:13:21 +07:00
|
|
|
throttle_vm_writeout(sc->gfp_mask);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static void shrink_zone(int priority, struct zone *zone,
|
|
|
|
struct scan_control *sc)
|
|
|
|
{
|
2012-01-13 08:17:59 +07:00
|
|
|
struct mem_cgroup *root = sc->target_mem_cgroup;
|
|
|
|
struct mem_cgroup_reclaim_cookie reclaim = {
|
2012-01-13 08:17:52 +07:00
|
|
|
.zone = zone,
|
2012-01-13 08:17:59 +07:00
|
|
|
.priority = priority,
|
2012-01-13 08:17:52 +07:00
|
|
|
};
|
2012-01-13 08:17:59 +07:00
|
|
|
struct mem_cgroup *memcg;
|
|
|
|
|
|
|
|
memcg = mem_cgroup_iter(root, NULL, &reclaim);
|
|
|
|
do {
|
|
|
|
struct mem_cgroup_zone mz = {
|
|
|
|
.mem_cgroup = memcg,
|
|
|
|
.zone = zone,
|
|
|
|
};
|
2012-01-13 08:17:52 +07:00
|
|
|
|
2012-01-13 08:17:59 +07:00
|
|
|
shrink_mem_cgroup_zone(priority, &mz, sc);
|
|
|
|
/*
|
|
|
|
* Limit reclaim has historically picked one memcg and
|
|
|
|
* scanned it with decreasing priority levels until
|
|
|
|
* nr_to_reclaim had been reclaimed. This priority
|
|
|
|
* cycle is thus over after a single memcg.
|
2012-01-13 08:18:06 +07:00
|
|
|
*
|
|
|
|
* Direct reclaim and kswapd, on the other hand, have
|
|
|
|
* to scan all memory cgroups to fulfill the overall
|
|
|
|
* scan target for the zone.
|
2012-01-13 08:17:59 +07:00
|
|
|
*/
|
|
|
|
if (!global_reclaim(sc)) {
|
|
|
|
mem_cgroup_iter_break(root, memcg);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
memcg = mem_cgroup_iter(root, memcg, &reclaim);
|
|
|
|
} while (memcg);
|
2012-01-13 08:17:52 +07:00
|
|
|
}
|
|
|
|
|
2012-01-13 08:19:45 +07:00
|
|
|
/* Returns true if compaction should go ahead for a high-order request */
|
|
|
|
static inline bool compaction_ready(struct zone *zone, struct scan_control *sc)
|
|
|
|
{
|
|
|
|
unsigned long balance_gap, watermark;
|
|
|
|
bool watermark_ok;
|
|
|
|
|
|
|
|
/* Do not consider compaction for orders reclaim is meant to satisfy */
|
|
|
|
if (sc->order <= PAGE_ALLOC_COSTLY_ORDER)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Compaction takes time to run and there are potentially other
|
|
|
|
* callers using the pages just freed. Continue reclaiming until
|
|
|
|
* there is a buffer of free pages available to give compaction
|
|
|
|
* a reasonable chance of completing and allocating the page
|
|
|
|
*/
|
|
|
|
balance_gap = min(low_wmark_pages(zone),
|
|
|
|
(zone->present_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
|
|
|
|
KSWAPD_ZONE_BALANCE_GAP_RATIO);
|
|
|
|
watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order);
|
|
|
|
watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If compaction is deferred, reclaim up to a point where
|
|
|
|
* compaction will have a chance of success when re-enabled
|
|
|
|
*/
|
2012-03-22 06:33:52 +07:00
|
|
|
if (compaction_deferred(zone, sc->order))
|
2012-01-13 08:19:45 +07:00
|
|
|
return watermark_ok;
|
|
|
|
|
|
|
|
/* If compaction is not ready to start, keep reclaiming */
|
|
|
|
if (!compaction_suitable(zone, sc->order))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
return watermark_ok;
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* This is the direct reclaim path, for page-allocating processes. We only
|
|
|
|
* try to reclaim pages from zones which will satisfy the caller's allocation
|
|
|
|
* request.
|
|
|
|
*
|
2009-06-17 05:32:12 +07:00
|
|
|
* We reclaim from a zone even if that zone is over high_wmark_pages(zone).
|
|
|
|
* Because:
|
2005-04-17 05:20:36 +07:00
|
|
|
* a) The caller may be trying to free *extra* pages to satisfy a higher-order
|
|
|
|
* allocation or
|
2009-06-17 05:32:12 +07:00
|
|
|
* b) The target zone may be at high_wmark_pages(zone) but the lower zones
|
|
|
|
* must go *over* high_wmark_pages(zone) to satisfy the `incremental min'
|
|
|
|
* zone defense algorithm.
|
2005-04-17 05:20:36 +07:00
|
|
|
*
|
|
|
|
* If a zone is deemed to be full of pinned pages then just give it a light
|
|
|
|
* scan then give up on it.
|
2011-11-01 07:09:33 +07:00
|
|
|
*
|
|
|
|
* This function returns true if a zone is being reclaimed for a costly
|
2012-01-13 08:19:45 +07:00
|
|
|
* high-order allocation and compaction is ready to begin. This indicates to
|
2012-01-13 08:19:49 +07:00
|
|
|
* the caller that it should consider retrying the allocation instead of
|
|
|
|
* further reclaim.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2011-11-01 07:09:33 +07:00
|
|
|
static bool shrink_zones(int priority, struct zonelist *zonelist,
|
2006-03-22 15:08:20 +07:00
|
|
|
struct scan_control *sc)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2008-04-28 16:12:17 +07:00
|
|
|
struct zoneref *z;
|
2008-04-28 16:12:16 +07:00
|
|
|
struct zone *zone;
|
2011-05-27 06:25:27 +07:00
|
|
|
unsigned long nr_soft_reclaimed;
|
|
|
|
unsigned long nr_soft_scanned;
|
2012-01-13 08:19:49 +07:00
|
|
|
bool aborted_reclaim = false;
|
2008-02-07 15:14:37 +07:00
|
|
|
|
2012-03-22 06:34:00 +07:00
|
|
|
/*
|
|
|
|
* If the number of buffer_heads in the machine exceeds the maximum
|
|
|
|
* allowed level, force direct reclaim to scan the highmem zone as
|
|
|
|
* highmem pages could be pinning lowmem pages storing buffer_heads
|
|
|
|
*/
|
|
|
|
if (buffer_heads_over_limit)
|
|
|
|
sc->gfp_mask |= __GFP_HIGHMEM;
|
|
|
|
|
2010-08-10 07:19:29 +07:00
|
|
|
for_each_zone_zonelist_nodemask(zone, z, zonelist,
|
|
|
|
gfp_zone(sc->gfp_mask), sc->nodemask) {
|
2006-01-06 15:11:15 +07:00
|
|
|
if (!populated_zone(zone))
|
2005-04-17 05:20:36 +07:00
|
|
|
continue;
|
2008-02-07 15:14:37 +07:00
|
|
|
/*
|
|
|
|
* Take care memory controller reclaiming has small influence
|
|
|
|
* to global LRU.
|
|
|
|
*/
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc)) {
|
2008-02-07 15:14:37 +07:00
|
|
|
if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
|
|
|
|
continue;
|
2010-03-06 04:41:55 +07:00
|
|
|
if (zone->all_unreclaimable && priority != DEF_PRIORITY)
|
2008-02-07 15:14:37 +07:00
|
|
|
continue; /* Let kswapd poll it */
|
2011-11-01 07:09:31 +07:00
|
|
|
if (COMPACTION_BUILD) {
|
|
|
|
/*
|
2011-11-01 07:09:33 +07:00
|
|
|
* If we already have plenty of memory free for
|
|
|
|
* compaction in this zone, don't free any more.
|
|
|
|
* Even though compaction is invoked for any
|
|
|
|
* non-zero order, only frequent costly order
|
|
|
|
* reclamation is disruptive enough to become a
|
2012-03-22 06:34:10 +07:00
|
|
|
* noticeable problem, like transparent huge
|
|
|
|
* page allocations.
|
2011-11-01 07:09:31 +07:00
|
|
|
*/
|
2012-01-13 08:19:45 +07:00
|
|
|
if (compaction_ready(zone, sc)) {
|
2012-01-13 08:19:49 +07:00
|
|
|
aborted_reclaim = true;
|
2011-11-01 07:09:31 +07:00
|
|
|
continue;
|
2011-11-01 07:09:33 +07:00
|
|
|
}
|
2011-11-01 07:09:31 +07:00
|
|
|
}
|
2011-06-28 06:18:12 +07:00
|
|
|
/*
|
|
|
|
* This steals pages from memory cgroups over softlimit
|
|
|
|
* and returns the number of reclaimed pages and
|
|
|
|
* scanned pages. This works for global memory pressure
|
|
|
|
* and balancing, not for a memcg's limit.
|
|
|
|
*/
|
|
|
|
nr_soft_scanned = 0;
|
|
|
|
nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone,
|
|
|
|
sc->order, sc->gfp_mask,
|
|
|
|
&nr_soft_scanned);
|
|
|
|
sc->nr_reclaimed += nr_soft_reclaimed;
|
|
|
|
sc->nr_scanned += nr_soft_scanned;
|
|
|
|
/* need some check for avoid more shrink_zone() */
|
2008-02-07 15:14:37 +07:00
|
|
|
}
|
2006-09-26 13:31:27 +07:00
|
|
|
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
shrink_zone(priority, zone, sc);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2011-11-01 07:09:33 +07:00
|
|
|
|
2012-01-13 08:19:49 +07:00
|
|
|
return aborted_reclaim;
|
2010-09-23 03:05:01 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
static bool zone_reclaimable(struct zone *zone)
|
|
|
|
{
|
|
|
|
return zone->pages_scanned < zone_reclaimable_pages(zone) * 6;
|
|
|
|
}
|
|
|
|
|
vmscan: all_unreclaimable() use zone->all_unreclaimable as a name
all_unreclaimable check in direct reclaim has been introduced at 2.6.19
by following commit.
2006 Sep 25; commit 408d8544; oom: use unreclaimable info
And it went through strange history. firstly, following commit broke
the logic unintentionally.
2008 Apr 29; commit a41f24ea; page allocator: smarter retry of
costly-order allocations
Two years later, I've found obvious meaningless code fragment and
restored original intention by following commit.
2010 Jun 04; commit bb21c7ce; vmscan: fix do_try_to_free_pages()
return value when priority==0
But, the logic didn't works when 32bit highmem system goes hibernation
and Minchan slightly changed the algorithm and fixed it .
2010 Sep 22: commit d1908362: vmscan: check all_unreclaimable
in direct reclaim path
But, recently, Andrey Vagin found the new corner case. Look,
struct zone {
..
int all_unreclaimable;
..
unsigned long pages_scanned;
..
}
zone->all_unreclaimable and zone->pages_scanned are neigher atomic
variables nor protected by lock. Therefore zones can become a state of
zone->page_scanned=0 and zone->all_unreclaimable=1. In this case, current
all_unreclaimable() return false even though zone->all_unreclaimabe=1.
This resulted in the kernel hanging up when executing a loop of the form
1. fork
2. mmap
3. touch memory
4. read memory
5. munmmap
as described in
http://www.gossamer-threads.com/lists/linux/kernel/1348725#1348725
Is this ignorable minor issue? No. Unfortunately, x86 has very small dma
zone and it become zone->all_unreclamble=1 easily. and if it become
all_unreclaimable=1, it never restore all_unreclaimable=0. Why? if
all_unreclaimable=1, vmscan only try DEF_PRIORITY reclaim and
a-few-lru-pages>>DEF_PRIORITY always makes 0. that mean no page scan at
all!
Eventually, oom-killer never works on such systems. That said, we can't
use zone->pages_scanned for this purpose. This patch restore
all_unreclaimable() use zone->all_unreclaimable as old. and in addition,
to add oom_killer_disabled check to avoid reintroduce the issue of commit
d1908362 ("vmscan: check all_unreclaimable in direct reclaim path").
Reported-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-15 05:22:12 +07:00
|
|
|
/* All zones in zonelist are unreclaimable? */
|
2010-09-23 03:05:01 +07:00
|
|
|
static bool all_unreclaimable(struct zonelist *zonelist,
|
|
|
|
struct scan_control *sc)
|
|
|
|
{
|
|
|
|
struct zoneref *z;
|
|
|
|
struct zone *zone;
|
|
|
|
|
|
|
|
for_each_zone_zonelist_nodemask(zone, z, zonelist,
|
|
|
|
gfp_zone(sc->gfp_mask), sc->nodemask) {
|
|
|
|
if (!populated_zone(zone))
|
|
|
|
continue;
|
|
|
|
if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
|
|
|
|
continue;
|
vmscan: all_unreclaimable() use zone->all_unreclaimable as a name
all_unreclaimable check in direct reclaim has been introduced at 2.6.19
by following commit.
2006 Sep 25; commit 408d8544; oom: use unreclaimable info
And it went through strange history. firstly, following commit broke
the logic unintentionally.
2008 Apr 29; commit a41f24ea; page allocator: smarter retry of
costly-order allocations
Two years later, I've found obvious meaningless code fragment and
restored original intention by following commit.
2010 Jun 04; commit bb21c7ce; vmscan: fix do_try_to_free_pages()
return value when priority==0
But, the logic didn't works when 32bit highmem system goes hibernation
and Minchan slightly changed the algorithm and fixed it .
2010 Sep 22: commit d1908362: vmscan: check all_unreclaimable
in direct reclaim path
But, recently, Andrey Vagin found the new corner case. Look,
struct zone {
..
int all_unreclaimable;
..
unsigned long pages_scanned;
..
}
zone->all_unreclaimable and zone->pages_scanned are neigher atomic
variables nor protected by lock. Therefore zones can become a state of
zone->page_scanned=0 and zone->all_unreclaimable=1. In this case, current
all_unreclaimable() return false even though zone->all_unreclaimabe=1.
This resulted in the kernel hanging up when executing a loop of the form
1. fork
2. mmap
3. touch memory
4. read memory
5. munmmap
as described in
http://www.gossamer-threads.com/lists/linux/kernel/1348725#1348725
Is this ignorable minor issue? No. Unfortunately, x86 has very small dma
zone and it become zone->all_unreclamble=1 easily. and if it become
all_unreclaimable=1, it never restore all_unreclaimable=0. Why? if
all_unreclaimable=1, vmscan only try DEF_PRIORITY reclaim and
a-few-lru-pages>>DEF_PRIORITY always makes 0. that mean no page scan at
all!
Eventually, oom-killer never works on such systems. That said, we can't
use zone->pages_scanned for this purpose. This patch restore
all_unreclaimable() use zone->all_unreclaimable as old. and in addition,
to add oom_killer_disabled check to avoid reintroduce the issue of commit
d1908362 ("vmscan: check all_unreclaimable in direct reclaim path").
Reported-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-15 05:22:12 +07:00
|
|
|
if (!zone->all_unreclaimable)
|
|
|
|
return false;
|
2010-09-23 03:05:01 +07:00
|
|
|
}
|
|
|
|
|
vmscan: all_unreclaimable() use zone->all_unreclaimable as a name
all_unreclaimable check in direct reclaim has been introduced at 2.6.19
by following commit.
2006 Sep 25; commit 408d8544; oom: use unreclaimable info
And it went through strange history. firstly, following commit broke
the logic unintentionally.
2008 Apr 29; commit a41f24ea; page allocator: smarter retry of
costly-order allocations
Two years later, I've found obvious meaningless code fragment and
restored original intention by following commit.
2010 Jun 04; commit bb21c7ce; vmscan: fix do_try_to_free_pages()
return value when priority==0
But, the logic didn't works when 32bit highmem system goes hibernation
and Minchan slightly changed the algorithm and fixed it .
2010 Sep 22: commit d1908362: vmscan: check all_unreclaimable
in direct reclaim path
But, recently, Andrey Vagin found the new corner case. Look,
struct zone {
..
int all_unreclaimable;
..
unsigned long pages_scanned;
..
}
zone->all_unreclaimable and zone->pages_scanned are neigher atomic
variables nor protected by lock. Therefore zones can become a state of
zone->page_scanned=0 and zone->all_unreclaimable=1. In this case, current
all_unreclaimable() return false even though zone->all_unreclaimabe=1.
This resulted in the kernel hanging up when executing a loop of the form
1. fork
2. mmap
3. touch memory
4. read memory
5. munmmap
as described in
http://www.gossamer-threads.com/lists/linux/kernel/1348725#1348725
Is this ignorable minor issue? No. Unfortunately, x86 has very small dma
zone and it become zone->all_unreclamble=1 easily. and if it become
all_unreclaimable=1, it never restore all_unreclaimable=0. Why? if
all_unreclaimable=1, vmscan only try DEF_PRIORITY reclaim and
a-few-lru-pages>>DEF_PRIORITY always makes 0. that mean no page scan at
all!
Eventually, oom-killer never works on such systems. That said, we can't
use zone->pages_scanned for this purpose. This patch restore
all_unreclaimable() use zone->all_unreclaimable as old. and in addition,
to add oom_killer_disabled check to avoid reintroduce the issue of commit
d1908362 ("vmscan: check all_unreclaimable in direct reclaim path").
Reported-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-15 05:22:12 +07:00
|
|
|
return true;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2008-10-19 10:26:32 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* This is the main entry point to direct page reclaim.
|
|
|
|
*
|
|
|
|
* If a full scan of the inactive list fails to free enough memory then we
|
|
|
|
* are "out of memory" and something needs to be killed.
|
|
|
|
*
|
|
|
|
* If the caller is !__GFP_FS then the probability of a failure is reasonably
|
|
|
|
* high - the zone may be full of dirty or under-writeback pages, which this
|
2009-09-24 00:37:09 +07:00
|
|
|
* caller can't do much about. We kick the writeback threads and take explicit
|
|
|
|
* naps in the hope that some of these pages can be written. But if the
|
|
|
|
* allocating task holds filesystem locks which prevent writeout this might not
|
|
|
|
* work, and the allocation attempt will fail.
|
page allocator: smarter retry of costly-order allocations
Because of page order checks in __alloc_pages(), hugepage (and similarly
large order) allocations will not retry unless explicitly marked
__GFP_REPEAT. However, the current retry logic is nearly an infinite
loop (or until reclaim does no progress whatsoever). For these costly
allocations, that seems like overkill and could potentially never
terminate. Mel observed that allowing current __GFP_REPEAT semantics for
hugepage allocations essentially killed the system. I believe this is
because we may continue to reclaim small orders of pages all over, but
never have enough to satisfy the hugepage allocation request. This is
clearly only a problem for large order allocations, of which hugepages
are the most obvious (to me).
Modify try_to_free_pages() to indicate how many pages were reclaimed.
Use that information in __alloc_pages() to eventually fail a large
__GFP_REPEAT allocation when we've reclaimed an order of pages equal to
or greater than the allocation's order. This relies on lumpy reclaim
functioning as advertised. Due to fragmentation, lumpy reclaim may not
be able to free up the order needed in one invocation, so multiple
iterations may be requred. In other words, the more fragmented memory
is, the more retry attempts __GFP_REPEAT will make (particularly for
higher order allocations).
This changes the semantics of __GFP_REPEAT subtly, but *only* for
allocations > PAGE_ALLOC_COSTLY_ORDER. With this patch, for those size
allocations, we will try up to some point (at least 1<<order reclaimed
pages), rather than forever (which is the case for allocations <=
PAGE_ALLOC_COSTLY_ORDER).
This change improves the /proc/sys/vm/nr_hugepages interface with a
follow-on patch that makes pool allocations use __GFP_REPEAT. Rather
than administrators repeatedly echo'ing a particular value into the
sysctl, and forcing reclaim into action manually, this change allows for
the sysctl to attempt a reasonable effort itself. Similarly, dynamic
pool growth should be more successful under load, as lumpy reclaim can
try to free up pages, rather than failing right away.
Choosing to reclaim only up to the order of the requested allocation
strikes a balance between not failing hugepage allocations and returning
to the caller when it's unlikely to every succeed. Because of lumpy
reclaim, if we have freed the order requested, hopefully it has been in
big chunks and those chunks will allow our allocation to succeed. If
that isn't the case after freeing up the current order, I don't think it
is likely to succeed in the future, although it is possible given a
particular fragmentation pattern.
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Tested-by: Mel Gorman <mel@csn.ul.ie>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-29 14:58:25 +07:00
|
|
|
*
|
|
|
|
* returns: 0, if no pages reclaimed
|
|
|
|
* else, the number of pages reclaimed
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2008-04-28 16:12:12 +07:00
|
|
|
static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
|
2011-05-25 07:12:26 +07:00
|
|
|
struct scan_control *sc,
|
|
|
|
struct shrink_control *shrink)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
int priority;
|
2006-03-22 15:08:19 +07:00
|
|
|
unsigned long total_scanned = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
struct reclaim_state *reclaim_state = current->reclaim_state;
|
2008-04-28 16:12:17 +07:00
|
|
|
struct zoneref *z;
|
2008-04-28 16:12:16 +07:00
|
|
|
struct zone *zone;
|
2009-12-15 08:59:10 +07:00
|
|
|
unsigned long writeback_threshold;
|
2012-01-13 08:19:49 +07:00
|
|
|
bool aborted_reclaim;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-07-25 15:48:52 +07:00
|
|
|
delayacct_freepages_start();
|
|
|
|
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc))
|
2008-02-07 15:14:37 +07:00
|
|
|
count_vm_event(ALLOCSTALL);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
for (priority = DEF_PRIORITY; priority >= 0; priority--) {
|
2008-02-07 15:13:56 +07:00
|
|
|
sc->nr_scanned = 0;
|
2005-11-29 04:44:07 +07:00
|
|
|
if (!priority)
|
2012-01-13 08:17:52 +07:00
|
|
|
disable_swap_token(sc->target_mem_cgroup);
|
2012-01-13 08:19:49 +07:00
|
|
|
aborted_reclaim = shrink_zones(priority, zonelist, sc);
|
2011-11-01 07:09:33 +07:00
|
|
|
|
2008-02-07 15:13:56 +07:00
|
|
|
/*
|
|
|
|
* Don't shrink slabs when reclaiming memory from
|
|
|
|
* over limit cgroups
|
|
|
|
*/
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc)) {
|
2010-08-10 07:19:14 +07:00
|
|
|
unsigned long lru_pages = 0;
|
2010-08-10 07:19:29 +07:00
|
|
|
for_each_zone_zonelist(zone, z, zonelist,
|
|
|
|
gfp_zone(sc->gfp_mask)) {
|
2010-08-10 07:19:14 +07:00
|
|
|
if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
lru_pages += zone_reclaimable_pages(zone);
|
|
|
|
}
|
|
|
|
|
2011-05-25 07:12:27 +07:00
|
|
|
shrink_slab(shrink, sc->nr_scanned, lru_pages);
|
2008-02-07 15:14:29 +07:00
|
|
|
if (reclaim_state) {
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
sc->nr_reclaimed += reclaim_state->reclaimed_slab;
|
2008-02-07 15:14:29 +07:00
|
|
|
reclaim_state->reclaimed_slab = 0;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2008-02-07 15:13:56 +07:00
|
|
|
total_scanned += sc->nr_scanned;
|
2010-06-05 04:15:05 +07:00
|
|
|
if (sc->nr_reclaimed >= sc->nr_to_reclaim)
|
2005-04-17 05:20:36 +07:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Try to write back as many pages as we just scanned. This
|
|
|
|
* tends to cause slow streaming writers to write data to the
|
|
|
|
* disk smoothly, at the dirtying rate, which is nice. But
|
|
|
|
* that's undesirable in laptop mode, where we *want* lumpy
|
|
|
|
* writeout. So in laptop mode, write out the whole world.
|
|
|
|
*/
|
2009-12-15 08:59:10 +07:00
|
|
|
writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2;
|
|
|
|
if (total_scanned > writeback_threshold) {
|
2011-10-08 10:54:10 +07:00
|
|
|
wakeup_flusher_threads(laptop_mode ? 0 : total_scanned,
|
|
|
|
WB_REASON_TRY_TO_FREE_PAGES);
|
2008-02-07 15:13:56 +07:00
|
|
|
sc->may_writepage = 1;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Take a nap, wait for some writeback to complete */
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
if (!sc->hibernation_mode && sc->nr_scanned &&
|
2010-10-27 04:21:45 +07:00
|
|
|
priority < DEF_PRIORITY - 2) {
|
|
|
|
struct zone *preferred_zone;
|
|
|
|
|
|
|
|
first_zones_zonelist(zonelist, gfp_zone(sc->gfp_mask),
|
2011-01-26 06:07:20 +07:00
|
|
|
&cpuset_current_mems_allowed,
|
|
|
|
&preferred_zone);
|
2010-10-27 04:21:45 +07:00
|
|
|
wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/10);
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2010-06-05 04:15:05 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
out:
|
2008-07-25 15:48:52 +07:00
|
|
|
delayacct_freepages_end();
|
|
|
|
|
2010-06-05 04:15:05 +07:00
|
|
|
if (sc->nr_reclaimed)
|
|
|
|
return sc->nr_reclaimed;
|
|
|
|
|
vmscan: all_unreclaimable() use zone->all_unreclaimable as a name
all_unreclaimable check in direct reclaim has been introduced at 2.6.19
by following commit.
2006 Sep 25; commit 408d8544; oom: use unreclaimable info
And it went through strange history. firstly, following commit broke
the logic unintentionally.
2008 Apr 29; commit a41f24ea; page allocator: smarter retry of
costly-order allocations
Two years later, I've found obvious meaningless code fragment and
restored original intention by following commit.
2010 Jun 04; commit bb21c7ce; vmscan: fix do_try_to_free_pages()
return value when priority==0
But, the logic didn't works when 32bit highmem system goes hibernation
and Minchan slightly changed the algorithm and fixed it .
2010 Sep 22: commit d1908362: vmscan: check all_unreclaimable
in direct reclaim path
But, recently, Andrey Vagin found the new corner case. Look,
struct zone {
..
int all_unreclaimable;
..
unsigned long pages_scanned;
..
}
zone->all_unreclaimable and zone->pages_scanned are neigher atomic
variables nor protected by lock. Therefore zones can become a state of
zone->page_scanned=0 and zone->all_unreclaimable=1. In this case, current
all_unreclaimable() return false even though zone->all_unreclaimabe=1.
This resulted in the kernel hanging up when executing a loop of the form
1. fork
2. mmap
3. touch memory
4. read memory
5. munmmap
as described in
http://www.gossamer-threads.com/lists/linux/kernel/1348725#1348725
Is this ignorable minor issue? No. Unfortunately, x86 has very small dma
zone and it become zone->all_unreclamble=1 easily. and if it become
all_unreclaimable=1, it never restore all_unreclaimable=0. Why? if
all_unreclaimable=1, vmscan only try DEF_PRIORITY reclaim and
a-few-lru-pages>>DEF_PRIORITY always makes 0. that mean no page scan at
all!
Eventually, oom-killer never works on such systems. That said, we can't
use zone->pages_scanned for this purpose. This patch restore
all_unreclaimable() use zone->all_unreclaimable as old. and in addition,
to add oom_killer_disabled check to avoid reintroduce the issue of commit
d1908362 ("vmscan: check all_unreclaimable in direct reclaim path").
Reported-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-15 05:22:12 +07:00
|
|
|
/*
|
|
|
|
* As hibernation is going on, kswapd is freezed so that it can't mark
|
|
|
|
* the zone into all_unreclaimable. Thus bypassing all_unreclaimable
|
|
|
|
* check.
|
|
|
|
*/
|
|
|
|
if (oom_killer_disabled)
|
|
|
|
return 0;
|
|
|
|
|
2012-01-13 08:19:49 +07:00
|
|
|
/* Aborted reclaim to try compaction? don't OOM, then */
|
|
|
|
if (aborted_reclaim)
|
2012-01-13 08:19:33 +07:00
|
|
|
return 1;
|
|
|
|
|
2010-06-05 04:15:05 +07:00
|
|
|
/* top priority shrink_zones still had more to do? don't OOM, then */
|
2012-01-13 08:17:50 +07:00
|
|
|
if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc))
|
2010-06-05 04:15:05 +07:00
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2008-04-28 16:12:12 +07:00
|
|
|
unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
|
2009-04-01 05:23:31 +07:00
|
|
|
gfp_t gfp_mask, nodemask_t *nodemask)
|
2008-02-07 15:13:56 +07:00
|
|
|
{
|
2010-08-10 07:19:16 +07:00
|
|
|
unsigned long nr_reclaimed;
|
2008-02-07 15:13:56 +07:00
|
|
|
struct scan_control sc = {
|
|
|
|
.gfp_mask = gfp_mask,
|
|
|
|
.may_writepage = !laptop_mode,
|
2009-12-15 08:59:10 +07:00
|
|
|
.nr_to_reclaim = SWAP_CLUSTER_MAX,
|
2009-04-01 05:19:30 +07:00
|
|
|
.may_unmap = 1,
|
2009-04-22 02:24:57 +07:00
|
|
|
.may_swap = 1,
|
2008-02-07 15:13:56 +07:00
|
|
|
.order = order,
|
2012-01-13 08:17:52 +07:00
|
|
|
.target_mem_cgroup = NULL,
|
2009-04-01 05:23:31 +07:00
|
|
|
.nodemask = nodemask,
|
2008-02-07 15:13:56 +07:00
|
|
|
};
|
2011-05-25 07:12:26 +07:00
|
|
|
struct shrink_control shrink = {
|
|
|
|
.gfp_mask = sc.gfp_mask,
|
|
|
|
};
|
2008-02-07 15:13:56 +07:00
|
|
|
|
2010-08-10 07:19:16 +07:00
|
|
|
trace_mm_vmscan_direct_reclaim_begin(order,
|
|
|
|
sc.may_writepage,
|
|
|
|
gfp_mask);
|
|
|
|
|
2011-05-25 07:12:26 +07:00
|
|
|
nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
|
2010-08-10 07:19:16 +07:00
|
|
|
|
|
|
|
trace_mm_vmscan_direct_reclaim_end(nr_reclaimed);
|
|
|
|
|
|
|
|
return nr_reclaimed;
|
2008-02-07 15:13:56 +07:00
|
|
|
}
|
|
|
|
|
2008-03-05 05:28:39 +07:00
|
|
|
#ifdef CONFIG_CGROUP_MEM_RES_CTLR
|
2008-02-07 15:13:56 +07:00
|
|
|
|
2012-01-13 08:18:32 +07:00
|
|
|
unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg,
|
2009-09-24 05:56:39 +07:00
|
|
|
gfp_t gfp_mask, bool noswap,
|
2011-05-27 06:25:25 +07:00
|
|
|
struct zone *zone,
|
|
|
|
unsigned long *nr_scanned)
|
2009-09-24 05:56:39 +07:00
|
|
|
{
|
|
|
|
struct scan_control sc = {
|
2011-05-27 06:25:25 +07:00
|
|
|
.nr_scanned = 0,
|
2010-08-11 08:03:02 +07:00
|
|
|
.nr_to_reclaim = SWAP_CLUSTER_MAX,
|
2009-09-24 05:56:39 +07:00
|
|
|
.may_writepage = !laptop_mode,
|
|
|
|
.may_unmap = 1,
|
|
|
|
.may_swap = !noswap,
|
|
|
|
.order = 0,
|
2012-01-13 08:18:32 +07:00
|
|
|
.target_mem_cgroup = memcg,
|
2009-09-24 05:56:39 +07:00
|
|
|
};
|
2012-01-13 08:17:59 +07:00
|
|
|
struct mem_cgroup_zone mz = {
|
2012-01-13 08:18:32 +07:00
|
|
|
.mem_cgroup = memcg,
|
2012-01-13 08:17:59 +07:00
|
|
|
.zone = zone,
|
|
|
|
};
|
2011-05-27 06:25:25 +07:00
|
|
|
|
2009-09-24 05:56:39 +07:00
|
|
|
sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
|
|
|
|
(GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
|
2010-08-10 07:19:56 +07:00
|
|
|
|
|
|
|
trace_mm_vmscan_memcg_softlimit_reclaim_begin(0,
|
|
|
|
sc.may_writepage,
|
|
|
|
sc.gfp_mask);
|
|
|
|
|
2009-09-24 05:56:39 +07:00
|
|
|
/*
|
|
|
|
* NOTE: Although we can get the priority field, using it
|
|
|
|
* here is not a good idea, since it limits the pages we can scan.
|
|
|
|
* if we don't reclaim here, the shrink_zone from balance_pgdat
|
|
|
|
* will pick up pages from other mem cgroup's as well. We hack
|
|
|
|
* the priority and make it zero.
|
|
|
|
*/
|
2012-01-13 08:17:59 +07:00
|
|
|
shrink_mem_cgroup_zone(0, &mz, &sc);
|
2010-08-10 07:19:56 +07:00
|
|
|
|
|
|
|
trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed);
|
|
|
|
|
2011-05-27 06:25:25 +07:00
|
|
|
*nr_scanned = sc.nr_scanned;
|
2009-09-24 05:56:39 +07:00
|
|
|
return sc.nr_reclaimed;
|
|
|
|
}
|
|
|
|
|
2012-01-13 08:18:32 +07:00
|
|
|
unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
|
2009-01-08 09:08:24 +07:00
|
|
|
gfp_t gfp_mask,
|
2011-09-15 06:21:58 +07:00
|
|
|
bool noswap)
|
2008-02-07 15:13:56 +07:00
|
|
|
{
|
2009-09-24 05:56:39 +07:00
|
|
|
struct zonelist *zonelist;
|
2010-08-10 07:19:56 +07:00
|
|
|
unsigned long nr_reclaimed;
|
2011-05-27 06:25:33 +07:00
|
|
|
int nid;
|
2008-02-07 15:13:56 +07:00
|
|
|
struct scan_control sc = {
|
|
|
|
.may_writepage = !laptop_mode,
|
2009-04-01 05:19:30 +07:00
|
|
|
.may_unmap = 1,
|
2009-04-22 02:24:57 +07:00
|
|
|
.may_swap = !noswap,
|
2009-12-15 08:59:10 +07:00
|
|
|
.nr_to_reclaim = SWAP_CLUSTER_MAX,
|
2008-02-07 15:13:56 +07:00
|
|
|
.order = 0,
|
2012-01-13 08:18:32 +07:00
|
|
|
.target_mem_cgroup = memcg,
|
2009-04-01 05:23:31 +07:00
|
|
|
.nodemask = NULL, /* we don't care the placement */
|
2011-05-25 07:12:26 +07:00
|
|
|
.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
|
|
|
|
(GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK),
|
|
|
|
};
|
|
|
|
struct shrink_control shrink = {
|
|
|
|
.gfp_mask = sc.gfp_mask,
|
2008-02-07 15:13:56 +07:00
|
|
|
};
|
|
|
|
|
2011-05-27 06:25:33 +07:00
|
|
|
/*
|
|
|
|
* Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't
|
|
|
|
* take care of from where we get pages. So the node where we start the
|
|
|
|
* scan does not need to be the current node.
|
|
|
|
*/
|
2012-01-13 08:18:32 +07:00
|
|
|
nid = mem_cgroup_select_victim_node(memcg);
|
2011-05-27 06:25:33 +07:00
|
|
|
|
|
|
|
zonelist = NODE_DATA(nid)->node_zonelists;
|
2010-08-10 07:19:56 +07:00
|
|
|
|
|
|
|
trace_mm_vmscan_memcg_reclaim_begin(0,
|
|
|
|
sc.may_writepage,
|
|
|
|
sc.gfp_mask);
|
|
|
|
|
2011-05-25 07:12:26 +07:00
|
|
|
nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
|
2010-08-10 07:19:56 +07:00
|
|
|
|
|
|
|
trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed);
|
|
|
|
|
|
|
|
return nr_reclaimed;
|
2008-02-07 15:13:56 +07:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2012-01-13 08:17:52 +07:00
|
|
|
static void age_active_anon(struct zone *zone, struct scan_control *sc,
|
|
|
|
int priority)
|
|
|
|
{
|
2012-01-13 08:18:06 +07:00
|
|
|
struct mem_cgroup *memcg;
|
2012-01-13 08:17:52 +07:00
|
|
|
|
2012-01-13 08:18:06 +07:00
|
|
|
if (!total_swap_pages)
|
|
|
|
return;
|
|
|
|
|
|
|
|
memcg = mem_cgroup_iter(NULL, NULL, NULL);
|
|
|
|
do {
|
|
|
|
struct mem_cgroup_zone mz = {
|
|
|
|
.mem_cgroup = memcg,
|
|
|
|
.zone = zone,
|
|
|
|
};
|
|
|
|
|
|
|
|
if (inactive_anon_is_low(&mz))
|
|
|
|
shrink_active_list(SWAP_CLUSTER_MAX, &mz,
|
|
|
|
sc, priority, 0);
|
|
|
|
|
|
|
|
memcg = mem_cgroup_iter(NULL, memcg, NULL);
|
|
|
|
} while (memcg);
|
2012-01-13 08:17:52 +07:00
|
|
|
}
|
|
|
|
|
2011-01-14 06:46:21 +07:00
|
|
|
/*
|
|
|
|
* pgdat_balanced is used when checking if a node is balanced for high-order
|
|
|
|
* allocations. Only zones that meet watermarks and are in a zone allowed
|
|
|
|
* by the callers classzone_idx are added to balanced_pages. The total of
|
|
|
|
* balanced pages must be at least 25% of the zones allowed by classzone_idx
|
|
|
|
* for the node to be considered balanced. Forcing all zones to be balanced
|
|
|
|
* for high orders can cause excessive reclaim when there are imbalanced zones.
|
|
|
|
* The choice of 25% is due to
|
|
|
|
* o a 16M DMA zone that is balanced will not balance a zone on any
|
|
|
|
* reasonable sized machine
|
|
|
|
* o On all other machines, the top zone must be at least a reasonable
|
2011-03-31 08:57:33 +07:00
|
|
|
* percentage of the middle zones. For example, on 32-bit x86, highmem
|
2011-01-14 06:46:21 +07:00
|
|
|
* would need to be at least 256M for it to be balance a whole node.
|
|
|
|
* Similarly, on x86-64 the Normal zone would need to be at least 1G
|
|
|
|
* to balance a node on its own. These seemed like reasonable ratios.
|
|
|
|
*/
|
|
|
|
static bool pgdat_balanced(pg_data_t *pgdat, unsigned long balanced_pages,
|
|
|
|
int classzone_idx)
|
|
|
|
{
|
|
|
|
unsigned long present_pages = 0;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i <= classzone_idx; i++)
|
|
|
|
present_pages += pgdat->node_zones[i].present_pages;
|
|
|
|
|
2011-07-19 22:49:26 +07:00
|
|
|
/* A special case here: if zone has no page, we think it's balanced */
|
|
|
|
return balanced_pages >= (present_pages >> 2);
|
2011-01-14 06:46:21 +07:00
|
|
|
}
|
|
|
|
|
2009-12-15 08:58:53 +07:00
|
|
|
/* is kswapd sleeping prematurely? */
|
2011-01-14 06:46:26 +07:00
|
|
|
static bool sleeping_prematurely(pg_data_t *pgdat, int order, long remaining,
|
|
|
|
int classzone_idx)
|
2009-12-15 08:58:53 +07:00
|
|
|
{
|
2009-12-15 08:58:55 +07:00
|
|
|
int i;
|
2011-01-14 06:46:21 +07:00
|
|
|
unsigned long balanced = 0;
|
|
|
|
bool all_zones_ok = true;
|
2009-12-15 08:58:53 +07:00
|
|
|
|
|
|
|
/* If a direct reclaimer woke kswapd within HZ/10, it's premature */
|
|
|
|
if (remaining)
|
2011-01-14 06:46:26 +07:00
|
|
|
return true;
|
2009-12-15 08:58:53 +07:00
|
|
|
|
2011-01-14 06:46:22 +07:00
|
|
|
/* Check the watermark levels */
|
mm: vmscan: correct check for kswapd sleeping in sleeping_prematurely
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
This seems to happen most with recent sandybridge laptops but it's
probably a co-incidence as some of these laptops just happen to have a
small Normal zone. The reproduction case is almost always during copying
large files that kswapd pegs at 100% CPU until the file is deleted or
cache is dropped.
The problem is mostly down to sleeping_prematurely() keeping kswapd awake
when the highest zone is small and unreclaimable and compounded by the
fact we shrink slabs even when not shrinking zones causing a lot of time
to be spent in shrinkers and a lot of memory to be reclaimed.
Patch 1 corrects sleeping_prematurely to check the zones matching
the classzone_idx instead of all zones.
Patch 2 avoids shrinking slab when we are not shrinking a zone.
Patch 3 notes that sleeping_prematurely is checking lower zones against
a high classzone which is not what allocators or balance_pgdat()
is doing leading to an artifical belief that kswapd should be
still awake.
Patch 4 notes that when balance_pgdat() gives up on a high zone that the
decision is not communicated to sleeping_prematurely()
This problem affects 2.6.38.8 for certain and is expected to affect 2.6.39
and 3.0-rc4 as well. If accepted, they need to go to -stable to be picked
up by distros and this series is against 3.0-rc4. I've cc'd people that
reported similar problems recently to see if they still suffer from the
problem and if this fixes it.
This patch: correct the check for kswapd sleeping in sleeping_prematurely()
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour.
A problem occurs if the highest zone is small. balance_pgdat() only
considers unreclaimable zones when priority is DEF_PRIORITY but
sleeping_prematurely considers all zones. It's possible for this sequence
to occur
1. kswapd wakes up and enters balance_pgdat()
2. At DEF_PRIORITY, marks highest zone unreclaimable
3. At DEF_PRIORITY-1, ignores highest zone setting end_zone
4. At DEF_PRIORITY-1, calls shrink_slab freeing memory from
highest zone, clearing all_unreclaimable. Highest zone
is still unbalanced
5. kswapd returns and calls sleeping_prematurely
6. sleeping_prematurely looks at *all* zones, not just the ones
being considered by balance_pgdat. The highest small zone
has all_unreclaimable cleared but the zone is not
balanced. all_zones_ok is false so kswapd stays awake
This patch corrects the behaviour of sleeping_prematurely to check the
zones balance_pgdat() checked.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:36 +07:00
|
|
|
for (i = 0; i <= classzone_idx; i++) {
|
2009-12-15 08:58:55 +07:00
|
|
|
struct zone *zone = pgdat->node_zones + i;
|
|
|
|
|
|
|
|
if (!populated_zone(zone))
|
|
|
|
continue;
|
|
|
|
|
2011-01-14 06:46:24 +07:00
|
|
|
/*
|
|
|
|
* balance_pgdat() skips over all_unreclaimable after
|
|
|
|
* DEF_PRIORITY. Effectively, it considers them balanced so
|
|
|
|
* they must be considered balanced here as well if kswapd
|
|
|
|
* is to sleep
|
|
|
|
*/
|
|
|
|
if (zone->all_unreclaimable) {
|
|
|
|
balanced += zone->present_pages;
|
2010-01-16 08:01:25 +07:00
|
|
|
continue;
|
2011-01-14 06:46:24 +07:00
|
|
|
}
|
2010-01-16 08:01:25 +07:00
|
|
|
|
mm: page allocator: adjust the per-cpu counter threshold when memory is low
Commit aa45484 ("calculate a better estimate of NR_FREE_PAGES when memory
is low") noted that watermarks were based on the vmstat NR_FREE_PAGES. To
avoid synchronization overhead, these counters are maintained on a per-cpu
basis and drained both periodically and when a threshold is above a
threshold. On large CPU systems, the difference between the estimate and
real value of NR_FREE_PAGES can be very high. The system can get into a
case where pages are allocated far below the min watermark potentially
causing livelock issues. The commit solved the problem by taking a better
reading of NR_FREE_PAGES when memory was low.
Unfortately, as reported by Shaohua Li this accurate reading can consume a
large amount of CPU time on systems with many sockets due to cache line
bouncing. This patch takes a different approach. For large machines
where counter drift might be unsafe and while kswapd is awake, the per-cpu
thresholds for the target pgdat are reduced to limit the level of drift to
what should be a safe level. This incurs a performance penalty in heavy
memory pressure by a factor that depends on the workload and the machine
but the machine should function correctly without accidentally exhausting
all memory on a node. There is an additional cost when kswapd wakes and
sleeps but the event is not expected to be frequent - in Shaohua's test
case, there was one recorded sleep and wake event at least.
To ensure that kswapd wakes up, a safe version of zone_watermark_ok() is
introduced that takes a more accurate reading of NR_FREE_PAGES when called
from wakeup_kswapd, when deciding whether it is really safe to go back to
sleep in sleeping_prematurely() and when deciding if a zone is really
balanced or not in balance_pgdat(). We are still using an expensive
function but limiting how often it is called.
When the test case is reproduced, the time spent in the watermark
functions is reduced. The following report is on the percentage of time
spent cumulatively spent in the functions zone_nr_free_pages(),
zone_watermark_ok(), __zone_watermark_ok(), zone_watermark_ok_safe(),
zone_page_state_snapshot(), zone_page_state().
vanilla 11.6615%
disable-threshold 0.2584%
David said:
: We had to pull aa454840 "mm: page allocator: calculate a better estimate
: of NR_FREE_PAGES when memory is low and kswapd is awake" from 2.6.36
: internally because tests showed that it would cause the machine to stall
: as the result of heavy kswapd activity. I merged it back with this fix as
: it is pending in the -mm tree and it solves the issue we were seeing, so I
: definitely think this should be pushed to -stable (and I would seriously
: consider it for 2.6.37 inclusion even at this late date).
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reported-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Tested-by: Nicolas Bareil <nico@chdir.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: <stable@kernel.org> [2.6.37.1, 2.6.36.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:45:41 +07:00
|
|
|
if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone),
|
2011-07-09 05:39:39 +07:00
|
|
|
i, 0))
|
2011-01-14 06:46:21 +07:00
|
|
|
all_zones_ok = false;
|
|
|
|
else
|
|
|
|
balanced += zone->present_pages;
|
2009-12-15 08:58:55 +07:00
|
|
|
}
|
2009-12-15 08:58:53 +07:00
|
|
|
|
2011-01-14 06:46:21 +07:00
|
|
|
/*
|
|
|
|
* For high-order requests, the balanced zones must contain at least
|
|
|
|
* 25% of the nodes pages for kswapd to sleep. For order-0, all zones
|
|
|
|
* must be balanced
|
|
|
|
*/
|
|
|
|
if (order)
|
2011-05-25 07:11:09 +07:00
|
|
|
return !pgdat_balanced(pgdat, balanced, classzone_idx);
|
2011-01-14 06:46:21 +07:00
|
|
|
else
|
|
|
|
return !all_zones_ok;
|
2009-12-15 08:58:53 +07:00
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* For kswapd, balance_pgdat() will work across all this node's zones until
|
2009-06-17 05:32:12 +07:00
|
|
|
* they are all at high_wmark_pages(zone).
|
2005-04-17 05:20:36 +07:00
|
|
|
*
|
2011-01-14 06:46:22 +07:00
|
|
|
* Returns the final order kswapd was reclaiming at
|
2005-04-17 05:20:36 +07:00
|
|
|
*
|
|
|
|
* There is special handling here for zones which are full of pinned pages.
|
|
|
|
* This can happen if the pages are all mlocked, or if they are all used by
|
|
|
|
* device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb.
|
|
|
|
* What we do is to detect the case where all pages in the zone have been
|
|
|
|
* scanned twice and there has been zero successful reclaim. Mark the zone as
|
|
|
|
* dead and from now on, only perform a short scan. Basically we're polling
|
|
|
|
* the zone for when the problem goes away.
|
|
|
|
*
|
|
|
|
* kswapd scans the zones in the highmem->normal->dma direction. It skips
|
2009-06-17 05:32:12 +07:00
|
|
|
* zones which have free_pages > high_wmark_pages(zone), but once a zone is
|
|
|
|
* found to have free_pages <= high_wmark_pages(zone), we scan that zone and the
|
|
|
|
* lower zones regardless of the number of free pages in the lower zones. This
|
|
|
|
* interoperates with the page allocator fallback scheme to ensure that aging
|
|
|
|
* of pages is balanced across the zones.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
static unsigned long balance_pgdat(pg_data_t *pgdat, int order,
|
2011-01-14 06:46:26 +07:00
|
|
|
int *classzone_idx)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
int all_zones_ok;
|
2011-01-14 06:46:21 +07:00
|
|
|
unsigned long balanced;
|
2005-04-17 05:20:36 +07:00
|
|
|
int priority;
|
|
|
|
int i;
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */
|
2006-03-22 15:08:19 +07:00
|
|
|
unsigned long total_scanned;
|
2005-04-17 05:20:36 +07:00
|
|
|
struct reclaim_state *reclaim_state = current->reclaim_state;
|
2011-05-27 06:25:25 +07:00
|
|
|
unsigned long nr_soft_reclaimed;
|
|
|
|
unsigned long nr_soft_scanned;
|
2006-03-22 15:08:18 +07:00
|
|
|
struct scan_control sc = {
|
|
|
|
.gfp_mask = GFP_KERNEL,
|
2009-04-01 05:19:30 +07:00
|
|
|
.may_unmap = 1,
|
2009-04-22 02:24:57 +07:00
|
|
|
.may_swap = 1,
|
2009-12-15 08:59:10 +07:00
|
|
|
/*
|
|
|
|
* kswapd doesn't want to be bailed out while reclaim. because
|
|
|
|
* we want to put equal scanning pressure on each zone.
|
|
|
|
*/
|
|
|
|
.nr_to_reclaim = ULONG_MAX,
|
2007-07-17 18:03:16 +07:00
|
|
|
.order = order,
|
2012-01-13 08:17:52 +07:00
|
|
|
.target_mem_cgroup = NULL,
|
2006-03-22 15:08:18 +07:00
|
|
|
};
|
2011-05-25 07:12:26 +07:00
|
|
|
struct shrink_control shrink = {
|
|
|
|
.gfp_mask = sc.gfp_mask,
|
|
|
|
};
|
2005-04-17 05:20:36 +07:00
|
|
|
loop_again:
|
|
|
|
total_scanned = 0;
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
sc.nr_reclaimed = 0;
|
2006-06-12 05:22:26 +07:00
|
|
|
sc.may_writepage = !laptop_mode;
|
2006-06-30 15:55:45 +07:00
|
|
|
count_vm_event(PAGEOUTRUN);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
for (priority = DEF_PRIORITY; priority >= 0; priority--) {
|
|
|
|
unsigned long lru_pages = 0;
|
2009-12-15 08:58:55 +07:00
|
|
|
int has_under_min_watermark_zone = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-11-29 04:44:07 +07:00
|
|
|
/* The swap token gets in the way of swapout... */
|
|
|
|
if (!priority)
|
2011-06-16 05:08:13 +07:00
|
|
|
disable_swap_token(NULL);
|
2005-11-29 04:44:07 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
all_zones_ok = 1;
|
2011-01-14 06:46:21 +07:00
|
|
|
balanced = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-06-23 16:03:18 +07:00
|
|
|
/*
|
|
|
|
* Scan in the highmem->dma direction for the highest
|
|
|
|
* zone which needs scanning
|
|
|
|
*/
|
|
|
|
for (i = pgdat->nr_zones - 1; i >= 0; i--) {
|
|
|
|
struct zone *zone = pgdat->node_zones + i;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-06-23 16:03:18 +07:00
|
|
|
if (!populated_zone(zone))
|
|
|
|
continue;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2010-03-06 04:41:55 +07:00
|
|
|
if (zone->all_unreclaimable && priority != DEF_PRIORITY)
|
2006-06-23 16:03:18 +07:00
|
|
|
continue;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-10-19 10:26:34 +07:00
|
|
|
/*
|
|
|
|
* Do some background aging of the anon list, to give
|
|
|
|
* pages a chance to be referenced before reclaiming.
|
|
|
|
*/
|
2012-01-13 08:17:52 +07:00
|
|
|
age_active_anon(zone, &sc, priority);
|
2008-10-19 10:26:34 +07:00
|
|
|
|
2012-03-22 06:34:00 +07:00
|
|
|
/*
|
|
|
|
* If the number of buffer_heads in the machine
|
|
|
|
* exceeds the maximum allowed level and this node
|
|
|
|
* has a highmem zone, force kswapd to reclaim from
|
|
|
|
* it to relieve lowmem pressure.
|
|
|
|
*/
|
|
|
|
if (buffer_heads_over_limit && is_highmem_idx(i)) {
|
|
|
|
end_zone = i;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
mm: page allocator: adjust the per-cpu counter threshold when memory is low
Commit aa45484 ("calculate a better estimate of NR_FREE_PAGES when memory
is low") noted that watermarks were based on the vmstat NR_FREE_PAGES. To
avoid synchronization overhead, these counters are maintained on a per-cpu
basis and drained both periodically and when a threshold is above a
threshold. On large CPU systems, the difference between the estimate and
real value of NR_FREE_PAGES can be very high. The system can get into a
case where pages are allocated far below the min watermark potentially
causing livelock issues. The commit solved the problem by taking a better
reading of NR_FREE_PAGES when memory was low.
Unfortately, as reported by Shaohua Li this accurate reading can consume a
large amount of CPU time on systems with many sockets due to cache line
bouncing. This patch takes a different approach. For large machines
where counter drift might be unsafe and while kswapd is awake, the per-cpu
thresholds for the target pgdat are reduced to limit the level of drift to
what should be a safe level. This incurs a performance penalty in heavy
memory pressure by a factor that depends on the workload and the machine
but the machine should function correctly without accidentally exhausting
all memory on a node. There is an additional cost when kswapd wakes and
sleeps but the event is not expected to be frequent - in Shaohua's test
case, there was one recorded sleep and wake event at least.
To ensure that kswapd wakes up, a safe version of zone_watermark_ok() is
introduced that takes a more accurate reading of NR_FREE_PAGES when called
from wakeup_kswapd, when deciding whether it is really safe to go back to
sleep in sleeping_prematurely() and when deciding if a zone is really
balanced or not in balance_pgdat(). We are still using an expensive
function but limiting how often it is called.
When the test case is reproduced, the time spent in the watermark
functions is reduced. The following report is on the percentage of time
spent cumulatively spent in the functions zone_nr_free_pages(),
zone_watermark_ok(), __zone_watermark_ok(), zone_watermark_ok_safe(),
zone_page_state_snapshot(), zone_page_state().
vanilla 11.6615%
disable-threshold 0.2584%
David said:
: We had to pull aa454840 "mm: page allocator: calculate a better estimate
: of NR_FREE_PAGES when memory is low and kswapd is awake" from 2.6.36
: internally because tests showed that it would cause the machine to stall
: as the result of heavy kswapd activity. I merged it back with this fix as
: it is pending in the -mm tree and it solves the issue we were seeing, so I
: definitely think this should be pushed to -stable (and I would seriously
: consider it for 2.6.37 inclusion even at this late date).
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reported-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Tested-by: Nicolas Bareil <nico@chdir.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: <stable@kernel.org> [2.6.37.1, 2.6.36.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:45:41 +07:00
|
|
|
if (!zone_watermark_ok_safe(zone, order,
|
2009-06-17 05:32:12 +07:00
|
|
|
high_wmark_pages(zone), 0, 0)) {
|
2006-06-23 16:03:18 +07:00
|
|
|
end_zone = i;
|
2006-12-07 11:32:01 +07:00
|
|
|
break;
|
2011-08-26 05:59:12 +07:00
|
|
|
} else {
|
|
|
|
/* If balanced, clear the congested flag */
|
|
|
|
zone_clear_flag(zone, ZONE_CONGESTED);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
}
|
2006-12-07 11:32:01 +07:00
|
|
|
if (i < 0)
|
|
|
|
goto out;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
for (i = 0; i <= end_zone; i++) {
|
|
|
|
struct zone *zone = pgdat->node_zones + i;
|
|
|
|
|
2009-09-22 07:01:42 +07:00
|
|
|
lru_pages += zone_reclaimable_pages(zone);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now scan the zone in the dma->highmem direction, stopping
|
|
|
|
* at the last zone which needs scanning.
|
|
|
|
*
|
|
|
|
* We do this because the page allocator works in the opposite
|
|
|
|
* direction. This prevents the page allocator from allocating
|
|
|
|
* pages behind kswapd's direction of progress, which would
|
|
|
|
* cause too much scanning of the lower zones.
|
|
|
|
*/
|
|
|
|
for (i = 0; i <= end_zone; i++) {
|
|
|
|
struct zone *zone = pgdat->node_zones + i;
|
2012-03-22 06:33:51 +07:00
|
|
|
int nr_slab, testorder;
|
mm: vmscan: kswapd should not free an excessive number of pages when balancing small zones
When reclaiming for order-0 pages, kswapd requires that all zones be
balanced. Each cycle through balance_pgdat() does background ageing on
all zones if necessary and applies equal pressure on the inactive zone
unless a lot of pages are free already.
A "lot of free pages" is defined as a "balance gap" above the high
watermark which is currently 7*high_watermark. Historically this was
reasonable as min_free_kbytes was small. However, on systems using huge
pages, it is recommended that min_free_kbytes is higher and it is tuned
with hugeadm --set-recommended-min_free_kbytes. With the introduction of
transparent huge page support, this recommended value is also applied. On
X86-64 with 4G of memory, min_free_kbytes becomes 67584 so one would
expect around 68M of memory to be free. The Normal zone is approximately
35000 pages so under even normal memory pressure such as copying a large
file, it gets exhausted quickly. As it is getting exhausted, kswapd
applies pressure equally to all zones, including the DMA32 zone. DMA32 is
approximately 700,000 pages with a high watermark of around 23,000 pages.
In this situation, kswapd will reclaim around (23000*8 where 8 is the high
watermark + balance gap of 7 * high watermark) pages or 718M of pages
before the zone is ignored. What the user sees is that free memory far
higher than it should be.
To avoid an excessive number of pages being reclaimed from the larger
zones, explicitely defines the "balance gap" to be either 1% of the zone
or the low watermark for the zone, whichever is smaller. While kswapd
will check all zones to apply pressure, it'll ignore zones that meets the
(high_wmark + balance_gap) watermark.
To test this, 80G were copied from a partition and the amount of memory
being used was recorded. A comparison of a patch and unpatched kernel can
be seen at
http://www.csn.ul.ie/~mel/postings/minfree-20110222/memory-usage-hydra.ps
and shows that kswapd is not reclaiming as much memory with the patch
applied.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: "Chen, Tim C" <tim.c.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 06:33:04 +07:00
|
|
|
unsigned long balance_gap;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-01-06 15:11:15 +07:00
|
|
|
if (!populated_zone(zone))
|
2005-04-17 05:20:36 +07:00
|
|
|
continue;
|
|
|
|
|
2010-03-06 04:41:55 +07:00
|
|
|
if (zone->all_unreclaimable && priority != DEF_PRIORITY)
|
2005-04-17 05:20:36 +07:00
|
|
|
continue;
|
|
|
|
|
|
|
|
sc.nr_scanned = 0;
|
2009-09-24 05:56:39 +07:00
|
|
|
|
2011-05-27 06:25:25 +07:00
|
|
|
nr_soft_scanned = 0;
|
2009-09-24 05:56:39 +07:00
|
|
|
/*
|
|
|
|
* Call soft limit reclaim before calling shrink_zone.
|
|
|
|
*/
|
2011-05-27 06:25:25 +07:00
|
|
|
nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone,
|
|
|
|
order, sc.gfp_mask,
|
|
|
|
&nr_soft_scanned);
|
|
|
|
sc.nr_reclaimed += nr_soft_reclaimed;
|
|
|
|
total_scanned += nr_soft_scanned;
|
2010-08-11 08:03:05 +07:00
|
|
|
|
mm: prevent kswapd from freeing excessive amounts of lowmem
The current VM can get itself into trouble fairly easily on systems with a
small ZONE_HIGHMEM, which is common on i686 computers with 1GB of memory.
On one side, page_alloc() will allocate down to zone->pages_low, while on
the other side, kswapd() and balance_pgdat() will try to free memory from
every zone, until every zone has more free pages than zone->pages_high.
Highmem can be filled up to zone->pages_low with page tables, ramfs,
vmalloc allocations and other unswappable things quite easily and without
many bad side effects, since we still have a huge ZONE_NORMAL to do future
allocations from.
However, as long as the number of free pages in the highmem zone is below
zone->pages_high, kswapd will continue swapping things out from
ZONE_NORMAL, too!
Sami Farin managed to get his system into a stage where kswapd had freed
about 700MB of low memory and was still "going strong".
The attached patch will make kswapd stop paging out data from zones when
there is more than enough memory free. We do go above zone->pages_high in
order to keep pressure between zones equal in normal circumstances, but the
patch should prevent the kind of excesses that made Sami's computer totally
unusable.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 15:24:50 +07:00
|
|
|
/*
|
mm: vmscan: kswapd should not free an excessive number of pages when balancing small zones
When reclaiming for order-0 pages, kswapd requires that all zones be
balanced. Each cycle through balance_pgdat() does background ageing on
all zones if necessary and applies equal pressure on the inactive zone
unless a lot of pages are free already.
A "lot of free pages" is defined as a "balance gap" above the high
watermark which is currently 7*high_watermark. Historically this was
reasonable as min_free_kbytes was small. However, on systems using huge
pages, it is recommended that min_free_kbytes is higher and it is tuned
with hugeadm --set-recommended-min_free_kbytes. With the introduction of
transparent huge page support, this recommended value is also applied. On
X86-64 with 4G of memory, min_free_kbytes becomes 67584 so one would
expect around 68M of memory to be free. The Normal zone is approximately
35000 pages so under even normal memory pressure such as copying a large
file, it gets exhausted quickly. As it is getting exhausted, kswapd
applies pressure equally to all zones, including the DMA32 zone. DMA32 is
approximately 700,000 pages with a high watermark of around 23,000 pages.
In this situation, kswapd will reclaim around (23000*8 where 8 is the high
watermark + balance gap of 7 * high watermark) pages or 718M of pages
before the zone is ignored. What the user sees is that free memory far
higher than it should be.
To avoid an excessive number of pages being reclaimed from the larger
zones, explicitely defines the "balance gap" to be either 1% of the zone
or the low watermark for the zone, whichever is smaller. While kswapd
will check all zones to apply pressure, it'll ignore zones that meets the
(high_wmark + balance_gap) watermark.
To test this, 80G were copied from a partition and the amount of memory
being used was recorded. A comparison of a patch and unpatched kernel can
be seen at
http://www.csn.ul.ie/~mel/postings/minfree-20110222/memory-usage-hydra.ps
and shows that kswapd is not reclaiming as much memory with the patch
applied.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: "Chen, Tim C" <tim.c.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 06:33:04 +07:00
|
|
|
* We put equal pressure on every zone, unless
|
|
|
|
* one zone has way too many pages free
|
|
|
|
* already. The "too many pages" is defined
|
|
|
|
* as the high wmark plus a "gap" where the
|
|
|
|
* gap is either the low watermark or 1%
|
|
|
|
* of the zone, whichever is smaller.
|
mm: prevent kswapd from freeing excessive amounts of lowmem
The current VM can get itself into trouble fairly easily on systems with a
small ZONE_HIGHMEM, which is common on i686 computers with 1GB of memory.
On one side, page_alloc() will allocate down to zone->pages_low, while on
the other side, kswapd() and balance_pgdat() will try to free memory from
every zone, until every zone has more free pages than zone->pages_high.
Highmem can be filled up to zone->pages_low with page tables, ramfs,
vmalloc allocations and other unswappable things quite easily and without
many bad side effects, since we still have a huge ZONE_NORMAL to do future
allocations from.
However, as long as the number of free pages in the highmem zone is below
zone->pages_high, kswapd will continue swapping things out from
ZONE_NORMAL, too!
Sami Farin managed to get his system into a stage where kswapd had freed
about 700MB of low memory and was still "going strong".
The attached patch will make kswapd stop paging out data from zones when
there is more than enough memory free. We do go above zone->pages_high in
order to keep pressure between zones equal in normal circumstances, but the
patch should prevent the kind of excesses that made Sami's computer totally
unusable.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 15:24:50 +07:00
|
|
|
*/
|
mm: vmscan: kswapd should not free an excessive number of pages when balancing small zones
When reclaiming for order-0 pages, kswapd requires that all zones be
balanced. Each cycle through balance_pgdat() does background ageing on
all zones if necessary and applies equal pressure on the inactive zone
unless a lot of pages are free already.
A "lot of free pages" is defined as a "balance gap" above the high
watermark which is currently 7*high_watermark. Historically this was
reasonable as min_free_kbytes was small. However, on systems using huge
pages, it is recommended that min_free_kbytes is higher and it is tuned
with hugeadm --set-recommended-min_free_kbytes. With the introduction of
transparent huge page support, this recommended value is also applied. On
X86-64 with 4G of memory, min_free_kbytes becomes 67584 so one would
expect around 68M of memory to be free. The Normal zone is approximately
35000 pages so under even normal memory pressure such as copying a large
file, it gets exhausted quickly. As it is getting exhausted, kswapd
applies pressure equally to all zones, including the DMA32 zone. DMA32 is
approximately 700,000 pages with a high watermark of around 23,000 pages.
In this situation, kswapd will reclaim around (23000*8 where 8 is the high
watermark + balance gap of 7 * high watermark) pages or 718M of pages
before the zone is ignored. What the user sees is that free memory far
higher than it should be.
To avoid an excessive number of pages being reclaimed from the larger
zones, explicitely defines the "balance gap" to be either 1% of the zone
or the low watermark for the zone, whichever is smaller. While kswapd
will check all zones to apply pressure, it'll ignore zones that meets the
(high_wmark + balance_gap) watermark.
To test this, 80G were copied from a partition and the amount of memory
being used was recorded. A comparison of a patch and unpatched kernel can
be seen at
http://www.csn.ul.ie/~mel/postings/minfree-20110222/memory-usage-hydra.ps
and shows that kswapd is not reclaiming as much memory with the patch
applied.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: "Chen, Tim C" <tim.c.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 06:33:04 +07:00
|
|
|
balance_gap = min(low_wmark_pages(zone),
|
|
|
|
(zone->present_pages +
|
|
|
|
KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
|
|
|
|
KSWAPD_ZONE_BALANCE_GAP_RATIO);
|
2012-03-22 06:33:51 +07:00
|
|
|
/*
|
|
|
|
* Kswapd reclaims only single pages with compaction
|
|
|
|
* enabled. Trying too hard to reclaim until contiguous
|
|
|
|
* free pages have become available can hurt performance
|
|
|
|
* by evicting too much useful data from memory.
|
|
|
|
* Do not reclaim more than needed for compaction.
|
|
|
|
*/
|
|
|
|
testorder = order;
|
|
|
|
if (COMPACTION_BUILD && order &&
|
|
|
|
compaction_suitable(zone, order) !=
|
|
|
|
COMPACT_SKIPPED)
|
|
|
|
testorder = 0;
|
|
|
|
|
2012-03-22 06:34:00 +07:00
|
|
|
if ((buffer_heads_over_limit && is_highmem_idx(i)) ||
|
2012-03-23 16:57:31 +07:00
|
|
|
!zone_watermark_ok_safe(zone, testorder,
|
mm: vmscan: kswapd should not free an excessive number of pages when balancing small zones
When reclaiming for order-0 pages, kswapd requires that all zones be
balanced. Each cycle through balance_pgdat() does background ageing on
all zones if necessary and applies equal pressure on the inactive zone
unless a lot of pages are free already.
A "lot of free pages" is defined as a "balance gap" above the high
watermark which is currently 7*high_watermark. Historically this was
reasonable as min_free_kbytes was small. However, on systems using huge
pages, it is recommended that min_free_kbytes is higher and it is tuned
with hugeadm --set-recommended-min_free_kbytes. With the introduction of
transparent huge page support, this recommended value is also applied. On
X86-64 with 4G of memory, min_free_kbytes becomes 67584 so one would
expect around 68M of memory to be free. The Normal zone is approximately
35000 pages so under even normal memory pressure such as copying a large
file, it gets exhausted quickly. As it is getting exhausted, kswapd
applies pressure equally to all zones, including the DMA32 zone. DMA32 is
approximately 700,000 pages with a high watermark of around 23,000 pages.
In this situation, kswapd will reclaim around (23000*8 where 8 is the high
watermark + balance gap of 7 * high watermark) pages or 718M of pages
before the zone is ignored. What the user sees is that free memory far
higher than it should be.
To avoid an excessive number of pages being reclaimed from the larger
zones, explicitely defines the "balance gap" to be either 1% of the zone
or the low watermark for the zone, whichever is smaller. While kswapd
will check all zones to apply pressure, it'll ignore zones that meets the
(high_wmark + balance_gap) watermark.
To test this, 80G were copied from a partition and the amount of memory
being used was recorded. A comparison of a patch and unpatched kernel can
be seen at
http://www.csn.ul.ie/~mel/postings/minfree-20110222/memory-usage-hydra.ps
and shows that kswapd is not reclaiming as much memory with the patch
applied.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: "Chen, Tim C" <tim.c.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 06:33:04 +07:00
|
|
|
high_wmark_pages(zone) + balance_gap,
|
2011-07-09 05:39:38 +07:00
|
|
|
end_zone, 0)) {
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
shrink_zone(priority, zone, &sc);
|
2011-01-14 06:47:11 +07:00
|
|
|
|
2011-07-09 05:39:38 +07:00
|
|
|
reclaim_state->reclaimed_slab = 0;
|
|
|
|
nr_slab = shrink_slab(&shrink, sc.nr_scanned, lru_pages);
|
|
|
|
sc.nr_reclaimed += reclaim_state->reclaimed_slab;
|
|
|
|
total_scanned += sc.nr_scanned;
|
|
|
|
|
|
|
|
if (nr_slab == 0 && !zone_reclaimable(zone))
|
|
|
|
zone->all_unreclaimable = 1;
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* If we've done a decent amount of scanning and
|
|
|
|
* the reclaim ratio is low, start doing writepage
|
|
|
|
* even in laptop mode
|
|
|
|
*/
|
|
|
|
if (total_scanned > SWAP_CLUSTER_MAX * 2 &&
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
total_scanned > sc.nr_reclaimed + sc.nr_reclaimed / 2)
|
2005-04-17 05:20:36 +07:00
|
|
|
sc.may_writepage = 1;
|
2009-12-15 08:58:55 +07:00
|
|
|
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
if (zone->all_unreclaimable) {
|
|
|
|
if (end_zone && end_zone == i)
|
|
|
|
end_zone--;
|
2011-07-09 05:39:38 +07:00
|
|
|
continue;
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
}
|
2011-07-09 05:39:38 +07:00
|
|
|
|
2012-03-22 06:33:51 +07:00
|
|
|
if (!zone_watermark_ok_safe(zone, testorder,
|
2010-03-06 04:41:45 +07:00
|
|
|
high_wmark_pages(zone), end_zone, 0)) {
|
|
|
|
all_zones_ok = 0;
|
|
|
|
/*
|
|
|
|
* We are still under min water mark. This
|
|
|
|
* means that we have a GFP_ATOMIC allocation
|
|
|
|
* failure risk. Hurry up!
|
|
|
|
*/
|
mm: page allocator: adjust the per-cpu counter threshold when memory is low
Commit aa45484 ("calculate a better estimate of NR_FREE_PAGES when memory
is low") noted that watermarks were based on the vmstat NR_FREE_PAGES. To
avoid synchronization overhead, these counters are maintained on a per-cpu
basis and drained both periodically and when a threshold is above a
threshold. On large CPU systems, the difference between the estimate and
real value of NR_FREE_PAGES can be very high. The system can get into a
case where pages are allocated far below the min watermark potentially
causing livelock issues. The commit solved the problem by taking a better
reading of NR_FREE_PAGES when memory was low.
Unfortately, as reported by Shaohua Li this accurate reading can consume a
large amount of CPU time on systems with many sockets due to cache line
bouncing. This patch takes a different approach. For large machines
where counter drift might be unsafe and while kswapd is awake, the per-cpu
thresholds for the target pgdat are reduced to limit the level of drift to
what should be a safe level. This incurs a performance penalty in heavy
memory pressure by a factor that depends on the workload and the machine
but the machine should function correctly without accidentally exhausting
all memory on a node. There is an additional cost when kswapd wakes and
sleeps but the event is not expected to be frequent - in Shaohua's test
case, there was one recorded sleep and wake event at least.
To ensure that kswapd wakes up, a safe version of zone_watermark_ok() is
introduced that takes a more accurate reading of NR_FREE_PAGES when called
from wakeup_kswapd, when deciding whether it is really safe to go back to
sleep in sleeping_prematurely() and when deciding if a zone is really
balanced or not in balance_pgdat(). We are still using an expensive
function but limiting how often it is called.
When the test case is reproduced, the time spent in the watermark
functions is reduced. The following report is on the percentage of time
spent cumulatively spent in the functions zone_nr_free_pages(),
zone_watermark_ok(), __zone_watermark_ok(), zone_watermark_ok_safe(),
zone_page_state_snapshot(), zone_page_state().
vanilla 11.6615%
disable-threshold 0.2584%
David said:
: We had to pull aa454840 "mm: page allocator: calculate a better estimate
: of NR_FREE_PAGES when memory is low and kswapd is awake" from 2.6.36
: internally because tests showed that it would cause the machine to stall
: as the result of heavy kswapd activity. I merged it back with this fix as
: it is pending in the -mm tree and it solves the issue we were seeing, so I
: definitely think this should be pushed to -stable (and I would seriously
: consider it for 2.6.37 inclusion even at this late date).
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reported-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Tested-by: Nicolas Bareil <nico@chdir.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: <stable@kernel.org> [2.6.37.1, 2.6.36.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:45:41 +07:00
|
|
|
if (!zone_watermark_ok_safe(zone, order,
|
2010-03-06 04:41:45 +07:00
|
|
|
min_wmark_pages(zone), end_zone, 0))
|
|
|
|
has_under_min_watermark_zone = 1;
|
2010-10-27 04:21:45 +07:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* If a zone reaches its high watermark,
|
|
|
|
* consider it to be no longer congested. It's
|
|
|
|
* possible there are dirty pages backed by
|
|
|
|
* congested BDIs but as pressure is relieved,
|
|
|
|
* spectulatively avoid congestion waits
|
|
|
|
*/
|
|
|
|
zone_clear_flag(zone, ZONE_CONGESTED);
|
2011-01-14 06:46:26 +07:00
|
|
|
if (i <= *classzone_idx)
|
2011-01-14 06:46:21 +07:00
|
|
|
balanced += zone->present_pages;
|
2010-03-06 04:41:45 +07:00
|
|
|
}
|
2009-12-15 08:58:55 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2011-01-14 06:46:26 +07:00
|
|
|
if (all_zones_ok || (order && pgdat_balanced(pgdat, balanced, *classzone_idx)))
|
2005-04-17 05:20:36 +07:00
|
|
|
break; /* kswapd: all done */
|
|
|
|
/*
|
|
|
|
* OK, kswapd is getting into trouble. Take a nap, then take
|
|
|
|
* another pass across the zones.
|
|
|
|
*/
|
2009-12-15 08:58:55 +07:00
|
|
|
if (total_scanned && (priority < DEF_PRIORITY - 2)) {
|
|
|
|
if (has_under_min_watermark_zone)
|
|
|
|
count_vm_event(KSWAPD_SKIP_CONGESTION_WAIT);
|
|
|
|
else
|
|
|
|
congestion_wait(BLK_RW_ASYNC, HZ/10);
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* We do this so kswapd doesn't build up large priorities for
|
|
|
|
* example when it is freeing in parallel with allocators. It
|
|
|
|
* matches the direct reclaim path behaviour in terms of impact
|
|
|
|
* on zone->*_priority.
|
|
|
|
*/
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
if (sc.nr_reclaimed >= SWAP_CLUSTER_MAX)
|
2005-04-17 05:20:36 +07:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
out:
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* order-0: All zones must meet high watermark for a balanced node
|
2011-01-14 06:46:21 +07:00
|
|
|
* high-order: Balanced zones must make up at least 25% of the node
|
|
|
|
* for the node to be balanced
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
*/
|
2011-01-14 06:46:26 +07:00
|
|
|
if (!(all_zones_ok || (order && pgdat_balanced(pgdat, balanced, *classzone_idx)))) {
|
2005-04-17 05:20:36 +07:00
|
|
|
cond_resched();
|
[PATCH] swsusp: Improve handling of highmem
Currently swsusp saves the contents of highmem pages by copying them to the
normal zone which is quite inefficient (eg. it requires two normal pages
to be used for saving one highmem page). This may be improved by using
highmem for saving the contents of saveable highmem pages.
Namely, during the suspend phase of the suspend-resume cycle we try to
allocate as many free highmem pages as there are saveable highmem pages.
If there are not enough highmem image pages to store the contents of all of
the saveable highmem pages, some of them will be stored in the "normal"
memory. Next, we allocate as many free "normal" pages as needed to store
the (remaining) image data. We use a memory bitmap to mark the allocated
free pages (ie. highmem as well as "normal" image pages).
Now, we use another memory bitmap to mark all of the saveable pages
(highmem as well as "normal") and the contents of the saveable pages are
copied into the image pages. Then, the second bitmap is used to save the
pfns corresponding to the saveable pages and the first one is used to save
their data.
During the resume phase the pfns of the pages that were saveable during the
suspend are loaded from the image and used to mark the "unsafe" page
frames. Next, we try to allocate as many free highmem page frames as to
load all of the image data that had been in the highmem before the suspend
and we allocate so many free "normal" page frames that the total number of
allocated free pages (highmem and "normal") is equal to the size of the
image. While doing this we have to make sure that there will be some extra
free "normal" and "safe" page frames for two lists of PBEs constructed
later.
Now, the image data are loaded, if possible, into their "original" page
frames. The image data that cannot be written into their "original" page
frames are loaded into "safe" page frames and their "original" kernel
virtual addresses, as well as the addresses of the "safe" pages containing
their copies, are stored in one of two lists of PBEs.
One list of PBEs is for the copies of "normal" suspend pages (ie. "normal"
pages that were saveable during the suspend) and it is used in the same way
as previously (ie. by the architecture-dependent parts of swsusp). The
other list of PBEs is for the copies of highmem suspend pages. The pages
in this list are restored (in a reversible way) right before the
arch-dependent code is called.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 11:34:18 +07:00
|
|
|
|
|
|
|
try_to_freeze();
|
|
|
|
|
2009-01-07 05:40:33 +07:00
|
|
|
/*
|
|
|
|
* Fragmentation may mean that the system cannot be
|
|
|
|
* rebalanced for high-order allocations in all zones.
|
|
|
|
* At this point, if nr_reclaimed < SWAP_CLUSTER_MAX,
|
|
|
|
* it means the zones have been fully scanned and are still
|
|
|
|
* not balanced. For high-order allocations, there is
|
|
|
|
* little point trying all over again as kswapd may
|
|
|
|
* infinite loop.
|
|
|
|
*
|
|
|
|
* Instead, recheck all watermarks at order-0 as they
|
|
|
|
* are the most important. If watermarks are ok, kswapd will go
|
|
|
|
* back to sleep. High-order users can still perform direct
|
|
|
|
* reclaim if they wish.
|
|
|
|
*/
|
|
|
|
if (sc.nr_reclaimed < SWAP_CLUSTER_MAX)
|
|
|
|
order = sc.order = 0;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
goto loop_again;
|
|
|
|
}
|
|
|
|
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
/*
|
|
|
|
* If kswapd was reclaiming at a higher order, it has the option of
|
|
|
|
* sleeping without all zones being balanced. Before it does, it must
|
|
|
|
* ensure that the watermarks for order-0 on *all* zones are met and
|
|
|
|
* that the congestion flags are cleared. The congestion flag must
|
|
|
|
* be cleared as kswapd is the only mechanism that clears the flag
|
|
|
|
* and it is potentially going to sleep here.
|
|
|
|
*/
|
|
|
|
if (order) {
|
2012-03-22 06:33:52 +07:00
|
|
|
int zones_need_compaction = 1;
|
|
|
|
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
for (i = 0; i <= end_zone; i++) {
|
|
|
|
struct zone *zone = pgdat->node_zones + i;
|
|
|
|
|
|
|
|
if (!populated_zone(zone))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (zone->all_unreclaimable && priority != DEF_PRIORITY)
|
|
|
|
continue;
|
|
|
|
|
2012-03-22 06:33:51 +07:00
|
|
|
/* Would compaction fail due to lack of free memory? */
|
2012-03-24 21:26:21 +07:00
|
|
|
if (COMPACTION_BUILD &&
|
|
|
|
compaction_suitable(zone, order) == COMPACT_SKIPPED)
|
2012-03-22 06:33:51 +07:00
|
|
|
goto loop_again;
|
|
|
|
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
/* Confirm the zone is balanced for order-0 */
|
|
|
|
if (!zone_watermark_ok(zone, 0,
|
|
|
|
high_wmark_pages(zone), 0, 0)) {
|
|
|
|
order = sc.order = 0;
|
|
|
|
goto loop_again;
|
|
|
|
}
|
|
|
|
|
2012-03-22 06:33:52 +07:00
|
|
|
/* Check if the memory needs to be defragmented. */
|
|
|
|
if (zone_watermark_ok(zone, order,
|
|
|
|
low_wmark_pages(zone), *classzone_idx, 0))
|
|
|
|
zones_need_compaction = 0;
|
|
|
|
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
/* If balanced, clear the congested flag */
|
|
|
|
zone_clear_flag(zone, ZONE_CONGESTED);
|
|
|
|
}
|
2012-03-22 06:33:52 +07:00
|
|
|
|
|
|
|
if (zones_need_compaction)
|
|
|
|
compact_pgdat(pgdat, order);
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
}
|
|
|
|
|
2011-01-14 06:46:22 +07:00
|
|
|
/*
|
|
|
|
* Return the order we were reclaiming at so sleeping_prematurely()
|
|
|
|
* makes a decision on the order we were last reclaiming at. However,
|
|
|
|
* if another caller entered the allocator slow path while kswapd
|
|
|
|
* was awake, order will remain at the higher level
|
|
|
|
*/
|
2011-01-14 06:46:26 +07:00
|
|
|
*classzone_idx = end_zone;
|
2011-01-14 06:46:22 +07:00
|
|
|
return order;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2011-01-14 06:46:26 +07:00
|
|
|
static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx)
|
2011-01-14 06:45:50 +07:00
|
|
|
{
|
|
|
|
long remaining = 0;
|
|
|
|
DEFINE_WAIT(wait);
|
|
|
|
|
|
|
|
if (freezing(current) || kthread_should_stop())
|
|
|
|
return;
|
|
|
|
|
|
|
|
prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
|
|
|
|
|
|
|
|
/* Try to sleep for a short interval */
|
2011-01-14 06:46:26 +07:00
|
|
|
if (!sleeping_prematurely(pgdat, order, remaining, classzone_idx)) {
|
2011-01-14 06:45:50 +07:00
|
|
|
remaining = schedule_timeout(HZ/10);
|
|
|
|
finish_wait(&pgdat->kswapd_wait, &wait);
|
|
|
|
prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* After a short sleep, check if it was a premature sleep. If not, then
|
|
|
|
* go fully to sleep until explicitly woken up.
|
|
|
|
*/
|
2011-01-14 06:46:26 +07:00
|
|
|
if (!sleeping_prematurely(pgdat, order, remaining, classzone_idx)) {
|
2011-01-14 06:45:50 +07:00
|
|
|
trace_mm_vmscan_kswapd_sleep(pgdat->node_id);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vmstat counters are not perfectly accurate and the estimated
|
|
|
|
* value for counters such as NR_FREE_PAGES can deviate from the
|
|
|
|
* true value by nr_online_cpus * threshold. To avoid the zone
|
|
|
|
* watermarks being breached while under pressure, we reduce the
|
|
|
|
* per-cpu vmstat threshold while kswapd is awake and restore
|
|
|
|
* them before going back to sleep.
|
|
|
|
*/
|
|
|
|
set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold);
|
|
|
|
schedule();
|
|
|
|
set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold);
|
|
|
|
} else {
|
|
|
|
if (remaining)
|
|
|
|
count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY);
|
|
|
|
else
|
|
|
|
count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY);
|
|
|
|
}
|
|
|
|
finish_wait(&pgdat->kswapd_wait, &wait);
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* The background pageout daemon, started as a kernel thread
|
2008-10-19 10:26:32 +07:00
|
|
|
* from the init process.
|
2005-04-17 05:20:36 +07:00
|
|
|
*
|
|
|
|
* This basically trickles out pages so that we have _some_
|
|
|
|
* free memory available even if there is no other activity
|
|
|
|
* that frees anything up. This is needed for things like routing
|
|
|
|
* etc, where we otherwise might have all activity going on in
|
|
|
|
* asynchronous contexts that cannot page things out.
|
|
|
|
*
|
|
|
|
* If there are applications that are active memory-allocators
|
|
|
|
* (most normal use), this basically shouldn't matter.
|
|
|
|
*/
|
|
|
|
static int kswapd(void *p)
|
|
|
|
{
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
unsigned long order, new_order;
|
kswapd: avoid unnecessary rebalance after an unsuccessful balancing
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing. But in the following scenario, kswapd do something that
is not matched our expectation. The patch fixes this issue.
1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
While the expectation behavior of kswapd should try to sleep.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:39 +07:00
|
|
|
unsigned balanced_order;
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
int classzone_idx, new_classzone_idx;
|
kswapd: avoid unnecessary rebalance after an unsuccessful balancing
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing. But in the following scenario, kswapd do something that
is not matched our expectation. The patch fixes this issue.
1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
While the expectation behavior of kswapd should try to sleep.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:39 +07:00
|
|
|
int balanced_classzone_idx;
|
2005-04-17 05:20:36 +07:00
|
|
|
pg_data_t *pgdat = (pg_data_t*)p;
|
|
|
|
struct task_struct *tsk = current;
|
2011-01-14 06:45:50 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
struct reclaim_state reclaim_state = {
|
|
|
|
.reclaimed_slab = 0,
|
|
|
|
};
|
2009-03-13 11:19:46 +07:00
|
|
|
const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
lockdep: annotate reclaim context (__GFP_NOFS)
Here is another version, with the incremental patch rolled up, and
added reclaim context annotation to kswapd, and allocation tracing
to slab allocators (which may only ever reach the page allocator
in rare cases, so it is good to put annotations here too).
Haven't tested this version as such, but it should be getting closer
to merge worthy ;)
--
After noticing some code in mm/filemap.c accidentally perform a __GFP_FS
allocation when it should not have been, I thought it might be a good idea to
try to catch this kind of thing with lockdep.
I coded up a little idea that seems to work. Unfortunately the system has to
actually be in __GFP_FS page reclaim, then take the lock, before it will mark
it. But at least that might still be some orders of magnitude more common
(and more debuggable) than an actual deadlock condition, so we have some
improvement I hope (the concept is no less complete than discovery of a lock's
interrupt contexts).
I guess we could even do the same thing with __GFP_IO (normal reclaim), and
even GFP_NOIO locks too... but filesystems will have the most locks and fiddly
code paths, so let's start there and see how it goes.
It *seems* to work. I did a quick test.
=================================
[ INFO: inconsistent lock state ]
2.6.28-rc6-00007-ged31348-dirty #26
---------------------------------
inconsistent {in-reclaim-W} -> {ov-reclaim-W} usage.
modprobe/8526 [HC0[0]:SC0[0]:HE1:SE1] takes:
(testlock){--..}, at: [<ffffffffa0020055>] brd_init+0x55/0x216 [brd]
{in-reclaim-W} state was registered at:
[<ffffffff80267bdb>] __lock_acquire+0x75b/0x1a60
[<ffffffff80268f71>] lock_acquire+0x91/0xc0
[<ffffffff8070f0e1>] mutex_lock_nested+0xb1/0x310
[<ffffffffa002002b>] brd_init+0x2b/0x216 [brd]
[<ffffffff8020903b>] _stext+0x3b/0x170
[<ffffffff80272ebf>] sys_init_module+0xaf/0x1e0
[<ffffffff8020c3fb>] system_call_fastpath+0x16/0x1b
[<ffffffffffffffff>] 0xffffffffffffffff
irq event stamp: 3929
hardirqs last enabled at (3929): [<ffffffff8070f2b5>] mutex_lock_nested+0x285/0x310
hardirqs last disabled at (3928): [<ffffffff8070f089>] mutex_lock_nested+0x59/0x310
softirqs last enabled at (3732): [<ffffffff8061f623>] sk_filter+0x83/0xe0
softirqs last disabled at (3730): [<ffffffff8061f5b6>] sk_filter+0x16/0xe0
other info that might help us debug this:
1 lock held by modprobe/8526:
#0: (testlock){--..}, at: [<ffffffffa0020055>] brd_init+0x55/0x216 [brd]
stack backtrace:
Pid: 8526, comm: modprobe Not tainted 2.6.28-rc6-00007-ged31348-dirty #26
Call Trace:
[<ffffffff80265483>] print_usage_bug+0x193/0x1d0
[<ffffffff80266530>] mark_lock+0xaf0/0xca0
[<ffffffff80266735>] mark_held_locks+0x55/0xc0
[<ffffffffa0020000>] ? brd_init+0x0/0x216 [brd]
[<ffffffff802667ca>] trace_reclaim_fs+0x2a/0x60
[<ffffffff80285005>] __alloc_pages_internal+0x475/0x580
[<ffffffff8070f29e>] ? mutex_lock_nested+0x26e/0x310
[<ffffffffa0020000>] ? brd_init+0x0/0x216 [brd]
[<ffffffffa002006a>] brd_init+0x6a/0x216 [brd]
[<ffffffffa0020000>] ? brd_init+0x0/0x216 [brd]
[<ffffffff8020903b>] _stext+0x3b/0x170
[<ffffffff8070f8b9>] ? mutex_unlock+0x9/0x10
[<ffffffff8070f83d>] ? __mutex_unlock_slowpath+0x10d/0x180
[<ffffffff802669ec>] ? trace_hardirqs_on_caller+0x12c/0x190
[<ffffffff80272ebf>] sys_init_module+0xaf/0x1e0
[<ffffffff8020c3fb>] system_call_fastpath+0x16/0x1b
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-01-21 14:12:39 +07:00
|
|
|
lockdep_set_current_reclaim_state(GFP_KERNEL);
|
|
|
|
|
2009-01-01 06:42:29 +07:00
|
|
|
if (!cpumask_empty(cpumask))
|
2008-04-05 08:11:10 +07:00
|
|
|
set_cpus_allowed_ptr(tsk, cpumask);
|
2005-04-17 05:20:36 +07:00
|
|
|
current->reclaim_state = &reclaim_state;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Tell the memory management that we're a "memory allocator",
|
|
|
|
* and that if we need more memory we should get access to it
|
|
|
|
* regardless (see "__alloc_pages()"). "kswapd" should
|
|
|
|
* never get caught in the normal page freeing logic.
|
|
|
|
*
|
|
|
|
* (Kswapd normally doesn't need memory anyway, but sometimes
|
|
|
|
* you need a small amount of memory in order to be able to
|
|
|
|
* page out something else, and this flag essentially protects
|
|
|
|
* us from recursively trying to free more memory as we're
|
|
|
|
* trying to free the first piece of memory in the first place).
|
|
|
|
*/
|
2006-01-08 16:00:47 +07:00
|
|
|
tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
|
2007-07-17 18:03:35 +07:00
|
|
|
set_freezable();
|
2005-04-17 05:20:36 +07:00
|
|
|
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
order = new_order = 0;
|
kswapd: avoid unnecessary rebalance after an unsuccessful balancing
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing. But in the following scenario, kswapd do something that
is not matched our expectation. The patch fixes this issue.
1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
While the expectation behavior of kswapd should try to sleep.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:39 +07:00
|
|
|
balanced_order = 0;
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
classzone_idx = new_classzone_idx = pgdat->nr_zones - 1;
|
kswapd: avoid unnecessary rebalance after an unsuccessful balancing
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing. But in the following scenario, kswapd do something that
is not matched our expectation. The patch fixes this issue.
1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
While the expectation behavior of kswapd should try to sleep.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:39 +07:00
|
|
|
balanced_classzone_idx = classzone_idx;
|
2005-04-17 05:20:36 +07:00
|
|
|
for ( ; ; ) {
|
2009-12-15 08:58:33 +07:00
|
|
|
int ret;
|
2005-06-25 13:13:50 +07:00
|
|
|
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
/*
|
|
|
|
* If the last balance_pgdat was unsuccessful it's unlikely a
|
|
|
|
* new request of a similar or harder type will succeed soon
|
|
|
|
* so consider going to sleep on the basis we reclaimed at
|
|
|
|
*/
|
kswapd: avoid unnecessary rebalance after an unsuccessful balancing
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing. But in the following scenario, kswapd do something that
is not matched our expectation. The patch fixes this issue.
1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
While the expectation behavior of kswapd should try to sleep.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:39 +07:00
|
|
|
if (balanced_classzone_idx >= new_classzone_idx &&
|
|
|
|
balanced_order == new_order) {
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
new_order = pgdat->kswapd_max_order;
|
|
|
|
new_classzone_idx = pgdat->classzone_idx;
|
|
|
|
pgdat->kswapd_max_order = 0;
|
|
|
|
pgdat->classzone_idx = pgdat->nr_zones - 1;
|
|
|
|
}
|
|
|
|
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
if (order < new_order || classzone_idx > new_classzone_idx) {
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* Don't sleep if someone wants a larger 'order'
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
* allocation or has tigher zone constraints
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
|
|
|
order = new_order;
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
classzone_idx = new_classzone_idx;
|
2005-04-17 05:20:36 +07:00
|
|
|
} else {
|
kswapd: avoid unnecessary rebalance after an unsuccessful balancing
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing. But in the following scenario, kswapd do something that
is not matched our expectation. The patch fixes this issue.
1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
While the expectation behavior of kswapd should try to sleep.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:39 +07:00
|
|
|
kswapd_try_to_sleep(pgdat, balanced_order,
|
|
|
|
balanced_classzone_idx);
|
2005-04-17 05:20:36 +07:00
|
|
|
order = pgdat->kswapd_max_order;
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
classzone_idx = pgdat->classzone_idx;
|
kswapd: assign new_order and new_classzone_idx after wakeup in sleeping
There 2 places to read pgdat in kswapd. One is return from a successful
balance, another is waked up from kswapd sleeping. The new_order and
new_classzone_idx represent the balance input order and classzone_idx.
But current new_order and new_classzone_idx are not assigned after
kswapd_try_to_sleep(), that will cause a bug in the following scenario.
1: after a successful balance, kswapd goes to sleep, and new_order = 0;
new_classzone_idx = __MAX_NR_ZONES - 1;
2: kswapd waked up with order = 3 and classzone_idx = ZONE_NORMAL
3: in the balance_pgdat() running, a new balance wakeup happened with
order = 5, and classzone_idx = ZONE_NORMAL
4: the first wakeup(order = 3) finished successufly, return order = 3
but, the new_order is still 0, so, this balancing will be treated as a
failed balance. And then the second tighter balancing will be missed.
So, to avoid the above problem, the new_order and new_classzone_idx need
to be assigned for later successful comparison.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:45 +07:00
|
|
|
new_order = order;
|
|
|
|
new_classzone_idx = classzone_idx;
|
2011-01-14 06:46:23 +07:00
|
|
|
pgdat->kswapd_max_order = 0;
|
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark. This
is expected behaviour. Unfortunately, if the highest zone is small, a
problem occurs.
When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable. Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1. It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again. As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.
There are two problems here. 1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling. 2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.
This patch does two things that are related. If the end_zone is
unreclaimable, this information is communicated back. Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep. Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-07-09 05:39:40 +07:00
|
|
|
pgdat->classzone_idx = pgdat->nr_zones - 1;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2009-12-15 08:58:33 +07:00
|
|
|
ret = try_to_freeze();
|
|
|
|
if (kthread_should_stop())
|
|
|
|
break;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We can speed up thawing tasks if we don't call balance_pgdat
|
|
|
|
* after returning from the refrigerator
|
|
|
|
*/
|
2010-08-10 07:19:16 +07:00
|
|
|
if (!ret) {
|
|
|
|
trace_mm_vmscan_kswapd_wake(pgdat->node_id, order);
|
kswapd: avoid unnecessary rebalance after an unsuccessful balancing
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing. But in the following scenario, kswapd do something that
is not matched our expectation. The patch fixes this issue.
1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
While the expectation behavior of kswapd should try to sleep.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-11-01 07:08:39 +07:00
|
|
|
balanced_classzone_idx = classzone_idx;
|
|
|
|
balanced_order = balance_pgdat(pgdat, order,
|
|
|
|
&balanced_classzone_idx);
|
2010-08-10 07:19:16 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* A zone is low on free memory, so wake its kswapd task to service it.
|
|
|
|
*/
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
pg_data_t *pgdat;
|
|
|
|
|
2006-01-06 15:11:15 +07:00
|
|
|
if (!populated_zone(zone))
|
2005-04-17 05:20:36 +07:00
|
|
|
return;
|
|
|
|
|
mm: page allocator: adjust the per-cpu counter threshold when memory is low
Commit aa45484 ("calculate a better estimate of NR_FREE_PAGES when memory
is low") noted that watermarks were based on the vmstat NR_FREE_PAGES. To
avoid synchronization overhead, these counters are maintained on a per-cpu
basis and drained both periodically and when a threshold is above a
threshold. On large CPU systems, the difference between the estimate and
real value of NR_FREE_PAGES can be very high. The system can get into a
case where pages are allocated far below the min watermark potentially
causing livelock issues. The commit solved the problem by taking a better
reading of NR_FREE_PAGES when memory was low.
Unfortately, as reported by Shaohua Li this accurate reading can consume a
large amount of CPU time on systems with many sockets due to cache line
bouncing. This patch takes a different approach. For large machines
where counter drift might be unsafe and while kswapd is awake, the per-cpu
thresholds for the target pgdat are reduced to limit the level of drift to
what should be a safe level. This incurs a performance penalty in heavy
memory pressure by a factor that depends on the workload and the machine
but the machine should function correctly without accidentally exhausting
all memory on a node. There is an additional cost when kswapd wakes and
sleeps but the event is not expected to be frequent - in Shaohua's test
case, there was one recorded sleep and wake event at least.
To ensure that kswapd wakes up, a safe version of zone_watermark_ok() is
introduced that takes a more accurate reading of NR_FREE_PAGES when called
from wakeup_kswapd, when deciding whether it is really safe to go back to
sleep in sleeping_prematurely() and when deciding if a zone is really
balanced or not in balance_pgdat(). We are still using an expensive
function but limiting how often it is called.
When the test case is reproduced, the time spent in the watermark
functions is reduced. The following report is on the percentage of time
spent cumulatively spent in the functions zone_nr_free_pages(),
zone_watermark_ok(), __zone_watermark_ok(), zone_watermark_ok_safe(),
zone_page_state_snapshot(), zone_page_state().
vanilla 11.6615%
disable-threshold 0.2584%
David said:
: We had to pull aa454840 "mm: page allocator: calculate a better estimate
: of NR_FREE_PAGES when memory is low and kswapd is awake" from 2.6.36
: internally because tests showed that it would cause the machine to stall
: as the result of heavy kswapd activity. I merged it back with this fix as
: it is pending in the -mm tree and it solves the issue we were seeing, so I
: definitely think this should be pushed to -stable (and I would seriously
: consider it for 2.6.37 inclusion even at this late date).
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reported-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Tested-by: Nicolas Bareil <nico@chdir.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: <stable@kernel.org> [2.6.37.1, 2.6.36.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:45:41 +07:00
|
|
|
if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
|
2005-04-17 05:20:36 +07:00
|
|
|
return;
|
mm: page allocator: adjust the per-cpu counter threshold when memory is low
Commit aa45484 ("calculate a better estimate of NR_FREE_PAGES when memory
is low") noted that watermarks were based on the vmstat NR_FREE_PAGES. To
avoid synchronization overhead, these counters are maintained on a per-cpu
basis and drained both periodically and when a threshold is above a
threshold. On large CPU systems, the difference between the estimate and
real value of NR_FREE_PAGES can be very high. The system can get into a
case where pages are allocated far below the min watermark potentially
causing livelock issues. The commit solved the problem by taking a better
reading of NR_FREE_PAGES when memory was low.
Unfortately, as reported by Shaohua Li this accurate reading can consume a
large amount of CPU time on systems with many sockets due to cache line
bouncing. This patch takes a different approach. For large machines
where counter drift might be unsafe and while kswapd is awake, the per-cpu
thresholds for the target pgdat are reduced to limit the level of drift to
what should be a safe level. This incurs a performance penalty in heavy
memory pressure by a factor that depends on the workload and the machine
but the machine should function correctly without accidentally exhausting
all memory on a node. There is an additional cost when kswapd wakes and
sleeps but the event is not expected to be frequent - in Shaohua's test
case, there was one recorded sleep and wake event at least.
To ensure that kswapd wakes up, a safe version of zone_watermark_ok() is
introduced that takes a more accurate reading of NR_FREE_PAGES when called
from wakeup_kswapd, when deciding whether it is really safe to go back to
sleep in sleeping_prematurely() and when deciding if a zone is really
balanced or not in balance_pgdat(). We are still using an expensive
function but limiting how often it is called.
When the test case is reproduced, the time spent in the watermark
functions is reduced. The following report is on the percentage of time
spent cumulatively spent in the functions zone_nr_free_pages(),
zone_watermark_ok(), __zone_watermark_ok(), zone_watermark_ok_safe(),
zone_page_state_snapshot(), zone_page_state().
vanilla 11.6615%
disable-threshold 0.2584%
David said:
: We had to pull aa454840 "mm: page allocator: calculate a better estimate
: of NR_FREE_PAGES when memory is low and kswapd is awake" from 2.6.36
: internally because tests showed that it would cause the machine to stall
: as the result of heavy kswapd activity. I merged it back with this fix as
: it is pending in the -mm tree and it solves the issue we were seeing, so I
: definitely think this should be pushed to -stable (and I would seriously
: consider it for 2.6.37 inclusion even at this late date).
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reported-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Tested-by: Nicolas Bareil <nico@chdir.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: <stable@kernel.org> [2.6.37.1, 2.6.36.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:45:41 +07:00
|
|
|
pgdat = zone->zone_pgdat;
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
if (pgdat->kswapd_max_order < order) {
|
2005-04-17 05:20:36 +07:00
|
|
|
pgdat->kswapd_max_order = order;
|
mm: kswapd: stop high-order balancing when any suitable zone is balanced
Simon Kirby reported the following problem
We're seeing cases on a number of servers where cache never fully
grows to use all available memory. Sometimes we see servers with 4 GB
of memory that never seem to have less than 1.5 GB free, even with a
constantly-active VM. In some cases, these servers also swap out while
this happens, even though they are constantly reading the working set
into memory. We have been seeing this happening for a long time; I
don't think it's anything recent, and it still happens on 2.6.36.
After some debugging work by Simon, Dave Hansen and others, the prevaling
theory became that kswapd is reclaiming order-3 pages requested by SLUB
too aggressive about it.
There are two apparent problems here. On the target machine, there is a
small Normal zone in comparison to DMA32. As kswapd tries to balance all
zones, it would continually try reclaiming for Normal even though DMA32
was balanced enough for callers. The second problem is that
sleeping_prematurely() does not use the same logic as balance_pgdat() when
deciding whether to sleep or not. This keeps kswapd artifically awake.
A number of tests were run and the figures from previous postings will
look very different for a few reasons. One, the old figures were forcing
my network card to use GFP_ATOMIC in attempt to replicate Simon's problem.
Second, I previous specified slub_min_order=3 again in an attempt to
reproduce Simon's problem. In this posting, I'm depending on Simon to say
whether his problem is fixed or not and these figures are to show the
impact to the ordinary cases. Finally, the "vmscan" figures are taken
from /proc/vmstat instead of the tracepoints. There is less information
but recording is less disruptive.
The first test of relevance was postmark with a process running in the
background reading a large amount of anonymous memory in blocks. The
objective was to vaguely simulate what was happening on Simon's machine
and it's memory intensive enough to have kswapd awake.
POSTMARK
traceonly kanyzone
Transactions per second: 156.00 ( 0.00%) 153.00 (-1.96%)
Data megabytes read per second: 21.51 ( 0.00%) 21.52 ( 0.05%)
Data megabytes written per second: 29.28 ( 0.00%) 29.11 (-0.58%)
Files created alone per second: 250.00 ( 0.00%) 416.00 (39.90%)
Files create/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
Files deleted alone per second: 520.00 ( 0.00%) 420.00 (-23.81%)
Files delete/transact per second: 79.00 ( 0.00%) 76.00 (-3.95%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds) 16.58 17.4
Total Elapsed Time (seconds) 218.48 222.47
VMstat Reclaim Statistics: vmscan
Direct reclaims 0 4
Direct reclaim pages scanned 0 203
Direct reclaim pages reclaimed 0 184
Kswapd pages scanned 326631 322018
Kswapd pages reclaimed 312632 309784
Kswapd low wmark quickly 1 4
Kswapd high wmark quickly 122 475
Kswapd skip congestion_wait 1 0
Pages activated 700040 705317
Pages deactivated 212113 203922
Pages written 9875 6363
Total pages scanned 326631 322221
Total pages reclaimed 312632 309968
%age total pages scanned/reclaimed 95.71% 96.20%
%age total pages scanned/written 3.02% 1.97%
proc vmstat: Faults
Major Faults 300 254
Minor Faults 645183 660284
Page ins 493588 486704
Page outs 4960088 4986704
Swap ins 1230 661
Swap outs 9869 6355
Performance is mildly affected because kswapd is no longer doing as much
work and the background memory consumer process is getting in the way.
Note that kswapd scanned and reclaimed fewer pages as it's less aggressive
and overall fewer pages were scanned and reclaimed. Swap in/out is
particularly reduced again reflecting kswapd throwing out fewer pages.
The slight performance impact is unfortunate here but it looks like a
direct result of kswapd being less aggressive. As the bug report is about
too many pages being freed by kswapd, it may have to be accepted for now.
The second test is a streaming IO benchmark that was previously used by
Johannes to show regressions in page reclaim.
MICRO
traceonly kanyzone
User/Sys Time Running Test (seconds) 29.29 28.87
Total Elapsed Time (seconds) 492.18 488.79
VMstat Reclaim Statistics: vmscan
Direct reclaims 2128 1460
Direct reclaim pages scanned 2284822 1496067
Direct reclaim pages reclaimed 148919 110937
Kswapd pages scanned 15450014 16202876
Kswapd pages reclaimed 8503697 8537897
Kswapd low wmark quickly 3100 3397
Kswapd high wmark quickly 1860 7243
Kswapd skip congestion_wait 708 801
Pages activated 9635 9573
Pages deactivated 1432 1271
Pages written 223 1130
Total pages scanned 17734836 17698943
Total pages reclaimed 8652616 8648834
%age total pages scanned/reclaimed 48.79% 48.87%
%age total pages scanned/written 0.00% 0.01%
proc vmstat: Faults
Major Faults 165 221
Minor Faults 9655785 9656506
Page ins 3880 7228
Page outs 37692940 37480076
Swap ins 0 69
Swap outs 19 15
Again fewer pages are scanned and reclaimed as expected and this time the
test completed faster. Note that kswapd is hitting its watermarks faster
(low and high wmark quickly) which I expect is due to kswapd reclaiming
fewer pages.
I also ran fs-mark, iozone and sysbench but there is nothing interesting
to report in the figures. Performance is not significantly changed and
the reclaim statistics look reasonable.
Tgis patch:
When the allocator enters its slow path, kswapd is woken up to balance the
node. It continues working until all zones within the node are balanced.
For order-0 allocations, this makes perfect sense but for higher orders it
can have unintended side-effects. If the zone sizes are imbalanced,
kswapd may reclaim heavily within a smaller zone discarding an excessive
number of pages. The user-visible behaviour is that kswapd is awake and
reclaiming even though plenty of pages are free from a suitable zone.
This patch alters the "balance" logic for high-order reclaim allowing
kswapd to stop if any suitable zone becomes balanced to reduce the number
of pages it reclaims from other zones. kswapd still tries to ensure that
order-0 watermarks for all zones are met before sleeping.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Eric B Munson <emunson@mgebm.net>
Cc: Simon Kirby <sim@hostway.ca>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:46:20 +07:00
|
|
|
pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx);
|
|
|
|
}
|
2005-09-13 15:25:07 +07:00
|
|
|
if (!waitqueue_active(&pgdat->kswapd_wait))
|
2005-04-17 05:20:36 +07:00
|
|
|
return;
|
mm: page allocator: adjust the per-cpu counter threshold when memory is low
Commit aa45484 ("calculate a better estimate of NR_FREE_PAGES when memory
is low") noted that watermarks were based on the vmstat NR_FREE_PAGES. To
avoid synchronization overhead, these counters are maintained on a per-cpu
basis and drained both periodically and when a threshold is above a
threshold. On large CPU systems, the difference between the estimate and
real value of NR_FREE_PAGES can be very high. The system can get into a
case where pages are allocated far below the min watermark potentially
causing livelock issues. The commit solved the problem by taking a better
reading of NR_FREE_PAGES when memory was low.
Unfortately, as reported by Shaohua Li this accurate reading can consume a
large amount of CPU time on systems with many sockets due to cache line
bouncing. This patch takes a different approach. For large machines
where counter drift might be unsafe and while kswapd is awake, the per-cpu
thresholds for the target pgdat are reduced to limit the level of drift to
what should be a safe level. This incurs a performance penalty in heavy
memory pressure by a factor that depends on the workload and the machine
but the machine should function correctly without accidentally exhausting
all memory on a node. There is an additional cost when kswapd wakes and
sleeps but the event is not expected to be frequent - in Shaohua's test
case, there was one recorded sleep and wake event at least.
To ensure that kswapd wakes up, a safe version of zone_watermark_ok() is
introduced that takes a more accurate reading of NR_FREE_PAGES when called
from wakeup_kswapd, when deciding whether it is really safe to go back to
sleep in sleeping_prematurely() and when deciding if a zone is really
balanced or not in balance_pgdat(). We are still using an expensive
function but limiting how often it is called.
When the test case is reproduced, the time spent in the watermark
functions is reduced. The following report is on the percentage of time
spent cumulatively spent in the functions zone_nr_free_pages(),
zone_watermark_ok(), __zone_watermark_ok(), zone_watermark_ok_safe(),
zone_page_state_snapshot(), zone_page_state().
vanilla 11.6615%
disable-threshold 0.2584%
David said:
: We had to pull aa454840 "mm: page allocator: calculate a better estimate
: of NR_FREE_PAGES when memory is low and kswapd is awake" from 2.6.36
: internally because tests showed that it would cause the machine to stall
: as the result of heavy kswapd activity. I merged it back with this fix as
: it is pending in the -mm tree and it solves the issue we were seeing, so I
: definitely think this should be pushed to -stable (and I would seriously
: consider it for 2.6.37 inclusion even at this late date).
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reported-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Tested-by: Nicolas Bareil <nico@chdir.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: <stable@kernel.org> [2.6.37.1, 2.6.36.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-01-14 06:45:41 +07:00
|
|
|
if (zone_watermark_ok_safe(zone, order, low_wmark_pages(zone), 0, 0))
|
|
|
|
return;
|
|
|
|
|
|
|
|
trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order);
|
2005-09-13 15:25:07 +07:00
|
|
|
wake_up_interruptible(&pgdat->kswapd_wait);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2009-09-22 07:01:42 +07:00
|
|
|
/*
|
|
|
|
* The reclaimable count would be mostly accurate.
|
|
|
|
* The less reclaimable pages may be
|
|
|
|
* - mlocked pages, which will be moved to unevictable list when encountered
|
|
|
|
* - mapped pages, which may require several travels to be reclaimed
|
|
|
|
* - dirty pages, which is not "instantly" reclaimable
|
|
|
|
*/
|
|
|
|
unsigned long global_reclaimable_pages(void)
|
2008-10-19 10:26:32 +07:00
|
|
|
{
|
2009-09-22 07:01:42 +07:00
|
|
|
int nr;
|
|
|
|
|
|
|
|
nr = global_page_state(NR_ACTIVE_FILE) +
|
|
|
|
global_page_state(NR_INACTIVE_FILE);
|
|
|
|
|
|
|
|
if (nr_swap_pages > 0)
|
|
|
|
nr += global_page_state(NR_ACTIVE_ANON) +
|
|
|
|
global_page_state(NR_INACTIVE_ANON);
|
|
|
|
|
|
|
|
return nr;
|
|
|
|
}
|
|
|
|
|
|
|
|
unsigned long zone_reclaimable_pages(struct zone *zone)
|
|
|
|
{
|
|
|
|
int nr;
|
|
|
|
|
|
|
|
nr = zone_page_state(zone, NR_ACTIVE_FILE) +
|
|
|
|
zone_page_state(zone, NR_INACTIVE_FILE);
|
|
|
|
|
|
|
|
if (nr_swap_pages > 0)
|
|
|
|
nr += zone_page_state(zone, NR_ACTIVE_ANON) +
|
|
|
|
zone_page_state(zone, NR_INACTIVE_ANON);
|
|
|
|
|
|
|
|
return nr;
|
2008-10-19 10:26:32 +07:00
|
|
|
}
|
|
|
|
|
2009-05-25 03:16:31 +07:00
|
|
|
#ifdef CONFIG_HIBERNATION
|
2005-04-17 05:20:36 +07:00
|
|
|
/*
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
* Try to free `nr_to_reclaim' of memory, system-wide, and return the number of
|
2006-06-23 16:03:18 +07:00
|
|
|
* freed pages.
|
|
|
|
*
|
|
|
|
* Rather than trying to age LRUs the aim is to preserve the overall
|
|
|
|
* LRU order by reclaiming preferentially
|
|
|
|
* inactive > active > active referenced > active mapped
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
unsigned long shrink_all_memory(unsigned long nr_to_reclaim)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2006-06-23 16:03:18 +07:00
|
|
|
struct reclaim_state reclaim_state;
|
|
|
|
struct scan_control sc = {
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
.gfp_mask = GFP_HIGHUSER_MOVABLE,
|
|
|
|
.may_swap = 1,
|
|
|
|
.may_unmap = 1,
|
2006-06-23 16:03:18 +07:00
|
|
|
.may_writepage = 1,
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
.nr_to_reclaim = nr_to_reclaim,
|
|
|
|
.hibernation_mode = 1,
|
|
|
|
.order = 0,
|
2005-04-17 05:20:36 +07:00
|
|
|
};
|
2011-05-25 07:12:26 +07:00
|
|
|
struct shrink_control shrink = {
|
|
|
|
.gfp_mask = sc.gfp_mask,
|
|
|
|
};
|
|
|
|
struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask);
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
struct task_struct *p = current;
|
|
|
|
unsigned long nr_reclaimed;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
p->flags |= PF_MEMALLOC;
|
|
|
|
lockdep_set_current_reclaim_state(sc.gfp_mask);
|
|
|
|
reclaim_state.reclaimed_slab = 0;
|
|
|
|
p->reclaim_state = &reclaim_state;
|
2006-06-23 16:03:18 +07:00
|
|
|
|
2011-05-25 07:12:26 +07:00
|
|
|
nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink);
|
2009-04-01 05:19:34 +07:00
|
|
|
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
p->reclaim_state = NULL;
|
|
|
|
lockdep_clear_current_reclaim_state();
|
|
|
|
p->flags &= ~PF_MEMALLOC;
|
2006-06-23 16:03:18 +07:00
|
|
|
|
vmscan: kill hibernation specific reclaim logic and unify it
shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
memory shrinker) for hibernate performance improvement. and
sc.swap_cluster_max was introduced by commit a06fe4d307 (Speed freeing
memory for suspend).
commit a06fe4d307 said
Without the patch:
Freed 14600 pages in 1749 jiffies = 32.61 MB/s (Anomolous!)
Freed 88563 pages in 14719 jiffies = 23.50 MB/s
Freed 205734 pages in 32389 jiffies = 24.81 MB/s
With the patch:
Freed 68252 pages in 496 jiffies = 537.52 MB/s
Freed 116464 pages in 569 jiffies = 798.54 MB/s
Freed 209699 pages in 705 jiffies = 1161.89 MB/s
At that time, their patch was pretty worth. However, Modern Hardware
trend and recent VM improvement broke its worth. From several reason, I
think we should remove shrink_all_zones() at all.
detail:
1) Old days, shrink_zone()'s slowness was mainly caused by stupid io-throttle
at no i/o congestion.
but current shrink_zone() is sane, not slow.
2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
fine on numa system.
example)
System has 4GB memory and each node have 2GB. and hibernate need 1GB.
optimal)
steal 500MB from each node.
shrink_all_zones)
steal 1GB from node-0.
Oh, Cache balancing logic was broken. ;)
Unfortunately, Desktop system moved ahead NUMA at nowadays.
(Side note, if hibernate require 2GB, shrink_all_zones() never success
on above machine)
3) if the node has several I/O flighting pages, shrink_all_zones() makes
pretty bad result.
schenario) hibernate need 1GB
1) shrink_all_zones() try to reclaim 1GB from Node-0
2) but it only reclaimed 990MB
3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
4) it reclaimed 990MB
Oh, well. it reclaimed twice much than required.
In the other hand, current shrink_zone() has sane baling out logic.
then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
Now, shrink_all_memory() is only the wrapper function of do_try_to_free_pages().
it bring good reviewability and debuggability, and solve above problems.
side note: Reclaim logic unificication makes two good side effect.
- Fix recursive reclaim bug on shrink_all_memory().
it did forgot to use PF_MEMALLOC. it mean the system be able to stuck into deadlock.
- Now, shrink_all_memory() got lockdep awareness. it bring good debuggability.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-15 08:59:12 +07:00
|
|
|
return nr_reclaimed;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2009-05-25 03:16:31 +07:00
|
|
|
#endif /* CONFIG_HIBERNATION */
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* It's optimal to keep kswapds on the same CPUs as their memory, but
|
|
|
|
not required for correctness. So if the last cpu in a node goes
|
|
|
|
away, we get changed to run anywhere: as the first one comes back,
|
|
|
|
restore their cpu bindings. */
|
2006-06-27 16:54:07 +07:00
|
|
|
static int __devinit cpu_callback(struct notifier_block *nfb,
|
2006-03-22 15:08:19 +07:00
|
|
|
unsigned long action, void *hcpu)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2007-10-16 15:25:40 +07:00
|
|
|
int nid;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-05-09 16:35:10 +07:00
|
|
|
if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) {
|
2007-10-16 15:25:40 +07:00
|
|
|
for_each_node_state(nid, N_HIGH_MEMORY) {
|
2008-04-05 08:11:10 +07:00
|
|
|
pg_data_t *pgdat = NODE_DATA(nid);
|
2009-03-13 11:19:46 +07:00
|
|
|
const struct cpumask *mask;
|
|
|
|
|
|
|
|
mask = cpumask_of_node(pgdat->node_id);
|
2008-04-05 08:11:10 +07:00
|
|
|
|
2009-01-01 06:42:24 +07:00
|
|
|
if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids)
|
2005-04-17 05:20:36 +07:00
|
|
|
/* One of our CPUs online: restore mask */
|
2008-04-05 08:11:10 +07:00
|
|
|
set_cpus_allowed_ptr(pgdat->kswapd, mask);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return NOTIFY_OK;
|
|
|
|
}
|
|
|
|
|
2006-06-27 16:53:33 +07:00
|
|
|
/*
|
|
|
|
* This kswapd start function will be called by init and node-hot-add.
|
|
|
|
* On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added.
|
|
|
|
*/
|
|
|
|
int kswapd_run(int nid)
|
|
|
|
{
|
|
|
|
pg_data_t *pgdat = NODE_DATA(nid);
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (pgdat->kswapd)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
|
|
|
|
if (IS_ERR(pgdat->kswapd)) {
|
|
|
|
/* failure at boot is fatal */
|
|
|
|
BUG_ON(system_state == SYSTEM_BOOTING);
|
|
|
|
printk("Failed to start kswapd on node %d\n",nid);
|
|
|
|
ret = -1;
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-12-15 08:58:33 +07:00
|
|
|
/*
|
|
|
|
* Called by memory hotplug when all memory in a node is offlined.
|
|
|
|
*/
|
|
|
|
void kswapd_stop(int nid)
|
|
|
|
{
|
|
|
|
struct task_struct *kswapd = NODE_DATA(nid)->kswapd;
|
|
|
|
|
|
|
|
if (kswapd)
|
|
|
|
kthread_stop(kswapd);
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
static int __init kswapd_init(void)
|
|
|
|
{
|
2006-06-27 16:53:33 +07:00
|
|
|
int nid;
|
2006-03-22 15:08:19 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
swap_setup();
|
2007-10-16 15:25:31 +07:00
|
|
|
for_each_node_state(nid, N_HIGH_MEMORY)
|
2006-06-27 16:53:33 +07:00
|
|
|
kswapd_run(nid);
|
2005-04-17 05:20:36 +07:00
|
|
|
hotcpu_notifier(cpu_callback, 0);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
module_init(kswapd_init)
|
2006-01-19 08:42:31 +07:00
|
|
|
|
|
|
|
#ifdef CONFIG_NUMA
|
|
|
|
/*
|
|
|
|
* Zone reclaim mode
|
|
|
|
*
|
|
|
|
* If non-zero call zone_reclaim when the number of free pages falls below
|
|
|
|
* the watermarks.
|
|
|
|
*/
|
|
|
|
int zone_reclaim_mode __read_mostly;
|
|
|
|
|
2006-02-01 18:05:34 +07:00
|
|
|
#define RECLAIM_OFF 0
|
2008-07-30 12:33:41 +07:00
|
|
|
#define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */
|
2006-02-01 18:05:34 +07:00
|
|
|
#define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */
|
|
|
|
#define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */
|
|
|
|
|
2006-02-01 18:05:32 +07:00
|
|
|
/*
|
|
|
|
* Priority for ZONE_RECLAIM. This determines the fraction of pages
|
|
|
|
* of a node considered for each zone_reclaim. 4 scans 1/16th of
|
|
|
|
* a zone.
|
|
|
|
*/
|
|
|
|
#define ZONE_RECLAIM_PRIORITY 4
|
|
|
|
|
2006-07-03 14:24:13 +07:00
|
|
|
/*
|
|
|
|
* Percentage of pages in a zone that must be unmapped for zone_reclaim to
|
|
|
|
* occur.
|
|
|
|
*/
|
|
|
|
int sysctl_min_unmapped_ratio = 1;
|
|
|
|
|
2006-09-26 13:31:52 +07:00
|
|
|
/*
|
|
|
|
* If the number of slab pages in a zone grows beyond this percentage then
|
|
|
|
* slab reclaim needs to occur.
|
|
|
|
*/
|
|
|
|
int sysctl_min_slab_ratio = 5;
|
|
|
|
|
vmscan: properly account for the number of page cache pages zone_reclaim() can reclaim
A bug was brought to my attention against a distro kernel but it affects
mainline and I believe problems like this have been reported in various
guises on the mailing lists although I don't have specific examples at the
moment.
The reported problem was that malloc() stalled for a long time (minutes in
some cases) if a large tmpfs mount was occupying a large percentage of
memory overall. The pages did not get cleaned or reclaimed by
zone_reclaim() because the zone_reclaim_mode was unsuitable, but the lists
are uselessly scanned frequencly making the CPU spin at near 100%.
This patchset intends to address that bug and bring the behaviour of
zone_reclaim() more in line with expectations which were noticed during
investigation. It is based on top of mmotm and takes advantage of
Kosaki's work with respect to zone_reclaim().
Patch 1 fixes the heuristics that zone_reclaim() uses to determine if the
scan should go ahead. The broken heuristic is what was causing the
malloc() stall as it uselessly scanned the LRU constantly. Currently,
zone_reclaim is assuming zone_reclaim_mode is 1 and historically it
could not deal with tmpfs pages at all. This fixes up the heuristic so
that an unnecessary scan is more likely to be correctly avoided.
Patch 2 notes that zone_reclaim() returning a failure automatically means
the zone is marked full. This is not always true. It could have
failed because the GFP mask or zone_reclaim_mode were unsuitable.
Patch 3 introduces a counter zreclaim_failed that will increment each
time the zone_reclaim scan-avoidance heuristics fail. If that
counter is rapidly increasing, then zone_reclaim_mode should be
set to 0 as a temporarily resolution and a bug reported because
the scan-avoidance heuristic is still broken.
This patch:
On NUMA machines, the administrator can configure zone_reclaim_mode that
is a more targetted form of direct reclaim. On machines with large NUMA
distances for example, a zone_reclaim_mode defaults to 1 meaning that
clean unmapped pages will be reclaimed if the zone watermarks are not
being met.
There is a heuristic that determines if the scan is worthwhile but the
problem is that the heuristic is not being properly applied and is
basically assuming zone_reclaim_mode is 1 if it is enabled. The lack of
proper detection can manfiest as high CPU usage as the LRU list is scanned
uselessly.
Historically, once enabled it was depending on NR_FILE_PAGES which may
include swapcache pages that the reclaim_mode cannot deal with. Patch
vmscan-change-the-number-of-the-unmapped-files-in-zone-reclaim.patch by
Kosaki Motohiro noted that zone_page_state(zone, NR_FILE_PAGES) included
pages that were not file-backed such as swapcache and made a calculation
based on the inactive, active and mapped files. This is far superior when
zone_reclaim==1 but if RECLAIM_SWAP is set, then NR_FILE_PAGES is a
reasonable starting figure.
This patch alters how zone_reclaim() works out how many pages it might be
able to reclaim given the current reclaim_mode. If RECLAIM_SWAP is set in
the reclaim_mode it will either consider NR_FILE_PAGES as potential
candidates or else use NR_{IN}ACTIVE}_PAGES-NR_FILE_MAPPED to discount
swapcache and other non-file-backed pages. If RECLAIM_WRITE is not set,
then NR_FILE_DIRTY number of pages are not candidates. If RECLAIM_SWAP is
not set, then NR_FILE_MAPPED are not.
[kosaki.motohiro@jp.fujitsu.com: Estimate unmapped pages minus tmpfs pages]
[fengguang.wu@intel.com: Fix underflow problem in Kosaki's estimate]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:20 +07:00
|
|
|
static inline unsigned long zone_unmapped_file_pages(struct zone *zone)
|
|
|
|
{
|
|
|
|
unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED);
|
|
|
|
unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) +
|
|
|
|
zone_page_state(zone, NR_ACTIVE_FILE);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* It's possible for there to be more file mapped pages than
|
|
|
|
* accounted for by the pages on the file LRU lists because
|
|
|
|
* tmpfs pages accounted for as ANON can also be FILE_MAPPED
|
|
|
|
*/
|
|
|
|
return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Work out how many page cache pages we can reclaim in this reclaim_mode */
|
|
|
|
static long zone_pagecache_reclaimable(struct zone *zone)
|
|
|
|
{
|
|
|
|
long nr_pagecache_reclaimable;
|
|
|
|
long delta = 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If RECLAIM_SWAP is set, then all file pages are considered
|
|
|
|
* potentially reclaimable. Otherwise, we have to worry about
|
|
|
|
* pages like swapcache and zone_unmapped_file_pages() provides
|
|
|
|
* a better estimate
|
|
|
|
*/
|
|
|
|
if (zone_reclaim_mode & RECLAIM_SWAP)
|
|
|
|
nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES);
|
|
|
|
else
|
|
|
|
nr_pagecache_reclaimable = zone_unmapped_file_pages(zone);
|
|
|
|
|
|
|
|
/* If we can't clean pages, remove dirty pages from consideration */
|
|
|
|
if (!(zone_reclaim_mode & RECLAIM_WRITE))
|
|
|
|
delta += zone_page_state(zone, NR_FILE_DIRTY);
|
|
|
|
|
|
|
|
/* Watch for any possible underflows due to delta */
|
|
|
|
if (unlikely(delta > nr_pagecache_reclaimable))
|
|
|
|
delta = nr_pagecache_reclaimable;
|
|
|
|
|
|
|
|
return nr_pagecache_reclaimable - delta;
|
|
|
|
}
|
|
|
|
|
2006-01-19 08:42:31 +07:00
|
|
|
/*
|
|
|
|
* Try to free up some pages from this zone through reclaim.
|
|
|
|
*/
|
2006-03-22 15:08:18 +07:00
|
|
|
static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
|
2006-01-19 08:42:31 +07:00
|
|
|
{
|
2006-03-22 15:08:22 +07:00
|
|
|
/* Minimum pages needed in order to stay on node */
|
2006-03-22 15:08:19 +07:00
|
|
|
const unsigned long nr_pages = 1 << order;
|
2006-01-19 08:42:31 +07:00
|
|
|
struct task_struct *p = current;
|
|
|
|
struct reclaim_state reclaim_state;
|
2006-03-22 15:08:18 +07:00
|
|
|
int priority;
|
2006-03-22 15:08:18 +07:00
|
|
|
struct scan_control sc = {
|
|
|
|
.may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE),
|
2009-04-01 05:19:30 +07:00
|
|
|
.may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP),
|
2009-04-22 02:24:57 +07:00
|
|
|
.may_swap = 1,
|
2009-12-15 08:59:10 +07:00
|
|
|
.nr_to_reclaim = max_t(unsigned long, nr_pages,
|
|
|
|
SWAP_CLUSTER_MAX),
|
2006-03-22 15:08:18 +07:00
|
|
|
.gfp_mask = gfp_mask,
|
2009-04-01 05:19:38 +07:00
|
|
|
.order = order,
|
2006-03-22 15:08:18 +07:00
|
|
|
};
|
2011-05-25 07:12:26 +07:00
|
|
|
struct shrink_control shrink = {
|
|
|
|
.gfp_mask = sc.gfp_mask,
|
|
|
|
};
|
2010-08-10 07:19:50 +07:00
|
|
|
unsigned long nr_slab_pages0, nr_slab_pages1;
|
2006-01-19 08:42:31 +07:00
|
|
|
|
|
|
|
cond_resched();
|
2006-02-25 04:04:22 +07:00
|
|
|
/*
|
|
|
|
* We need to be able to allocate from the reserves for RECLAIM_SWAP
|
|
|
|
* and we also need to be able to write out pages for RECLAIM_WRITE
|
|
|
|
* and RECLAIM_SWAP.
|
|
|
|
*/
|
|
|
|
p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
|
2010-03-06 04:41:47 +07:00
|
|
|
lockdep_set_current_reclaim_state(gfp_mask);
|
2006-01-19 08:42:31 +07:00
|
|
|
reclaim_state.reclaimed_slab = 0;
|
|
|
|
p->reclaim_state = &reclaim_state;
|
2006-02-01 18:05:29 +07:00
|
|
|
|
vmscan: properly account for the number of page cache pages zone_reclaim() can reclaim
A bug was brought to my attention against a distro kernel but it affects
mainline and I believe problems like this have been reported in various
guises on the mailing lists although I don't have specific examples at the
moment.
The reported problem was that malloc() stalled for a long time (minutes in
some cases) if a large tmpfs mount was occupying a large percentage of
memory overall. The pages did not get cleaned or reclaimed by
zone_reclaim() because the zone_reclaim_mode was unsuitable, but the lists
are uselessly scanned frequencly making the CPU spin at near 100%.
This patchset intends to address that bug and bring the behaviour of
zone_reclaim() more in line with expectations which were noticed during
investigation. It is based on top of mmotm and takes advantage of
Kosaki's work with respect to zone_reclaim().
Patch 1 fixes the heuristics that zone_reclaim() uses to determine if the
scan should go ahead. The broken heuristic is what was causing the
malloc() stall as it uselessly scanned the LRU constantly. Currently,
zone_reclaim is assuming zone_reclaim_mode is 1 and historically it
could not deal with tmpfs pages at all. This fixes up the heuristic so
that an unnecessary scan is more likely to be correctly avoided.
Patch 2 notes that zone_reclaim() returning a failure automatically means
the zone is marked full. This is not always true. It could have
failed because the GFP mask or zone_reclaim_mode were unsuitable.
Patch 3 introduces a counter zreclaim_failed that will increment each
time the zone_reclaim scan-avoidance heuristics fail. If that
counter is rapidly increasing, then zone_reclaim_mode should be
set to 0 as a temporarily resolution and a bug reported because
the scan-avoidance heuristic is still broken.
This patch:
On NUMA machines, the administrator can configure zone_reclaim_mode that
is a more targetted form of direct reclaim. On machines with large NUMA
distances for example, a zone_reclaim_mode defaults to 1 meaning that
clean unmapped pages will be reclaimed if the zone watermarks are not
being met.
There is a heuristic that determines if the scan is worthwhile but the
problem is that the heuristic is not being properly applied and is
basically assuming zone_reclaim_mode is 1 if it is enabled. The lack of
proper detection can manfiest as high CPU usage as the LRU list is scanned
uselessly.
Historically, once enabled it was depending on NR_FILE_PAGES which may
include swapcache pages that the reclaim_mode cannot deal with. Patch
vmscan-change-the-number-of-the-unmapped-files-in-zone-reclaim.patch by
Kosaki Motohiro noted that zone_page_state(zone, NR_FILE_PAGES) included
pages that were not file-backed such as swapcache and made a calculation
based on the inactive, active and mapped files. This is far superior when
zone_reclaim==1 but if RECLAIM_SWAP is set, then NR_FILE_PAGES is a
reasonable starting figure.
This patch alters how zone_reclaim() works out how many pages it might be
able to reclaim given the current reclaim_mode. If RECLAIM_SWAP is set in
the reclaim_mode it will either consider NR_FILE_PAGES as potential
candidates or else use NR_{IN}ACTIVE}_PAGES-NR_FILE_MAPPED to discount
swapcache and other non-file-backed pages. If RECLAIM_WRITE is not set,
then NR_FILE_DIRTY number of pages are not candidates. If RECLAIM_SWAP is
not set, then NR_FILE_MAPPED are not.
[kosaki.motohiro@jp.fujitsu.com: Estimate unmapped pages minus tmpfs pages]
[fengguang.wu@intel.com: Fix underflow problem in Kosaki's estimate]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:20 +07:00
|
|
|
if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) {
|
2006-09-26 13:31:52 +07:00
|
|
|
/*
|
|
|
|
* Free memory by calling shrink zone with increasing
|
|
|
|
* priorities until we have enough memory freed.
|
|
|
|
*/
|
|
|
|
priority = ZONE_RECLAIM_PRIORITY;
|
|
|
|
do {
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
shrink_zone(priority, zone, &sc);
|
2006-09-26 13:31:52 +07:00
|
|
|
priority--;
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
} while (priority >= 0 && sc.nr_reclaimed < nr_pages);
|
2006-09-26 13:31:52 +07:00
|
|
|
}
|
2006-02-01 18:05:29 +07:00
|
|
|
|
2010-08-10 07:19:50 +07:00
|
|
|
nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
|
|
|
|
if (nr_slab_pages0 > zone->min_slab_pages) {
|
2006-02-01 18:05:35 +07:00
|
|
|
/*
|
2006-03-22 15:08:22 +07:00
|
|
|
* shrink_slab() does not currently allow us to determine how
|
2006-09-26 13:31:52 +07:00
|
|
|
* many pages were freed in this zone. So we take the current
|
|
|
|
* number of slab pages and shake the slab until it is reduced
|
|
|
|
* by the same nr_pages that we used for reclaiming unmapped
|
|
|
|
* pages.
|
2006-02-01 18:05:35 +07:00
|
|
|
*
|
2006-09-26 13:31:52 +07:00
|
|
|
* Note that shrink_slab will free memory on all zones and may
|
|
|
|
* take a long time.
|
2006-02-01 18:05:35 +07:00
|
|
|
*/
|
2010-08-10 07:19:54 +07:00
|
|
|
for (;;) {
|
|
|
|
unsigned long lru_pages = zone_reclaimable_pages(zone);
|
|
|
|
|
|
|
|
/* No reclaimable slab or very low memory pressure */
|
2011-05-25 07:12:27 +07:00
|
|
|
if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages))
|
2010-08-10 07:19:54 +07:00
|
|
|
break;
|
|
|
|
|
|
|
|
/* Freed enough memory */
|
|
|
|
nr_slab_pages1 = zone_page_state(zone,
|
|
|
|
NR_SLAB_RECLAIMABLE);
|
|
|
|
if (nr_slab_pages1 + nr_pages <= nr_slab_pages0)
|
|
|
|
break;
|
|
|
|
}
|
2006-09-26 13:31:53 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Update nr_reclaimed by the number of slab pages we
|
|
|
|
* reclaimed from this zone.
|
|
|
|
*/
|
2010-08-10 07:19:50 +07:00
|
|
|
nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
|
|
|
|
if (nr_slab_pages1 < nr_slab_pages0)
|
|
|
|
sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1;
|
2006-02-01 18:05:35 +07:00
|
|
|
}
|
|
|
|
|
2006-01-19 08:42:31 +07:00
|
|
|
p->reclaim_state = NULL;
|
2006-02-25 04:04:22 +07:00
|
|
|
current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE);
|
2010-03-06 04:41:47 +07:00
|
|
|
lockdep_clear_current_reclaim_state();
|
vmscan: bail out of direct reclaim after swap_cluster_max pages
When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously. It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.
This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.
This in turn can result in each direct reclaim process freeing
many pages. Together, they can end up freeing way too many pages.
This kicks useful data out of memory (in some cases more than half
of all memory is swapped out). It also impacts performance by
keeping tasks stuck in the pageout code for too long.
A 30% improvement in hackbench has been observed with this patch.
The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.
We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.
However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.
akpm: a historical interlude...
We tried this in 2004:
:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date: Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly hit
: a large number of reclaimable pages on the LRU.
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.
And we reverted it in 2006:
:commit 210fe530305ee50cd889fe9250168228b2994f32
:Author: Andrew Morton <akpm@osdl.org>
:Date: Fri Jan 6 00:11:14 2006 -0800
:
: [PATCH] vmscan: balancing fix
:
: Revert a patch which went into 2.6.8-rc1. The changelog for that patch was:
:
: The shrink_zone() logic can, under some circumstances, cause far too many
: pages to be reclaimed. Say, we're scanning at high priority and suddenly
: hit a large number of reclaimable pages on the LRU.
:
: Change things so we bale out when SWAP_CLUSTER_MAX pages have been
: reclaimed.
:
: Problem is, this change caused significant imbalance in inter-zone scan
: balancing by truncating scans of larger zones.
:
: Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL. The zone
: balancing algorithm would require that if we're scanning 100 pages of
: ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL. But this logic will
: cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
: reclaimed. Thus effectively causing smaller zones to be scanned relatively
: harder than large ones.
:
: Now I need to remember what the workload was which caused me to write this
: patch originally, then fix it up in a different way...
And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-07 05:40:01 +07:00
|
|
|
return sc.nr_reclaimed >= nr_pages;
|
2006-01-19 08:42:31 +07:00
|
|
|
}
|
2006-03-22 15:08:18 +07:00
|
|
|
|
|
|
|
int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
|
|
|
|
{
|
|
|
|
int node_id;
|
2007-10-17 13:26:01 +07:00
|
|
|
int ret;
|
2006-03-22 15:08:18 +07:00
|
|
|
|
|
|
|
/*
|
2006-09-26 13:31:52 +07:00
|
|
|
* Zone reclaim reclaims unmapped file backed pages and
|
|
|
|
* slab pages if we are over the defined limits.
|
2006-06-30 15:55:37 +07:00
|
|
|
*
|
2006-07-03 14:24:13 +07:00
|
|
|
* A small portion of unmapped file backed pages is needed for
|
|
|
|
* file I/O otherwise pages read by file I/O will be immediately
|
|
|
|
* thrown out if the zone is overallocated. So we do not reclaim
|
|
|
|
* if less than a specified percentage of the zone is used by
|
|
|
|
* unmapped file backed pages.
|
2006-03-22 15:08:18 +07:00
|
|
|
*/
|
vmscan: properly account for the number of page cache pages zone_reclaim() can reclaim
A bug was brought to my attention against a distro kernel but it affects
mainline and I believe problems like this have been reported in various
guises on the mailing lists although I don't have specific examples at the
moment.
The reported problem was that malloc() stalled for a long time (minutes in
some cases) if a large tmpfs mount was occupying a large percentage of
memory overall. The pages did not get cleaned or reclaimed by
zone_reclaim() because the zone_reclaim_mode was unsuitable, but the lists
are uselessly scanned frequencly making the CPU spin at near 100%.
This patchset intends to address that bug and bring the behaviour of
zone_reclaim() more in line with expectations which were noticed during
investigation. It is based on top of mmotm and takes advantage of
Kosaki's work with respect to zone_reclaim().
Patch 1 fixes the heuristics that zone_reclaim() uses to determine if the
scan should go ahead. The broken heuristic is what was causing the
malloc() stall as it uselessly scanned the LRU constantly. Currently,
zone_reclaim is assuming zone_reclaim_mode is 1 and historically it
could not deal with tmpfs pages at all. This fixes up the heuristic so
that an unnecessary scan is more likely to be correctly avoided.
Patch 2 notes that zone_reclaim() returning a failure automatically means
the zone is marked full. This is not always true. It could have
failed because the GFP mask or zone_reclaim_mode were unsuitable.
Patch 3 introduces a counter zreclaim_failed that will increment each
time the zone_reclaim scan-avoidance heuristics fail. If that
counter is rapidly increasing, then zone_reclaim_mode should be
set to 0 as a temporarily resolution and a bug reported because
the scan-avoidance heuristic is still broken.
This patch:
On NUMA machines, the administrator can configure zone_reclaim_mode that
is a more targetted form of direct reclaim. On machines with large NUMA
distances for example, a zone_reclaim_mode defaults to 1 meaning that
clean unmapped pages will be reclaimed if the zone watermarks are not
being met.
There is a heuristic that determines if the scan is worthwhile but the
problem is that the heuristic is not being properly applied and is
basically assuming zone_reclaim_mode is 1 if it is enabled. The lack of
proper detection can manfiest as high CPU usage as the LRU list is scanned
uselessly.
Historically, once enabled it was depending on NR_FILE_PAGES which may
include swapcache pages that the reclaim_mode cannot deal with. Patch
vmscan-change-the-number-of-the-unmapped-files-in-zone-reclaim.patch by
Kosaki Motohiro noted that zone_page_state(zone, NR_FILE_PAGES) included
pages that were not file-backed such as swapcache and made a calculation
based on the inactive, active and mapped files. This is far superior when
zone_reclaim==1 but if RECLAIM_SWAP is set, then NR_FILE_PAGES is a
reasonable starting figure.
This patch alters how zone_reclaim() works out how many pages it might be
able to reclaim given the current reclaim_mode. If RECLAIM_SWAP is set in
the reclaim_mode it will either consider NR_FILE_PAGES as potential
candidates or else use NR_{IN}ACTIVE}_PAGES-NR_FILE_MAPPED to discount
swapcache and other non-file-backed pages. If RECLAIM_WRITE is not set,
then NR_FILE_DIRTY number of pages are not candidates. If RECLAIM_SWAP is
not set, then NR_FILE_MAPPED are not.
[kosaki.motohiro@jp.fujitsu.com: Estimate unmapped pages minus tmpfs pages]
[fengguang.wu@intel.com: Fix underflow problem in Kosaki's estimate]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-06-17 05:33:20 +07:00
|
|
|
if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages &&
|
|
|
|
zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages)
|
2009-06-17 05:33:22 +07:00
|
|
|
return ZONE_RECLAIM_FULL;
|
2006-03-22 15:08:18 +07:00
|
|
|
|
2010-03-06 04:41:55 +07:00
|
|
|
if (zone->all_unreclaimable)
|
2009-06-17 05:33:22 +07:00
|
|
|
return ZONE_RECLAIM_FULL;
|
2007-10-17 13:26:01 +07:00
|
|
|
|
2006-03-22 15:08:18 +07:00
|
|
|
/*
|
2007-10-17 13:26:01 +07:00
|
|
|
* Do not scan if the allocation should not be delayed.
|
2006-03-22 15:08:18 +07:00
|
|
|
*/
|
2007-10-17 13:26:01 +07:00
|
|
|
if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC))
|
2009-06-17 05:33:22 +07:00
|
|
|
return ZONE_RECLAIM_NOSCAN;
|
2006-03-22 15:08:18 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Only run zone reclaim on the local zone or on zones that do not
|
|
|
|
* have associated processors. This will favor the local processor
|
|
|
|
* over remote processors and spread off node memory allocations
|
|
|
|
* as wide as possible.
|
|
|
|
*/
|
2006-09-26 13:31:55 +07:00
|
|
|
node_id = zone_to_nid(zone);
|
2007-10-16 15:25:36 +07:00
|
|
|
if (node_state(node_id, N_CPU) && node_id != numa_node_id())
|
2009-06-17 05:33:22 +07:00
|
|
|
return ZONE_RECLAIM_NOSCAN;
|
2007-10-17 13:26:01 +07:00
|
|
|
|
|
|
|
if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED))
|
2009-06-17 05:33:22 +07:00
|
|
|
return ZONE_RECLAIM_NOSCAN;
|
|
|
|
|
2007-10-17 13:26:01 +07:00
|
|
|
ret = __zone_reclaim(zone, gfp_mask, order);
|
|
|
|
zone_clear_flag(zone, ZONE_RECLAIM_LOCKED);
|
|
|
|
|
2009-06-17 05:33:23 +07:00
|
|
|
if (!ret)
|
|
|
|
count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED);
|
|
|
|
|
2007-10-17 13:26:01 +07:00
|
|
|
return ret;
|
2006-03-22 15:08:18 +07:00
|
|
|
}
|
2006-01-19 08:42:31 +07:00
|
|
|
#endif
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* page_evictable - test whether a page is evictable
|
|
|
|
* @page: the page to test
|
|
|
|
* @vma: the VMA in which the page is or will be mapped, may be NULL
|
|
|
|
*
|
|
|
|
* Test whether page is evictable--i.e., should be placed on active/inactive
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
* lists vs unevictable list. The vma argument is !NULL when called from the
|
|
|
|
* fault path to determine how to instantate a new page.
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
*
|
|
|
|
* Reasons page might not be evictable:
|
2008-10-19 10:26:42 +07:00
|
|
|
* (1) page's mapping marked unevictable
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
* (2) page is part of an mlocked VMA
|
2008-10-19 10:26:42 +07:00
|
|
|
*
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
*/
|
|
|
|
int page_evictable(struct page *page, struct vm_area_struct *vma)
|
|
|
|
{
|
|
|
|
|
2008-10-19 10:26:42 +07:00
|
|
|
if (mapping_unevictable(page_mapping(page)))
|
|
|
|
return 0;
|
|
|
|
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:44 +07:00
|
|
|
if (PageMlocked(page) || (vma && is_mlocked_vma(vma, page)))
|
|
|
|
return 0;
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 10:26:39 +07:00
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
2008-10-19 10:26:43 +07:00
|
|
|
|
SHM_UNLOCK: fix long unpreemptible section
scan_mapping_unevictable_pages() is used to make SysV SHM_LOCKed pages
evictable again once the shared memory is unlocked. It does this with
pagevec_lookup()s across the whole object (which might occupy most of
memory), and takes 300ms to unlock 7GB here. A cond_resched() every
PAGEVEC_SIZE pages would be good.
However, KOSAKI-san points out that this is called under shmem.c's
info->lock, and it's also under shm.c's shm_lock(), both spinlocks.
There is no strong reason for that: we need to take these pages off the
unevictable list soonish, but those locks are not required for it.
So move the call to scan_mapping_unevictable_pages() from shmem.c's
unlock handling up to shm.c's unlock handling. Remove the recently
added barrier, not needed now we have spin_unlock() before the scan.
Use get_file(), with subsequent fput(), to make sure we have a reference
to mapping throughout scan_mapping_unevictable_pages(): that's something
that was previously guaranteed by the shm_lock().
Remove shmctl's lru_add_drain_all(): we don't fault in pages at SHM_LOCK
time, and we lazily discover them to be Unevictable later, so it serves
no purpose for SHM_LOCK; and serves no purpose for SHM_UNLOCK, since
pages still on pagevec are not marked Unevictable.
The original code avoided redundant rescans by checking VM_LOCKED flag
at its level: now avoid them by checking shp's SHM_LOCKED.
The original code called scan_mapping_unevictable_pages() on a locked
area at shm_destroy() time: perhaps we once had accounting cross-checks
which required that, but not now, so skip the overhead and just let
inode eviction deal with them.
Put check_move_unevictable_page() and scan_mapping_unevictable_pages()
under CONFIG_SHMEM (with stub for the TINY case when ramfs is used),
more as comment than to save space; comment them used for SHM_UNLOCK.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:19 +07:00
|
|
|
#ifdef CONFIG_SHMEM
|
2008-10-19 10:26:43 +07:00
|
|
|
/**
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
* check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list
|
|
|
|
* @pages: array of pages to check
|
|
|
|
* @nr_pages: number of pages to check
|
2008-10-19 10:26:43 +07:00
|
|
|
*
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
* Checks pages for evictability and moves them to the appropriate lru list.
|
SHM_UNLOCK: fix long unpreemptible section
scan_mapping_unevictable_pages() is used to make SysV SHM_LOCKed pages
evictable again once the shared memory is unlocked. It does this with
pagevec_lookup()s across the whole object (which might occupy most of
memory), and takes 300ms to unlock 7GB here. A cond_resched() every
PAGEVEC_SIZE pages would be good.
However, KOSAKI-san points out that this is called under shmem.c's
info->lock, and it's also under shm.c's shm_lock(), both spinlocks.
There is no strong reason for that: we need to take these pages off the
unevictable list soonish, but those locks are not required for it.
So move the call to scan_mapping_unevictable_pages() from shmem.c's
unlock handling up to shm.c's unlock handling. Remove the recently
added barrier, not needed now we have spin_unlock() before the scan.
Use get_file(), with subsequent fput(), to make sure we have a reference
to mapping throughout scan_mapping_unevictable_pages(): that's something
that was previously guaranteed by the shm_lock().
Remove shmctl's lru_add_drain_all(): we don't fault in pages at SHM_LOCK
time, and we lazily discover them to be Unevictable later, so it serves
no purpose for SHM_LOCK; and serves no purpose for SHM_UNLOCK, since
pages still on pagevec are not marked Unevictable.
The original code avoided redundant rescans by checking VM_LOCKED flag
at its level: now avoid them by checking shp's SHM_LOCKED.
The original code called scan_mapping_unevictable_pages() on a locked
area at shm_destroy() time: perhaps we once had accounting cross-checks
which required that, but not now, so skip the overhead and just let
inode eviction deal with them.
Put check_move_unevictable_page() and scan_mapping_unevictable_pages()
under CONFIG_SHMEM (with stub for the TINY case when ramfs is used),
more as comment than to save space; comment them used for SHM_UNLOCK.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:19 +07:00
|
|
|
*
|
|
|
|
* This function is only used for SysV IPC SHM_UNLOCK.
|
2008-10-19 10:26:43 +07:00
|
|
|
*/
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
void check_move_unevictable_pages(struct page **pages, int nr_pages)
|
2008-10-19 10:26:43 +07:00
|
|
|
{
|
2012-01-13 08:18:15 +07:00
|
|
|
struct lruvec *lruvec;
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
struct zone *zone = NULL;
|
|
|
|
int pgscanned = 0;
|
|
|
|
int pgrescued = 0;
|
|
|
|
int i;
|
2008-10-19 10:26:43 +07:00
|
|
|
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
for (i = 0; i < nr_pages; i++) {
|
|
|
|
struct page *page = pages[i];
|
|
|
|
struct zone *pagezone;
|
2008-10-19 10:26:43 +07:00
|
|
|
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
pgscanned++;
|
|
|
|
pagezone = page_zone(page);
|
|
|
|
if (pagezone != zone) {
|
|
|
|
if (zone)
|
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
|
|
|
zone = pagezone;
|
|
|
|
spin_lock_irq(&zone->lru_lock);
|
|
|
|
}
|
2008-10-19 10:26:43 +07:00
|
|
|
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
if (!PageLRU(page) || !PageUnevictable(page))
|
|
|
|
continue;
|
2008-10-19 10:26:43 +07:00
|
|
|
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
if (page_evictable(page, NULL)) {
|
|
|
|
enum lru_list lru = page_lru_base_type(page);
|
|
|
|
|
|
|
|
VM_BUG_ON(PageActive(page));
|
|
|
|
ClearPageUnevictable(page);
|
|
|
|
__dec_zone_state(zone, NR_UNEVICTABLE);
|
|
|
|
lruvec = mem_cgroup_lru_move_lists(zone, page,
|
|
|
|
LRU_UNEVICTABLE, lru);
|
|
|
|
list_move(&page->lru, &lruvec->lists[lru]);
|
|
|
|
__inc_zone_state(zone, NR_INACTIVE_ANON + lru);
|
|
|
|
pgrescued++;
|
2008-10-19 10:26:43 +07:00
|
|
|
}
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
}
|
2008-10-19 10:26:43 +07:00
|
|
|
|
SHM_UNLOCK: fix Unevictable pages stranded after swap
Commit cc39c6a9bbde ("mm: account skipped entries to avoid looping in
find_get_pages") correctly fixed an infinite loop; but left a problem
that find_get_pages() on shmem would return 0 (appearing to callers to
mean end of tree) when it meets a run of nr_pages swap entries.
The only uses of find_get_pages() on shmem are via pagevec_lookup(),
called from invalidate_mapping_pages(), and from shmctl SHM_UNLOCK's
scan_mapping_unevictable_pages(). The first is already commented, and
not worth worrying about; but the second can leave pages on the
Unevictable list after an unusual sequence of swapping and locking.
Fix that by using shmem_find_get_pages_and_swap() (then ignoring the
swap) instead of pagevec_lookup().
But I don't want to contaminate vmscan.c with shmem internals, nor
shmem.c with LRU locking. So move scan_mapping_unevictable_pages() into
shmem.c, renaming it shmem_unlock_mapping(); and rename
check_move_unevictable_page() to check_move_unevictable_pages(), looping
down an array of pages, oftentimes under the same lock.
Leave out the "rotate unevictable list" block: that's a leftover from
when this was used for /proc/sys/vm/scan_unevictable_pages, whose flawed
handling involved looking at pages at tail of LRU.
Was there significance to the sequence first ClearPageUnevictable, then
test page_evictable, then SetPageUnevictable here? I think not, we're
under LRU lock, and have no barriers between those.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: <stable@vger.kernel.org> [back to 3.1 but will need respins]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:21 +07:00
|
|
|
if (zone) {
|
|
|
|
__count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued);
|
|
|
|
__count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned);
|
|
|
|
spin_unlock_irq(&zone->lru_lock);
|
2008-10-19 10:26:43 +07:00
|
|
|
}
|
|
|
|
}
|
SHM_UNLOCK: fix long unpreemptible section
scan_mapping_unevictable_pages() is used to make SysV SHM_LOCKed pages
evictable again once the shared memory is unlocked. It does this with
pagevec_lookup()s across the whole object (which might occupy most of
memory), and takes 300ms to unlock 7GB here. A cond_resched() every
PAGEVEC_SIZE pages would be good.
However, KOSAKI-san points out that this is called under shmem.c's
info->lock, and it's also under shm.c's shm_lock(), both spinlocks.
There is no strong reason for that: we need to take these pages off the
unevictable list soonish, but those locks are not required for it.
So move the call to scan_mapping_unevictable_pages() from shmem.c's
unlock handling up to shm.c's unlock handling. Remove the recently
added barrier, not needed now we have spin_unlock() before the scan.
Use get_file(), with subsequent fput(), to make sure we have a reference
to mapping throughout scan_mapping_unevictable_pages(): that's something
that was previously guaranteed by the shm_lock().
Remove shmctl's lru_add_drain_all(): we don't fault in pages at SHM_LOCK
time, and we lazily discover them to be Unevictable later, so it serves
no purpose for SHM_LOCK; and serves no purpose for SHM_UNLOCK, since
pages still on pagevec are not marked Unevictable.
The original code avoided redundant rescans by checking VM_LOCKED flag
at its level: now avoid them by checking shp's SHM_LOCKED.
The original code called scan_mapping_unevictable_pages() on a locked
area at shm_destroy() time: perhaps we once had accounting cross-checks
which required that, but not now, so skip the overhead and just let
inode eviction deal with them.
Put check_move_unevictable_page() and scan_mapping_unevictable_pages()
under CONFIG_SHMEM (with stub for the TINY case when ramfs is used),
more as comment than to save space; comment them used for SHM_UNLOCK.
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-01-21 05:34:19 +07:00
|
|
|
#endif /* CONFIG_SHMEM */
|
2008-10-19 10:26:53 +07:00
|
|
|
|
2011-11-01 07:09:13 +07:00
|
|
|
static void warn_scan_unevictable_pages(void)
|
2008-10-19 10:26:53 +07:00
|
|
|
{
|
2011-11-01 07:09:13 +07:00
|
|
|
printk_once(KERN_WARNING
|
2012-01-11 06:07:40 +07:00
|
|
|
"%s: The scan_unevictable_pages sysctl/node-interface has been "
|
2011-11-01 07:09:13 +07:00
|
|
|
"disabled for lack of a legitimate use case. If you have "
|
2012-01-11 06:07:40 +07:00
|
|
|
"one, please send an email to linux-mm@kvack.org.\n",
|
|
|
|
current->comm);
|
2008-10-19 10:26:53 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* scan_unevictable_pages [vm] sysctl handler. On demand re-scan of
|
|
|
|
* all nodes' unevictable lists for evictable pages
|
|
|
|
*/
|
|
|
|
unsigned long scan_unevictable_pages;
|
|
|
|
|
|
|
|
int scan_unevictable_handler(struct ctl_table *table, int write,
|
2009-09-24 05:57:19 +07:00
|
|
|
void __user *buffer,
|
2008-10-19 10:26:53 +07:00
|
|
|
size_t *length, loff_t *ppos)
|
|
|
|
{
|
2011-11-01 07:09:13 +07:00
|
|
|
warn_scan_unevictable_pages();
|
2009-09-24 05:57:19 +07:00
|
|
|
proc_doulongvec_minmax(table, write, buffer, length, ppos);
|
2008-10-19 10:26:53 +07:00
|
|
|
scan_unevictable_pages = 0;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-10-27 04:21:28 +07:00
|
|
|
#ifdef CONFIG_NUMA
|
2008-10-19 10:26:53 +07:00
|
|
|
/*
|
|
|
|
* per node 'scan_unevictable_pages' attribute. On demand re-scan of
|
|
|
|
* a specified node's per zone unevictable lists for evictable pages.
|
|
|
|
*/
|
|
|
|
|
2011-12-22 05:48:43 +07:00
|
|
|
static ssize_t read_scan_unevictable_node(struct device *dev,
|
|
|
|
struct device_attribute *attr,
|
2008-10-19 10:26:53 +07:00
|
|
|
char *buf)
|
|
|
|
{
|
2011-11-01 07:09:13 +07:00
|
|
|
warn_scan_unevictable_pages();
|
2008-10-19 10:26:53 +07:00
|
|
|
return sprintf(buf, "0\n"); /* always zero; should fit... */
|
|
|
|
}
|
|
|
|
|
2011-12-22 05:48:43 +07:00
|
|
|
static ssize_t write_scan_unevictable_node(struct device *dev,
|
|
|
|
struct device_attribute *attr,
|
2008-10-19 10:26:53 +07:00
|
|
|
const char *buf, size_t count)
|
|
|
|
{
|
2011-11-01 07:09:13 +07:00
|
|
|
warn_scan_unevictable_pages();
|
2008-10-19 10:26:53 +07:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2011-12-22 05:48:43 +07:00
|
|
|
static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR,
|
2008-10-19 10:26:53 +07:00
|
|
|
read_scan_unevictable_node,
|
|
|
|
write_scan_unevictable_node);
|
|
|
|
|
|
|
|
int scan_unevictable_register_node(struct node *node)
|
|
|
|
{
|
2011-12-22 05:48:43 +07:00
|
|
|
return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages);
|
2008-10-19 10:26:53 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
void scan_unevictable_unregister_node(struct node *node)
|
|
|
|
{
|
2011-12-22 05:48:43 +07:00
|
|
|
device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages);
|
2008-10-19 10:26:53 +07:00
|
|
|
}
|
2010-10-27 04:21:28 +07:00
|
|
|
#endif
|