Add operations to allocate/release p4ds.
Xen requires more work. We will need to come back to it.
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-arch@vger.kernel.org
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/20170330080731.65421-5-kirill.shutemov@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch converts x86 to use proper folding of a new (fifth) page table level
with <asm-generic/pgtable-nop4d.h>.
That's a bit of a kitchen sink patch, but I don't see how to split it further
without hurting bisectability.
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: linux-arch@vger.kernel.org
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/20170317185515.8636-7-kirill.shutemov@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Page tables can bite a relatively big chunk off system memory and their
allocations are easy to trigger from userspace, so they should be
accounted to kmemcg.
This patch marks page table allocations as __GFP_ACCOUNT for x86. Note
we must not charge allocations of kernel page tables, because they can
be shared among processes from different cgroups so accounting them to a
particular one can pin other cgroups for indefinitely long. So we clear
__GFP_ACCOUNT flag if a page table is allocated for the kernel.
Link: http://lkml.kernel.org/r/7d5c54f6a2bcbe76f03171689440003d87e6c742.1464079538.git.vdavydov@virtuozzo.com
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the third version of the patchset previously sent [1]. I have
basically only rebased it on top of 4.7-rc1 tree and dropped "dm: get
rid of superfluous gfp flags" which went through dm tree. I am sending
it now because it is tree wide and chances for conflicts are reduced
considerably when we want to target rc2. I plan to send the next step
and rename the flag and move to a better semantic later during this
release cycle so we will have a new semantic ready for 4.8 merge window
hopefully.
Motivation:
While working on something unrelated I've checked the current usage of
__GFP_REPEAT in the tree. It seems that a majority of the usage is and
always has been bogus because __GFP_REPEAT has always been about costly
high order allocations while we are using it for order-0 or very small
orders very often. It seems that a big pile of them is just a
copy&paste when a code has been adopted from one arch to another.
I think it makes some sense to get rid of them because they are just
making the semantic more unclear. Please note that GFP_REPEAT is
documented as
* __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
* _might_ fail. This depends upon the particular VM implementation.
while !costly requests have basically nofail semantic. So one could
reasonably expect that order-0 request with __GFP_REPEAT will not loop
for ever. This is not implemented right now though.
I would like to move on with __GFP_REPEAT and define a better semantic
for it.
$ git grep __GFP_REPEAT origin/master | wc -l
111
$ git grep __GFP_REPEAT | wc -l
36
So we are down to the third after this patch series. The remaining
places really seem to be relying on __GFP_REPEAT due to large allocation
requests. This still needs some double checking which I will do later
after all the simple ones are sorted out.
I am touching a lot of arch specific code here and I hope I got it right
but as a matter of fact I even didn't compile test for some archs as I
do not have cross compiler for them. Patches should be quite trivial to
review for stupid compile mistakes though. The tricky parts are usually
hidden by macro definitions and thats where I would appreciate help from
arch maintainers.
[1] http://lkml.kernel.org/r/1461849846-27209-1-git-send-email-mhocko@kernel.org
This patch (of 19):
__GFP_REPEAT has a rather weak semantic but since it has been introduced
around 2.6.12 it has been ignored for low order allocations. Yet we
have the full kernel tree with its usage for apparently order-0
allocations. This is really confusing because __GFP_REPEAT is
explicitly documented to allow allocation failures which is a weaker
semantic than the current order-0 has (basically nofail).
Let's simply drop __GFP_REPEAT from those places. This would allow to
identify place which really need allocator to retry harder and formulate
a more specific semantic for what the flag is supposed to do actually.
Link: http://lkml.kernel.org/r/1464599699-30131-2-git-send-email-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chen Liqin <liqin.linux@gmail.com>
Cc: Chris Metcalf <cmetcalf@mellanox.com> [for tile]
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: John Crispin <blogic@openwrt.org>
Cc: Lennox Wu <lennox.wu@gmail.com>
Cc: Ley Foon Tan <lftan@altera.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vineet Gupta <vgupta@synopsys.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We would want to use number of page table level to define mm_struct.
Let's expose it as CONFIG_PGTABLE_LEVELS.
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Distros generally (I looked at Debian, RHEL5 and SLES11) seem to
enable CONFIG_HIGHPTE for any x86 configuration which has highmem
enabled. This means that the overhead applies even to machines which
have a fairly modest amount of high memory and which therefore do not
really benefit from allocating PTEs in high memory but still pay the
price of the additional mapping operations.
Running kernbench on a 4G box I found that with CONFIG_HIGHPTE=y but
no actual highptes being allocated there was a reduction in system
time used from 59.737s to 55.9s.
With CONFIG_HIGHPTE=y and highmem PTEs being allocated:
Average Optimal load -j 4 Run (std deviation):
Elapsed Time 175.396 (0.238914)
User Time 515.983 (5.85019)
System Time 59.737 (1.26727)
Percent CPU 263.8 (71.6796)
Context Switches 39989.7 (4672.64)
Sleeps 42617.7 (246.307)
With CONFIG_HIGHPTE=y but with no highmem PTEs being allocated:
Average Optimal load -j 4 Run (std deviation):
Elapsed Time 174.278 (0.831968)
User Time 515.659 (6.07012)
System Time 55.9 (1.07799)
Percent CPU 263.8 (71.266)
Context Switches 39929.6 (4485.13)
Sleeps 42583.7 (373.039)
This patch allows the user to control the allocation of PTEs in
highmem from the command line ("userpte=nohigh") but retains the
status-quo as the default.
It is possible that some simple heuristic could be developed which
allows auto-tuning of this option however I don't have a sufficiently
large machine available to me to perform any particularly meaningful
experiments. We could probably handwave up an argument for a threshold
at 16G of total RAM.
Assuming 768M of lowmem we have 196608 potential lowmem PTE
pages. Each page can map 2M of RAM in a PAE-enabled configuration,
meaning a maximum of 384G of RAM could potentially be mapped using
lowmem PTEs.
Even allowing generous factor of 10 to account for other required
lowmem allocations, generous slop to account for page sharing (which
reduces the total amount of RAM mappable by a given number of PT
pages) and other innacuracies in the estimations it would seem that
even a 32G machine would not have a particularly pressing need for
highmem PTEs. I think 32G could be considered to be at the upper bound
of what might be sensible on a 32 bit machine (although I think in
practice 64G is still supported).
It's seems questionable if HIGHPTE is even a win for any amount of RAM
you would sensibly run a 32 bit kernel on rather than going 64 bit.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
LKML-Reference: <1266403090-20162-1-git-send-email-ian.campbell@citrix.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
mm: Pass virtual address to [__]p{te,ud,md}_free_tlb()
Upcoming paches to support the new 64-bit "BookE" powerpc architecture
will need to have the virtual address corresponding to PTE page when
freeing it, due to the way the HW table walker works.
Basically, the TLB can be loaded with "large" pages that cover the whole
virtual space (well, sort-of, half of it actually) represented by a PTE
page, and which contain an "indirect" bit indicating that this TLB entry
RPN points to an array of PTEs from which the TLB can then create direct
entries. Thus, in order to invalidate those when PTE pages are deleted,
we need the virtual address to pass to tlbilx or tlbivax instructions.
The old trick of sticking it somewhere in the PTE page struct page sucks
too much, the address is almost readily available in all call sites and
almost everybody implemets these as macros, so we may as well add the
argument everywhere. I added it to the pmd and pud variants for consistency.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: David Howells <dhowells@redhat.com> [MN10300 & FRV]
Acked-by: Nick Piggin <npiggin@suse.de>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com> [s390]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
On -rt we were seeing spurious bad page states like:
Bad page state in process 'firefox'
page:c1bc2380 flags:0x40000000 mapping:c1bc2390 mapcount:0 count:0
Trying to fix it up, but a reboot is needed
Backtrace:
Pid: 503, comm: firefox Not tainted 2.6.26.8-rt13 #3
[<c043d0f3>] ? printk+0x14/0x19
[<c0272d4e>] bad_page+0x4e/0x79
[<c0273831>] free_hot_cold_page+0x5b/0x1d3
[<c02739f6>] free_hot_page+0xf/0x11
[<c0273a18>] __free_pages+0x20/0x2b
[<c027d170>] __pte_alloc+0x87/0x91
[<c027d25e>] handle_mm_fault+0xe4/0x733
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c043f680>] ? rt_mutex_down_read_trylock+0x57/0x63
[<c0218875>] do_page_fault+0x36f/0x88a
This is the case where a concurrent fault already installed the PTE and
we get to free the newly allocated one.
This is due to pgtable_page_ctor() doing the spin_lock_init(&page->ptl)
which is overlaid with the {private, mapping} struct.
union {
struct {
unsigned long private;
struct address_space *mapping;
};
spinlock_t ptl;
struct kmem_cache *slab;
struct page *first_page;
};
Normally the spinlock is small enough to not stomp on page->mapping, but
PREEMPT_RT=y has huge 'spin'locks.
But lockdep kernels should also be able to trigger this splat, as the
lock tracking code grows the spinlock to cover page->mapping.
The obvious fix is calling pgtable_page_dtor() like the regular pte free
path __pte_free_tlb() does.
It seems all architectures except x86 and nm10300 already do this, and
nm10300 doesn't seem to use pgtable_page_ctor(), which suggests it
doesn't do SMP or simply doesnt do MMU at all or something.
Signed-off-by: Peter Zijlstra <a.p.zijlsta@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: <stable@kernel.org>
Change header guards named "ASM_X86__*" to "_ASM_X86_*" since:
a. the double underscore is ugly and pointless.
b. no leading underscore violates namespace constraints.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
