The "mm: sched: numa: Control enabling and disabling of NUMA balancing"
depends on scheduling debug being enabled but it's perfectly legimate to
disable automatic NUMA balancing even without this option. This should
take care of it.
Signed-off-by: Mel Gorman <mgorman@suse.de>
This patch adds Kconfig options and kernel parameters to allow the
enabling and disabling of automatic NUMA balancing. The existance
of such a switch was and is very important when debugging problems
related to transparent hugepages and we should have the same for
automatic NUMA placement.
Signed-off-by: Mel Gorman <mgorman@suse.de>
The PTE scanning rate and fault rates are two of the biggest sources of
system CPU overhead with automatic NUMA placement. Ideally a proper policy
would detect if a workload was properly placed, schedule and adjust the
PTE scanning rate accordingly. We do not track the necessary information
to do that but we at least know if we migrated or not.
This patch scans slower if a page was not migrated as the result of a
NUMA hinting fault up to sysctl_numa_balancing_scan_period_max which is
now higher than the previous default. Once every minute it will reset
the scanner in case of phase changes.
This is hilariously crude and the numbers are arbitrary. Workloads will
converge quite slowly in comparison to what a proper policy should be able
to do. On the plus side, we will chew up less CPU for workloads that have
no need for automatic balancing.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Currently the rate of scanning for an address space is controlled
by the individual tasks. The next scan is simply determined by
2*p->numa_scan_period.
The 2*p->numa_scan_period is arbitrary and never changes. At this point
there is still no proper policy that decides if a task or process is
properly placed. It just scans and assumes the next NUMA fault will
place it properly. As it is assumed that pages will get properly placed
over time, increase the scan window each time a fault is incurred. This
is a big assumption as noted in the comments.
It should be noted that changing to p->numa_scan_period will increase
system CPU usage because now the scanning rate has effectively doubled.
If that is a problem then the min_rate should be made 200ms instead of
restoring the 2* logic.
Signed-off-by: Mel Gorman <mgorman@suse.de>
If there are a large number of NUMA hinting faults and all of them
are resulting in migrations it may indicate that memory is just
bouncing uselessly around. NUMA balancing cost is likely exceeding
any benefit from locality. Rate limit the PTE updates if the node
is migration rate-limited. As noted in the comments, this distorts
the NUMA faulting statistics.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Add a 1 second delay before starting to scan the working set of
a task and starting to balance it amongst nodes.
[ note that before the constant per task WSS sampling rate patch
the initial scan would happen much later still, in effect that
patch caused this regression. ]
The theory is that short-run tasks benefit very little from NUMA
placement: they come and go, and they better stick to the node
they were started on. As tasks mature and rebalance to other CPUs
and nodes, so does their NUMA placement have to change and so
does it start to matter more and more.
In practice this change fixes an observable kbuild regression:
# [ a perf stat --null --repeat 10 test of ten bzImage builds to /dev/shm ]
!NUMA:
45.291088843 seconds time elapsed ( +- 0.40% )
45.154231752 seconds time elapsed ( +- 0.36% )
+NUMA, no slow start:
46.172308123 seconds time elapsed ( +- 0.30% )
46.343168745 seconds time elapsed ( +- 0.25% )
+NUMA, 1 sec slow start:
45.224189155 seconds time elapsed ( +- 0.25% )
45.160866532 seconds time elapsed ( +- 0.17% )
and it also fixes an observable perf bench (hackbench) regression:
# perf stat --null --repeat 10 perf bench sched messaging
-NUMA:
-NUMA: 0.246225691 seconds time elapsed ( +- 1.31% )
+NUMA no slow start: 0.252620063 seconds time elapsed ( +- 1.13% )
+NUMA 1sec delay: 0.248076230 seconds time elapsed ( +- 1.35% )
The implementation is simple and straightforward, most of the patch
deals with adding the /proc/sys/kernel/numa_balancing_scan_delay_ms tunable
knob.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
[ Wrote the changelog, ran measurements, tuned the default. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Previously, to probe the working set of a task, we'd use
a very simple and crude method: mark all of its address
space PROT_NONE.
That method has various (obvious) disadvantages:
- it samples the working set at dissimilar rates,
giving some tasks a sampling quality advantage
over others.
- creates performance problems for tasks with very
large working sets
- over-samples processes with large address spaces but
which only very rarely execute
Improve that method by keeping a rotating offset into the
address space that marks the current position of the scan,
and advance it by a constant rate (in a CPU cycles execution
proportional manner). If the offset reaches the last mapped
address of the mm then it then it starts over at the first
address.
The per-task nature of the working set sampling functionality in this tree
allows such constant rate, per task, execution-weight proportional sampling
of the working set, with an adaptive sampling interval/frequency that
goes from once per 100ms up to just once per 8 seconds. The current
sampling volume is 256 MB per interval.
As tasks mature and converge their working set, so does the
sampling rate slow down to just a trickle, 256 MB per 8
seconds of CPU time executed.
This, beyond being adaptive, also rate-limits rarely
executing systems and does not over-sample on overloaded
systems.
[ In AutoNUMA speak, this patch deals with the effective sampling
rate of the 'hinting page fault'. AutoNUMA's scanning is
currently rate-limited, but it is also fundamentally
single-threaded, executing in the knuma_scand kernel thread,
so the limit in AutoNUMA is global and does not scale up with
the number of CPUs, nor does it scan tasks in an execution
proportional manner.
So the idea of rate-limiting the scanning was first implemented
in the AutoNUMA tree via a global rate limit. This patch goes
beyond that by implementing an execution rate proportional
working set sampling rate that is not implemented via a single
global scanning daemon. ]
[ Dan Carpenter pointed out a possible NULL pointer dereference in the
first version of this patch. ]
Based-on-idea-by: Andrea Arcangeli <aarcange@redhat.com>
Bug-Found-By: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
[ Wrote changelog and fixed bug. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
NOTE: This patch is based on "sched, numa, mm: Add fault driven
placement and migration policy" but as it throws away all the policy
to just leave a basic foundation I had to drop the signed-offs-by.
This patch creates a bare-bones method for setting PTEs pte_numa in the
context of the scheduler that when faulted later will be faulted onto the
node the CPU is running on. In itself this does nothing useful but any
placement policy will fundamentally depend on receiving hints on placement
from fault context and doing something intelligent about it.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Pull scheduler fixes from Ingo Molnar:
"A CPU hotplug related crash fix and a nohz accounting fixlet."
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched: Update sched_domains_numa_masks[][] when new cpus are onlined
sched: Ensure 'sched_domains_numa_levels' is safe to use in other functions
nohz: Fix one jiffy count too far in idle cputime
Pull pile 2 of execve and kernel_thread unification work from Al Viro:
"Stuff in there: kernel_thread/kernel_execve/sys_execve conversions for
several more architectures plus assorted signal fixes and cleanups.
There'll be more (in particular, real fixes for the alpha
do_notify_resume() irq mess)..."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal: (43 commits)
alpha: don't open-code trace_report_syscall_{enter,exit}
Uninclude linux/freezer.h
m32r: trim masks
avr32: trim masks
tile: don't bother with SIGTRAP in setup_frame
microblaze: don't bother with SIGTRAP in setup_rt_frame()
mn10300: don't bother with SIGTRAP in setup_frame()
frv: no need to raise SIGTRAP in setup_frame()
x86: get rid of duplicate code in case of CONFIG_VM86
unicore32: remove pointless test
h8300: trim _TIF_WORK_MASK
parisc: decide whether to go to slow path (tracesys) based on thread flags
parisc: don't bother looping in do_signal()
parisc: fix double restarts
bury the rest of TIF_IRET
sanitize tsk_is_polling()
bury _TIF_RESTORE_SIGMASK
unicore32: unobfuscate _TIF_WORK_MASK
mips: NOTIFY_RESUME is not needed in TIF masks
mips: merge the identical "return from syscall" per-ABI code
...
Conflicts:
arch/arm/include/asm/thread_info.h
Once array sched_domains_numa_masks[] []is defined, it is never updated.
When a new cpu on a new node is onlined, the coincident member in
sched_domains_numa_masks[][] is not initialized, and all the masks are 0.
As a result, the build_overlap_sched_groups() will initialize a NULL
sched_group for the new cpu on the new node, which will lead to kernel panic:
[ 3189.403280] Call Trace:
[ 3189.403286] [<ffffffff8106c36f>] warn_slowpath_common+0x7f/0xc0
[ 3189.403289] [<ffffffff8106c3ca>] warn_slowpath_null+0x1a/0x20
[ 3189.403292] [<ffffffff810b1d57>] build_sched_domains+0x467/0x470
[ 3189.403296] [<ffffffff810b2067>] partition_sched_domains+0x307/0x510
[ 3189.403299] [<ffffffff810b1ea2>] ? partition_sched_domains+0x142/0x510
[ 3189.403305] [<ffffffff810fcc93>] cpuset_update_active_cpus+0x83/0x90
[ 3189.403308] [<ffffffff810b22a8>] cpuset_cpu_active+0x38/0x70
[ 3189.403316] [<ffffffff81674b87>] notifier_call_chain+0x67/0x150
[ 3189.403320] [<ffffffff81664647>] ? native_cpu_up+0x18a/0x1b5
[ 3189.403328] [<ffffffff810a044e>] __raw_notifier_call_chain+0xe/0x10
[ 3189.403333] [<ffffffff81070470>] __cpu_notify+0x20/0x40
[ 3189.403337] [<ffffffff8166663e>] _cpu_up+0xe9/0x131
[ 3189.403340] [<ffffffff81666761>] cpu_up+0xdb/0xee
[ 3189.403348] [<ffffffff8165667c>] store_online+0x9c/0xd0
[ 3189.403355] [<ffffffff81437640>] dev_attr_store+0x20/0x30
[ 3189.403361] [<ffffffff8124aa63>] sysfs_write_file+0xa3/0x100
[ 3189.403368] [<ffffffff811ccbe0>] vfs_write+0xd0/0x1a0
[ 3189.403371] [<ffffffff811ccdb4>] sys_write+0x54/0xa0
[ 3189.403375] [<ffffffff81679c69>] system_call_fastpath+0x16/0x1b
[ 3189.403377] ---[ end trace 1e6cf85d0859c941 ]---
[ 3189.403398] BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
This patch registers a new notifier for cpu hotplug notify chain, and
updates sched_domains_numa_masks every time a new cpu is onlined or offlined.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
[ fixed compile warning ]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1348578751-16904-3-git-send-email-tangchen@cn.fujitsu.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We should temporarily reset 'sched_domains_numa_levels' to 0 after
it is reset to 'level' in sched_init_numa(). If it fails to allocate
memory for array sched_domains_numa_masks[][], the array will contain
less then 'level' members. This could be dangerous when we use it to
iterate array sched_domains_numa_masks[][] in other functions.
This patch set sched_domains_numa_levels to 0 before initializing
array sched_domains_numa_masks[][], and reset it to 'level' when
sched_domains_numa_masks[][] is fully initialized.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1348578751-16904-2-git-send-email-tangchen@cn.fujitsu.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull scheduler changes from Ingo Molnar:
"Continued quest to clean up and enhance the cputime code by Frederic
Weisbecker, in preparation for future tickless kernel features.
Other than that, smallish changes."
Fix up trivial conflicts due to additions next to each other in arch/{x86/}Kconfig
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (24 commits)
cputime: Make finegrained irqtime accounting generally available
cputime: Gather time/stats accounting config options into a single menu
ia64: Reuse system and user vtime accounting functions on task switch
ia64: Consolidate user vtime accounting
vtime: Consolidate system/idle context detection
cputime: Use a proper subsystem naming for vtime related APIs
sched: cpu_power: enable ARCH_POWER
sched/nohz: Clean up select_nohz_load_balancer()
sched: Fix load avg vs. cpu-hotplug
sched: Remove __ARCH_WANT_INTERRUPTS_ON_CTXSW
sched: Fix nohz_idle_balance()
sched: Remove useless code in yield_to()
sched: Add time unit suffix to sched sysctl knobs
sched/debug: Limit sd->*_idx range on sysctl
sched: Remove AFFINE_WAKEUPS feature flag
s390: Remove leftover account_tick_vtime() header
cputime: Consolidate vtime handling on context switch
sched: Move cputime code to its own file
cputime: Generalize CONFIG_VIRT_CPU_ACCOUNTING
tile: Remove SD_PREFER_LOCAL leftover
...
Make default just return 0. The current default (checking
TIF_POLLING_NRFLAG) is taken to architectures that need it;
ones that don't do polling in their idle threads don't need
to defined TIF_POLLING_NRFLAG at all.
ia64 defined both TS_POLLING (used by its tsk_is_polling())
and TIF_POLLING_NRFLAG (not used at all). Killed the latter...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
When exceptions or irq are about to resume userspace, if
the task needs to be rescheduled, the arch low level code
calls schedule() directly.
If we call it, it is because we have the TIF_RESCHED flag:
- It can be set after random local calls to set_need_resched()
(RCU, drm, ...)
- A wake up happened and the CPU needs preemption. This can
happen in several ways:
* Remotely: the remote waking CPU has set TIF_RESCHED and send the
wakee an IPI to schedule the new task.
* Remotely enqueued: the remote waking CPU sends an IPI to the target
and the wake up is made by the target.
* Locally: waking CPU == wakee CPU and the wakeup is done locally.
set_need_resched() is called without IPI.
In the case of local and remotely enqueued wake ups, the tick can
be restarted when we enqueue the new task and RCU can exit the
extended quiescent state at the same time. Then by the time we reach
irq exit path and we call schedule, we are not in RCU user mode.
But if we call schedule() only because something called set_need_resched(),
RCU may still be in user mode when we reach schedule.
Also if a wake up is done remotely, the CPU might see the TIF_RESCHED
flag and call schedule while the IPI has not yet happen to restart the
tick and exit RCU user mode.
We need to manually protect against these corner cases.
Create a new API schedule_user() that calls schedule() inside
rcu_user_exit()-rcu_user_enter() in order to protect it. Archs
will need to rely on it now to implement user preemption safely.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Avi Kivity <avi@redhat.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Geoff Levand <geoff@infradead.org>
Cc: Gilad Ben Yossef <gilad@benyossef.com>
Cc: Hakan Akkan <hakanakkan@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Kevin Hilman <khilman@ti.com>
Cc: Max Krasnyansky <maxk@qualcomm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Sven-Thorsten Dietrich <thebigcorporation@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
When an exception or an irq exits, and we are going to resume into
interrupted kernel code, the low level architecture code calls
preempt_schedule_irq() if there is a need to reschedule.
If the interrupt/exception occured between a call to rcu_user_enter()
(from syscall exit, exception exit, do_notify_resume exit, ...) and
a real resume to userspace (iret,...), preempt_schedule_irq() can be
called whereas RCU thinks we are in userspace. But preempt_schedule_irq()
is going to run kernel code and may be some RCU read side critical
section. We must exit the userspace extended quiescent state before
we call it.
To solve this, just call rcu_user_exit() in the beginning of
preempt_schedule_irq().
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Avi Kivity <avi@redhat.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Geoff Levand <geoff@infradead.org>
Cc: Gilad Ben Yossef <gilad@benyossef.com>
Cc: Hakan Akkan <hakanakkan@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Kevin Hilman <khilman@ti.com>
Cc: Max Krasnyansky <maxk@qualcomm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Sven-Thorsten Dietrich <thebigcorporation@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Clear the syscalls hook of a task when it's scheduled out so that if
the task migrates, it doesn't run the syscall slow path on a CPU
that might not need it.
Also set the syscalls hook on the next task if needed.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Avi Kivity <avi@redhat.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Geoff Levand <geoff@infradead.org>
Cc: Gilad Ben Yossef <gilad@benyossef.com>
Cc: Hakan Akkan <hakanakkan@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Kevin Hilman <khilman@ti.com>
Cc: Max Krasnyansky <maxk@qualcomm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Sven-Thorsten Dietrich <thebigcorporation@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Move the code that finds out to which context we account the
cputime into generic layer.
Archs that consider the whole time spent in the idle task as idle
time (ia64, powerpc) can rely on the generic vtime_account()
and implement vtime_account_system() and vtime_account_idle(),
letting the generic code to decide when to call which API.
Archs that have their own meaning of idle time, such as s390
that only considers the time spent in CPU low power mode as idle
time, can just override vtime_account().
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Use a naming based on vtime as a prefix for virtual based
cputime accounting APIs:
- account_system_vtime() -> vtime_account()
- account_switch_vtime() -> vtime_task_switch()
It makes it easier to allow for further declension such
as vtime_account_system(), vtime_account_idle(), ... if we
want to find out the context we account to from generic code.
This also make it better to know on which subsystem these APIs
refer to.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Rabik and Paul reported two different issues related to the same few
lines of code.
Rabik's issue is that the nr_uninterruptible migration code is wrong in
that he sees artifacts due to this (Rabik please do expand in more
detail).
Paul's issue is that this code as it stands relies on us using
stop_machine() for unplug, we all would like to remove this assumption
so that eventually we can remove this stop_machine() usage altogether.
The only reason we'd have to migrate nr_uninterruptible is so that we
could use for_each_online_cpu() loops in favour of
for_each_possible_cpu() loops, however since nr_uninterruptible() is the
only such loop and its using possible lets not bother at all.
The problem Rabik sees is (probably) caused by the fact that by
migrating nr_uninterruptible we screw rq->calc_load_active for both rqs
involved.
So don't bother with fancy migration schemes (meaning we now have to
keep using for_each_possible_cpu()) and instead fold any nr_active delta
after we migrate all tasks away to make sure we don't have any skewed
nr_active accounting.
[ paulmck: Move call to calc_load_migration to CPU_DEAD to avoid
miscounting noted by Rakib. ]
Reported-by: Rakib Mullick <rakib.mullick@gmail.com>
Reported-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Paul E. McKenney <paul.mckenney@linaro.org>
This reverts commit 970e178985.
Nikolay Ulyanitsky reported thatthe 3.6-rc5 kernel has a 15-20%
performance drop on PostgreSQL 9.2 on his machine (running "pgbench").
Borislav Petkov was able to reproduce this, and bisected it to this
commit 970e178985 ("sched: Improve scalability via 'CPU buddies' ...")
apparently because the new single-idle-buddy model simply doesn't find
idle CPU's to reschedule on aggressively enough.
Mike Galbraith suspects that it is likely due to the user-mode spinlocks
in PostgreSQL not reacting well to preemption, but we don't really know
the details - I'll just revert the commit for now.
There are hopefully other approaches to improve scheduler scalability
without it causing these kinds of downsides.
Reported-by: Nikolay Ulyanitsky <lystor@gmail.com>
Bisected-by: Borislav Petkov <bp@alien8.de>
Acked-by: Mike Galbraith <efault@gmx.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Heteregeneous ARM platform uses arch_scale_freq_power function
to reflect the relative capacity of each core
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1341826026-6504-6-git-send-email-vincent.guittot@linaro.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There is no load_balancer to be selected now. It just sets the
state of the nohz tick to stop.
So rename the function, pass the 'cpu' as a parameter and then
remove the useless call from tick_nohz_restart_sched_tick().
[ s/set_nohz_tick_stopped/nohz_balance_enter_idle/g
s/clear_nohz_tick_stopped/nohz_balance_exit_idle/g ]
Signed-off-by: Alex Shi <alex.shi@intel.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1347261059-24747-1-git-send-email-alex.shi@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commit f319da0c68 ("sched: Fix load avg vs cpu-hotplug") was an
incomplete fix:
In particular, the problem is that at the point it calls
calc_load_migrate() nr_running := 1 (the stopper thread), so move the
call to CPU_DEAD where we're sure that nr_running := 0.
Also note that we can call calc_load_migrate() without serialization, we
know the state of rq is stable since its cpu is dead, and we modify the
global state using appropriate atomic ops.
Suggested-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1346882630.2600.59.camel@twins
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Now that the last architecture to use this has stopped doing so (ARM,
thanks Catalin!) we can remove this complexity from the scheduler
core.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Link: http://lkml.kernel.org/n/tip-g9p2a1w81xxbrze25v9zpzbf@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
On tickless systems, one CPU runs load balance for all idle CPUs.
The cpu_load of this CPU is updated before starting the load balance
of each other idle CPUs. We should instead update the cpu_load of
the balance_cpu.
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Venkatesh Pallipadi <venki@google.com>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1347509486-8688-1-git-send-email-vincent.guittot@linaro.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
It's impossible to enter the else branch if we have set
skip_clock_update in task_yield_fair(), as yield_to_task_fair()
will directly return true after invoke task_yield_fair().
Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/4FF2925A.9060005@linux.vnet.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Various sd->*_idx's are used for refering the rq's load average table
when selecting a cpu to run. However they can be set to any number
with sysctl knobs so that it can crash the kernel if something bad is
given. Fix it by limiting them into the actual range.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1345104204-8317-1-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commit beac4c7e4a ("sched: Remove AFFINE_WAKEUPS feature") removed
use of the flag but left the definition. Get rid of it.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Link: http://lkml.kernel.org/r/1345090865-20851-1-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Fix two kernel-doc warnings in kernel/sched/fair.c:
Warning(kernel/sched/fair.c:3660): Excess function parameter 'cpus' description in 'update_sg_lb_stats'
Warning(kernel/sched/fair.c:3806): Excess function parameter 'cpus' description in 'update_sd_lb_stats'
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/50303714.3090204@xenotime.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
migrate_tasks() uses _pick_next_task_rt() to get tasks from the
real-time runqueues to be migrated. When rt_rq is throttled
_pick_next_task_rt() won't return anything, in which case
migrate_tasks() can't move all threads over and gets stuck in an
infinite loop.
Instead unthrottle rt runqueues before migrating tasks.
Additionally: move unthrottle_offline_cfs_rqs() to rq_offline_fair()
Signed-off-by: Peter Boonstoppel <pboonstoppel@nvidia.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Turner <pjt@google.com>
Link: http://lkml.kernel.org/r/5FBF8E85CA34454794F0F7ECBA79798F379D3648B7@HQMAIL04.nvidia.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Rabik and Paul reported two different issues related to the same few
lines of code.
Rabik's issue is that the nr_uninterruptible migration code is wrong in
that he sees artifacts due to this (Rabik please do expand in more
detail).
Paul's issue is that this code as it stands relies on us using
stop_machine() for unplug, we all would like to remove this assumption
so that eventually we can remove this stop_machine() usage altogether.
The only reason we'd have to migrate nr_uninterruptible is so that we
could use for_each_online_cpu() loops in favour of
for_each_possible_cpu() loops, however since nr_uninterruptible() is the
only such loop and its using possible lets not bother at all.
The problem Rabik sees is (probably) caused by the fact that by
migrating nr_uninterruptible we screw rq->calc_load_active for both rqs
involved.
So don't bother with fancy migration schemes (meaning we now have to
keep using for_each_possible_cpu()) and instead fold any nr_active delta
after we migrate all tasks away to make sure we don't have any skewed
nr_active accounting.
Reported-by: Rakib Mullick <rakib.mullick@gmail.com>
Reported-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1345454817.23018.27.camel@twins
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull scheduler fixes from Ingo Molnar.
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched: Fix migration thread runtime bogosity
sched,rt: fix isolated CPUs leaving root_task_group indefinitely throttled
sched,cgroup: Fix up task_groups list
sched: fix divide by zero at {thread_group,task}_times
sched, cgroup: Reduce rq->lock hold times for large cgroup hierarchies
The archs that implement virtual cputime accounting all
flush the cputime of a task when it gets descheduled
and sometimes set up some ground initialization for the
next task to account its cputime.
These archs all put their own hooks in their context
switch callbacks and handle the off-case themselves.
Consolidate this by creating a new account_switch_vtime()
callback called in generic code right after a context switch
and that these archs must implement to flush the prev task
cputime and initialize the next task cputime related state.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Extract cputime code from the giant sched/core.c and
put it in its own file. This make it easier to deal with
this particular area and de-bloat a bit more core.c
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Since power saving code was removed from sched now, the implement
code is out of service in this function, and even pollute other logical.
like, 'want_sd' never has chance to be set '0', that remove the effect
of SD_WAKE_AFFINE here.
So, clean up the obsolete code, includes SD_PREFER_LOCAL.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/5028F431.6000306@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
As we already have dst_rq in lb_env, using or changing "this_rq" do not
make sense.
This patch will replace "this_rq" with dst_rq in load_balance, and we
don't need to change "this_rq" while process LBF_SOME_PINNED any more.
Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/501F8357.3070102@linux.vnet.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
This patch adds a comment on top of the schedule() function to explain
to scheduler newbies how the main scheduler function is entered.
Acked-by: Randy Dunlap <rdunlap@xenotime.net>
Explained-by: Ingo Molnar <mingo@kernel.org>
Explained-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344070187-2420-1-git-send-email-penberg@kernel.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
It should be sched_nr_latency so fix it before it annoys me more.
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344435364-18632-1-git-send-email-bp@amd64.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Make stop scheduler class do the same accounting as other classes,
Migration threads can be caught in the act while doing exec balancing,
leading to the below due to use of unmaintained ->se.exec_start. The
load that triggered this particular instance was an apparently out of
control heavily threaded application that does system monitoring in
what equated to an exec bomb, with one of the VERY frequently migrated
tasks being ps.
%CPU PID USER CMD
99.3 45 root [migration/10]
97.7 53 root [migration/12]
97.0 57 root [migration/13]
90.1 49 root [migration/11]
89.6 65 root [migration/15]
88.7 17 root [migration/3]
80.4 37 root [migration/8]
78.1 41 root [migration/9]
44.2 13 root [migration/2]
Signed-off-by: Mike Galbraith <mgalbraith@suse.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344051854.6739.19.camel@marge.simpson.net
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Root task group bandwidth replenishment must service all CPUs, regardless of
where the timer was last started, and regardless of the isolation mechanism,
lest 'Quoth the Raven, "Nevermore"' become rt scheduling policy.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344326558.6968.25.camel@marge.simpson.net
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
With multiple instances of task_groups, for_each_rt_rq() is a noop,
no task groups having been added to the rt.c list instance. This
renders __enable/disable_runtime() and print_rt_stats() noop, the
user (non) visible effect being that rt task groups are missing in
/proc/sched_debug.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Cc: stable@kernel.org # v3.3+
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344308413.6846.7.camel@marge.simpson.net
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
On architectures where cputime_t is 64 bit type, is possible to trigger
divide by zero on do_div(temp, (__force u32) total) line, if total is a
non zero number but has lower 32 bit's zeroed. Removing casting is not
a good solution since some do_div() implementations do cast to u32
internally.
This problem can be triggered in practice on very long lived processes:
PID: 2331 TASK: ffff880472814b00 CPU: 2 COMMAND: "oraagent.bin"
#0 [ffff880472a51b70] machine_kexec at ffffffff8103214b
#1 [ffff880472a51bd0] crash_kexec at ffffffff810b91c2
#2 [ffff880472a51ca0] oops_end at ffffffff814f0b00
#3 [ffff880472a51cd0] die at ffffffff8100f26b
#4 [ffff880472a51d00] do_trap at ffffffff814f03f4
#5 [ffff880472a51d60] do_divide_error at ffffffff8100cfff
#6 [ffff880472a51e00] divide_error at ffffffff8100be7b
[exception RIP: thread_group_times+0x56]
RIP: ffffffff81056a16 RSP: ffff880472a51eb8 RFLAGS: 00010046
RAX: bc3572c9fe12d194 RBX: ffff880874150800 RCX: 0000000110266fad
RDX: 0000000000000000 RSI: ffff880472a51eb8 RDI: 001038ae7d9633dc
RBP: ffff880472a51ef8 R8: 00000000b10a3a64 R9: ffff880874150800
R10: 00007fcba27ab680 R11: 0000000000000202 R12: ffff880472a51f08
R13: ffff880472a51f10 R14: 0000000000000000 R15: 0000000000000007
ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
#7 [ffff880472a51f00] do_sys_times at ffffffff8108845d
#8 [ffff880472a51f40] sys_times at ffffffff81088524
#9 [ffff880472a51f80] system_call_fastpath at ffffffff8100b0f2
RIP: 0000003808caac3a RSP: 00007fcba27ab6d8 RFLAGS: 00000202
RAX: 0000000000000064 RBX: ffffffff8100b0f2 RCX: 0000000000000000
RDX: 00007fcba27ab6e0 RSI: 000000000076d58e RDI: 00007fcba27ab6e0
RBP: 00007fcba27ab700 R8: 0000000000000020 R9: 000000000000091b
R10: 00007fcba27ab680 R11: 0000000000000202 R12: 00007fff9ca41940
R13: 0000000000000000 R14: 00007fcba27ac9c0 R15: 00007fff9ca41940
ORIG_RAX: 0000000000000064 CS: 0033 SS: 002b
Cc: stable@vger.kernel.org
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20120808092714.GA3580@redhat.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Peter Portante reported that for large cgroup hierarchies (and or on
large CPU counts) we get immense lock contention on rq->lock and stuff
stops working properly.
His workload was a ton of processes, each in their own cgroup,
everybody idling except for a sporadic wakeup once every so often.
It was found that:
schedule()
idle_balance()
load_balance()
local_irq_save()
double_rq_lock()
update_h_load()
walk_tg_tree(tg_load_down)
tg_load_down()
Results in an entire cgroup hierarchy walk under rq->lock for every
new-idle balance and since new-idle balance isn't throttled this
results in a lot of work while holding the rq->lock.
This patch does two things, it removes the work from under rq->lock
based on the good principle of race and pray which is widely employed
in the load-balancer as a whole. And secondly it throttles the
update_h_load() calculation to max once per jiffy.
I considered excluding update_h_load() for new-idle balance
all-together, but purely relying on regular balance passes to update
this data might not work out under some rare circumstances where the
new-idle busiest isn't the regular busiest for a while (unlikely, but
a nightmare to debug if someone hits it and suffers).
Cc: pjt@google.com
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Reported-by: Peter Portante <pportant@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-aaarrzfpnaam7pqrekofu8a6@git.kernel.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Pull scheduler fixes from Ingo Molnar:
"Fixes and two late cleanups"
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/cleanups: Add load balance cpumask pointer to 'struct lb_env'
sched: Fix comment about PREEMPT_ACTIVE bit location
sched: Fix minor code style issues
sched: Use task_rq_unlock() in __sched_setscheduler()
sched/numa: Add SD_PERFER_SIBLING to CPU domain
Pull perf updates from Ingo Molnar:
"The biggest changes are Intel Nehalem-EX PMU uncore support, uprobes
updates/cleanups/fixes from Oleg and diverse tooling updates (mostly
fixes) now that Arnaldo is back from vacation."
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (40 commits)
uprobes: __replace_page() needs munlock_vma_page()
uprobes: Rename vma_address() and make it return "unsigned long"
uprobes: Fix register_for_each_vma()->vma_address() check
uprobes: Introduce vaddr_to_offset(vma, vaddr)
uprobes: Teach build_probe_list() to consider the range
uprobes: Remove insert_vm_struct()->uprobe_mmap()
uprobes: Remove copy_vma()->uprobe_mmap()
uprobes: Fix overflow in vma_address()/find_active_uprobe()
uprobes: Suppress uprobe_munmap() from mmput()
uprobes: Uprobe_mmap/munmap needs list_for_each_entry_safe()
uprobes: Clean up and document write_opcode()->lock_page(old_page)
uprobes: Kill write_opcode()->lock_page(new_page)
uprobes: __replace_page() should not use page_address_in_vma()
uprobes: Don't recheck vma/f_mapping in write_opcode()
perf/x86: Fix missing struct before structure name
perf/x86: Fix format definition of SNB-EP uncore QPI box
perf/x86: Make bitfield unsigned
perf/x86: Fix LLC-* and node-* events on Intel SandyBridge
perf/x86: Add Intel Nehalem-EX uncore support
perf/x86: Fix typo in format definition of uncore PCU filter
...
