If a cpu is offline, its nid will be set to -1, and cpu_to_node(cpu)
will return -1. As a result, cpumask_of_node(nid) will return NULL. In
this case, find_next_bit() in for_each_cpu will get a NULL pointer and
cause panic.
Here is a call trace:
Call Trace:
<IRQ>
select_fallback_rq+0x71/0x190
try_to_wake_up+0x2cb/0x2f0
wake_up_process+0x15/0x20
hrtimer_wakeup+0x22/0x30
__run_hrtimer+0x83/0x320
hrtimer_interrupt+0x106/0x280
smp_apic_timer_interrupt+0x69/0x99
apic_timer_interrupt+0x6f/0x80
There is a hrtimer process sleeping, whose cpu has already been
offlined. When it is waken up, it tries to find another cpu to run, and
get a -1 nid. As a result, cpumask_of_node(-1) returns NULL, and causes
ernel panic.
This patch fixes this problem by judging if the nid is -1. If nid is
not -1, a cpu on the same node will be picked. Else, a online cpu on
another node will be picked.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Jiang Liu <liuj97@gmail.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull cgroup changes from Tejun Heo:
"Nothing too drastic.
- Removal of synchronize_rcu() from userland visible paths.
- Various fixes and cleanups from Li.
- cgroup_rightmost_descendant() added which will be used by cpuset
changes (it will be a separate pull request)."
* 'for-3.9' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
cgroup: fail if monitored file and event_control are in different cgroup
cgroup: fix cgroup_rmdir() vs close(eventfd) race
cpuset: fix cpuset_print_task_mems_allowed() vs rename() race
cgroup: fix exit() vs rmdir() race
cgroup: remove bogus comments in cgroup_diput()
cgroup: remove synchronize_rcu() from cgroup_diput()
cgroup: remove duplicate RCU free on struct cgroup
sched: remove redundant NULL cgroup check in task_group_path()
sched: split out css_online/css_offline from tg creation/destruction
cgroup: initialize cgrp->dentry before css_alloc()
cgroup: remove a NULL check in cgroup_exit()
cgroup: fix bogus kernel warnings when cgroup_create() failed
cgroup: remove synchronize_rcu() from rebind_subsystems()
cgroup: remove synchronize_rcu() from cgroup_attach_{task|proc}()
cgroup: use new hashtable implementation
cgroups: fix cgroup_event_listener error handling
cgroups: move cgroup_event_listener.c to tools/cgroup
cgroup: implement cgroup_rightmost_descendant()
cgroup: remove unused dummy cgroup_fork_callbacks()
Pull workqueue changes from Tejun Heo:
"A lot of reorganization is going on mostly to prepare for worker pools
with custom attributes so that workqueue can replace custom pool
implementations in places including writeback and btrfs and make CPU
assignment in crypto more flexible.
workqueue evolved from purely per-cpu design and implementation, so
there are a lot of assumptions regarding being bound to CPUs and even
unbound workqueues are implemented as an extension of the model -
workqueues running on the special unbound CPU. Bulk of changes this
round are about promoting worker_pools as the top level abstraction
replacing global_cwq (global cpu workqueue). At this point, I'm
fairly confident about getting custom worker pools working pretty soon
and ready for the next merge window.
Lai's patches are replacing the convoluted mb() dancing workqueue has
been doing with much simpler mechanism which only depends on
assignment atomicity of long. For details, please read the commit
message of 0b3dae68ac ("workqueue: simplify is-work-item-queued-here
test"). While the change ends up adding one pointer to struct
delayed_work, the inflation in percentage is less than five percent
and it decouples delayed_work logic a lot more cleaner from usual work
handling, removes the unusual memory barrier dancing, and allows for
further simplification, so I think the trade-off is acceptable.
There will be two more workqueue related pull requests and there are
some shared commits among them. I'll write further pull requests
assuming this pull request is pulled first."
* 'for-3.9' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq: (37 commits)
workqueue: un-GPL function delayed_work_timer_fn()
workqueue: rename cpu_workqueue to pool_workqueue
workqueue: reimplement is_chained_work() using current_wq_worker()
workqueue: fix is_chained_work() regression
workqueue: pick cwq instead of pool in __queue_work()
workqueue: make get_work_pool_id() cheaper
workqueue: move nr_running into worker_pool
workqueue: cosmetic update in try_to_grab_pending()
workqueue: simplify is-work-item-queued-here test
workqueue: make work->data point to pool after try_to_grab_pending()
workqueue: add delayed_work->wq to simplify reentrancy handling
workqueue: make work_busy() test WORK_STRUCT_PENDING first
workqueue: replace WORK_CPU_NONE/LAST with WORK_CPU_END
workqueue: post global_cwq removal cleanups
workqueue: rename nr_running variables
workqueue: remove global_cwq
workqueue: remove worker_pool->gcwq
workqueue: replace for_each_worker_pool() with for_each_std_worker_pool()
workqueue: make freezing/thawing per-pool
workqueue: make hotplug processing per-pool
...
Pull scheduler changes from Ingo Molnar:
"Main changes:
- scheduler side full-dynticks (user-space execution is undisturbed
and receives no timer IRQs) preparation changes that convert the
cputime accounting code to be full-dynticks ready, from Frederic
Weisbecker.
- Initial sched.h split-up changes, by Clark Williams
- select_idle_sibling() performance improvement by Mike Galbraith:
" 1 tbench pair (worst case) in a 10 core + SMT package:
pre 15.22 MB/sec 1 procs
post 252.01 MB/sec 1 procs "
- sched_rr_get_interval() ABI fix/change. We think this detail is not
used by apps (so it's not an ABI in practice), but lets keep it
under observation.
- misc RT scheduling cleanups, optimizations"
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (24 commits)
sched/rt: Add <linux/sched/rt.h> header to <linux/init_task.h>
cputime: Remove irqsave from seqlock readers
sched, powerpc: Fix sched.h split-up build failure
cputime: Restore CPU_ACCOUNTING config defaults for PPC64
sched/rt: Move rt specific bits into new header file
sched/rt: Add a tuning knob to allow changing SCHED_RR timeslice
sched: Move sched.h sysctl bits into separate header
sched: Fix signedness bug in yield_to()
sched: Fix select_idle_sibling() bouncing cow syndrome
sched/rt: Further simplify pick_rt_task()
sched/rt: Do not account zero delta_exec in update_curr_rt()
cputime: Safely read cputime of full dynticks CPUs
kvm: Prepare to add generic guest entry/exit callbacks
cputime: Use accessors to read task cputime stats
cputime: Allow dynamic switch between tick/virtual based cputime accounting
cputime: Generic on-demand virtual cputime accounting
cputime: Move default nsecs_to_cputime() to jiffies based cputime file
cputime: Librarize per nsecs resolution cputime definitions
cputime: Avoid multiplication overflow on utime scaling
context_tracking: Export context state for generic vtime
...
Fix up conflict in kernel/context_tracking.c due to comment additions.
The reader side code has no requirement to disable interrupts while
sampling data. The sequence counter is enough to ensure consistency.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Move rt scheduler definitions out of include/linux/sched.h into
new file include/linux/sched/rt.h
Signed-off-by: Clark Williams <williams@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/20130207094707.7b9f825f@riff.lan
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add a /proc/sys/kernel scheduler knob named
sched_rr_timeslice_ms that allows global changing of the
SCHED_RR timeslice value. User visable value is in milliseconds
but is stored as jiffies. Setting to 0 (zero) resets to the
default (currently 100ms).
Signed-off-by: Clark Williams <williams@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/20130207094704.13751796@riff.lan
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Move the sysctl-related bits from include/linux/sched.h into
a new file: include/linux/sched/sysctl.h. Then update source
files requiring access to those bits by including the new
header file.
Signed-off-by: Clark Williams <williams@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/20130207094659.06dced96@riff.lan
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Typical cputime stats infrastructure relies on the timer tick and
its periodic polling on the CPU to account the amount of time
spent by the CPUs and the tasks per high level domains such as
userspace, kernelspace, guest, ...
Now we are preparing to implement full dynticks capability on
Linux for Real Time and HPC users who want full CPU isolation.
This feature requires a cputime accounting that doesn't depend
on the timer tick.
To implement it, this new cputime infrastructure plugs into
kernel/user/guest boundaries to take snapshots of cputime and
flush these to the stats when needed. This performs pretty
much like CONFIG_VIRT_CPU_ACCOUNTING except that context location
and cputime snaphots are synchronized between write and read
side such that the latter can safely retrieve the pending tickless
cputime of a task and add it to its latest cputime snapshot to
return the correct result to the user.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)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=1Tzp
-----END PGP SIGNATURE-----
Merge tag 'full-dynticks-cputime-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks into sched/core
Pull full-dynticks (user-space execution is undisturbed and
receives no timer IRQs) preparation changes that convert the
cputime accounting code to be full-dynticks ready,
from Frederic Weisbecker:
"This implements the cputime accounting on full dynticks CPUs.
Typical cputime stats infrastructure relies on the timer tick and
its periodic polling on the CPU to account the amount of time
spent by the CPUs and the tasks per high level domains such as
userspace, kernelspace, guest, ...
Now we are preparing to implement full dynticks capability on
Linux for Real Time and HPC users who want full CPU isolation.
This feature requires a cputime accounting that doesn't depend
on the timer tick.
To implement it, this new cputime infrastructure plugs into
kernel/user/guest boundaries to take snapshots of cputime and
flush these to the stats when needed. This performs pretty
much like CONFIG_VIRT_CPU_ACCOUNTING except that context location
and cputime snaphots are synchronized between write and read
side such that the latter can safely retrieve the pending tickless
cputime of a task and add it to its latest cputime snapshot to
return the correct result to the user."
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In 7b270f6099 "sched: Bail out of yield_to when source and
target runqueue has one task" we changed this to store -ESRCH so
it needs to be signed.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: kbuild@01.org
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Mike Galbraith <efault@gmx.de>
Link: http://lkml.kernel.org/r/20130205113751.GA20521@elgon.mountain
Signed-off-by: Ingo Molnar <mingo@kernel.org>
If the previous CPU is cache affine and idle, select it.
The current implementation simply traverses the sd_llc domain,
taking the first idle CPU encountered, which walks buddy pairs
hand in hand over the package, inflicting excruciating pain.
1 tbench pair (worst case) in a 10 core + SMT package:
pre 15.22 MB/sec 1 procs
post 252.01 MB/sec 1 procs
Signed-off-by: Mike Galbraith <bitbucket@online.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1359371965.5783.127.camel@marge.simpson.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Function next_prio() has been removed and pull_rt_task() is the
only user of pick_next_highest_task_rt() at the moment.
pull_rt_task is not interested in p->nr_cpus_allowed, its only
interest is the fact that cpu is allowed to execute p. If
nr_cpus_allowed == 1, cpu != task_cpu(p) and cpu is allowed then
it means that task p is in the middle of the migration
techniques; the task waits until it is moved by migration
thread. So, lets pull it earlier.
Signed-off-by: Kirill V Tkhai <tkhai@yandex.ru>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
CC: linux-rt-users <linux-rt-users@vger.kernel.org>
Link: http://lkml.kernel.org/r/70871359644177@web16d.yandex.ru
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There are several places of consecutive calls of
dequeue_task_rt() and put_prev_task_rt() in the scheduler.
For example, function rt_mutex_setprio() does it.
The both calls lead to update_curr_rt(), the second of it
receives zeroed delta_exec. The only effective action in this
case is call of sched_rt_avg_update(), which can change
rq->age_stamp and rq->rt_avg. But it is possible in case of
""floating"" rq->clock. This fact is not reasonable to be
accounted. Another actions do nothing.
Signed-off-by: Kirill V Tkhai <tkhai@yandex.ru>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
CC: linux-rt-users <linux-rt-users@vger.kernel.org>
Link: http://lkml.kernel.org/r/931541359550236@web1g.yandex.ru
Signed-off-by: Ingo Molnar <mingo@kernel.org>
In case of undercomitted scenarios, especially in large guests
yield_to overhead is significantly high. when run queue length of
source and target is one, take an opportunity to bail out and return
-ESRCH. This return condition can be further exploited to quickly come
out of PLE handler.
(History: Raghavendra initially worked on break out of kvm ple handler upon
seeing source runqueue length = 1, but it had to export rq length).
Peter came up with the elegant idea of return -ESRCH in scheduler core.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Raghavendra, Checking the rq length of target vcpu condition added.(thanks Avi)
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Acked-by: Andrew Jones <drjones@redhat.com>
Tested-by: Chegu Vinod <chegu_vinod@hp.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
While remotely reading the cputime of a task running in a
full dynticks CPU, the values stored in utime/stime fields
of struct task_struct may be stale. Its values may be those
of the last kernel <-> user transition time snapshot and
we need to add the tickless time spent since this snapshot.
To fix this, flush the cputime of the dynticks CPUs on
kernel <-> user transition and record the time / context
where we did this. Then on top of this snapshot and the current
time, perform the fixup on the reader side from task_times()
accessors.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
[fixed kvm module related build errors]
Signed-off-by: Sedat Dilek <sedat.dilek@gmail.com>
Do some ground preparatory work before adding guest_enter()
and guest_exit() context tracking callbacks. Those will
be later used to read the guest cputime safely when we
run in full dynticks mode.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
This is in preparation for the full dynticks feature. While
remotely reading the cputime of a task running in a full
dynticks CPU, we'll need to do some extra-computation. This
way we can account the time it spent tickless in userspace
since its last cputime snapshot.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Allow to dynamically switch between tick and virtual based
cputime accounting. This way we can provide a kind of "on-demand"
virtual based cputime accounting. In this mode, the kernel relies
on the context tracking subsystem to dynamically probe on kernel
boundaries.
This is in preparation for being able to stop the timer tick in
more places than just the idle state. Doing so will depend on
CONFIG_VIRT_CPU_ACCOUNTING_GEN which makes it possible to account
the cputime without the tick by hooking on kernel/user boundaries.
Depending whether the tick is stopped or not, we can switch between
tick and vtime based accounting anytime in order to minimize the
overhead associated to user hooks.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
If we want to stop the tick further idle, we need to be
able to account the cputime without using the tick.
Virtual based cputime accounting solves that problem by
hooking into kernel/user boundaries.
However implementing CONFIG_VIRT_CPU_ACCOUNTING require
low level hooks and involves more overhead. But we already
have a generic context tracking subsystem that is required
for RCU needs by archs which plan to shut down the tick
outside idle.
This patch implements a generic virtual based cputime
accounting that relies on these generic kernel/user hooks.
There are some upsides of doing this:
- This requires no arch code to implement CONFIG_VIRT_CPU_ACCOUNTING
if context tracking is already built (already necessary for RCU in full
tickless mode).
- We can rely on the generic context tracking subsystem to dynamically
(de)activate the hooks, so that we can switch anytime between virtual
and tick based accounting. This way we don't have the overhead
of the virtual accounting when the tick is running periodically.
And one downside:
- There is probably more overhead than a native virtual based cputime
accounting. But this relies on hooks that are already set anyway.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
If the architecture doesn't provide an implementation of
nsecs_to_cputime(), the cputime accounting core uses a
default one that converts the nanoseconds to jiffies. However
this only makes sense if we use the jiffies based cputime.
For now it doesn't matter much because this API is only
called on code that uses jiffies based cputime accounting.
But the code may evolve and this API may be used more
broadly in the future. Keeping this default implementation
around is very error prone as it may introduce a bug and
hide it on architectures that don't override this API.
Fix this by moving this definition to the jiffies based
cputime headers as it is the only place where it belongs to.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
We scale stime, utime values based on rtime (sum_exec_runtime
converted to jiffies). During scaling we multiple rtime * utime,
which seems to be fine, since both values are converted to u64,
but it's not.
Let assume HZ is 1000 - 1ms tick. Process consist of 64 threads,
run for 1 day, threads utilize 100% cpu on user space. Machine
has 64 cpus.
Process rtime = utime will be 64 * 24 * 60 * 60 * 1000 jiffies,
which is 0x149970000. Multiplication rtime * utime result is
0x1a855771100000000, which can not be covered in 64 bits.
Result of overflow is stall of utime values visible in user
space (prev_utime in kernel), even if application still consume
lot of CPU time.
A solution to solve this is to perform the multiplication on
stime instead of utime. It's easy to grow the utime value fast
with a CPU bound thread in userspace for example. Now we assume
that doing so with stime is much harder. In most cases a task
shouldn't ever spend much time in kernel space as it tends to
sleep waiting for jobs completion when they take long to
achieve. IO is the typical example of that.
Hence scaling the cputime by performing the multiplication on
stime instead of utime should considerably reduce the chances of
an overflow on most workloads.
This is largely inspired by a patch from Stanislaw Gruszka:
http://lkml.kernel.org/r/20130107113144.GA7544@redhat.com
Inspired-by: Stanislaw Gruszka <sgruszka@redhat.com>
Reported-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/1359217182-25184-1-git-send-email-fweisbec@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The type returned from atomic64_t can be either unsigned
long or unsigned long long, depending on the architecture.
Using a cast to unsigned long long lets us use the same
format string for all architectures.
Without this patch, building with scheduler debugging
enabled results in:
kernel/sched/debug.c: In function 'print_cfs_rq':
kernel/sched/debug.c:225:2: warning: format '%ld' expects argument of type 'long int', but argument 4 has type 'long long int' [-Wformat]
kernel/sched/debug.c:225:2: warning: format '%ld' expects argument of type 'long int', but argument 3 has type 'long long int' [-Wformat]
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Turner <pjt@google.com>
Cc: linux-arm-kernel@list.infradead.org
Link: http://lkml.kernel.org/r/1359123276-15833-7-git-send-email-arnd@arndb.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
a4c96ae319 "sched: Unthrottle rt runqueues in
__disable_runtime()" turned the unthrottle_offline_cfs_rqs
function into a static symbol, which now triggers a warning
about it being potentially unused:
kernel/sched/fair.c:2055:13: warning: 'unthrottle_offline_cfs_rqs' defined but not used [-Wunused-function]
Marking it __maybe_unused shuts up the gcc warning and lets the
compiler safely drop the function body when it's not being used.
To reproduce, build the ARM bcm2835_defconfig.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Peter Boonstoppel <pboonstoppel@nvidia.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Turner <pjt@google.com>
Cc: linux-arm-kernel@list.infradead.org
Link: http://lkml.kernel.org/r/1359123276-15833-6-git-send-email-arnd@arndb.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The issue below was found in 2.6.34-rt rather than mainline rt
kernel, but the issue still exists upstream as well.
So please let me describe how it was noticed on 2.6.34-rt:
On this version, each softirq has its own thread, it means there
is at least one RT FIFO task per cpu. The priority of these
tasks is set to 49 by default. If user launches an RT FIFO task
with priority lower than 49 of softirq RT tasks, it's possible
there are two RT FIFO tasks enqueued one cpu runqueue at one
moment. By current strategy of balancing RT tasks, when it comes
to RT tasks, we really need to put them off to a CPU that they
can run on as soon as possible. Even if it means a bit of cache
line flushing, we want RT tasks to be run with the least latency.
When the user RT FIFO task which just launched before is
running, the sched timer tick of the current cpu happens. In this
tick period, the timeout value of the user RT task will be
updated once. Subsequently, we try to wake up one softirq RT
task on its local cpu. As the priority of current user RT task
is lower than the softirq RT task, the current task will be
preempted by the higher priority softirq RT task. Before
preemption, we check to see if current can readily move to a
different cpu. If so, we will reschedule to allow the RT push logic
to try to move current somewhere else. Whenever the woken
softirq RT task runs, it first tries to migrate the user FIFO RT
task over to a cpu that is running a task of lesser priority. If
migration is done, it will send a reschedule request to the found
cpu by IPI interrupt. Once the target cpu responds the IPI
interrupt, it will pick the migrated user RT task to preempt its
current task. When the user RT task is running on the new cpu,
the sched timer tick of the cpu fires. So it will tick the user
RT task again. This also means the RT task timeout value will be
updated again. As the migration may be done in one tick period,
it means the user RT task timeout value will be updated twice
within one tick.
If we set a limit on the amount of cpu time for the user RT task
by setrlimit(RLIMIT_RTTIME), the SIGXCPU signal should be posted
upon reaching the soft limit.
But exactly when the SIGXCPU signal should be sent depends on the
RT task timeout value. In fact the timeout mechanism of sending
the SIGXCPU signal assumes the RT task timeout is increased once
every tick.
However, currently the timeout value may be added twice per
tick. So it results in the SIGXCPU signal being sent earlier
than expected.
To solve this issue, we prevent the timeout value from increasing
twice within one tick time by remembering the jiffies value of
last updating the timeout. As long as the RT task's jiffies is
different with the global jiffies value, we allow its timeout to
be updated.
Signed-off-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Fan Du <fan.du@windriver.com>
Reviewed-by: Yong Zhang <yong.zhang0@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1342508623-2887-1-git-send-email-ying.xue@windriver.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When the system has multiple domains do_sched_rt_period_timer()
can run on any CPU and may iterate over all rt_rq in
cpu_online_mask. This means when balance_runtime() is run for a
given rt_rq that rt_rq may be in a different rd than the current
processor. Thus if we use smp_processor_id() to get rd in
do_balance_runtime() we may borrow runtime from a rt_rq that is
not part of our rd.
This changes do_balance_runtime to get the rd from the passed in
rt_rq ensuring that we borrow runtime only from the correct rd
for the given rt_rq.
This fixes a BUG at kernel/sched/rt.c:687! in __disable_runtime
when we try reclaim runtime lent to other rt_rq but runtime has
been lent to a rt_rq in another rd.
Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Mike Galbraith <bitbucket@online.de>
Cc: peterz@infradead.org
Cc: <stable@kernel.org>
Link: http://lkml.kernel.org/r/1358186131-29494-1-git-send-email-sbohrer@rgmadvisors.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
A task_group won't be online (thus no one can see it) until
cpu_cgroup_css_online(), and at that time tg->css.cgroup has
been initialized, so this NULL check is redundant.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
This is a preparaton for later patches.
- What do we gain from cpu_cgroup_css_online():
After ss->css_alloc() and before ss->css_online(), there's a small
window that tg->css.cgroup is NULL. With this change, tg won't be seen
before ss->css_online(), where it's added to the global list, so we're
guaranteed we'll never see NULL tg->css.cgroup.
- What do we gain from cpu_cgroup_css_offline():
tg is freed via RCU, so is cgroup. Without this change, This is how
synchronization works:
cgroup_rmdir()
no ss->css_offline()
diput()
syncornize_rcu()
ss->css_free() <-- unregister tg, and free it via call_rcu()
kfree_rcu(cgroup) <-- wait possible refs to cgroup, and free cgroup
We can't just kfree(cgroup), because tg might access tg->css.cgroup.
With this change:
cgroup_rmdir()
ss->css_offline() <-- unregister tg
diput()
synchronize_rcu() <-- wait possible refs to tg and cgroup
ss->css_free() <-- free tg
kfree_rcu(cgroup) <-- free cgroup
As you see, kfree_rcu() is redundant now.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
Reschedule rq->curr if the first RT task has just been
pulled to the rq.
Signed-off-by: Kirill V Tkhai <tkhai@yandex.ru>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tkhai Kirill <tkhai@yandex.ru>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/118761353614535@web28f.yandex.ru
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The caller of sched_sliced() should pass se.cfs_rq and se as the
arguments, however in sched_rr_get_interval() we gave it
rq.cfs_rq and se, which made the following computation obviously
wrong.
The change was introduced by commit:
77034937dc sched: fix crash in sys_sched_rr_get_interval()
... 5 years ago, while it had been the correct 'cfs_rq_of' before
the commit. The change seems to be irrelevant to the commit
msg, which was to return a 0 timeslice for tasks that are on an
idle runqueue. So I believe that was just a plain typo.
Signed-off-by: Zhu Yanhai <gaoyang.zyh@taobao.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Turner <pjt@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1357621012-15039-1-git-send-email-gaoyang.zyh@taobao.com
[ Since this is an ABI and an old bug, we'll test this via a
slow upstream route, to hopefully discover any app breakage. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
wake_up_process() should never wakeup a TASK_STOPPED/TRACED task.
Change it to use TASK_NORMAL and add the WARN_ON().
TASK_ALL has no other users, probably can be killed.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Workqueue wants to expose more interface internal to kernel/. Instead
of adding a new header file, repurpose kernel/workqueue_sched.h.
Rename it to workqueue_internal.h and add include protector.
This patch doesn't introduce any functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
task_numa_placement() oopsed on NULL p->mm when task_numa_fault() got
called in the handling of break_ksm() for ksmd. That might be a
peculiar case, which perhaps KSM could takes steps to avoid? but it's
more robust if task_numa_placement() allows for such a possibility.
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull user namespace changes from Eric Biederman:
"While small this set of changes is very significant with respect to
containers in general and user namespaces in particular. The user
space interface is now complete.
This set of changes adds support for unprivileged users to create user
namespaces and as a user namespace root to create other namespaces.
The tyranny of supporting suid root preventing unprivileged users from
using cool new kernel features is broken.
This set of changes completes the work on setns, adding support for
the pid, user, mount namespaces.
This set of changes includes a bunch of basic pid namespace
cleanups/simplifications. Of particular significance is the rework of
the pid namespace cleanup so it no longer requires sending out
tendrils into all kinds of unexpected cleanup paths for operation. At
least one case of broken error handling is fixed by this cleanup.
The files under /proc/<pid>/ns/ have been converted from regular files
to magic symlinks which prevents incorrect caching by the VFS,
ensuring the files always refer to the namespace the process is
currently using and ensuring that the ptrace_mayaccess permission
checks are always applied.
The files under /proc/<pid>/ns/ have been given stable inode numbers
so it is now possible to see if different processes share the same
namespaces.
Through the David Miller's net tree are changes to relax many of the
permission checks in the networking stack to allowing the user
namespace root to usefully use the networking stack. Similar changes
for the mount namespace and the pid namespace are coming through my
tree.
Two small changes to add user namespace support were commited here adn
in David Miller's -net tree so that I could complete the work on the
/proc/<pid>/ns/ files in this tree.
Work remains to make it safe to build user namespaces and 9p, afs,
ceph, cifs, coda, gfs2, ncpfs, nfs, nfsd, ocfs2, and xfs so the
Kconfig guard remains in place preventing that user namespaces from
being built when any of those filesystems are enabled.
Future design work remains to allow root users outside of the initial
user namespace to mount more than just /proc and /sys."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace: (38 commits)
proc: Usable inode numbers for the namespace file descriptors.
proc: Fix the namespace inode permission checks.
proc: Generalize proc inode allocation
userns: Allow unprivilged mounts of proc and sysfs
userns: For /proc/self/{uid,gid}_map derive the lower userns from the struct file
procfs: Print task uids and gids in the userns that opened the proc file
userns: Implement unshare of the user namespace
userns: Implent proc namespace operations
userns: Kill task_user_ns
userns: Make create_new_namespaces take a user_ns parameter
userns: Allow unprivileged use of setns.
userns: Allow unprivileged users to create new namespaces
userns: Allow setting a userns mapping to your current uid.
userns: Allow chown and setgid preservation
userns: Allow unprivileged users to create user namespaces.
userns: Ignore suid and sgid on binaries if the uid or gid can not be mapped
userns: fix return value on mntns_install() failure
vfs: Allow unprivileged manipulation of the mount namespace.
vfs: Only support slave subtrees across different user namespaces
vfs: Add a user namespace reference from struct mnt_namespace
...
Michal Hocko reported that the following build error occurs if
CONFIG_NUMA_BALANCING is set without THP support
kernel/sched/fair.c: In function ‘task_numa_work’:
kernel/sched/fair.c:932:55: error: call to ‘__build_bug_failed’ declared with attribute error: BUILD_BUG failed
The problem is that HPAGE_PMD_SHIFT triggers a BUILD_BUG() on
!CONFIG_TRANSPARENT_HUGEPAGE. This patch addresses the problem.
Reported-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.18 (GNU/Linux)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=ivQa
-----END PGP SIGNATURE-----
Merge tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma
Pull Automatic NUMA Balancing bare-bones from Mel Gorman:
"There are three implementations for NUMA balancing, this tree
(balancenuma), numacore which has been developed in tip/master and
autonuma which is in aa.git.
In almost all respects balancenuma is the dumbest of the three because
its main impact is on the VM side with no attempt to be smart about
scheduling. In the interest of getting the ball rolling, it would be
desirable to see this much merged for 3.8 with the view to building
scheduler smarts on top and adapting the VM where required for 3.9.
The most recent set of comparisons available from different people are
mel: https://lkml.org/lkml/2012/12/9/108
mingo: https://lkml.org/lkml/2012/12/7/331
tglx: https://lkml.org/lkml/2012/12/10/437
srikar: https://lkml.org/lkml/2012/12/10/397
The results are a mixed bag. In my own tests, balancenuma does
reasonably well. It's dumb as rocks and does not regress against
mainline. On the other hand, Ingo's tests shows that balancenuma is
incapable of converging for this workloads driven by perf which is bad
but is potentially explained by the lack of scheduler smarts. Thomas'
results show balancenuma improves on mainline but falls far short of
numacore or autonuma. Srikar's results indicate we all suffer on a
large machine with imbalanced node sizes.
My own testing showed that recent numacore results have improved
dramatically, particularly in the last week but not universally.
We've butted heads heavily on system CPU usage and high levels of
migration even when it shows that overall performance is better.
There are also cases where it regresses. Of interest is that for
specjbb in some configurations it will regress for lower numbers of
warehouses and show gains for higher numbers which is not reported by
the tool by default and sometimes missed in treports. Recently I
reported for numacore that the JVM was crashing with
NullPointerExceptions but currently it's unclear what the source of
this problem is. Initially I thought it was in how numacore batch
handles PTEs but I'm no longer think this is the case. It's possible
numacore is just able to trigger it due to higher rates of migration.
These reports were quite late in the cycle so I/we would like to start
with this tree as it contains much of the code we can agree on and has
not changed significantly over the last 2-3 weeks."
* tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma: (50 commits)
mm/rmap, migration: Make rmap_walk_anon() and try_to_unmap_anon() more scalable
mm/rmap: Convert the struct anon_vma::mutex to an rwsem
mm: migrate: Account a transhuge page properly when rate limiting
mm: numa: Account for failed allocations and isolations as migration failures
mm: numa: Add THP migration for the NUMA working set scanning fault case build fix
mm: numa: Add THP migration for the NUMA working set scanning fault case.
mm: sched: numa: Delay PTE scanning until a task is scheduled on a new node
mm: sched: numa: Control enabling and disabling of NUMA balancing if !SCHED_DEBUG
mm: sched: numa: Control enabling and disabling of NUMA balancing
mm: sched: Adapt the scanning rate if a NUMA hinting fault does not migrate
mm: numa: Use a two-stage filter to restrict pages being migrated for unlikely task<->node relationships
mm: numa: migrate: Set last_nid on newly allocated page
mm: numa: split_huge_page: Transfer last_nid on tail page
mm: numa: Introduce last_nid to the page frame
sched: numa: Slowly increase the scanning period as NUMA faults are handled
mm: numa: Rate limit setting of pte_numa if node is saturated
mm: numa: Rate limit the amount of memory that is migrated between nodes
mm: numa: Structures for Migrate On Fault per NUMA migration rate limiting
mm: numa: Migrate pages handled during a pmd_numa hinting fault
mm: numa: Migrate on reference policy
...
This reverts commit f269ae0469.
It turns out it causes a very noticeable interactivity regression with
CONFIG_SCHED_AUTOGROUP (test-case: "make -j32" of the kernel in a
terminal window, while scrolling in a browser - the autogrouping means
that the two end up in separate cgroups, and the browser should be
smooth as silk despite the high load).
Says Paul Turner:
"It seems that the update-throttling on the wake-side is reducing the
interactive tasks' ability to preempt. While I suspect the right
longer term answer here is force these updates only in the
cross-cgroup case; this is less trivial. For this release I believe
the right answer is either going to be a revert or restore the updates
on the enqueue-side."
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Bisected-by: Mike Galbraith <efault@gmx.de>
Acked-by: Paul Turner <pjt@google.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull KVM updates from Marcelo Tosatti:
"Considerable KVM/PPC work, x86 kvmclock vsyscall support,
IA32_TSC_ADJUST MSR emulation, amongst others."
Fix up trivial conflict in kernel/sched/core.c due to cross-cpu
migration notifier added next to rq migration call-back.
* tag 'kvm-3.8-1' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (156 commits)
KVM: emulator: fix real mode segment checks in address linearization
VMX: remove unneeded enable_unrestricted_guest check
KVM: VMX: fix DPL during entry to protected mode
x86/kexec: crash_vmclear_local_vmcss needs __rcu
kvm: Fix irqfd resampler list walk
KVM: VMX: provide the vmclear function and a bitmap to support VMCLEAR in kdump
x86/kexec: VMCLEAR VMCSs loaded on all cpus if necessary
KVM: MMU: optimize for set_spte
KVM: PPC: booke: Get/set guest EPCR register using ONE_REG interface
KVM: PPC: bookehv: Add EPCR support in mtspr/mfspr emulation
KVM: PPC: bookehv: Add guest computation mode for irq delivery
KVM: PPC: Make EPCR a valid field for booke64 and bookehv
KVM: PPC: booke: Extend MAS2 EPN mask for 64-bit
KVM: PPC: e500: Mask MAS2 EPN high 32-bits in 32/64 tlbwe emulation
KVM: PPC: Mask ea's high 32-bits in 32/64 instr emulation
KVM: PPC: e500: Add emulation helper for getting instruction ea
KVM: PPC: bookehv64: Add support for interrupt handling
KVM: PPC: bookehv: Remove GET_VCPU macro from exception handler
KVM: PPC: booke: Fix get_tb() compile error on 64-bit
KVM: PPC: e500: Silence bogus GCC warning in tlb code
...
Pull cgroup changes from Tejun Heo:
"A lot of activities on cgroup side. The big changes are focused on
making cgroup hierarchy handling saner.
- cgroup_rmdir() had peculiar semantics - it allowed cgroup
destruction to be vetoed by individual controllers and tried to
drain refcnt synchronously. The vetoing never worked properly and
caused good deal of contortions in cgroup. memcg was the last
reamining user. Michal Hocko removed the usage and cgroup_rmdir()
path has been simplified significantly. This was done in a
separate branch so that the memcg people can base further memcg
changes on top.
- The above allowed cleaning up cgroup lifecycle management and
implementation of generic cgroup iterators which are used to
improve hierarchy support.
- cgroup_freezer updated to allow migration in and out of a frozen
cgroup and handle hierarchy. If a cgroup is frozen, all descendant
cgroups are frozen.
- netcls_cgroup and netprio_cgroup updated to handle hierarchy
properly.
- Various fixes and cleanups.
- Two merge commits. One to pull in memcg and rmdir cleanups (needed
to build iterators). The other pulled in cgroup/for-3.7-fixes for
device_cgroup fixes so that further device_cgroup patches can be
stacked on top."
Fixed up a trivial conflict in mm/memcontrol.c as per Tejun (due to
commit bea8c150a7 ("memcg: fix hotplugged memory zone oops") in master
touching code close to commit 2ef37d3fe4 ("memcg: Simplify
mem_cgroup_force_empty_list error handling") in for-3.8)
* 'for-3.8' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup: (65 commits)
cgroup: update Documentation/cgroups/00-INDEX
cgroup_rm_file: don't delete the uncreated files
cgroup: remove subsystem files when remounting cgroup
cgroup: use cgroup_addrm_files() in cgroup_clear_directory()
cgroup: warn about broken hierarchies only after css_online
cgroup: list_del_init() on removed events
cgroup: fix lockdep warning for event_control
cgroup: move list add after list head initilization
netprio_cgroup: allow nesting and inherit config on cgroup creation
netprio_cgroup: implement netprio[_set]_prio() helpers
netprio_cgroup: use cgroup->id instead of cgroup_netprio_state->prioidx
netprio_cgroup: reimplement priomap expansion
netprio_cgroup: shorten variable names in extend_netdev_table()
netprio_cgroup: simplify write_priomap()
netcls_cgroup: move config inheritance to ->css_online() and remove .broken_hierarchy marking
cgroup: remove obsolete guarantee from cgroup_task_migrate.
cgroup: add cgroup->id
cgroup, cpuset: remove cgroup_subsys->post_clone()
cgroup: s/CGRP_CLONE_CHILDREN/CGRP_CPUSET_CLONE_CHILDREN/
cgroup: rename ->create/post_create/pre_destroy/destroy() to ->css_alloc/online/offline/free()
...
Pull scheduler updates from Ingo Molnar:
"The biggest change affects group scheduling: we now track the runnable
average on a per-task entity basis, allowing a smoother, exponential
decay average based load/weight estimation instead of the previous
binary on-the-runqueue/off-the-runqueue load weight method.
This will inevitably disturb workloads that were in some sort of
borderline balancing state or unstable equilibrium, so an eye has to
be kept on regressions.
For that reason the new load average is only limited to group
scheduling (shares distribution) at the moment (which was also hurting
the most from the prior, crude weight calculation and whose scheduling
quality wins most from this change) - but we plan to extend this to
regular SMP balancing as well in the future, which will simplify and
speed up things a bit.
Other changes involve ongoing preparatory work to extend NOHZ to the
scheduler as well, eventually allowing completely irq-free user-space
execution."
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (33 commits)
Revert "sched/autogroup: Fix crash on reboot when autogroup is disabled"
cputime: Comment cputime's adjusting code
cputime: Consolidate cputime adjustment code
cputime: Rename thread_group_times to thread_group_cputime_adjusted
cputime: Move thread_group_cputime() to sched code
vtime: Warn if irqs aren't disabled on system time accounting APIs
vtime: No need to disable irqs on vtime_account()
vtime: Consolidate a bit the ctx switch code
vtime: Explicitly account pending user time on process tick
vtime: Remove the underscore prefix invasion
sched/autogroup: Fix crash on reboot when autogroup is disabled
cputime: Separate irqtime accounting from generic vtime
cputime: Specialize irq vtime hooks
kvm: Directly account vtime to system on guest switch
vtime: Make vtime_account_system() irqsafe
vtime: Gather vtime declarations to their own header file
sched: Describe CFS load-balancer
sched: Introduce temporary FAIR_GROUP_SCHED dependency for load-tracking
sched: Make __update_entity_runnable_avg() fast
sched: Update_cfs_shares at period edge
...
Pull RCU update from Ingo Molnar:
"The major features of this tree are:
1. A first version of no-callbacks CPUs. This version prohibits
offlining CPU 0, but only when enabled via CONFIG_RCU_NOCB_CPU=y.
Relaxing this constraint is in progress, but not yet ready
for prime time. These commits were posted to LKML at
https://lkml.org/lkml/2012/10/30/724.
2. Changes to SRCU that allows statically initialized srcu_struct
structures. These commits were posted to LKML at
https://lkml.org/lkml/2012/10/30/296.
3. Restructuring of RCU's debugfs output. These commits were posted
to LKML at https://lkml.org/lkml/2012/10/30/341.
4. Additional CPU-hotplug/RCU improvements, posted to LKML at
https://lkml.org/lkml/2012/10/30/327.
Note that the commit eliminating __stop_machine() was judged to
be too-high of risk, so is deferred to 3.9.
5. Changes to RCU's idle interface, most notably a new module
parameter that redirects normal grace-period operations to
their expedited equivalents. These were posted to LKML at
https://lkml.org/lkml/2012/10/30/739.
6. Additional diagnostics for RCU's CPU stall warning facility,
posted to LKML at https://lkml.org/lkml/2012/10/30/315.
The most notable change reduces the
default RCU CPU stall-warning time from 60 seconds to 21 seconds,
so that it once again happens sooner than the softlockup timeout.
7. Documentation updates, which were posted to LKML at
https://lkml.org/lkml/2012/10/30/280.
A couple of late-breaking changes were posted at
https://lkml.org/lkml/2012/11/16/634 and
https://lkml.org/lkml/2012/11/16/547.
8. Miscellaneous fixes, which were posted to LKML at
https://lkml.org/lkml/2012/10/30/309.
9. Finally, a fix for an lockdep-RCU splat was posted to LKML
at https://lkml.org/lkml/2012/11/7/486."
* 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (49 commits)
context_tracking: New context tracking susbsystem
sched: Mark RCU reader in sched_show_task()
rcu: Separate accounting of callbacks from callback-free CPUs
rcu: Add callback-free CPUs
rcu: Add documentation for the new rcuexp debugfs trace file
rcu: Update documentation for TREE_RCU debugfs tracing
rcu: Reduce default RCU CPU stall warning timeout
rcu: Fix TINY_RCU rcu_is_cpu_rrupt_from_idle check
rcu: Clarify memory-ordering properties of grace-period primitives
rcu: Add new rcutorture module parameters to start/end test messages
rcu: Remove list_for_each_continue_rcu()
rcu: Fix batch-limit size problem
rcu: Add tracing for synchronize_sched_expedited()
rcu: Remove old debugfs interfaces and also RCU flavor name
rcu: split 'rcuhier' to each flavor
rcu: split 'rcugp' to each flavor
rcu: split 'rcuboost' to each flavor
rcu: split 'rcubarrier' to each flavor
rcu: Fix tracing formatting
rcu: Remove the interface "rcudata.csv"
...
Due to the fact that migrations are driven by the CPU a task is running
on there is no point tracking NUMA faults until one task runs on a new
node. This patch tracks the first node used by an address space. Until
it changes, PTE scanning is disabled and no NUMA hinting faults are
trapped. This should help workloads that are short-lived, do not care
about NUMA placement or have bound themselves to a single node.
This takes advantage of the logic in "mm: sched: numa: Implement slow
start for working set sampling" to delay when the checks are made. This
will take advantage of processes that set their CPU and node bindings
early in their lifetime. It will also potentially allow any initial load
balancing to take place.
Signed-off-by: Mel Gorman <mgorman@suse.de>
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>
