Normally the RT bandwidth scheme will share bandwidth across the
entire root_domain. However sometimes its convenient to disable this
sharing for debug purposes. Provide a simple feature switch to this
end.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Remove the WAKEUP_PREEMPT feature, disabling it doesn't make any sense
and its outlived its use by a long long while.
Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110729082033.GB12106@zhy
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Allow for sched_domain spans that overlap by giving such domains their
own sched_group list instead of sharing the sched_groups amongst
each-other.
This is needed for machines with more than 16 nodes, because
sched_domain_node_span() will generate a node mask from the
16 nearest nodes without regard if these masks have any overlap.
Currently sched_domains have a sched_group that maps to their child
sched_domain span, and since there is no overlap we share the
sched_group between the sched_domains of the various CPUs. If however
there is overlap, we would need to link the sched_group list in
different ways for each cpu, and hence sharing isn't possible.
In order to solve this, allocate private sched_groups for each CPU's
sched_domain but have the sched_groups share a sched_group_power
structure such that we can uniquely track the power.
Reported-and-tested-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/n/tip-08bxqw9wis3qti9u5inifh3y@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch makes update_rq_clock() aware of steal time.
The mechanism of operation is not different from irq_time,
and follows the same principles. This lives in a CONFIG
option itself, and can be compiled out independently of
the rest of steal time reporting. The effect of disabling it
is that the scheduler will still report steal time (that cannot be
disabled), but won't use this information for cpu power adjustments.
Everytime update_rq_clock_task() is invoked, we query information
about how much time was stolen since last call, and feed it into
sched_rt_avg_update().
Although steal time reporting in account_process_tick() keeps
track of the last time we read the steal clock, in prev_steal_time,
this patch do it independently using another field,
prev_steal_time_rq. This is because otherwise, information about time
accounted in update_process_tick() would never reach us in update_rq_clock().
Signed-off-by: Glauber Costa <glommer@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Eric B Munson <emunson@mgebm.net>
CC: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
CC: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Now that we've removed the rq->lock requirement from the first part of
ttwu() and can compute placement without holding any rq->lock, ensure
we execute the second half of ttwu() on the actual cpu we want the
task to run on.
This avoids having to take rq->lock and doing the task enqueue
remotely, saving lots on cacheline transfers.
As measured using: http://oss.oracle.com/~mason/sembench.c
$ for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i; done
$ echo 4096 32000 64 128 > /proc/sys/kernel/sem
$ ./sembench -t 2048 -w 1900 -o 0
unpatched: run time 30 seconds 647278 worker burns per second
patched: run time 30 seconds 816715 worker burns per second
Reviewed-by: Frank Rowand <frank.rowand@am.sony.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110405152729.515897185@chello.nl
By tracking a per-cpu load-avg for each cfs_rq and folding it into a
global task_group load on each tick we can rework tg_shares_up to be
strictly per-cpu.
This should improve cpu-cgroup performance for smp systems
significantly.
[ Paul: changed to use queueing cfs_rq + bug fixes ]
Signed-off-by: Paul Turner <pjt@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20101115234937.580480400@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The idea was suggested by Peter Zijlstra here:
http://marc.info/?l=linux-kernel&m=127476934517534&w=2
irq time is technically not available to the tasks running on the CPU.
This patch removes irq time from CPU power piggybacking on
sched_rt_avg_update().
Tested this by keeping CPU X busy with a network intensive task having 75%
oa a single CPU irq processing (hard+soft) on a 4-way system. And start seven
cycle soakers on the system. Without this change, there will be two tasks on
each CPU. With this change, there is a single task on irq busy CPU X and
remaining 7 tasks are spread around among other 3 CPUs.
Signed-off-by: Venkatesh Pallipadi <venki@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1286237003-12406-8-git-send-email-venki@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This features has been enabled for quite a while, after testing showed that
easing preemption for light tasks was harmful to high priority threads.
Remove the feature flag.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301675.6785.44.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Sync wakeups are critical functionality with a long history. Remove it, we don't
need the branch or icache footprint.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301817.6785.47.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This feature never earned its keep, remove it.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301591.6785.42.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Our preemption model relies too heavily on sleeper fairness to disable it
without dire consequences. Remove the feature, and save a branch or two.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301520.6785.40.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This feature hasn't been enabled in a long time, remove effectively dead code.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301447.6785.38.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Both avg_overlap and avg_wakeup had an inherent problem in that their accuracy
was detrimentally affected by cross-cpu wakeups, this because we are missing
the necessary call to update_curr(). This can't be fixed without increasing
overhead in our already too fat fastpath.
Additionally, with recent load balancing changes making us prefer to place tasks
in an idle cache domain (which is good for compute bound loads), communicating
tasks suffer when a sync wakeup, which would enable affine placement, is turned
into a non-sync wakeup by SYNC_LESS. With one task on the runqueue, wake_affine()
rejects the affine wakeup request, leaving the unfortunate where placed, taking
frequent cache misses.
Remove it, and recover some fastpath cycles.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301121.6785.30.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Testing the load which led to this heuristic (nfs4 kbuild) shows that it has
outlived it's usefullness. With intervening load balancing changes, I cannot
see any difference with/without, so recover there fastpath cycles.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1268301062.6785.29.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
WAKEUP_RUNNING was an experiment, not sure why that ever ended up being
merged...
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Create a new wakeup preemption mode, preempt towards tasks that run
shorter on avg. It sets next buddy to be sure we actually run the task
we preempted for.
Test results:
root@twins:~# while :; do :; done &
[1] 6537
root@twins:~# while :; do :; done &
[2] 6538
root@twins:~# while :; do :; done &
[3] 6539
root@twins:~# while :; do :; done &
[4] 6540
root@twins:/home/peter# ./latt -c4 sleep 4
Entries: 48 (clients=4)
Averages:
------------------------------
Max 4750 usec
Avg 497 usec
Stdev 737 usec
root@twins:/home/peter# echo WAKEUP_RUNNING > /debug/sched_features
root@twins:/home/peter# ./latt -c4 sleep 4
Entries: 48 (clients=4)
Averages:
------------------------------
Max 14 usec
Avg 5 usec
Stdev 3 usec
Disabled by default - needs more testing.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
LKML-Reference: <new-submission>
We don't need to call update_shares() for each domain we iterate,
just got the largets one.
However, we should call it before wake_affine() as well, so that
that can use up-to-date values too.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add back FAIR_SLEEPERS and GENTLE_FAIR_SLEEPERS.
FAIR_SLEEPERS is the old logic: credit sleepers with their sleep time.
GENTLE_FAIR_SLEEPERS dampens this a bit: 50% of their sleep time gets
credited.
The hope here is to still give the benefits of fair-sleepers logic
(quick wakeups, etc.) while not allow them to have 100% of their
sleep time as if they were running.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Currently we use overlap to weaken the SYNC hint, but allow it to
set the hint as well.
echo NO_SYNC_WAKEUP > /debug/sched_features
echo SYNC_MORE > /debug/sched_features
preserves pipe-test behaviour without using the WF_SYNC hint.
Worth playing with on more workloads...
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
I suspect a feed-back loop between cpuidle and the aperf/mperf
cpu_power bits, where when we have idle C-states lower the ratio,
which leads to lower cpu_power and then less load, which generates
more idle time, etc..
Put in a knob to disable it.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Make the idle balancer more agressive, to improve a
x264 encoding workload provided by Jason Garrett-Glaser:
NEXT_BUDDY NO_LB_BIAS
encoded 600 frames, 252.82 fps, 22096.60 kb/s
encoded 600 frames, 250.69 fps, 22096.60 kb/s
encoded 600 frames, 245.76 fps, 22096.60 kb/s
NO_NEXT_BUDDY LB_BIAS
encoded 600 frames, 344.44 fps, 22096.60 kb/s
encoded 600 frames, 346.66 fps, 22096.60 kb/s
encoded 600 frames, 352.59 fps, 22096.60 kb/s
NO_NEXT_BUDDY NO_LB_BIAS
encoded 600 frames, 425.75 fps, 22096.60 kb/s
encoded 600 frames, 425.45 fps, 22096.60 kb/s
encoded 600 frames, 422.49 fps, 22096.60 kb/s
Peter pointed out that this is better done via newidle_idx,
not via LB_BIAS, newidle balancing should look for where
there is load _now_, not where there was load 2 ticks ago.
Worst-case latencies are improved as well as no buddies
means less vruntime spread. (as per prior lkml discussions)
This change improves kbuild-peak parallelism as well.
Reported-by: Jason Garrett-Glaser <darkshikari@gmail.com>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1253011667.9128.16.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add a NEXT_BUDDY feature flag to aid in debugging.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
It consists of two conditions, split them out in separate toggles
so we can test them independently.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Nikos Chantziaras and Jens Axboe reported that turning off
NEW_FAIR_SLEEPERS improves desktop interactivity visibly.
Nikos described his experiences the following way:
" With this setting, I can do "nice -n 19 make -j20" and
still have a very smooth desktop and watch a movie at
the same time. Various other annoyances (like the
"logout/shutdown/restart" dialog of KDE not appearing
at all until the background fade-out effect has finished)
are also gone. So this seems to be the single most
important setting that vastly improves desktop behavior,
at least here. "
Jens described it the following way, referring to a 10-seconds
xmodmap scheduling delay he was trying to debug:
" Then I tried switching NO_NEW_FAIR_SLEEPERS on, and then
I get:
Performance counter stats for 'xmodmap .xmodmap-carl':
9.009137 task-clock-msecs # 0.447 CPUs
18 context-switches # 0.002 M/sec
1 CPU-migrations # 0.000 M/sec
315 page-faults # 0.035 M/sec
0.020167093 seconds time elapsed
Woot! "
So disable it for now. In perf trace output i can see weird
delta timestamps:
cc1-9943 [001] 2802.059479616: sched_stat_wait: task: as:9944 wait: 2801938766276 [ns]
That nsec field is not supposed to be that large. More digging
is needed - but lets turn it off while the real bug is found.
Reported-by: Nikos Chantziaras <realnc@arcor.de>
Tested-by: Nikos Chantziaras <realnc@arcor.de>
Reported-by: Jens Axboe <jens.axboe@oracle.com>
Tested-by: Jens Axboe <jens.axboe@oracle.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <4AA93D34.8040500@arcor.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Prefer tasks that wake other tasks to preempt quickly. This improves
performance because more work is available sooner.
The workload that prompted this patch was a kernel build over NFS4 (for some
curious and not understood reason we had to revert commit:
18de973530 to make any progress at all)
Without this patch a make -j8 bzImage (of x86-64 defconfig) would take
3m30-ish, with this patch we're down to 2m50-ish.
psql-sysbench/mysql-sysbench show a slight improvement in peak performance as
well, tbench and vmark seemed to not care.
It is possible to improve upon the build time (to 2m20-ish) but that seriously
destroys other benchmarks (just shows that there's more room for tinkering).
Much thanks to Mike who put in a lot of effort to benchmark things and proved
a worthy opponent with a competing patch.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Change mutex contention behaviour such that it will sometimes busy wait on
acquisition - moving its behaviour closer to that of spinlocks.
This concept got ported to mainline from the -rt tree, where it was originally
implemented for rtmutexes by Steven Rostedt, based on work by Gregory Haskins.
Testing with Ingo's test-mutex application (http://lkml.org/lkml/2006/1/8/50)
gave a 345% boost for VFS scalability on my testbox:
# ./test-mutex-shm V 16 10 | grep "^avg ops"
avg ops/sec: 296604
# ./test-mutex-shm V 16 10 | grep "^avg ops"
avg ops/sec: 85870
The key criteria for the busy wait is that the lock owner has to be running on
a (different) cpu. The idea is that as long as the owner is running, there is a
fair chance it'll release the lock soon, and thus we'll be better off spinning
instead of blocking/scheduling.
Since regular mutexes (as opposed to rtmutexes) do not atomically track the
owner, we add the owner in a non-atomic fashion and deal with the races in
the slowpath.
Furthermore, to ease the testing of the performance impact of this new code,
there is means to disable this behaviour runtime (without having to reboot
the system), when scheduler debugging is enabled (CONFIG_SCHED_DEBUG=y),
by issuing the following command:
# echo NO_OWNER_SPIN > /debug/sched_features
This command re-enables spinning again (this is also the default):
# echo OWNER_SPIN > /debug/sched_features
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: improve/change/fix wakeup-buddy scheduling
Currently we only have a forward looking buddy, that is, we prefer to
schedule to the task we last woke up, under the presumption that its
going to consume the data we just produced, and therefore will have
cache hot benefits.
This allows co-waking producer/consumer task pairs to run ahead of the
pack for a little while, keeping their cache warm. Without this, we
would interleave all pairs, utterly trashing the cache.
This patch introduces a backward looking buddy, that is, suppose that
in the above scenario, the consumer preempts the producer before it
can go to sleep, we will therefore miss the wakeup from consumer to
producer (its already running, after all), breaking the cycle and
reverting to the cache-trashing interleaved schedule pattern.
The backward buddy will try to schedule back to the task that woke us
up in case the forward buddy is not available, under the assumption
that the last task will be the one with the most cache hot task around
barring current.
This will basically allow a task to continue after it got preempted.
In order to avoid starvation, we allow either buddy to get wakeup_gran
ahead of the pack.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
David Miller reported that hrtick update overhead has tripled the
wakeup overhead on Sparc64.
That is too much - disable the HRTICK feature for now by default,
until a faster implementation is found.
Reported-by: David Miller <davem@davemloft.net>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Lin Ming reported a 10% OLTP regression against 2.6.27-rc4.
The difference seems to come from different preemption agressiveness,
which affects the cache footprint of the workload and its effective
cache trashing.
Aggresively preempt a task if its avg overlap is very small, this should
avoid the task going to sleep and find it still running when we schedule
back to it - saving a wakeup.
Reported-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Yanmin reported a significant regression on his 16-core machine due to:
commit 93b75217df
Author: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Fri Jun 27 13:41:33 2008 +0200
Flip back to the old behaviour.
Reported-by: "Zhang, Yanmin" <yanmin_zhang@linux.intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Measurement shows that the difference between cgroup:/ and cgroup:/foo
wake_affine() results is that the latter succeeds significantly more.
Therefore bias the calculations towards failing the test.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We found that the affine wakeup code needs rather accurate load figures
to be effective. The trouble is that updating the load figures is fairly
expensive with group scheduling. Therefore ratelimit the updating.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The bias given by source/target_load functions can be very large, disable
it by default to get faster convergence.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
calc_delta_asym() is supposed to do the same as calc_delta_fair() except
linearly shrink the result for negative nice processes - this causes them
to have a smaller preemption threshold so that they are more easily preempted.
The problem is that for task groups se->load.weight is the per cpu share of
the actual task group weight; take that into account.
Also provide a debug switch to disable the asymmetry (which I still don't
like - but it does greatly benefit some workloads)
This would explain the interactivity issues reported against group scheduling.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
provide a text based interface to the scheduler features; this saves the
'user' from setting bits using decimal arithmetic.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Peter Zijlstra
Yong Zhang
Linus Torvalds
Glauber Costa
Venkatesh Pallipadi
Mike Galbraith
Peter Zijlstra
Ingo Molnar