Clean up the new affine to idle sibling bits while trying to
grok them. Should not have any function differences.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20091112145610.832503781@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When waking affine, check for an idle shared cache, and if
found, wake to that CPU/sibling instead of the waker's CPU.
This improves pgsql+oltp ramp up by roughly 8%. Possibly more
for other loads, depending on overlap. The trade-off is a
roughly 1% peak downturn if tasks are truly synchronous.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: <stable@kernel.org>
LKML-Reference: <1256654138.17752.7.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch restores the effectiveness of LAST_BUDDY in preventing
pgsql+oltp from collapsing due to wakeup preemption. It also
switches LAST_BUDDY to exclusively do what it does best, namely
mitigate the effects of aggressive wakeup preemption, which
improves vmark throughput markedly, and restores mysql+oltp
scalability.
Since buddies are about scalability, enable them beginning at the
point where we begin expanding sched_latency, namely
sched_nr_latency. Previously, buddies were cleared aggressively,
which seriously reduced their effectiveness. Not clearing
aggressively however, produces a small drop in mysql+oltp
throughput immediately after peak, indicating that LAST_BUDDY is
actually doing some harm. This is right at the point where X on the
desktop in competition with another load wants low latency service.
Ergo, do not enable until we need to scale.
To mitigate latency induced by buddies, or by a task just missing
wakeup preemption, check latency at tick time.
Last hunk prevents buddies from stymieing BALANCE_NEWIDLE via
CACHE_HOT_BUDDY.
Supporting performance tests:
tip = v2.6.32-rc5-1497-ga525b32
tipx = NO_GENTLE_FAIR_SLEEPERS NEXT_BUDDY granularity knobs = 31 knobs + 31 buddies
tip+x = NO_GENTLE_FAIR_SLEEPERS granularity knobs = 31 knobs
(Three run averages except where noted.)
vmark:
------
tip 108466 messages per second
tip+ 125307 messages per second
tip+x 125335 messages per second
tipx 117781 messages per second
2.6.31.3 122729 messages per second
mysql+oltp:
-----------
clients 1 2 4 8 16 32 64 128 256
..........................................................................................
tip 9949.89 18690.20 34801.24 34460.04 32682.88 30765.97 28305.27 25059.64 19548.08
tip+ 10013.90 18526.84 34900.38 34420.14 33069.83 32083.40 30578.30 28010.71 25605.47
tipx 9698.71 18002.70 34477.56 33420.01 32634.30 31657.27 29932.67 26827.52 21487.18
2.6.31.3 8243.11 18784.20 34404.83 33148.38 31900.32 31161.90 29663.81 25995.94 18058.86
pgsql+oltp:
-----------
clients 1 2 4 8 16 32 64 128 256
..........................................................................................
tip 13686.37 26609.25 51934.28 51347.81 49479.51 45312.65 36691.91 26851.57 24145.35
tip+ (1x) 13907.85 27135.87 52951.98 52514.04 51742.52 50705.43 49947.97 48374.19 46227.94
tip+x 13906.78 27065.81 52951.19 52542.59 52176.11 51815.94 50838.90 49439.46 46891.00
tipx 13742.46 26769.81 52351.99 51891.73 51320.79 50938.98 50248.65 48908.70 46553.84
2.6.31.3 13815.35 26906.46 52683.34 52061.31 51937.10 51376.80 50474.28 49394.47 47003.25
Signed-off-by: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Yanmin reported a hackbench regression due to:
> commit de69a80be3
> Author: Peter Zijlstra <a.p.zijlstra@chello.nl>
> Date: Thu Sep 17 09:01:20 2009 +0200
>
> sched: Stop buddies from hogging the system
I really liked de69a80b, and it affecting hackbench shows I wasn't
crazy ;-)
So hackbench is a multi-cast, with one sender spraying multiple
receivers, who in their turn don't spray back.
This would be exactly the scenario that patch 'cures'. Previously
we would not clear the last buddy after running the next task,
allowing the sender to get back to work sooner than it otherwise
ought to have been, increasing latencies for other tasks.
Now, since those receivers don't poke back, they don't enforce the
buddy relation, which means there's nothing to re-elect the sender.
Cure this by less agressively clearing the buddy stats. Only clear
buddies when they were not chosen. It should still avoid a buddy
sticking around long after its served its time.
Reported-by: "Zhang, Yanmin" <yanmin_zhang@linux.intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
CC: Mike Galbraith <efault@gmx.de>
LKML-Reference: <1255084986.8802.46.camel@laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
It's unused.
It isn't needed -- read or write flag is already passed and sysctl
shouldn't care about the rest.
It _was_ used in two places at arch/frv for some reason.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: James Morris <jmorris@namei.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: Simplify sys_sched_rr_get_interval() system call
sched: Fix potential NULL derference of doms_cur
sched: Fix raciness in runqueue_is_locked()
sched: Re-add lost cpu_allowed check to sched_fair.c::select_task_rq_fair()
sched: Remove unneeded indentation in sched_fair.c::place_entity()
By removing the need for it to know details of scheduling classes.
This allows PlugSched to define orthogonal scheduling classes.
Signed-off-by: Peter Williams <pwil3058@bigpond.net.au>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <06d1b89ee15a0eef82d7.1253496713@mudlark.pw.nest>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
While doing some testing, I pinned mplayer, only to find it
following X around like a puppy. Looking at commit c88d591, I found
a cpu_allowed check that went AWOL. I plugged it back in where it
looks like it needs to go, and now when I say "sit, stay!", mplayer
obeys again.
'c88d591 sched: Merge select_task_rq_fair() and
sched_balance_self()' accidentally dropped the check, causing
wake_affine() to pull pinned tasks - put it back.
[ v2: use a cheaper version from Peter ]
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The SD_POWERSAVING_BALANCE|SD_PREFER_LOCAL code can break out of
the domain iteration early, making us miss the SD_WAKE_AFFINE bits.
Fix this by continuing iteration until there is no need for a
larger domain.
This also cleans up the cgroup stuff a bit, but not having two
update_shares() invocations.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Clear buddies more agressively.
The (theoretical, haven't actually observed any of this) problem is
that when we do not select either buddy in pick_next_entity()
because they are too far ahead of the left-most task, we do not
clear the buddies.
This means that as soon as we service the left-most task, these
same buddies will be tried again on the next schedule. Now if the
left-most task was a pure hog, it wouldn't have done any wakeups
and it wouldn't have set buddies of its own. That leads to the old
buddies dominating, which would lead to bad latencies.
Signed-off-by: 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>
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>
Clean up the code a little.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
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>
And turn it on for NUMA and MC domains. This improves
locality in balancing decisions by keeping up to
capacity amount of tasks local before looking for idle
CPUs. (and twice the capacity if SD_POWERSAVINGS_BALANCE
is set.)
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
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>
Avoid the cache buddies from biasing the time distribution away
from fork()ers. Normally the next buddy will be the preferred
scheduling target, but this makes fork()s prefer to run the new
child, whereas we prefer to run the parent, since that will
generate more work.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to extend the functions to have more than 1 flag (sync),
rename the argument to flags, and explicitly define a WF_ space for
individual flags.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to be able to rename the sync argument, we need to rename
the current flag argument.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When merging select_task_rq_fair() and sched_balance_self() we lost
the use of wake_idx, restore that and set them to 0 to make wake
balancing more aggressive.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
While merging select_task_rq_fair() and sched_balance_self() I made
a mistake that leads to testing the wrong task affinty.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
for_each_domain() uses RCU to serialize the sched_domains, except
it doesn't actually use rcu_read_lock() and instead relies on
disabling preemption -> FAIL.
XXX: audit other sched_domain code.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
One of the problems of power-saving balancing is that under certain
scenarios it is too slow and allows tons of real work to pile up.
Avoid this by ignoring the powersave stuff when there's real work
to be done.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The problem with wake_idle() is that is doesn't respect things like
cpu_power, which means it doesn't deal well with SMT nor the recent
RT interaction.
To cure this, it needs to do what sched_balance_self() does, which
leads to the possibility of merging select_task_rq_fair() and
sched_balance_self().
Modify sched_balance_self() to:
- update_shares() when walking up the domain tree,
(it only called it for the top domain, but it should
have done this anyway), which allows us to remove
this ugly bit from try_to_wake_up().
- do wake_affine() on the smallest domain that contains
both this (the waking) and the prev (the wakee) cpu for
WAKE invocations.
Then use the top-down balance steps it had to replace wake_idle().
This leads to the dissapearance of SD_WAKE_BALANCE and
SD_WAKE_IDLE_FAR, with SD_WAKE_IDLE replaced with SD_BALANCE_WAKE.
SD_WAKE_AFFINE needs SD_BALANCE_WAKE to be effective.
Touch all topology bits to replace the old with new SD flags --
platforms might need re-tuning, enabling SD_BALANCE_WAKE
conditionally on a NUMA distance seems like a good additional
feature, magny-core and small nehalem systems would want this
enabled, systems with slow interconnects would not.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Rather ugly patch to fully place the sched_balance_self() code
inside the fair class.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Move the sched_balance_self() code into sched_fair.c
This facilitates the merger of sched_balance_self() and
sched_fair::select_task_rq().
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
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>
This allows more precise tracking of how the scheduler accounts
(and acts upon) a task having spent N nanoseconds of CPU time.
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This weird perf trace output:
cc1-9943 [001] 2802.059479616: sched_stat_wait: task: as:9944 wait: 2801938766276 [ns]
Is caused by setting one component field of the delta to zero
a bit too early. Move it to later.
( Note, this does not affect the NEW_FAIR_SLEEPERS interactivity bug,
it's just a reporting bug in essence. )
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Nikos Chantziaras <realnc@arcor.de>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <4AA93D34.8040500@arcor.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Reduce the latency target from 20 msecs to 5 msecs.
Why? Larger latencies increase spread, which is good for scaling,
but bad for worst case latency.
We still have the ilog(nr_cpus) rule to scale up on bigger
server boxes.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1252486344.28645.18.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Set child_runs_first default to off.
It hurts 'optimal' make -j<NR_CPUS> workloads as make jobs
get preempted by child tasks, reducing parallelism.
Note, this patch might make existing races in user
applications more prominent than before - so breakages
might be bisected to this commit.
Child-runs-first is broken on SMP to begin with, and we
already had it off briefly in v2.6.23 so most of the
offenders ought to be fixed. Would be nice not to revert
this commit but fix those apps finally ...
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1252486344.28645.18.camel@marge.simson.net>
[ made the sysctl independent of CONFIG_SCHED_DEBUG, in case
people want to work around broken apps. ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
A fork/exec load is usually "pass the baton", so the child
should never be placed behind the parent. With START_DEBIT we
make room for the new task, but with child_runs_first, that
room comes out of the _parent's_ hide. There's nothing to say
that the parent wasn't ahead of min_vruntime at fork() time,
which means that the "baton carrier", who is essentially the
parent in drag, can gain time and increase scheduling latencies
for waiters.
With NEW_FAIR_SLEEPERS + START_DEBIT + child_runs_first
enabled, we essentially pass the sleeper fairness off to the
child, which is fine, but if we don't base placement on the
parent's updated vruntime, we can end up compounding latency
woes if the child itself then does fork/exec. The debit
incurred at fork doesn't hurt the parent who is then going to
sleep and maybe exit, but the child who acquires the error
harms all comers.
This improves latencies of make -j<n> kernel build workloads.
Reported-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
wake_affine() would always fail under low-load situations where
both prev and this were idle, because adding a single task will
always be a significant imbalance, even if there's nothing
around that could balance it.
Deal with this by allowing imbalance when there's nothing you
can do about it.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
select_task_rq_fair() incorrectly skips the wake_affine()
logic, remove this.
When prev_cpu == this_cpu, the code jumps straight to the
wake_idle() logic, this doesn't give the wake_affine() logic
the chance to pin the task to this cpu.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add 3 schedstat tracepoints to help account for wait-time,
sleep-time and iowait-time.
They can also be used as a perf-counter source to profile tasks
on these clocks.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
LKML-Reference: <new-submission>
[ build fix for the !CONFIG_SCHEDSTATS case ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
For counting how long an application has been waiting for
(disk) IO, there currently is only the HZ sample driven
information available, while for all other counters in this
class, a high resolution version is available via
CONFIG_SCHEDSTATS.
In order to make an improved bootchart tool possible, we also
need a higher resolution version of the iowait time.
This patch below adds this scheduler statistic to the kernel.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <4A64B813.1080506@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
A frequent mistake appears to be to call task_of() on a
scheduler entity that is not actually a task, which can result
in a wild pointer.
Add a check to catch these mistakes.
Suggested-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Reflect "active" cpus in the rq->rd->online field, instead of
the online_map.
The motivation is that things that use the root-domain code
(such as cpupri) only care about cpus classified as "active"
anyway. By synchronizing the root-domain state with the active
map, we allow several optimizations.
For instance, we can remove an extra cpumask_and from the
scheduler hotpath by utilizing rq->rd->online (since it is now
a cached version of cpu_active_map & rq->rd->span).
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20090730145723.25226.24493.stgit@dev.haskins.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The latencytop and sleep accounting code assumes that any
scheduler entity represents a task, this is not so.
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
I spotted two sites that didn't take vruntime wrap-around into
account. Fix these by creating a comparison helper that does do
so.
Signed-off-by: Fabio Checconi <fabio@gandalf.sssup.it>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
One of the isolation modifications for SCHED_IDLE is the
unitization of sleeper credit. However the check for this
assumes that the sched_entity we're placing always belongs to a
task.
This is potentially not true with group scheduling and leaves
us rummaging randomly when we try to pull the policy.
Signed-off-by: Paul Turner <pjt@google.com>
Cc: peterz@infradead.org
LKML-Reference: <alpine.DEB.1.00.0907101649570.29914@kitami.corp.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Access to local variable lw is aliased by usage of pointer load.
Access to pointer load in calc_delta_mine() happens when lw is
already out of scope.
[ Reported by static code analysis. ]
Signed-off-by: Christian Engelmayer <christian.engelmayer@frequentis.com>
LKML-Reference: <20090616103512.0c846e51@frequentis.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: micro-optimization
Under group scheduling we traverse up until we are at common siblings
to make the wakeup comparison on.
At this point however, they should have the same parent so continuing
to check up the tree is redundant.
Signed-off-by: Paul Turner <pjt@google.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <alpine.DEB.1.00.0904081520320.30317@kitami.corp.google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Intel reported a 10% regression (mysql+sysbench) on a 16-way machine
with these patches:
1596e29: sched: symmetric sync vs avg_overlap
d942fb6: sched: fix sync wakeups
Revert them.
Reported-by: "Zhang, Yanmin" <yanmin_zhang@linux.intel.com>
Bisected-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>
Similar to the previous patch, by not clearing buddies we can select entities
past their run quota, which can increase latency. This means we have to clear
group buddies as well.
Do not use the group clear for pick_next_task(), otherwise that'll get O(n^2).
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
It was noticed that a task could get re-elected past its run quota due to buddy
affinities. This could increase latency a little. Cure it by more aggresively
clearing buddy state.
We do so in two situations:
- when we force preempt
- when we select a buddy to run
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Pawel Dziekonski reported that the openssl benchmark and his
quantum chemistry application both show slowdowns due to the
scheduler under-parallelizing execution.
The reason are pipe wakeups still doing 'sync' wakeups which
overrides the normal buddy wakeup logic - even if waker and
wakee are loosely coupled.
Fix an inversion of logic in the buddy wakeup code.
Reported-by: Pawel Dziekonski <dzieko@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Mike's change: 0a582440f "sched: fix sched_slice())" broke group
scheduling by forgetting to reload cfs_rq on each loop.
This patch fixes aim7 regression and specjbb2005 regression becomes
less than 1.5% on 8-core stokley.
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Jayson King <dev@jaysonking.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: fix SCHED_IDLE latency problems
OK, so we have 1 running task A (which is obviously curr and the tree is
equally obviously empty).
'A' nicely chugs along, doing its thing, carrying min_vruntime along as it
goes.
Then some whacko speed freak SCHED_IDLE task gets inserted due to SMP
balancing, which is very likely far right, in that case
update_curr
update_min_vruntime
cfs_rq->rb_leftmost := true (the crazy task sitting in a tree)
vruntime = se->vruntime
and voila, min_vruntime is waaay right of where it ought to be.
OK, so why did I write it like that to begin with...
Aah, yes.
Say we've just dequeued current
schedule
deactivate_task(prev)
dequeue_entity
update_min_vruntime
Then we'll set
vruntime = cfs_rq->min_vruntime;
we find !cfs_rq->curr, but do find someone in the tree. Then we _must_
do vruntime = se->vruntime, because
vruntime = min_vruntime(vruntime := cfs_rq->min_vruntime, se->vruntime)
will not advance vruntime, and cause lags the other way around (which we
fixed with that initial patch: 1af5f730fc
(sched: more accurate min_vruntime accounting).
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Mike Galbraith <efault@gmx.de>
Acked-by: Mike Galbraith <efault@gmx.de>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Stronger SCHED_IDLE isolation:
- no SCHED_IDLE buddies
- never let SCHED_IDLE preempt on wakeup
- always preempt SCHED_IDLE on wakeup
- limit SLEEPER fairness for SCHED_IDLE.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
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>
* 'cpus4096-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (66 commits)
x86: export vector_used_by_percpu_irq
x86: use logical apicid in x2apic_cluster's x2apic_cpu_mask_to_apicid_and()
sched: nominate preferred wakeup cpu, fix
x86: fix lguest used_vectors breakage, -v2
x86: fix warning in arch/x86/kernel/io_apic.c
sched: fix warning in kernel/sched.c
sched: move test_sd_parent() to an SMP section of sched.h
sched: add SD_BALANCE_NEWIDLE at MC and CPU level for sched_mc>0
sched: activate active load balancing in new idle cpus
sched: bias task wakeups to preferred semi-idle packages
sched: nominate preferred wakeup cpu
sched: favour lower logical cpu number for sched_mc balance
sched: framework for sched_mc/smt_power_savings=N
sched: convert BALANCE_FOR_xx_POWER to inline functions
x86: use possible_cpus=NUM to extend the possible cpus allowed
x86: fix cpu_mask_to_apicid_and to include cpu_online_mask
x86: update io_apic.c to the new cpumask code
x86: Introduce topology_core_cpumask()/topology_thread_cpumask()
x86: xen: use smp_call_function_many()
x86: use work_on_cpu in x86/kernel/cpu/mcheck/mce_amd_64.c
...
Fixed up trivial conflict in kernel/time/tick-sched.c manually
Impact: fix bad-interactivity buglet
Fix sched_slice() to emit a sane result whether a task is currently
enqueued or not.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Tested-by: Jayson King <dev@jaysonking.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched_fair.c | 30 ++++++++++++------------------
1 file changed, 12 insertions(+), 18 deletions(-)
Impact: tweak task wakeup to save power more agressively
Preferred wakeup cpu (from a semi idle package) has been
nominated in find_busiest_group() in the previous patch. Use
this information in sched_mc_preferred_wakeup_cpu in function
wake_idle() to bias task wakeups if the following conditions
are satisfied:
- The present cpu that is trying to wakeup the process is
idle and waking the target process on this cpu will
potentially wakeup a completely idle package
- The previous cpu on which the target process ran is
also idle and hence selecting the previous cpu may
wakeup a semi idle cpu package
- The task being woken up is allowed to run in the
nominated cpu (cpu affinity and restrictions)
Basically if both the current cpu and the previous cpu on
which the task ran is idle, select the nominated cpu from semi
idle cpu package for running the new task that is waking up.
Cache hotness is considered since the actual biasing happens
in wake_idle() only if the application is cache cold.
This technique will effectively move short running bursty jobs in
a mostly idle system.
Wakeup biasing for power savings gets automatically disabled if
system utilisation increases due to the fact that the probability
of finding both this_cpu and prev_cpu idle decreases.
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: micro-optimization
Skip the hard work when there is none.
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>
Impact: sharpen the wakeup-granularity to always be against current scheduler time
It was possible to do the preemption check against an old time stamp.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: Trivial API conversion
NR_CPUS -> nr_cpu_ids
cpumask_t -> struct cpumask
sizeof(cpumask_t) -> cpumask_size()
cpumask_a = cpumask_b -> cpumask_copy(&cpumask_a, &cpumask_b)
cpu_set() -> cpumask_set_cpu()
first_cpu() -> cpumask_first()
cpumask_of_cpu() -> cpumask_of()
cpus_* -> cpumask_*
There are some FIXMEs where we all archs to complete infrastructure
(patches have been sent):
cpu_coregroup_map -> cpu_coregroup_mask
node_to_cpumask* -> cpumask_of_node
There is also one FIXME where we pass an array of cpumasks to
partition_sched_domains(): this implies knowing the definition of
'struct cpumask' and the size of a cpumask. This will be fixed in a
future patch.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: trivial wrap of member accesses
This eases the transition in the next patch.
We also get rid of a temporary cpumask in find_idlest_cpu() thanks to
for_each_cpu_and, and sched_balance_self() due to getting weight before
setting sd to NULL.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Clear buddies on yield, so that the buddy rules don't schedule them
despite them being placed right-most.
This fixed a performance regression with yield-happy binary JVMs.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Impact: scheduling order fix for group scheduling
For each level in the hierarchy, set the buddy to point to the right entity.
Therefore, when we do the hierarchical schedule, we have a fair chance of
ending up where we meant to.
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>
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>
Impact: fix cross-class preemption
Inter-class wakeup preemptions should go on class order.
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>
Since we moved wakeup preemption back to virtual time, it makes sense to move
the buddy stuff back as well. The purpose of the buddy scheduling is to allow
a quickly scheduling pair of tasks to run away from the group as far as a
regular busy task would be allowed under wakeup preemption.
This has the advantage that the pair can ping-pong for a while, enjoying
cache-hotness. Without buddy scheduling other tasks would interleave destroying
the cache.
Also, it saves a word in cfs_rq.
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>
The advantage is that vruntime based wakeup preemption has a better
conceptual model. Here wakeup_gran = 0 means: preempt when 'fair'.
Therefore wakeup_gran is the granularity of unfairness we allow in order
to make progress.
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>
Mysql+oltp and pgsql+oltp peaks are still shifted right. The below puts
the peaks back to 1 client/server pair per core.
Use the avg_overlap information to weaken the sync hint.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Mike noticed the current min_vruntime tracking can go wrong and skip the
current task. If the only remaining task in the tree is a nice 19 task
with huge vruntime, new tasks will be inserted too far to the right too,
causing some interactibity issues.
min_vruntime can only change due to the leftmost entry disappearing
(dequeue_entity()), or by the leftmost entry being incremented past the
next entry, which elects a new leftmost (__update_curr())
Due to the current entry not being part of the actual tree, we have to
compare the leftmost tree entry with the current entry, and take the
leftmost of these two.
So create a update_min_vruntime() function that takes computes the
leftmost vruntime in the system (either tree of current) and increases
the cfs_rq->min_vruntime if the computed value is larger than the
previously found min_vruntime. And call this from the two sites we've
identified that can change min_vruntime.
Reported-by: Mike Galbraith <efault@gmx.de>
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>
* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: disable the hrtick for now
sched: revert back to per-rq vruntime
sched: fair scheduler should not resched rt tasks
sched: optimize group load balancer
sched: minor fast-path overhead reduction
sched: fix the wrong mask_len, cleanup
sched: kill unused scheduler decl.
sched: fix the wrong mask_len
sched: only update rq->clock while holding rq->lock
a patch from Henrik Austad did this:
>> Do not declare select_task_rq as part of sched_class when CONFIG_SMP is
>> not set.
Peter observed:
> While a proper cleanup, could you do it by re-arranging the methods so
> as to not create an additional ifdef?
Do not declare select_task_rq and some other methods as part of sched_class
when CONFIG_SMP is not set.
Also gather those methods to avoid CONFIG_SMP mess.
Idea-by: Henrik Austad <henrik.austad@gmail.com>
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Henrik Austad <henrik@austad.us>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Vatsa rightly points out that having the runqueue weight in the vruntime
calculations can cause unfairness in the face of task joins/leaves.
Suppose: dv = dt * rw / w
Then take 10 tasks t_n, each of similar weight. If the first will run 1
then its vruntime will increase by 10. Now, if the next 8 tasks leave after
having run their 1, then the last task will get a vruntime increase of 2
after having run 1.
Which will leave us with 2 tasks of equal weight and equal runtime, of which
one will not be scheduled for 8/2=4 units of time.
Ergo, we cannot do that and must use: dv = dt / w.
This means we cannot have a global vruntime based on effective priority, but
must instead go back to the vruntime per rq model we started out with.
This patch was lightly tested by doing starting while loops on each nice level
and observing their execution time, and a simple group scenario of 1:2:3 pinned
to a single cpu.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
With use of ftrace Steven noticed that some RT tasks got rescheduled due
to sched_fair interaction.
What happens is that we reprogram the hrtick from enqueue/dequeue_fair_task()
because that can change nr_running, and thus a current tasks ideal runtime.
However, its possible the current task isn't a fair_sched_class task, and thus
doesn't have a hrtick set to change.
Fix this by wrapping those hrtick_start_fair() calls in a hrtick_update()
function, which will check for the right conditions.
Reported-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Greetings,
103638d added a bit of avoidable overhead to the fast-path.
Use sysctl_sched_min_granularity instead of sched_slice() to restrict buddy wakeups.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
While looking at the code I wondered why we always do:
sync && avg_overlap < migration_cost
Which is a bit odd, since the overlap test was meant to detect sync wakeups
so using it to specialize sync wakeups doesn't make much sense.
Hence change the code to do:
sync || avg_overlap < migration_cost
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch does following:
o Removes unused variable and argument "rq".
o Optimizes one of the "if" conditions in wake_affine() - i.e. if
"balanced" is true, we need not do rest of the calculations in the
condition.
o If this cpu is same as the previous cpu (on which woken up task
was running when it went to sleep), no need to call wake_affine at all.
Signed-off-by: Amit K Arora <aarora@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
cfs_rq->tasks list is used by the load balancer to iterate
over all the tasks. Currently it holds all the entities
(both task and group entities) because of which there is
a need to check for group entities explicitly during load
balancing. This patch changes the cfs_rq->tasks list to
hold only task entities.
Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We should set the buddy even though we might already have the
TIF_RESCHED flag set.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
We should not only correct the increment for the initial group, but should
be consistent and do so for all the groups we encounter.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Rework the wakeup preemption to work on real runtime instead of
the virtual runtime. This greatly simplifies the code.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
load_balance_fair() calls rcu_read_lock() but then traverses the list
using the regular list traversal routine. This patch converts the
list traversal to use the _rcu version.
Signed-off-by: Chris Friesen <cfriesen@nortel.com>
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>
Overview
This patch reworks the handling of POSIX CPU timers, including the
ITIMER_PROF, ITIMER_VIRT timers and rlimit handling. It was put together
with the help of Roland McGrath, the owner and original writer of this code.
The problem we ran into, and the reason for this rework, has to do with using
a profiling timer in a process with a large number of threads. It appears
that the performance of the old implementation of run_posix_cpu_timers() was
at least O(n*3) (where "n" is the number of threads in a process) or worse.
Everything is fine with an increasing number of threads until the time taken
for that routine to run becomes the same as or greater than the tick time, at
which point things degrade rather quickly.
This patch fixes bug 9906, "Weird hang with NPTL and SIGPROF."
Code Changes
This rework corrects the implementation of run_posix_cpu_timers() to make it
run in constant time for a particular machine. (Performance may vary between
one machine and another depending upon whether the kernel is built as single-
or multiprocessor and, in the latter case, depending upon the number of
running processors.) To do this, at each tick we now update fields in
signal_struct as well as task_struct. The run_posix_cpu_timers() function
uses those fields to make its decisions.
We define a new structure, "task_cputime," to contain user, system and
scheduler times and use these in appropriate places:
struct task_cputime {
cputime_t utime;
cputime_t stime;
unsigned long long sum_exec_runtime;
};
This is included in the structure "thread_group_cputime," which is a new
substructure of signal_struct and which varies for uniprocessor versus
multiprocessor kernels. For uniprocessor kernels, it uses "task_cputime" as
a simple substructure, while for multiprocessor kernels it is a pointer:
struct thread_group_cputime {
struct task_cputime totals;
};
struct thread_group_cputime {
struct task_cputime *totals;
};
We also add a new task_cputime substructure directly to signal_struct, to
cache the earliest expiration of process-wide timers, and task_cputime also
replaces the it_*_expires fields of task_struct (used for earliest expiration
of thread timers). The "thread_group_cputime" structure contains process-wide
timers that are updated via account_user_time() and friends. In the non-SMP
case the structure is a simple aggregator; unfortunately in the SMP case that
simplicity was not achievable due to cache-line contention between CPUs (in
one measured case performance was actually _worse_ on a 16-cpu system than
the same test on a 4-cpu system, due to this contention). For SMP, the
thread_group_cputime counters are maintained as a per-cpu structure allocated
using alloc_percpu(). The timer functions update only the timer field in
the structure corresponding to the running CPU, obtained using per_cpu_ptr().
We define a set of inline functions in sched.h that we use to maintain the
thread_group_cputime structure and hide the differences between UP and SMP
implementations from the rest of the kernel. The thread_group_cputime_init()
function initializes the thread_group_cputime structure for the given task.
The thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the
out-of-line function thread_group_cputime_alloc_smp() to allocate and fill
in the per-cpu structures and fields. The thread_group_cputime_free()
function, also a no-op for UP, in SMP frees the per-cpu structures. The
thread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls
thread_group_cputime_alloc() if the per-cpu structures haven't yet been
allocated. The thread_group_cputime() function fills the task_cputime
structure it is passed with the contents of the thread_group_cputime fields;
in UP it's that simple but in SMP it must also safely check that tsk->signal
is non-NULL (if it is it just uses the appropriate fields of task_struct) and,
if so, sums the per-cpu values for each online CPU. Finally, the three
functions account_group_user_time(), account_group_system_time() and
account_group_exec_runtime() are used by timer functions to update the
respective fields of the thread_group_cputime structure.
Non-SMP operation is trivial and will not be mentioned further.
The per-cpu structure is always allocated when a task creates its first new
thread, via a call to thread_group_cputime_clone_thread() from copy_signal().
It is freed at process exit via a call to thread_group_cputime_free() from
cleanup_signal().
All functions that formerly summed utime/stime/sum_sched_runtime values from
from all threads in the thread group now use thread_group_cputime() to
snapshot the values in the thread_group_cputime structure or the values in
the task structure itself if the per-cpu structure hasn't been allocated.
Finally, the code in kernel/posix-cpu-timers.c has changed quite a bit.
The run_posix_cpu_timers() function has been split into a fast path and a
slow path; the former safely checks whether there are any expired thread
timers and, if not, just returns, while the slow path does the heavy lifting.
With the dedicated thread group fields, timers are no longer "rebalanced" and
the process_timer_rebalance() function and related code has gone away. All
summing loops are gone and all code that used them now uses the
thread_group_cputime() inline. When process-wide timers are set, the new
task_cputime structure in signal_struct is used to cache the earliest
expiration; this is checked in the fast path.
Performance
The fix appears not to add significant overhead to existing operations. It
generally performs the same as the current code except in two cases, one in
which it performs slightly worse (Case 5 below) and one in which it performs
very significantly better (Case 2 below). Overall it's a wash except in those
two cases.
I've since done somewhat more involved testing on a dual-core Opteron system.
Case 1: With no itimer running, for a test with 100,000 threads, the fixed
kernel took 1428.5 seconds, 513 seconds more than the unfixed system,
all of which was spent in the system. There were twice as many
voluntary context switches with the fix as without it.
Case 2: With an itimer running at .01 second ticks and 4000 threads (the most
an unmodified kernel can handle), the fixed kernel ran the test in
eight percent of the time (5.8 seconds as opposed to 70 seconds) and
had better tick accuracy (.012 seconds per tick as opposed to .023
seconds per tick).
Case 3: A 4000-thread test with an initial timer tick of .01 second and an
interval of 10,000 seconds (i.e. a timer that ticks only once) had
very nearly the same performance in both cases: 6.3 seconds elapsed
for the fixed kernel versus 5.5 seconds for the unfixed kernel.
With fewer threads (eight in these tests), the Case 1 test ran in essentially
the same time on both the modified and unmodified kernels (5.2 seconds versus
5.8 seconds). The Case 2 test ran in about the same time as well, 5.9 seconds
versus 5.4 seconds but again with much better tick accuracy, .013 seconds per
tick versus .025 seconds per tick for the unmodified kernel.
Since the fix affected the rlimit code, I also tested soft and hard CPU limits.
Case 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer
running), the modified kernel was very slightly favored in that while
it killed the process in 19.997 seconds of CPU time (5.002 seconds of
wall time), only .003 seconds of that was system time, the rest was
user time. The unmodified kernel killed the process in 20.001 seconds
of CPU (5.014 seconds of wall time) of which .016 seconds was system
time. Really, though, the results were too close to call. The results
were essentially the same with no itimer running.
Case 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds
(where the hard limit would never be reached) and an itimer running,
the modified kernel exhibited worse tick accuracy than the unmodified
kernel: .050 seconds/tick versus .028 seconds/tick. Otherwise,
performance was almost indistinguishable. With no itimer running this
test exhibited virtually identical behavior and times in both cases.
In times past I did some limited performance testing. those results are below.
On a four-cpu Opteron system without this fix, a sixteen-thread test executed
in 3569.991 seconds, of which user was 3568.435s and system was 1.556s. On
the same system with the fix, user and elapsed time were about the same, but
system time dropped to 0.007 seconds. Performance with eight, four and one
thread were comparable. Interestingly, the timer ticks with the fix seemed
more accurate: The sixteen-thread test with the fix received 149543 ticks
for 0.024 seconds per tick, while the same test without the fix received 58720
for 0.061 seconds per tick. Both cases were configured for an interval of
0.01 seconds. Again, the other tests were comparable. Each thread in this
test computed the primes up to 25,000,000.
I also did a test with a large number of threads, 100,000 threads, which is
impossible without the fix. In this case each thread computed the primes only
up to 10,000 (to make the runtime manageable). System time dominated, at
1546.968 seconds out of a total 2176.906 seconds (giving a user time of
629.938s). It received 147651 ticks for 0.015 seconds per tick, still quite
accurate. There is obviously no comparable test without the fix.
Signed-off-by: Frank Mayhar <fmayhar@google.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The __load_balance_iterator() returns a NULL when there's only one
sched_entity which is a task. It is caused by the following code-path.
/* Skip over entities that are not tasks */
do {
se = list_entry(next, struct sched_entity, group_node);
next = next->next;
} while (next != &cfs_rq->tasks && !entity_is_task(se));
if (next == &cfs_rq->tasks)
return NULL;
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This will return NULL even when se is a task.
As a side-effect, there was a regression in sched_mc behavior since 2.6.25,
since iter_move_one_task() when it calls load_balance_start_fair(),
would not get any tasks to move!
Fix this by checking if the last entity was a task or not.
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
- During wake up of a new task, task_new_fair() can do a resched_task()
on the current task. Later in the code path, check_preempt_curr() also ends
up doing the same, which can be avoided. Check if TIF_NEED_RESCHED is
already set for the current task.
- task_new_fair() does a resched_task() on the current task unconditionally.
This can be done only in case when child runs before the parent.
So this is a small speedup.
Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Defer commit 6d299f1b53 to the next release.
Testing of the tip/sched/clock tree revealed a mysql+oltp regression
which bisection eventually traced back to this commit in mainline.
Pertinent test results: Three run sysbench averages, throughput units
in read/write requests/sec.
clients 1 2 4 8 16 32 64
6e0534f 9646 17876 34774 33868 32230 30767 29441
2.6.26.1 9112 17936 34652 33383 31929 30665 29232
6d299f1 9112 14637 28370 33339 32038 30762 29204
Note: subsequent commits hide the majority of this regression until you
apply the clock fixes, at which time it reemerges at full magnitude.
We cannot see anything bad about the change itself so we defer it to the
next release until this problem is fully analysed.
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Gregory Haskins <ghaskins@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Benjamin Herrenschmidt reported:
> I get that on ppc64 ...
>
> In file included from kernel/sched.c:1595:
> kernel/sched_fair.c: In function ‘hrtick_start_fair’:
> kernel/sched_fair.c:902: warning: comparison of distinct pointer types lacks a cast
>
> Probably harmless but annoying.
s64 delta = slice - ran;
--> delta = max(10000LL, delta);
Probably ppc64's s64 is long vs long long..
I think hpa was looking at sanitizing all these 64bit types across the
architectures.
Use max_t with an explicit type meanwhile.
Reported-by: Benjamin Herrenschmid <benh@kernel.crashing.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'sched/for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
sched: hrtick_enabled() should use cpu_active()
sched, x86: clean up hrtick implementation
sched: fix build error, provide partition_sched_domains() unconditionally
sched: fix warning in inc_rt_tasks() to not declare variable 'rq' if it's not needed
cpu hotplug: Make cpu_active_map synchronization dependency clear
cpu hotplug, sched: Introduce cpu_active_map and redo sched domain managment (take 2)
sched: rework of "prioritize non-migratable tasks over migratable ones"
sched: reduce stack size in isolated_cpu_setup()
Revert parts of "ftrace: do not trace scheduler functions"
Fixed up conflicts in include/asm-x86/thread_info.h (due to the
TIF_SINGLESTEP unification vs TIF_HRTICK_RESCHED removal) and
kernel/sched_fair.c (due to cpu_active_map vs for_each_cpu_mask_nr()
introduction).
random uvesafb failures were reported against Gentoo:
http://bugs.gentoo.org/show_bug.cgi?id=222799
and Mihai Moldovan bisected it back to:
> 8f4d37ec07 is first bad commit
> commit 8f4d37ec07
> Author: Peter Zijlstra <a.p.zijlstra@chello.nl>
> Date: Fri Jan 25 21:08:29 2008 +0100
>
> sched: high-res preemption tick
Linus suspected it to be hrtick + vm86 interaction and observed:
> Btw, Peter, Ingo: I think that commit is doing bad things. They aren't
> _incorrect_ per se, but they are definitely bad.
>
> Why?
>
> Using random _TIF_WORK_MASK flags is really impolite for doing
> "scheduling" work. There's a reason that arch/x86/kernel/entry_32.S
> special-cases the _TIF_NEED_RESCHED flag: we don't want to exit out of
> vm86 mode unnecessarily.
>
> See the "work_notifysig_v86" label, and how it does that
> "save_v86_state()" thing etc etc.
Right, I never liked having to fiddle with those TIF flags. Initially I
needed it because the hrtimer base lock could not nest in the rq lock.
That however is fixed these days.
Currently the only reason left to fiddle with the TIF flags is remote
wakeups. We cannot program a remote cpu's hrtimer. I've been thinking
about using the new and improved IPI function call stuff to implement
hrtimer_start_on().
However that does require that smp_call_function_single(.wait=0) works
from interrupt context - /me looks at the latest series from Jens - Yes
that does seem to be supported, good.
Here's a stab at cleaning this stuff up ...
Mihai reported test success as well.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Mihai Moldovan <ionic@ionic.de>
Cc: Michal Januszewski <spock@gentoo.org>
Cc: Antonino Daplas <adaplas@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This is based on Linus' idea of creating cpu_active_map that prevents
scheduler load balancer from migrating tasks to the cpu that is going
down.
It allows us to simplify domain management code and avoid unecessary
domain rebuilds during cpu hotplug event handling.
Please ignore the cpusets part for now. It needs some more work in order
to avoid crazy lock nesting. Although I did simplfy and unify domain
reinitialization logic. We now simply call partition_sched_domains() in
all the cases. This means that we're using exact same code paths as in
cpusets case and hence the test below cover cpusets too.
Cpuset changes to make rebuild_sched_domains() callable from various
contexts are in the separate patch (right next after this one).
This not only boots but also easily handles
while true; do make clean; make -j 8; done
and
while true; do on-off-cpu 1; done
at the same time.
(on-off-cpu 1 simple does echo 0/1 > /sys/.../cpu1/online thing).
Suprisingly the box (dual-core Core2) is quite usable. In fact I'm typing
this on right now in gnome-terminal and things are moving just fine.
Also this is running with most of the debug features enabled (lockdep,
mutex, etc) no BUG_ONs or lockdep complaints so far.
I believe I addressed all of the Dmitry's comments for original Linus'
version. I changed both fair and rt balancer to mask out non-active cpus.
And replaced cpu_is_offline() with !cpu_active() in the main scheduler
code where it made sense (to me).
Signed-off-by: Max Krasnyanskiy <maxk@qualcomm.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Gregory Haskins <ghaskins@novell.com>
Cc: dmitry.adamushko@gmail.com
Cc: pj@sgi.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
