* [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running
@ 2024-09-25 8:54 Chen Yu
2024-09-27 1:54 ` Honglei Wang
` (2 more replies)
0 siblings, 3 replies; 12+ messages in thread
From: Chen Yu @ 2024-09-25 8:54 UTC (permalink / raw)
To: Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot
Cc: Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel,
Chen Yu, Chen Yu, kernel test robot, K Prateek Nayak
Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt")
introduced a mechanism that a wakee with shorter slice could preempt
the current running task. It also lower the bar for the current task
to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running
when the current task has ran out of time slice. But there is a scenario
that is problematic. Say, if there is 1 cfs task and 1 rt task, before
85e511df3cec, update_deadline() will not trigger a reschedule, and after
85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr()
would happen.
Some workloads (like the hackbench reported by lkp) do not like
over-scheduling. We can see that the preemption rate has been
increased by 2.2%:
1.654e+08 +2.2% 1.69e+08 hackbench.time.involuntary_context_switches
Restore its previous check criterion.
Fixes: 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt")
Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202409231416.9403c2e9-oliver.sang@intel.com
Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
---
v1->v2:
Check cfs_rq->nr_running instead of rq->nr_running(K Prateek Nayak)
---
kernel/sched/fair.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 225b31aaee55..53a351b18740 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq)
account_cfs_rq_runtime(cfs_rq, delta_exec);
- if (rq->nr_running == 1)
+ if (cfs_rq->nr_running == 1)
return;
if (resched || did_preempt_short(cfs_rq, curr)) {
--
2.25.1
^ permalink raw reply related [flat|nested] 12+ messages in thread* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-25 8:54 [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running Chen Yu @ 2024-09-27 1:54 ` Honglei Wang 2024-09-27 6:56 ` Chen Yu 2024-09-30 7:50 ` Peter Zijlstra 2024-10-04 10:16 ` [tip: sched/urgent] " tip-bot2 for Chen Yu 2 siblings, 1 reply; 12+ messages in thread From: Honglei Wang @ 2024-09-27 1:54 UTC (permalink / raw) To: Chen Yu, Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot Cc: Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, kernel test robot, K Prateek Nayak On 2024/9/25 16:54, Chen Yu wrote: > Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > introduced a mechanism that a wakee with shorter slice could preempt > the current running task. It also lower the bar for the current task > to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running > when the current task has ran out of time slice. But there is a scenario > that is problematic. Say, if there is 1 cfs task and 1 rt task, before > 85e511df3cec, update_deadline() will not trigger a reschedule, and after > 85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr() > would happen. > > Some workloads (like the hackbench reported by lkp) do not like > over-scheduling. We can see that the preemption rate has been > increased by 2.2%: > > 1.654e+08 +2.2% 1.69e+08 hackbench.time.involuntary_context_switches > > Restore its previous check criterion. > > Fixes: 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > Reported-by: kernel test robot <oliver.sang@intel.com> > Closes: https://lore.kernel.org/oe-lkp/202409231416.9403c2e9-oliver.sang@intel.com > Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com> > Tested-by: K Prateek Nayak <kprateek.nayak@amd.com> > Signed-off-by: Chen Yu <yu.c.chen@intel.com> > --- > v1->v2: > Check cfs_rq->nr_running instead of rq->nr_running(K Prateek Nayak) > --- > kernel/sched/fair.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c > index 225b31aaee55..53a351b18740 100644 > --- a/kernel/sched/fair.c > +++ b/kernel/sched/fair.c > @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) > > account_cfs_rq_runtime(cfs_rq, delta_exec); > > - if (rq->nr_running == 1) > + if (cfs_rq->nr_running == 1) > return; > Hi Yu, I'm wondering if commit 85e511df3cec wants to give more chances to do resched just in case there are 'short slices' tasks ready in the other cfs hierarchy. Does something like rq->cfs->nr_running == 1 make more sense? But maybe it helps less than 'cfs_rq->nr_running == 1' in this hackbench case. Thanks, Honglei > if (resched || did_preempt_short(cfs_rq, curr)) { ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-27 1:54 ` Honglei Wang @ 2024-09-27 6:56 ` Chen Yu 2024-09-27 13:38 ` Honglei Wang 0 siblings, 1 reply; 12+ messages in thread From: Chen Yu @ 2024-09-27 6:56 UTC (permalink / raw) To: Honglei Wang Cc: Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, kernel test robot, K Prateek Nayak Hello Honglei, On 2024-09-27 at 09:54:28 +0800, Honglei Wang wrote: > > > On 2024/9/25 16:54, Chen Yu wrote: > > Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > > introduced a mechanism that a wakee with shorter slice could preempt > > the current running task. It also lower the bar for the current task > > to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running > > when the current task has ran out of time slice. But there is a scenario > > that is problematic. Say, if there is 1 cfs task and 1 rt task, before > > 85e511df3cec, update_deadline() will not trigger a reschedule, and after > > 85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr() > > would happen. > > > > Some workloads (like the hackbench reported by lkp) do not like > > over-scheduling. We can see that the preemption rate has been > > increased by 2.2%: > > > > 1.654e+08 +2.2% 1.69e+08 hackbench.time.involuntary_context_switches > > > > Restore its previous check criterion. > > > > Fixes: 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > > Reported-by: kernel test robot <oliver.sang@intel.com> > > Closes: https://lore.kernel.org/oe-lkp/202409231416.9403c2e9-oliver.sang@intel.com > > Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com> > > Tested-by: K Prateek Nayak <kprateek.nayak@amd.com> > > Signed-off-by: Chen Yu <yu.c.chen@intel.com> > > --- > > v1->v2: > > Check cfs_rq->nr_running instead of rq->nr_running(K Prateek Nayak) > > --- > > kernel/sched/fair.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c > > index 225b31aaee55..53a351b18740 100644 > > --- a/kernel/sched/fair.c > > +++ b/kernel/sched/fair.c > > @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) > > account_cfs_rq_runtime(cfs_rq, delta_exec); > > - if (rq->nr_running == 1) > > + if (cfs_rq->nr_running == 1) > > return; > Hi Yu, > > I'm wondering if commit 85e511df3cec wants to give more chances to do > resched just in case there are 'short slices' tasks ready in the other cfs > hierarchy. > Does something like rq->cfs->nr_running == 1 make more sense? But > maybe it helps less than 'cfs_rq->nr_running == 1' in this hackbench case. > Thanks for taking a look. It could be possible that Peter wanted the short tasks to preempt other quickly. If I understand correctly, when we say preemption, we usually consider two tasks which are in the same cfs_rq(level). For example, check_preempt_wakeup_fair() iterates the hierarchy from down-up until the current task and the wakee are in the same level via find_matching_se(&se, &pse), then check if the wakee can preempt the current. This should be consistent with the tick preemption in update_curr(). And whether the short task should preempt the current is checked by update_curr() -> did_preempt_short(), rather than checking the cfs_rq->nr_running/nr_h_running I suppose. Thanks, Chenyu ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-27 6:56 ` Chen Yu @ 2024-09-27 13:38 ` Honglei Wang 2024-09-28 6:28 ` Chen Yu 0 siblings, 1 reply; 12+ messages in thread From: Honglei Wang @ 2024-09-27 13:38 UTC (permalink / raw) To: Chen Yu Cc: Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, kernel test robot, K Prateek Nayak On 2024/9/27 14:56, Chen Yu wrote: > Hello Honglei, > > On 2024-09-27 at 09:54:28 +0800, Honglei Wang wrote: >> >> >> On 2024/9/25 16:54, Chen Yu wrote: >>> Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") >>> introduced a mechanism that a wakee with shorter slice could preempt >>> the current running task. It also lower the bar for the current task >>> to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running >>> when the current task has ran out of time slice. But there is a scenario >>> that is problematic. Say, if there is 1 cfs task and 1 rt task, before >>> 85e511df3cec, update_deadline() will not trigger a reschedule, and after >>> 85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr() >>> would happen. >>> >>> Some workloads (like the hackbench reported by lkp) do not like >>> over-scheduling. We can see that the preemption rate has been >>> increased by 2.2%: >>> >>> 1.654e+08 +2.2% 1.69e+08 hackbench.time.involuntary_context_switches >>> >>> Restore its previous check criterion. >>> >>> Fixes: 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") >>> Reported-by: kernel test robot <oliver.sang@intel.com> >>> Closes: https://lore.kernel.org/oe-lkp/202409231416.9403c2e9-oliver.sang@intel.com >>> Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com> >>> Tested-by: K Prateek Nayak <kprateek.nayak@amd.com> >>> Signed-off-by: Chen Yu <yu.c.chen@intel.com> >>> --- >>> v1->v2: >>> Check cfs_rq->nr_running instead of rq->nr_running(K Prateek Nayak) >>> --- >>> kernel/sched/fair.c | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c >>> index 225b31aaee55..53a351b18740 100644 >>> --- a/kernel/sched/fair.c >>> +++ b/kernel/sched/fair.c >>> @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) >>> account_cfs_rq_runtime(cfs_rq, delta_exec); >>> - if (rq->nr_running == 1) >>> + if (cfs_rq->nr_running == 1) >>> return; >> Hi Yu, >> >> I'm wondering if commit 85e511df3cec wants to give more chances to do >> resched just in case there are 'short slices' tasks ready in the other cfs >> hierarchy. >> Does something like rq->cfs->nr_running == 1 make more sense? But >> maybe it helps less than 'cfs_rq->nr_running == 1' in this hackbench case. >> > > Thanks for taking a look. > > It could be possible that Peter wanted the short tasks to preempt other quickly. > If I understand correctly, when we say preemption, we usually consider two > tasks which are in the same cfs_rq(level). For example, check_preempt_wakeup_fair() > iterates the hierarchy from down-up until the current task and the wakee are in the > same level via find_matching_se(&se, &pse), then check if the wakee can preempt the > current. This should be consistent with the tick preemption in update_curr(). And > whether the short task should preempt the current is checked by > update_curr() -> did_preempt_short(), rather than checking the cfs_rq->nr_running/nr_h_running > I suppose. > Hi Yu, Yep, I understand the preemption should happen in the same cfs level. Just not sure the purpose of the 'nr_running check' stuff. Perhaps its role is just to judge whether it’s necessary to do the preemption check. If there is at least one more ready (cfs) task in the rq and current is not eligible, we take care of the waiting tasks. Thoughts? Thanks, Honglei > Thanks, > Chenyu ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-27 13:38 ` Honglei Wang @ 2024-09-28 6:28 ` Chen Yu 2024-09-29 13:51 ` Oliver Sang 0 siblings, 1 reply; 12+ messages in thread From: Chen Yu @ 2024-09-28 6:28 UTC (permalink / raw) To: Honglei Wang, Oliver Sang Cc: Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, kernel test robot, K Prateek Nayak Hi Honglei, On 2024-09-27 at 21:38:53 +0800, Honglei Wang wrote: > > > On 2024/9/27 14:56, Chen Yu wrote: > > Hello Honglei, > > > > On 2024-09-27 at 09:54:28 +0800, Honglei Wang wrote: > > > > > > > > > On 2024/9/25 16:54, Chen Yu wrote: > > > > Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > > > > introduced a mechanism that a wakee with shorter slice could preempt > > > > the current running task. It also lower the bar for the current task > > > > to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running > > > > when the current task has ran out of time slice. But there is a scenario > > > > that is problematic. Say, if there is 1 cfs task and 1 rt task, before > > > > 85e511df3cec, update_deadline() will not trigger a reschedule, and after > > > > 85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr() > > > > would happen. > > > > > > > > Some workloads (like the hackbench reported by lkp) do not like > > > > over-scheduling. We can see that the preemption rate has been > > > > increased by 2.2%: > > > > > > > > 1.654e+08 +2.2% 1.69e+08 hackbench.time.involuntary_context_switches > > > > > > > > Restore its previous check criterion. > > > > > > > > Fixes: 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > > > > Reported-by: kernel test robot <oliver.sang@intel.com> > > > > Closes: https://lore.kernel.org/oe-lkp/202409231416.9403c2e9-oliver.sang@intel.com > > > > Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com> > > > > Tested-by: K Prateek Nayak <kprateek.nayak@amd.com> > > > > Signed-off-by: Chen Yu <yu.c.chen@intel.com> > > > > --- > > > > v1->v2: > > > > Check cfs_rq->nr_running instead of rq->nr_running(K Prateek Nayak) > > > > --- > > > > kernel/sched/fair.c | 2 +- > > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c > > > > index 225b31aaee55..53a351b18740 100644 > > > > --- a/kernel/sched/fair.c > > > > +++ b/kernel/sched/fair.c > > > > @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) > > > > account_cfs_rq_runtime(cfs_rq, delta_exec); > > > > - if (rq->nr_running == 1) > > > > + if (cfs_rq->nr_running == 1) > > > > return; > > > Hi Yu, > > > > > > I'm wondering if commit 85e511df3cec wants to give more chances to do > > > resched just in case there are 'short slices' tasks ready in the other cfs > > > hierarchy. > > > Does something like rq->cfs->nr_running == 1 make more sense? But > > > maybe it helps less than 'cfs_rq->nr_running == 1' in this hackbench case. > > > > > > > Thanks for taking a look. > > > > It could be possible that Peter wanted the short tasks to preempt other quickly. > > If I understand correctly, when we say preemption, we usually consider two > > tasks which are in the same cfs_rq(level). For example, check_preempt_wakeup_fair() > > iterates the hierarchy from down-up until the current task and the wakee are in the > > same level via find_matching_se(&se, &pse), then check if the wakee can preempt the > > current. This should be consistent with the tick preemption in update_curr(). And > > whether the short task should preempt the current is checked by > > update_curr() -> did_preempt_short(), rather than checking the cfs_rq->nr_running/nr_h_running > > I suppose. > > > Hi Yu, > > Yep, I understand the preemption should happen in the same cfs level. Just > not sure the purpose of the 'nr_running check' stuff. Perhaps its role is > just to judge whether it’s necessary to do the preemption check. If there is > at least one more ready (cfs) task in the rq and current is not eligible, we > take care of the waiting tasks. Thoughts? I got your point and it makes sense. Whether the preemption check should be triggered seems to be a heuristic trade-off to me. I'm ok with using more aggressive preemption strategy as it depends on whether that workload prefers latency or throughput, and as long as it did not introduce regression :-) Oliver, may I know if you happen to have time for a test if the following change suggested by Honglei would make the regression go away? Thanks. thanks, Chenyu diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index fd2f3831c74e..290e5fdfc267 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) account_cfs_rq_runtime(cfs_rq, delta_exec); - if (rq->nr_running == 1) + if (rq->cfs.nr_running == 1) return; if (resched || did_preempt_short(cfs_rq, curr)) { -- 2.25.1 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-28 6:28 ` Chen Yu @ 2024-09-29 13:51 ` Oliver Sang 2024-09-30 2:04 ` Honglei Wang 0 siblings, 1 reply; 12+ messages in thread From: Oliver Sang @ 2024-09-29 13:51 UTC (permalink / raw) To: Chen Yu Cc: Honglei Wang, Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, K Prateek Nayak, oliver.sang hi, Chenyu and Honglei, On Sat, Sep 28, 2024 at 02:28:30PM +0800, Chen Yu wrote: > Hi Honglei, > > On 2024-09-27 at 21:38:53 +0800, Honglei Wang wrote: > > Hi Yu, > > > > Yep, I understand the preemption should happen in the same cfs level. Just > > not sure the purpose of the 'nr_running check' stuff. Perhaps its role is > > just to judge whether it’s necessary to do the preemption check. If there is > > at least one more ready (cfs) task in the rq and current is not eligible, we > > take care of the waiting tasks. Thoughts? > > I got your point and it makes sense. Whether the preemption check should be triggered > seems to be a heuristic trade-off to me. I'm ok with using more aggressive preemption > strategy as it depends on whether that workload prefers latency or throughput, and as > long as it did not introduce regression :-) > > Oliver, may I know if you happen to have time for a test if the following change > suggested by Honglei would make the regression go away? Thanks. I applied the patch directly upon 85e511df3cec, the test found it can reduce the regression but not totally reovered ========================================================================================= compiler/cpufreq_governor/ipc/iterations/kconfig/mode/nr_threads/rootfs/tbox_group/testcase: gcc-12/performance/socket/4/x86_64-rhel-8.3/process/50%/debian-12-x86_64-20240206.cgz/lkp-spr-r02/hackbench commit: 82e9d0456e06 ("sched/fair: Avoid re-setting virtual deadline on 'migrations'") 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") 8079496d311b <--- patch from Honglei 82e9d0456e06cebe 85e511df3cec46021024176672a 8079496d311b6b0d4dae973f4df ---------------- --------------------------- --------------------------- %stddev %change %stddev %change %stddev \ | \ | \ 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput > > thanks, > Chenyu > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c > index fd2f3831c74e..290e5fdfc267 100644 > --- a/kernel/sched/fair.c > +++ b/kernel/sched/fair.c > @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) > > account_cfs_rq_runtime(cfs_rq, delta_exec); > > - if (rq->nr_running == 1) > + if (rq->cfs.nr_running == 1) > return; > > if (resched || did_preempt_short(cfs_rq, curr)) { > -- > 2.25.1 below is full comparison FYI. ========================================================================================= compiler/cpufreq_governor/ipc/iterations/kconfig/mode/nr_threads/rootfs/tbox_group/testcase: gcc-12/performance/socket/4/x86_64-rhel-8.3/process/50%/debian-12-x86_64-20240206.cgz/lkp-spr-r02/hackbench commit: 82e9d0456e06 ("sched/fair: Avoid re-setting virtual deadline on 'migrations'") 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") 8079496d311b <--- patch from Honglei 82e9d0456e06cebe 85e511df3cec46021024176672a 8079496d311b6b0d4dae973f4df ---------------- --------------------------- --------------------------- %stddev %change %stddev %change %stddev \ | \ | \ 150791 ± 3% -3.9% 144868 ± 2% -8.4% 138189 ± 3% perf-c2c.HITM.local 217.40 +13.5% 246.74 +1.9% 221.55 uptime.boot 4679401 -15.8% 3938145 +35.5% 6342833 vmstat.system.cs 854648 -15.2% 724774 +35.0% 1153448 vmstat.system.in 0.46 ± 2% -0.1 0.40 +0.1 0.51 ± 2% mpstat.cpu.all.irq% 0.03 ± 3% -0.0 0.03 +0.0 0.04 mpstat.cpu.all.soft% 3.35 -0.6 2.75 +0.3 3.65 mpstat.cpu.all.usr% 922628 ± 16% +17.2% 1080933 ± 14% +87.9% 1733278 ± 14% numa-numastat.node0.local_node 1027104 ± 13% +17.6% 1207529 ± 15% +82.3% 1872582 ± 14% numa-numastat.node0.numa_hit 2553206 ± 16% +0.4% 2563309 ± 10% +46.3% 3735450 ± 4% numa-numastat.node1.local_node 2684151 ± 16% -0.4% 2672834 ± 8% +43.2% 3843864 ± 5% numa-numastat.node1.numa_hit 174.58 +16.3% 203.09 +2.1% 178.28 time.elapsed_time 174.58 +16.3% 203.09 +2.1% 178.28 time.elapsed_time.max 1.654e+08 +2.2% 1.69e+08 +50.7% 2.492e+08 time.involuntary_context_switches 36869 +17.6% 43340 +1.6% 37461 time.system_time 1172 -5.5% 1107 +8.9% 1276 time.user_time 6.478e+08 -3.5% 6.255e+08 +36.2% 8.824e+08 ± 2% time.voluntary_context_switches 8177075 ± 14% +8.6% 8880848 ± 13% +47.8% 12086160 ± 2% meminfo.Cached 7771720 ± 15% +9.5% 8506289 ± 13% +50.8% 11720513 ± 2% meminfo.Committed_AS 4992416 ± 31% +17.9% 5883734 ± 19% +83.6% 9165655 ± 2% meminfo.Inactive 4992203 ± 31% +17.9% 5883521 ± 19% +83.6% 9165427 ± 2% meminfo.Inactive(anon) 2281337 ± 19% +17.2% 2673418 ± 4% +38.0% 3147813 ± 3% meminfo.Mapped 12824178 ± 10% +5.9% 13577581 ± 8% +30.4% 16717964 meminfo.Memused 5084291 ± 22% +13.8% 5788029 ± 20% +76.9% 8993338 ± 2% meminfo.Shmem 13135346 ± 9% +5.2% 13824666 ± 8% +29.2% 16968485 meminfo.max_used_kB 3089827 ± 31% +15.4% 3565461 ± 7% +49.8% 4628448 ± 10% numa-meminfo.node0.FilePages 5391461 ± 19% +16.5% 6281524 ± 6% +31.1% 7067023 ± 10% numa-meminfo.node0.MemUsed 352581 ± 13% +24.6% 439472 ± 16% +11.3% 392488 ± 14% numa-meminfo.node0.SUnreclaim 5090008 ± 22% +4.5% 5317103 ± 18% +46.6% 7462041 ± 3% numa-meminfo.node1.FilePages 4237429 ± 32% +13.8% 4822257 ± 21% +66.3% 7045772 ± 7% numa-meminfo.node1.Inactive 4237311 ± 32% +13.8% 4822122 ± 21% +66.3% 7045629 ± 7% numa-meminfo.node1.Inactive(anon) 1875959 ± 20% +18.1% 2215441 ± 5% +44.6% 2711735 ± 3% numa-meminfo.node1.Mapped 7434746 ± 16% -1.8% 7298189 ± 12% +29.9% 9655790 ± 5% numa-meminfo.node1.MemUsed 4692017 ± 22% +10.5% 5186768 ± 18% +55.8% 7310040 ± 4% numa-meminfo.node1.Shmem 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput 606251 -14.1% 520789 -2.1% 593622 hackbench.throughput_avg 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput_best 580034 -14.8% 494354 +0.4% 582377 hackbench.throughput_worst 174.58 +16.3% 203.09 +2.1% 178.28 hackbench.time.elapsed_time 174.58 +16.3% 203.09 +2.1% 178.28 hackbench.time.elapsed_time.max 1.654e+08 +2.2% 1.69e+08 +50.7% 2.492e+08 hackbench.time.involuntary_context_switches 36869 +17.6% 43340 +1.6% 37461 hackbench.time.system_time 1172 -5.5% 1107 +8.9% 1276 hackbench.time.user_time 6.478e+08 -3.5% 6.255e+08 +36.2% 8.824e+08 ± 2% hackbench.time.voluntary_context_switches 772255 ± 31% +15.4% 891384 ± 7% +49.8% 1156899 ± 10% numa-vmstat.node0.nr_file_pages 1025412 ± 13% +17.7% 1206842 ± 15% +82.6% 1871921 ± 14% numa-vmstat.node0.numa_hit 920942 ± 16% +17.3% 1080247 ± 14% +88.1% 1732619 ± 14% numa-vmstat.node0.numa_local 1271682 ± 22% +4.5% 1328902 ± 18% +46.6% 1863679 ± 3% numa-vmstat.node1.nr_file_pages 1059760 ± 32% +13.7% 1205295 ± 21% +66.1% 1759754 ± 7% numa-vmstat.node1.nr_inactive_anon 469078 ± 20% +18.0% 553505 ± 5% +44.2% 676623 ± 3% numa-vmstat.node1.nr_mapped 1172184 ± 22% +10.6% 1296319 ± 18% +55.8% 1825678 ± 4% numa-vmstat.node1.nr_shmem 1059759 ± 32% +13.7% 1205294 ± 21% +66.1% 1759752 ± 7% numa-vmstat.node1.nr_zone_inactive_anon 2682254 ± 16% -0.4% 2671316 ± 8% +43.3% 3842866 ± 5% numa-vmstat.node1.numa_hit 2551311 ± 16% +0.4% 2561794 ± 10% +46.4% 3734454 ± 4% numa-vmstat.node1.numa_local 6233000 -0.3% 6214123 -1.6% 6135664 proc-vmstat.nr_dirty_background_threshold 12481241 -0.3% 12443440 -1.6% 12286330 proc-vmstat.nr_dirty_threshold 2044616 ± 14% +8.6% 2220456 ± 13% +47.8% 3022899 ± 2% proc-vmstat.nr_file_pages 62706078 -0.3% 62517035 -1.6% 61731279 proc-vmstat.nr_free_pages 1247557 ± 31% +17.9% 1471046 ± 19% +83.8% 2292396 ± 2% proc-vmstat.nr_inactive_anon 570709 ± 19% +17.2% 668608 ± 4% +38.0% 787666 ± 3% proc-vmstat.nr_mapped 1271419 ± 23% +13.8% 1447250 ± 20% +76.9% 2249692 ± 2% proc-vmstat.nr_shmem 44542 +2.7% 45755 +5.1% 46796 proc-vmstat.nr_slab_reclaimable 188881 -0.7% 187493 +1.4% 191440 proc-vmstat.nr_slab_unreclaimable 1247557 ± 31% +17.9% 1471046 ± 19% +83.8% 2292396 ± 2% proc-vmstat.nr_zone_inactive_anon 3714588 ± 9% +6.5% 3954189 ± 12% +54.0% 5719994 proc-vmstat.numa_hit 3479163 ± 10% +4.8% 3645729 ± 9% +57.3% 5472276 proc-vmstat.numa_local 642130 ± 68% -71.2% 184909 ± 11% -73.9% 167609 ± 14% proc-vmstat.pgactivate 4593887 ± 7% +1.2% 4650141 ± 7% +131.4% 10629624 proc-vmstat.pgalloc_normal 2170433 ± 2% +6.8% 2318318 ± 2% +6.4% 2308275 ± 5% proc-vmstat.pgfault 2631626 ± 7% -8.8% 2400780 +177.3% 7298788 proc-vmstat.pgfree 138302 ± 4% +6.7% 147631 ± 3% -0.7% 137382 ± 4% proc-vmstat.pgreuse 0.30 ± 3% -4.9% 0.28 ± 2% +20.7% 0.36 perf-stat.i.MPKI 6.354e+10 -11.4% 5.63e+10 +2.2% 6.497e+10 perf-stat.i.branch-instructions 0.53 -0.0 0.52 +0.0 0.57 perf-stat.i.branch-miss-rate% 3.226e+08 -12.5% 2.822e+08 +9.2% 3.523e+08 perf-stat.i.branch-misses 3.91 ± 3% -0.2 3.74 ± 2% +0.9 4.80 ± 2% perf-stat.i.cache-miss-rate% 94557935 ± 3% -15.2% 80197744 ± 2% +25.1% 1.183e+08 perf-stat.i.cache-misses 2.563e+09 -13.7% 2.212e+09 +1.3% 2.596e+09 perf-stat.i.cache-references 4710895 -15.9% 3959720 +36.7% 6439647 perf-stat.i.context-switches 1.86 +14.3% 2.13 -0.8% 1.85 perf-stat.i.cpi 224291 +0.0% 224353 +1.4% 227521 perf-stat.i.cpu-clock 6.345e+11 +0.4% 6.368e+11 +1.3% 6.424e+11 perf-stat.i.cpu-cycles 601598 -15.0% 511540 +13.7% 683759 ± 2% perf-stat.i.cpu-migrations 7390 ± 5% +28.4% 9492 ± 2% -23.5% 5655 perf-stat.i.cycles-between-cache-misses 3.408e+11 -12.1% 2.997e+11 +2.1% 3.478e+11 perf-stat.i.instructions 0.54 -12.2% 0.47 +0.8% 0.55 perf-stat.i.ipc 23.73 -15.9% 19.95 +32.3% 31.39 perf-stat.i.metric.K/sec 224291 +0.0% 224353 +1.4% 227521 perf-stat.i.task-clock 0.25 ± 35% +8.1% 0.27 ± 2% +38.0% 0.34 perf-stat.overall.MPKI 0.45 ± 35% +0.1 0.50 +0.1 0.54 perf-stat.overall.branch-miss-rate% 3.26 ± 35% +0.3 3.60 +1.3 4.54 perf-stat.overall.cache-miss-rate% 1.66 ± 35% +28.5% 2.13 +11.6% 1.85 perf-stat.overall.cpi 6006 ± 35% +33.5% 8020 ± 2% -9.1% 5458 perf-stat.overall.cycles-between-cache-misses 0.48 ± 35% -1.5% 0.47 +13.4% 0.54 perf-stat.overall.ipc 2.847e+08 ± 35% -1.6% 2.802e+08 +21.1% 3.448e+08 perf-stat.ps.branch-misses 83193631 ± 35% -5.0% 79040711 ± 2% +38.7% 1.154e+08 perf-stat.ps.cache-misses 4153465 ± 35% -5.4% 3930147 +52.2% 6323104 perf-stat.ps.context-switches 531469 ± 35% -4.7% 506531 +26.2% 670740 ± 2% perf-stat.ps.cpu-migrations 5.287e+13 ± 35% +15.1% 6.083e+13 +15.6% 6.114e+13 perf-stat.total.instructions 13829361 +51.8% 20989754 +6.0% 14655040 ± 15% sched_debug.cfs_rq:/.avg_vruntime.avg 18756074 ± 5% +44.2% 27055241 ± 3% +3.8% 19470555 ± 13% sched_debug.cfs_rq:/.avg_vruntime.max 12499623 ± 2% +52.4% 19043277 ± 2% +8.0% 13493604 ± 14% sched_debug.cfs_rq:/.avg_vruntime.min 8.93 ± 2% +14.1% 10.19 -5.2% 8.47 ± 7% sched_debug.cfs_rq:/.h_nr_running.avg 4.68 ± 3% +10.0% 5.15 ± 2% -1.5% 4.62 ± 8% sched_debug.cfs_rq:/.h_nr_running.stddev 0.44 ± 35% +75.8% 0.78 ± 19% -1.6% 0.44 ± 50% sched_debug.cfs_rq:/.load_avg.min 13829361 +51.8% 20989754 +6.0% 14655040 ± 15% sched_debug.cfs_rq:/.min_vruntime.avg 18756074 ± 5% +44.2% 27055241 ± 3% +3.8% 19470555 ± 13% sched_debug.cfs_rq:/.min_vruntime.max 12499623 ± 2% +52.4% 19043277 ± 2% +8.0% 13493604 ± 14% sched_debug.cfs_rq:/.min_vruntime.min 0.68 +11.7% 0.76 -1.3% 0.67 ± 2% sched_debug.cfs_rq:/.nr_running.avg 176.30 ± 3% -22.8% 136.16 ± 4% -2.9% 171.17 ± 10% sched_debug.cfs_rq:/.removed.runnable_avg.max 176.30 ± 3% -22.8% 136.16 ± 4% -2.9% 171.17 ± 10% sched_debug.cfs_rq:/.removed.util_avg.max 8995 +16.0% 10437 -5.7% 8484 ± 7% sched_debug.cfs_rq:/.runnable_avg.avg 18978 ± 6% +13.7% 21579 ± 6% -10.1% 17066 ± 11% sched_debug.cfs_rq:/.runnable_avg.max 2890 ± 4% +13.9% 3292 ± 3% -1.5% 2848 ± 9% sched_debug.cfs_rq:/.runnable_avg.stddev 1094 ± 3% +2.9% 1125 ± 3% +72.4% 1886 ± 5% sched_debug.cfs_rq:/.util_est.avg 4561 ± 14% +19.8% 5464 ± 19% +50.2% 6850 ± 7% sched_debug.cfs_rq:/.util_est.max 816.44 ± 6% +7.4% 877.16 ± 4% +78.4% 1456 ± 4% sched_debug.cfs_rq:/.util_est.stddev 415209 ± 22% -23.3% 318311 ± 3% -8.8% 378602 ± 6% sched_debug.cpu.avg_idle.avg 102333 ± 2% +30.5% 133496 ± 2% +4.0% 106415 ± 9% sched_debug.cpu.clock.avg 102519 ± 2% +30.4% 133722 ± 2% +4.0% 106590 ± 8% sched_debug.cpu.clock.max 102127 ± 2% +30.5% 133254 ± 2% +4.0% 106199 ± 9% sched_debug.cpu.clock.min 101839 ± 2% +30.5% 132880 ± 2% +4.0% 105862 ± 8% sched_debug.cpu.clock_task.avg 102169 ± 2% +30.4% 133268 ± 2% +3.9% 106186 ± 8% sched_debug.cpu.clock_task.max 87129 ± 2% +35.6% 118117 ± 2% +4.6% 91152 ± 10% sched_debug.cpu.clock_task.min 11573 +32.4% 15327 -0.3% 11534 sched_debug.cpu.curr->pid.avg 14704 +23.9% 18214 +3.2% 15178 ± 7% sched_debug.cpu.curr->pid.max 1516 ± 9% +16.4% 1765 ± 10% +33.3% 2021 ± 44% sched_debug.cpu.curr->pid.stddev 8.92 ± 2% +14.1% 10.18 -5.2% 8.46 ± 7% sched_debug.cpu.nr_running.avg 4.69 ± 2% +10.0% 5.16 ± 2% -1.7% 4.61 ± 8% sched_debug.cpu.nr_running.stddev 1232815 ± 2% +27.3% 1569099 +45.1% 1789014 ± 17% sched_debug.cpu.nr_switches.avg 1411362 ± 5% +26.8% 1789325 ± 3% +48.6% 2097828 ± 20% sched_debug.cpu.nr_switches.max 1045767 ± 2% +27.3% 1331341 ± 3% +42.5% 1489877 ± 17% sched_debug.cpu.nr_switches.min 102127 ± 2% +30.5% 133250 ± 2% +4.0% 106198 ± 9% sched_debug.cpu_clk 101071 ± 2% +30.8% 132194 ± 2% +4.0% 105143 ± 9% sched_debug.ktime 0.00 -25.0% 0.00 -3.1% 0.00 ± 8% sched_debug.rt_rq:.rt_nr_running.avg 0.33 -25.0% 0.25 -3.1% 0.32 ± 8% sched_debug.rt_rq:.rt_nr_running.max 0.02 -25.0% 0.02 -3.1% 0.02 ± 8% sched_debug.rt_rq:.rt_nr_running.stddev 102997 ± 2% +30.2% 134142 ± 2% +4.0% 107069 ± 9% sched_debug.sched_clk 16347631 +100.0% 32695263 +100.0% 32695263 sched_debug.sysctl_sched.sysctl_sched_features 3.31 ±123% -42.6% 1.90 ± 86% -89.1% 0.36 ± 88% perf-sched.sch_delay.avg.ms.__cond_resched.__alloc_pages_noprof.alloc_pages_mpol_noprof.folio_alloc_mpol_noprof.shmem_alloc_folio 8.46 ± 38% +100.8% 16.98 ± 82% -85.1% 1.26 ±134% perf-sched.sch_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof 5.72 ± 38% +21.2% 6.93 ± 80% -85.6% 0.83 ± 85% perf-sched.sch_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc 7.88 ± 21% +11.3% 8.77 ± 59% -89.0% 0.87 ± 38% perf-sched.sch_delay.avg.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags 7.16 ± 36% -24.4% 5.42 ± 73% -70.7% 2.10 ± 66% perf-sched.sch_delay.avg.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity 28.76 ±142% -62.8% 10.70 ±198% -99.4% 0.18 ±233% perf-sched.sch_delay.avg.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate 3.75 ± 24% +0.6% 3.77 ± 59% -62.1% 1.42 ± 39% perf-sched.sch_delay.avg.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter 10.26 ± 19% -9.5% 9.29 ± 62% -74.3% 2.64 ± 40% perf-sched.sch_delay.avg.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write 2.82 ± 43% -22.6% 2.18 ± 73% -83.9% 0.45 ± 59% perf-sched.sch_delay.avg.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write 2.30 ± 29% -4.8% 2.19 ± 61% -72.3% 0.64 ± 38% perf-sched.sch_delay.avg.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb 1.75 ± 49% -5.1% 1.66 ± 65% -61.1% 0.68 ± 39% perf-sched.sch_delay.avg.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg 4.13 ± 87% +0.3% 4.15 ± 88% -86.0% 0.58 ±133% perf-sched.sch_delay.avg.ms.__cond_resched.shmem_inode_acct_blocks.shmem_alloc_and_add_folio.shmem_get_folio_gfp.shmem_write_begin 2.81 ± 59% -20.7% 2.23 ± 61% -72.8% 0.76 ± 38% perf-sched.sch_delay.avg.ms.__cond_resched.task_work_run.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe 9.13 ± 30% +0.5% 9.17 ± 67% -87.4% 1.15 ± 53% perf-sched.sch_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] 9.94 ± 32% -38.3% 6.13 ± 62% -56.9% 4.28 ± 37% perf-sched.sch_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_reschedule_ipi.[unknown].[unknown] 11.06 ± 57% -45.0% 6.08 ± 75% -90.4% 1.06 ± 60% perf-sched.sch_delay.avg.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread 16.36 ± 24% -23.9% 12.45 ± 59% -53.7% 7.57 ± 38% perf-sched.sch_delay.avg.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter 1.77 ± 23% -28.5% 1.26 ± 60% -66.1% 0.60 ± 38% perf-sched.sch_delay.avg.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg 2.66 ± 21% -15.5% 2.25 ± 58% -63.8% 0.96 ± 37% perf-sched.sch_delay.avg.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] 226.74 ±108% -40.0% 136.06 ±120% -95.1% 11.09 ± 91% perf-sched.sch_delay.avg.ms.worker_thread.kthread.ret_from_fork.ret_from_fork_asm 1412 ± 25% -7.0% 1313 ± 66% -83.6% 231.23 ± 47% perf-sched.sch_delay.max.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags 1221 ± 4% -23.0% 940.87 ± 58% -51.1% 596.88 ± 89% perf-sched.sch_delay.max.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity 262.66 ±176% -72.1% 73.22 ±239% -99.9% 0.35 ±248% perf-sched.sch_delay.max.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate 779.32 ± 29% -15.9% 655.34 ± 72% -80.4% 152.55 ± 45% perf-sched.sch_delay.max.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter 1042 ± 23% -8.4% 954.56 ± 63% -81.8% 190.25 ± 44% perf-sched.sch_delay.max.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write 463.55 ± 83% -3.7% 446.18 ± 87% -85.8% 65.73 ± 93% perf-sched.sch_delay.max.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write 1128 ± 17% -17.7% 928.10 ± 62% -81.0% 214.51 ± 40% perf-sched.sch_delay.max.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb 500.79 ± 50% -25.9% 370.90 ± 79% -76.0% 120.40 ± 46% perf-sched.sch_delay.max.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg 744.38 ± 27% +0.4% 747.54 ± 78% -86.3% 102.18 ± 71% perf-sched.sch_delay.max.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] 579.24 ± 43% -10.1% 520.76 ± 62% -65.3% 201.13 ± 62% perf-sched.sch_delay.max.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread 1945 ± 16% -15.9% 1635 ± 61% -73.4% 516.65 ± 40% perf-sched.sch_delay.max.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter 1530 ± 22% -7.2% 1420 ± 63% -75.5% 375.37 ± 40% perf-sched.sch_delay.max.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg 3.03 ± 23% -26.5% 2.23 ± 59% -65.2% 1.05 ± 38% perf-sched.total_sch_delay.average.ms 8.72 ± 23% -28.5% 6.23 ± 59% -64.2% 3.12 ± 38% perf-sched.total_wait_and_delay.average.ms 8284 ± 27% -38.2% 5118 ± 60% -41.6% 4839 ± 44% perf-sched.total_wait_and_delay.max.ms 5.68 ± 23% -29.6% 4.00 ± 59% -63.7% 2.06 ± 38% perf-sched.total_wait_time.average.ms 5811 ± 8% -28.5% 4155 ± 59% -27.8% 4194 ± 38% perf-sched.total_wait_time.max.ms 37.25 ± 91% +17.8% 43.88 ± 72% -90.7% 3.45 ±136% perf-sched.wait_and_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof 44.86 ± 74% +10.0% 49.33 ±117% -95.6% 1.99 ±132% perf-sched.wait_and_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc 22.30 ± 21% +6.1% 23.66 ± 59% -100.0% 0.00 perf-sched.wait_and_delay.avg.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags 17.29 ± 21% -28.7% 12.33 ± 66% -60.1% 6.90 ± 45% perf-sched.wait_and_delay.avg.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity 28.98 ± 21% -13.8% 24.98 ± 61% -74.3% 7.45 ± 39% perf-sched.wait_and_delay.avg.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write 27.08 ± 28% -10.4% 24.27 ± 64% -93.8% 1.69 ±129% perf-sched.wait_and_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] 27.52 ± 28% -37.8% 17.11 ± 61% -54.9% 12.40 ± 37% perf-sched.wait_and_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_reschedule_ipi.[unknown].[unknown] 38.19 ± 43% -40.4% 22.76 ± 65% -80.4% 7.50 ± 40% perf-sched.wait_and_delay.avg.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread 45.63 ± 24% -25.8% 33.84 ± 59% -53.4% 21.28 ± 38% perf-sched.wait_and_delay.avg.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter 5.06 ± 24% -30.6% 3.51 ± 60% -65.6% 1.74 ± 38% perf-sched.wait_and_delay.avg.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg 7.23 ± 22% -18.4% 5.90 ± 59% -64.2% 2.59 ± 38% perf-sched.wait_and_delay.avg.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] 17219 ± 26% +18.6% 20422 ± 60% -100.0% 0.00 perf-sched.wait_and_delay.count.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags 287.02 ± 72% -56.9% 123.78 ± 90% -90.0% 28.73 ±130% perf-sched.wait_and_delay.max.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc 2851 ± 25% -5.3% 2699 ± 64% -100.0% 0.00 perf-sched.wait_and_delay.max.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags 2443 ± 4% -23.0% 1881 ± 58% -37.8% 1519 ± 53% perf-sched.wait_and_delay.max.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity 2132 ± 23% -9.7% 1925 ± 63% -82.0% 384.48 ± 45% perf-sched.wait_and_delay.max.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write 3449 ± 45% -35.6% 2220 ± 71% -59.1% 1410 ± 49% perf-sched.wait_and_delay.max.ms.do_wait.kernel_wait4.__do_sys_wait4.do_syscall_64 1507 ± 27% +0.2% 1510 ± 79% -93.3% 101.36 ±137% perf-sched.wait_and_delay.max.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] 1312 ± 35% -16.4% 1097 ± 61% -65.2% 456.55 ± 64% perf-sched.wait_and_delay.max.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread 3951 ± 16% -17.0% 3278 ± 61% -73.7% 1038 ± 40% perf-sched.wait_and_delay.max.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter 3088 ± 22% -7.9% 2844 ± 62% -75.1% 768.96 ± 41% perf-sched.wait_and_delay.max.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg 3.31 ±123% -42.6% 1.90 ± 86% -89.1% 0.36 ± 88% perf-sched.wait_time.avg.ms.__cond_resched.__alloc_pages_noprof.alloc_pages_mpol_noprof.folio_alloc_mpol_noprof.shmem_alloc_folio 28.79 ±114% -6.6% 26.89 ± 70% -90.2% 2.82 ± 96% perf-sched.wait_time.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof 39.14 ± 89% +8.3% 42.40 ±140% -95.1% 1.92 ± 78% perf-sched.wait_time.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc 14.42 ± 21% +3.3% 14.89 ± 59% -88.8% 1.61 ± 37% perf-sched.wait_time.avg.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags 10.13 ± 17% -31.7% 6.92 ± 63% -52.6% 4.80 ± 41% perf-sched.wait_time.avg.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity 40.86 ±110% -46.6% 21.80 ±152% -99.1% 0.35 ±248% perf-sched.wait_time.avg.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate 6.56 ± 23% -5.6% 6.20 ± 59% -61.9% 2.50 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter 18.72 ± 22% -16.2% 15.69 ± 61% -74.3% 4.81 ± 39% perf-sched.wait_time.avg.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write 2.82 ± 43% -22.6% 2.18 ± 73% -83.9% 0.45 ± 59% perf-sched.wait_time.avg.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write 5.03 ± 24% -6.1% 4.72 ± 59% -63.4% 1.84 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb 3.16 ± 33% -19.7% 2.54 ± 65% -60.4% 1.25 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg 4.13 ± 87% +0.3% 4.15 ± 88% -86.0% 0.58 ±133% perf-sched.wait_time.avg.ms.__cond_resched.shmem_inode_acct_blocks.shmem_alloc_and_add_folio.shmem_get_folio_gfp.shmem_write_begin 4.42 ± 44% -21.0% 3.49 ± 65% -71.3% 1.27 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.task_work_run.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe 224.55 ± 53% -20.5% 178.59 ± 71% -71.8% 63.25 ± 68% perf-sched.wait_time.avg.ms.do_wait.kernel_wait4.__do_sys_wait4.do_syscall_64 17.95 ± 30% -15.9% 15.10 ± 63% -89.3% 1.91 ± 46% perf-sched.wait_time.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] 17.58 ± 26% -37.5% 10.98 ± 61% -53.8% 8.12 ± 38% perf-sched.wait_time.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_reschedule_ipi.[unknown].[unknown] 27.13 ± 39% -38.5% 16.68 ± 63% -76.3% 6.43 ± 38% perf-sched.wait_time.avg.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread 29.27 ± 24% -26.9% 21.39 ± 59% -53.2% 13.71 ± 38% perf-sched.wait_time.avg.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter 3.29 ± 24% -31.7% 2.25 ± 60% -65.3% 1.14 ± 38% perf-sched.wait_time.avg.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg 4.57 ± 22% -20.1% 3.66 ± 59% -64.4% 1.63 ± 38% perf-sched.wait_time.avg.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] 1446 ± 25% -3.4% 1396 ± 63% -80.6% 280.50 ± 42% perf-sched.wait_time.max.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags 292.35 ±159% -68.1% 93.21 ±217% -99.8% 0.69 ±256% perf-sched.wait_time.max.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate 932.87 ± 20% -20.6% 740.44 ± 65% -81.7% 170.42 ± 45% perf-sched.wait_time.max.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter 1110 ± 24% -11.8% 979.30 ± 62% -81.7% 203.13 ± 43% perf-sched.wait_time.max.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write 463.55 ± 83% -3.7% 446.18 ± 87% -85.8% 65.73 ± 93% perf-sched.wait_time.max.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write 1203 ± 20% -19.5% 969.24 ± 62% -80.5% 234.88 ± 42% perf-sched.wait_time.max.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb 614.81 ± 41% -27.8% 443.79 ± 79% -79.3% 127.43 ± 46% perf-sched.wait_time.max.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg 822.39 ± 32% -24.1% 624.54 ± 74% -82.9% 140.70 ± 43% perf-sched.wait_time.max.ms.__cond_resched.task_work_run.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe 817.75 ± 28% -6.8% 762.54 ± 79% -85.7% 117.26 ± 58% perf-sched.wait_time.max.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] 774.10 ± 27% -22.5% 600.11 ± 61% -63.0% 286.19 ± 57% perf-sched.wait_time.max.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread 2023 ± 16% -18.8% 1643 ± 60% -74.2% 521.59 ± 40% perf-sched.wait_time.max.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter 1624 ± 25% -12.1% 1428 ± 62% -75.0% 405.54 ± 42% perf-sched.wait_time.max.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg 2801 ± 45% -40.4% 1669 ± 81% -54.5% 1273 ± 52% perf-sched.wait_time.max.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] 55.19 -0.7 54.46 -0.6 54.55 perf-profile.calltrace.cycles-pp.read 53.91 -0.6 53.30 -0.8 53.13 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.read 53.76 -0.6 53.16 -0.8 52.98 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.read 52.41 -0.5 51.92 -1.1 51.32 perf-profile.calltrace.cycles-pp.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe.read 52.02 -0.5 51.56 -1.1 50.92 perf-profile.calltrace.cycles-pp.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe.read 50.82 -0.4 50.46 -1.2 49.67 perf-profile.calltrace.cycles-pp.sock_read_iter.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe 50.29 -0.3 49.98 -1.2 49.12 perf-profile.calltrace.cycles-pp.sock_recvmsg.sock_read_iter.vfs_read.ksys_read.do_syscall_64 49.61 -0.3 49.36 -1.2 48.42 perf-profile.calltrace.cycles-pp.unix_stream_recvmsg.sock_recvmsg.sock_read_iter.vfs_read.ksys_read 49.37 -0.2 49.13 -1.2 48.17 perf-profile.calltrace.cycles-pp.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg.sock_read_iter.vfs_read 24.50 -0.2 24.29 ± 4% -1.1 23.39 perf-profile.calltrace.cycles-pp.alloc_skb_with_frags.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter.vfs_write 5.54 ± 2% -0.2 5.34 ± 6% -0.3 5.25 ± 2% perf-profile.calltrace.cycles-pp.unix_stream_read_actor.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg.sock_read_iter 5.49 ± 2% -0.2 5.29 ± 6% -0.3 5.20 ± 2% perf-profile.calltrace.cycles-pp.skb_copy_datagram_iter.unix_stream_read_actor.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg 5.40 ± 2% -0.2 5.21 ± 6% -0.3 5.11 ± 2% perf-profile.calltrace.cycles-pp.__skb_datagram_iter.skb_copy_datagram_iter.unix_stream_read_actor.unix_stream_read_generic.unix_stream_recvmsg 24.31 -0.2 24.12 ± 4% -1.1 23.20 perf-profile.calltrace.cycles-pp.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter 2.86 ± 2% -0.2 2.68 ± 5% -0.2 2.65 ± 2% perf-profile.calltrace.cycles-pp.__memcg_slab_free_hook.kmem_cache_free.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg 1.60 ± 2% -0.1 1.45 ± 5% -0.0 1.57 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.read 1.50 ± 2% -0.1 1.36 ± 6% -0.0 1.50 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.write 0.62 ± 2% -0.1 0.50 ± 38% +0.3 0.95 perf-profile.calltrace.cycles-pp.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function.__wake_up_common 2.76 ± 2% -0.1 2.66 ± 6% -0.3 2.45 ± 2% perf-profile.calltrace.cycles-pp.simple_copy_to_iter.__skb_datagram_iter.skb_copy_datagram_iter.unix_stream_read_actor.unix_stream_read_generic 2.65 ± 2% -0.1 2.55 ± 6% -0.3 2.34 ± 2% perf-profile.calltrace.cycles-pp.__check_object_size.simple_copy_to_iter.__skb_datagram_iter.skb_copy_datagram_iter.unix_stream_read_actor 0.97 ± 2% -0.1 0.89 ± 5% +0.3 1.26 perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.read 0.78 ± 2% -0.1 0.71 ± 7% +0.4 1.18 perf-profile.calltrace.cycles-pp.ttwu_do_activate.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_sync_key 1.24 ± 2% -0.1 1.16 ± 5% +0.1 1.31 ± 2% perf-profile.calltrace.cycles-pp.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb 2.20 ± 2% -0.1 2.14 ± 6% -0.3 1.90 ± 2% perf-profile.calltrace.cycles-pp.check_heap_object.__check_object_size.simple_copy_to_iter.__skb_datagram_iter.skb_copy_datagram_iter 26.13 -0.0 26.09 ± 3% -1.0 25.11 perf-profile.calltrace.cycles-pp.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter.vfs_write.ksys_write 0.53 ± 2% -0.0 0.49 ± 38% +0.3 0.79 ± 2% perf-profile.calltrace.cycles-pp.dequeue_task_fair.__schedule.schedule.schedule_timeout.unix_stream_data_wait 1.03 ± 3% -0.0 1.02 ± 9% +0.1 1.12 perf-profile.calltrace.cycles-pp.skb_set_owner_w.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter.vfs_write 0.66 ± 2% -0.0 0.65 ± 6% +0.1 0.79 perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.write 0.00 +0.0 0.00 +0.6 0.61 perf-profile.calltrace.cycles-pp.enqueue_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function 1.83 ± 2% +0.1 1.88 ± 6% +0.8 2.61 perf-profile.calltrace.cycles-pp.schedule.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg 1.87 ± 2% +0.1 1.93 ± 6% +0.8 2.68 perf-profile.calltrace.cycles-pp.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg 1.79 ± 2% +0.1 1.85 ± 6% +0.8 2.56 perf-profile.calltrace.cycles-pp.__schedule.schedule.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic 2.30 ± 2% +0.1 2.37 ± 6% +1.0 3.30 perf-profile.calltrace.cycles-pp.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg.sock_read_iter 0.40 ± 53% +0.1 0.48 ± 38% +0.4 0.77 ± 2% perf-profile.calltrace.cycles-pp.dequeue_entities.dequeue_task_fair.__schedule.schedule.schedule_timeout 0.00 +0.3 0.26 ±100% +0.6 0.61 perf-profile.calltrace.cycles-pp.pick_next_task_fair.__schedule.schedule.schedule_timeout.unix_stream_data_wait 1.62 ± 3% +0.3 1.93 ± 24% +0.9 2.48 perf-profile.calltrace.cycles-pp.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_sync_key.sock_def_readable 1.63 ± 3% +0.3 1.94 ± 24% +0.9 2.50 perf-profile.calltrace.cycles-pp.autoremove_wake_function.__wake_up_common.__wake_up_sync_key.sock_def_readable.unix_stream_sendmsg 1.71 ± 3% +0.3 2.03 ± 23% +0.9 2.61 perf-profile.calltrace.cycles-pp.__wake_up_common.__wake_up_sync_key.sock_def_readable.unix_stream_sendmsg.sock_write_iter 39.00 +0.5 39.50 +0.1 39.13 perf-profile.calltrace.cycles-pp.sock_write_iter.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe 37.56 +0.6 38.15 +0.1 37.67 perf-profile.calltrace.cycles-pp.unix_stream_sendmsg.sock_write_iter.vfs_write.ksys_write.do_syscall_64 1.96 ± 3% +0.7 2.63 ± 24% +1.1 3.08 perf-profile.calltrace.cycles-pp.__wake_up_sync_key.sock_def_readable.unix_stream_sendmsg.sock_write_iter.vfs_write 3.52 ± 2% +0.7 4.23 ± 15% +1.1 4.62 ± 2% perf-profile.calltrace.cycles-pp.sock_def_readable.unix_stream_sendmsg.sock_write_iter.vfs_write.ksys_write 55.93 -0.8 55.12 -0.7 55.27 perf-profile.children.cycles-pp.read 52.45 -0.5 51.95 -1.1 51.36 perf-profile.children.cycles-pp.ksys_read 52.06 -0.5 51.60 -1.1 50.96 perf-profile.children.cycles-pp.vfs_read 50.85 -0.4 50.48 -1.2 49.69 perf-profile.children.cycles-pp.sock_read_iter 50.34 -0.3 50.03 -1.2 49.17 perf-profile.children.cycles-pp.sock_recvmsg 4.68 ± 2% -0.3 4.38 ± 5% -0.2 4.49 perf-profile.children.cycles-pp.__memcg_slab_free_hook 49.64 -0.3 49.38 -1.2 48.45 perf-profile.children.cycles-pp.unix_stream_recvmsg 49.50 -0.2 49.25 -1.2 48.30 perf-profile.children.cycles-pp.unix_stream_read_generic 24.52 -0.2 24.32 ± 4% -1.1 23.41 perf-profile.children.cycles-pp.alloc_skb_with_frags 5.57 ± 2% -0.2 5.36 ± 6% -0.3 5.27 ± 2% perf-profile.children.cycles-pp.unix_stream_read_actor 5.52 ± 2% -0.2 5.31 ± 6% -0.3 5.22 ± 2% perf-profile.children.cycles-pp.skb_copy_datagram_iter 24.38 -0.2 24.18 ± 4% -1.1 23.27 perf-profile.children.cycles-pp.__alloc_skb 5.43 ± 2% -0.2 5.24 ± 6% -0.3 5.14 ± 2% perf-profile.children.cycles-pp.__skb_datagram_iter 95.87 -0.2 95.70 -0.5 95.42 perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe 3.46 ± 2% -0.2 3.29 ± 6% -0.3 3.16 perf-profile.children.cycles-pp.__check_object_size 1.77 ± 2% -0.2 1.61 ± 5% -0.0 1.76 perf-profile.children.cycles-pp.entry_SYSCALL_64 1.87 ± 2% -0.2 1.72 ± 5% +0.0 1.87 ± 2% perf-profile.children.cycles-pp.mod_objcg_state 95.57 -0.1 95.43 -0.5 95.11 perf-profile.children.cycles-pp.do_syscall_64 2.79 ± 2% -0.1 2.68 ± 6% -0.3 2.48 perf-profile.children.cycles-pp.simple_copy_to_iter 2.62 ± 2% -0.1 2.53 ± 6% -0.3 2.33 ± 2% perf-profile.children.cycles-pp.check_heap_object 1.69 ± 2% -0.1 1.60 ± 5% +0.4 2.12 perf-profile.children.cycles-pp.syscall_exit_to_user_mode 0.90 ± 2% -0.1 0.83 ± 5% -0.0 0.90 perf-profile.children.cycles-pp.entry_SYSRETQ_unsafe_stack 0.85 ± 2% -0.1 0.78 ± 5% +0.0 0.86 ± 2% perf-profile.children.cycles-pp.obj_cgroup_charge 0.10 ± 4% -0.1 0.04 ±104% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.__handle_mm_fault 0.10 ± 4% -0.1 0.04 ±104% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.handle_mm_fault 0.10 ± 4% -0.1 0.04 ±102% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.do_user_addr_fault 0.10 ± 4% -0.1 0.04 ±102% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.exc_page_fault 0.63 ± 2% -0.1 0.57 ± 5% +0.0 0.64 perf-profile.children.cycles-pp.__cond_resched 26.17 -0.0 26.13 ± 3% -1.0 25.15 perf-profile.children.cycles-pp.sock_alloc_send_pskb 0.10 ± 4% -0.0 0.05 ± 63% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.asm_exc_page_fault 0.32 ± 2% -0.0 0.28 ± 3% +0.0 0.33 perf-profile.children.cycles-pp.task_mm_cid_work 0.32 ± 2% -0.0 0.28 ± 3% +0.0 0.33 perf-profile.children.cycles-pp.task_work_run 0.34 ± 2% -0.0 0.30 ± 5% -0.0 0.34 ± 2% perf-profile.children.cycles-pp.rcu_all_qs 0.32 ± 3% -0.0 0.29 ± 6% +0.0 0.33 ± 2% perf-profile.children.cycles-pp.__virt_addr_valid 0.32 ± 3% -0.0 0.29 ± 5% +0.0 0.36 ± 2% perf-profile.children.cycles-pp.syscall_return_via_sysret 0.05 -0.0 0.02 ±129% +0.0 0.07 ± 4% perf-profile.children.cycles-pp.avg_vruntime 0.18 ± 4% -0.0 0.16 ± 7% +0.1 0.27 perf-profile.children.cycles-pp.__enqueue_entity 0.47 ± 2% -0.0 0.44 ± 5% +0.0 0.50 perf-profile.children.cycles-pp.mutex_lock 0.23 ± 3% -0.0 0.21 ± 6% +0.1 0.33 ± 2% perf-profile.children.cycles-pp.set_next_entity 0.14 ± 4% -0.0 0.12 ± 4% +0.1 0.20 ± 2% perf-profile.children.cycles-pp.__dequeue_entity 1.05 ± 3% -0.0 1.04 ± 9% +0.1 1.14 perf-profile.children.cycles-pp.skb_set_owner_w 0.28 ± 2% -0.0 0.26 ± 7% +0.1 0.38 perf-profile.children.cycles-pp.switch_fpu_return 0.48 ± 2% -0.0 0.47 ± 6% +0.2 0.70 perf-profile.children.cycles-pp.enqueue_entity 0.08 ± 3% -0.0 0.07 ± 9% +0.0 0.13 ± 6% perf-profile.children.cycles-pp.asm_sysvec_reschedule_ipi 0.24 ± 4% -0.0 0.22 ± 6% +0.1 0.31 ± 4% perf-profile.children.cycles-pp.__rseq_handle_notify_resume 0.06 -0.0 0.05 +0.0 0.08 ± 4% perf-profile.children.cycles-pp.cpuacct_charge 0.29 ± 3% -0.0 0.28 ± 7% +0.1 0.40 ± 4% perf-profile.children.cycles-pp.switch_mm_irqs_off 0.16 ± 2% -0.0 0.15 ± 4% +0.0 0.18 perf-profile.children.cycles-pp.syscall_exit_to_user_mode_prepare 0.12 ± 3% -0.0 0.12 ± 6% +0.1 0.18 perf-profile.children.cycles-pp.update_rq_clock_task 0.18 ± 3% -0.0 0.17 ± 6% +0.1 0.24 ± 2% perf-profile.children.cycles-pp.restore_fpregs_from_fpstate 0.18 ± 4% -0.0 0.17 ± 6% +0.1 0.23 ± 3% perf-profile.children.cycles-pp.rseq_ip_fixup 0.05 ± 6% -0.0 0.04 ± 37% +0.0 0.07 ± 5% perf-profile.children.cycles-pp.rseq_update_cpu_node_id 0.05 -0.0 0.04 ± 37% +0.0 0.06 ± 7% perf-profile.children.cycles-pp.sched_clock 0.07 ± 7% -0.0 0.06 ± 9% +0.0 0.09 ± 5% perf-profile.children.cycles-pp.os_xsave 0.13 ± 6% -0.0 0.12 ± 6% +0.0 0.17 ± 3% perf-profile.children.cycles-pp.rseq_get_rseq_cs 0.12 ± 6% -0.0 0.12 ± 7% +0.0 0.16 ± 4% perf-profile.children.cycles-pp.__get_user_8 0.14 ± 3% -0.0 0.13 ± 7% +0.1 0.20 ± 3% perf-profile.children.cycles-pp.wakeup_preempt 0.09 ± 5% -0.0 0.09 ± 5% +0.0 0.11 ± 4% perf-profile.children.cycles-pp.cgroup_rstat_updated 0.14 ± 3% -0.0 0.13 ± 4% +0.0 0.18 ± 3% perf-profile.children.cycles-pp.__update_load_avg_se 0.18 ± 2% -0.0 0.17 ± 2% -0.0 0.14 ± 2% perf-profile.children.cycles-pp.put_cpu_partial 0.13 ± 3% -0.0 0.13 ± 7% +0.0 0.16 ± 3% perf-profile.children.cycles-pp.pick_eevdf 0.07 ± 4% -0.0 0.07 ± 7% +0.0 0.10 ± 5% perf-profile.children.cycles-pp.check_preempt_wakeup_fair 0.08 ± 5% -0.0 0.08 ± 8% +0.0 0.13 perf-profile.children.cycles-pp.update_curr_se 0.06 ± 5% -0.0 0.06 ± 5% +0.0 0.07 ± 6% perf-profile.children.cycles-pp.sched_clock_cpu 0.07 +0.0 0.07 ± 7% +0.0 0.11 perf-profile.children.cycles-pp.put_prev_entity 0.00 +0.0 0.00 +0.1 0.05 perf-profile.children.cycles-pp.__put_user_8 0.06 ± 10% +0.0 0.06 ± 21% +0.4 0.43 ± 4% perf-profile.children.cycles-pp.get_any_partial 0.22 ± 3% +0.0 0.22 ± 7% +0.1 0.31 ± 2% perf-profile.children.cycles-pp.update_cfs_group 0.06 ± 6% +0.0 0.06 ± 10% +0.0 0.09 ± 5% perf-profile.children.cycles-pp.___perf_sw_event 0.12 ± 2% +0.0 0.12 ± 7% +0.1 0.17 ± 2% perf-profile.children.cycles-pp.wake_affine 0.07 +0.0 0.07 ± 11% +0.0 0.10 ± 4% perf-profile.children.cycles-pp.update_min_vruntime 0.10 ± 4% +0.0 0.11 ± 7% +0.0 0.15 ± 3% perf-profile.children.cycles-pp.__update_load_avg_cfs_rq 0.05 ± 35% +0.0 0.06 ± 5% +0.0 0.08 ± 5% perf-profile.children.cycles-pp.perf_tp_event 0.09 ± 35% +0.0 0.10 ± 7% +0.0 0.14 ± 3% perf-profile.children.cycles-pp.perf_trace_sched_wakeup_template 0.05 ± 36% +0.0 0.05 ± 6% +0.0 0.08 ± 4% perf-profile.children.cycles-pp.place_entity 0.07 ± 4% +0.0 0.08 ± 9% +0.0 0.10 ± 3% perf-profile.children.cycles-pp.task_h_load 0.00 +0.0 0.01 ±264% +0.1 0.06 ± 9% perf-profile.children.cycles-pp.native_irq_return_iret 0.12 ± 4% +0.0 0.12 ± 8% +0.1 0.17 ± 4% perf-profile.children.cycles-pp.__switch_to 0.12 ± 3% +0.0 0.13 ± 6% +0.0 0.16 ± 2% perf-profile.children.cycles-pp.__switch_to_asm 0.75 ± 2% +0.0 0.76 ± 7% +0.3 1.08 perf-profile.children.cycles-pp.enqueue_task_fair 0.16 ± 5% +0.0 0.17 ± 7% +0.0 0.20 ± 5% perf-profile.children.cycles-pp.task_tick_fair 0.47 ± 2% +0.0 0.49 ± 6% +0.2 0.64 perf-profile.children.cycles-pp.update_load_avg 0.64 ± 2% +0.0 0.65 ± 6% +0.3 0.90 perf-profile.children.cycles-pp.dequeue_task_fair 0.76 ± 2% +0.0 0.78 ± 4% +0.2 0.98 perf-profile.children.cycles-pp.pick_next_task_fair 0.22 ± 3% +0.0 0.23 ± 7% +0.1 0.28 ± 2% perf-profile.children.cycles-pp.pick_task_fair 0.12 ± 4% +0.0 0.13 ± 14% +0.1 0.17 perf-profile.children.cycles-pp.reweight_entity 0.20 ± 3% +0.0 0.22 ± 6% +0.0 0.24 ± 4% perf-profile.children.cycles-pp.sched_tick 0.88 ± 2% +0.0 0.90 ± 7% +0.4 1.29 perf-profile.children.cycles-pp.ttwu_do_activate 0.30 ± 4% +0.0 0.32 ± 7% +0.1 0.38 ± 2% perf-profile.children.cycles-pp.tick_nohz_handler 0.35 ± 2% +0.0 0.37 ± 5% +0.1 0.49 perf-profile.children.cycles-pp.dequeue_entity 0.31 ± 4% +0.0 0.33 ± 6% +0.1 0.38 ± 2% perf-profile.children.cycles-pp.__hrtimer_run_queues 0.29 ± 4% +0.0 0.31 ± 6% +0.1 0.36 ± 3% perf-profile.children.cycles-pp.update_process_times 0.13 ± 2% +0.0 0.15 ± 11% +0.1 0.19 ± 2% perf-profile.children.cycles-pp.finish_task_switch 0.18 ± 2% +0.0 0.21 ± 8% +0.1 0.26 perf-profile.children.cycles-pp.prepare_task_switch 0.35 ± 3% +0.0 0.37 ± 5% +0.1 0.43 ± 2% perf-profile.children.cycles-pp.hrtimer_interrupt 0.35 ± 3% +0.0 0.37 ± 6% +0.1 0.44 ± 2% perf-profile.children.cycles-pp.__sysvec_apic_timer_interrupt 0.63 ± 2% +0.0 0.65 ± 7% +0.3 0.89 ± 2% perf-profile.children.cycles-pp.dequeue_entities 0.09 ± 8% +0.0 0.12 ± 7% +0.0 0.11 ± 11% perf-profile.children.cycles-pp.generic_perform_write 2.21 ± 2% +0.0 2.23 ± 6% +0.8 3.04 perf-profile.children.cycles-pp.schedule_timeout 0.07 ± 10% +0.0 0.10 ± 11% +0.0 0.08 ± 20% perf-profile.children.cycles-pp.sched_balance_find_src_group 0.06 ± 10% +0.0 0.09 ± 11% +0.0 0.06 ± 40% perf-profile.children.cycles-pp.update_sg_lb_stats 0.07 ± 10% +0.0 0.10 ± 11% +0.0 0.07 ± 19% perf-profile.children.cycles-pp.update_sd_lb_stats 0.10 ± 3% +0.0 0.13 ± 14% +0.0 0.13 ± 5% perf-profile.children.cycles-pp.update_rq_clock 0.01 ±282% +0.0 0.04 ± 57% +0.1 0.06 ± 7% perf-profile.children.cycles-pp.native_sched_clock 0.43 ± 3% +0.0 0.47 ± 7% +0.2 0.64 perf-profile.children.cycles-pp.update_curr 0.13 ± 8% +0.0 0.16 ± 8% +0.0 0.16 ± 12% perf-profile.children.cycles-pp.writen 0.39 ± 3% +0.0 0.42 ± 6% +0.1 0.48 ± 2% perf-profile.children.cycles-pp.sysvec_apic_timer_interrupt 0.42 ± 3% +0.0 0.46 ± 6% +0.1 0.52 ± 2% perf-profile.children.cycles-pp.asm_sysvec_apic_timer_interrupt 0.02 ±111% +0.0 0.06 ± 13% +0.0 0.04 ± 38% perf-profile.children.cycles-pp.set_task_cpu 0.02 ±141% +0.0 0.06 ± 11% +0.0 0.06 ± 11% perf-profile.children.cycles-pp.ring_buffer_read_head 2.45 ± 2% +0.0 2.50 ± 6% +1.0 3.43 perf-profile.children.cycles-pp.schedule 0.00 +0.1 0.06 ± 8% +0.0 0.03 ± 77% perf-profile.children.cycles-pp.vruntime_eligible 0.22 ± 8% +0.1 0.28 ± 9% +0.1 0.27 ± 12% perf-profile.children.cycles-pp.perf_mmap__push 0.23 ± 8% +0.1 0.29 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.record__mmap_read_evlist 0.23 ± 7% +0.1 0.29 ± 9% +0.1 0.28 ± 11% perf-profile.children.cycles-pp.cmd_record 0.23 ± 7% +0.1 0.30 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.handle_internal_command 0.23 ± 7% +0.1 0.30 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.main 0.23 ± 7% +0.1 0.30 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.run_builtin 2.31 ± 2% +0.1 2.38 ± 7% +1.0 3.32 perf-profile.children.cycles-pp.unix_stream_data_wait 0.18 ± 38% +0.1 0.26 ± 17% +0.2 0.41 ± 3% perf-profile.children.cycles-pp.ordered_events__queue 0.18 ± 38% +0.1 0.26 ± 17% +0.2 0.41 ± 3% perf-profile.children.cycles-pp.queue_event 0.19 ± 37% +0.1 0.26 ± 16% +0.2 0.41 ± 3% perf-profile.children.cycles-pp.process_simple 0.00 +0.1 0.07 ± 19% +0.0 0.00 perf-profile.children.cycles-pp.schedule_idle 0.19 ± 37% +0.1 0.29 ± 17% +0.2 0.42 ± 4% perf-profile.children.cycles-pp.perf_session__process_events 0.19 ± 37% +0.1 0.29 ± 17% +0.2 0.42 ± 4% perf-profile.children.cycles-pp.reader__read_event 0.19 ± 37% +0.1 0.29 ± 17% +0.2 0.42 ± 4% perf-profile.children.cycles-pp.record__finish_output 2.45 ± 2% +0.1 2.56 ± 6% +1.0 3.43 perf-profile.children.cycles-pp.__schedule 0.00 +0.1 0.12 ± 19% +0.0 0.00 perf-profile.children.cycles-pp.flush_smp_call_function_queue 0.42 ± 18% +0.1 0.54 ± 23% +0.3 0.70 ± 4% perf-profile.children.cycles-pp.__cmd_record 0.00 +0.1 0.14 ± 20% +0.0 0.00 perf-profile.children.cycles-pp.sched_ttwu_pending 0.00 +0.2 0.15 ± 17% +0.0 0.00 perf-profile.children.cycles-pp.intel_idle 0.00 +0.2 0.17 ± 18% +0.0 0.00 perf-profile.children.cycles-pp.__flush_smp_call_function_queue 0.01 ±282% +0.2 0.18 ± 70% +0.1 0.06 perf-profile.children.cycles-pp.available_idle_cpu 0.23 ± 3% +0.2 0.43 ± 16% +0.1 0.31 perf-profile.children.cycles-pp.prepare_to_wait 0.00 +0.2 0.20 ± 18% +0.0 0.03 ±104% perf-profile.children.cycles-pp.cpuidle_enter 0.00 +0.2 0.20 ± 18% +0.0 0.03 ±104% perf-profile.children.cycles-pp.cpuidle_enter_state 0.01 ±282% +0.2 0.22 ± 16% +0.0 0.04 ± 81% perf-profile.children.cycles-pp.cpuidle_idle_call 3.50 ± 2% +0.2 3.73 ± 8% +0.4 3.87 ± 2% perf-profile.children.cycles-pp._raw_spin_lock 1.81 ± 3% +0.3 2.12 ± 22% +0.9 2.68 perf-profile.children.cycles-pp.autoremove_wake_function 1.80 ± 3% +0.3 2.12 ± 22% +0.9 2.66 perf-profile.children.cycles-pp.try_to_wake_up 1.89 ± 3% +0.3 2.22 ± 21% +0.9 2.80 perf-profile.children.cycles-pp.__wake_up_common 0.00 +0.3 0.35 ±110% +0.0 0.00 perf-profile.children.cycles-pp.select_idle_cpu 0.06 ± 6% +0.4 0.41 ± 95% +0.0 0.09 ± 3% perf-profile.children.cycles-pp.select_idle_sibling 0.22 ± 2% +0.4 0.57 ± 70% +0.1 0.31 perf-profile.children.cycles-pp.select_task_rq_fair 0.26 ± 5% +0.4 0.62 ± 65% +0.1 0.38 perf-profile.children.cycles-pp.select_task_rq 0.04 ± 77% +0.4 0.43 ± 18% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.start_secondary 0.04 ± 77% +0.4 0.43 ± 18% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.do_idle 0.04 ± 77% +0.4 0.43 ± 17% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.common_startup_64 0.04 ± 77% +0.4 0.43 ± 17% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.cpu_startup_entry 40.28 +0.4 40.71 +0.2 40.46 perf-profile.children.cycles-pp.vfs_write 39.04 +0.5 39.54 +0.1 39.18 perf-profile.children.cycles-pp.sock_write_iter 37.73 +0.6 38.31 +0.1 37.84 perf-profile.children.cycles-pp.unix_stream_sendmsg 2.15 ± 3% +0.7 2.82 ± 23% +1.1 3.27 perf-profile.children.cycles-pp.__wake_up_sync_key 3.55 ± 2% +0.7 4.25 ± 15% +1.1 4.65 ± 2% perf-profile.children.cycles-pp.sock_def_readable 2.90 ± 2% -0.2 2.74 ± 5% -0.2 2.69 ± 2% perf-profile.self.cycles-pp.__memcg_slab_free_hook 1.50 ± 2% -0.1 1.37 ± 5% +0.0 1.51 ± 2% perf-profile.self.cycles-pp.mod_objcg_state 1.41 ± 2% -0.1 1.30 ± 5% -0.2 1.20 ± 2% perf-profile.self.cycles-pp.kmem_cache_free 0.83 ± 3% -0.1 0.74 ± 5% -0.0 0.81 perf-profile.self.cycles-pp.read 0.88 ± 2% -0.1 0.80 ± 5% -0.0 0.88 perf-profile.self.cycles-pp.entry_SYSRETQ_unsafe_stack 0.74 ± 2% -0.1 0.67 ± 5% +0.0 0.75 ± 2% perf-profile.self.cycles-pp.write 0.70 ± 2% -0.1 0.64 ± 6% +0.0 0.73 ± 2% perf-profile.self.cycles-pp.vfs_read 0.67 ± 2% -0.1 0.61 ± 5% +0.0 0.68 perf-profile.self.cycles-pp.vfs_write 2.19 ± 2% -0.1 2.13 ± 6% -0.3 1.88 ± 2% perf-profile.self.cycles-pp.check_heap_object 0.67 ± 2% -0.1 0.62 ± 4% -0.0 0.65 perf-profile.self.cycles-pp.__kmalloc_node_track_caller_noprof 0.52 ± 3% -0.1 0.47 ± 4% -0.0 0.52 perf-profile.self.cycles-pp.obj_cgroup_charge 0.51 ± 2% -0.0 0.46 ± 5% -0.0 0.51 ± 2% perf-profile.self.cycles-pp.kmem_cache_alloc_node_noprof 0.29 ± 2% -0.0 0.25 ± 3% +0.0 0.31 perf-profile.self.cycles-pp.task_mm_cid_work 0.43 ± 2% -0.0 0.40 ± 4% +0.0 0.44 perf-profile.self.cycles-pp.do_syscall_64 0.34 ± 3% -0.0 0.31 ± 6% +0.0 0.35 ± 2% perf-profile.self.cycles-pp.__skb_datagram_iter 0.29 ± 2% -0.0 0.26 ± 6% +0.0 0.30 ± 2% perf-profile.self.cycles-pp.__virt_addr_valid 0.33 ± 2% -0.0 0.30 ± 6% +0.0 0.33 ± 2% perf-profile.self.cycles-pp.__cond_resched 0.28 ± 3% -0.0 0.25 ± 5% +0.0 0.31 ± 2% perf-profile.self.cycles-pp.syscall_return_via_sysret 0.37 ± 2% -0.0 0.34 ± 6% -0.0 0.36 perf-profile.self.cycles-pp.__check_object_size 0.18 ± 5% -0.0 0.15 ± 6% +0.1 0.27 perf-profile.self.cycles-pp.__enqueue_entity 0.70 -0.0 0.67 ± 2% -0.1 0.60 perf-profile.self.cycles-pp.get_partial_node 0.21 ± 3% -0.0 0.18 ± 6% -0.0 0.21 ± 2% perf-profile.self.cycles-pp.rcu_all_qs 0.22 ± 3% -0.0 0.20 ± 6% -0.0 0.22 ± 2% perf-profile.self.cycles-pp.x64_sys_call 0.34 ± 2% -0.0 0.32 ± 5% +0.0 0.37 perf-profile.self.cycles-pp.syscall_exit_to_user_mode 1.75 ± 2% -0.0 1.72 ± 5% -0.1 1.63 ± 2% perf-profile.self.cycles-pp.__alloc_skb 0.95 -0.0 0.93 -0.1 0.87 perf-profile.self.cycles-pp.___slab_alloc 0.42 ± 2% -0.0 0.40 ± 7% +0.0 0.45 ± 2% perf-profile.self.cycles-pp.skb_release_data 0.19 ± 2% -0.0 0.17 ± 6% -0.0 0.18 ± 2% perf-profile.self.cycles-pp.rw_verify_area 1.03 ± 3% -0.0 1.02 ± 9% +0.1 1.12 perf-profile.self.cycles-pp.skb_set_owner_w 0.03 ± 89% -0.0 0.01 ±173% +0.0 0.06 perf-profile.self.cycles-pp.avg_vruntime 0.06 ± 5% -0.0 0.04 ± 37% +0.0 0.08 ± 4% perf-profile.self.cycles-pp.cpuacct_charge 0.32 ± 2% -0.0 0.30 ± 5% +0.0 0.34 ± 2% perf-profile.self.cycles-pp.mutex_lock 0.11 ± 4% -0.0 0.10 ± 7% +0.0 0.15 ± 2% perf-profile.self.cycles-pp.__dequeue_entity 0.06 ± 8% -0.0 0.05 ± 38% +0.0 0.08 ± 6% perf-profile.self.cycles-pp.prepare_to_wait 0.18 ± 3% -0.0 0.17 ± 7% +0.1 0.24 ± 2% perf-profile.self.cycles-pp.restore_fpregs_from_fpstate 0.11 ± 4% -0.0 0.10 ± 6% +0.1 0.17 perf-profile.self.cycles-pp.update_rq_clock_task 0.29 ± 3% -0.0 0.28 ± 7% +0.1 0.40 ± 4% perf-profile.self.cycles-pp.switch_mm_irqs_off 0.12 ± 3% -0.0 0.11 ± 7% +0.0 0.15 ± 3% perf-profile.self.cycles-pp.__get_user_8 0.05 -0.0 0.04 ± 37% +0.0 0.07 ± 4% perf-profile.self.cycles-pp.rseq_update_cpu_node_id 0.01 ±282% -0.0 0.00 +0.1 0.06 perf-profile.self.cycles-pp.place_entity 0.09 ± 3% -0.0 0.08 ± 8% +0.0 0.13 ± 2% perf-profile.self.cycles-pp.unix_stream_data_wait 0.22 ± 3% -0.0 0.22 ± 7% +0.1 0.31 ± 3% perf-profile.self.cycles-pp.update_cfs_group 0.05 ± 8% -0.0 0.05 ± 40% +0.0 0.07 perf-profile.self.cycles-pp.dequeue_entities 0.10 ± 5% -0.0 0.10 ± 7% +0.0 0.15 ± 3% perf-profile.self.cycles-pp.switch_fpu_return 0.05 -0.0 0.05 ± 38% +0.0 0.06 ± 7% perf-profile.self.cycles-pp.___perf_sw_event 0.05 -0.0 0.05 ± 38% +0.0 0.07 ± 4% perf-profile.self.cycles-pp.enqueue_task_fair 0.13 ± 3% -0.0 0.12 ± 5% +0.0 0.17 ± 2% perf-profile.self.cycles-pp.__update_load_avg_se 0.10 ± 3% -0.0 0.10 ± 7% +0.0 0.14 ± 3% perf-profile.self.cycles-pp.try_to_wake_up 0.17 -0.0 0.17 -0.0 0.13 ± 2% perf-profile.self.cycles-pp.put_cpu_partial 0.06 ± 7% -0.0 0.06 ± 9% +0.0 0.09 ± 5% perf-profile.self.cycles-pp.os_xsave 0.32 -0.0 0.32 ± 2% -0.0 0.28 perf-profile.self.cycles-pp.__put_partials 0.08 ± 3% -0.0 0.08 ± 6% +0.0 0.12 ± 3% perf-profile.self.cycles-pp.update_curr_se 0.09 ± 5% -0.0 0.09 ± 7% +0.0 0.14 ± 3% perf-profile.self.cycles-pp.schedule_timeout 0.11 ± 4% -0.0 0.11 ± 6% +0.0 0.13 perf-profile.self.cycles-pp.pick_eevdf 0.06 ± 5% -0.0 0.06 ± 5% +0.0 0.08 ± 4% perf-profile.self.cycles-pp.dequeue_entity 0.06 -0.0 0.06 ± 8% +0.0 0.08 ± 5% perf-profile.self.cycles-pp.pick_task_fair 0.00 +0.0 0.00 +0.1 0.05 ± 6% perf-profile.self.cycles-pp.select_task_rq_fair 0.00 +0.0 0.00 +0.1 0.06 perf-profile.self.cycles-pp.reweight_entity 0.09 ± 3% +0.0 0.09 ± 5% +0.0 0.13 perf-profile.self.cycles-pp.finish_task_switch 0.07 ± 7% +0.0 0.07 ± 8% +0.0 0.10 perf-profile.self.cycles-pp.update_min_vruntime 0.23 ± 2% +0.0 0.23 ± 7% +0.1 0.30 perf-profile.self.cycles-pp.update_load_avg 0.07 ± 9% +0.0 0.07 ± 9% +0.0 0.10 ± 4% perf-profile.self.cycles-pp.task_h_load 0.10 ± 3% +0.0 0.10 ± 6% +0.0 0.14 ± 3% perf-profile.self.cycles-pp.__update_load_avg_cfs_rq 0.04 ± 53% +0.0 0.04 ± 38% +0.0 0.07 ± 4% perf-profile.self.cycles-pp.select_task_rq 0.00 +0.0 0.01 ±264% +0.1 0.06 ± 9% perf-profile.self.cycles-pp.native_irq_return_iret 0.12 ± 2% +0.0 0.13 ± 6% +0.0 0.16 ± 2% perf-profile.self.cycles-pp.__switch_to_asm 0.01 ±187% +0.0 0.02 ±129% +0.0 0.06 perf-profile.self.cycles-pp.perf_tp_event 0.11 ± 3% +0.0 0.12 ± 8% +0.0 0.16 ± 3% perf-profile.self.cycles-pp.__switch_to 0.08 ± 5% +0.0 0.09 ± 7% +0.0 0.12 ± 2% perf-profile.self.cycles-pp.__wake_up_common 0.24 ± 2% +0.0 0.26 ± 7% +0.1 0.35 perf-profile.self.cycles-pp.__schedule 0.12 ± 5% +0.0 0.14 ± 11% +0.1 0.18 ± 3% perf-profile.self.cycles-pp.update_curr 0.00 +0.0 0.02 ±129% +0.1 0.05 ± 9% perf-profile.self.cycles-pp.native_sched_clock 0.13 ± 3% +0.0 0.15 ± 7% +0.1 0.18 ± 2% perf-profile.self.cycles-pp.prepare_task_switch 0.05 +0.0 0.08 ± 23% +0.0 0.06 ± 6% perf-profile.self.cycles-pp.update_rq_clock 0.00 +0.0 0.04 ± 38% +0.1 0.05 ± 9% perf-profile.self.cycles-pp.enqueue_entity 0.00 +0.1 0.06 ± 11% +0.0 0.05 ± 38% perf-profile.self.cycles-pp.ring_buffer_read_head 0.18 ± 37% +0.1 0.25 ± 17% +0.2 0.40 ± 3% perf-profile.self.cycles-pp.queue_event 0.00 +0.2 0.15 ± 17% +0.0 0.00 perf-profile.self.cycles-pp.intel_idle 0.01 ±282% +0.2 0.18 ± 70% +0.0 0.05 ± 8% perf-profile.self.cycles-pp.available_idle_cpu ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-29 13:51 ` Oliver Sang @ 2024-09-30 2:04 ` Honglei Wang 2024-09-30 2:22 ` Chen Yu 0 siblings, 1 reply; 12+ messages in thread From: Honglei Wang @ 2024-09-30 2:04 UTC (permalink / raw) To: Oliver Sang, Chen Yu Cc: Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, K Prateek Nayak Hi Oliver, On 2024/9/29 21:51, Oliver Sang wrote: > hi, Chenyu and Honglei, > > On Sat, Sep 28, 2024 at 02:28:30PM +0800, Chen Yu wrote: >> Hi Honglei, >> >> On 2024-09-27 at 21:38:53 +0800, Honglei Wang wrote: >>> Hi Yu, >>> >>> Yep, I understand the preemption should happen in the same cfs level. Just >>> not sure the purpose of the 'nr_running check' stuff. Perhaps its role is >>> just to judge whether it’s necessary to do the preemption check. If there is >>> at least one more ready (cfs) task in the rq and current is not eligible, we >>> take care of the waiting tasks. Thoughts? >> >> I got your point and it makes sense. Whether the preemption check should be triggered >> seems to be a heuristic trade-off to me. I'm ok with using more aggressive preemption >> strategy as it depends on whether that workload prefers latency or throughput, and as >> long as it did not introduce regression :-) >> >> Oliver, may I know if you happen to have time for a test if the following change >> suggested by Honglei would make the regression go away? Thanks. > > I applied the patch directly upon 85e511df3cec, the test found it can reduce the > regression but not totally reovered > > ========================================================================================= > compiler/cpufreq_governor/ipc/iterations/kconfig/mode/nr_threads/rootfs/tbox_group/testcase: > gcc-12/performance/socket/4/x86_64-rhel-8.3/process/50%/debian-12-x86_64-20240206.cgz/lkp-spr-r02/hackbench > > commit: > 82e9d0456e06 ("sched/fair: Avoid re-setting virtual deadline on 'migrations'") > 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > 8079496d311b <--- patch from Honglei > > 82e9d0456e06cebe 85e511df3cec46021024176672a 8079496d311b6b0d4dae973f4df > ---------------- --------------------------- --------------------------- > %stddev %change %stddev %change %stddev > \ | \ | \ > 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput > Thanks a lot for running the test. The result is as expectation, as the strategy of short slices tends to favor more frequent scheduling to help delay-sensitive tasks acquire CPU as early as possible. I suspect that the current test environment does not have any special configurations for slices. In this case, a 3.2% regression is still somewhat significant. As Yu mentioned, this is a heuristic adjustment. In this particular case, it seems that Yu's patch is more effective in solving the problem. Let's delegate the preemption opportunity to the higher-level update_curr() function. Thanks, Honglei >> >> thanks, >> Chenyu >> >> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c >> index fd2f3831c74e..290e5fdfc267 100644 >> --- a/kernel/sched/fair.c >> +++ b/kernel/sched/fair.c >> @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) >> >> account_cfs_rq_runtime(cfs_rq, delta_exec); >> >> - if (rq->nr_running == 1) >> + if (rq->cfs.nr_running == 1) >> return; >> >> if (resched || did_preempt_short(cfs_rq, curr)) { >> -- >> 2.25.1 > > > below is full comparison FYI. > > > ========================================================================================= > compiler/cpufreq_governor/ipc/iterations/kconfig/mode/nr_threads/rootfs/tbox_group/testcase: > gcc-12/performance/socket/4/x86_64-rhel-8.3/process/50%/debian-12-x86_64-20240206.cgz/lkp-spr-r02/hackbench > > commit: > 82e9d0456e06 ("sched/fair: Avoid re-setting virtual deadline on 'migrations'") > 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > 8079496d311b <--- patch from Honglei > > 82e9d0456e06cebe 85e511df3cec46021024176672a 8079496d311b6b0d4dae973f4df > ---------------- --------------------------- --------------------------- > %stddev %change %stddev %change %stddev > \ | \ | \ > 150791 ± 3% -3.9% 144868 ± 2% -8.4% 138189 ± 3% perf-c2c.HITM.local > 217.40 +13.5% 246.74 +1.9% 221.55 uptime.boot > 4679401 -15.8% 3938145 +35.5% 6342833 vmstat.system.cs > 854648 -15.2% 724774 +35.0% 1153448 vmstat.system.in > 0.46 ± 2% -0.1 0.40 +0.1 0.51 ± 2% mpstat.cpu.all.irq% > 0.03 ± 3% -0.0 0.03 +0.0 0.04 mpstat.cpu.all.soft% > 3.35 -0.6 2.75 +0.3 3.65 mpstat.cpu.all.usr% > 922628 ± 16% +17.2% 1080933 ± 14% +87.9% 1733278 ± 14% numa-numastat.node0.local_node > 1027104 ± 13% +17.6% 1207529 ± 15% +82.3% 1872582 ± 14% numa-numastat.node0.numa_hit > 2553206 ± 16% +0.4% 2563309 ± 10% +46.3% 3735450 ± 4% numa-numastat.node1.local_node > 2684151 ± 16% -0.4% 2672834 ± 8% +43.2% 3843864 ± 5% numa-numastat.node1.numa_hit > 174.58 +16.3% 203.09 +2.1% 178.28 time.elapsed_time > 174.58 +16.3% 203.09 +2.1% 178.28 time.elapsed_time.max > 1.654e+08 +2.2% 1.69e+08 +50.7% 2.492e+08 time.involuntary_context_switches > 36869 +17.6% 43340 +1.6% 37461 time.system_time > 1172 -5.5% 1107 +8.9% 1276 time.user_time > 6.478e+08 -3.5% 6.255e+08 +36.2% 8.824e+08 ± 2% time.voluntary_context_switches > 8177075 ± 14% +8.6% 8880848 ± 13% +47.8% 12086160 ± 2% meminfo.Cached > 7771720 ± 15% +9.5% 8506289 ± 13% +50.8% 11720513 ± 2% meminfo.Committed_AS > 4992416 ± 31% +17.9% 5883734 ± 19% +83.6% 9165655 ± 2% meminfo.Inactive > 4992203 ± 31% +17.9% 5883521 ± 19% +83.6% 9165427 ± 2% meminfo.Inactive(anon) > 2281337 ± 19% +17.2% 2673418 ± 4% +38.0% 3147813 ± 3% meminfo.Mapped > 12824178 ± 10% +5.9% 13577581 ± 8% +30.4% 16717964 meminfo.Memused > 5084291 ± 22% +13.8% 5788029 ± 20% +76.9% 8993338 ± 2% meminfo.Shmem > 13135346 ± 9% +5.2% 13824666 ± 8% +29.2% 16968485 meminfo.max_used_kB > 3089827 ± 31% +15.4% 3565461 ± 7% +49.8% 4628448 ± 10% numa-meminfo.node0.FilePages > 5391461 ± 19% +16.5% 6281524 ± 6% +31.1% 7067023 ± 10% numa-meminfo.node0.MemUsed > 352581 ± 13% +24.6% 439472 ± 16% +11.3% 392488 ± 14% numa-meminfo.node0.SUnreclaim > 5090008 ± 22% +4.5% 5317103 ± 18% +46.6% 7462041 ± 3% numa-meminfo.node1.FilePages > 4237429 ± 32% +13.8% 4822257 ± 21% +66.3% 7045772 ± 7% numa-meminfo.node1.Inactive > 4237311 ± 32% +13.8% 4822122 ± 21% +66.3% 7045629 ± 7% numa-meminfo.node1.Inactive(anon) > 1875959 ± 20% +18.1% 2215441 ± 5% +44.6% 2711735 ± 3% numa-meminfo.node1.Mapped > 7434746 ± 16% -1.8% 7298189 ± 12% +29.9% 9655790 ± 5% numa-meminfo.node1.MemUsed > 4692017 ± 22% +10.5% 5186768 ± 18% +55.8% 7310040 ± 4% numa-meminfo.node1.Shmem > 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput > 606251 -14.1% 520789 -2.1% 593622 hackbench.throughput_avg > 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput_best > 580034 -14.8% 494354 +0.4% 582377 hackbench.throughput_worst > 174.58 +16.3% 203.09 +2.1% 178.28 hackbench.time.elapsed_time > 174.58 +16.3% 203.09 +2.1% 178.28 hackbench.time.elapsed_time.max > 1.654e+08 +2.2% 1.69e+08 +50.7% 2.492e+08 hackbench.time.involuntary_context_switches > 36869 +17.6% 43340 +1.6% 37461 hackbench.time.system_time > 1172 -5.5% 1107 +8.9% 1276 hackbench.time.user_time > 6.478e+08 -3.5% 6.255e+08 +36.2% 8.824e+08 ± 2% hackbench.time.voluntary_context_switches > 772255 ± 31% +15.4% 891384 ± 7% +49.8% 1156899 ± 10% numa-vmstat.node0.nr_file_pages > 1025412 ± 13% +17.7% 1206842 ± 15% +82.6% 1871921 ± 14% numa-vmstat.node0.numa_hit > 920942 ± 16% +17.3% 1080247 ± 14% +88.1% 1732619 ± 14% numa-vmstat.node0.numa_local > 1271682 ± 22% +4.5% 1328902 ± 18% +46.6% 1863679 ± 3% numa-vmstat.node1.nr_file_pages > 1059760 ± 32% +13.7% 1205295 ± 21% +66.1% 1759754 ± 7% numa-vmstat.node1.nr_inactive_anon > 469078 ± 20% +18.0% 553505 ± 5% +44.2% 676623 ± 3% numa-vmstat.node1.nr_mapped > 1172184 ± 22% +10.6% 1296319 ± 18% +55.8% 1825678 ± 4% numa-vmstat.node1.nr_shmem > 1059759 ± 32% +13.7% 1205294 ± 21% +66.1% 1759752 ± 7% numa-vmstat.node1.nr_zone_inactive_anon > 2682254 ± 16% -0.4% 2671316 ± 8% +43.3% 3842866 ± 5% numa-vmstat.node1.numa_hit > 2551311 ± 16% +0.4% 2561794 ± 10% +46.4% 3734454 ± 4% numa-vmstat.node1.numa_local > 6233000 -0.3% 6214123 -1.6% 6135664 proc-vmstat.nr_dirty_background_threshold > 12481241 -0.3% 12443440 -1.6% 12286330 proc-vmstat.nr_dirty_threshold > 2044616 ± 14% +8.6% 2220456 ± 13% +47.8% 3022899 ± 2% proc-vmstat.nr_file_pages > 62706078 -0.3% 62517035 -1.6% 61731279 proc-vmstat.nr_free_pages > 1247557 ± 31% +17.9% 1471046 ± 19% +83.8% 2292396 ± 2% proc-vmstat.nr_inactive_anon > 570709 ± 19% +17.2% 668608 ± 4% +38.0% 787666 ± 3% proc-vmstat.nr_mapped > 1271419 ± 23% +13.8% 1447250 ± 20% +76.9% 2249692 ± 2% proc-vmstat.nr_shmem > 44542 +2.7% 45755 +5.1% 46796 proc-vmstat.nr_slab_reclaimable > 188881 -0.7% 187493 +1.4% 191440 proc-vmstat.nr_slab_unreclaimable > 1247557 ± 31% +17.9% 1471046 ± 19% +83.8% 2292396 ± 2% proc-vmstat.nr_zone_inactive_anon > 3714588 ± 9% +6.5% 3954189 ± 12% +54.0% 5719994 proc-vmstat.numa_hit > 3479163 ± 10% +4.8% 3645729 ± 9% +57.3% 5472276 proc-vmstat.numa_local > 642130 ± 68% -71.2% 184909 ± 11% -73.9% 167609 ± 14% proc-vmstat.pgactivate > 4593887 ± 7% +1.2% 4650141 ± 7% +131.4% 10629624 proc-vmstat.pgalloc_normal > 2170433 ± 2% +6.8% 2318318 ± 2% +6.4% 2308275 ± 5% proc-vmstat.pgfault > 2631626 ± 7% -8.8% 2400780 +177.3% 7298788 proc-vmstat.pgfree > 138302 ± 4% +6.7% 147631 ± 3% -0.7% 137382 ± 4% proc-vmstat.pgreuse > 0.30 ± 3% -4.9% 0.28 ± 2% +20.7% 0.36 perf-stat.i.MPKI > 6.354e+10 -11.4% 5.63e+10 +2.2% 6.497e+10 perf-stat.i.branch-instructions > 0.53 -0.0 0.52 +0.0 0.57 perf-stat.i.branch-miss-rate% > 3.226e+08 -12.5% 2.822e+08 +9.2% 3.523e+08 perf-stat.i.branch-misses > 3.91 ± 3% -0.2 3.74 ± 2% +0.9 4.80 ± 2% perf-stat.i.cache-miss-rate% > 94557935 ± 3% -15.2% 80197744 ± 2% +25.1% 1.183e+08 perf-stat.i.cache-misses > 2.563e+09 -13.7% 2.212e+09 +1.3% 2.596e+09 perf-stat.i.cache-references > 4710895 -15.9% 3959720 +36.7% 6439647 perf-stat.i.context-switches > 1.86 +14.3% 2.13 -0.8% 1.85 perf-stat.i.cpi > 224291 +0.0% 224353 +1.4% 227521 perf-stat.i.cpu-clock > 6.345e+11 +0.4% 6.368e+11 +1.3% 6.424e+11 perf-stat.i.cpu-cycles > 601598 -15.0% 511540 +13.7% 683759 ± 2% perf-stat.i.cpu-migrations > 7390 ± 5% +28.4% 9492 ± 2% -23.5% 5655 perf-stat.i.cycles-between-cache-misses > 3.408e+11 -12.1% 2.997e+11 +2.1% 3.478e+11 perf-stat.i.instructions > 0.54 -12.2% 0.47 +0.8% 0.55 perf-stat.i.ipc > 23.73 -15.9% 19.95 +32.3% 31.39 perf-stat.i.metric.K/sec > 224291 +0.0% 224353 +1.4% 227521 perf-stat.i.task-clock > 0.25 ± 35% +8.1% 0.27 ± 2% +38.0% 0.34 perf-stat.overall.MPKI > 0.45 ± 35% +0.1 0.50 +0.1 0.54 perf-stat.overall.branch-miss-rate% > 3.26 ± 35% +0.3 3.60 +1.3 4.54 perf-stat.overall.cache-miss-rate% > 1.66 ± 35% +28.5% 2.13 +11.6% 1.85 perf-stat.overall.cpi > 6006 ± 35% +33.5% 8020 ± 2% -9.1% 5458 perf-stat.overall.cycles-between-cache-misses > 0.48 ± 35% -1.5% 0.47 +13.4% 0.54 perf-stat.overall.ipc > 2.847e+08 ± 35% -1.6% 2.802e+08 +21.1% 3.448e+08 perf-stat.ps.branch-misses > 83193631 ± 35% -5.0% 79040711 ± 2% +38.7% 1.154e+08 perf-stat.ps.cache-misses > 4153465 ± 35% -5.4% 3930147 +52.2% 6323104 perf-stat.ps.context-switches > 531469 ± 35% -4.7% 506531 +26.2% 670740 ± 2% perf-stat.ps.cpu-migrations > 5.287e+13 ± 35% +15.1% 6.083e+13 +15.6% 6.114e+13 perf-stat.total.instructions > 13829361 +51.8% 20989754 +6.0% 14655040 ± 15% sched_debug.cfs_rq:/.avg_vruntime.avg > 18756074 ± 5% +44.2% 27055241 ± 3% +3.8% 19470555 ± 13% sched_debug.cfs_rq:/.avg_vruntime.max > 12499623 ± 2% +52.4% 19043277 ± 2% +8.0% 13493604 ± 14% sched_debug.cfs_rq:/.avg_vruntime.min > 8.93 ± 2% +14.1% 10.19 -5.2% 8.47 ± 7% sched_debug.cfs_rq:/.h_nr_running.avg > 4.68 ± 3% +10.0% 5.15 ± 2% -1.5% 4.62 ± 8% sched_debug.cfs_rq:/.h_nr_running.stddev > 0.44 ± 35% +75.8% 0.78 ± 19% -1.6% 0.44 ± 50% sched_debug.cfs_rq:/.load_avg.min > 13829361 +51.8% 20989754 +6.0% 14655040 ± 15% sched_debug.cfs_rq:/.min_vruntime.avg > 18756074 ± 5% +44.2% 27055241 ± 3% +3.8% 19470555 ± 13% sched_debug.cfs_rq:/.min_vruntime.max > 12499623 ± 2% +52.4% 19043277 ± 2% +8.0% 13493604 ± 14% sched_debug.cfs_rq:/.min_vruntime.min > 0.68 +11.7% 0.76 -1.3% 0.67 ± 2% sched_debug.cfs_rq:/.nr_running.avg > 176.30 ± 3% -22.8% 136.16 ± 4% -2.9% 171.17 ± 10% sched_debug.cfs_rq:/.removed.runnable_avg.max > 176.30 ± 3% -22.8% 136.16 ± 4% -2.9% 171.17 ± 10% sched_debug.cfs_rq:/.removed.util_avg.max > 8995 +16.0% 10437 -5.7% 8484 ± 7% sched_debug.cfs_rq:/.runnable_avg.avg > 18978 ± 6% +13.7% 21579 ± 6% -10.1% 17066 ± 11% sched_debug.cfs_rq:/.runnable_avg.max > 2890 ± 4% +13.9% 3292 ± 3% -1.5% 2848 ± 9% sched_debug.cfs_rq:/.runnable_avg.stddev > 1094 ± 3% +2.9% 1125 ± 3% +72.4% 1886 ± 5% sched_debug.cfs_rq:/.util_est.avg > 4561 ± 14% +19.8% 5464 ± 19% +50.2% 6850 ± 7% sched_debug.cfs_rq:/.util_est.max > 816.44 ± 6% +7.4% 877.16 ± 4% +78.4% 1456 ± 4% sched_debug.cfs_rq:/.util_est.stddev > 415209 ± 22% -23.3% 318311 ± 3% -8.8% 378602 ± 6% sched_debug.cpu.avg_idle.avg > 102333 ± 2% +30.5% 133496 ± 2% +4.0% 106415 ± 9% sched_debug.cpu.clock.avg > 102519 ± 2% +30.4% 133722 ± 2% +4.0% 106590 ± 8% sched_debug.cpu.clock.max > 102127 ± 2% +30.5% 133254 ± 2% +4.0% 106199 ± 9% sched_debug.cpu.clock.min > 101839 ± 2% +30.5% 132880 ± 2% +4.0% 105862 ± 8% sched_debug.cpu.clock_task.avg > 102169 ± 2% +30.4% 133268 ± 2% +3.9% 106186 ± 8% sched_debug.cpu.clock_task.max > 87129 ± 2% +35.6% 118117 ± 2% +4.6% 91152 ± 10% sched_debug.cpu.clock_task.min > 11573 +32.4% 15327 -0.3% 11534 sched_debug.cpu.curr->pid.avg > 14704 +23.9% 18214 +3.2% 15178 ± 7% sched_debug.cpu.curr->pid.max > 1516 ± 9% +16.4% 1765 ± 10% +33.3% 2021 ± 44% sched_debug.cpu.curr->pid.stddev > 8.92 ± 2% +14.1% 10.18 -5.2% 8.46 ± 7% sched_debug.cpu.nr_running.avg > 4.69 ± 2% +10.0% 5.16 ± 2% -1.7% 4.61 ± 8% sched_debug.cpu.nr_running.stddev > 1232815 ± 2% +27.3% 1569099 +45.1% 1789014 ± 17% sched_debug.cpu.nr_switches.avg > 1411362 ± 5% +26.8% 1789325 ± 3% +48.6% 2097828 ± 20% sched_debug.cpu.nr_switches.max > 1045767 ± 2% +27.3% 1331341 ± 3% +42.5% 1489877 ± 17% sched_debug.cpu.nr_switches.min > 102127 ± 2% +30.5% 133250 ± 2% +4.0% 106198 ± 9% sched_debug.cpu_clk > 101071 ± 2% +30.8% 132194 ± 2% +4.0% 105143 ± 9% sched_debug.ktime > 0.00 -25.0% 0.00 -3.1% 0.00 ± 8% sched_debug.rt_rq:.rt_nr_running.avg > 0.33 -25.0% 0.25 -3.1% 0.32 ± 8% sched_debug.rt_rq:.rt_nr_running.max > 0.02 -25.0% 0.02 -3.1% 0.02 ± 8% sched_debug.rt_rq:.rt_nr_running.stddev > 102997 ± 2% +30.2% 134142 ± 2% +4.0% 107069 ± 9% sched_debug.sched_clk > 16347631 +100.0% 32695263 +100.0% 32695263 sched_debug.sysctl_sched.sysctl_sched_features > 3.31 ±123% -42.6% 1.90 ± 86% -89.1% 0.36 ± 88% perf-sched.sch_delay.avg.ms.__cond_resched.__alloc_pages_noprof.alloc_pages_mpol_noprof.folio_alloc_mpol_noprof.shmem_alloc_folio > 8.46 ± 38% +100.8% 16.98 ± 82% -85.1% 1.26 ±134% perf-sched.sch_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof > 5.72 ± 38% +21.2% 6.93 ± 80% -85.6% 0.83 ± 85% perf-sched.sch_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc > 7.88 ± 21% +11.3% 8.77 ± 59% -89.0% 0.87 ± 38% perf-sched.sch_delay.avg.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags > 7.16 ± 36% -24.4% 5.42 ± 73% -70.7% 2.10 ± 66% perf-sched.sch_delay.avg.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity > 28.76 ±142% -62.8% 10.70 ±198% -99.4% 0.18 ±233% perf-sched.sch_delay.avg.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate > 3.75 ± 24% +0.6% 3.77 ± 59% -62.1% 1.42 ± 39% perf-sched.sch_delay.avg.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter > 10.26 ± 19% -9.5% 9.29 ± 62% -74.3% 2.64 ± 40% perf-sched.sch_delay.avg.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write > 2.82 ± 43% -22.6% 2.18 ± 73% -83.9% 0.45 ± 59% perf-sched.sch_delay.avg.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write > 2.30 ± 29% -4.8% 2.19 ± 61% -72.3% 0.64 ± 38% perf-sched.sch_delay.avg.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb > 1.75 ± 49% -5.1% 1.66 ± 65% -61.1% 0.68 ± 39% perf-sched.sch_delay.avg.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg > 4.13 ± 87% +0.3% 4.15 ± 88% -86.0% 0.58 ±133% perf-sched.sch_delay.avg.ms.__cond_resched.shmem_inode_acct_blocks.shmem_alloc_and_add_folio.shmem_get_folio_gfp.shmem_write_begin > 2.81 ± 59% -20.7% 2.23 ± 61% -72.8% 0.76 ± 38% perf-sched.sch_delay.avg.ms.__cond_resched.task_work_run.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe > 9.13 ± 30% +0.5% 9.17 ± 67% -87.4% 1.15 ± 53% perf-sched.sch_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] > 9.94 ± 32% -38.3% 6.13 ± 62% -56.9% 4.28 ± 37% perf-sched.sch_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_reschedule_ipi.[unknown].[unknown] > 11.06 ± 57% -45.0% 6.08 ± 75% -90.4% 1.06 ± 60% perf-sched.sch_delay.avg.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread > 16.36 ± 24% -23.9% 12.45 ± 59% -53.7% 7.57 ± 38% perf-sched.sch_delay.avg.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter > 1.77 ± 23% -28.5% 1.26 ± 60% -66.1% 0.60 ± 38% perf-sched.sch_delay.avg.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg > 2.66 ± 21% -15.5% 2.25 ± 58% -63.8% 0.96 ± 37% perf-sched.sch_delay.avg.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] > 226.74 ±108% -40.0% 136.06 ±120% -95.1% 11.09 ± 91% perf-sched.sch_delay.avg.ms.worker_thread.kthread.ret_from_fork.ret_from_fork_asm > 1412 ± 25% -7.0% 1313 ± 66% -83.6% 231.23 ± 47% perf-sched.sch_delay.max.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags > 1221 ± 4% -23.0% 940.87 ± 58% -51.1% 596.88 ± 89% perf-sched.sch_delay.max.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity > 262.66 ±176% -72.1% 73.22 ±239% -99.9% 0.35 ±248% perf-sched.sch_delay.max.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate > 779.32 ± 29% -15.9% 655.34 ± 72% -80.4% 152.55 ± 45% perf-sched.sch_delay.max.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter > 1042 ± 23% -8.4% 954.56 ± 63% -81.8% 190.25 ± 44% perf-sched.sch_delay.max.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write > 463.55 ± 83% -3.7% 446.18 ± 87% -85.8% 65.73 ± 93% perf-sched.sch_delay.max.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write > 1128 ± 17% -17.7% 928.10 ± 62% -81.0% 214.51 ± 40% perf-sched.sch_delay.max.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb > 500.79 ± 50% -25.9% 370.90 ± 79% -76.0% 120.40 ± 46% perf-sched.sch_delay.max.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg > 744.38 ± 27% +0.4% 747.54 ± 78% -86.3% 102.18 ± 71% perf-sched.sch_delay.max.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] > 579.24 ± 43% -10.1% 520.76 ± 62% -65.3% 201.13 ± 62% perf-sched.sch_delay.max.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread > 1945 ± 16% -15.9% 1635 ± 61% -73.4% 516.65 ± 40% perf-sched.sch_delay.max.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter > 1530 ± 22% -7.2% 1420 ± 63% -75.5% 375.37 ± 40% perf-sched.sch_delay.max.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg > 3.03 ± 23% -26.5% 2.23 ± 59% -65.2% 1.05 ± 38% perf-sched.total_sch_delay.average.ms > 8.72 ± 23% -28.5% 6.23 ± 59% -64.2% 3.12 ± 38% perf-sched.total_wait_and_delay.average.ms > 8284 ± 27% -38.2% 5118 ± 60% -41.6% 4839 ± 44% perf-sched.total_wait_and_delay.max.ms > 5.68 ± 23% -29.6% 4.00 ± 59% -63.7% 2.06 ± 38% perf-sched.total_wait_time.average.ms > 5811 ± 8% -28.5% 4155 ± 59% -27.8% 4194 ± 38% perf-sched.total_wait_time.max.ms > 37.25 ± 91% +17.8% 43.88 ± 72% -90.7% 3.45 ±136% perf-sched.wait_and_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof > 44.86 ± 74% +10.0% 49.33 ±117% -95.6% 1.99 ±132% perf-sched.wait_and_delay.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc > 22.30 ± 21% +6.1% 23.66 ± 59% -100.0% 0.00 perf-sched.wait_and_delay.avg.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags > 17.29 ± 21% -28.7% 12.33 ± 66% -60.1% 6.90 ± 45% perf-sched.wait_and_delay.avg.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity > 28.98 ± 21% -13.8% 24.98 ± 61% -74.3% 7.45 ± 39% perf-sched.wait_and_delay.avg.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write > 27.08 ± 28% -10.4% 24.27 ± 64% -93.8% 1.69 ±129% perf-sched.wait_and_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] > 27.52 ± 28% -37.8% 17.11 ± 61% -54.9% 12.40 ± 37% perf-sched.wait_and_delay.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_reschedule_ipi.[unknown].[unknown] > 38.19 ± 43% -40.4% 22.76 ± 65% -80.4% 7.50 ± 40% perf-sched.wait_and_delay.avg.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread > 45.63 ± 24% -25.8% 33.84 ± 59% -53.4% 21.28 ± 38% perf-sched.wait_and_delay.avg.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter > 5.06 ± 24% -30.6% 3.51 ± 60% -65.6% 1.74 ± 38% perf-sched.wait_and_delay.avg.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg > 7.23 ± 22% -18.4% 5.90 ± 59% -64.2% 2.59 ± 38% perf-sched.wait_and_delay.avg.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] > 17219 ± 26% +18.6% 20422 ± 60% -100.0% 0.00 perf-sched.wait_and_delay.count.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags > 287.02 ± 72% -56.9% 123.78 ± 90% -90.0% 28.73 ±130% perf-sched.wait_and_delay.max.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc > 2851 ± 25% -5.3% 2699 ± 64% -100.0% 0.00 perf-sched.wait_and_delay.max.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags > 2443 ± 4% -23.0% 1881 ± 58% -37.8% 1519 ± 53% perf-sched.wait_and_delay.max.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity > 2132 ± 23% -9.7% 1925 ± 63% -82.0% 384.48 ± 45% perf-sched.wait_and_delay.max.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write > 3449 ± 45% -35.6% 2220 ± 71% -59.1% 1410 ± 49% perf-sched.wait_and_delay.max.ms.do_wait.kernel_wait4.__do_sys_wait4.do_syscall_64 > 1507 ± 27% +0.2% 1510 ± 79% -93.3% 101.36 ±137% perf-sched.wait_and_delay.max.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] > 1312 ± 35% -16.4% 1097 ± 61% -65.2% 456.55 ± 64% perf-sched.wait_and_delay.max.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread > 3951 ± 16% -17.0% 3278 ± 61% -73.7% 1038 ± 40% perf-sched.wait_and_delay.max.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter > 3088 ± 22% -7.9% 2844 ± 62% -75.1% 768.96 ± 41% perf-sched.wait_and_delay.max.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg > 3.31 ±123% -42.6% 1.90 ± 86% -89.1% 0.36 ± 88% perf-sched.wait_time.avg.ms.__cond_resched.__alloc_pages_noprof.alloc_pages_mpol_noprof.folio_alloc_mpol_noprof.shmem_alloc_folio > 28.79 ±114% -6.6% 26.89 ± 70% -90.2% 2.82 ± 96% perf-sched.wait_time.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof > 39.14 ± 89% +8.3% 42.40 ±140% -95.1% 1.92 ± 78% perf-sched.wait_time.avg.ms.__cond_resched.__kmalloc_node_noprof.alloc_slab_obj_exts.allocate_slab.___slab_alloc > 14.42 ± 21% +3.3% 14.89 ± 59% -88.8% 1.61 ± 37% perf-sched.wait_time.avg.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags > 10.13 ± 17% -31.7% 6.92 ± 63% -52.6% 4.80 ± 41% perf-sched.wait_time.avg.ms.__cond_resched.__wait_for_common.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity > 40.86 ±110% -46.6% 21.80 ±152% -99.1% 0.35 ±248% perf-sched.wait_time.avg.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate > 6.56 ± 23% -5.6% 6.20 ± 59% -61.9% 2.50 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter > 18.72 ± 22% -16.2% 15.69 ± 61% -74.3% 4.81 ± 39% perf-sched.wait_time.avg.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write > 2.82 ± 43% -22.6% 2.18 ± 73% -83.9% 0.45 ± 59% perf-sched.wait_time.avg.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write > 5.03 ± 24% -6.1% 4.72 ± 59% -63.4% 1.84 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb > 3.16 ± 33% -19.7% 2.54 ± 65% -60.4% 1.25 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg > 4.13 ± 87% +0.3% 4.15 ± 88% -86.0% 0.58 ±133% perf-sched.wait_time.avg.ms.__cond_resched.shmem_inode_acct_blocks.shmem_alloc_and_add_folio.shmem_get_folio_gfp.shmem_write_begin > 4.42 ± 44% -21.0% 3.49 ± 65% -71.3% 1.27 ± 38% perf-sched.wait_time.avg.ms.__cond_resched.task_work_run.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe > 224.55 ± 53% -20.5% 178.59 ± 71% -71.8% 63.25 ± 68% perf-sched.wait_time.avg.ms.do_wait.kernel_wait4.__do_sys_wait4.do_syscall_64 > 17.95 ± 30% -15.9% 15.10 ± 63% -89.3% 1.91 ± 46% perf-sched.wait_time.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] > 17.58 ± 26% -37.5% 10.98 ± 61% -53.8% 8.12 ± 38% perf-sched.wait_time.avg.ms.irqentry_exit_to_user_mode.asm_sysvec_reschedule_ipi.[unknown].[unknown] > 27.13 ± 39% -38.5% 16.68 ± 63% -76.3% 6.43 ± 38% perf-sched.wait_time.avg.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread > 29.27 ± 24% -26.9% 21.39 ± 59% -53.2% 13.71 ± 38% perf-sched.wait_time.avg.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter > 3.29 ± 24% -31.7% 2.25 ± 60% -65.3% 1.14 ± 38% perf-sched.wait_time.avg.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg > 4.57 ± 22% -20.1% 3.66 ± 59% -64.4% 1.63 ± 38% perf-sched.wait_time.avg.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] > 1446 ± 25% -3.4% 1396 ± 63% -80.6% 280.50 ± 42% perf-sched.wait_time.max.ms.__cond_resched.__kmalloc_node_track_caller_noprof.kmalloc_reserve.__alloc_skb.alloc_skb_with_frags > 292.35 ±159% -68.1% 93.21 ±217% -99.8% 0.69 ±256% perf-sched.wait_time.max.ms.__cond_resched.__wait_for_common.stop_two_cpus.migrate_swap.task_numa_migrate > 932.87 ± 20% -20.6% 740.44 ± 65% -81.7% 170.42 ± 45% perf-sched.wait_time.max.ms.__cond_resched.aa_sk_perm.security_socket_recvmsg.sock_recvmsg.sock_read_iter > 1110 ± 24% -11.8% 979.30 ± 62% -81.7% 203.13 ± 43% perf-sched.wait_time.max.ms.__cond_resched.aa_sk_perm.security_socket_sendmsg.sock_write_iter.vfs_write > 463.55 ± 83% -3.7% 446.18 ± 87% -85.8% 65.73 ± 93% perf-sched.wait_time.max.ms.__cond_resched.generic_perform_write.shmem_file_write_iter.vfs_write.ksys_write > 1203 ± 20% -19.5% 969.24 ± 62% -80.5% 234.88 ± 42% perf-sched.wait_time.max.ms.__cond_resched.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb > 614.81 ± 41% -27.8% 443.79 ± 79% -79.3% 127.43 ± 46% perf-sched.wait_time.max.ms.__cond_resched.mutex_lock.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg > 822.39 ± 32% -24.1% 624.54 ± 74% -82.9% 140.70 ± 43% perf-sched.wait_time.max.ms.__cond_resched.task_work_run.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe > 817.75 ± 28% -6.8% 762.54 ± 79% -85.7% 117.26 ± 58% perf-sched.wait_time.max.ms.irqentry_exit_to_user_mode.asm_sysvec_apic_timer_interrupt.[unknown].[unknown] > 774.10 ± 27% -22.5% 600.11 ± 61% -63.0% 286.19 ± 57% perf-sched.wait_time.max.ms.schedule_timeout.rcu_gp_fqs_loop.rcu_gp_kthread.kthread > 2023 ± 16% -18.8% 1643 ± 60% -74.2% 521.59 ± 40% perf-sched.wait_time.max.ms.schedule_timeout.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter > 1624 ± 25% -12.1% 1428 ± 62% -75.0% 405.54 ± 42% perf-sched.wait_time.max.ms.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg > 2801 ± 45% -40.4% 1669 ± 81% -54.5% 1273 ± 52% perf-sched.wait_time.max.ms.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.[unknown] > 55.19 -0.7 54.46 -0.6 54.55 perf-profile.calltrace.cycles-pp.read > 53.91 -0.6 53.30 -0.8 53.13 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.read > 53.76 -0.6 53.16 -0.8 52.98 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.read > 52.41 -0.5 51.92 -1.1 51.32 perf-profile.calltrace.cycles-pp.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe.read > 52.02 -0.5 51.56 -1.1 50.92 perf-profile.calltrace.cycles-pp.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe.read > 50.82 -0.4 50.46 -1.2 49.67 perf-profile.calltrace.cycles-pp.sock_read_iter.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe > 50.29 -0.3 49.98 -1.2 49.12 perf-profile.calltrace.cycles-pp.sock_recvmsg.sock_read_iter.vfs_read.ksys_read.do_syscall_64 > 49.61 -0.3 49.36 -1.2 48.42 perf-profile.calltrace.cycles-pp.unix_stream_recvmsg.sock_recvmsg.sock_read_iter.vfs_read.ksys_read > 49.37 -0.2 49.13 -1.2 48.17 perf-profile.calltrace.cycles-pp.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg.sock_read_iter.vfs_read > 24.50 -0.2 24.29 ± 4% -1.1 23.39 perf-profile.calltrace.cycles-pp.alloc_skb_with_frags.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter.vfs_write > 5.54 ± 2% -0.2 5.34 ± 6% -0.3 5.25 ± 2% perf-profile.calltrace.cycles-pp.unix_stream_read_actor.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg.sock_read_iter > 5.49 ± 2% -0.2 5.29 ± 6% -0.3 5.20 ± 2% perf-profile.calltrace.cycles-pp.skb_copy_datagram_iter.unix_stream_read_actor.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg > 5.40 ± 2% -0.2 5.21 ± 6% -0.3 5.11 ± 2% perf-profile.calltrace.cycles-pp.__skb_datagram_iter.skb_copy_datagram_iter.unix_stream_read_actor.unix_stream_read_generic.unix_stream_recvmsg > 24.31 -0.2 24.12 ± 4% -1.1 23.20 perf-profile.calltrace.cycles-pp.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter > 2.86 ± 2% -0.2 2.68 ± 5% -0.2 2.65 ± 2% perf-profile.calltrace.cycles-pp.__memcg_slab_free_hook.kmem_cache_free.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg > 1.60 ± 2% -0.1 1.45 ± 5% -0.0 1.57 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.read > 1.50 ± 2% -0.1 1.36 ± 6% -0.0 1.50 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.write > 0.62 ± 2% -0.1 0.50 ± 38% +0.3 0.95 perf-profile.calltrace.cycles-pp.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function.__wake_up_common > 2.76 ± 2% -0.1 2.66 ± 6% -0.3 2.45 ± 2% perf-profile.calltrace.cycles-pp.simple_copy_to_iter.__skb_datagram_iter.skb_copy_datagram_iter.unix_stream_read_actor.unix_stream_read_generic > 2.65 ± 2% -0.1 2.55 ± 6% -0.3 2.34 ± 2% perf-profile.calltrace.cycles-pp.__check_object_size.simple_copy_to_iter.__skb_datagram_iter.skb_copy_datagram_iter.unix_stream_read_actor > 0.97 ± 2% -0.1 0.89 ± 5% +0.3 1.26 perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.read > 0.78 ± 2% -0.1 0.71 ± 7% +0.4 1.18 perf-profile.calltrace.cycles-pp.ttwu_do_activate.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_sync_key > 1.24 ± 2% -0.1 1.16 ± 5% +0.1 1.31 ± 2% perf-profile.calltrace.cycles-pp.__memcg_slab_post_alloc_hook.kmem_cache_alloc_node_noprof.__alloc_skb.alloc_skb_with_frags.sock_alloc_send_pskb > 2.20 ± 2% -0.1 2.14 ± 6% -0.3 1.90 ± 2% perf-profile.calltrace.cycles-pp.check_heap_object.__check_object_size.simple_copy_to_iter.__skb_datagram_iter.skb_copy_datagram_iter > 26.13 -0.0 26.09 ± 3% -1.0 25.11 perf-profile.calltrace.cycles-pp.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter.vfs_write.ksys_write > 0.53 ± 2% -0.0 0.49 ± 38% +0.3 0.79 ± 2% perf-profile.calltrace.cycles-pp.dequeue_task_fair.__schedule.schedule.schedule_timeout.unix_stream_data_wait > 1.03 ± 3% -0.0 1.02 ± 9% +0.1 1.12 perf-profile.calltrace.cycles-pp.skb_set_owner_w.sock_alloc_send_pskb.unix_stream_sendmsg.sock_write_iter.vfs_write > 0.66 ± 2% -0.0 0.65 ± 6% +0.1 0.79 perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.write > 0.00 +0.0 0.00 +0.6 0.61 perf-profile.calltrace.cycles-pp.enqueue_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function > 1.83 ± 2% +0.1 1.88 ± 6% +0.8 2.61 perf-profile.calltrace.cycles-pp.schedule.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg > 1.87 ± 2% +0.1 1.93 ± 6% +0.8 2.68 perf-profile.calltrace.cycles-pp.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg > 1.79 ± 2% +0.1 1.85 ± 6% +0.8 2.56 perf-profile.calltrace.cycles-pp.__schedule.schedule.schedule_timeout.unix_stream_data_wait.unix_stream_read_generic > 2.30 ± 2% +0.1 2.37 ± 6% +1.0 3.30 perf-profile.calltrace.cycles-pp.unix_stream_data_wait.unix_stream_read_generic.unix_stream_recvmsg.sock_recvmsg.sock_read_iter > 0.40 ± 53% +0.1 0.48 ± 38% +0.4 0.77 ± 2% perf-profile.calltrace.cycles-pp.dequeue_entities.dequeue_task_fair.__schedule.schedule.schedule_timeout > 0.00 +0.3 0.26 ±100% +0.6 0.61 perf-profile.calltrace.cycles-pp.pick_next_task_fair.__schedule.schedule.schedule_timeout.unix_stream_data_wait > 1.62 ± 3% +0.3 1.93 ± 24% +0.9 2.48 perf-profile.calltrace.cycles-pp.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_sync_key.sock_def_readable > 1.63 ± 3% +0.3 1.94 ± 24% +0.9 2.50 perf-profile.calltrace.cycles-pp.autoremove_wake_function.__wake_up_common.__wake_up_sync_key.sock_def_readable.unix_stream_sendmsg > 1.71 ± 3% +0.3 2.03 ± 23% +0.9 2.61 perf-profile.calltrace.cycles-pp.__wake_up_common.__wake_up_sync_key.sock_def_readable.unix_stream_sendmsg.sock_write_iter > 39.00 +0.5 39.50 +0.1 39.13 perf-profile.calltrace.cycles-pp.sock_write_iter.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe > 37.56 +0.6 38.15 +0.1 37.67 perf-profile.calltrace.cycles-pp.unix_stream_sendmsg.sock_write_iter.vfs_write.ksys_write.do_syscall_64 > 1.96 ± 3% +0.7 2.63 ± 24% +1.1 3.08 perf-profile.calltrace.cycles-pp.__wake_up_sync_key.sock_def_readable.unix_stream_sendmsg.sock_write_iter.vfs_write > 3.52 ± 2% +0.7 4.23 ± 15% +1.1 4.62 ± 2% perf-profile.calltrace.cycles-pp.sock_def_readable.unix_stream_sendmsg.sock_write_iter.vfs_write.ksys_write > 55.93 -0.8 55.12 -0.7 55.27 perf-profile.children.cycles-pp.read > 52.45 -0.5 51.95 -1.1 51.36 perf-profile.children.cycles-pp.ksys_read > 52.06 -0.5 51.60 -1.1 50.96 perf-profile.children.cycles-pp.vfs_read > 50.85 -0.4 50.48 -1.2 49.69 perf-profile.children.cycles-pp.sock_read_iter > 50.34 -0.3 50.03 -1.2 49.17 perf-profile.children.cycles-pp.sock_recvmsg > 4.68 ± 2% -0.3 4.38 ± 5% -0.2 4.49 perf-profile.children.cycles-pp.__memcg_slab_free_hook > 49.64 -0.3 49.38 -1.2 48.45 perf-profile.children.cycles-pp.unix_stream_recvmsg > 49.50 -0.2 49.25 -1.2 48.30 perf-profile.children.cycles-pp.unix_stream_read_generic > 24.52 -0.2 24.32 ± 4% -1.1 23.41 perf-profile.children.cycles-pp.alloc_skb_with_frags > 5.57 ± 2% -0.2 5.36 ± 6% -0.3 5.27 ± 2% perf-profile.children.cycles-pp.unix_stream_read_actor > 5.52 ± 2% -0.2 5.31 ± 6% -0.3 5.22 ± 2% perf-profile.children.cycles-pp.skb_copy_datagram_iter > 24.38 -0.2 24.18 ± 4% -1.1 23.27 perf-profile.children.cycles-pp.__alloc_skb > 5.43 ± 2% -0.2 5.24 ± 6% -0.3 5.14 ± 2% perf-profile.children.cycles-pp.__skb_datagram_iter > 95.87 -0.2 95.70 -0.5 95.42 perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe > 3.46 ± 2% -0.2 3.29 ± 6% -0.3 3.16 perf-profile.children.cycles-pp.__check_object_size > 1.77 ± 2% -0.2 1.61 ± 5% -0.0 1.76 perf-profile.children.cycles-pp.entry_SYSCALL_64 > 1.87 ± 2% -0.2 1.72 ± 5% +0.0 1.87 ± 2% perf-profile.children.cycles-pp.mod_objcg_state > 95.57 -0.1 95.43 -0.5 95.11 perf-profile.children.cycles-pp.do_syscall_64 > 2.79 ± 2% -0.1 2.68 ± 6% -0.3 2.48 perf-profile.children.cycles-pp.simple_copy_to_iter > 2.62 ± 2% -0.1 2.53 ± 6% -0.3 2.33 ± 2% perf-profile.children.cycles-pp.check_heap_object > 1.69 ± 2% -0.1 1.60 ± 5% +0.4 2.12 perf-profile.children.cycles-pp.syscall_exit_to_user_mode > 0.90 ± 2% -0.1 0.83 ± 5% -0.0 0.90 perf-profile.children.cycles-pp.entry_SYSRETQ_unsafe_stack > 0.85 ± 2% -0.1 0.78 ± 5% +0.0 0.86 ± 2% perf-profile.children.cycles-pp.obj_cgroup_charge > 0.10 ± 4% -0.1 0.04 ±104% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.__handle_mm_fault > 0.10 ± 4% -0.1 0.04 ±104% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.handle_mm_fault > 0.10 ± 4% -0.1 0.04 ±102% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.do_user_addr_fault > 0.10 ± 4% -0.1 0.04 ±102% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.exc_page_fault > 0.63 ± 2% -0.1 0.57 ± 5% +0.0 0.64 perf-profile.children.cycles-pp.__cond_resched > 26.17 -0.0 26.13 ± 3% -1.0 25.15 perf-profile.children.cycles-pp.sock_alloc_send_pskb > 0.10 ± 4% -0.0 0.05 ± 63% +0.0 0.10 ± 8% perf-profile.children.cycles-pp.asm_exc_page_fault > 0.32 ± 2% -0.0 0.28 ± 3% +0.0 0.33 perf-profile.children.cycles-pp.task_mm_cid_work > 0.32 ± 2% -0.0 0.28 ± 3% +0.0 0.33 perf-profile.children.cycles-pp.task_work_run > 0.34 ± 2% -0.0 0.30 ± 5% -0.0 0.34 ± 2% perf-profile.children.cycles-pp.rcu_all_qs > 0.32 ± 3% -0.0 0.29 ± 6% +0.0 0.33 ± 2% perf-profile.children.cycles-pp.__virt_addr_valid > 0.32 ± 3% -0.0 0.29 ± 5% +0.0 0.36 ± 2% perf-profile.children.cycles-pp.syscall_return_via_sysret > 0.05 -0.0 0.02 ±129% +0.0 0.07 ± 4% perf-profile.children.cycles-pp.avg_vruntime > 0.18 ± 4% -0.0 0.16 ± 7% +0.1 0.27 perf-profile.children.cycles-pp.__enqueue_entity > 0.47 ± 2% -0.0 0.44 ± 5% +0.0 0.50 perf-profile.children.cycles-pp.mutex_lock > 0.23 ± 3% -0.0 0.21 ± 6% +0.1 0.33 ± 2% perf-profile.children.cycles-pp.set_next_entity > 0.14 ± 4% -0.0 0.12 ± 4% +0.1 0.20 ± 2% perf-profile.children.cycles-pp.__dequeue_entity > 1.05 ± 3% -0.0 1.04 ± 9% +0.1 1.14 perf-profile.children.cycles-pp.skb_set_owner_w > 0.28 ± 2% -0.0 0.26 ± 7% +0.1 0.38 perf-profile.children.cycles-pp.switch_fpu_return > 0.48 ± 2% -0.0 0.47 ± 6% +0.2 0.70 perf-profile.children.cycles-pp.enqueue_entity > 0.08 ± 3% -0.0 0.07 ± 9% +0.0 0.13 ± 6% perf-profile.children.cycles-pp.asm_sysvec_reschedule_ipi > 0.24 ± 4% -0.0 0.22 ± 6% +0.1 0.31 ± 4% perf-profile.children.cycles-pp.__rseq_handle_notify_resume > 0.06 -0.0 0.05 +0.0 0.08 ± 4% perf-profile.children.cycles-pp.cpuacct_charge > 0.29 ± 3% -0.0 0.28 ± 7% +0.1 0.40 ± 4% perf-profile.children.cycles-pp.switch_mm_irqs_off > 0.16 ± 2% -0.0 0.15 ± 4% +0.0 0.18 perf-profile.children.cycles-pp.syscall_exit_to_user_mode_prepare > 0.12 ± 3% -0.0 0.12 ± 6% +0.1 0.18 perf-profile.children.cycles-pp.update_rq_clock_task > 0.18 ± 3% -0.0 0.17 ± 6% +0.1 0.24 ± 2% perf-profile.children.cycles-pp.restore_fpregs_from_fpstate > 0.18 ± 4% -0.0 0.17 ± 6% +0.1 0.23 ± 3% perf-profile.children.cycles-pp.rseq_ip_fixup > 0.05 ± 6% -0.0 0.04 ± 37% +0.0 0.07 ± 5% perf-profile.children.cycles-pp.rseq_update_cpu_node_id > 0.05 -0.0 0.04 ± 37% +0.0 0.06 ± 7% perf-profile.children.cycles-pp.sched_clock > 0.07 ± 7% -0.0 0.06 ± 9% +0.0 0.09 ± 5% perf-profile.children.cycles-pp.os_xsave > 0.13 ± 6% -0.0 0.12 ± 6% +0.0 0.17 ± 3% perf-profile.children.cycles-pp.rseq_get_rseq_cs > 0.12 ± 6% -0.0 0.12 ± 7% +0.0 0.16 ± 4% perf-profile.children.cycles-pp.__get_user_8 > 0.14 ± 3% -0.0 0.13 ± 7% +0.1 0.20 ± 3% perf-profile.children.cycles-pp.wakeup_preempt > 0.09 ± 5% -0.0 0.09 ± 5% +0.0 0.11 ± 4% perf-profile.children.cycles-pp.cgroup_rstat_updated > 0.14 ± 3% -0.0 0.13 ± 4% +0.0 0.18 ± 3% perf-profile.children.cycles-pp.__update_load_avg_se > 0.18 ± 2% -0.0 0.17 ± 2% -0.0 0.14 ± 2% perf-profile.children.cycles-pp.put_cpu_partial > 0.13 ± 3% -0.0 0.13 ± 7% +0.0 0.16 ± 3% perf-profile.children.cycles-pp.pick_eevdf > 0.07 ± 4% -0.0 0.07 ± 7% +0.0 0.10 ± 5% perf-profile.children.cycles-pp.check_preempt_wakeup_fair > 0.08 ± 5% -0.0 0.08 ± 8% +0.0 0.13 perf-profile.children.cycles-pp.update_curr_se > 0.06 ± 5% -0.0 0.06 ± 5% +0.0 0.07 ± 6% perf-profile.children.cycles-pp.sched_clock_cpu > 0.07 +0.0 0.07 ± 7% +0.0 0.11 perf-profile.children.cycles-pp.put_prev_entity > 0.00 +0.0 0.00 +0.1 0.05 perf-profile.children.cycles-pp.__put_user_8 > 0.06 ± 10% +0.0 0.06 ± 21% +0.4 0.43 ± 4% perf-profile.children.cycles-pp.get_any_partial > 0.22 ± 3% +0.0 0.22 ± 7% +0.1 0.31 ± 2% perf-profile.children.cycles-pp.update_cfs_group > 0.06 ± 6% +0.0 0.06 ± 10% +0.0 0.09 ± 5% perf-profile.children.cycles-pp.___perf_sw_event > 0.12 ± 2% +0.0 0.12 ± 7% +0.1 0.17 ± 2% perf-profile.children.cycles-pp.wake_affine > 0.07 +0.0 0.07 ± 11% +0.0 0.10 ± 4% perf-profile.children.cycles-pp.update_min_vruntime > 0.10 ± 4% +0.0 0.11 ± 7% +0.0 0.15 ± 3% perf-profile.children.cycles-pp.__update_load_avg_cfs_rq > 0.05 ± 35% +0.0 0.06 ± 5% +0.0 0.08 ± 5% perf-profile.children.cycles-pp.perf_tp_event > 0.09 ± 35% +0.0 0.10 ± 7% +0.0 0.14 ± 3% perf-profile.children.cycles-pp.perf_trace_sched_wakeup_template > 0.05 ± 36% +0.0 0.05 ± 6% +0.0 0.08 ± 4% perf-profile.children.cycles-pp.place_entity > 0.07 ± 4% +0.0 0.08 ± 9% +0.0 0.10 ± 3% perf-profile.children.cycles-pp.task_h_load > 0.00 +0.0 0.01 ±264% +0.1 0.06 ± 9% perf-profile.children.cycles-pp.native_irq_return_iret > 0.12 ± 4% +0.0 0.12 ± 8% +0.1 0.17 ± 4% perf-profile.children.cycles-pp.__switch_to > 0.12 ± 3% +0.0 0.13 ± 6% +0.0 0.16 ± 2% perf-profile.children.cycles-pp.__switch_to_asm > 0.75 ± 2% +0.0 0.76 ± 7% +0.3 1.08 perf-profile.children.cycles-pp.enqueue_task_fair > 0.16 ± 5% +0.0 0.17 ± 7% +0.0 0.20 ± 5% perf-profile.children.cycles-pp.task_tick_fair > 0.47 ± 2% +0.0 0.49 ± 6% +0.2 0.64 perf-profile.children.cycles-pp.update_load_avg > 0.64 ± 2% +0.0 0.65 ± 6% +0.3 0.90 perf-profile.children.cycles-pp.dequeue_task_fair > 0.76 ± 2% +0.0 0.78 ± 4% +0.2 0.98 perf-profile.children.cycles-pp.pick_next_task_fair > 0.22 ± 3% +0.0 0.23 ± 7% +0.1 0.28 ± 2% perf-profile.children.cycles-pp.pick_task_fair > 0.12 ± 4% +0.0 0.13 ± 14% +0.1 0.17 perf-profile.children.cycles-pp.reweight_entity > 0.20 ± 3% +0.0 0.22 ± 6% +0.0 0.24 ± 4% perf-profile.children.cycles-pp.sched_tick > 0.88 ± 2% +0.0 0.90 ± 7% +0.4 1.29 perf-profile.children.cycles-pp.ttwu_do_activate > 0.30 ± 4% +0.0 0.32 ± 7% +0.1 0.38 ± 2% perf-profile.children.cycles-pp.tick_nohz_handler > 0.35 ± 2% +0.0 0.37 ± 5% +0.1 0.49 perf-profile.children.cycles-pp.dequeue_entity > 0.31 ± 4% +0.0 0.33 ± 6% +0.1 0.38 ± 2% perf-profile.children.cycles-pp.__hrtimer_run_queues > 0.29 ± 4% +0.0 0.31 ± 6% +0.1 0.36 ± 3% perf-profile.children.cycles-pp.update_process_times > 0.13 ± 2% +0.0 0.15 ± 11% +0.1 0.19 ± 2% perf-profile.children.cycles-pp.finish_task_switch > 0.18 ± 2% +0.0 0.21 ± 8% +0.1 0.26 perf-profile.children.cycles-pp.prepare_task_switch > 0.35 ± 3% +0.0 0.37 ± 5% +0.1 0.43 ± 2% perf-profile.children.cycles-pp.hrtimer_interrupt > 0.35 ± 3% +0.0 0.37 ± 6% +0.1 0.44 ± 2% perf-profile.children.cycles-pp.__sysvec_apic_timer_interrupt > 0.63 ± 2% +0.0 0.65 ± 7% +0.3 0.89 ± 2% perf-profile.children.cycles-pp.dequeue_entities > 0.09 ± 8% +0.0 0.12 ± 7% +0.0 0.11 ± 11% perf-profile.children.cycles-pp.generic_perform_write > 2.21 ± 2% +0.0 2.23 ± 6% +0.8 3.04 perf-profile.children.cycles-pp.schedule_timeout > 0.07 ± 10% +0.0 0.10 ± 11% +0.0 0.08 ± 20% perf-profile.children.cycles-pp.sched_balance_find_src_group > 0.06 ± 10% +0.0 0.09 ± 11% +0.0 0.06 ± 40% perf-profile.children.cycles-pp.update_sg_lb_stats > 0.07 ± 10% +0.0 0.10 ± 11% +0.0 0.07 ± 19% perf-profile.children.cycles-pp.update_sd_lb_stats > 0.10 ± 3% +0.0 0.13 ± 14% +0.0 0.13 ± 5% perf-profile.children.cycles-pp.update_rq_clock > 0.01 ±282% +0.0 0.04 ± 57% +0.1 0.06 ± 7% perf-profile.children.cycles-pp.native_sched_clock > 0.43 ± 3% +0.0 0.47 ± 7% +0.2 0.64 perf-profile.children.cycles-pp.update_curr > 0.13 ± 8% +0.0 0.16 ± 8% +0.0 0.16 ± 12% perf-profile.children.cycles-pp.writen > 0.39 ± 3% +0.0 0.42 ± 6% +0.1 0.48 ± 2% perf-profile.children.cycles-pp.sysvec_apic_timer_interrupt > 0.42 ± 3% +0.0 0.46 ± 6% +0.1 0.52 ± 2% perf-profile.children.cycles-pp.asm_sysvec_apic_timer_interrupt > 0.02 ±111% +0.0 0.06 ± 13% +0.0 0.04 ± 38% perf-profile.children.cycles-pp.set_task_cpu > 0.02 ±141% +0.0 0.06 ± 11% +0.0 0.06 ± 11% perf-profile.children.cycles-pp.ring_buffer_read_head > 2.45 ± 2% +0.0 2.50 ± 6% +1.0 3.43 perf-profile.children.cycles-pp.schedule > 0.00 +0.1 0.06 ± 8% +0.0 0.03 ± 77% perf-profile.children.cycles-pp.vruntime_eligible > 0.22 ± 8% +0.1 0.28 ± 9% +0.1 0.27 ± 12% perf-profile.children.cycles-pp.perf_mmap__push > 0.23 ± 8% +0.1 0.29 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.record__mmap_read_evlist > 0.23 ± 7% +0.1 0.29 ± 9% +0.1 0.28 ± 11% perf-profile.children.cycles-pp.cmd_record > 0.23 ± 7% +0.1 0.30 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.handle_internal_command > 0.23 ± 7% +0.1 0.30 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.main > 0.23 ± 7% +0.1 0.30 ± 9% +0.1 0.28 ± 12% perf-profile.children.cycles-pp.run_builtin > 2.31 ± 2% +0.1 2.38 ± 7% +1.0 3.32 perf-profile.children.cycles-pp.unix_stream_data_wait > 0.18 ± 38% +0.1 0.26 ± 17% +0.2 0.41 ± 3% perf-profile.children.cycles-pp.ordered_events__queue > 0.18 ± 38% +0.1 0.26 ± 17% +0.2 0.41 ± 3% perf-profile.children.cycles-pp.queue_event > 0.19 ± 37% +0.1 0.26 ± 16% +0.2 0.41 ± 3% perf-profile.children.cycles-pp.process_simple > 0.00 +0.1 0.07 ± 19% +0.0 0.00 perf-profile.children.cycles-pp.schedule_idle > 0.19 ± 37% +0.1 0.29 ± 17% +0.2 0.42 ± 4% perf-profile.children.cycles-pp.perf_session__process_events > 0.19 ± 37% +0.1 0.29 ± 17% +0.2 0.42 ± 4% perf-profile.children.cycles-pp.reader__read_event > 0.19 ± 37% +0.1 0.29 ± 17% +0.2 0.42 ± 4% perf-profile.children.cycles-pp.record__finish_output > 2.45 ± 2% +0.1 2.56 ± 6% +1.0 3.43 perf-profile.children.cycles-pp.__schedule > 0.00 +0.1 0.12 ± 19% +0.0 0.00 perf-profile.children.cycles-pp.flush_smp_call_function_queue > 0.42 ± 18% +0.1 0.54 ± 23% +0.3 0.70 ± 4% perf-profile.children.cycles-pp.__cmd_record > 0.00 +0.1 0.14 ± 20% +0.0 0.00 perf-profile.children.cycles-pp.sched_ttwu_pending > 0.00 +0.2 0.15 ± 17% +0.0 0.00 perf-profile.children.cycles-pp.intel_idle > 0.00 +0.2 0.17 ± 18% +0.0 0.00 perf-profile.children.cycles-pp.__flush_smp_call_function_queue > 0.01 ±282% +0.2 0.18 ± 70% +0.1 0.06 perf-profile.children.cycles-pp.available_idle_cpu > 0.23 ± 3% +0.2 0.43 ± 16% +0.1 0.31 perf-profile.children.cycles-pp.prepare_to_wait > 0.00 +0.2 0.20 ± 18% +0.0 0.03 ±104% perf-profile.children.cycles-pp.cpuidle_enter > 0.00 +0.2 0.20 ± 18% +0.0 0.03 ±104% perf-profile.children.cycles-pp.cpuidle_enter_state > 0.01 ±282% +0.2 0.22 ± 16% +0.0 0.04 ± 81% perf-profile.children.cycles-pp.cpuidle_idle_call > 3.50 ± 2% +0.2 3.73 ± 8% +0.4 3.87 ± 2% perf-profile.children.cycles-pp._raw_spin_lock > 1.81 ± 3% +0.3 2.12 ± 22% +0.9 2.68 perf-profile.children.cycles-pp.autoremove_wake_function > 1.80 ± 3% +0.3 2.12 ± 22% +0.9 2.66 perf-profile.children.cycles-pp.try_to_wake_up > 1.89 ± 3% +0.3 2.22 ± 21% +0.9 2.80 perf-profile.children.cycles-pp.__wake_up_common > 0.00 +0.3 0.35 ±110% +0.0 0.00 perf-profile.children.cycles-pp.select_idle_cpu > 0.06 ± 6% +0.4 0.41 ± 95% +0.0 0.09 ± 3% perf-profile.children.cycles-pp.select_idle_sibling > 0.22 ± 2% +0.4 0.57 ± 70% +0.1 0.31 perf-profile.children.cycles-pp.select_task_rq_fair > 0.26 ± 5% +0.4 0.62 ± 65% +0.1 0.38 perf-profile.children.cycles-pp.select_task_rq > 0.04 ± 77% +0.4 0.43 ± 18% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.start_secondary > 0.04 ± 77% +0.4 0.43 ± 18% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.do_idle > 0.04 ± 77% +0.4 0.43 ± 17% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.common_startup_64 > 0.04 ± 77% +0.4 0.43 ± 17% +0.1 0.10 ± 30% perf-profile.children.cycles-pp.cpu_startup_entry > 40.28 +0.4 40.71 +0.2 40.46 perf-profile.children.cycles-pp.vfs_write > 39.04 +0.5 39.54 +0.1 39.18 perf-profile.children.cycles-pp.sock_write_iter > 37.73 +0.6 38.31 +0.1 37.84 perf-profile.children.cycles-pp.unix_stream_sendmsg > 2.15 ± 3% +0.7 2.82 ± 23% +1.1 3.27 perf-profile.children.cycles-pp.__wake_up_sync_key > 3.55 ± 2% +0.7 4.25 ± 15% +1.1 4.65 ± 2% perf-profile.children.cycles-pp.sock_def_readable > 2.90 ± 2% -0.2 2.74 ± 5% -0.2 2.69 ± 2% perf-profile.self.cycles-pp.__memcg_slab_free_hook > 1.50 ± 2% -0.1 1.37 ± 5% +0.0 1.51 ± 2% perf-profile.self.cycles-pp.mod_objcg_state > 1.41 ± 2% -0.1 1.30 ± 5% -0.2 1.20 ± 2% perf-profile.self.cycles-pp.kmem_cache_free > 0.83 ± 3% -0.1 0.74 ± 5% -0.0 0.81 perf-profile.self.cycles-pp.read > 0.88 ± 2% -0.1 0.80 ± 5% -0.0 0.88 perf-profile.self.cycles-pp.entry_SYSRETQ_unsafe_stack > 0.74 ± 2% -0.1 0.67 ± 5% +0.0 0.75 ± 2% perf-profile.self.cycles-pp.write > 0.70 ± 2% -0.1 0.64 ± 6% +0.0 0.73 ± 2% perf-profile.self.cycles-pp.vfs_read > 0.67 ± 2% -0.1 0.61 ± 5% +0.0 0.68 perf-profile.self.cycles-pp.vfs_write > 2.19 ± 2% -0.1 2.13 ± 6% -0.3 1.88 ± 2% perf-profile.self.cycles-pp.check_heap_object > 0.67 ± 2% -0.1 0.62 ± 4% -0.0 0.65 perf-profile.self.cycles-pp.__kmalloc_node_track_caller_noprof > 0.52 ± 3% -0.1 0.47 ± 4% -0.0 0.52 perf-profile.self.cycles-pp.obj_cgroup_charge > 0.51 ± 2% -0.0 0.46 ± 5% -0.0 0.51 ± 2% perf-profile.self.cycles-pp.kmem_cache_alloc_node_noprof > 0.29 ± 2% -0.0 0.25 ± 3% +0.0 0.31 perf-profile.self.cycles-pp.task_mm_cid_work > 0.43 ± 2% -0.0 0.40 ± 4% +0.0 0.44 perf-profile.self.cycles-pp.do_syscall_64 > 0.34 ± 3% -0.0 0.31 ± 6% +0.0 0.35 ± 2% perf-profile.self.cycles-pp.__skb_datagram_iter > 0.29 ± 2% -0.0 0.26 ± 6% +0.0 0.30 ± 2% perf-profile.self.cycles-pp.__virt_addr_valid > 0.33 ± 2% -0.0 0.30 ± 6% +0.0 0.33 ± 2% perf-profile.self.cycles-pp.__cond_resched > 0.28 ± 3% -0.0 0.25 ± 5% +0.0 0.31 ± 2% perf-profile.self.cycles-pp.syscall_return_via_sysret > 0.37 ± 2% -0.0 0.34 ± 6% -0.0 0.36 perf-profile.self.cycles-pp.__check_object_size > 0.18 ± 5% -0.0 0.15 ± 6% +0.1 0.27 perf-profile.self.cycles-pp.__enqueue_entity > 0.70 -0.0 0.67 ± 2% -0.1 0.60 perf-profile.self.cycles-pp.get_partial_node > 0.21 ± 3% -0.0 0.18 ± 6% -0.0 0.21 ± 2% perf-profile.self.cycles-pp.rcu_all_qs > 0.22 ± 3% -0.0 0.20 ± 6% -0.0 0.22 ± 2% perf-profile.self.cycles-pp.x64_sys_call > 0.34 ± 2% -0.0 0.32 ± 5% +0.0 0.37 perf-profile.self.cycles-pp.syscall_exit_to_user_mode > 1.75 ± 2% -0.0 1.72 ± 5% -0.1 1.63 ± 2% perf-profile.self.cycles-pp.__alloc_skb > 0.95 -0.0 0.93 -0.1 0.87 perf-profile.self.cycles-pp.___slab_alloc > 0.42 ± 2% -0.0 0.40 ± 7% +0.0 0.45 ± 2% perf-profile.self.cycles-pp.skb_release_data > 0.19 ± 2% -0.0 0.17 ± 6% -0.0 0.18 ± 2% perf-profile.self.cycles-pp.rw_verify_area > 1.03 ± 3% -0.0 1.02 ± 9% +0.1 1.12 perf-profile.self.cycles-pp.skb_set_owner_w > 0.03 ± 89% -0.0 0.01 ±173% +0.0 0.06 perf-profile.self.cycles-pp.avg_vruntime > 0.06 ± 5% -0.0 0.04 ± 37% +0.0 0.08 ± 4% perf-profile.self.cycles-pp.cpuacct_charge > 0.32 ± 2% -0.0 0.30 ± 5% +0.0 0.34 ± 2% perf-profile.self.cycles-pp.mutex_lock > 0.11 ± 4% -0.0 0.10 ± 7% +0.0 0.15 ± 2% perf-profile.self.cycles-pp.__dequeue_entity > 0.06 ± 8% -0.0 0.05 ± 38% +0.0 0.08 ± 6% perf-profile.self.cycles-pp.prepare_to_wait > 0.18 ± 3% -0.0 0.17 ± 7% +0.1 0.24 ± 2% perf-profile.self.cycles-pp.restore_fpregs_from_fpstate > 0.11 ± 4% -0.0 0.10 ± 6% +0.1 0.17 perf-profile.self.cycles-pp.update_rq_clock_task > 0.29 ± 3% -0.0 0.28 ± 7% +0.1 0.40 ± 4% perf-profile.self.cycles-pp.switch_mm_irqs_off > 0.12 ± 3% -0.0 0.11 ± 7% +0.0 0.15 ± 3% perf-profile.self.cycles-pp.__get_user_8 > 0.05 -0.0 0.04 ± 37% +0.0 0.07 ± 4% perf-profile.self.cycles-pp.rseq_update_cpu_node_id > 0.01 ±282% -0.0 0.00 +0.1 0.06 perf-profile.self.cycles-pp.place_entity > 0.09 ± 3% -0.0 0.08 ± 8% +0.0 0.13 ± 2% perf-profile.self.cycles-pp.unix_stream_data_wait > 0.22 ± 3% -0.0 0.22 ± 7% +0.1 0.31 ± 3% perf-profile.self.cycles-pp.update_cfs_group > 0.05 ± 8% -0.0 0.05 ± 40% +0.0 0.07 perf-profile.self.cycles-pp.dequeue_entities > 0.10 ± 5% -0.0 0.10 ± 7% +0.0 0.15 ± 3% perf-profile.self.cycles-pp.switch_fpu_return > 0.05 -0.0 0.05 ± 38% +0.0 0.06 ± 7% perf-profile.self.cycles-pp.___perf_sw_event > 0.05 -0.0 0.05 ± 38% +0.0 0.07 ± 4% perf-profile.self.cycles-pp.enqueue_task_fair > 0.13 ± 3% -0.0 0.12 ± 5% +0.0 0.17 ± 2% perf-profile.self.cycles-pp.__update_load_avg_se > 0.10 ± 3% -0.0 0.10 ± 7% +0.0 0.14 ± 3% perf-profile.self.cycles-pp.try_to_wake_up > 0.17 -0.0 0.17 -0.0 0.13 ± 2% perf-profile.self.cycles-pp.put_cpu_partial > 0.06 ± 7% -0.0 0.06 ± 9% +0.0 0.09 ± 5% perf-profile.self.cycles-pp.os_xsave > 0.32 -0.0 0.32 ± 2% -0.0 0.28 perf-profile.self.cycles-pp.__put_partials > 0.08 ± 3% -0.0 0.08 ± 6% +0.0 0.12 ± 3% perf-profile.self.cycles-pp.update_curr_se > 0.09 ± 5% -0.0 0.09 ± 7% +0.0 0.14 ± 3% perf-profile.self.cycles-pp.schedule_timeout > 0.11 ± 4% -0.0 0.11 ± 6% +0.0 0.13 perf-profile.self.cycles-pp.pick_eevdf > 0.06 ± 5% -0.0 0.06 ± 5% +0.0 0.08 ± 4% perf-profile.self.cycles-pp.dequeue_entity > 0.06 -0.0 0.06 ± 8% +0.0 0.08 ± 5% perf-profile.self.cycles-pp.pick_task_fair > 0.00 +0.0 0.00 +0.1 0.05 ± 6% perf-profile.self.cycles-pp.select_task_rq_fair > 0.00 +0.0 0.00 +0.1 0.06 perf-profile.self.cycles-pp.reweight_entity > 0.09 ± 3% +0.0 0.09 ± 5% +0.0 0.13 perf-profile.self.cycles-pp.finish_task_switch > 0.07 ± 7% +0.0 0.07 ± 8% +0.0 0.10 perf-profile.self.cycles-pp.update_min_vruntime > 0.23 ± 2% +0.0 0.23 ± 7% +0.1 0.30 perf-profile.self.cycles-pp.update_load_avg > 0.07 ± 9% +0.0 0.07 ± 9% +0.0 0.10 ± 4% perf-profile.self.cycles-pp.task_h_load > 0.10 ± 3% +0.0 0.10 ± 6% +0.0 0.14 ± 3% perf-profile.self.cycles-pp.__update_load_avg_cfs_rq > 0.04 ± 53% +0.0 0.04 ± 38% +0.0 0.07 ± 4% perf-profile.self.cycles-pp.select_task_rq > 0.00 +0.0 0.01 ±264% +0.1 0.06 ± 9% perf-profile.self.cycles-pp.native_irq_return_iret > 0.12 ± 2% +0.0 0.13 ± 6% +0.0 0.16 ± 2% perf-profile.self.cycles-pp.__switch_to_asm > 0.01 ±187% +0.0 0.02 ±129% +0.0 0.06 perf-profile.self.cycles-pp.perf_tp_event > 0.11 ± 3% +0.0 0.12 ± 8% +0.0 0.16 ± 3% perf-profile.self.cycles-pp.__switch_to > 0.08 ± 5% +0.0 0.09 ± 7% +0.0 0.12 ± 2% perf-profile.self.cycles-pp.__wake_up_common > 0.24 ± 2% +0.0 0.26 ± 7% +0.1 0.35 perf-profile.self.cycles-pp.__schedule > 0.12 ± 5% +0.0 0.14 ± 11% +0.1 0.18 ± 3% perf-profile.self.cycles-pp.update_curr > 0.00 +0.0 0.02 ±129% +0.1 0.05 ± 9% perf-profile.self.cycles-pp.native_sched_clock > 0.13 ± 3% +0.0 0.15 ± 7% +0.1 0.18 ± 2% perf-profile.self.cycles-pp.prepare_task_switch > 0.05 +0.0 0.08 ± 23% +0.0 0.06 ± 6% perf-profile.self.cycles-pp.update_rq_clock > 0.00 +0.0 0.04 ± 38% +0.1 0.05 ± 9% perf-profile.self.cycles-pp.enqueue_entity > 0.00 +0.1 0.06 ± 11% +0.0 0.05 ± 38% perf-profile.self.cycles-pp.ring_buffer_read_head > 0.18 ± 37% +0.1 0.25 ± 17% +0.2 0.40 ± 3% perf-profile.self.cycles-pp.queue_event > 0.00 +0.2 0.15 ± 17% +0.0 0.00 perf-profile.self.cycles-pp.intel_idle > 0.01 ±282% +0.2 0.18 ± 70% +0.0 0.05 ± 8% perf-profile.self.cycles-pp.available_idle_cpu ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-30 2:04 ` Honglei Wang @ 2024-09-30 2:22 ` Chen Yu 2024-09-30 6:11 ` Honglei Wang 0 siblings, 1 reply; 12+ messages in thread From: Chen Yu @ 2024-09-30 2:22 UTC (permalink / raw) To: Honglei Wang Cc: Oliver Sang, Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, K Prateek Nayak On 2024-09-30 at 10:04:13 +0800, Honglei Wang wrote: > Hi Oliver, > > On 2024/9/29 21:51, Oliver Sang wrote: > > hi, Chenyu and Honglei, > > > > On Sat, Sep 28, 2024 at 02:28:30PM +0800, Chen Yu wrote: > >> Hi Honglei, > >> > >> On 2024-09-27 at 21:38:53 +0800, Honglei Wang wrote: > >>> Hi Yu, > >>> > >>> Yep, I understand the preemption should happen in the same cfs level. Just > >>> not sure the purpose of the 'nr_running check' stuff. Perhaps its role is > >>> just to judge whether it’s necessary to do the preemption check. If there is > >>> at least one more ready (cfs) task in the rq and current is not eligible, we > >>> take care of the waiting tasks. Thoughts? > >> > >> I got your point and it makes sense. Whether the preemption check should be triggered > >> seems to be a heuristic trade-off to me. I'm ok with using more aggressive preemption > >> strategy as it depends on whether that workload prefers latency or throughput, and as > >> long as it did not introduce regression :-) > >> > >> Oliver, may I know if you happen to have time for a test if the following change > >> suggested by Honglei would make the regression go away? Thanks. > > > > I applied the patch directly upon 85e511df3cec, the test found it can reduce the > > regression but not totally reovered > > > > ========================================================================================= > > compiler/cpufreq_governor/ipc/iterations/kconfig/mode/nr_threads/rootfs/tbox_group/testcase: > > gcc-12/performance/socket/4/x86_64-rhel-8.3/process/50%/debian-12-x86_64-20240206.cgz/lkp-spr-r02/hackbench > > > > commit: > > 82e9d0456e06 ("sched/fair: Avoid re-setting virtual deadline on 'migrations'") > > 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > > 8079496d311b <--- patch from Honglei > > > > 82e9d0456e06cebe 85e511df3cec46021024176672a 8079496d311b6b0d4dae973f4df > > ---------------- --------------------------- --------------------------- > > %stddev %change %stddev %change %stddev > > \ | \ | \ > > 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput > > > > Thanks a lot for running the test. The result is as expectation, as the > strategy of short slices tends to favor more frequent scheduling to help > delay-sensitive tasks acquire CPU as early as possible. > > I suspect that the current test environment does not have any special > configurations for slices. In this case, a 3.2% regression is still > somewhat significant. As Yu mentioned, this is a heuristic adjustment. > In this particular case, it seems that Yu's patch is more effective in > solving the problem. Let's delegate the preemption opportunity to the > higher-level update_curr() function. > Thanks Oliver and Hongwei. Hi Hongwei, do you mind if I added your Reviewed-by tag on this? thanks, Chenyu ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-30 2:22 ` Chen Yu @ 2024-09-30 6:11 ` Honglei Wang 0 siblings, 0 replies; 12+ messages in thread From: Honglei Wang @ 2024-09-30 6:11 UTC (permalink / raw) To: Chen Yu Cc: Oliver Sang, Ingo Molnar, Peter Zijlstra, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, K Prateek Nayak On 2024/9/30 10:22, Chen Yu wrote: > On 2024-09-30 at 10:04:13 +0800, Honglei Wang wrote: >> Hi Oliver, >> >> On 2024/9/29 21:51, Oliver Sang wrote: >>> hi, Chenyu and Honglei, >>> >>> On Sat, Sep 28, 2024 at 02:28:30PM +0800, Chen Yu wrote: >>>> Hi Honglei, >>>> >>>> On 2024-09-27 at 21:38:53 +0800, Honglei Wang wrote: >>>>> Hi Yu, >>>>> >>>>> Yep, I understand the preemption should happen in the same cfs level. Just >>>>> not sure the purpose of the 'nr_running check' stuff. Perhaps its role is >>>>> just to judge whether it’s necessary to do the preemption check. If there is >>>>> at least one more ready (cfs) task in the rq and current is not eligible, we >>>>> take care of the waiting tasks. Thoughts? >>>> >>>> I got your point and it makes sense. Whether the preemption check should be triggered >>>> seems to be a heuristic trade-off to me. I'm ok with using more aggressive preemption >>>> strategy as it depends on whether that workload prefers latency or throughput, and as >>>> long as it did not introduce regression :-) >>>> >>>> Oliver, may I know if you happen to have time for a test if the following change >>>> suggested by Honglei would make the regression go away? Thanks. >>> >>> I applied the patch directly upon 85e511df3cec, the test found it can reduce the >>> regression but not totally reovered >>> >>> ========================================================================================= >>> compiler/cpufreq_governor/ipc/iterations/kconfig/mode/nr_threads/rootfs/tbox_group/testcase: >>> gcc-12/performance/socket/4/x86_64-rhel-8.3/process/50%/debian-12-x86_64-20240206.cgz/lkp-spr-r02/hackbench >>> >>> commit: >>> 82e9d0456e06 ("sched/fair: Avoid re-setting virtual deadline on 'migrations'") >>> 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") >>> 8079496d311b <--- patch from Honglei >>> >>> 82e9d0456e06cebe 85e511df3cec46021024176672a 8079496d311b6b0d4dae973f4df >>> ---------------- --------------------------- --------------------------- >>> %stddev %change %stddev %change %stddev >>> \ | \ | \ >>> 623219 -13.1% 541887 -3.2% 603080 hackbench.throughput >>> >> >> Thanks a lot for running the test. The result is as expectation, as the >> strategy of short slices tends to favor more frequent scheduling to help >> delay-sensitive tasks acquire CPU as early as possible. >> >> I suspect that the current test environment does not have any special >> configurations for slices. In this case, a 3.2% regression is still >> somewhat significant. As Yu mentioned, this is a heuristic adjustment. >> In this particular case, it seems that Yu's patch is more effective in >> solving the problem. Let's delegate the preemption opportunity to the >> higher-level update_curr() function. >> > > Thanks Oliver and Hongwei. > > Hi Hongwei, do you mind if I added your Reviewed-by tag on this? > OK. Feel free to include: Reviewed-by: Honglei Wang <jameshongleiwang@126.com> Thanks, Honglei > thanks, > Chenyu ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-25 8:54 [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running Chen Yu 2024-09-27 1:54 ` Honglei Wang @ 2024-09-30 7:50 ` Peter Zijlstra 2024-09-30 11:01 ` Chen Yu 2024-10-04 10:16 ` [tip: sched/urgent] " tip-bot2 for Chen Yu 2 siblings, 1 reply; 12+ messages in thread From: Peter Zijlstra @ 2024-09-30 7:50 UTC (permalink / raw) To: Chen Yu Cc: Ingo Molnar, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, kernel test robot, K Prateek Nayak On Wed, Sep 25, 2024 at 04:54:40PM +0800, Chen Yu wrote: > Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > introduced a mechanism that a wakee with shorter slice could preempt > the current running task. It also lower the bar for the current task > to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running > when the current task has ran out of time slice. But there is a scenario > that is problematic. Say, if there is 1 cfs task and 1 rt task, before > 85e511df3cec, update_deadline() will not trigger a reschedule, and after > 85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr() > would happen. > So I'm applying the patch (that change was not indended), but I'm a little confused by the above description. If we already have TIF_RESCHED set, then doing resched_curr() should be a NO-OP. At the same time, how can we have an RT task and not get TIF_RESCHED set. ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-30 7:50 ` Peter Zijlstra @ 2024-09-30 11:01 ` Chen Yu 0 siblings, 0 replies; 12+ messages in thread From: Chen Yu @ 2024-09-30 11:01 UTC (permalink / raw) To: Peter Zijlstra Cc: Ingo Molnar, Juri Lelli, Vincent Guittot, Dietmar Eggemann, Valentin Schneider, Chunxin Zang, linux-kernel, Chen Yu, kernel test robot, K Prateek Nayak On 2024-09-30 at 09:50:45 +0200, Peter Zijlstra wrote: > On Wed, Sep 25, 2024 at 04:54:40PM +0800, Chen Yu wrote: > > Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") > > introduced a mechanism that a wakee with shorter slice could preempt > > the current running task. It also lower the bar for the current task > > to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running > > when the current task has ran out of time slice. But there is a scenario > > that is problematic. Say, if there is 1 cfs task and 1 rt task, before > > 85e511df3cec, update_deadline() will not trigger a reschedule, and after > > 85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr() > > would happen. > > > > So I'm applying the patch (that change was not indended), but I'm a > little confused by the above description. If we already have TIF_RESCHED > set, then doing resched_curr() should be a NO-OP. > > At the same time, how can we have an RT task and not get TIF_RESCHED > set. You are right, the curr should have TIF_RESCHED set if there is RT in the rq, because after the RT task is enqueued, wakeup_preempt() should set TIF_RESCHED for current CFS task. The 1 RT + 1 CFS task was just suspicion when trying to figure out why commit 85e511df3cec increased preemption. (besides the suspicion of did_preempt_short() -> entity_eligible().) and the test patch seems to recover the performance 0day reported, so we did not chase from did_preempt_short(). Another scenario I'm thinking of to increase preemption regarding the nr_running check might be(and per Honglei's feedback): There is 1 cfs task p1 in the current cfs_rq1, and 1 cfs task p2 in the parent cfs_rq2. The rq->nr_running is 2, and the p1's cfs_rq1->nr_running is 1. Before the commit, there is no preemption on cfs_rq1, after the commit, p1 is set to TIF_RESCHED and the CPU reschedule from topdown and pick p2. thanks, Chenyu ^ permalink raw reply [flat|nested] 12+ messages in thread
* [tip: sched/urgent] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running 2024-09-25 8:54 [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running Chen Yu 2024-09-27 1:54 ` Honglei Wang 2024-09-30 7:50 ` Peter Zijlstra @ 2024-10-04 10:16 ` tip-bot2 for Chen Yu 2 siblings, 0 replies; 12+ messages in thread From: tip-bot2 for Chen Yu @ 2024-10-04 10:16 UTC (permalink / raw) To: linux-tip-commits Cc: kernel test robot, K Prateek Nayak, Chen Yu, Peter Zijlstra (Intel), Honglei Wang, x86, linux-kernel The following commit has been merged into the sched/urgent branch of tip: Commit-ID: d4ac164bde7a12ec0a238a7ead5aa26819bbb1c1 Gitweb: https://git.kernel.org/tip/d4ac164bde7a12ec0a238a7ead5aa26819bbb1c1 Author: Chen Yu <yu.c.chen@intel.com> AuthorDate: Wed, 25 Sep 2024 16:54:40 +08:00 Committer: Peter Zijlstra <peterz@infradead.org> CommitterDate: Wed, 02 Oct 2024 11:27:54 +02:00 sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running Commit 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") introduced a mechanism that a wakee with shorter slice could preempt the current running task. It also lower the bar for the current task to be preempted, by checking the rq->nr_running instead of cfs_rq->nr_running when the current task has ran out of time slice. But there is a scenario that is problematic. Say, if there is 1 cfs task and 1 rt task, before 85e511df3cec, update_deadline() will not trigger a reschedule, and after 85e511df3cec, since rq->nr_running is 2 and resched is true, a resched_curr() would happen. Some workloads (like the hackbench reported by lkp) do not like over-scheduling. We can see that the preemption rate has been increased by 2.2%: 1.654e+08 +2.2% 1.69e+08 hackbench.time.involuntary_context_switches Restore its previous check criterion. Fixes: 85e511df3cec ("sched/eevdf: Allow shorter slices to wakeup-preempt") Closes: https://lore.kernel.org/oe-lkp/202409231416.9403c2e9-oliver.sang@intel.com Reported-by: kernel test robot <oliver.sang@intel.com> Suggested-by: K Prateek Nayak <kprateek.nayak@amd.com> Signed-off-by: Chen Yu <yu.c.chen@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Honglei Wang <jameshongleiwang@126.com> Tested-by: K Prateek Nayak <kprateek.nayak@amd.com> Link: https://lore.kernel.org/r/20240925085440.358138-1-yu.c.chen@intel.com --- kernel/sched/fair.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index b63a7ac..ab497fa 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -1247,7 +1247,7 @@ static void update_curr(struct cfs_rq *cfs_rq) account_cfs_rq_runtime(cfs_rq, delta_exec); - if (rq->nr_running == 1) + if (cfs_rq->nr_running == 1) return; if (resched || did_preempt_short(cfs_rq, curr)) { ^ permalink raw reply related [flat|nested] 12+ messages in thread
end of thread, other threads:[~2024-10-04 10:16 UTC | newest] Thread overview: 12+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2024-09-25 8:54 [PATCH v2] sched/eevdf: Fix wakeup-preempt by checking cfs_rq->nr_running Chen Yu 2024-09-27 1:54 ` Honglei Wang 2024-09-27 6:56 ` Chen Yu 2024-09-27 13:38 ` Honglei Wang 2024-09-28 6:28 ` Chen Yu 2024-09-29 13:51 ` Oliver Sang 2024-09-30 2:04 ` Honglei Wang 2024-09-30 2:22 ` Chen Yu 2024-09-30 6:11 ` Honglei Wang 2024-09-30 7:50 ` Peter Zijlstra 2024-09-30 11:01 ` Chen Yu 2024-10-04 10:16 ` [tip: sched/urgent] " tip-bot2 for Chen Yu
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox