[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANRm+CwGEsjJKnjiMPVOTjPUjYLK2-WFJhVvQ44qOi6SM16RVw@mail.gmail.com>
Date: Wed, 11 May 2016 08:53:05 +0800
From: Wanpeng Li <kernellwp@...il.com>
To: Tomasz Grabiec <tgrabiec@...lladb.com>
Cc: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Ingo Molnar <mingo@...hat.com>,
Peter Zijlstra <peterz@...radead.org>
Subject: Re: PROBLEM: Scheduling latency not respected in CFS with HRTICK enabled
2016-05-06 23:43 GMT+08:00 Tomasz Grabiec <tgrabiec@...lladb.com>:
[...]
> prev_comm=python prev_pid=873 prev_prio=120 prev_state=R ==>
> next_comm=python next_pid=874 next_prio=120
> python 874 [002] 3273334.932741: probe:hrtick: (ffffffff8109b430)
> python 874 [002] 3273334.932742: sched:sched_stat_runtime:
> comm=python pid=874 runtime=503174 [ns] vruntime=1549788942230752 [ns]
hrtick fire after almost 0.5ms(the time slice process 874 get).
> python 874 [002] 3273334.932743: sched:sched_stat_runtime:
> comm=python pid=874 runtime=948 [ns] vruntime=1549788942231700 [ns]
> python 874 [002] 3273334.936240: probe:scheduler_tick: (ffffffff810a1f30)
> python 874 [002] 3273334.936241: sched:sched_stat_runtime:
> comm=python pid=874 runtime=3497530 [ns] vruntime=1549788945729230
> [ns]
continue to run another 3.4ms until next normal scheduler tick fire.
> python 874 [002] 3273334.936249: sched:sched_stat_runtime:
> comm=python pid=874 runtime=8553 [ns] vruntime=1549788945737783 [ns]
> python 874 [002] 3273334.936250: sched:sched_stat_wait:
> comm=python pid=873 delay=4010205 [ns]
> python 874 [002] 3273334.936250: probe:hrtick_start: (ffffffff8109f4f0)
> python 874 [002] 3273334.936251: sched:sched_switch:
> prev_comm=python prev_pid=874 prev_prio=120 prev_state=R ==>
> next_comm=python next_pid=873 next_prio=120
So the process 874 continue to run util next scheduler fick event if
hrtick has already fired. I think Peterz can give any light.
Ping Peterz :-)
Regards,
Wanpeng Li
Powered by blists - more mailing lists