Re: Potential problem with 31e77c93e432dec7 ("sched/fair: Update blocked load when newly idle")

[Geert Uytterhoeven]

Hi Vincent,

On Thu, Apr 26, 2018 at 12:31 PM, Vincent Guittot
<vincent.guittot@xxxxxxxxxx> wrote:
> Le Thursday 26 Apr 2018 Ã 00:56:03 (+0200), Niklas SÃderlund a Ãcrit :
>> Here are the result, sorry for the delay.
>>
>> On 2018-04-23 11:54:20 +0200, Vincent Guittot wrote:
>>
>> [snip]
>>
>> >
>> > Thanks for the report. Can you re run with the following trace-cmd sequence ? My previous sequence disables ftrace events
>> >
>> > trace-cmd reset > /dev/null
>> > trace-cmd start -b 40000 -p function -l dump_backtrace:traceoff -e sched -e cpu_idle -e cpu_frequency -e timer -e ipi -e irq -e printk
>> > trace-cmd start -b 40000 -p function -l dump_backtrace -e sched -e cpu_idle -e cpu_frequency -e timer -e ipi -e irq -e printk
>> >
>> > I have updated the patch and added traces to check that scheduler returns from idle_balance function and doesn't stay stuck
>>
>> Once more I applied the change bellow on-top of c18bb396d3d261eb ("Merge
>> git://git.kernel.org/pub/scm/linux/kernel/git/davem/net").
>>
>> This time the result of 'trace-cmd report' is so large I do not include
>> it here, but I attach the trace.dat file. Not sure why but the timing of
>> sending the NMI to the backtrace print is different (but content the
>> same AFIK) so in the odd change it can help figure this out:
>>
>
> Thanks for the trace, I have been able to catch a problem with it.
> Could you test the patch below to confirm that the problem is solved ?
> The patch apply on-top of
> c18bb396d3d261eb ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net")
>
> From: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> Date: Thu, 26 Apr 2018 12:19:32 +0200
> Subject: [PATCH] sched/fair: fix the update of blocked load when newly idle
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
>
> With commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle"),
> we release the rq->lock when updating blocked load of idle CPUs. This open
> a time window during which another CPU can add a task to this CPU's cfs_rq.
> The check for newly added task of idle_balance() is not in the common path.
> Move the out label to include this check.
>
> Fixes: 31e77c93e432 ("sched/fair: Update blocked load when newly idle")
> Reported-by: Heiner Kallweit <hkallweit1@xxxxxxxxx>
> Reported-by: Niklas SÃderlund <niklas.soderlund@xxxxxxxxxxxx>
> Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>

Thanks!
Tested-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds