Re: [PATCH] sched: update_rq_clock() must skip ONE update

From: Linus Torvalds
Date: Sun Mar 30 2014 - 20:13:04 EST


The patch looks fine, but the changelog is so chock-full of odd
language that I don't know what to do with the patch.

Is this actually a problem in real life, or just in the drug-induced
wonderland that Mike was in when writing the changelog?

Linus

On Sun, Mar 30, 2014 at 12:24 AM, Mike Galbraith
<umgwanakikbuti@xxxxxxxxx> wrote:
> NEWSFLASH: Wimpy latent scheduler buglet turns axe-murderer...
>
> Root cause being us staying in kernel for weeks at a time during boot:
>
> Subverting ftrace, we see below that updates _can_ be skipped for AGES,
> which can make very bad things happen. It's supposed to skip ONE, as an
> optimization. not lord knows how many. Here we see we've been skipping
> since 4915630446, resulting in the watchdog being credited with a 137ms
> delta_exec. Bogus, and..
>
> modprobe-134 [000] 5.027903: update_rq_clock <-scheduler_tick
> modprobe-134 [000] 5.037857: update_rq_clock <-scheduler_tick
> modprobe-134 [000] 5.037864: update_rq_clock <-enqueue_task
> modprobe-134 [000] 5.037866: enqueue_task_watchdog: WATCHDOG/0 enqueue 4915630446
> modprobe-134 [000] 5.044965: update_rq_clock <-scheduler_tick
> modprobe-134 [000] 5.046275: update_rq_clock <-scheduler_tick
> modprobe-134 [000] 5.050274: update_rq_clock <-scheduler_tick
> modprobe-134 [000] 5.052862: update_rq_clock <-dequeue_task
> modprobe-134 [000] 5.052866: pick_next_task_watchdog: WATCHDOG/0 select 4915630446
> watchdog/0-7 [000] 5.052868: finish_task_switch: WATCHDOG/0 runs 5052867644
> watchdog/0-7 [000] 5.052869: watchdog: WATCHDOG/0 DELAYED 137238048 max: 137238048
> watchdog/0-7 [000] 5.052872: update_rq_clock <-dequeue_task
> watchdog/0-7 [000] 5.052873: dequeue_task_watchdog: WATCHDOG/0 dequeue 5052872018
> watchdog/0-7 [000] 5.052874: put_prev_task_watchdog: WATCHDOG/0 STOP trace : delta_exec 137241572 sum_exec 137254287
>
> ..delta_exec was sufficient in a NOPREEMPT kernel on zillion disk box
> that disk discovery combined with swsusp wanting to find out if it needs
> to resume, that the featherweight watchdog was credited with delta_exec
> HUGE enough to be incapacitated by the rt throttle long enough for the
> hard lockup detector to trigger.
>
> Throttled watchdog is bad juju, led to -ENOBOOT for IO beast.
>
> Cc: <stable@xxxxxxxxxxxxxxx>
> Signed-off-by: Mike Galbraith <umgwanakikbuti@xxxxxxxxx>
> ---
> kernel/sched/core.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -118,7 +118,7 @@ void update_rq_clock(struct rq *rq)
> {
> s64 delta;
>
> - if (rq->skip_clock_update > 0)
> + if (rq->skip_clock_update-- > 0)
> return;
>
> delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/