Re: [PATCH 4/4] nohz: Fix iowait overcounting if iowait task migrates

From: Denys Vlasenko
Date: Mon May 05 2014 - 14:06:57 EST


On 04/29/2014 04:47 PM, Frederic Weisbecker wrote:
> On Thu, Apr 24, 2014 at 08:45:58PM +0200, Denys Vlasenko wrote:
>> +void tick_nohz_iowait_to_idle(int cpu)
>> +{
>> + struct tick_sched *ts = tick_get_tick_sched(cpu);
>> + ktime_t now = ktime_get();
>> +
>> + write_seqcount_begin(&ts->idle_sleeptime_seq);
>> + ts->iowait_exittime = now;
>> + write_seqcount_end(&ts->idle_sleeptime_seq);
>
> So now you have two concurrent updaters using the seqcount, which is
> very dangerous as the counters aren't updated atomically.
>
> seqcount is only suitable when there is a single sequential updater.
> Once you deal with concurrent updaters you need seqlock.
>
> And once you add seqlock in the hot scheduler path, you're hitting
> a big scalability issue.

What I need here is merely an atomic store.
The complication is, of course, that, ktime_t is not atomic[64]_t.

How do you think I can do an atomic store?


--
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/