Re: [PATCH] sched: tg_set_cfs_bandwidth() causes rq->lock deadlock

[bsegall]

Peter Zijlstra <peterz@xxxxxxxxxxxxx> writes:

> On Fri, May 16, 2014 at 12:38:21PM +0400, Roman Gushchin wrote:
>
>> I still think, there is a deadlock. I'll try to explain.
>> Three CPUs must be involved:
>> CPU0				CPU1				CPU2
>> take rq->lock			period timer fired		
>> ...				take cfs_b lock
>> ...				...				tg_set_cfs_bandwidth()
>> throttle_cfs_rq()		release cfs_b lock		take cfs_b lock
>> ...				distribute_cfs_runtime()	timer_active = 0
>> take cfs_b->lock		wait for rq->lock		...
>> __start_cfs_bandwidth()	
>> {wait for timer callback
>>  break if timer_active == 1}
>> 
>> So, CPU0 and CPU1 are deadlocked.
>
> OK, so can someone explain this ->timer_active thing? esp. what's the
> 'obvious' difference with hrtimer_active()?

timer_active is only changed or checked under cfs_b lock, so that we can
be certain that the timer is active whenever runtime isn't full. We
can't use hrtimer_active because the timer might have unlocked the cfs_b
on another cpu and be about to return HRTIMER_NORESTART.
>
>
> Ideally we'd change the lot to not have this, but if we have to keep it
> we'll need to make it lockdep visible because all this stinks
--
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/