Re: [PATCH 2/2] mlock use lru_add_drain_all_async()

From: KOSAKI Motohiro
Date: Tue Oct 06 2009 - 23:39:02 EST


2009/10/7 Christoph Lameter <cl@xxxxxxxxxxxxxxxxxxxx>:
> On Tue, 6 Oct 2009, KOSAKI Motohiro wrote:
>
>>   Suppose you have 2 cpus, cpu1 is busy doing a SCHED_FIFO-99 while(1),
>>   cpu0 does mlock()->lru_add_drain_all(), which does
>>   schedule_on_each_cpu(), which then waits for all cpus to complete the
>>   work. Except that cpu1, which is busy with the RT task, will never run
>>   keventd until the RT load goes away.
>>
>>   This is not so much an actual deadlock as a serious starvation case.
>>
>> Actually, mlock() doesn't need to wait to finish lru_add_drain_all().
>> Thus, this patch replace it with lru_add_drain_all_async().
>
> Ok so this will queue up lots of events for the cpu doing a RT task. If
> the RT task is continuous then they will be queued there forever?

Yes. this patch solved very specific issue only.
In original bug-report case, the system has two cpuset and the RT task
own one cpuset as monopoly. Thus, your worried thing doesn't occur.

Perhaps, we need complete solution. but I don't think this patch have
bad side effect. then, I hope to push it into mainline.
--
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/