Andrew Morton wrote:

>Nick Piggin <piggin@cyberone.com.au> wrote:
>
>>But I'm guessing that you have no requests in flight by the time
>> blk_congestion_wait gets called, so nothing ever gets kicked.
>>
>
>That's why blk_congestion_wait() in -mm propagates the schedule_timeout()
>return value.   You can do:
>
>	if (blk_congestion_wait(...))
>		printk("ouch\n");
>
>If your kernel says ouch much, we have a problem.
>
>

Martin, have you tried adding this printk?

Andrew, could you take the following patch (even though it didn't fix
the problem).

I think the smp_mb isn't needed because the rl waitqueue stuff is
serialised by the queue spinlocks.

The addition of the smp_mb and the other change is to try to close the
window for races a bit. Obviously they can still happen, it's a racy
interface and it doesn't matter much.