Re: blk_congestion_wait racy?

[Nick Piggin]

Martin Schwidefsky wrote:

> Hi Nick,
>
>
> > Another problem is that if there are no requests anywhere in the system,
> > sleepers in blk_congestion_wait will not get kicked. blk_congestion_wait
> > could probably have blk_run_queues moved after prepare_to_wait, which
> > might help.
> I tried putting blk_run_queues after prepare_to_wait, it worked but it
> didn't help. The test still needs close to a minute.

OK. This was *with* the memory barrier changes too, was it? Not that
they should make that much difference. The test is still racy, but
the window just gets smaller.

But I'm guessing that you have no requests in flight by the time
blk_congestion_wait gets called, so nothing ever gets kicked.

I prefer something more like this model: if 'current' submits a request
to a congested queue then it gets put on the congestion waitqueue.
You can then run blk_congestion_wait afterwards and it won't block if
the queue you've written to has come out of congestion at any time.

This also means that you can (should, in fact) stop uncongested queues
from waking up the waiters every time they complete a request. Hmm, I
like it.

-
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/