On Thu, 2017-12-14 at 21:20 +0100, Peter Zijlstra wrote:
> On Thu, Dec 14, 2017 at 06:51:11PM +0000, Bart Van Assche wrote:
> > On Tue, 2017-12-12 at 11:01 -0800, Tejun Heo wrote:
> > > + write_seqcount_begin(&rq->gstate_seq);
> > > + blk_mq_rq_update_state(rq, MQ_RQ_IN_FLIGHT);
> > > + blk_add_timer(rq);
> > > + write_seqcount_end(&rq->gstate_seq);
> >
> > My understanding is that both write_seqcount_begin() and
> > write_seqcount_end()
> > trigger a write memory barrier. Is a seqcount really faster than a spinlock?
>
> Yes lots, no atomic operations and no waiting.
>
> The only constraint for write_seqlock is that there must not be any
> concurrency.
>
> But now that I look at this again, TJ, why can't the below happen?
>
> write_seqlock_begin();
> blk_mq_rq_update_state(rq, IN_FLIGHT);
> blk_add_timer(rq);
> <timer-irq>
> read_seqcount_begin()
> while (seq & 1)
> cpurelax();
> // life-lock
> </timer-irq>
> write_seqlock_end();
Hello Peter,
Some time ago the block layer was changed to handle timeouts in thread context
instead of interrupt context. See also commit 287922eb0b18 ("block: defer
timeouts to a workqueue").
Bart.
