Hello, Bart.
On Mon, Apr 09, 2018 at 09:30:27PM +0000, Bart Van Assche wrote:
> On Mon, 2018-04-09 at 11:56 -0700, tj@xxxxxxxxxx wrote:
> > On Mon, Apr 09, 2018 at 05:03:05PM +0000, Bart Van Assche wrote:
> > > exist today in the blk-mq timeout handling code cannot be fixed completely
> > > using RCU only.
> >
> > I really don't think that is that complicated. Let's first confirm
> > the race fix and get to narrowing / closing that window.
>
> Two months ago it was reported for the first time that commit 1d9bd5161ba3
> ("blk-mq: replace timeout synchronization with a RCU and generation based
> scheme") introduces a regression. Since that report nobody has posted a
> patch that fixes all races related to blk-mq timeout handling and that only
The two patches using RCU were posted a long time ago. It was just
that the repro that only you had at the time didn't work anymore so we
couldn't confirm the fix. If we now have a different repro, awesome.
Let's see whether the fix works.
> uses RCU. If you want to continue working on this that's fine with me. But
> since my opinion is that it is impossible to fix these races using RCU only
> I will continue working on an alternative approach. See also "[PATCH]
> blk-mq: Fix a race between resetting the timer and completion handling"
> (https://www.mail-archive.com/linux-block@xxxxxxxxxxxxxxx/msg18089.html).
ISTR discussing that patch earlier. Didn't the RCU based fix get
posted after that discussion?
Thanks.
--
tejun