Re: [RFC v2 5/5] block: revert back to synchronous request_queue removal

[Luis Chamberlain <mcgrof@kernel.org>]

On Sat, Apr 11, 2020 at 04:21:17PM -0700, Bart Van Assche wrote:
> On 2020-04-10 14:27, Luis Chamberlain wrote:
> > On Fri, Apr 10, 2020 at 2:50 PM Luis Chamberlain <mcgrof@kernel.org> wrote:
> >>
> >> On Fri, Apr 10, 2020 at 8:34 AM Luis Chamberlain <mcgrof@kernel.org> wrote:
> >>> On Thu, Apr 09, 2020 at 08:12:21PM -0700, Bart Van Assche wrote:
> >>>> Please add a might_sleep() call in blk_put_queue() since with this patch
> >>>> applied it is no longer allowed to call blk_put_queue() from atomic context.
> >>>
> >>> Sure thing.
> >>
> >> On second though, I don't think blk_put_queue() would be the right
> >> place for might_sleep(), given we really only care about the *last*
> >> refcount decrement to 0. So I'll move it to blk_release_queue().
> >> Granted, at that point we are too late, and we'd get a splat about
> >> this issue *iff* we really sleep. So yeah, I do suppose that forcing
> >> this check there still makes sense.
> > 
> > I'll add might_sleep() to both blk_release_queue() *and* blk_cleanup_queue().
> 
> Since there is already an unconditional mutex_lock() call in
> blk_cleanup_queue(), do we really need to add a might_sleep() call in
> blk_cleanup_queue()?

You are right, mutex_lock() already has a might_sleep() sprinkled on it.

  Luis