Re: [PATCH] hw/nvme: fix handling of over-committed queues

[Keith Busch]

On Mon, Oct 28, 2024 at 10:01:50AM +0100, Klaus Jensen wrote:
> On Oct 25 10:45, Keith Busch wrote:
> > On Fri, Oct 25, 2024 at 12:50:45PM +0200, Klaus Jensen wrote:
> > > @@ -1520,9 +1520,16 @@ static void nvme_post_cqes(void *opaque)
> > >          nvme_inc_cq_tail(cq);
> > >          nvme_sg_unmap(&req->sg);
> > > +
> > > +        if (QTAILQ_EMPTY(&sq->req_list) && !nvme_sq_empty(sq)) {
> > > +            qemu_bh_schedule(sq->bh);
> > > +        }
> > > +
> > >          QTAILQ_INSERT_TAIL(&sq->req_list, req, entry);
> > >      }
> > 
> > Shouldn't we schedule the bottom half after the req has been added to
> > the list? I think everything the callback needs to be written prior to
> > calling qemu_bh_schedule().
> > 
> 
> Not as far as I know. It is only queued up; it won't be executed
> immediately. It might run next (ASAP) if we are already in a bottom
> half, but not before whatever context we are in returns.

Okay. I was trying to come up with an explanation for why Waldek was
still able to reproduce the problem, and that was all I have so far.