[Intel-gfx] [PATCH] drm/i915/gvt: Remove bogus retry around i915_wait_request

[Chris Wilson]

On Mon, Mar 13, 2017 at 09:21:58AM +0000, Dong, Chuanxiao wrote:
> 
> 
> > -----Original Message-----
> > From: intel-gvt-dev [mailto:intel-gvt-dev-bounces at lists.freedesktop.org] On
> > Behalf Of Dong, Chuanxiao
> > Sent: Thursday, March 9, 2017 9:28 AM
> > To: Chris Wilson <chris at chris-wilson.co.uk>; intel-gfx at lists.freedesktop.org
> > Cc: intel-gvt-dev at lists.freedesktop.org; Wang, Zhi A
> > <zhi.a.wang at intel.com>; Zhenyu Wang <zhenyuw at linux.intel.com>
> > Subject: RE: [PATCH] drm/i915/gvt: Remove bogus retry around
> > i915_wait_request
> > 
> > 
> > 
> > > -----Original Message-----
> > > From: Chris Wilson [mailto:chris.ickle.wilson at gmail.com] On Behalf Of
> > > Chris Wilson
> > > Sent: Thursday, March 9, 2017 6:08 AM
> > > To: intel-gfx at lists.freedesktop.org
> > > Cc: intel-gvt-dev at lists.freedesktop.org; Chris Wilson <chris at chris-
> > > wilson.co.uk>; Dong, Chuanxiao <chuanxiao.dong at intel.com>; Zhenyu
> > Wang
> > > <zhenyuw at linux.intel.com>; Wang, Zhi A <zhi.a.wang at intel.com>
> > > Subject: [PATCH] drm/i915/gvt: Remove bogus retry around
> > > i915_wait_request
> > >
> > > commit 8f1117abb408 ("drm/i915/gvt: handle workload lifecycle
> > > properly") includes some nonsense to retry a indefinite wait -
> > > i915_wait_request() does not return until the request is completed
> > > when used from an uninterruptible context.
> > >
> > > Fixes: 8f1117abb408 ("drm/i915/gvt: handle workload lifecycle properly"
> > > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > > Cc: Chuanxiao Dong <chuanxiao.dong at intel.com>
> > > Cc: Zhenyu Wang <zhenyuw at linux.intel.com>
> > > Cc: Zhi Wang <zhi.a.wang at intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/gvt/scheduler.c | 14 +-------------
> > >  1 file changed, 1 insertion(+), 13 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c
> > > b/drivers/gpu/drm/i915/gvt/scheduler.c
> > > index cfd9f7a38c27..8024904c0d58 100644
> > > --- a/drivers/gpu/drm/i915/gvt/scheduler.c
> > > +++ b/drivers/gpu/drm/i915/gvt/scheduler.c
> > > @@ -465,19 +465,7 @@ static int workload_thread(void *priv)
> > >
> > >  		gvt_dbg_sched("ring id %d wait workload %p\n",
> > >  				workload->ring_id, workload);
> > > -retry:
> > > -		i915_wait_request(workload->req,
> > > -					 0, MAX_SCHEDULE_TIMEOUT);
> > > -		/* I915 has replay mechanism and a request will be replayed
> > > -		 * if there is i915 reset. So the seqno will be updated anyway.
> > > -		 * If the seqno is not updated yet after waiting, which means
> > > -		 * the replay may still be in progress and we can wait again.
> > > -		 */
> > > -		if (!i915_gem_request_completed(workload->req)) {
> > > -			gvt_dbg_sched("workload %p not completed, wait
> > > again\n",
> > > -					workload);
> > > -			goto retry;
> > > -		}
> > > +		i915_wait_request(workload->req, 0,
> > > MAX_SCHEDULE_TIMEOUT);
> > 
> > Thanks Chris for pointing this out. So per the comment, i915_wait_request
> > will make sure to return until the request completed, even if a GPU hang
> > happened and replay occurred during i915_wait_request(), this function will
> > only return after the replay completed, right?
> 
> Hi Chris, ping for the above question. I guess the answer is "yes"?

Yes. Given MAX_SCHEDULE_TIMEOUT, no INTERRUPTIBLE flags, it will only
return once the request is completed - that is the hw breadcrumb is
advanced past the request seqno. That includes waiting across a GPU
reset. It is not allowed to return another error.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre