[Intel-gfx] [PATCH 03/14] drm/i915/execlists: Flush the tasklet on parking

[Chris Wilson]

Quoting Tvrtko Ursulin (2019-05-02 15:14:08)
> 
> On 02/05/2019 14:53, Chris Wilson wrote:
> > Quoting Tvrtko Ursulin (2019-05-02 14:48:18)
> >>
> >> On 01/05/2019 12:45, Chris Wilson wrote:
> >>> Tidy up the cleanup sequence by always ensure that the tasklet is
> >>> flushed on parking (before we cleanup). The parking provides a
> >>> convenient point to ensure that the backend is truly idle.
> >>>
> >>> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> >>> ---
> >>>    drivers/gpu/drm/i915/gt/intel_lrc.c         | 7 ++++++-
> >>>    drivers/gpu/drm/i915/intel_guc_submission.c | 1 +
> >>>    2 files changed, 7 insertions(+), 1 deletion(-)
> >>>
> >>> diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
> >>> index 851e62ddcb87..7be54b868d8e 100644
> >>> --- a/drivers/gpu/drm/i915/gt/intel_lrc.c
> >>> +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
> >>> @@ -2331,6 +2331,11 @@ static int gen8_init_rcs_context(struct i915_request *rq)
> >>>        return i915_gem_render_state_emit(rq);
> >>>    }
> >>>    
> >>> +static void execlists_park(struct intel_engine_cs *engine)
> >>> +{
> >>> +     tasklet_kill(&engine->execlists.tasklet);
> >>
> >> Isn't it actually a problem if tasklet is scheduled and unstarted, or
> >> even in progress at the point of engine getting parked?
> > 
> > That would be a broken driver. :|
> > 
> > We must be quite sure that engine isn't going to send an interrupt as we
> > are just about to drop the wakeref we need to service that interrupt.
> > 
> > tasklet_kill()
> > GEM_BUG_ON(engine->execlists.active);
> 
> Or instead of both:
> 
> /* Tasklet must not be running or scheduled at this point. */
> GEM_BUG_ON(engine->execlists.tasklet.state);

There's the dilemma that we start parking based on retirement not
final CS event.
-Chris