[Intel-gfx] [PATCH 01/28] drm/i915: Adjust the sentinel assert to match implementation

Chris Wilson chris at chris-wilson.co.uk
Tue Jun 9 10:29:10 UTC 2020


Quoting Tvrtko Ursulin (2020-06-09 07:59:27)
> 666
> On 08/06/2020 10:33, Chris Wilson wrote:
> > Quoting Tvrtko Ursulin (2020-06-08 08:44:01)
> >>
> >> On 07/06/2020 23:20, Chris Wilson wrote:
> >>> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> >>>
> >>> Sentinels are supposed to be last reqeusts in the elsp queue, not the
> >>> only one, so adjust the assert accordingly.
> >>>
> >>> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> >>> ---
> >>>    drivers/gpu/drm/i915/gt/intel_lrc.c | 14 +++-----------
> >>>    1 file changed, 3 insertions(+), 11 deletions(-)
> >>>
> >>> diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
> >>> index d55a5e0466e5..db8a170b0e5c 100644
> >>> --- a/drivers/gpu/drm/i915/gt/intel_lrc.c
> >>> +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
> >>> @@ -1635,9 +1635,9 @@ assert_pending_valid(const struct intel_engine_execlists *execlists,
> >>>                ccid = ce->lrc.ccid;
> >>>    
> >>>                /*
> >>> -              * Sentinels are supposed to be lonely so they flush the
> >>> -              * current exection off the HW. Check that they are the
> >>> -              * only request in the pending submission.
> >>> +              * Sentinels are supposed to be the last request so they flush
> >>> +              * the current exection off the HW. Check that they are the only
> >>> +              * request in the pending submission.
> >>>                 */
> >>>                if (sentinel) {
> >>>                        GEM_TRACE_ERR("%s: context:%llx after sentinel in pending[%zd]\n",
> >>> @@ -1646,15 +1646,7 @@ assert_pending_valid(const struct intel_engine_execlists *execlists,
> >>>                                      port - execlists->pending);
> >>>                        return false;
> >>>                }
> >>> -
> >>>                sentinel = i915_request_has_sentinel(rq);
> >>
> >> FWIW I was changing it to "sentinel |= ..." so it keeps working if we
> >> decide to use more than 2 elsp ports on Icelake one day.
> > 
> > But it will always fail on the next port...
> 
> I don't follow. Sentinel has to be last so if it fails on the next port 
> it is correct to do so, no?

Exactly. We only check the first port after setting sentinel, if that
port is occupied we fail. Hence why we don't need |=, since there is no
continuation.
-Chris


More information about the Intel-gfx mailing list