[Intel-gfx] [PATCH 08/28] drm/i915/gt: Show all active timelines for debugging
Chris Wilson
chris at chris-wilson.co.uk
Thu Nov 19 10:47:27 UTC 2020
Quoting Tvrtko Ursulin (2020-11-18 15:51:41)
>
> On 17/11/2020 13:25, Chris Wilson wrote:
> > Quoting Tvrtko Ursulin (2020-11-17 12:59:44)
> >>
> >> On 17/11/2020 11:30, Chris Wilson wrote:
> >>> + if (show_request) {
> >>> + list_for_each_entry_safe(rq, rn, &tl->requests, link)
> >>> + show_request(m, rq,
> >>> + i915_request_is_active(rq) ? " E" :
> >>> + i915_request_is_ready(rq) ? " Q" :
> >>> + " U");
> >>
> >> Can we get some consistency between the category counts and flags.
> >>
> >> s/count/queued/ -> Q
> >
> > Hmm, if you are sure. Q would then not match with the engine info.
>
> Sure? Not really. What do we have there? You mean "!/*/+/-" flags? Or
> "E/Q/V" from intel_execlists_show_requests? Right, 'Q' there means
> runnable and it doesn't show queued at all. Yes, why not change everything.
>
> > Still favouring count over queued; I think count indicates more clearly
> > that it is the superset, but queued may imply it excludes ready and
> > definitely sounds like it should not include inflight.
>
> I am okay with that.
>
> >> ready -> R (also matches with term runnable)
> >> active -> E ? hmmm E is consistent with the engine info dump.
> >>
> >> Not ideal but perhaps every bit of more consistency is good.
> >
> > Not sold yet, but not happy with the current flags either.
> >
> > If we go with 'R' for ready, we should also update engine info.
>
> Okay we seem to have plenty of options.
>
> U or Q - queued/unready
> R or Q - ready/queued (to backend) (Rv/Qv for virtual?)
> E or R, or I - executing/running/in-flight
>
> Q -> R -> E
> U -> R -> E
> U -> Q -> E/R/I
> U -> R -> E/I
>
> I don't know.. either one as long as all places use the same.
URE.
Unready -> ready -> executing.
Unready -> ready is a good pairing, and executing over inflight to avoid
being confused for an infection.
And unready opens up a plethora of jokes by just dropping the 'y'.
-Chris
More information about the Intel-gfx
mailing list