[Intel-gfx] [PATCH 2/7] drm/i915/selftests: Exercise timeslice rewinding

[Chris Wilson]

Quoting Mika Kuoppala (2020-02-11 14:50:08)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> > +             /* Release the hounds! */
> > +             slot[0] = 1;
> > +             wmb();
> > +
> > +             for (i = 1; i <= 3; i++) {
> > +                     unsigned long timeout = jiffies + HZ / 2;
> > +
> > +                     while (!READ_ONCE(slot[i]) &&
> > +                            time_before(jiffies, timeout))
> 
> you pushed with wmb so you could expect with rmb() and cpu_relax();
> I guess it works fine without :O.

The wmb() "pairs" with GPU; just paranoia.

> > +                             ;
> > +
> > +                     if (!time_before(jiffies, timeout)) {
> > +                             pr_err("%s: rq[%d] timed out\n",
> > +                                    engine->name, i - 1);
> > +                             err = -ETIME;
> > +                             goto err;
> > +                     }
> > +
> > +                     pr_debug("%s: slot[%d]:%x\n", engine->name, i, slot[i]);
> > +             }
> > +
> > +             /* XZY: XZ < XY */
> > +             if (slot[3] - slot[1] >= slot[2] - slot[1]) {
> > +                     pr_err("%s: timeslicing did not run context B [%u] before A [%u]!\n",
> > +                            engine->name,
> > +                            slot[3] - slot[1],
> > +                            slot[2] - slot[1]);
> > +                     err = -EINVAL;
> > +             }
> > +
> > +err:
> > +             memset(slot, 0xff, 4 * sizeof(u32));
> 
> was expecting slot[0] = 
memset32(&slot[0], -1, 4); /* weirdo */
-Chris