[igt-dev] [PATCH i-g-t 38/93] tests/i915/gem_ctx_exec: Convert to intel_ctx_t
Zbigniew Kempczyński
zbigniew.kempczynski at intel.com
Mon Jun 14 16:52:23 UTC 2021
On Mon, Jun 14, 2021 at 10:17:05AM -0500, Jason Ekstrand wrote:
> On Mon, Jun 14, 2021 at 7:34 AM Zbigniew Kempczyński
> <zbigniew.kempczynski at intel.com> wrote:
> >
> > On Wed, Jun 09, 2021 at 12:36:21PM -0500, Jason Ekstrand wrote:
> > > Signed-off-by: Jason Ekstrand <jason at jlekstrand.net>
> > > ---
> > > tests/i915/gem_ctx_exec.c | 51 ++++++++++++++++++++++++---------------
> > > 1 file changed, 31 insertions(+), 20 deletions(-)
> > >
> > > diff --git a/tests/i915/gem_ctx_exec.c b/tests/i915/gem_ctx_exec.c
> > > index 6db2d598..fb18bb01 100644
> > > --- a/tests/i915/gem_ctx_exec.c
> > > +++ b/tests/i915/gem_ctx_exec.c
> > > @@ -268,7 +268,7 @@ static void nohangcheck_hostile(int i915)
> > > const struct intel_execution_engine2 *e;
> > > igt_hang_t hang;
> > > int fence = -1;
> > > - uint32_t ctx;
> > > + const intel_ctx_t *ctx;
> > > int err = 0;
> > > int dir;
> > >
> > > @@ -282,12 +282,12 @@ static void nohangcheck_hostile(int i915)
> > > dir = igt_params_open(i915);
> > > igt_require(dir != -1);
> > >
> > > - ctx = gem_context_create(i915);
> > > - hang = igt_allow_hang(i915, ctx, 0);
> > > + ctx = intel_ctx_create_all_physical(i915);
> > > + hang = igt_allow_hang(i915, ctx->id, 0);
> > >
> > > igt_require(__enable_hangcheck(dir, false));
> > >
> > > - ____for_each_physical_engine(i915, ctx, e) {
> > > + for_each_ctx_engine(i915, ctx, e) {
> > > igt_spin_t *spin;
> > > int new;
> > >
> > > @@ -295,7 +295,7 @@ static void nohangcheck_hostile(int i915)
> > > gem_engine_property_printf(i915, e->name,
> > > "preempt_timeout_ms", "%d", 50);
> > >
> > > - spin = __igt_spin_new(i915, ctx,
> > > + spin = __igt_spin_new(i915, .ctx = ctx,
> > > .engine = e->flags,
> > > .flags = (IGT_SPIN_NO_PREEMPTION |
> > > IGT_SPIN_FENCE_OUT));
> > > @@ -316,7 +316,7 @@ static void nohangcheck_hostile(int i915)
> > > fence = tmp;
> > > }
> > > }
> > > - gem_context_destroy(i915, ctx);
> > > + intel_ctx_destroy(i915, ctx);
> > > igt_assert(fence != -1);
> >
> > Here I got no objections.
> >
> > >
> > > if (sync_fence_wait(fence, MSEC_PER_SEC)) { /* 640ms preempt-timeout */
> > > @@ -341,30 +341,38 @@ static void nohangcheck_hostile(int i915)
> > > static void close_race(int i915)
> > > {
> > > const int ncpus = sysconf(_SC_NPROCESSORS_ONLN);
> > > - uint32_t *contexts;
> > > + const intel_ctx_t *base_ctx;
> > > + const intel_ctx_t **ctx;
> > > + uint32_t *ctx_id;
> > > igt_spin_t *spin;
> > >
> > > /* Check we can execute a polling spinner */
> > > - igt_spin_free(i915, igt_spin_new(i915, .flags = IGT_SPIN_POLL_RUN));
> > > + base_ctx = intel_ctx_create(i915, NULL);
> > > + igt_spin_free(i915, igt_spin_new(i915, .ctx = base_ctx,
> > > + .flags = IGT_SPIN_POLL_RUN));
> > >
> > > - contexts = mmap(NULL, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
> > > - igt_assert(contexts != MAP_FAILED);
> > > + ctx = calloc(ncpus, sizeof(*ctx));
> > > + ctx_id = mmap(NULL, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
> > > + igt_assert(ctx_id != MAP_FAILED);
> > >
> > > - for (int child = 0; child < ncpus; child++)
> > > - contexts[child] = gem_context_create(i915);
> > > + for (int child = 0; child < ncpus; child++) {
> > > + ctx[child] = intel_ctx_create(i915, NULL);
> > > + ctx_id[child] = ctx[child]->id;
> > > + }
> > >
> > > igt_fork(child, ncpus) {
> > > - spin = __igt_spin_new(i915, .flags = IGT_SPIN_POLL_RUN);
> > > + spin = __igt_spin_new(i915, .ctx = base_ctx,
> > > + .flags = IGT_SPIN_POLL_RUN);
> > > igt_spin_end(spin);
> > > gem_sync(i915, spin->handle);
> > >
> > > - while (!READ_ONCE(contexts[ncpus])) {
> > > + while (!READ_ONCE(ctx_id[ncpus])) {
> > > int64_t timeout = 1;
> > >
> > > igt_spin_reset(spin);
> > > igt_assert(!igt_spin_has_started(spin));
> > >
> > > - spin->execbuf.rsvd1 = READ_ONCE(contexts[child]);
> > > + spin->execbuf.rsvd1 = READ_ONCE(ctx_id[child]);
> > > if (__gem_execbuf(i915, &spin->execbuf))
> > > continue;
> > >
> > > @@ -404,19 +412,22 @@ static void close_race(int i915)
> > > * and the kernel's context/request handling.
> > > */
> > > for (int child = 0; child < ncpus; child++) {
> > > - gem_context_destroy(i915, contexts[child]);
> > > - contexts[child] = gem_context_create(i915);
> > > + intel_ctx_destroy(i915, ctx[child]);
> > > + ctx[child] = intel_ctx_create(i915, NULL);
> > > + ctx_id[child] = ctx[child]->id;
> > > }
> > > usleep(1000 + hars_petruska_f54_1_random_unsafe() % 2000);
> > > }
> > >
> > > - contexts[ncpus] = 1;
> > > + ctx_id[ncpus] = 1;
> > > igt_waitchildren();
> > >
> > > + intel_ctx_destroy(i915, base_ctx);
> > > for (int child = 0; child < ncpus; child++)
> > > - gem_context_destroy(i915, contexts[child]);
> > > + intel_ctx_destroy(i915, ctx[child]);
> > >
> > > - munmap(contexts, 4096);
> > > + free(ctx);
> > > + munmap(ctx_id, 4096);
> > > }
> >
> > But here I got. What for is all intel_ctx_t usage if we just need
> > to create contexts and use in the code? There's no necessity to
> > walk through engines here so I would stay on ordinary gem_context_create()/
> > gem_context_destroy(). Spinner can run here in ctx0 context.
> > Or I'm missed something?
>
> It's because, at the end of the series, I start requiring an
> intel_ctx_t in the spinner for POLL_RUN. This is to let us avoid some
> context introspection. As for the other contexts that aren't the one
> we're spinning on. I think that was mostly for consistency. I can go
> back to using uint32_t context IDs if you'd like.
I see POLL_RUN requires intel_ctx_t for ALL_ENGINES, not for default engine
so we can live without intel_ctx_t here. But to be consistent with other
code maybe it is better to left as it is.
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski at intel.com>
--
Zbigniew
>
> --Jason
>
>
> > --
> > Zbigniew
> >
> >
> > >
> > > igt_main
> > > --
> > > 2.31.1
> > >
> > > _______________________________________________
> > > igt-dev mailing list
> > > igt-dev at lists.freedesktop.org
> > > https://lists.freedesktop.org/mailman/listinfo/igt-dev
More information about the igt-dev
mailing list