[igt-dev] [PATCH i-g-t] tests/i915/i915_hangman: add DG2 wa for hang subtest

[Kamil Konieczny]

Hi Umesh,

On 2022-07-13 at 16:30:04 -0700, Umesh Nerlige Ramappa wrote:
> On Wed, Jul 13, 2022 at 05:48:16PM +0200, Kamil Konieczny wrote:
> > DG2 requires a workaround that prevents context switches on CCS when
> > RCS is not idle. That breaks the hangman test because a hang on RCS
> > means CCS cannot switch. Thus, the well behaved background task on CCS
> > is seen as hanging and thus the test fails. The same problem is seen
> > on RCS when CCS is hang, so do not use the other one when causing hang.
> > 
> > Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5884
> > Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa at intel.com>
> > Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
> > Signed-off-by: Kamil Konieczny <kamil.konieczny at linux.intel.com>
> > ---
> > tests/i915/i915_hangman.c | 9 +++++++++
> > 1 file changed, 9 insertions(+)
> > 
> > diff --git a/tests/i915/i915_hangman.c b/tests/i915/i915_hangman.c
> > index 687dee14..9e244a10 100644
> > --- a/tests/i915/i915_hangman.c
> > +++ b/tests/i915/i915_hangman.c
> > @@ -304,11 +304,17 @@ test_engine_hang(const intel_ctx_t *ctx,
> > 	igt_spin_t *spin, *next;
> > 	IGT_LIST_HEAD(list);
> > 	uint64_t ahnd = get_reloc_ahnd(device, ctx->id), ahndN;
> > +	int skip_class = -1;
> > 	int num_ctx;
> > 
> > 	igt_skip_on(flags & IGT_SPIN_INVALID_CS &&
> > 		    gem_engine_has_cmdparser(device, &ctx->cfg, e->flags));
> > 
> > +	/* Wa_14014494547:DG2 hang on RCS means CCS cannot switch */
> > +	if (IS_DG2(intel_get_drm_devid(device)))
> > +		skip_class = e->class == I915_ENGINE_CLASS_RENDER ? I915_ENGINE_CLASS_COMPUTE :
> > +					 I915_ENGINE_CLASS_RENDER;
> > +
> 
> This works, but also has a side effect of skipping background batch on
> render when testing media engines. Based on the commit msg, I think you want
> to skip render for compute specific hangs only.

Thank you for replay, I need to debug this more. 
I checked this on different dg2 with different drm-tip build
and it fails.

Regards,
Kamil

> > 	/*
> > 	 * Fill all engines with background load.
> > 	 * This verifies that independent engines are unaffected and gives
> > @@ -317,6 +323,9 @@ test_engine_hang(const intel_ctx_t *ctx,
> > 	 */
> > 	num_ctx = 0;
> > 	for_each_ctx_engine(device, ctx, other) {
> > +		if (other->class == skip_class)
> > +			continue;
> > +
> > 		local_ctx[num_ctx] = intel_ctx_create(device, &ctx->cfg);
> > 		context_unban(device, local_ctx[num_ctx]->id);
> > 		ahndN = get_reloc_ahnd(device, local_ctx[num_ctx]->id);
> > -- 
> > 2.34.1
> >