[PATCH I-G-T v5] tests/intel/i915_hangman: Fix spin engine hang

Umesh Nerlige Ramappa umesh.nerlige.ramappa at intel.com
Thu Dec 19 21:19:01 UTC 2024


On Tue, Dec 17, 2024 at 02:45:53PM +0100, Mikolaj Wasiak wrote:
>On DG2 when RCS engine is active, the CCS engine freezes which leads to
>hangman test failures.
>This patch disables RCS and CCS background spins when testing any engine
>on hangman.
>
>Signed-off-by: Mikolaj Wasiak <mikolaj.wasiak at intel.com>
>---
> tests/intel/i915_hangman.c | 14 ++++++++++++++
> 1 file changed, 14 insertions(+)
>
>diff --git a/tests/intel/i915_hangman.c b/tests/intel/i915_hangman.c
>index a556eec53..6f1fdc788 100644
>--- a/tests/intel/i915_hangman.c
>+++ b/tests/intel/i915_hangman.c
>@@ -365,6 +365,12 @@ static void chk_err(int *dst, int err, int expected)
> #define ERR_FENCE_END  3
> #define ERR_FENCE_STAT 4
>
>+static bool is_rcs_ccs(const struct intel_execution_engine2 *e)
>+{
>+	return e->class == I915_ENGINE_CLASS_COMPUTE ||
>+	       e->class == I915_ENGINE_CLASS_RENDER;
>+}
>+
> static void
> test_engine_hang(const intel_ctx_t *ctx,
> 		 const struct intel_execution_engine2 *e, unsigned int flags)
>@@ -388,6 +394,14 @@ test_engine_hang(const intel_ctx_t *ctx,
> 	 */
> 	num_ctx = 0;
> 	for_each_ctx_engine(device, ctx, other) {
>+		/* Due to Wa_14014494547:DG2 we can't run RCS and CCS
>+		 * in parallel. We still want background spin of the
>+		 * same type as hanged spin to be run.
>+		 */

Ideally this should be (IS_DG2 && is_rcs_ccs(e) && is_rcs_ccs(other)), 
but if this is working for you, then

Acked-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa at intel.com>

>+		if (IS_DG2(intel_get_drm_devid(device)) && 
>is_rcs_ccs(other) &&
>+		    other->class != e->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);
>-- 
>2.47.1
>


More information about the igt-dev mailing list