[Intel-gfx] [PATCH] drm/i915/selftests: Update live.evict to wait on requests / idle GPU after each loop

[Thomas Hellström]

On 10/21/21 23:40, Matthew Brost wrote:
> Update live.evict to wait on last request and idle GPU after each loop.
> This not only enhances the test to fill the GGTT on each engine class
> but also avoid timeouts from igt_flush_test when using GuC submission.
> igt_flush_test (idle GPU) can take a long time with GuC submission if
> losts of contexts are created due to H2G / G2H required to destroy
> contexts.
>
> Signed-off-by: Matthew Brost <matthew.brost at intel.com>

LGTM,

Reviewed-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>


> ---
>   .../gpu/drm/i915/selftests/i915_gem_evict.c   | 19 +++++++++++++++++++
>   1 file changed, 19 insertions(+)
>
> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> index f99bb0113726..7e0658a77659 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> @@ -442,6 +442,7 @@ static int igt_evict_contexts(void *arg)
>   	/* Overfill the GGTT with context objects and so try to evict one. */
>   	for_each_engine(engine, gt, id) {
>   		struct i915_sw_fence fence;
> +		struct i915_request *last = NULL;
>   
>   		count = 0;
>   		onstack_fence_init(&fence);
> @@ -479,6 +480,9 @@ static int igt_evict_contexts(void *arg)
>   
>   			i915_request_add(rq);
>   			count++;
> +			if (last)
> +				i915_request_put(last);
> +			last = i915_request_get(rq);
>   			err = 0;
>   		} while(1);
>   		onstack_fence_fini(&fence);
> @@ -486,6 +490,21 @@ static int igt_evict_contexts(void *arg)
>   			count, engine->name);
>   		if (err)
>   			break;
> +		if (last) {
> +			if (i915_request_wait(last, 0, HZ) < 0) {
> +				err = -EIO;
> +				i915_request_put(last);
> +				pr_err("Failed waiting for last request (on %s)",
> +				       engine->name);
> +				break;
> +			}
> +			i915_request_put(last);
> +		}
> +		err = intel_gt_wait_for_idle(engine->gt, HZ * 3);
> +		if (err) {
> +			pr_err("Failed to idle GT (on %s)", engine->name);
> +			break;
> +		}
>   	}
>   
>   	mutex_lock(&ggtt->vm.mutex);