[Intel-gfx] [PATCH i-g-t] i915/perf_pmu: Exercise I915_PMU_SOFTWARE_GT_AWAKE_TIME

Tvrtko Ursulin tvrtko.ursulin at linux.intel.com
Thu Dec 17 17:33:05 UTC 2020


On 17/12/2020 17:26, Chris Wilson wrote:
> Measure the sample gt-awake time while each engine and every engine is
> busy. They should all report the same duration, the elapsed runtime of
> the batch.
> 
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com>
> ---
>   include/drm-uapi/i915_drm.h |  3 ++-
>   tests/i915/perf_pmu.c       | 42 +++++++++++++++++++++++++++++++++++++
>   2 files changed, 44 insertions(+), 1 deletion(-)
> 
> diff --git a/include/drm-uapi/i915_drm.h b/include/drm-uapi/i915_drm.h
> index ef696d1a5..bf9ea471c 100644
> --- a/include/drm-uapi/i915_drm.h
> +++ b/include/drm-uapi/i915_drm.h
> @@ -177,8 +177,9 @@ enum drm_i915_pmu_engine_sample {
>   #define I915_PMU_REQUESTED_FREQUENCY	__I915_PMU_OTHER(1)
>   #define I915_PMU_INTERRUPTS		__I915_PMU_OTHER(2)
>   #define I915_PMU_RC6_RESIDENCY		__I915_PMU_OTHER(3)
> +#define I915_PMU_SOFTWARE_GT_AWAKE_TIME	__I915_PMU_OTHER(4)
>   
> -#define I915_PMU_LAST I915_PMU_RC6_RESIDENCY
> +#define I915_PMU_LAST /* Deprecated - do not use */ I915_PMU_RC6_RESIDENCY
>   
>   /* Each region is a minimum of 16k, and there are at most 255 of them.
>    */
> diff --git a/tests/i915/perf_pmu.c b/tests/i915/perf_pmu.c
> index c2b186d39..90182074b 100644
> --- a/tests/i915/perf_pmu.c
> +++ b/tests/i915/perf_pmu.c
> @@ -883,6 +883,42 @@ sema_busy(int gem_fd,
>   	close(fd[1]);
>   }
>   
> +static void test_awake(int i915)
> +{
> +	const struct intel_execution_engine2 *e;
> +	unsigned long slept;
> +	uint64_t val;
> +	int fd;
> +
> +	fd = perf_i915_open(i915, I915_PMU_SOFTWARE_GT_AWAKE_TIME);
> +	igt_skip_on(fd < 0);
> +
> +	/* Check that each engine is captured by the GT wakeref */
> +	__for_each_physical_engine(i915, e) {
> +		igt_spin_new(i915, .engine = e->flags);
> +
> +		val = pmu_read_single(fd);
> +		slept = measured_usleep(batch_duration_ns / 1000);
> +		val = pmu_read_single(fd) - val;
> +
> +		gem_quiescent_gpu(i915);
> +		assert_within_epsilon(val, slept, tolerance);
> +	}
> +
> +	/* And that the total GT wakeref matches walltime not summation */
> +	__for_each_physical_engine(i915, e)
> +		igt_spin_new(i915, .engine = e->flags);
> +
> +	val = pmu_read_single(fd);
> +	slept = measured_usleep(batch_duration_ns / 1000);
> +	val = pmu_read_single(fd) - val;
> +
> +	gem_quiescent_gpu(i915);
> +	assert_within_epsilon(val, slept, tolerance);
> +

Erm who terminates all those spinners? :)

Regards,

Tvrtko

> +	close(fd);
> +}
> +
>   #define   MI_WAIT_FOR_PIPE_C_VBLANK (1<<21)
>   #define   MI_WAIT_FOR_PIPE_B_VBLANK (1<<11)
>   #define   MI_WAIT_FOR_PIPE_A_VBLANK (1<<3)
> @@ -2250,6 +2286,12 @@ igt_main
>   	igt_subtest("rc6-suspend")
>   		test_rc6(fd, TEST_S3);
>   
> +	/**
> +	 * Test GT wakeref tracking (similar to RC0, opposite of RC6)
> +	 */
> +	igt_subtest("gt-awake")
> +		test_awake(fd);
> +
>   	/**
>   	 * Check render nodes are counted.
>   	 */
> 


More information about the Intel-gfx mailing list