[igt-dev] [PATCH i-g-t v2 2/2] tests/i915_pm_rpm: use device coherent mapping instead of mmap_gtt

[Dixit, Ashutosh]

On Fri, 07 Feb 2020 05:45:27 -0800, Ramalingam C wrote:
>
> Since on new discrete GPUs we dont have the mappable aperture, if that
> is acceptable for the test purpose, we should use GEM_MMAP_OFFSET.
> Hence using gem_mmap_device_coherent() which wraps the mmap options in
> the order of gem_mmap_offset / gem_mmap / mmap_gtt
>
> Incase of fencing/gtt related tests, we mandate the existance of the
> mmap_gtt.
>
> v2:
>   3 tests are skipped based on mappable aperture and tiling support
>
> Signed-off-by: Ramalingam C <ramalingam.c at intel.com>
> ---
>  tests/i915/i915_pm_rpm.c | 26 +++++++++++++++++++-------
>  1 file changed, 19 insertions(+), 7 deletions(-)
>
> diff --git a/tests/i915/i915_pm_rpm.c b/tests/i915/i915_pm_rpm.c
> index 3d7343240802..5ad2d833d53b 100644
> --- a/tests/i915/i915_pm_rpm.c
> +++ b/tests/i915/i915_pm_rpm.c
> @@ -1535,7 +1535,7 @@ static void fill_igt_fb(struct igt_fb *fb, uint32_t color)
>	int i;
>	uint32_t *ptr;
>
> -	ptr = gem_mmap__gtt(drm_fd, fb->gem_handle, fb->size, PROT_WRITE);
> +	ptr = gem_mmap__device_coherent(drm_fd, fb->gem_handle, 0, fb->size, PROT_WRITE);

My earlier comment on this is still valid: even though fill_igt_fb() is
called with tiling set, I think it is still ok > to do this since all we
are doing is filling a solid color for which tiling is immaterial.

cursor_subtest: fix set_tiling() call.

>	for (i = 0; i < fb->size/sizeof(uint32_t); i++)
>		ptr[i] = color;
>	igt_assert(munmap(ptr, fb->size) == 0);
> @@ -2016,16 +2016,20 @@ igt_main_args("", long_options, help_str, opt_handler, NULL)
>	/* GEM */
>	igt_subtest("gem-mmap-cpu")
>		gem_mmap_subtest(false);
> -	igt_subtest("gem-mmap-gtt")
> +	igt_subtest("gem-mmap-gtt") {
> +		gem_require_mappable_ggtt(drm_fd);

OK.

>		gem_mmap_subtest(true);
> +	}
>	igt_subtest("gem-pread")
>		gem_pread_subtest();
>	igt_subtest("gem-execbuf")
>		gem_execbuf_subtest();
>	igt_subtest("gem-idle")
>		gem_idle_subtest();
> -	igt_subtest("gem-evict-pwrite")
> +	igt_subtest("gem-evict-pwrite") {
> +		gem_require_mappable_ggtt(drm_fd);

Sorry I think I changed my mind on this: for this one it is possible to set
num_trash_bos to some constant value when we don't have aperture and use
device_coherent. I am not completely sure if there's value in that but
maybe let's just go ahead and do it.

>		gem_evict_pwrite_subtest();
> +	}
>
>	/* Planes and cursors */
>	igt_subtest("cursor")
> @@ -2058,10 +2062,14 @@ igt_main_args("", long_options, help_str, opt_handler, NULL)
>		dpms_mode_unset_subtest(SCREEN_TYPE_LPSP);
>	igt_subtest("dpms-mode-unset-non-lpsp")
>		dpms_mode_unset_subtest(SCREEN_TYPE_NON_LPSP);
> -	igt_subtest("fences")
> +	igt_subtest("fences") {
> +		gem_require_mappable_ggtt(drm_fd);
>		fences_subtest(false);

Now I am thinking if fences_subtest() should be skipped or we just fix
tiling and mmap and retain it?

> -	igt_subtest("fences-dpms")
> +	}
> +	igt_subtest("fences-dpms") {
> +		gem_require_mappable_ggtt(drm_fd);
>		fences_subtest(true);

Ditto.

> +	}
>
>	/* Modeset stress */
>	igt_subtest("modeset-lpsp-stress")
> @@ -2102,10 +2110,14 @@ igt_main_args("", long_options, help_str, opt_handler, NULL)
>		gem_execbuf_stress_subtest(rounds, WAIT_STATUS | WAIT_EXTRA);
>
>	/* power-wake reference tests */
> -	igt_subtest("pm-tiling")
> +	igt_subtest("pm-tiling") {
> +		gem_require_mappable_ggtt(drm_fd);
>		pm_test_tiling();

Maybe fix and retain?

> -	igt_subtest("pm-caching")
> +	}
> +	igt_subtest("pm-caching") {
> +		gem_require_mappable_ggtt(drm_fd);

Again maybe fix and retain?