On Wed, Mar 02, 2016 at 05:50:28PM +0200, David Weinehall wrote:
> On machines that lack an LLC the pm-caching subtest will
> terminate with sigbus and thus CRASH during the
> I915_CACHING_CACHED iteration. This patch adds a check for
> this and uses I915_CACHING_NONE instead.
>
> v2: Various improvements based on feedback from Chris Wilson
>
> Signed-off-by: David Weinehall@xxxxxxxxxxxxxxx
> ---
> tests/pm_rpm.c | 33 +++++++++++----------------------
> 1 file changed, 11 insertions(+), 22 deletions(-)
>
> diff --git a/tests/pm_rpm.c b/tests/pm_rpm.c
> index 2aa6c1018aa2..add8e2eb457f 100644
> --- a/tests/pm_rpm.c
> +++ b/tests/pm_rpm.c
> @@ -1800,7 +1800,7 @@ static void pm_test_caching(void)
> uint32_t handle;
> uint8_t *gem_buf;
>
> - uint32_t i, got_caching;
> + uint32_t i;
> uint32_t gtt_obj_max_size = (16 * 1024);
> uint32_t cache_levels[3] = {
> I915_CACHING_NONE,
> @@ -1808,42 +1808,31 @@ static void pm_test_caching(void)
> I915_CACHING_DISPLAY, /* eDRAM caching */
> };
>
> + disable_all_screens(&ms_data);
>
> handle = gem_create(drm_fd, gtt_obj_max_size);
> gem_buf = gem_mmap__gtt(drm_fd, handle, gtt_obj_max_size, PROT_WRITE);
>
> for (i = 0; i < ARRAY_SIZE(cache_levels); i++) {
> - memset(gem_buf, 16 << i, gtt_obj_max_size);
> + igt_assert(wait_for_suspended());
>
> - disable_all_screens_and_wait(&ms_data);
> + memset(gem_buf, 16 << i, gtt_obj_max_size);
>
> igt_debug("Setting cache level %u\n", cache_levels[i]);
>
> - gem_set_caching(drm_fd, handle, cache_levels[i]);
> -
> - got_caching = gem_get_caching(drm_fd, handle);
> -
> - igt_debug("Got back %u\n", got_caching);
> -
> - /*
> - * Allow fall-back to CACHING_NONE in case the platform does
> - * not support it.
> - */
> - if (cache_levels[i] == I915_CACHING_DISPLAY)
> - igt_assert(got_caching == I915_CACHING_NONE ||
> - got_caching == I915_CACHING_DISPLAY);
> - else
> - igt_assert(got_caching == cache_levels[i]);
> -
> - enable_one_screen_and_wait(&ms_data);
> + /* If we lack an LLC cache we use I915_CACHING_NONE instead */
> + if (cache_levels[i] == I915_CACHING_CACHED &&
> + !gem_has_llc(drm_fd)) {
> + igt_debug("!gem_has_llc(); using I915_CACHING_NONE\n");
> + gem_set_caching(drm_fd, handle, I915_CACHING_NONE);
> + } else
> + gem_set_caching(drm_fd, handle, cache_levels[i]);
> }
>
> igt_assert(munmap(gem_buf, gtt_obj_max_size) == 0);
> gem_close(drm_fd, handle);
> }
My argument is that:
unsigned default_level;
disable_all_screens(&ms_data);
handle = gem_create();
default_level = gem_get_caching();
ptr = gem_mmap__gtt()
for_each_level() {
igt_assert(wait_for_suspended());
gem_set_caching(default_level);
/* Ensure we bind the vma into the GGTT */
memset(ptr, 0, OBJECT_SIZE);
/* Now try changing the cache-level on the bound object.
* This will either unlikely unbind the object from the GGTT,
* or more likely just change the PTEs inside the GGTT. Either
* way the driver must take the rpm wakelock around the GSM
* access.
*/
igt_assert(wait_for_suspended());
gem_set_caching(this_level);
//igt_assert_eq(intel_detect_and_clear_invalid_rpm_access(), 0);
}
is clearer to follow.
We should also add partial-vma exercising. For that, change OBJECT_SIZE
to be greater than the mappable aperture.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
