On 31/10/15 01:45, Vivek Kasireddy wrote:
Hi Tvrtko,
On Fri, 30 Oct 2015 10:22:08 +0000
Tvrtko Ursulin <tvrtko.ursu...@linux.intel.com> wrote:
On 30/10/15 01:44, Vivek Kasireddy wrote:
The main goal of this subtest is to trigger the following warning in
the function i915_gem_object_get_fence():
if (WARN_ON(!obj->map_and_fenceable))
To trigger this warning, the subtest first creates a Y-tiled object
and an associated framebuffer with the Y-fb modifier. Furthermore,
to prevent the map_and_fenceable from being set, we make sure that
the object does not have a normal VMA by refraining from rendering
to the object and by setting the rotation property upfront before
calling commit.
v2: Do not call paint_squares and just use one output.
v3: Convert an if condition to igt_require and move the plane
rotation requirement further up before the fb allocation.
v4: After setting rotation to 90 and committing, change the
rotation to 0 and commit once more. This is to test if the i915
driver hits any warnings while pinning and unpinning an object that
has both normal and rotated views.
v5:
- Add another subtest to toggle the order of rotation
- Exhaustively test the i915 driver's pinning and unpinning code
paths for any fence leaks by iterating until MAX available fences.
Cc: Tvrtko Ursulin <tvrtko.ursu...@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasire...@intel.com>
---
tests/kms_rotation_crc.c | 84
++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 84
insertions(+)
diff --git a/tests/kms_rotation_crc.c b/tests/kms_rotation_crc.c
index cc9847e..34f8150 100644
--- a/tests/kms_rotation_crc.c
+++ b/tests/kms_rotation_crc.c
@@ -264,6 +264,80 @@ static void test_plane_rotation(data_t *data,
enum igt_plane plane_type) igt_require_f(valid_tests, "no valid
crtc/connector combinations found\n"); }
+static void test_plane_rotation_ytiled_obj(data_t *data, enum
igt_plane plane_type,
+ int toggle)
+{
+ igt_display_t *display = &data->display;
+ uint64_t tiling = LOCAL_I915_FORMAT_MOD_Y_TILED;
+ uint32_t format = DRM_FORMAT_XRGB8888;
+ int bpp = igt_drm_format_to_bpp(format);
+ enum igt_commit_style commit = COMMIT_LEGACY;
+ int fd = data->gfx_fd;
+ igt_output_t *output = &display->outputs[0];
+ igt_plane_t *plane;
+ drmModeModeInfo *mode;
+ unsigned int stride, size, w, h;
+ uint32_t gem_handle;
+ int num_fences = gem_available_fences(fd);
+ int i, ret;
+
+ igt_require(output != NULL && output->valid == true);
+
+ plane = igt_output_get_plane(output, plane_type);
+ igt_require(igt_plane_supports_rotation(plane));
+
+ if (plane_type == IGT_PLANE_PRIMARY || plane_type ==
IGT_PLANE_CURSOR) {
+ igt_require(data->display.has_universal_planes);
+ commit = COMMIT_UNIVERSAL;
+ }
+
+ mode = igt_output_get_mode(output);
+ w = mode->hdisplay;
+ h = mode->vdisplay;
+
+ for (stride = 512; stride < (w * bpp / 8); stride *= 2)
+ ;
+ for (size = 1024*1024; size < stride * h; size *= 2)
+ ;
+
+ gem_handle = gem_create(fd, size);
+ ret = __gem_set_tiling(fd, gem_handle, I915_TILING_Y,
stride);
+ igt_assert(ret == 0);
+
+ do_or_die(__kms_addfb(fd, gem_handle, w, h, stride,
+ format, tiling, LOCAL_DRM_MODE_FB_MODIFIERS,
+ &data->fb.fb_id));
+ data->fb.width = w;
+ data->fb.height = h;
+ data->fb.gem_handle = gem_handle;
+
+ igt_plane_set_fb(plane, NULL);
+ igt_display_commit(display);
+
+ igt_plane_set_fb(plane, &data->fb);
+
+ for (i = 0; i < num_fences + 1; i++) {
+ igt_plane_set_rotation(plane, toggle ?
IGT_ROTATION_0 : IGT_ROTATION_90);
+ drmModeObjectSetProperty(fd,
plane->drm_plane->plane_id,
+ DRM_MODE_OBJECT_PLANE,
+ plane->rotation_property,
+ plane->rotation);
+ ret = igt_display_try_commit2(display, commit);
+ igt_assert(ret == 0);
+
+ igt_plane_set_rotation(plane, toggle ?
IGT_ROTATION_90 : IGT_ROTATION_0);
+ drmModeObjectSetProperty(fd,
plane->drm_plane->plane_id,
+ DRM_MODE_OBJECT_PLANE,
+ plane->rotation_property,
+ plane->rotation);
+ ret = igt_display_try_commit2(display, commit);
+ igt_assert(ret == 0);
+ }
It manages to exhaust the fences with only one object? I was
expecting that multiple objects will be required since if it is only
one how come it allocates more than one fence register?
Before I sent out this version, I did try with two objects to see if it
triggers any WARNs but it didn't. I am not sure if I did it right
though, I'll take a look at it again. However, when I tried this
version, that is, single-object-multiple-views, it did trigger the
following WARNS:
WARN_ON(!ggtt_vma || dev_priv->fence_regs[obj->fence_reg].pin_count >
ggtt_vma->pin_count)
if (WARN_ON(fence->pin_count))
I tested these with v3 of my i915 driver patch. Of-course, with v4,
I don't seen any WARNs. I'll take a closer look at triggering
any WARNs with 2 or more objects. However, one thing that was not clear
I thought you would need number of framebuffers / objects > num_fences
and display all of them to trigger the running out of fences ref leak.
to me was when you said unpin both (in your previous comment), I was
wondering how I can do that from the test. Did you mean something like
this:
igt_plane_set_fb(plane, NULL);
igt_display_commit(display);
Couldn't find in this thread where I said "unpin both" so don't know.
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
