[igt-dev] [PATCH] tests/kms_cursor_legacy: allow an extra vblank in atomic-transitions tests

[Alex Hung]

On i915 devices:
kmstest_get_vblank() calls a drm_ioctl(DRM_IOCTL_WAIT_VBLANK), but
vblanks are not updated until a following read() op triggers kernel to
update vblank, ex. "[drm:drm_handle_vblank [drm]] vblank event on xxx".

However, this is not the case for all devices:
kmstest_get_vblank() triggers a drm_ioctl(DRM_IOCTL_WAIT_VBLANK),
and then drm_wait_vblank_ioctl() in drm_vblank.c is called and
returns vblank, ex. "[drm]] crtc 0 returning xxx to client" right away.
In this scenario, the read() op does nothing related to vblank.

Since the vblank can be updated before the read(), the extra vblank
in commit da0b6abeshould also be allowed before the read().

Signed-off-by: Alex Hung <alex.hung at amd.com>
---
 tests/kms_cursor_legacy.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/tests/kms_cursor_legacy.c b/tests/kms_cursor_legacy.c
index 1ad92eaa..868cfb97 100644
--- a/tests/kms_cursor_legacy.c
+++ b/tests/kms_cursor_legacy.c
@@ -802,6 +802,8 @@ static void flip_vs_cursor(igt_display_t *display, enum flip_test mode, int nloo
 		/* Nor should it have delayed the following cursor update */
 		if (!cursor_slowpath(display, mode))
 			igt_assert_eq(kmstest_get_vblank(display->drm_fd, pipe, 0), vblank_start);
+		else if (mode_requires_extra_vblank(mode))
+			igt_assert_lte(kmstest_get_vblank(display->drm_fd, pipe, 0), vblank_start + 2);
 		else
 			igt_assert_lte(kmstest_get_vblank(display->drm_fd, pipe, 0), vblank_start + 1);
 
-- 
2.34.1