On Tue, Sep 19, 2017 at 01:31:13PM +0200, Maarten Lankhorst wrote: > Collecting CRC may force a modeset, which is a bad idea after we just > forced a hang. The hang is intended to make sure the page flip doesn't > complete before the cursor, making sure that works. > > Signed-off-by: Maarten Lankhorst <maarten.lankhorst@xxxxxxxxxxxxxxx> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102259 > Cc: Marta Lofstedt <marta.lofstedt@xxxxxxxxx> > --- > tests/kms_cursor_legacy.c | 25 +++++++++++++++++++++---- > 1 file changed, 21 insertions(+), 4 deletions(-) > > diff --git a/tests/kms_cursor_legacy.c b/tests/kms_cursor_legacy.c > index 2d32d3a91570..c3149535b838 100644 > --- a/tests/kms_cursor_legacy.c > +++ b/tests/kms_cursor_legacy.c > @@ -1334,7 +1334,7 @@ static void flip_vs_cursor_busy_crc(igt_display_t *display, bool atomic) > igt_pipe_crc_t *pipe_crc; > igt_pipe_t *pipe_connected = &display->pipes[pipe]; > igt_plane_t *plane_primary = igt_pipe_get_plane_type(pipe_connected, DRM_PLANE_TYPE_PRIMARY); > - igt_crc_t crcs[3]; > + igt_crc_t crcs[2]; > > if (atomic) > igt_require(display->is_atomic); > @@ -1348,7 +1348,7 @@ static void flip_vs_cursor_busy_crc(igt_display_t *display, bool atomic) > > igt_display_commit2(display, display->is_atomic ? COMMIT_ATOMIC : COMMIT_LEGACY); > > - pipe_crc = igt_pipe_crc_new(display->drm_fd, pipe, INTEL_PIPE_CRC_SOURCE_AUTO); > + pipe_crc = igt_pipe_crc_new_nonblock(display->drm_fd, pipe, INTEL_PIPE_CRC_SOURCE_AUTO); > > set_cursor_on_pipe(display, pipe, &cursor_fb); > igt_display_commit2(display, COMMIT_UNIVERSAL); > @@ -1371,9 +1371,17 @@ static void flip_vs_cursor_busy_crc(igt_display_t *display, bool atomic) > igt_plane_set_fb(plane_primary, &fb_info[0]); > igt_display_commit2(display, COMMIT_UNIVERSAL); > > + /* > + * We must enable CRC collecting here since this may force > + * a modeset, and this loop is timing sensitive. > + */ > + igt_pipe_crc_start(pipe_crc); > + > /* Disable cursor, and immediately queue a flip. Check if resulting crc is correct. */ > for (int i = 1; i >= 0; i--) { > uint32_t *busy; > + igt_crc_t *received_crcs = NULL; > + int ncrcs; > > busy = make_fb_busy(display->drm_fd, &fb_info[1]); > > @@ -1384,7 +1392,7 @@ static void flip_vs_cursor_busy_crc(igt_display_t *display, bool atomic) > > igt_assert_eq(get_vblank(display->drm_fd, pipe, 0), vblank_start); > > - igt_pipe_crc_collect_crc(pipe_crc, &crcs[2]); > + ncrcs = igt_pipe_crc_get_crcs(pipe_crc, 8, &received_crcs); > > finish_fb_busy(busy); > > @@ -1397,13 +1405,22 @@ static void flip_vs_cursor_busy_crc(igt_display_t *display, bool atomic) > igt_plane_set_fb(plane_primary, &fb_info[0]); > igt_display_commit2(display, COMMIT_UNIVERSAL); > > - igt_assert_crc_equal(&crcs[i], &crcs[2]); > + igt_assert(ncrcs > 0); Should we perhaps assign some arbitrary upper limit on the crcs we get? If the kernel already buffered 8 or more crcs, we may get a crc mismatch even though the problem is that the test blocked for too long and we're now looking at a stale crc rather than the current crc being wrong. Might make it slightly easier to diagnose the failure correctly. Anyways Reviewed-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx> > + > + igt_assert_crc_equal(&crcs[i], &received_crcs[ncrcs - 1]); > + free(received_crcs); > } > > do_cleanup_display(display); > igt_remove_fb(display->drm_fd, &fb_info[1]); > igt_remove_fb(display->drm_fd, &fb_info[0]); > igt_remove_fb(display->drm_fd, &cursor_fb); > + > + /* > + * igt_pipe_crc_stop() may force a modeset for workarounds, call > + * it after do_cleanup_display since we disable the display anyway. > + */ > + igt_pipe_crc_stop(pipe_crc); > igt_pipe_crc_free(pipe_crc); > } > > -- > 2.14.1 > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@xxxxxxxxxxxxxxxxxxxxx > https://lists.freedesktop.org/mailman/listinfo/intel-gfx -- Ville Syrjälä Intel OTC _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
