drm/i915: Unset legacy_cursor_update early in intel_atomic_commit, v3.
authorMaarten Lankhorst <maarten.lankhorst@linux.intel.com>
Tue, 19 Sep 2017 12:14:18 +0000 (14:14 +0200)
committerMaarten Lankhorst <maarten.lankhorst@linux.intel.com>
Wed, 4 Oct 2017 07:16:21 +0000 (09:16 +0200)
Commit b44d5c0c105a ("drm/i915: Always wait for flip_done, v2.") removed
the call to wait_for_vblanks and replaced it with flip_done.

Unfortunately legacy_cursor_update was unset too late, and the
replacement call drm_atomic_helper_wait_for_flip_done() was
a noop. Make sure that its unset before setup_commit() is
called to fix this issue.

Changes since v1:
- Force vblank wait for watermarks not yet converted to atomic too. (Ville)
- Use for_each_new_intel_crtc_in_state. (Ville)
Changes since v2:
- Move the optimization to a separate commit. (Ville)

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Fixes: b44d5c0c105a ("drm/i915: Always wait for flip_done, v2.")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102675
Testcase: kms_cursor_crc
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Reported-by: Marta Löfstedt <marta.lofstedt@intel.com>
Cc: Marta Löfstedt <marta.lofstedt@intel.com>
Tested-by: Marta Löfstedt <marta.lofstedt@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170919121419.13708-1-maarten.lankhorst@linux.intel.com
drivers/gpu/drm/i915/intel_display.c

index b7a6ddc..e55f68c 100644 (file)
@@ -12541,21 +12541,10 @@ static int intel_atomic_commit(struct drm_device *dev,
        struct drm_i915_private *dev_priv = to_i915(dev);
        int ret = 0;
 
-       ret = drm_atomic_helper_setup_commit(state, nonblock);
-       if (ret)
-               return ret;
-
        drm_atomic_state_get(state);
        i915_sw_fence_init(&intel_state->commit_ready,
                           intel_atomic_commit_ready);
 
-       ret = intel_atomic_prepare_commit(dev, state);
-       if (ret) {
-               DRM_DEBUG_ATOMIC("Preparing state failed with %i\n", ret);
-               i915_sw_fence_commit(&intel_state->commit_ready);
-               return ret;
-       }
-
        /*
         * The intel_legacy_cursor_update() fast path takes care
         * of avoiding the vblank waits for simple cursor
@@ -12564,11 +12553,11 @@ static int intel_atomic_commit(struct drm_device *dev,
         * updates happen during the correct frames. Gen9+ have
         * double buffered watermarks and so shouldn't need this.
         *
-        * Do this after drm_atomic_helper_setup_commit() and
-        * intel_atomic_prepare_commit() because we still want
-        * to skip the flip and fb cleanup waits. Although that
-        * does risk yanking the mapping from under the display
-        * engine.
+        * Unset state->legacy_cursor_update before the call to
+        * drm_atomic_helper_setup_commit() because otherwise
+        * drm_atomic_helper_wait_for_flip_done() is a noop and
+        * we get FIFO underruns because we didn't wait
+        * for vblank.
         *
         * FIXME doing watermarks and fb cleanup from a vblank worker
         * (assuming we had any) would solve these problems.
@@ -12576,7 +12565,17 @@ static int intel_atomic_commit(struct drm_device *dev,
        if (INTEL_GEN(dev_priv) < 9)
                state->legacy_cursor_update = false;
 
-       ret = drm_atomic_helper_swap_state(state, true);
+       ret = intel_atomic_prepare_commit(dev, state);
+       if (ret) {
+               DRM_DEBUG_ATOMIC("Preparing state failed with %i\n", ret);
+               i915_sw_fence_commit(&intel_state->commit_ready);
+               return ret;
+       }
+
+       ret = drm_atomic_helper_setup_commit(state, nonblock);
+       if (!ret)
+               ret = drm_atomic_helper_swap_state(state, true);
+
        if (ret) {
                i915_sw_fence_commit(&intel_state->commit_ready);