On 2020-09-11 12:27 p.m., Aurabindo Pillai wrote:
[Why&How]
Since there is no need for accessing crtc state in the interrupt
handler, interrupts need not be disabled well in advance, and
can be moved to commit_tail where it should be.
Signed-off-by: Aurabindo Pillai <aurabindo.pillai@xxxxxxx>
---
.../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 41 ++++++-------------
1 file changed, 13 insertions(+), 28 deletions(-)
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index 1455acf5beca..b5af1f692499 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -7488,34 +7488,6 @@ static int amdgpu_dm_atomic_commit(struct drm_device *dev,
struct drm_atomic_state *state,
bool nonblock)
{
- struct drm_crtc *crtc;
- struct drm_crtc_state *old_crtc_state, *new_crtc_state;
- struct amdgpu_device *adev = drm_to_adev(dev);
- int i;
-
- /*
- * We evade vblank and pflip interrupts on CRTCs that are undergoing
- * a modeset, being disabled, or have no active planes.
- *
- * It's done in atomic commit rather than commit tail for now since
- * some of these interrupt handlers access the current CRTC state and
- * potentially the stream pointer itself.
- *
- * Since the atomic state is swapped within atomic commit and not within
- * commit tail this would leave to new state (that hasn't been committed yet)
- * being accesssed from within the handlers.
- *
- * TODO: Fix this so we can do this in commit tail and not have to block
- * in atomic check.
- */
- for_each_oldnew_crtc_in_state(state, crtc, old_crtc_state, new_crtc_state, i) {
- struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
-
- if (old_crtc_state->active &&
- (!new_crtc_state->active ||
- drm_atomic_crtc_needs_modeset(new_crtc_state)))
- manage_dm_interrupts(adev, acrtc, false);
- }
/*
* Add check here for SoC's that support hardware cursor plane, to
* unset legacy_cursor_update
@@ -7566,6 +7538,19 @@ static void amdgpu_dm_atomic_commit_tail(struct drm_atomic_state *state)
dc_resource_state_copy_construct_current(dm->dc, dc_state);
}
+ for_each_oldnew_crtc_in_state (state, crtc, old_crtc_state,
+ new_crtc_state, i) {
+ acrtc = to_amdgpu_crtc(crtc);
+ dm_old_crtc_state = to_dm_crtc_state(old_crtc_state);
+
+ if (old_crtc_state->active &&
+ (!new_crtc_state->active ||
+ drm_atomic_crtc_needs_modeset(new_crtc_state))) {
+ manage_dm_interrupts(adev, acrtc, false);
+ dc_stream_release(dm_old_crtc_state->stream);
Please include this dc_stream_release() in patch #2 as well to prevent
memory leaks during bisections.
With that change, this series is Reviewed-by: Nicholas Kazlauskas
<nicholas.kazlauskas@xxxxxxx>
Regards,
Nicholas Kazlauskas
+ }
+ }
+
/* update changed items */
for_each_oldnew_crtc_in_state(state, crtc, old_crtc_state, new_crtc_state, i) {
struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
_______________________________________________
amd-gfx mailing list
amd-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
