Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [High] drm/msm/dpu: unbalanced frame_pending decrements in clone mode --
commit 8cc8441c63a33addc5000cc2ffb1005c0a8221b2 Author: Michael Burch <[email protected]> drm/msm/dpu: count the pending frame before kicking it off This commit moves the frame_pending counter increment to precede the encoder kickoff to fix an issue where a frame completing quickly could miss the increment. It introduces dpu_crtc_count_pending_frame() and calls it before kickoffs in both clone mode and normal mode. > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c > b/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c > index 42d0a529b4d51..81282a8a5d9d1 100644 > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c [ ... ] > @@ -1062,6 +1075,8 @@ static int dpu_crtc_kickoff_clone_mode(struct drm_crtc > *crtc) > > dpu_vbif_clear_errors(dpu_kms); > > + dpu_crtc_count_pending_frame(crtc); > + > /* > * Kickoff real time encoder last as it's the encoder that > * will do the flush > */ > dpu_encoder_kickoff(wb_encoder); > dpu_encoder_kickoff(rt_encoder); [Severity: High] Does this code introduce an unbalanced decrement of frame_pending? In dpu_crtc_kickoff_clone_mode(), dpu_crtc_count_pending_frame() increments frame_pending once per CRTC commit. However, the function subsequently kicks off both wb_encoder and rt_encoder independently. When they complete, both encoders independently trigger dpu_encoder_frame_done_callback(), queueing two separate DPU_ENCODER_FRAME_EVENT_DONE events to the CRTC: drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c:dpu_encoder_frame_done_callback() { if (!dpu_enc->frame_busy_mask[0]) { ... dpu_encoder_resource_control(drm_enc, DPU_ENC_RC_EVENT_FRAME_DONE); if (dpu_enc->crtc) dpu_crtc_frame_event_cb(dpu_enc->crtc, event); } } The worker then drops the counter on each DONE event: drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c:dpu_crtc_frame_event_work() { ... } else if (atomic_dec_return(&dpu_crtc->frame_pending) == 0) { /* release bandwidth and other resources */ trace_dpu_crtc_frame_event_done(DRMID(crtc), fevent->event); dpu_core_perf_crtc_release_bw(crtc); } ... } If multiple clone-mode commits are interleaved, could the second event from an earlier commit arrive after the next commit sets frame_pending to 1? This would cause the worker to decrement the count to 0 and release system memory bandwidth prematurely before the real-time encoder has actually finished rendering the next commit. This might lead to FIFO underflows, visual artifacts, or display hangs. -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=1
