The issue scenario is as follows:
1. User space issues CWB commit N-1, frame got picked up and wr_ptr_irq
is received.
2. Next commit N CWB disable commit is programmed waits for N-1 wb_done
irq.
3. The kickoff count is decremented on wb_done_irq of commit N-1
and wb wait_for_idle is exited.
4. wb_frame_done irq thread execution stalled before populating fences
and commit thread execution continues.
5. wb_reset disables in_clone_mode flag, the stalled wb_done_irq thread
resumes its execution and signals the release fences on primary crtc.
6. Commit N-1 frame_done irq is received and release fences is
signaled again.
Made changes to avoid the race between irq thread and commit thread
over in_clone_mode flag, by adding a lock over wb physical encoder.
Change-Id: Iba9b6613c49d413239c9603228fe16b0d64c0ab6
Signed-off-by: Andhavarapu Karthik <quic_kartkart@quicinc.com>
13d31e5111