Commit f558d747 authored Nov 16, 2015 by Dhaval Patel

msm: mdss: flush retire work before ctl_stop operation



The vsync_retire worker registers for vsync_handler and
removes it when client requests power_off. It may
possible that vsync_handler is removed from list
through two different contexts. One from worker thread
and other from ctl_stop_sub call. Such race condition
can lead to list corruption. Ideally, ov_off should
wait for retire worker flush before calling ctl_stop
to avoid such race condition.

Change-Id: I7d68d67d1fe1df07e568a5f40db745ce155d7d14
Signed-off-by: Dhaval Patel <pdhaval@codeaurora.org>

parent bfbfa771

drivers/video/msm/mdss/mdss_mdp_overlay.c

+10 −0

Original line number	Diff line number	Diff line
		@@ -4686,6 +4686,16 @@ static int mdss_mdp_overlay_off(struct msm_fb_data_type *mfd)
		msleep(vsync_time);

		__vsync_retire_signal(mfd, mdp5_data->retire_cnt);

		/*
		* the retire work can still schedule after above retire_signal
		* api call. Flush workqueue guarantees that current caller
		* context is blocked till retire_work finishes. Any work
		* schedule after flush call should not cause any issue because
		* retire_signal api checks for retire_cnt with sync_mutex lock.
		*/

		flush_work(&mdp5_data->retire_work);
		}

		ctl_stop: