gpu: nvgpu: fix deadlock on timeout lock

In gk20a_channel_timeout_stop(), we take the channel's
timeout lock and then cancel the timeout worker thread

Timeout worker thread also tries to acquire same timeout
lock.

Hence, while cancelling the timeout in gk20a_channel_timeout_stop()
if the timeout_handler is already scheduled, we will have a deadlock

Fix this by moving cancel_delayed_work_sync() out of the locks

Bug 200133289
Bug 1695481

Change-Id: Iea78770180b483a63e5e176efba27831174e9dde
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/815922
Reviewed-by: Ishan Mittal <imittal@nvidia.com>
Tested-by: Ishan Mittal <imittal@nvidia.com>

1 files changed, 3 insertions, 2 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 6dad412a..6f0d7375 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -1561,15 +1561,16 @@ static void gk20a_channel_timeout_start(struct channel_gk20a *ch,
 static void gk20a_channel_timeout_stop(struct channel_gk20a *ch)
 {
         mutex_lock(&ch->timeout.lock);
-
         if (!ch->timeout.initialized) {
                 mutex_unlock(&ch->timeout.lock);
                 return;
         }
+        mutex_unlock(&ch->timeout.lock);
 
-        ch->timeout.initialized = false;
         cancel_delayed_work_sync(&ch->timeout.wq);
 
+        mutex_lock(&ch->timeout.lock);
+        ch->timeout.initialized = false;
         mutex_unlock(&ch->timeout.lock);
 }