gpu: nvgpu: move resetup_ramfc() out of sync_lock

We currently have this sequence :
- acquire sync_lock
- sync_create
- resetup_ramfc()
- release sync_lock

but this can lead to deadlock in case resetup_ramfc()
triggers below stack :
- resetup_ramfc()
  - channel_preempt() - preemption fails
  - trigger recovery
  - channel_abort()
    - acquire sync_lock

Fix this by moving resetup_ramfc() out of sync_lock.

resetup_ramfc() is still protected by submit_lock and
hence we cannot free sync after allocation and
before resetup

Bug 200165811

Change-Id: Iebf74d950d6f6902b6d180c2cd8cd2d50493062c
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/931726
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>

1 files changed, 11 insertions, 4 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 0c28d0bb..45501d4f 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -1936,6 +1936,7 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
         bool skip_buffer_refcounting = (flags &
                         NVGPU_SUBMIT_GPFIFO_FLAGS_SKIP_BUFFER_REFCOUNTING);
         bool need_sync_fence = false;
+        bool new_sync_created = false;
 
         /*
          * If user wants to allocate sync_fence_fd always, then respect that;
@@ -2033,16 +2034,22 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
                 c->sync = gk20a_channel_sync_create(c);
                 if (!c->sync) {
                         err = -ENOMEM;
+                        mutex_unlock(&c->sync_lock);
                         mutex_unlock(&c->submit_lock);
                         goto clean_up;
                 }
-                if (g->ops.fifo.resetup_ramfc)
-                        err = g->ops.fifo.resetup_ramfc(c);
-                if (err)
-                        return err;
+                new_sync_created = true;
         }
         mutex_unlock(&c->sync_lock);
 
+        if (g->ops.fifo.resetup_ramfc && new_sync_created) {
+                err = g->ops.fifo.resetup_ramfc(c);
+                if (err) {
+                        mutex_unlock(&c->submit_lock);
+                        goto clean_up;
+                }
+        }
+
         /*
          * optionally insert syncpt wait in the beginning of gpfifo submission
          * when user requested and the wait hasn't expired.