gpu: nvgpu: disable sync_fence for CE jobs

We do not need sync_fence for CE jobs submitted in
gk20a_ce_execute_ops() since all the waiters of
fence are in kernel space only

Jira DNVGPU-84

Change-Id: Idad6c40abcefb86e60a5327bbbff6827b1ca33cc
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/1201347
(cherry picked from commit e294b2d37cf79182bb9a255adb188eb6afa47c27)
Reviewed-on: http://git-master/r/1210951
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

1 files changed, 2 insertions, 1 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/ce2_gk20a.c b/drivers/gpu/nvgpu/gk20a/ce2_gk20a.c
index 9be37a5e..8548ae84 100644
--- a/drivers/gpu/nvgpu/gk20a/ce2_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/ce2_gk20a.c
@@ -652,7 +652,8 @@ int gk20a_ce_execute_ops(struct device *dev,
                 wmb();
 
                 ret = gk20a_submit_channel_gpfifo(ce_ctx->ch, &gpfifo, NULL,
-                                        1, submit_flags, &fence, &ce_cmd_buf_fence_out, true);
+                                        1, submit_flags, &fence,
+                                        &ce_cmd_buf_fence_out, false);
 
                 if (!ret) {
                         memcpy((void *)(cmd_buf_cpu_va + fence_index),