gpu: nvgpu: Optimize sync fence creation

Only create sync-fences in the semaphore synchronization path
when they are actually needed (i.e requested by userspace).

Bug 1795076

Reviewed-on: http://git-master/r/1201564
(cherry picked from commit dc52d424a839e6c064c02b7f02905dd6a59a50af)
Change-Id: Ieac6aef415678d4ea982683a955897c64959436e
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: http://git-master/r/1221041
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>

3 files changed, 11 insertions, 8 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
index e4972610..98363c88 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
@@ -605,7 +605,7 @@ static int gk20a_channel_semaphore_wait_fd(
                         *fence = gk20a_fence_from_semaphore(sema->timeline,
                                                             fp_sema,
                                                             &c->semaphore_wq,
-                                                            NULL, false);
+                                                            NULL, false, false);
                 else
                         /*
                          * Allocate an empty fence. It will instantly return
@@ -676,7 +676,7 @@ static int gk20a_channel_semaphore_wait_fd(
            semaphore while the job is in flight. */
         *fence = gk20a_fence_from_semaphore(sema->timeline, w->sema,
                                             &c->semaphore_wq,
-                                            NULL, false);
+                                            NULL, false, false);
 
 skip_slow_path:
         *entry = wait_cmd;
@@ -734,7 +734,8 @@ static int __gk20a_channel_semaphore_incr(
 
         *fence = gk20a_fence_from_semaphore(sp->timeline, semaphore,
                                             &c->semaphore_wq,
-                                            dependency, wfi_cmd);
+                                            dependency, wfi_cmd,
+                                            need_sync_fence);
         *entry = incr_cmd;
         return 0;
 }
diff --git a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
index fbbaa2a7..596dc549 100644
--- a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
@@ -148,17 +148,19 @@ struct gk20a_fence *gk20a_fence_from_semaphore(
                 struct gk20a_semaphore *semaphore,
                 wait_queue_head_t *semaphore_wq,
                 struct sync_fence *dependency,
-                bool wfi)
+                bool wfi, bool need_sync_fence)
 {
         struct gk20a_fence *f;
         struct sync_fence *sync_fence = NULL;
 
 #ifdef CONFIG_SYNC
-        sync_fence = gk20a_sync_fence_create(timeline, semaphore,
+        if (need_sync_fence) {
+                sync_fence = gk20a_sync_fence_create(timeline, semaphore,
                                         dependency, "f-gk20a-0x%04x",
                                         gk20a_semaphore_gpu_ro_va(semaphore));
-        if (!sync_fence)
-                return NULL;
+                if (!sync_fence)
+                        return NULL;
+        }
 #endif
 
         f  = gk20a_alloc_fence(&gk20a_semaphore_fence_ops, sync_fence, wfi);
diff --git a/drivers/gpu/nvgpu/gk20a/fence_gk20a.h b/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
index ff00bcf1..35488ea3 100644
--- a/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
@@ -52,7 +52,7 @@ struct gk20a_fence *gk20a_fence_from_semaphore(
                 struct gk20a_semaphore *semaphore,
                 wait_queue_head_t *semaphore_wq,
                 struct sync_fence *dependency,
-                bool wfi);
+                bool wfi, bool need_sync_fence);
 
 struct gk20a_fence *gk20a_fence_from_syncpt(
                 struct platform_device *host1x_pdev,