gpu: nvgpu: create sync_fence only if needed

Currently, we create sync_fence (from nvhost_sync_create_fence())
for every submit
But not all submits request for a sync_fence.

Also, nvhost_sync_create_fence() API takes about 1/3rd of the total
submit path.

Hence to optimize, we can allocate sync_fence
only when user explicitly asks for it using
(NVGPU_SUBMIT_GPFIFO_FLAGS_FENCE_GET &&
NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE)

Also, in CDE path from gk20a_prepare_compressible_read(),
we reuse existing fence stored in "state" and that can
result into not returning sync_fence_fd when user asked
for it
Hence, force allocation of sync_fence when job submission
comes from CDE path

Bug 200141116

Change-Id: Ia921701bf0e2432d6b8a5e8b7d91160e7f52db1e
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/812845
(cherry picked from commit 5fd47015eeed00352cc8473eff969a66c94fee98)
Reviewed-on: http://git-master/r/837662
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Sachin Nikam <snikam@nvidia.com>

1 files changed, 16 insertions, 4 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 59c3e31d..98c8760e 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -1830,7 +1830,8 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
                                 u32 num_entries,
                                 u32 flags,
                                 struct nvgpu_fence *fence,
-                                struct gk20a_fence **fence_out)
+                                struct gk20a_fence **fence_out,
+                                bool force_need_sync_fence)
 {
         struct gk20a *g = c->g;
         struct device *d = dev_from_gk20a(g);
@@ -1848,6 +1849,14 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
         struct nvgpu_gpfifo *gpfifo_mem = c->gpfifo.mem.cpu_va;
         bool skip_buffer_refcounting = (flags &
                         NVGPU_SUBMIT_GPFIFO_FLAGS_SKIP_BUFFER_REFCOUNTING);
+        bool need_sync_fence = false;
+
+        /*
+         * If user wants to allocate sync_fence_fd always, then respect that;
+         * otherwise, allocate sync_fence_fd based on user flags only
+         */
+        if (force_need_sync_fence)
+                need_sync_fence = true;
 
         if (c->has_timedout)
                 return -ETIMEDOUT;
@@ -1970,15 +1979,18 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
                 goto clean_up;
         }
 
+        if ((flags & NVGPU_SUBMIT_GPFIFO_FLAGS_FENCE_GET) &&
+                        (flags & NVGPU_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE))
+                need_sync_fence = true;
 
         /* always insert syncpt increment at end of gpfifo submission
            to keep track of method completion for idle railgating */
         if (flags & NVGPU_SUBMIT_GPFIFO_FLAGS_FENCE_GET)
                 err = c->sync->incr_user(c->sync, wait_fence_fd, &incr_cmd,
-                                         &post_fence, need_wfi);
+                                 &post_fence, need_wfi, need_sync_fence);
         else
                 err = c->sync->incr(c->sync, &incr_cmd,
-                                    &post_fence);
+                                    &post_fence, need_sync_fence);
         if (err) {
                 mutex_unlock(&c->submit_lock);
                 goto clean_up;
@@ -2578,7 +2590,7 @@ static int gk20a_ioctl_channel_submit_gpfifo(
 
         ret = gk20a_submit_channel_gpfifo(ch, NULL, args, args->num_entries,
                                           args->flags, &args->fence,
-                                          &fence_out);
+                                          &fence_out, false);
 
         if (ret)
                 goto clean_up;