gpu: nvgpu: Allow suppressing WFI on submit

Allow suppressing WFI when submitting work and requesting a fence
back.

Bug 1491545

Change-Id: Ic3d061bb4f116cf7ea68dbd6a1b2ace9f11d0ab5
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/390457
Reviewed-on: http://git-master/r/671029
GVS: Gerrit_Virtual_Submit
Reviewed-by: Sibashis Mohapatra <sibashism@nvidia.com>
Tested-by: Sibashis Mohapatra <sibashism@nvidia.com>
Reviewed-by: Yogesh Kini <ykini@nvidia.com>
Reviewed-by: Winnie Hsu <whsu@nvidia.com>

4 files changed, 18 insertions, 5 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index f145b661241b..07137c28ea6e 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -1413,6 +1413,7 @@ static int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
 	/* we might need two extra gpfifo entries - one for pre fence
 	 * and one for post fence. */
 	const int extra_entries = 2;
+	bool need_wfi = !(flags & NVHOST_SUBMIT_GPFIFO_FLAGS_SUPPRESS_WFI);
 
 	if (c->has_timedout)
 		return -ETIMEDOUT;
@@ -1505,10 +1506,12 @@ static int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
 			flags & NVHOST_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE)
 		err = c->sync->incr_user_fd(c->sync, &incr_cmd,
 					    &c->last_submit_fence,
+					    need_wfi,
 					    &fence->syncpt_id);
 	else if (flags & NVHOST_SUBMIT_GPFIFO_FLAGS_FENCE_GET)
 		err = c->sync->incr_user_syncpt(c->sync, &incr_cmd,
 						&c->last_submit_fence,
+						need_wfi,
 						&fence->syncpt_id,
 						&fence->value);
 	else
@@ -1523,7 +1526,8 @@ static int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
 		c->gpfifo.cpu_va[c->gpfifo.put].entry1 =
 			u64_hi32(wait_cmd->gva) |
 			pbdma_gp_entry1_length_f(wait_cmd->size);
-		trace_write_pushbuffer(c, &c->gpfifo.cpu_va[c->gpfifo.put]);
+		trace_gk20a_push_cmdbuf(c->g->dev->name,
+			0, wait_cmd->size, 0, wait_cmd->ptr);
 
 		c->gpfifo.put = (c->gpfifo.put + 1) &
 			(c->gpfifo.entry_num - 1);
@@ -1548,7 +1552,8 @@ static int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
 		c->gpfifo.cpu_va[c->gpfifo.put].entry1 =
 			u64_hi32(incr_cmd->gva) |
 			pbdma_gp_entry1_length_f(incr_cmd->size);
-		trace_write_pushbuffer(c, &c->gpfifo.cpu_va[c->gpfifo.put]);
+		trace_gk20a_push_cmdbuf(c->g->dev->name,
+			0, incr_cmd->size, 0, incr_cmd->ptr);
 
 		c->gpfifo.put = (c->gpfifo.put + 1) &
 			(c->gpfifo.entry_num - 1);
diff --git a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
index a8f57830f8ad..952087e5e07a 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
@@ -261,6 +261,7 @@ int gk20a_channel_syncpt_incr(struct gk20a_channel_sync *s,
 int gk20a_channel_syncpt_incr_user_syncpt(struct gk20a_channel_sync *s,
 					  struct priv_cmd_entry **entry,
 					  struct gk20a_channel_fence *fence,
+					  bool wfi,
 					  u32 *id, u32 *thresh)
 {
 	struct gk20a_channel_syncpt *sp =
@@ -268,8 +269,10 @@ int gk20a_channel_syncpt_incr_user_syncpt(struct gk20a_channel_sync *s,
 	/* Need to do 'host incr + wfi' or 'gfx incr' since we return the fence
 	 * to user space. */
 	int err = __gk20a_channel_syncpt_incr(s,
-			sp->c->obj_class == KEPLER_C /* use gfx class? */,
-			sp->c->obj_class != KEPLER_C /* wfi if host class */,
+			wfi &&
+			  sp->c->obj_class == KEPLER_C /* use gfx class? */,
+			wfi &&
+			  sp->c->obj_class != KEPLER_C /* wfi if host class */,
 			true /* register irq */,
 			entry, fence);
 	if (err)
@@ -282,6 +285,7 @@ int gk20a_channel_syncpt_incr_user_syncpt(struct gk20a_channel_sync *s,
 int gk20a_channel_syncpt_incr_user_fd(struct gk20a_channel_sync *s,
 				      struct priv_cmd_entry **entry,
 				      struct gk20a_channel_fence *fence,
+				      bool wfi,
 				      int *fd)
 {
 #ifdef CONFIG_SYNC
@@ -289,7 +293,7 @@ int gk20a_channel_syncpt_incr_user_fd(struct gk20a_channel_sync *s,
 	struct nvhost_ctrl_sync_fence_info pt;
 	struct gk20a_channel_syncpt *sp =
 		container_of(s, struct gk20a_channel_syncpt, ops);
-	err = gk20a_channel_syncpt_incr_user_syncpt(s, entry, fence,
+	err = gk20a_channel_syncpt_incr_user_syncpt(s, entry, fence, wfi,
 						    &pt.id, &pt.thresh);
 	if (err)
 		return err;
diff --git a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h
index 80f38b266089..90b61bfd7131 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h
@@ -77,6 +77,7 @@ struct gk20a_channel_sync {
 	int (*incr_user_syncpt)(struct gk20a_channel_sync *s,
 				struct priv_cmd_entry **entry,
 				struct gk20a_channel_fence *fence,
+				bool wfi,
 				u32 *id, u32 *thresh);
 
 	/* Increment syncpoint/semaphore, so that the returned fence represents
@@ -89,6 +90,7 @@ struct gk20a_channel_sync {
 	int (*incr_user_fd)(struct gk20a_channel_sync *s,
 			    struct priv_cmd_entry **entry,
 			    struct gk20a_channel_fence *fence,
+			    bool wfi,
 			    int *fd);
 
 	/* Reset the channel syncpoint/semaphore. */
diff --git a/include/linux/nvhost_ioctl.h b/include/linux/nvhost_ioctl.h
index 4b6e1a2fb132..b060864ff1d1 100644
--- a/include/linux/nvhost_ioctl.h
+++ b/include/linux/nvhost_ioctl.h
@@ -143,6 +143,8 @@ struct nvhost_fence {
 #define NVHOST_SUBMIT_GPFIFO_FLAGS_HW_FORMAT	BIT(2)
 /* create a sync fence fd instead of raw fence */
 #define NVHOST_SUBMIT_GPFIFO_FLAGS_SYNC_FENCE	BIT(3)
+/* suppress WFI before fence trigger */
+#define NVHOST_SUBMIT_GPFIFO_FLAGS_SUPPRESS_WFI	BIT(4)
 
 struct nvhost_submit_gpfifo_args {
 	__u64 gpfifo;