drm/amdgpu: Enable tunneling on high-priority compute queues

This improves latency if the GPU is already busy with other work.
This is useful for VR compositors that submit highly latency-sensitive
compositing work on high-priority compute queues while the GPU is busy
rendering the next frame.

Userspace merge request:
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26462

v2: bump driver version (Alex)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Friedrich Vock <friedrich.vock@gmx.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -791,6 +791,7 @@ struct amdgpu_mqd_prop {
 	uint64_t eop_gpu_addr;
 	uint32_t hqd_pipe_priority;
 	uint32_t hqd_queue_priority;
+	bool allow_tunneling;
 	bool hqd_active;
 };
 

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -115,9 +115,10 @@
  *   3.54.0 - Add AMDGPU_CTX_QUERY2_FLAGS_RESET_IN_PROGRESS support
  * - 3.55.0 - Add AMDGPU_INFO_GPUVM_FAULT query
  * - 3.56.0 - Update IB start address and size alignment for decode and encode
+ * - 3.57.0 - Compute tunneling on GFX10+
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	56
+#define KMS_DRIVER_MINOR	57
 #define KMS_DRIVER_PATCHLEVEL	0
 
 /*

drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c

@@ -642,6 +642,10 @@ static void amdgpu_ring_to_mqd_prop(struct amdgpu_ring *ring,
 				    struct amdgpu_mqd_prop *prop)
 {
 	struct amdgpu_device *adev = ring->adev;
+	bool is_high_prio_compute = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE &&
+				    amdgpu_gfx_is_high_priority_compute_queue(adev, ring);
+	bool is_high_prio_gfx = ring->funcs->type == AMDGPU_RING_TYPE_GFX &&
+				amdgpu_gfx_is_high_priority_graphics_queue(adev, ring);
 
 	memset(prop, 0, sizeof(*prop));
 
@@ -659,10 +663,8 @@ static void amdgpu_ring_to_mqd_prop(struct amdgpu_ring *ring,
 	 */
 	prop->hqd_active = ring->funcs->type == AMDGPU_RING_TYPE_KIQ;
 
-	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE &&
-	     amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) ||
-	    (ring->funcs->type == AMDGPU_RING_TYPE_GFX &&
-	     amdgpu_gfx_is_high_priority_graphics_queue(adev, ring))) {
+	prop->allow_tunneling = is_high_prio_compute;
+	if (is_high_prio_compute || is_high_prio_gfx) {
 		prop->hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
 		prop->hqd_queue_priority = AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
 	}

drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c

@@ -6593,7 +6593,8 @@ static int gfx_v10_0_compute_mqd_init(struct amdgpu_device *adev, void *m,
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
 #endif
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
-	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
+	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH,
+			    prop->allow_tunneling);
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
 	mqd->cp_hqd_pq_control = tmp;

drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c

@@ -3847,7 +3847,8 @@ static int gfx_v11_0_compute_mqd_init(struct amdgpu_device *adev, void *m,
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
 			    (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
-	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
+	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH,
+			    prop->allow_tunneling);
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
 	mqd->cp_hqd_pq_control = tmp;