drm/amdgpu: apply AMDGPU_IB_FLAG_EMIT_MEM_SYNC to compute IBs too (v3)

Compute IBs need this too.

v2: split out version bump
v3: squash in emit frame count fixes
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -450,7 +450,7 @@
 #              define PACKET3_DMA_DATA_CMD_SAIC    (1 << 28)
 #              define PACKET3_DMA_DATA_CMD_DAIC    (1 << 29)
 #              define PACKET3_DMA_DATA_CMD_RAW_WAIT  (1 << 30)
-#define	PACKET3_AQUIRE_MEM				0x58
+#define	PACKET3_ACQUIRE_MEM				0x58
 #define	PACKET3_REWIND					0x59
 #define	PACKET3_LOAD_UCONFIG_REG			0x5E
 #define	PACKET3_LOAD_SH_REG				0x5F

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

@@ -8133,7 +8133,8 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_compute = {
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
 		2 + /* gfx_v10_0_ring_emit_vm_flush */
-		8 + 8 + 8, /* gfx_v10_0_ring_emit_fence x3 for user fence, vm fence */
+		8 + 8 + 8 + /* gfx_v10_0_ring_emit_fence x3 for user fence, vm fence */
+		8, /* gfx_v10_0_emit_mem_sync */
 	.emit_ib_size =	7, /* gfx_v10_0_ring_emit_ib_compute */
 	.emit_ib = gfx_v10_0_ring_emit_ib_compute,
 	.emit_fence = gfx_v10_0_ring_emit_fence,
@@ -8148,6 +8149,7 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_compute = {
 	.emit_wreg = gfx_v10_0_ring_emit_wreg,
 	.emit_reg_wait = gfx_v10_0_ring_emit_reg_wait,
 	.emit_reg_write_reg_wait = gfx_v10_0_ring_emit_reg_write_reg_wait,
+	.emit_mem_sync = gfx_v10_0_emit_mem_sync,
 };
 
 static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_kiq = {

drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c

@@ -3533,7 +3533,8 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_compute = {
 		5 + 5 + /* hdp flush / invalidate */
 		7 + /* gfx_v6_0_ring_emit_pipeline_sync */
 		SI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v6_0_ring_emit_vm_flush */
-		14 + 14 + 14, /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */
+		14 + 14 + 14 + /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */
+		5, /* SURFACE_SYNC */
 	.emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */
 	.emit_ib = gfx_v6_0_ring_emit_ib,
 	.emit_fence = gfx_v6_0_ring_emit_fence,
@@ -3543,6 +3544,7 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_compute = {
 	.test_ib = gfx_v6_0_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
 	.emit_wreg = gfx_v6_0_ring_emit_wreg,
+	.emit_mem_sync = gfx_v6_0_emit_mem_sync,
 };
 
 static void gfx_v6_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -5010,6 +5010,20 @@ static void gfx_v7_0_emit_mem_sync(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
 }
 
+static void gfx_v7_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
+	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
+			  PACKET3_TC_ACTION_ENA |
+			  PACKET3_SH_KCACHE_ACTION_ENA |
+			  PACKET3_SH_ICACHE_ACTION_ENA);  /* CP_COHER_CNTL */
+	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
+	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
+	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
+	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
+	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
+}
+
 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = {
 	.name = "gfx_v7_0",
 	.early_init = gfx_v7_0_early_init,
@@ -5075,7 +5089,8 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 		5 + /* hdp invalidate */
 		7 + /* gfx_v7_0_ring_emit_pipeline_sync */
 		CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */
-		7 + 7 + 7, /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
+		7 + 7 + 7 + /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */
+		7, /* gfx_v7_0_emit_mem_sync_compute */
 	.emit_ib_size =	7, /* gfx_v7_0_ring_emit_ib_compute */
 	.emit_ib = gfx_v7_0_ring_emit_ib_compute,
 	.emit_fence = gfx_v7_0_ring_emit_fence_compute,
@@ -5088,6 +5103,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = {
 	.insert_nop = amdgpu_ring_insert_nop,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
 	.emit_wreg = gfx_v7_0_ring_emit_wreg,
+	.emit_mem_sync = gfx_v7_0_emit_mem_sync_compute,
 };
 
 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev)

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -6830,6 +6830,21 @@ static void gfx_v8_0_emit_mem_sync(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
 }
 
+static void gfx_v8_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
+{
+	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
+	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
+			  PACKET3_TC_ACTION_ENA |
+			  PACKET3_SH_KCACHE_ACTION_ENA |
+			  PACKET3_SH_ICACHE_ACTION_ENA |
+			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
+	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
+	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
+	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
+	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
+	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
+}
+
 static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
 	.name = "gfx_v8_0",
 	.early_init = gfx_v8_0_early_init,
@@ -6912,7 +6927,8 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 		5 + /* hdp_invalidate */
 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
-		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
+		7 + 7 + 7 + /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
+		7, /* gfx_v8_0_emit_mem_sync_compute */
 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
 	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
@@ -6925,6 +6941,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.insert_nop = amdgpu_ring_insert_nop,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
+	.emit_mem_sync = gfx_v8_0_emit_mem_sync_compute,
 };
 
 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c

@@ -6741,7 +6741,8 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
 		2 + /* gfx_v9_0_ring_emit_vm_flush */
-		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
+		8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
+		7, /* gfx_v9_0_emit_mem_sync */
 	.emit_ib_size =	7, /* gfx_v9_0_ring_emit_ib_compute */
 	.emit_ib = gfx_v9_0_ring_emit_ib_compute,
 	.emit_fence = gfx_v9_0_ring_emit_fence,
@@ -6756,6 +6757,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
 	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
+	.emit_mem_sync = gfx_v9_0_emit_mem_sync,
 };
 
 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {

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

@@ -332,7 +332,7 @@
 #              define PACKET3_DMA_DATA_CMD_SAIC    (1 << 28)
 #              define PACKET3_DMA_DATA_CMD_DAIC    (1 << 29)
 #              define PACKET3_DMA_DATA_CMD_RAW_WAIT  (1 << 30)
-#define	PACKET3_AQUIRE_MEM				0x58
+#define	PACKET3_ACQUIRE_MEM				0x58
 #define	PACKET3_REWIND					0x59
 #define	PACKET3_LOAD_UCONFIG_REG			0x5E
 #define	PACKET3_LOAD_SH_REG				0x5F