drm/amdgpu: cleanup conditional execution

First of all calculating the number of dw to patch into a
conditional execution is not something HW generation specific.
This is just standard ring buffer calculations. While at it also
reduce the BUG_ON() into WARN_ON().

Then instead of a random bit pattern use 0 as default value for
the number of dw skipped, this way it's not mandatory any more
to patch the conditional execution.

And last make the address to check a parameter of the
conditional execution instead of getting this from the ring.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Author: ChristianKoenigAMD

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

@@ -131,18 +131,18 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned int num_ibs,
 	struct amdgpu_ib *ib = &ibs[0];
 	struct dma_fence *tmp = NULL;
 	bool need_ctx_switch;
-	unsigned int patch_offset = ~0;
 	struct amdgpu_vm *vm;
 	uint64_t fence_ctx;
 	uint32_t status = 0, alloc_size;
 	unsigned int fence_flags = 0;
 	bool secure, init_shadow;
 	u64 shadow_va, csa_va, gds_va;
 	int vmid = AMDGPU_JOB_GET_VMID(job);
+	bool need_pipe_sync = false;
+	unsigned int cond_exec;
 
 	unsigned int i;
 	int r = 0;
-	bool need_pipe_sync = false;
 
 	if (num_ibs == 0)
 		return -EINVAL;
@@ -228,7 +228,8 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned int num_ibs,
 					    init_shadow, vmid);
 
 	if (ring->funcs->init_cond_exec)
-		patch_offset = amdgpu_ring_init_cond_exec(ring);
+		cond_exec = amdgpu_ring_init_cond_exec(ring,
+						       ring->cond_exe_gpu_addr);
 
 	amdgpu_device_flush_hdp(adev, ring);
 
@@ -278,16 +279,9 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned int num_ibs,
 				       fence_flags | AMDGPU_FENCE_FLAG_64BIT);
 	}
 
-	if (ring->funcs->emit_gfx_shadow) {
+	if (ring->funcs->emit_gfx_shadow && ring->funcs->init_cond_exec) {
 		amdgpu_ring_emit_gfx_shadow(ring, 0, 0, 0, false, 0);
-
-		if (ring->funcs->init_cond_exec) {
-			unsigned int ce_offset = ~0;
-
-			ce_offset = amdgpu_ring_init_cond_exec(ring);
-			if (ce_offset != ~0 && ring->funcs->patch_cond_exec)
-				amdgpu_ring_patch_cond_exec(ring, ce_offset);
-		}
+		amdgpu_ring_init_cond_exec(ring, ring->cond_exe_gpu_addr);
 	}
 
 	r = amdgpu_fence_emit(ring, f, job, fence_flags);
@@ -302,8 +296,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned int num_ibs,
 	if (ring->funcs->insert_end)
 		ring->funcs->insert_end(ring);
 
-	if (patch_offset != ~0 && ring->funcs->patch_cond_exec)
-		amdgpu_ring_patch_cond_exec(ring, patch_offset);
+	amdgpu_ring_patch_cond_exec(ring, cond_exec);
 
 	ring->current_ctx = fence_ctx;
 	if (vm && ring->funcs->emit_switch_buffer)

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

@@ -209,8 +209,7 @@ struct amdgpu_ring_funcs {
 	void (*insert_end)(struct amdgpu_ring *ring);
 	/* pad the indirect buffer to the necessary number of dw */
 	void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
-	unsigned (*init_cond_exec)(struct amdgpu_ring *ring);
-	void (*patch_cond_exec)(struct amdgpu_ring *ring, unsigned offset);
+	unsigned (*init_cond_exec)(struct amdgpu_ring *ring, uint64_t addr);
 	/* note usage for clock and power gating */
 	void (*begin_use)(struct amdgpu_ring *ring);
 	void (*end_use)(struct amdgpu_ring *ring);
@@ -327,8 +326,7 @@ struct amdgpu_ring {
 #define amdgpu_ring_emit_reg_write_reg_wait(r, d0, d1, v, m) (r)->funcs->emit_reg_write_reg_wait((r), (d0), (d1), (v), (m))
 #define amdgpu_ring_emit_frame_cntl(r, b, s) (r)->funcs->emit_frame_cntl((r), (b), (s))
 #define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib)))
-#define amdgpu_ring_init_cond_exec(r) (r)->funcs->init_cond_exec((r))
-#define amdgpu_ring_patch_cond_exec(r,o) (r)->funcs->patch_cond_exec((r),(o))
+#define amdgpu_ring_init_cond_exec(r, a) (r)->funcs->init_cond_exec((r), (a))
 #define amdgpu_ring_preempt_ib(r) (r)->funcs->preempt_ib(r)
 #define amdgpu_ring_patch_cntl(r, o) ((r)->funcs->patch_cntl((r), (o)))
 #define amdgpu_ring_patch_ce(r, o) ((r)->funcs->patch_ce((r), (o)))
@@ -411,6 +409,30 @@ static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring,
 	ring->count_dw -= count_dw;
 }
 
+/**
+ * amdgpu_ring_patch_cond_exec - patch dw count of conditional execute
+ * @ring: amdgpu_ring structure
+ * @offset: offset returned by amdgpu_ring_init_cond_exec
+ *
+ * Calculate the dw count and patch it into a cond_exec command.
+ */
+static inline void amdgpu_ring_patch_cond_exec(struct amdgpu_ring *ring,
+					       unsigned int offset)
+{
+	unsigned cur;
+
+	if (!ring->funcs->init_cond_exec)
+		return;
+
+	WARN_ON(offset > ring->buf_mask);
+	WARN_ON(ring->ring[offset] != 0);
+
+	cur = (ring->wptr - 1) & ring->buf_mask;
+	if (cur < offset)
+		cur += ring->ring_size >> 2;
+	ring->ring[offset] = cur - offset;
+}
+
 #define amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset)			\
 	(ring->is_mes_queue && ring->mes_ctx ?				\
 	 (ring->mes_ctx->meta_data_gpu_addr + offset) : 0)

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

@@ -658,7 +658,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 	bool vm_flush_needed = job->vm_needs_flush;
 	struct dma_fence *fence = NULL;
 	bool pasid_mapping_needed = false;
-	unsigned patch_offset = 0;
+	unsigned int patch;
 	int r;
 
 	if (amdgpu_vmid_had_gpu_reset(adev, id)) {
@@ -685,7 +685,8 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 
 	amdgpu_ring_ib_begin(ring);
 	if (ring->funcs->init_cond_exec)
-		patch_offset = amdgpu_ring_init_cond_exec(ring);
+		patch = amdgpu_ring_init_cond_exec(ring,
+						   ring->cond_exe_gpu_addr);
 
 	if (need_pipe_sync)
 		amdgpu_ring_emit_pipeline_sync(ring);
@@ -733,8 +734,7 @@ int amdgpu_vm_flush(struct amdgpu_ring *ring, struct amdgpu_job *job,
 	}
 	dma_fence_put(fence);
 
-	if (ring->funcs->patch_cond_exec)
-		amdgpu_ring_patch_cond_exec(ring, patch_offset);
+	amdgpu_ring_patch_cond_exec(ring, patch);
 
 	/* the double SWITCH_BUFFER here *cannot* be skipped by COND_EXEC */
 	if (ring->funcs->emit_switch_buffer) {

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

@@ -546,34 +546,21 @@ static void vpe_ring_emit_vm_flush(struct amdgpu_ring *ring, unsigned int vmid,
 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
 }
 
-static unsigned int vpe_ring_init_cond_exec(struct amdgpu_ring *ring)
+static unsigned int vpe_ring_init_cond_exec(struct amdgpu_ring *ring,
+					    uint64_t addr)
 {
 	unsigned int ret;
 
 	amdgpu_ring_write(ring, VPE_CMD_HEADER(VPE_CMD_OPCODE_COND_EXE, 0));
-	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
+	amdgpu_ring_write(ring, lower_32_bits(addr));
+	amdgpu_ring_write(ring, upper_32_bits(addr));
 	amdgpu_ring_write(ring, 1);
-	ret = ring->wptr & ring->buf_mask;/* this is the offset we need patch later */
-	amdgpu_ring_write(ring, 0x55aa55aa);/* insert dummy here and patch it later */
+	ret = ring->wptr & ring->buf_mask;
+	amdgpu_ring_write(ring, 0);
 
 	return ret;
 }
 
-static void vpe_ring_patch_cond_exec(struct amdgpu_ring *ring, unsigned int offset)
-{
-	unsigned int cur;
-
-	WARN_ON_ONCE(offset > ring->buf_mask);
-	WARN_ON_ONCE(ring->ring[offset] != 0x55aa55aa);
-
-	cur = (ring->wptr - 1) & ring->buf_mask;
-	if (cur > offset)
-		ring->ring[offset] = cur - offset;
-	else
-		ring->ring[offset] = (ring->buf_mask + 1) - offset + cur;
-}
-
 static int vpe_ring_preempt_ib(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
@@ -864,7 +851,6 @@ static const struct amdgpu_ring_funcs vpe_ring_funcs = {
 	.test_ring = vpe_ring_test_ring,
 	.test_ib = vpe_ring_test_ib,
 	.init_cond_exec = vpe_ring_init_cond_exec,
-	.patch_cond_exec = vpe_ring_patch_cond_exec,
 	.preempt_ib = vpe_ring_preempt_ib,
 	.begin_use = vpe_ring_begin_use,
 	.end_use = vpe_ring_end_use,

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

@@ -8542,34 +8542,23 @@ static void gfx_v10_0_ring_emit_cntxcntl(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, 0);
 }
 
-static unsigned int gfx_v10_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
+static unsigned int gfx_v10_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
+						       uint64_t addr)
 {
 	unsigned int ret;
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
-	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, lower_32_bits(addr));
+	amdgpu_ring_write(ring, upper_32_bits(addr));
+	/* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, 0);
 	ret = ring->wptr & ring->buf_mask;
-	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
+	/* patch dummy value later */
+	amdgpu_ring_write(ring, 0);
 
 	return ret;
 }
 
-static void gfx_v10_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned int offset)
-{
-	unsigned int cur;
-
-	BUG_ON(offset > ring->buf_mask);
-	BUG_ON(ring->ring[offset] != 0x55aa55aa);
-
-	cur = (ring->wptr - 1) & ring->buf_mask;
-	if (likely(cur > offset))
-		ring->ring[offset] = cur - offset;
-	else
-		ring->ring[offset] = (ring->buf_mask + 1) - offset + cur;
-}
-
 static int gfx_v10_0_ring_preempt_ib(struct amdgpu_ring *ring)
 {
 	int i, r = 0;
@@ -9224,7 +9213,6 @@ static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_gfx = {
 	.emit_switch_buffer = gfx_v10_0_ring_emit_sb,
 	.emit_cntxcntl = gfx_v10_0_ring_emit_cntxcntl,
 	.init_cond_exec = gfx_v10_0_ring_emit_init_cond_exec,
-	.patch_cond_exec = gfx_v10_0_ring_emit_patch_cond_exec,
 	.preempt_ib = gfx_v10_0_ring_preempt_ib,
 	.emit_frame_cntl = gfx_v10_0_ring_emit_frame_cntl,
 	.emit_wreg = gfx_v10_0_ring_emit_wreg,

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

@@ -5533,33 +5533,23 @@ static void gfx_v11_0_ring_emit_gfx_shadow(struct amdgpu_ring *ring,
 			  PACKET3_SET_Q_PREEMPTION_MODE_INIT_SHADOW_MEM : 0);
 }
 
-static unsigned gfx_v11_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
+static unsigned gfx_v11_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
+						   uint64_t addr)
 {
 	unsigned ret;
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
-	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, lower_32_bits(addr));
+	amdgpu_ring_write(ring, upper_32_bits(addr));
+	/* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, 0);
 	ret = ring->wptr & ring->buf_mask;
-	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
+	/* patch dummy value later */
+	amdgpu_ring_write(ring, 0);
 
 	return ret;
 }
 
-static void gfx_v11_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
-{
-	unsigned cur;
-	BUG_ON(offset > ring->buf_mask);
-	BUG_ON(ring->ring[offset] != 0x55aa55aa);
-
-	cur = (ring->wptr - 1) & ring->buf_mask;
-	if (likely(cur > offset))
-		ring->ring[offset] = cur - offset;
-	else
-		ring->ring[offset] = (ring->buf_mask + 1) - offset + cur;
-}
-
 static int gfx_v11_0_ring_preempt_ib(struct amdgpu_ring *ring)
 {
 	int i, r = 0;
@@ -6153,7 +6143,6 @@ static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_gfx = {
 	.emit_cntxcntl = gfx_v11_0_ring_emit_cntxcntl,
 	.emit_gfx_shadow = gfx_v11_0_ring_emit_gfx_shadow,
 	.init_cond_exec = gfx_v11_0_ring_emit_init_cond_exec,
-	.patch_cond_exec = gfx_v11_0_ring_emit_patch_cond_exec,
 	.preempt_ib = gfx_v11_0_ring_preempt_ib,
 	.emit_frame_cntl = gfx_v11_0_ring_emit_frame_cntl,
 	.emit_wreg = gfx_v11_0_ring_emit_wreg,

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

@@ -6326,33 +6326,22 @@ static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
 	amdgpu_ring_write(ring, 0);
 }
 
-static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
+static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
+						  uint64_t addr)
 {
 	unsigned ret;
 
 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
-	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, lower_32_bits(addr));
+	amdgpu_ring_write(ring, upper_32_bits(addr));
+	/* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, 0);
 	ret = ring->wptr & ring->buf_mask;
-	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
+	/* patch dummy value later */
+	amdgpu_ring_write(ring, 0);
 	return ret;
 }
 
-static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
-{
-	unsigned cur;
-
-	BUG_ON(offset > ring->buf_mask);
-	BUG_ON(ring->ring[offset] != 0x55aa55aa);
-
-	cur = (ring->wptr & ring->buf_mask) - 1;
-	if (likely(cur > offset))
-		ring->ring[offset] = cur - offset;
-	else
-		ring->ring[offset] = (ring->ring_size >> 2) - offset + cur;
-}
-
 static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
 				    uint32_t reg_val_offs)
 {
@@ -6932,7 +6921,6 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.emit_switch_buffer = gfx_v8_ring_emit_sb,
 	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
 	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
-	.patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec,
 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
 	.soft_recovery = gfx_v8_0_ring_soft_recovery,
 	.emit_mem_sync = gfx_v8_0_emit_mem_sync,

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

@@ -5610,31 +5610,21 @@ static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
 	amdgpu_ring_write(ring, 0);
 }
 
-static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
+static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
+						  uint64_t addr)
 {
 	unsigned ret;
 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
-	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
-	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, lower_32_bits(addr));
+	amdgpu_ring_write(ring, upper_32_bits(addr));
+	/* discard following DWs if *cond_exec_gpu_addr==0 */
+	amdgpu_ring_write(ring, 0);
 	ret = ring->wptr & ring->buf_mask;
-	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
+	/* patch dummy value later */
+	amdgpu_ring_write(ring, 0);
 	return ret;
 }
 
-static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
-{
-	unsigned cur;
-	BUG_ON(offset > ring->buf_mask);
-	BUG_ON(ring->ring[offset] != 0x55aa55aa);
-
-	cur = (ring->wptr - 1) & ring->buf_mask;
-	if (likely(cur > offset))
-		ring->ring[offset] = cur - offset;
-	else
-		ring->ring[offset] = (ring->ring_size>>2) - offset + cur;
-}
-
 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
 				    uint32_t reg_val_offs)
 {
@@ -6908,7 +6898,6 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
 	.emit_switch_buffer = gfx_v9_ring_emit_sb,
 	.emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
 	.init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
-	.patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
 	.preempt_ib = gfx_v9_0_ring_preempt_ib,
 	.emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
@@ -6963,7 +6952,6 @@ static const struct amdgpu_ring_funcs gfx_v9_0_sw_ring_funcs_gfx = {
 	.emit_switch_buffer = gfx_v9_ring_emit_sb,
 	.emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
 	.init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
-	.patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
 	.emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,