Skip to contentSkip to conversation

drm/amdgpu: update cp cmd pkt for gfx v12_1

PATCH New Alex Deucher 1 participant 3 messages Apr 22

Subsystems (2)
lore
Share & tools
lore

Alex Deucher

author maintainer
#0 lore raw

PATCH 1/1 drm/amdgpu: update cp cmd pkt for gfx v12_1

From: Likun Gao <[email protected]>

Update CP command package for gfx v12.1.

v2: squash in update (Alex)

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c     |  34 +-
 drivers/gpu/drm/amd/amdgpu/gfx_v12_1_pkt.h | 347 +++++++++++++++------
 2 files changed, 273 insertions(+), 108 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c +17 -17
index 948758b51b5cc..f4089ab108474 100644
@@ -243,9 +243,9 @@ static void gfx_v12_1_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
 	amdgpu_ring_write(ring,
 			  /* memory (1) or register (0) */
-			  (WAIT_REG_MEM_MEM_SPACE(mem_space) |
-			   WAIT_REG_MEM_OPERATION(opt) | /* wait */
-			   WAIT_REG_MEM_FUNCTION(3)));  /* equal */
+			  (PACKET3_WAIT_REG_MEM__MEM_SPACE(mem_space) |
+			   PACKET3_WAIT_REG_MEM__OPERATION(opt) | /* wait */
+			   PACKET3_WAIT_REG_MEM__FUNCTION(3)));  /* equal */
 
 	if (mem_space)
 		BUG_ON(addr0 & 0x3); /* Dword align */
@@ -335,7 +335,7 @@ static int gfx_v12_1_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 	}
 
 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
-	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
+	ib.ptr[1] = PACKET3_WRITE_DATA__DST_SEL(5) | PACKET3_WRITE_DATA__WR_CONFIRM(1);
 	ib.ptr[2] = lower_32_bits(gpu_addr);
 	ib.ptr[3] = upper_32_bits(gpu_addr);
 	ib.ptr[4] = 0xDEADBEEF;
@@ -3366,7 +3366,7 @@ static void gfx_v12_1_ring_emit_ib_compute(struct amdgpu_ring *ring,
 					   uint32_t flags)
 {
 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
-	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
+	u32 control = PACKET3_INDIRECT_BUFFER__VALID(1) | ib->length_dw | (vmid << 24);
 
 	/* Currently, there is a high possibility to get wave ID mismatch
 	 * between ME and GDS, leading to a hw deadlock, because ME generates
@@ -3402,15 +3402,15 @@ static void gfx_v12_1_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
 
 	/* RELEASE_MEM - flush caches, send int */
 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
-	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_GCR_SEQ(1) |
-				 PACKET3_RELEASE_MEM_GCR_GLV_WB |
-				 PACKET3_RELEASE_MEM_GCR_GL2_WB |
-				 PACKET3_RELEASE_MEM_GCR_GL2_SCOPE(2) |
-				 PACKET3_RELEASE_MEM_TEMPORAL(3) |
-				 PACKET3_RELEASE_MEM_EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
-				 PACKET3_RELEASE_MEM_EVENT_INDEX(5)));
-	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_DATA_SEL(write64bit ? 2 : 1) |
-				 PACKET3_RELEASE_MEM_INT_SEL(int_sel ? 2 : 0)));
+	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM__GCR_SEQ(1) |
+				 PACKET3_RELEASE_MEM__GCR_GLV_WB |
+				 PACKET3_RELEASE_MEM__GCR_GL2_WB |
+				 PACKET3_RELEASE_MEM__GCR_GL2_SCOPE(2) |
+				 PACKET3_RELEASE_MEM__TEMPORAL(3) |
+				 PACKET3_RELEASE_MEM__EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
+				 PACKET3_RELEASE_MEM__EVENT_INDEX(5)));
+	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM__DATA_SEL(write64bit ? 2 : 1) |
+				 PACKET3_RELEASE_MEM__INT_SEL(int_sel ? 2 : 0)));
 
 	/*
 	 * the address should be Qword aligned if 64bit write, Dword
@@ -3471,7 +3471,7 @@ static void gfx_v12_1_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
 
 	/* write fence seq to the "addr" */
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-	amdgpu_ring_write(ring, (WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
+	amdgpu_ring_write(ring, (PACKET3_WRITE_DATA__DST_SEL(5) | PACKET3_WRITE_DATA__WR_CONFIRM(1)));
 	amdgpu_ring_write(ring, lower_32_bits(addr));
 	amdgpu_ring_write(ring, upper_32_bits(addr));
 	amdgpu_ring_write(ring, lower_32_bits(seq));
@@ -3479,7 +3479,7 @@ static void gfx_v12_1_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
 	if (flags & AMDGPU_FENCE_FLAG_INT) {
 		/* set register to trigger INT */
 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
-		amdgpu_ring_write(ring, (WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
+		amdgpu_ring_write(ring, (PACKET3_WRITE_DATA__DST_SEL(0) | PACKET3_WRITE_DATA__WR_CONFIRM(1)));
 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regCPC_INT_STATUS));
 		amdgpu_ring_write(ring, 0);
 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
@@ -3518,7 +3518,7 @@ static void gfx_v12_1_ring_emit_wreg(struct amdgpu_ring *ring,
 		cmd = (1 << 16); /* no inc addr */
 		break;
 	default:
-		cmd = WR_CONFIRM;
+		cmd = PACKET3_WRITE_DATA__WR_CONFIRM(1);
 		break;
 	}
 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
drivers/gpu/drm/amd/amdgpu/gfx_v12_1_pkt.h +256 -91

424 lines not on this page · raw message

Signature · 1 line
2.53.0
author maintainer
#1 lore raw

PATCH 1/1 drm/amdgpu: In GFX12.1 CU is same as WGP

From: Harish Kasiviswanathan <[email protected]>

Fix this for current ip discovery table.

v2: Move the change to gfx12 file
v3: Change only for the current version of ip discovery table
v4: Squash in build fixes

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 25 +++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.h |  2 ++
 drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c        | 10 ++++++++
 3 files changed, 37 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c +25 -0
index 193091583847b..fef90a1d69441 100644
@@ -3423,3 +3423,28 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 	return 0;
 }
 
+int amdgpu_discovery_get_gc_major_minor_version(struct amdgpu_device *adev,
+						uint16_t *major, uint16_t *minor)
+{
+	uint8_t *discovery_bin = adev->discovery.bin;
+	struct table_info *info;
+	union gc_info *gc_info;
+	u16 offset;
+
+	if (!discovery_bin)
+		return -EINVAL;
+	if (amdgpu_discovery_get_table_info(adev, &info, GC))
+		return -EINVAL;
+
+	offset = le16_to_cpu(info->offset);
+	if (!offset)
+		return -EINVAL;
+
+	gc_info = (union gc_info *)(discovery_bin + offset);
+
+	if (major)
+		*major = le16_to_cpu(gc_info->v1.header.version_major);
+	if (minor)
+		*minor = le16_to_cpu(gc_info->v1.header.version_minor);
+	return 0;
+}
drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.h +2 -0
index 0ff1a7923eedf..e0010f6a3eda5 100644
@@ -48,6 +48,8 @@ int amdgpu_discovery_get_nps_info(struct amdgpu_device *adev,
 				  uint32_t *nps_type,
 				  struct amdgpu_gmc_memrange *ranges,
 				  int *range_cnt, bool refresh);
+int amdgpu_discovery_get_gc_major_minor_version(struct amdgpu_device *adev,
+						uint16_t *major, uint16_t *minor);
 
 void amdgpu_discovery_dump(struct amdgpu_device *adev, struct drm_printer *p);
 
drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c +10 -0
index f4089ab108474..6872ce3f3ebbe 100644
@@ -1142,6 +1142,7 @@ static int gfx_v12_1_rlc_backdoor_autoload_enable(struct amdgpu_device *adev)
 
 static int gfx_v12_1_sw_init(struct amdgpu_ip_block *ip_block)
 {
+	uint16_t major_ver, minor_ver;
 	int i, j, k, r, ring_id = 0;
 	unsigned num_compute_rings;
 	int xcc_id, num_xcc;
@@ -1152,6 +1153,15 @@ static int gfx_v12_1_sw_init(struct amdgpu_ip_block *ip_block)
 		adev->gfx.mec.num_mec = 1;
 		adev->gfx.mec.num_pipe_per_mec = 4;
 		adev->gfx.mec.num_queue_per_pipe = 8;
+
+		if (!amdgpu_discovery_get_gc_major_minor_version(
+			    adev, &major_ver, &minor_ver)) {
+			if (major_ver == 1 && minor_ver == 3) {
+				adev->gfx.config.max_cu_per_sh /= 2;
+				dev_dbg(adev->dev, "Halving max_cu_per_sh for GC Discovery table v1:3 %d\n",
+					adev->gfx.config.max_cu_per_sh);
+			}
+		}
 		break;
 	default:
 		adev->gfx.mec.num_mec = 2;
Signature · 1 line
2.53.0
author maintainer
#2 lore raw

PATCH 1/1 drm/amdgpu: Switch to gfx_v12_1_get_xccs_per_xcp

In reply to Alex Deucher

From: Hawking Zhang <[email protected]>

Use gfx v12_1 callback to query the numbers of xccs per xcp

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c +11 -2
index 6872ce3f3ebbe..5412a528f78be 100644
@@ -718,10 +718,19 @@ static void gfx_v12_1_select_me_pipe_q(struct amdgpu_device *adev,
 	soc_v1_0_grbm_select(adev, me, pipe, q, vm, GET_INST(GC, xcc_id));
 }
 
+#define regGFX_IMU_PARTITION_SWITCH		0x5f8c
+#define regGFX_IMU_PARTITION_SWITCH_BASE_IDX	1
+#define GFX_IMU_PARTITION_SWITCH__TOTAL_XCCS_IN_XCP__SHIFT	0x2
+#define GFX_IMU_PARTITION_SWITCH__TOTAL_XCCS_IN_XCP_MASK		0x0000003CL
+
 static int gfx_v12_1_get_xccs_per_xcp(struct amdgpu_device *adev)
 {
-	/* Fill this in when the interface is ready */
-	return 1;
+	u32 reg_data;
+
+	/* the register data is expected to be the same on all instances */
+	reg_data = RREG32_SOC15(GC, GET_INST(GC, 0), regGFX_IMU_PARTITION_SWITCH);
+
+	return REG_GET_FIELD(reg_data, GFX_IMU_PARTITION_SWITCH, TOTAL_XCCS_IN_XCP);
 }
 
 static int gfx_v12_1_ih_to_xcc_inst(struct amdgpu_device *adev, int ih_node)
Signature · 1 line
2.53.0