Skip to contentSkip to conversation

GFX12 ipdump of gfx IP

PATCH In review Sunil Khatri 2 participants 6 messages Jul 3, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover GFX12 ipdump of gfx IP

Validated on navi48

Sunil Khatri (4):
drm/amdgpu: add gfx12 register support in ipdump drm/amdgpu: add print support for gfx12 ipdump drm/amdgpu: add cp queue registers for gfx12 ipdump drm/amdgpu: add gfx queue support for gfx12 ipdump

 drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 316 +++++++++++++++++++++++++
 1 file changed, 316 insertions(+)
Signature · 1 line
2.34.1
#1 lore raw

PATCH 1/4 drm/amdgpu: add gfx12 register support in ipdump

Add general registers of gfx12 in ipdump for
devcoredump support.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 101 +++++++++++++++++++++++++
 1 file changed, 101 insertions(+)
drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +101 -0
index ccb26f78252a..c06d0a2a03e9 100644
@@ -63,6 +63,73 @@ MODULE_FIRMWARE("amdgpu/gc_12_0_1_mec.bin");
 MODULE_FIRMWARE("amdgpu/gc_12_0_1_rlc.bin");
 MODULE_FIRMWARE("amdgpu/gc_12_0_1_toc.bin");
 
+static const struct amdgpu_hwip_reg_entry gc_reg_list_12_0[] = {
+	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS2),
+	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS3),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_STALLED_STAT1),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_STALLED_STAT2),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_STALLED_STAT3),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_STALLED_STAT1),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_STALLED_STAT1),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_BUSY_STAT),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_BUSY_STAT),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_BUSY_STAT),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_BUSY_STAT2),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_BUSY_STAT2),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPF_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_ERROR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HPD_STATUS0),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_BASE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_RPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_WPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB0_BASE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB0_RPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB0_WPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_CMD_BUFSZ),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB2_CMD_BUFSZ),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BASE_LO),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BASE_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BUFSZ),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB2_BASE_LO),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB2_BASE_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB2_BUFSZ),
+	SOC15_REG_ENTRY_STR(GC, 0, regCPF_UTCL1_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regCPC_UTCL1_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regCPG_UTCL1_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regIA_UTCL1_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regIA_UTCL1_STATUS_2),
+	SOC15_REG_ENTRY_STR(GC, 0, regPA_CL_CNTL_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regRMI_UTCL1_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regSQC_CACHES),
+	SOC15_REG_ENTRY_STR(GC, 0, regSQG_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regWD_UTCL1_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regGCVM_L2_PROTECTION_FAULT_CNTL),
+	SOC15_REG_ENTRY_STR(GC, 0, regGCVM_L2_PROTECTION_FAULT_STATUS_LO32),
+	SOC15_REG_ENTRY_STR(GC, 0, regGCVM_L2_PROTECTION_FAULT_STATUS_HI32),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_DEBUG),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_CNTL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_MES_CNTL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_MES_INSTR_PNTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_ME_INSTR_PNTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_PFP_INSTR_PNTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_CPC_STATUS),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_RS64_INSTR_PNTR0),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_RS64_INSTR_PNTR1),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_RS64_INSTR_PNTR),
+
+	/* cp header registers */
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_MEC_ME1_HEADER_DUMP),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_PFP_HEADER_DUMP),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_ME_HEADER_DUMP),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_MES_HEADER_DUMP),
+	/* SE status registers */
+	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE0),
+	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE1),
+	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE2),
+	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE3)
+};
+
 #define DEFAULT_SH_MEM_CONFIG \
 	((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
 	 (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
@@ -1129,6 +1196,20 @@ static int gfx_v12_0_rlc_backdoor_autoload_enable(struct amdgpu_device *adev)
 	return 0;
 }
 
+static void gfx_v12_0_alloc_ip_dump(struct amdgpu_device *adev)
+{
+	uint32_t reg_count = ARRAY_SIZE(gc_reg_list_12_0);
+	uint32_t *ptr;
+
+	ptr = kcalloc(reg_count, sizeof(uint32_t), GFP_KERNEL);
+	if (ptr == NULL) {
+		DRM_ERROR("Failed to allocate memory for GFX IP Dump\n");
+		adev->gfx.ip_dump_core = NULL;
+	} else {
+		adev->gfx.ip_dump_core = ptr;
+	}
+}
+
 static int gfx_v12_0_sw_init(void *handle)
 {
 	int i, j, k, r, ring_id = 0;
@@ -1261,6 +1342,8 @@ static int gfx_v12_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	gfx_v12_0_alloc_ip_dump(adev);
+
 	return 0;
 }
 
@@ -1320,6 +1403,8 @@ static int gfx_v12_0_sw_fini(void *handle)
 
 	gfx_v12_0_free_microcode(adev);
 
+	kfree(adev->gfx.ip_dump_core);
+
 	return 0;
 }
 
@@ -4673,6 +4758,21 @@ static void gfx_v12_0_emit_mem_sync(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, gcr_cntl); /* GCR_CNTL */
 }
 
+static void gfx_v12_ip_dump(void *handle)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	uint32_t i;
+	uint32_t reg_count = ARRAY_SIZE(gc_reg_list_12_0);
+
+	if (!adev->gfx.ip_dump_core)
+		return;
+
+	amdgpu_gfx_off_ctrl(adev, false);
+	for (i = 0; i < reg_count; i++)
+		adev->gfx.ip_dump_core[i] = RREG32(SOC15_REG_ENTRY_OFFSET(gc_reg_list_12_0[i]));
+	amdgpu_gfx_off_ctrl(adev, true);
+}
+
 static const struct amd_ip_funcs gfx_v12_0_ip_funcs = {
 	.name = "gfx_v12_0",
 	.early_init = gfx_v12_0_early_init,
@@ -4688,6 +4788,7 @@ static const struct amd_ip_funcs gfx_v12_0_ip_funcs = {
 	.set_clockgating_state = gfx_v12_0_set_clockgating_state,
 	.set_powergating_state = gfx_v12_0_set_powergating_state,
 	.get_clockgating_state = gfx_v12_0_get_clockgating_state,
+	.dump_ip_state = gfx_v12_ip_dump,
 };
 
 static const struct amdgpu_ring_funcs gfx_v12_0_ring_funcs_gfx = {
Signature · 1 line
2.34.1
#2 lore raw

PATCH 2/4 drm/amdgpu: add print support for gfx12 ipdump

In reply to Sunil Khatri

Add support of gfx12 ipdump print so devcoredump
could trigger it to dump the captured registers
in devcoredump.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +16 -0
index c06d0a2a03e9..19532409aec3 100644
@@ -4758,6 +4758,21 @@ static void gfx_v12_0_emit_mem_sync(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, gcr_cntl); /* GCR_CNTL */
 }
 
+static void gfx_v12_ip_print(void *handle, struct drm_printer *p)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	uint32_t i;
+	uint32_t reg_count = ARRAY_SIZE(gc_reg_list_12_0);
+
+	if (!adev->gfx.ip_dump_core)
+		return;
+
+	for (i = 0; i < reg_count; i++)
+		drm_printf(p, "%-50s \t 0x%08x\n",
+			   gc_reg_list_12_0[i].reg_name,
+			   adev->gfx.ip_dump_core[i]);
+}
+
 static void gfx_v12_ip_dump(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -4789,6 +4804,7 @@ static const struct amd_ip_funcs gfx_v12_0_ip_funcs = {
 	.set_powergating_state = gfx_v12_0_set_powergating_state,
 	.get_clockgating_state = gfx_v12_0_get_clockgating_state,
 	.dump_ip_state = gfx_v12_ip_dump,
+	.print_ip_state = gfx_v12_ip_print,
 };
 
 static const struct amdgpu_ring_funcs gfx_v12_0_ring_funcs_gfx = {
Signature · 1 line
2.34.1
#3 lore raw

PATCH 3/4 drm/amdgpu: add cp queue registers for gfx12 ipdump

In reply to Sunil Khatri

Add gfx12 support of CP queue registers for all queues to be used by devcoredump.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 111 ++++++++++++++++++++++++-
 1 file changed, 109 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +109 -2
index 19532409aec3..21b8167f39bd 100644
@@ -130,6 +130,49 @@ static const struct amdgpu_hwip_reg_entry gc_reg_list_12_0[] = {
 	SOC15_REG_ENTRY_STR(GC, 0, regGRBM_STATUS_SE3)
 };
 
+static const struct amdgpu_hwip_reg_entry gc_cp_reg_list_12[] = {
+	/* compute registers */
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_VMID),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PERSISTENT_STATE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PIPE_PRIORITY),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_QUEUE_PRIORITY),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_QUANTUM),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_BASE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_BASE_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_RPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_CONTROL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_BASE_ADDR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_BASE_ADDR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_RPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_IB_CONTROL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_DEQUEUE_REQUEST),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_BASE_ADDR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_BASE_ADDR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_CONTROL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_RPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_WPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_EVENTS),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_BASE_ADDR_LO),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_BASE_ADDR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_CONTROL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CNTL_STACK_OFFSET),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CNTL_STACK_SIZE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_WG_STATE_OFFSET),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_CTX_SAVE_SIZE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_GDS_RESOURCE_STATE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_ERROR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_EOP_WPTR_MEM),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_LO),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_PQ_WPTR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_SUSPEND_CNTL_STACK_OFFSET),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_SUSPEND_CNTL_STACK_DW_CNT),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_SUSPEND_WG_STATE_OFFSET),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_DEQUEUE_STATUS)
+};
+
 #define DEFAULT_SH_MEM_CONFIG \
 	((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
 	 (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
@@ -1200,6 +1243,7 @@ static void gfx_v12_0_alloc_ip_dump(struct amdgpu_device *adev)
 {
 	uint32_t reg_count = ARRAY_SIZE(gc_reg_list_12_0);
 	uint32_t *ptr;
+	uint32_t inst;
 
 	ptr = kcalloc(reg_count, sizeof(uint32_t), GFP_KERNEL);
 	if (ptr == NULL) {
@@ -1208,6 +1252,19 @@ static void gfx_v12_0_alloc_ip_dump(struct amdgpu_device *adev)
 	} else {
 		adev->gfx.ip_dump_core = ptr;
 	}
+
+	/* Allocate memory for compute queue registers for all the instances */
+	reg_count = ARRAY_SIZE(gc_cp_reg_list_12);
+	inst = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec *
+		adev->gfx.mec.num_queue_per_pipe;
+
+	ptr = kcalloc(reg_count * inst, sizeof(uint32_t), GFP_KERNEL);
+	if (ptr == NULL) {
+		DRM_ERROR("Failed to allocate memory for Compute Queues IP Dump\n");
+		adev->gfx.ip_dump_compute_queues = NULL;
+	} else {
+		adev->gfx.ip_dump_compute_queues = ptr;
+	}
 }
 
 static int gfx_v12_0_sw_init(void *handle)
@@ -1404,6 +1461,7 @@ static int gfx_v12_0_sw_fini(void *handle)
 	gfx_v12_0_free_microcode(adev);
 
 	kfree(adev->gfx.ip_dump_core);
+	kfree(adev->gfx.ip_dump_compute_queues);
 
 	return 0;
 }
@@ -4761,7 +4819,7 @@ static void gfx_v12_0_emit_mem_sync(struct amdgpu_ring *ring)
 static void gfx_v12_ip_print(void *handle, struct drm_printer *p)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-	uint32_t i;
+	uint32_t i, j, k, reg, index = 0;
 	uint32_t reg_count = ARRAY_SIZE(gc_reg_list_12_0);
 
 	if (!adev->gfx.ip_dump_core)
@@ -4771,12 +4829,36 @@ static void gfx_v12_ip_print(void *handle, struct drm_printer *p)
 		drm_printf(p, "%-50s \t 0x%08x\n",
 			   gc_reg_list_12_0[i].reg_name,
 			   adev->gfx.ip_dump_core[i]);
+
+	/* print compute queue registers for all instances */
+	if (!adev->gfx.ip_dump_compute_queues)
+		return;
+
+	reg_count = ARRAY_SIZE(gc_cp_reg_list_12);
+	drm_printf(p, "\nnum_mec: %d num_pipe: %d num_queue: %d\n",
+		   adev->gfx.mec.num_mec,
+		   adev->gfx.mec.num_pipe_per_mec,
+		   adev->gfx.mec.num_queue_per_pipe);
+
+	for (i = 0; i < adev->gfx.mec.num_mec; i++) {
+		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) {
+			for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) {
+				drm_printf(p, "\nmec %d, pipe %d, queue %d\n", i, j, k);
+				for (reg = 0; reg < reg_count; reg++) {
+					drm_printf(p, "%-50s \t 0x%08x\n",
+						   gc_cp_reg_list_12[reg].reg_name,
+						   adev->gfx.ip_dump_compute_queues[index + reg]);
+				}
+				index += reg_count;
+			}
+		}
+	}
 }
 
 static void gfx_v12_ip_dump(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-	uint32_t i;
+	uint32_t i, j, k, reg, index = 0;
 	uint32_t reg_count = ARRAY_SIZE(gc_reg_list_12_0);
 
 	if (!adev->gfx.ip_dump_core)
@@ -4786,6 +4868,31 @@ static void gfx_v12_ip_dump(void *handle)
 	for (i = 0; i < reg_count; i++)
 		adev->gfx.ip_dump_core[i] = RREG32(SOC15_REG_ENTRY_OFFSET(gc_reg_list_12_0[i]));
 	amdgpu_gfx_off_ctrl(adev, true);
+
+	/* dump compute queue registers for all instances */
+	if (!adev->gfx.ip_dump_compute_queues)
+		return;
+
+	reg_count = ARRAY_SIZE(gc_cp_reg_list_12);
+	amdgpu_gfx_off_ctrl(adev, false);
+	mutex_lock(&adev->srbm_mutex);
+	for (i = 0; i < adev->gfx.mec.num_mec; i++) {
+		for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) {
+			for (k = 0; k < adev->gfx.mec.num_queue_per_pipe; k++) {
+				/* ME0 is for GFX so start from 1 for CP */
+				soc24_grbm_select(adev, 1+i, j, k, 0);
+				for (reg = 0; reg < reg_count; reg++) {
+					adev->gfx.ip_dump_compute_queues[index + reg] =
+						RREG32(SOC15_REG_ENTRY_OFFSET(
+							gc_cp_reg_list_12[reg]));
+				}
+				index += reg_count;
+			}
+		}
+	}
+	soc24_grbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+	amdgpu_gfx_off_ctrl(adev, true);
 }
 
 static const struct amd_ip_funcs gfx_v12_0_ip_funcs = {
Signature · 1 line
2.34.1
#4 lore raw

PATCH 4/4 drm/amdgpu: add gfx queue support for gfx12 ipdump

In reply to Sunil Khatri

Add support of all the CP GFX queues for gfx12 ipdump to be used by devcoredump.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 94 ++++++++++++++++++++++++++
 1 file changed, 94 insertions(+)
drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +94 -0
index 21b8167f39bd..cf7209b93617 100644
@@ -173,6 +173,35 @@ static const struct amdgpu_hwip_reg_entry gc_cp_reg_list_12[] = {
 	SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_DEQUEUE_STATUS)
 };
 
+static const struct amdgpu_hwip_reg_entry gc_gfx_queue_reg_list_12[] = {
+	/* gfx queue registers */
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_ACTIVE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_VMID),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_QUEUE_PRIORITY),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_QUANTUM),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_BASE),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_BASE_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_OFFSET),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_CNTL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_CSMD_RPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_WPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_WPTR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_DEQUEUE_REQUEST),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_MAPPED),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_QUE_MGR_CONTROL),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_HQ_CONTROL0),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_HQ_STATUS0),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_MQD_BASE_ADDR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_MQD_BASE_ADDR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_RPTR),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BASE_LO),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BASE_HI),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_CMD_BUFSZ),
+	SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BUFSZ)
+};
+
 #define DEFAULT_SH_MEM_CONFIG \
 	((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
 	 (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
@@ -1265,6 +1294,19 @@ static void gfx_v12_0_alloc_ip_dump(struct amdgpu_device *adev)
 	} else {
 		adev->gfx.ip_dump_compute_queues = ptr;
 	}
+
+	/* Allocate memory for gfx queue registers for all the instances */
+	reg_count = ARRAY_SIZE(gc_gfx_queue_reg_list_12);
+	inst = adev->gfx.me.num_me * adev->gfx.me.num_pipe_per_me *
+		adev->gfx.me.num_queue_per_pipe;
+
+	ptr = kcalloc(reg_count * inst, sizeof(uint32_t), GFP_KERNEL);
+	if (ptr == NULL) {
+		DRM_ERROR("Failed to allocate memory for GFX Queues IP Dump\n");
+		adev->gfx.ip_dump_gfx_queues = NULL;
+	} else {
+		adev->gfx.ip_dump_gfx_queues = ptr;
+	}
 }
 
 static int gfx_v12_0_sw_init(void *handle)
@@ -1462,6 +1504,7 @@ static int gfx_v12_0_sw_fini(void *handle)
 
 	kfree(adev->gfx.ip_dump_core);
 	kfree(adev->gfx.ip_dump_compute_queues);
+	kfree(adev->gfx.ip_dump_gfx_queues);
 
 	return 0;
 }
@@ -4853,6 +4896,31 @@ static void gfx_v12_ip_print(void *handle, struct drm_printer *p)
 			}
 		}
 	}
+
+	/* print gfx queue registers for all instances */
+	if (!adev->gfx.ip_dump_gfx_queues)
+		return;
+
+	index = 0;
+	reg_count = ARRAY_SIZE(gc_gfx_queue_reg_list_12);
+	drm_printf(p, "\nnum_me: %d num_pipe: %d num_queue: %d\n",
+		   adev->gfx.me.num_me,
+		   adev->gfx.me.num_pipe_per_me,
+		   adev->gfx.me.num_queue_per_pipe);
+
+	for (i = 0; i < adev->gfx.me.num_me; i++) {
+		for (j = 0; j < adev->gfx.me.num_pipe_per_me; j++) {
+			for (k = 0; k < adev->gfx.me.num_queue_per_pipe; k++) {
+				drm_printf(p, "\nme %d, pipe %d, queue %d\n", i, j, k);
+				for (reg = 0; reg < reg_count; reg++) {
+					drm_printf(p, "%-50s \t 0x%08x\n",
+						   gc_gfx_queue_reg_list_12[reg].reg_name,
+						   adev->gfx.ip_dump_gfx_queues[index + reg]);
+				}
+				index += reg_count;
+			}
+		}
+	}
 }
 
 static void gfx_v12_ip_dump(void *handle)
@@ -4893,6 +4961,32 @@ static void gfx_v12_ip_dump(void *handle)
 	soc24_grbm_select(adev, 0, 0, 0, 0);
 	mutex_unlock(&adev->srbm_mutex);
 	amdgpu_gfx_off_ctrl(adev, true);
+
+	/* dump gfx queue registers for all instances */
+	if (!adev->gfx.ip_dump_gfx_queues)
+		return;
+
+	index = 0;
+	reg_count = ARRAY_SIZE(gc_gfx_queue_reg_list_12);
+	amdgpu_gfx_off_ctrl(adev, false);
+	mutex_lock(&adev->srbm_mutex);
+	for (i = 0; i < adev->gfx.me.num_me; i++) {
+		for (j = 0; j < adev->gfx.me.num_pipe_per_me; j++) {
+			for (k = 0; k < adev->gfx.me.num_queue_per_pipe; k++) {
+				soc24_grbm_select(adev, i, j, k, 0);
+
+				for (reg = 0; reg < reg_count; reg++) {
+					adev->gfx.ip_dump_gfx_queues[index + reg] =
+						RREG32(SOC15_REG_ENTRY_OFFSET(
+							gc_gfx_queue_reg_list_12[reg]));
+				}
+				index += reg_count;
+			}
+		}
+	}
+	soc24_grbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+	amdgpu_gfx_off_ctrl(adev, true);
 }
 
 static const struct amd_ip_funcs gfx_v12_0_ip_funcs = {
Signature · 1 line
2.34.1
Reviewed-by on v1 4/4
#5 lore raw

Re: [PATCH v1 4/4] drm/amdgpu: add gfx queue support for gfx12 ipdump

Series is:

Quoted 145 lines · Sunil Khatri

Add support of all the CP GFX queues for gfx12 ipdump to be used by devcoredump.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 94 ++++++++++++++++++++++++++
 1 file changed, 94 insertions(+)
drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +94 -1
index 21b8167f39bd..cf7209b93617 100644
@@ -173,6 +173,35 @@ static const struct amdgpu_hwip_reg_entry gc_cp_reg_list_12[] = {
        SOC15_REG_ENTRY_STR(GC, 0, regCP_HQD_DEQUEUE_STATUS)
 };
 
+static const struct amdgpu_hwip_reg_entry gc_gfx_queue_reg_list_12[] = {
+       /* gfx queue registers */
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_ACTIVE),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_VMID),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_QUEUE_PRIORITY),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_QUANTUM),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_BASE),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_BASE_HI),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_OFFSET),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_CNTL),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_CSMD_RPTR),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_WPTR),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_WPTR_HI),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_DEQUEUE_REQUEST),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_MAPPED),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_QUE_MGR_CONTROL),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_HQ_CONTROL0),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_HQD_HQ_STATUS0),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_MQD_BASE_ADDR),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_GFX_MQD_BASE_ADDR_HI),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_RB_RPTR),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BASE_LO),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BASE_HI),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_CMD_BUFSZ),
+       SOC15_REG_ENTRY_STR(GC, 0, regCP_IB1_BUFSZ)
+};
+
 #define DEFAULT_SH_MEM_CONFIG \
        ((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
         (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
@@ -1265,6 +1294,19 @@ static void gfx_v12_0_alloc_ip_dump(struct amdgpu_device *adev)
        } else {
                adev->gfx.ip_dump_compute_queues = ptr;
        }
+
+       /* Allocate memory for gfx queue registers for all the instances */
+       reg_count = ARRAY_SIZE(gc_gfx_queue_reg_list_12);
+       inst = adev->gfx.me.num_me * adev->gfx.me.num_pipe_per_me *
+               adev->gfx.me.num_queue_per_pipe;
+
+       ptr = kcalloc(reg_count * inst, sizeof(uint32_t), GFP_KERNEL);
+       if (ptr == NULL) {
+               DRM_ERROR("Failed to allocate memory for GFX Queues IP Dump\n");
+               adev->gfx.ip_dump_gfx_queues = NULL;
+       } else {
+               adev->gfx.ip_dump_gfx_queues = ptr;
+       }
 }
 
 static int gfx_v12_0_sw_init(void *handle)
@@ -1462,6 +1504,7 @@ static int gfx_v12_0_sw_fini(void *handle)
 
        kfree(adev->gfx.ip_dump_core);
        kfree(adev->gfx.ip_dump_compute_queues);
+       kfree(adev->gfx.ip_dump_gfx_queues);
 
        return 0;
 }
@@ -4853,6 +4896,31 @@ static void gfx_v12_ip_print(void *handle, struct drm_printer *p)
                        }
                }
        }
+
+       /* print gfx queue registers for all instances */
+       if (!adev->gfx.ip_dump_gfx_queues)
+               return;
+
+       index = 0;
+       reg_count = ARRAY_SIZE(gc_gfx_queue_reg_list_12);
+       drm_printf(p, "\nnum_me: %d num_pipe: %d num_queue: %d\n",
+                  adev->gfx.me.num_me,
+                  adev->gfx.me.num_pipe_per_me,
+                  adev->gfx.me.num_queue_per_pipe);
+
+       for (i = 0; i < adev->gfx.me.num_me; i++) {
+               for (j = 0; j < adev->gfx.me.num_pipe_per_me; j++) {
+                       for (k = 0; k < adev->gfx.me.num_queue_per_pipe; k++) {
+                               drm_printf(p, "\nme %d, pipe %d, queue %d\n", i, j, k);
+                               for (reg = 0; reg < reg_count; reg++) {
+                                       drm_printf(p, "%-50s \t 0x%08x\n",
+                                                  gc_gfx_queue_reg_list_12[reg].reg_name,
+                                                  adev->gfx.ip_dump_gfx_queues[index + reg]);
+                               }
+                               index += reg_count;
+                       }
+               }
+       }
 }
 
 static void gfx_v12_ip_dump(void *handle)
@@ -4893,6 +4961,32 @@ static void gfx_v12_ip_dump(void *handle)
        soc24_grbm_select(adev, 0, 0, 0, 0);
        mutex_unlock(&adev->srbm_mutex);
        amdgpu_gfx_off_ctrl(adev, true);
+
+       /* dump gfx queue registers for all instances */
+       if (!adev->gfx.ip_dump_gfx_queues)
+               return;
+
+       index = 0;
+       reg_count = ARRAY_SIZE(gc_gfx_queue_reg_list_12);
+       amdgpu_gfx_off_ctrl(adev, false);
+       mutex_lock(&adev->srbm_mutex);
+       for (i = 0; i < adev->gfx.me.num_me; i++) {
+               for (j = 0; j < adev->gfx.me.num_pipe_per_me; j++) {
+                       for (k = 0; k < adev->gfx.me.num_queue_per_pipe; k++) {
+                               soc24_grbm_select(adev, i, j, k, 0);
+
+                               for (reg = 0; reg < reg_count; reg++) {
+                                       adev->gfx.ip_dump_gfx_queues[index + reg] =
+                                               RREG32(SOC15_REG_ENTRY_OFFSET(
+                                                       gc_gfx_queue_reg_list_12[reg]));
+                               }
+                               index += reg_count;
+                       }
+               }
+       }
+       soc24_grbm_select(adev, 0, 0, 0, 0);
+       mutex_unlock(&adev->srbm_mutex);
+       amdgpu_gfx_off_ctrl(adev, true);
 }
 
 static const struct amd_ip_funcs gfx_v12_0_ip_funcs = {
--

2.34.1