Skip to contentSkip to conversation

drm/amdgpu/mes: change gfx_hqd mask set for mes v12_1

PATCH New Alex Deucher 1 participant 4 messages Apr 14

Subsystems (2)
lore
Share & tools
lore

Alex Deucher

author maintainer
#0 lore raw

PATCH 1/4 drm/amdgpu/mes: change gfx_hqd mask set for mes v12_1

From: Likun Gao <[email protected]>

Use mes->gfx_hqd_mask to set gfx_hqd_mask on mes hw res pkt for mes v12_1 just same as other mes ip.

---
 drivers/gpu/drm/amd/amdgpu/mes_v12_1.c | 15 +++------------
 1 file changed, 3 insertions(+), 12 deletions(-)
drivers/gpu/drm/amd/amdgpu/mes_v12_1.c +3 -12
index cec8012781265..456dbbcfce1e8 100644
@@ -633,17 +633,6 @@ static int mes_v12_1_set_hw_resources_1(struct amdgpu_mes *mes,
 			offsetof(union MESAPI_SET_HW_RESOURCES_1, api_status));
 }
 
-static void mes_v12_1_set_gfx_hqd_mask(union MESAPI_SET_HW_RESOURCES *pkt)
-{
-	/*
-	 * GFX V12 has only one GFX pipe, but 8 queues in it.
-	 * GFX pipe 0 queue 0 is being used by Kernel queue.
-	 * Set GFX pipe 0 queue 1-7 for MES scheduling
-	 * mask = 1111 1110b
-	 */
-	pkt->gfx_hqd_mask[0] = 0xFE;
-}
-
 static int mes_v12_1_set_hw_resources(struct amdgpu_mes *mes,
 					int pipe, int xcc_id)
 {
@@ -667,7 +656,9 @@ static int mes_v12_1_set_hw_resources(struct amdgpu_mes *mes,
 			mes_set_hw_res_pkt.compute_hqd_mask[i] =
 				mes->compute_hqd_mask[i];
 
-		mes_v12_1_set_gfx_hqd_mask(&mes_set_hw_res_pkt);
+		for (i = 0; i < MAX_GFX_PIPES; i++)
+			mes_set_hw_res_pkt.gfx_hqd_mask[i] =
+				mes->gfx_hqd_mask[i];
 
 		for (i = 0; i < MAX_SDMA_PIPES; i++)
 			mes_set_hw_res_pkt.sdma_hqd_mask[i] =
Signature · 1 line
2.53.0
author maintainer
#1 lore raw

PATCH 2/4 drm/amd/amdgpu : Distribute inv_tlbs on individual XCC

From: Shaoyun Liu <[email protected]>

From MES version 0x74 and above, pipe 1(kiq) support use the shared command buffer to distribute the tasks among the xccs within the same partition. It's more efficient for tlb invalidation been done individually on the each XCC at the same time compare to let master xcc loop all xccs.

This requires the coop_mode been set before kiq set_hw_resource.

---
 drivers/gpu/drm/amd/amdgpu/mes_v12_1.c | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/mes_v12_1.c +11 -8
index 456dbbcfce1e8..3023cf6ab1ba5 100644
@@ -44,6 +44,7 @@ static int mes_v12_1_hw_fini(struct amdgpu_ip_block *ip_block);
 static int mes_v12_1_kiq_hw_init(struct amdgpu_device *adev, uint32_t xcc_id);
 static int mes_v12_1_kiq_hw_fini(struct amdgpu_device *adev, uint32_t xcc_id);
 static int mes_v12_1_self_test(struct amdgpu_device *adev, int xcc_id);
+static int mes_v12_1_setup_coop_mode(struct amdgpu_device *adev, int xcc_id);
 
 #define MES_EOP_SIZE   2048
 
@@ -621,11 +622,15 @@ static int mes_v12_1_set_hw_resources_1(struct amdgpu_mes *mes,
 	mes_set_hw_res_1_pkt.header.dwsize = API_FRAME_SIZE_IN_DWORDS;
 	mes_set_hw_res_1_pkt.mes_kiq_unmap_timeout = 100;
 
-	if (mes->enable_coop_mode && pipe == AMDGPU_MES_SCHED_PIPE) {
+	/* From version 0x74 above, pipe1 support use shared command buffer
+	   to distribute some tasks on individual XCCs*/
+	if (mes->enable_coop_mode &&
+	    ((pipe == AMDGPU_MES_SCHED_PIPE) ||
+	    ((mes->sched_version & AMDGPU_MES_VERSION_MASK) >= 0x74))) {
 		master_xcc_id = mes->master_xcc_ids[inst];
 		mes_set_hw_res_1_pkt.mes_coop_mode = 1;
 		mes_set_hw_res_1_pkt.coop_sch_shared_mc_addr =
-			mes->shared_cmd_buf_gpu_addr[master_xcc_id];
+			mes->shared_cmd_buf_gpu_addr[master_xcc_id + pipe];
 	}
 
 	return mes_v12_1_submit_pkt_and_poll_completion(mes, xcc_id, pipe,
@@ -1148,9 +1153,6 @@ static int mes_v12_1_allocate_shared_cmd_buf(struct amdgpu_device *adev,
 {
 	int r, inst = MES_PIPE_INST(xcc_id, pipe);
 
-	if (pipe == AMDGPU_MES_KIQ_PIPE)
-		return 0;
-
 	r = amdgpu_bo_create_kernel(adev, PAGE_SIZE, PAGE_SIZE,
 				    AMDGPU_GEM_DOMAIN_VRAM,
 				    &adev->mes.shared_cmd_buf_obj[inst],
@@ -1714,6 +1716,10 @@ static int mes_v12_1_kiq_hw_init(struct amdgpu_device *adev, uint32_t xcc_id)
 		goto failure;
 
 	if (adev->enable_uni_mes) {
+		r = mes_v12_1_setup_coop_mode(adev, xcc_id);
+		if (r)
+			goto failure;
+
 		r = mes_v12_1_set_hw_resources(&adev->mes,
 						 AMDGPU_MES_KIQ_PIPE, xcc_id);
 		if (r)
@@ -1836,9 +1842,6 @@ static int mes_v12_1_xcc_hw_init(struct amdgpu_ip_block *ip_block, int xcc_id)
 		goto failure;
 
 	if (adev->enable_uni_mes) {
-		r = mes_v12_1_setup_coop_mode(adev, xcc_id);
-		if (r)
-			goto failure;
 		mes_v12_1_set_hw_resources_1(&adev->mes,
 					       AMDGPU_MES_SCHED_PIPE, xcc_id);
 	}
Signature · 1 line
2.53.0
author maintainer
#2 lore raw

PATCH 3/4 drm/amdgpu: Move get mes version info to the end of set_hw_resource

In reply to Alex Deucher

From: Shaoyun Liu <[email protected]>

After init mes queue, it require some time for mes to execute init code including setup version info in CP_MES_GP3 register. Put it after first MES call(set_hw_resource package).

---
 drivers/gpu/drm/amd/amdgpu/mes_v12_1.c | 32 ++++++++++++++------------
 1 file changed, 17 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdgpu/mes_v12_1.c +17 -15
index 3023cf6ab1ba5..d87ec61cac88b 100644
@@ -626,7 +626,7 @@ static int mes_v12_1_set_hw_resources_1(struct amdgpu_mes *mes,
 	   to distribute some tasks on individual XCCs*/
 	if (mes->enable_coop_mode &&
 	    ((pipe == AMDGPU_MES_SCHED_PIPE) ||
-	    ((mes->sched_version & AMDGPU_MES_VERSION_MASK) >= 0x74))) {
+	    ((mes->kiq_version & AMDGPU_MES_VERSION_MASK) >= 0x74))) {
 		master_xcc_id = mes->master_xcc_ids[inst];
 		mes_set_hw_res_1_pkt.mes_coop_mode = 1;
 		mes_set_hw_res_1_pkt.coop_sch_shared_mc_addr =
@@ -641,7 +641,7 @@ static int mes_v12_1_set_hw_resources_1(struct amdgpu_mes *mes,
 static int mes_v12_1_set_hw_resources(struct amdgpu_mes *mes,
 					int pipe, int xcc_id)
 {
-	int i;
+	int i, status;
 	struct amdgpu_device *adev = mes->adev;
 	union MESAPI_SET_HW_RESOURCES mes_set_hw_res_pkt;
 
@@ -711,9 +711,23 @@ static int mes_v12_1_set_hw_resources(struct amdgpu_mes *mes,
 	if (adev->enforce_isolation[0] == AMDGPU_ENFORCE_ISOLATION_ENABLE)
 		mes_set_hw_res_pkt.limit_single_process = 1;
 
-	return mes_v12_1_submit_pkt_and_poll_completion(mes, xcc_id, pipe,
+	status = mes_v12_1_submit_pkt_and_poll_completion(mes, xcc_id, pipe,
 			&mes_set_hw_res_pkt, sizeof(mes_set_hw_res_pkt),
 			offsetof(union MESAPI_SET_HW_RESOURCES, api_status));
+
+	/* get MES scheduler versions */
+	mutex_lock(&adev->srbm_mutex);
+	soc_v1_0_grbm_select(adev, 3, pipe, 0, 0, GET_INST(GC, xcc_id));
+
+	if (pipe == AMDGPU_MES_SCHED_PIPE)
+		adev->mes.sched_version = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_GP3_LO);
+	else if (pipe == AMDGPU_MES_KIQ_PIPE && adev->enable_mes_kiq)
+		adev->mes.kiq_version = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_GP3_LO);
+
+	soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
+	mutex_unlock(&adev->srbm_mutex);
+
+	return status;
 }
 
 static void mes_v12_1_init_aggregated_doorbell(struct amdgpu_mes *mes,
@@ -1398,18 +1412,6 @@ static int mes_v12_1_queue_init(struct amdgpu_device *adev,
 		mes_v12_1_queue_init_register(ring, xcc_id);
 	}
 
-	/* get MES scheduler/KIQ versions */
-	mutex_lock(&adev->srbm_mutex);
-	soc_v1_0_grbm_select(adev, 3, pipe, 0, 0, GET_INST(GC, xcc_id));
-
-	if (pipe == AMDGPU_MES_SCHED_PIPE)
-		adev->mes.sched_version = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_GP3_LO);
-	else if (pipe == AMDGPU_MES_KIQ_PIPE && adev->enable_mes_kiq)
-		adev->mes.kiq_version = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MES_GP3_LO);
-
-	soc_v1_0_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
-	mutex_unlock(&adev->srbm_mutex);
-
 	return 0;
 }
 
Signature · 1 line
2.53.0
author maintainer
#3 lore raw

PATCH 4/4 drm/amdgpu: update cp cmd pkt for gfx v12_1 v2

In reply to Alex Deucher

From: Likun Gao <[email protected]>

Update CP command package for gfx v12.1.
1. Update to common command pkg format for PACKET3_ACQUIRE_MEM.
2. Update to definition PACKET3_PRED_EXEC command pkt.

---
 drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c     | 12 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v12_1_pkt.h | 68 ++++++++++++++--------
 2 files changed, 51 insertions(+), 29 deletions(-)
drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c +6 -6
index 3bef36f5a487e..885183eb7e7b2 100644
@@ -3825,12 +3825,12 @@ static int gfx_v12_1_rlc_poison_irq(struct amdgpu_device *adev,
 static void gfx_v12_1_emit_mem_sync(struct amdgpu_ring *ring)
 {
 	const unsigned int gcr_cntl =
-			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_INV(1) |
-			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_WB(1) |
-			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLV_INV(1) |
-			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLK_INV(1) |
-			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLI_INV(1) |
-			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_SCOPE(2);
+			PACKET3_ACQUIRE_MEM__GCR_CNTL__GL2_INV(1) |
+			PACKET3_ACQUIRE_MEM__GCR_CNTL__GL2_WB(1) |
+			PACKET3_ACQUIRE_MEM__GCR_CNTL__GLV_INV(1) |
+			PACKET3_ACQUIRE_MEM__GCR_CNTL__GLK_INV(1) |
+			PACKET3_ACQUIRE_MEM__GCR_CNTL__GLI_INV(1) |
+			PACKET3_ACQUIRE_MEM__GCR_CNTL__GL2_SCOPE(2);
 
 	/* ACQUIRE_MEM - make one or more surfaces valid for use by the subsequent operations */
 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 6));
drivers/gpu/drm/amd/amdgpu/gfx_v12_1_pkt.h +45 -23

109 lines not on this page · raw message

Signature · 1 line
2.53.0