drm/amdkfd: Update MQD management on multi XCC setup

Update MQD management for both HIQ and user-mode compute
queues on a multi XCC setup. MQDs needs to be allocated,
initialized, loaded and destroyed for each XCC in the KFD
node.

v2: squash in fix "drm/amdkfd: Fix SDMA+HIQ HQD allocation on GFX9.4.3"
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Tested-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -800,6 +800,41 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_device *bdev,
 	sg_free_table(ttm->sg);
 }
 
+/*
+ * total_pages is constructed as MQD0+CtrlStack0 + MQD1+CtrlStack1 + ...
+ * MQDn+CtrlStackn where n is the number of XCCs per partition.
+ * pages_per_xcc is the size of one MQD+CtrlStack. The first page is MQD
+ * and uses memory type default, UC. The rest of pages_per_xcc are
+ * Ctrl stack and modify their memory type to NC.
+ */
+static void amdgpu_ttm_gart_bind_gfx9_mqd(struct amdgpu_device *adev,
+				struct ttm_tt *ttm, uint64_t flags)
+{
+	struct amdgpu_ttm_tt *gtt = (void *)ttm;
+	uint64_t total_pages = ttm->num_pages;
+	int num_xcc = max(1U, adev->gfx.num_xcc_per_xcp);
+	uint64_t page_idx, pages_per_xcc = total_pages / num_xcc;
+	int i;
+	uint64_t ctrl_flags = (flags & ~AMDGPU_PTE_MTYPE_VG10_MASK) |
+			AMDGPU_PTE_MTYPE_VG10(AMDGPU_MTYPE_NC);
+
+	for (i = 0, page_idx = 0; i < num_xcc; i++, page_idx += pages_per_xcc) {
+		/* MQD page: use default flags */
+		amdgpu_gart_bind(adev,
+				gtt->offset + (page_idx << PAGE_SHIFT),
+				1, &gtt->ttm.dma_address[page_idx], flags);
+		/*
+		 * Ctrl pages - modify the memory type to NC (ctrl_flags) from
+		 * the second page of the BO onward.
+		 */
+		amdgpu_gart_bind(adev,
+				gtt->offset + ((page_idx + 1) << PAGE_SHIFT),
+				pages_per_xcc - 1,
+				&gtt->ttm.dma_address[page_idx + 1],
+				ctrl_flags);
+	}
+}
+
 static void amdgpu_ttm_gart_bind(struct amdgpu_device *adev,
 				 struct ttm_buffer_object *tbo,
 				 uint64_t flags)
@@ -812,21 +847,7 @@ static void amdgpu_ttm_gart_bind(struct amdgpu_device *adev,
 		flags |= AMDGPU_PTE_TMZ;
 
 	if (abo->flags & AMDGPU_GEM_CREATE_CP_MQD_GFX9) {
-		uint64_t page_idx = 1;
-
-		amdgpu_gart_bind(adev, gtt->offset, page_idx,
-				 gtt->ttm.dma_address, flags);
-
-		/* The memory type of the first page defaults to UC. Now
-		 * modify the memory type to NC from the second page of
-		 * the BO onward.
-		 */
-		flags &= ~AMDGPU_PTE_MTYPE_VG10_MASK;
-		flags |= AMDGPU_PTE_MTYPE_VG10(AMDGPU_MTYPE_NC);
-
-		amdgpu_gart_bind(adev, gtt->offset + (page_idx << PAGE_SHIFT),
-				 ttm->num_pages - page_idx,
-				 &(gtt->ttm.dma_address[page_idx]), flags);
+		amdgpu_ttm_gart_bind_gfx9_mqd(adev, ttm, flags);
 	} else {
 		amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
 				 gtt->ttm.dma_address, flags);

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c

@@ -2247,7 +2247,8 @@ static int allocate_hiq_sdma_mqd(struct device_queue_manager *dqm)
 	uint32_t size = dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size *
 		get_num_all_sdma_engines(dqm) *
 		dev->kfd->device_info.num_sdma_queues_per_engine +
-		dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size;
+		(dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size *
+		dqm->dev->num_xcc_per_node);
 
 	retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev, size,
 		&(mem_obj->gtt_mem), &(mem_obj->gpu_addr),

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c

@@ -76,7 +76,8 @@ struct kfd_mem_obj *allocate_sdma_mqd(struct kfd_node *dev,
 		q->sdma_queue_id) *
 		dev->dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size;
 
-	offset += dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size;
+	offset += dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size *
+		  dev->num_xcc_per_node;
 
 	mqd_mem_obj->gtt_mem = (void *)((uint64_t)dev->dqm->hiq_sdma_mqd.gtt_mem
 				+ offset);
@@ -246,3 +247,28 @@ bool kfd_is_occupied_sdma(struct mqd_manager *mm, void *mqd,
 {
 	return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->adev, mqd);
 }
+
+uint64_t kfd_hiq_mqd_stride(struct kfd_node *dev)
+{
+	return dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size;
+}
+
+void kfd_get_hiq_xcc_mqd(struct kfd_node *dev, struct kfd_mem_obj *mqd_mem_obj,
+		     uint32_t virtual_xcc_id)
+{
+	uint64_t offset;
+
+	offset = kfd_hiq_mqd_stride(dev) * virtual_xcc_id;
+
+	mqd_mem_obj->gtt_mem = (virtual_xcc_id == 0) ?
+			dev->dqm->hiq_sdma_mqd.gtt_mem : NULL;
+	mqd_mem_obj->gpu_addr = dev->dqm->hiq_sdma_mqd.gpu_addr + offset;
+	mqd_mem_obj->cpu_ptr = (uint32_t *)((uintptr_t)
+				dev->dqm->hiq_sdma_mqd.cpu_ptr + offset);
+}
+
+uint64_t kfd_mqd_stride(struct mqd_manager *mm,
+			struct queue_properties *q)
+{
+	return mm->mqd_size;
+}

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h

@@ -119,6 +119,8 @@ struct mqd_manager {
 	int	(*debugfs_show_mqd)(struct seq_file *m, void *data);
 #endif
 	uint32_t (*read_doorbell_id)(void *mqd);
+	uint64_t (*mqd_stride)(struct mqd_manager *mm,
+				struct queue_properties *p);
 
 	struct mutex	mqd_mutex;
 	struct kfd_node	*dev;
@@ -164,4 +166,10 @@ bool kfd_is_occupied_sdma(struct mqd_manager *mm, void *mqd,
 		uint64_t queue_address, uint32_t pipe_id,
 		uint32_t queue_id);
 
+void kfd_get_hiq_xcc_mqd(struct kfd_node *dev,
+		struct kfd_mem_obj *mqd_mem_obj, uint32_t virtual_xcc_id);
+
+uint64_t kfd_hiq_mqd_stride(struct kfd_node *dev);
+uint64_t kfd_mqd_stride(struct mqd_manager *mm,
+			struct queue_properties *q);
 #endif /* KFD_MQD_MANAGER_H_ */

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c

@@ -428,6 +428,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
 		mqd->destroy_mqd = kfd_destroy_mqd_cp;
 		mqd->is_occupied = kfd_is_occupied_cp;
 		mqd->mqd_size = sizeof(struct cik_mqd);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd;
 #endif
@@ -442,6 +443,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
 		mqd->destroy_mqd = kfd_destroy_mqd_cp;
 		mqd->is_occupied = kfd_is_occupied_cp;
 		mqd->mqd_size = sizeof(struct cik_mqd);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd;
 #endif
@@ -457,6 +459,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
 		mqd->checkpoint_mqd = checkpoint_mqd_sdma;
 		mqd->restore_mqd = restore_mqd_sdma;
 		mqd->mqd_size = sizeof(struct cik_sdma_rlc_registers);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
 #endif

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c

@@ -432,6 +432,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
 		mqd->get_wave_state = get_wave_state;
 		mqd->checkpoint_mqd = checkpoint_mqd;
 		mqd->restore_mqd = restore_mqd;
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd;
 #endif
@@ -447,6 +448,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
 		mqd->destroy_mqd = kfd_destroy_mqd_cp;
 		mqd->is_occupied = kfd_is_occupied_cp;
 		mqd->mqd_size = sizeof(struct v10_compute_mqd);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd;
 #endif
@@ -478,6 +480,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
 		mqd->checkpoint_mqd = checkpoint_mqd_sdma;
 		mqd->restore_mqd = restore_mqd_sdma;
 		mqd->mqd_size = sizeof(struct v10_sdma_mqd);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
 #endif

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c

@@ -486,6 +486,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
 		mqd->destroy_mqd = kfd_destroy_mqd_cp;
 		mqd->is_occupied = kfd_is_occupied_cp;
 		mqd->mqd_size = sizeof(struct vi_mqd);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd;
 #endif
@@ -500,6 +501,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
 		mqd->destroy_mqd = kfd_destroy_mqd_cp;
 		mqd->is_occupied = kfd_is_occupied_cp;
 		mqd->mqd_size = sizeof(struct vi_mqd);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd;
 #endif
@@ -515,6 +517,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
 		mqd->checkpoint_mqd = checkpoint_mqd_sdma;
 		mqd->restore_mqd = restore_mqd_sdma;
 		mqd->mqd_size = sizeof(struct vi_sdma_mqd);
+		mqd->mqd_stride = kfd_mqd_stride;
 #if defined(CONFIG_DEBUG_FS)
 		mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
 #endif

drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c

@@ -927,7 +927,9 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
 	struct queue *q;
 	enum KFD_MQD_TYPE mqd_type;
 	struct mqd_manager *mqd_mgr;
-	int r = 0;
+	int r = 0, xcc, num_xccs = 1;
+	void *mqd;
+	uint64_t size = 0;
 
 	list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
 		if (pqn->q) {
@@ -943,6 +945,7 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
 				seq_printf(m, "  Compute queue on device %x\n",
 					   q->device->id);
 				mqd_type = KFD_MQD_TYPE_CP;
+				num_xccs = q->device->num_xcc_per_node;
 				break;
 			default:
 				seq_printf(m,
@@ -951,6 +954,8 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
 				continue;
 			}
 			mqd_mgr = q->device->dqm->mqd_mgrs[mqd_type];
+			size = mqd_mgr->mqd_stride(mqd_mgr,
+							&q->properties);
 		} else if (pqn->kq) {
 			q = pqn->kq->queue;
 			mqd_mgr = pqn->kq->mqd_mgr;
@@ -972,10 +977,13 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
 			continue;
 		}
 
-		r = mqd_mgr->debugfs_show_mqd(m, q->mqd);
+		for (xcc = 0; xcc < num_xccs; xcc++) {
+			mqd = q->mqd + size * xcc;
+			r = mqd_mgr->debugfs_show_mqd(m, mqd);
 			if (r != 0)
 				break;
 		}
+	}
 
 	return r;
 }

drivers/gpu/drm/amd/include/v9_structs.h

@@ -196,10 +196,20 @@ struct v9_mqd {
 	uint32_t compute_wave_restore_addr_lo;
 	uint32_t compute_wave_restore_addr_hi;
 	uint32_t compute_wave_restore_control;
+	union {
+		struct {
 			uint32_t compute_static_thread_mgmt_se4;
 			uint32_t compute_static_thread_mgmt_se5;
 			uint32_t compute_static_thread_mgmt_se6;
 			uint32_t compute_static_thread_mgmt_se7;
+		};
+		struct {
+			uint32_t compute_current_logic_xcc_id; // offset: 39  (0x27)
+			uint32_t compute_restart_cg_tg_id; // offset: 40  (0x28)
+			uint32_t compute_tg_chunk_size; // offset: 41  (0x29)
+			uint32_t compute_restore_tg_chunk_size; // offset: 42  (0x2A)
+		};
+	};
 	uint32_t reserved_43;
 	uint32_t reserved_44;
 	uint32_t reserved_45;
@@ -382,8 +392,16 @@ struct v9_mqd {
 	uint32_t iqtimer_pkt_dw29;
 	uint32_t iqtimer_pkt_dw30;
 	uint32_t iqtimer_pkt_dw31;
+	union {
+		struct {
 			uint32_t reserved_225;
 			uint32_t reserved_226;
+		};
+		struct {
+			uint32_t pm4_target_xcc_in_xcp; // offset: 225  (0xE1)
+			uint32_t cp_mqd_stride_size; // offset: 226  (0xE2)
+		};
+	};
 	uint32_t reserved_227;
 	uint32_t set_resources_header;
 	uint32_t set_resources_dw1;