Merge tag 'amd-drm-fixes-6.13-2024-12-11' of https://gitlab.freedesktop.org/agd5f/linux into drm-fixes

amd-drm-fixes-6.13-2024-12-11:

amdgpu:
- ISP hw init fix
- SR-IOV fixes
- Fix contiguous VRAM mapping for UVD on older GPUs
- Fix some regressions due to drm scheduler changes
- Workload profile fixes
- Cleaner shader fix

amdkfd:
- Fix DMA map direction for migration
- Fix a potential null pointer dereference
- Cacheline size fixes
- Runtime PM fix

Signed-off-by: Dave Airlie <[e-mail]>

From: Alex Deucher <[e-mail]>
Link: https://patchwork.freedesktop.org/patch/msgid/[e-mail]

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

···
 	if (dma_resv_locking_ctx((*bo)->tbo.base.resv) != &parser->exec.ticket)
 		return -EINVAL;
 
+	/* Make sure VRAM is allocated contigiously */
 	(*bo)->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-	amdgpu_bo_placement_from_domain(*bo, (*bo)->allowed_domains);
-	for (i = 0; i < (*bo)->placement.num_placement; i++)
-		(*bo)->placements[i].flags |= TTM_PL_FLAG_CONTIGUOUS;
-	r = ttm_bo_validate(&(*bo)->tbo, &(*bo)->placement, &ctx);
-	if (r)
-		return r;
+	if ((*bo)->tbo.resource->mem_type == TTM_PL_VRAM &&
+	    !((*bo)->tbo.resource->placement & TTM_PL_FLAG_CONTIGUOUS)) {
+
+		amdgpu_bo_placement_from_domain(*bo, (*bo)->allowed_domains);
+		for (i = 0; i < (*bo)->placement.num_placement; i++)
+			(*bo)->placements[i].flags |= TTM_PL_FLAG_CONTIGUOUS;
+		r = ttm_bo_validate(&(*bo)->tbo, &(*bo)->placement, &ctx);
+		if (r)
+			return r;
+	}
 
 	return amdgpu_ttm_alloc_gart(&(*bo)->tbo);
 }

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

···
 	"LAST",
 };
 
-#define AMDGPU_IP_BLK_MASK_ALL GENMASK(AMDGPU_MAX_IP_NUM, 0)
+#define AMDGPU_IP_BLK_MASK_ALL GENMASK(AMD_IP_BLOCK_TYPE_NUM  - 1, 0)
 /*
  * Default init level where all blocks are expected to be initialized. This is
  * the level of initialization expected by default and also after a full reset

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

···
 	for (i = 0; i < abo->placement.num_placement; ++i) {
 		abo->placements[i].fpfn = 0 >> PAGE_SHIFT;
 		abo->placements[i].lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT;
+		if (abo->placements[i].mem_type == TTM_PL_VRAM)
+			abo->placements[i].flags |= TTM_PL_FLAG_CONTIGUOUS;
 	}
 }
 

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

···
 	pasid_mapping_needed &= adev->gmc.gmc_funcs->emit_pasid_mapping &&
 		ring->funcs->emit_wreg;
 
-	if (adev->gfx.enable_cleaner_shader &&
-	    ring->funcs->emit_cleaner_shader &&
-	    job->enforce_isolation)
-		ring->funcs->emit_cleaner_shader(ring);
-
-	if (!vm_flush_needed && !gds_switch_needed && !need_pipe_sync)
+	if (!vm_flush_needed && !gds_switch_needed && !need_pipe_sync &&
+	    !(job->enforce_isolation && !job->vmid))
 		return 0;
 
 	amdgpu_ring_ib_begin(ring);
···
 
 	if (need_pipe_sync)
 		amdgpu_ring_emit_pipeline_sync(ring);
+
+	if (adev->gfx.enable_cleaner_shader &&
+	    ring->funcs->emit_cleaner_shader &&
+	    job->enforce_isolation)
+		ring->funcs->emit_cleaner_shader(ring);
 
 	if (vm_flush_needed) {
 		trace_amdgpu_vm_flush(ring, job->vmid, job->vm_pd_addr);

drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c

···
 MODULE_FIRMWARE("amdgpu/gc_9_4_4_mec.bin");
 MODULE_FIRMWARE("amdgpu/gc_9_4_3_rlc.bin");
 MODULE_FIRMWARE("amdgpu/gc_9_4_4_rlc.bin");
+MODULE_FIRMWARE("amdgpu/gc_9_4_3_sjt_mec.bin");
+MODULE_FIRMWARE("amdgpu/gc_9_4_4_sjt_mec.bin");
 
 #define GFX9_MEC_HPD_SIZE 4096
 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
···
 {
 	int err;
 
-	err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
-				   "amdgpu/%s_mec.bin", chip_name);
+	if (amdgpu_sriov_vf(adev))
+		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
+				"amdgpu/%s_sjt_mec.bin", chip_name);
+	else
+		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
+				"amdgpu/%s_mec.bin", chip_name);
 	if (err)
 		goto out;
 	amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1);

drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c

···
 					   struct amdgpu_job *job,
 					   struct amdgpu_ib *ib)
 {
-	struct amdgpu_ring *ring = to_amdgpu_ring(job->base.sched);
+	struct amdgpu_ring *ring = amdgpu_job_ring(job);
 	unsigned i;
 
 	/* No patching necessary for the first instance */

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

···
 
 
 static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev,
+						   bool cache_line_size_missing,
 						   struct kfd_gpu_cache_info *pcache_info)
 {
 	struct amdgpu_device *adev = kdev->adev;
···
 					CRAT_CACHE_FLAGS_SIMD_CACHE);
 		pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_tcp_per_wpg / 2;
 		pcache_info[i].cache_line_size = adev->gfx.config.gc_tcp_cache_line_size;
+		if (cache_line_size_missing && !pcache_info[i].cache_line_size)
+			pcache_info[i].cache_line_size = 128;
 		i++;
 	}
 	/* Scalar L1 Instruction Cache per SQC */
···
 					CRAT_CACHE_FLAGS_SIMD_CACHE);
 		pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_sqc_per_wgp * 2;
 		pcache_info[i].cache_line_size = adev->gfx.config.gc_instruction_cache_line_size;
+		if (cache_line_size_missing && !pcache_info[i].cache_line_size)
+			pcache_info[i].cache_line_size = 128;
 		i++;
 	}
 	/* Scalar L1 Data Cache per SQC */
···
 					CRAT_CACHE_FLAGS_SIMD_CACHE);
 		pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_sqc_per_wgp * 2;
 		pcache_info[i].cache_line_size = adev->gfx.config.gc_scalar_data_cache_line_size;
+		if (cache_line_size_missing && !pcache_info[i].cache_line_size)
+			pcache_info[i].cache_line_size = 64;
 		i++;
 	}
 	/* GL1 Data Cache per SA */
···
 					CRAT_CACHE_FLAGS_DATA_CACHE |
 					CRAT_CACHE_FLAGS_SIMD_CACHE);
 		pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh;
-		pcache_info[i].cache_line_size = 0;
+		if (cache_line_size_missing)
+			pcache_info[i].cache_line_size = 128;
 		i++;
 	}
 	/* L2 Data Cache per GPU (Total Tex Cache) */
···
 					CRAT_CACHE_FLAGS_SIMD_CACHE);
 		pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh;
 		pcache_info[i].cache_line_size = adev->gfx.config.gc_tcc_cache_line_size;
+		if (cache_line_size_missing && !pcache_info[i].cache_line_size)
+			pcache_info[i].cache_line_size = 128;
 		i++;
 	}
 	/* L3 Data Cache per GPU */
···
 					CRAT_CACHE_FLAGS_DATA_CACHE |
 					CRAT_CACHE_FLAGS_SIMD_CACHE);
 		pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh;
-		pcache_info[i].cache_line_size = 0;
+		pcache_info[i].cache_line_size = 64;
 		i++;
 	}
 	return i;
···
 int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pcache_info)
 {
 	int num_of_cache_types = 0;
+	bool cache_line_size_missing = false;
 
 	switch (kdev->adev->asic_type) {
 	case CHIP_KAVERI:
···
 		case IP_VERSION(11, 5, 0):
 		case IP_VERSION(11, 5, 1):
 		case IP_VERSION(11, 5, 2):
+			/* Cacheline size not available in IP discovery for gc11.
+			 * kfd_fill_gpu_cache_info_from_gfx_config to hard code it
+			 */
+			cache_line_size_missing = true;
+			fallthrough;
 		case IP_VERSION(12, 0, 0):
 		case IP_VERSION(12, 0, 1):
 			num_of_cache_types =
-				kfd_fill_gpu_cache_info_from_gfx_config(kdev->kfd, *pcache_info);
+				kfd_fill_gpu_cache_info_from_gfx_config(kdev->kfd,
+									cache_line_size_missing,
+									*pcache_info);
 			break;
 		default:
 			*pcache_info = dummy_cache_info;

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

···
 	if (!down_read_trylock(&adev->reset_domain->sem))
 		return -EIO;
 
+	if (!pdd->proc_ctx_cpu_ptr) {
+		r = amdgpu_amdkfd_alloc_gtt_mem(adev,
+				AMDGPU_MES_PROC_CTX_SIZE,
+				&pdd->proc_ctx_bo,
+				&pdd->proc_ctx_gpu_addr,
+				&pdd->proc_ctx_cpu_ptr,
+				false);
+		if (r) {
+			dev_err(adev->dev,
+				"failed to allocate process context bo\n");
+			return r;
+		}
+		memset(pdd->proc_ctx_cpu_ptr, 0, AMDGPU_MES_PROC_CTX_SIZE);
+	}
+
 	memset(&queue_input, 0x0, sizeof(struct mes_add_queue_input));
 	queue_input.process_id = qpd->pqm->process->pasid;
 	queue_input.page_table_base_addr =  qpd->page_table_base;

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

···
 		spage = migrate_pfn_to_page(migrate->src[i]);
 		if (spage && !is_zone_device_page(spage)) {
 			src[i] = dma_map_page(dev, spage, 0, PAGE_SIZE,
-					      DMA_TO_DEVICE);
+					      DMA_BIDIRECTIONAL);
 			r = dma_mapping_error(dev, src[i]);
 			if (r) {
 				dev_err(dev, "%s: fail %d dma_map_page\n",
···
 			goto out_oom;
 		}
 
-		dst[i] = dma_map_page(dev, dpage, 0, PAGE_SIZE, DMA_FROM_DEVICE);
+		dst[i] = dma_map_page(dev, dpage, 0, PAGE_SIZE, DMA_BIDIRECTIONAL);
 		r = dma_mapping_error(dev, dst[i]);
 		if (r) {
 			dev_err(adev->dev, "%s: fail %d dma_map_page\n", __func__, r);

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

···
 
 		kfd_free_process_doorbells(pdd->dev->kfd, pdd);
 
-		if (pdd->dev->kfd->shared_resources.enable_mes)
+		if (pdd->dev->kfd->shared_resources.enable_mes &&
+			pdd->proc_ctx_cpu_ptr)
 			amdgpu_amdkfd_free_gtt_mem(pdd->dev->adev,
 						   &pdd->proc_ctx_bo);
 		/*
···
 							struct kfd_process *p)
 {
 	struct kfd_process_device *pdd = NULL;
-	int retval = 0;
 
 	if (WARN_ON_ONCE(p->n_pdds >= MAX_GPU_INSTANCE))
 		return NULL;
···
 	pdd->user_gpu_id = dev->id;
 	atomic64_set(&pdd->evict_duration_counter, 0);
 
-	if (dev->kfd->shared_resources.enable_mes) {
-		retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev,
-						AMDGPU_MES_PROC_CTX_SIZE,
-						&pdd->proc_ctx_bo,
-						&pdd->proc_ctx_gpu_addr,
-						&pdd->proc_ctx_cpu_ptr,
-						false);
-		if (retval) {
-			dev_err(dev->adev->dev,
-				"failed to allocate process context bo\n");
-			goto err_free_pdd;
-		}
-		memset(pdd->proc_ctx_cpu_ptr, 0, AMDGPU_MES_PROC_CTX_SIZE);
-	}
-
 	p->pdds[p->n_pdds++] = pdd;
 	if (kfd_dbg_is_per_vmid_supported(pdd->dev))
 		pdd->spi_dbg_override = pdd->dev->kfd2kgd->disable_debug_trap(
···
 	idr_init(&pdd->alloc_idr);
 
 	return pdd;
-
-err_free_pdd:
-	kfree(pdd);
-	return NULL;
 }
 
 /**

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

···
 void pqm_uninit(struct process_queue_manager *pqm)
 {
 	struct process_queue_node *pqn, *next;
-	struct kfd_process_device *pdd;
 
 	list_for_each_entry_safe(pqn, next, &pqm->queues, process_queue_list) {
 		if (pqn->q) {
-			pdd = kfd_get_process_device_data(pqn->q->device, pqm->process);
-			kfd_queue_unref_bo_vas(pdd, &pqn->q->properties);
-			kfd_queue_release_buffers(pdd, &pqn->q->properties);
+			struct kfd_process_device *pdd = kfd_get_process_device_data(pqn->q->device,
+										     pqm->process);
+			if (pdd) {
+				kfd_queue_unref_bo_vas(pdd, &pqn->q->properties);
+				kfd_queue_release_buffers(pdd, &pqn->q->properties);
+			} else {
+				WARN_ON(!pdd);
+			}
 			pqm_clean_queue_resource(pqm, pqn);
 		}
 

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

···
 };
 
 enum PP_SMC_POWER_PROFILE {
+	PP_SMC_POWER_PROFILE_UNKNOWN = -1,
 	PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT = 0x0,
 	PP_SMC_POWER_PROFILE_FULLSCREEN3D = 0x1,
 	PP_SMC_POWER_PROFILE_POWERSAVING  = 0x2,

drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c

···
 	smu->smu_baco.platform_support = false;
 	smu->smu_baco.maco_support = false;
 	smu->user_dpm_profile.fan_mode = -1;
+	smu->power_profile_mode = PP_SMC_POWER_PROFILE_UNKNOWN;
 
 	mutex_init(&smu->message_lock);
 
···
 	return smu->workload_map && smu->workload_map[profile].valid_mapping;
 }
 
+static void smu_init_power_profile(struct smu_context *smu)
+{
+	if (smu->power_profile_mode == PP_SMC_POWER_PROFILE_UNKNOWN) {
+		if (smu->is_apu ||
+		    !smu_is_workload_profile_available(
+			    smu, PP_SMC_POWER_PROFILE_FULLSCREEN3D))
+			smu->power_profile_mode =
+				PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT;
+		else
+			smu->power_profile_mode =
+				PP_SMC_POWER_PROFILE_FULLSCREEN3D;
+	}
+	smu_power_profile_mode_get(smu, smu->power_profile_mode);
+}
+
 static int smu_sw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
···
 	atomic_set(&smu->smu_power.power_gate.vpe_gated, 1);
 	atomic_set(&smu->smu_power.power_gate.umsch_mm_gated, 1);
 
-	if (smu->is_apu ||
-	    !smu_is_workload_profile_available(smu, PP_SMC_POWER_PROFILE_FULLSCREEN3D))
-		smu->power_profile_mode = PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT;
-	else
-		smu->power_profile_mode = PP_SMC_POWER_PROFILE_FULLSCREEN3D;
-	smu_power_profile_mode_get(smu, smu->power_profile_mode);
-
+	smu_init_power_profile(smu);
 	smu->display_config = &adev->pm.pm_display_cfg;
 
 	smu->smu_dpm.dpm_level = AMD_DPM_FORCED_LEVEL_AUTO;

drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c

···
 	smu->workload_map = smu_v13_0_7_workload_map;
 	smu->smc_driver_if_version = SMU13_0_7_DRIVER_IF_VERSION;
 	smu_v13_0_set_smu_mailbox_registers(smu);
+	smu->power_profile_mode = PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT;
 }