diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 30 | ||||
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device.c | 3 | ||||
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 15 | ||||
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 4 | ||||
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 23 | ||||
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 12 |
6 files changed, 57 insertions, 30 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c index 723f1220e1cc..e5324c5bc6c7 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c @@ -1423,6 +1423,7 @@ err: 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; @@ -1437,6 +1438,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, 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 */ @@ -1449,6 +1452,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, 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 */ @@ -1460,6 +1465,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, 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 */ @@ -1472,7 +1479,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, 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) */ @@ -1484,6 +1492,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, 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 */ @@ -1494,7 +1504,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev, 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; @@ -1510,6 +1520,8 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, if (adev->gfx.config.gc_tcp_size_per_cu) { pcache_info[i].cache_size = adev->gfx.config.gc_tcp_size_per_cu; pcache_info[i].cache_level = 1; + /* Cacheline size not available in IP discovery for gc943,gc944 */ + pcache_info[i].cache_line_size = 128; pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); @@ -1521,6 +1533,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, pcache_info[i].cache_size = adev->gfx.config.gc_l1_instruction_cache_size_per_sqc; pcache_info[i].cache_level = 1; + pcache_info[i].cache_line_size = 64; pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_INST_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); @@ -1531,6 +1544,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, if (adev->gfx.config.gc_l1_data_cache_size_per_sqc) { pcache_info[i].cache_size = adev->gfx.config.gc_l1_data_cache_size_per_sqc; pcache_info[i].cache_level = 1; + pcache_info[i].cache_line_size = 64; pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); @@ -1541,6 +1555,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, if (adev->gfx.config.gc_tcc_size) { pcache_info[i].cache_size = adev->gfx.config.gc_tcc_size; pcache_info[i].cache_level = 2; + pcache_info[i].cache_line_size = 128; pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); @@ -1551,6 +1566,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, if (adev->gmc.mall_size) { pcache_info[i].cache_size = adev->gmc.mall_size / 1024; pcache_info[i].cache_level = 3; + pcache_info[i].cache_line_size = 64; pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED | CRAT_CACHE_FLAGS_DATA_CACHE | CRAT_CACHE_FLAGS_SIMD_CACHE); @@ -1563,6 +1579,7 @@ static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev, 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: @@ -1686,10 +1703,17 @@ int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pc 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; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index 956198da7859..9b51dd75fefc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -235,6 +235,9 @@ static void kfd_device_info_init(struct kfd_dev *kfd, */ kfd->device_info.needs_pci_atomics = true; kfd->device_info.no_atomic_fw_version = kfd->adev->gfx.rs64_enable ? 509 : 0; + } else if (gc_version < IP_VERSION(13, 0, 0)) { + kfd->device_info.needs_pci_atomics = true; + kfd->device_info.no_atomic_fw_version = 2090; } else { kfd->device_info.needs_pci_atomics = true; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index c79fe9069e22..16b5daaa272f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -207,6 +207,21 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q, 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; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index eacfeb32f35d..4b275937d05e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -306,7 +306,7 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, 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", @@ -629,7 +629,7 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange, 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); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 87cd52cf4ee9..d0ee173acf82 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -1076,7 +1076,8 @@ static void kfd_process_destroy_pdds(struct kfd_process *p) 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); /* @@ -1608,7 +1609,6 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev, 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; @@ -1632,21 +1632,6 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev, 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( @@ -1658,10 +1643,6 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev, idr_init(&pdd->alloc_idr); return pdd; - -err_free_pdd: - kfree(pdd); - return NULL; } /** diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c index c76db22a1000..59b92d66e958 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c @@ -212,13 +212,17 @@ static void pqm_clean_queue_resource(struct process_queue_manager *pqm, 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); } |