diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 123d4a09114d..fe6d988e7f24 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -1093,9 +1093,16 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data, * If that number is larger than the size of the array, the ioctl must * be retried. */ + if (args->num_entries > INT_MAX / sizeof(*vm_entries)) { + r = -EINVAL; + goto out_exec; + } + vm_entries = kvcalloc(args->num_entries, sizeof(*vm_entries), GFP_KERNEL); - if (!vm_entries) - return -ENOMEM; + if (!vm_entries) { + r = -ENOMEM; + goto out_exec; + } amdgpu_vm_bo_va_for_each_valid_mapping(bo_va, mapping) { if (num_mappings < args->num_entries) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c index f72990ac046e..5bfa5a84b09c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c @@ -51,8 +51,6 @@ #include "amdgpu_amdkfd.h" #include "amdgpu_hmm.h" -#define MAX_WALK_BYTE (2UL << 30) - /** * amdgpu_hmm_invalidate_gfx - callback to notify about mm change * @@ -78,6 +76,7 @@ static bool amdgpu_hmm_invalidate_gfx(struct mmu_interval_notifier *mni, mmu_interval_set_seq(mni, cur_seq); + amdgpu_vm_bo_invalidate(bo, false); r = dma_resv_wait_timeout(bo->tbo.base.resv, DMA_RESV_USAGE_BOOKKEEP, false, MAX_SCHEDULE_TIMEOUT); mutex_unlock(&adev->notifier_lock); @@ -170,11 +169,13 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier, void *owner, struct amdgpu_hmm_range *range) { - unsigned long end; + const u64 max_bytes = SZ_2G; + + struct hmm_range *hmm_range = &range->hmm_range; unsigned long timeout; unsigned long *pfns; - int r = 0; - struct hmm_range *hmm_range = &range->hmm_range; + unsigned long end; + int r; pfns = kvmalloc_array(npages, sizeof(*pfns), GFP_KERNEL); if (unlikely(!pfns)) { @@ -191,8 +192,9 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier, end = start + npages * PAGE_SIZE; hmm_range->dev_private_owner = owner; + hmm_range->notifier_seq = mmu_interval_read_begin(notifier); do { - hmm_range->end = min(hmm_range->start + MAX_WALK_BYTE, end); + hmm_range->end = min(hmm_range->start + max_bytes, end); pr_debug("hmm range: start = 0x%lx, end = 0x%lx", hmm_range->start, hmm_range->end); @@ -200,7 +202,6 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier, timeout = jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT); retry: - hmm_range->notifier_seq = mmu_interval_read_begin(notifier); r = hmm_range_fault(hmm_range); if (unlikely(r)) { if (r == -EBUSY && !time_after(jiffies, timeout)) @@ -210,7 +211,7 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier, if (hmm_range->end == end) break; - hmm_range->hmm_pfns += MAX_WALK_BYTE >> PAGE_SHIFT; + hmm_range->hmm_pfns += max_bytes >> PAGE_SHIFT; hmm_range->start = hmm_range->end; } while (hmm_range->end < end); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 6c644cfe6695..fc9f3adf9912 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2280,7 +2280,8 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device *adev) list_for_each_entry(obj, &con->head, node) { if (amdgpu_ras_is_supported(adev, obj->head.block) && (obj->attr_inuse == 1)) { - sprintf(fs_info.debugfs_name, "%s_err_inject", + snprintf(fs_info.debugfs_name, sizeof(fs_info.debugfs_name), + "%s_err_inject", get_ras_block_str(&obj->head)); fs_info.head = obj->head; amdgpu_ras_debugfs_create(adev, &fs_info, dir); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c index f070ea37d918..cf192500800f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c @@ -215,33 +215,15 @@ void amdgpu_userq_process_fence_irq(struct amdgpu_device *adev, u32 doorbell) xa_unlock_irqrestore(xa, flags); } -static int amdgpu_userq_buffer_va_list_add(struct amdgpu_usermode_queue *queue, - struct amdgpu_bo_va_mapping *va_map, u64 addr) -{ - struct amdgpu_userq_va_cursor *va_cursor; - struct userq_va_list; - - va_cursor = kzalloc_obj(*va_cursor); - if (!va_cursor) - return -ENOMEM; - - INIT_LIST_HEAD(&va_cursor->list); - va_cursor->gpu_addr = addr; - va_map->bo_va->userq_va_mapped = true; - list_add(&va_cursor->list, &queue->userq_va_list); - - return 0; -} - int amdgpu_userq_input_va_validate(struct amdgpu_device *adev, struct amdgpu_usermode_queue *queue, - u64 addr, u64 expected_size) + u64 addr, u64 expected_size, + u64 *va_out) { struct amdgpu_bo_va_mapping *va_map; struct amdgpu_vm *vm = queue->vm; u64 user_addr; u64 size; - int r = 0; /* Caller must hold vm->root.bo reservation */ dma_resv_assert_held(queue->vm->root.bo->tbo.base.resv); @@ -250,20 +232,18 @@ int amdgpu_userq_input_va_validate(struct amdgpu_device *adev, size = expected_size >> AMDGPU_GPU_PAGE_SHIFT; va_map = amdgpu_vm_bo_lookup_mapping(vm, user_addr); - if (!va_map) { - r = -EINVAL; - goto out_err; - } + if (!va_map) + return -EINVAL; + /* Only validate the userq whether resident in the VM mapping range */ if (user_addr >= va_map->start && va_map->last - user_addr + 1 >= size) { - amdgpu_userq_buffer_va_list_add(queue, va_map, user_addr); + va_map->bo_va->userq_va_mapped = true; + *va_out = user_addr; return 0; } - r = -EINVAL; -out_err: - return r; + return -EINVAL; } static bool amdgpu_userq_buffer_va_mapped(struct amdgpu_vm *vm, u64 addr) @@ -284,14 +264,16 @@ static bool amdgpu_userq_buffer_va_mapped(struct amdgpu_vm *vm, u64 addr) static bool amdgpu_userq_buffer_vas_mapped(struct amdgpu_usermode_queue *queue) { - struct amdgpu_userq_va_cursor *va_cursor, *tmp; - int r = 0; + int i, r = 0; - list_for_each_entry_safe(va_cursor, tmp, &queue->userq_va_list, list) { - r += amdgpu_userq_buffer_va_mapped(queue->vm, va_cursor->gpu_addr); + for (i = 0; i < ARRAY_SIZE(queue->userq_vas.va_array); i++) { + if (!queue->userq_vas.va_array[i]) + continue; + r += amdgpu_userq_buffer_va_mapped(queue->vm, + queue->userq_vas.va_array[i]); dev_dbg(queue->userq_mgr->adev->dev, "validate the userq mapping:%p va:%llx r:%d\n", - queue, va_cursor->gpu_addr, r); + queue, queue->userq_vas.va_array[i], r); } if (r != 0) @@ -300,24 +282,7 @@ static bool amdgpu_userq_buffer_vas_mapped(struct amdgpu_usermode_queue *queue) return false; } -static void amdgpu_userq_buffer_vas_list_cleanup(struct amdgpu_device *adev, - struct amdgpu_usermode_queue *queue) -{ - struct amdgpu_userq_va_cursor *va_cursor, *tmp; - struct amdgpu_bo_va_mapping *mapping; - /* Caller must hold vm->root.bo reservation */ - dma_resv_assert_held(queue->vm->root.bo->tbo.base.resv); - - list_for_each_entry_safe(va_cursor, tmp, &queue->userq_va_list, list) { - mapping = amdgpu_vm_bo_lookup_mapping(queue->vm, va_cursor->gpu_addr); - if (mapping) - dev_dbg(adev->dev, "delete the userq:%p va:%llx\n", - queue, va_cursor->gpu_addr); - list_del(&va_cursor->list); - kfree(va_cursor); - } -} static int amdgpu_userq_preempt_helper(struct amdgpu_usermode_queue *queue) { @@ -417,18 +382,14 @@ static void amdgpu_userq_cleanup(struct amdgpu_usermode_queue *queue) { struct amdgpu_userq_mgr *uq_mgr = queue->userq_mgr; struct amdgpu_device *adev = uq_mgr->adev; - const struct amdgpu_userq_funcs *uq_funcs = adev->userq_funcs[queue->queue_type]; /* Wait for mode-1 reset to complete */ down_read(&adev->reset_domain->sem); - uq_funcs->mqd_destroy(queue); /* Use interrupt-safe locking since IRQ handlers may access these XArrays */ xa_erase_irq(&adev->userq_doorbell_xa, queue->doorbell_index); amdgpu_userq_fence_driver_free(queue); queue->fence_drv = NULL; - queue->userq_mgr = NULL; - list_del(&queue->userq_va_list); up_read(&adev->reset_domain->sem); } @@ -467,81 +428,15 @@ amdgpu_userq_ensure_ev_fence(struct amdgpu_userq_mgr *uq_mgr, dma_fence_put(ev_fence); } -int amdgpu_userq_create_object(struct amdgpu_userq_mgr *uq_mgr, - struct amdgpu_userq_obj *userq_obj, - int size) -{ - struct amdgpu_device *adev = uq_mgr->adev; - struct amdgpu_bo_param bp; - int r; - memset(&bp, 0, sizeof(bp)); - bp.byte_align = PAGE_SIZE; - bp.domain = AMDGPU_GEM_DOMAIN_GTT; - bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS | - AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED; - bp.type = ttm_bo_type_kernel; - bp.size = size; - bp.resv = NULL; - bp.bo_ptr_size = sizeof(struct amdgpu_bo); - r = amdgpu_bo_create(adev, &bp, &userq_obj->obj); - if (r) { - drm_file_err(uq_mgr->file, "Failed to allocate BO for userqueue (%d)", r); - return r; - } - - r = amdgpu_bo_reserve(userq_obj->obj, true); - if (r) { - drm_file_err(uq_mgr->file, "Failed to reserve BO to map (%d)", r); - goto free_obj; - } - - r = amdgpu_bo_pin(userq_obj->obj, AMDGPU_GEM_DOMAIN_GTT); - if (r) - goto unresv; - - r = amdgpu_ttm_alloc_gart(&(userq_obj->obj)->tbo); - if (r) { - drm_file_err(uq_mgr->file, "Failed to alloc GART for userqueue object (%d)", r); - goto unpin_bo; - } - - r = amdgpu_bo_kmap(userq_obj->obj, &userq_obj->cpu_ptr); - if (r) { - drm_file_err(uq_mgr->file, "Failed to map BO for userqueue (%d)", r); - goto unpin_bo; - } - - userq_obj->gpu_addr = amdgpu_bo_gpu_offset(userq_obj->obj); - amdgpu_bo_unreserve(userq_obj->obj); - memset(userq_obj->cpu_ptr, 0, size); - return 0; - -unpin_bo: - amdgpu_bo_unpin(userq_obj->obj); -unresv: - amdgpu_bo_unreserve(userq_obj->obj); -free_obj: - amdgpu_bo_unref(&userq_obj->obj); - - return r; -} - -void amdgpu_userq_destroy_object(struct amdgpu_userq_mgr *uq_mgr, - struct amdgpu_userq_obj *userq_obj) -{ - amdgpu_bo_kunmap(userq_obj->obj); - amdgpu_bo_unpin(userq_obj->obj); - amdgpu_bo_unref(&userq_obj->obj); -} - -uint64_t +static int amdgpu_userq_get_doorbell_index(struct amdgpu_userq_mgr *uq_mgr, struct amdgpu_db_info *db_info, - struct drm_file *filp) + struct drm_file *filp, + u64 *index) { - uint64_t index; + u64 doorbell_index; struct drm_gem_object *gobj; struct amdgpu_userq_obj *db_obj = db_info->db_obj; int r, db_size; @@ -588,12 +483,13 @@ amdgpu_userq_get_doorbell_index(struct amdgpu_userq_mgr *uq_mgr, goto unpin_bo; } - index = amdgpu_doorbell_index_on_bar(uq_mgr->adev, db_obj->obj, - db_info->doorbell_offset, db_size); + doorbell_index = amdgpu_doorbell_index_on_bar(uq_mgr->adev, db_obj->obj, + db_info->doorbell_offset, db_size); drm_dbg_driver(adev_to_drm(uq_mgr->adev), - "[Usermode queues] doorbell index=%lld\n", index); + "[Usermode queues] doorbell index=%lld\n", doorbell_index); amdgpu_bo_unreserve(db_obj->obj); - return index; + *index = doorbell_index; + return 0; unpin_bo: amdgpu_bo_unpin(db_obj->obj); @@ -608,9 +504,7 @@ static int amdgpu_userq_destroy(struct amdgpu_userq_mgr *uq_mgr, struct amdgpu_usermode_queue *queue) { struct amdgpu_device *adev = uq_mgr->adev; - struct amdgpu_fpriv *fpriv = uq_mgr_to_fpriv(uq_mgr); - struct amdgpu_vm *vm = &fpriv->vm; - + const struct amdgpu_userq_funcs *uq_funcs = adev->userq_funcs[queue->queue_type]; int r = 0; cancel_delayed_work_sync(&uq_mgr->resume_work); @@ -618,14 +512,6 @@ amdgpu_userq_destroy(struct amdgpu_userq_mgr *uq_mgr, struct amdgpu_usermode_que /* Cancel any pending hang detection work and cleanup */ cancel_delayed_work_sync(&queue->hang_detect_work); - r = amdgpu_bo_reserve(vm->root.bo, false); - if (r) { - drm_file_err(uq_mgr->file, "Failed to reserve root bo during userqueue destroy\n"); - return r; - } - amdgpu_userq_buffer_vas_list_cleanup(adev, queue); - amdgpu_bo_unreserve(vm->root.bo); - mutex_lock(&uq_mgr->userq_mutex); amdgpu_userq_wait_for_last_fence(queue); @@ -637,6 +523,10 @@ amdgpu_userq_destroy(struct amdgpu_userq_mgr *uq_mgr, struct amdgpu_usermode_que amdgpu_userq_cleanup(queue); mutex_unlock(&uq_mgr->userq_mutex); + cancel_delayed_work_sync(&queue->hang_detect_work); + uq_funcs->mqd_destroy(queue); + queue->userq_mgr = NULL; + amdgpu_bo_reserve(queue->db_obj.obj, true); amdgpu_bo_unpin(queue->db_obj.obj); amdgpu_bo_unreserve(queue->db_obj.obj); @@ -739,7 +629,6 @@ amdgpu_userq_create(struct drm_file *filp, union drm_amdgpu_userq *args) } kref_init(&queue->refcount); - INIT_LIST_HEAD(&queue->userq_va_list); queue->doorbell_handle = args->in.doorbell_handle; queue->queue_type = args->in.ip_type; queue->vm = &fpriv->vm; @@ -748,26 +637,29 @@ amdgpu_userq_create(struct drm_file *filp, union drm_amdgpu_userq *args) INIT_DELAYED_WORK(&queue->hang_detect_work, amdgpu_userq_hang_detect_work); - mutex_init(&queue->fence_drv_lock); - xa_init_flags(&queue->fence_drv_xa, XA_FLAGS_ALLOC); r = amdgpu_userq_fence_driver_alloc(adev, &queue->fence_drv); if (r) goto free_queue; + xa_init_flags(&queue->fence_drv_xa, XA_FLAGS_ALLOC); + mutex_init(&queue->fence_drv_lock); /* Make sure the queue can actually run with those virtual addresses. */ r = amdgpu_bo_reserve(fpriv->vm.root.bo, false); if (r) goto free_fence_drv; if (amdgpu_userq_input_va_validate(adev, queue, args->in.queue_va, - args->in.queue_size) || + args->in.queue_size, + &queue->userq_vas.va.queue_rb) || amdgpu_userq_input_va_validate(adev, queue, args->in.rptr_va, - AMDGPU_GPU_PAGE_SIZE) || + AMDGPU_GPU_PAGE_SIZE, + &queue->userq_vas.va.rptr) || amdgpu_userq_input_va_validate(adev, queue, args->in.wptr_va, - AMDGPU_GPU_PAGE_SIZE)) { + AMDGPU_GPU_PAGE_SIZE, + &queue->userq_vas.va.wptr)) { r = -EINVAL; amdgpu_bo_unreserve(fpriv->vm.root.bo); - goto clean_mapping; + goto free_fence_drv; } amdgpu_bo_unreserve(fpriv->vm.root.bo); @@ -776,18 +668,17 @@ amdgpu_userq_create(struct drm_file *filp, union drm_amdgpu_userq *args) db_info.doorbell_handle = queue->doorbell_handle; db_info.db_obj = &queue->db_obj; db_info.doorbell_offset = args->in.doorbell_offset; - index = amdgpu_userq_get_doorbell_index(uq_mgr, &db_info, filp); - if (index == (uint64_t)-EINVAL) { + r = amdgpu_userq_get_doorbell_index(uq_mgr, &db_info, filp, &index); + if (r) { drm_file_err(uq_mgr->file, "Failed to get doorbell for queue\n"); - r = -EINVAL; - goto clean_mapping; + goto free_fence_drv; } queue->doorbell_index = index; r = uq_funcs->mqd_create(queue, &args->in); if (r) { drm_file_err(uq_mgr->file, "Failed to create Queue\n"); - goto clean_mapping; + goto clean_doorbell_bo; } /* Update VM owner at userq submit-time for page-fault attribution. */ @@ -808,7 +699,7 @@ amdgpu_userq_create(struct drm_file *filp, union drm_amdgpu_userq *args) if (r) { drm_file_err(uq_mgr->file, "Failed to map Queue\n"); mutex_unlock(&uq_mgr->userq_mutex); - goto clean_doorbell; + goto erase_doorbell; } } @@ -831,15 +722,15 @@ amdgpu_userq_create(struct drm_file *filp, union drm_amdgpu_userq *args) args->out.queue_id = qid; return 0; -clean_doorbell: +erase_doorbell: xa_erase_irq(&adev->userq_doorbell_xa, index); clean_mqd: uq_funcs->mqd_destroy(queue); -clean_mapping: - amdgpu_bo_reserve(fpriv->vm.root.bo, true); - amdgpu_userq_buffer_vas_list_cleanup(adev, queue); - amdgpu_bo_unreserve(fpriv->vm.root.bo); - mutex_destroy(&queue->fence_drv_lock); +clean_doorbell_bo: + amdgpu_bo_reserve(queue->db_obj.obj, true); + amdgpu_bo_unpin(queue->db_obj.obj); + amdgpu_bo_unreserve(queue->db_obj.obj); + amdgpu_bo_unref(&queue->db_obj.obj); free_fence_drv: amdgpu_userq_fence_driver_free(queue); free_queue: diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h index 49b33e2d6932..28cfc6682333 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h @@ -48,11 +48,6 @@ struct amdgpu_userq_obj { struct amdgpu_bo *obj; }; -struct amdgpu_userq_va_cursor { - u64 gpu_addr; - struct list_head list; -}; - struct amdgpu_usermode_queue { int queue_type; enum amdgpu_userq_state state; @@ -93,7 +88,17 @@ struct amdgpu_usermode_queue { struct delayed_work hang_detect_work; struct kref refcount; - struct list_head userq_va_list; + union { + struct { + u64 queue_rb; + u64 wptr; + u64 rptr; + u64 eop; + u64 shadow; + u64 csa; + } va; + u64 va_array[6]; + } userq_vas; }; struct amdgpu_userq_funcs { @@ -151,22 +156,11 @@ void amdgpu_userq_mgr_cancel_reset_work(struct amdgpu_device *adev); void amdgpu_userq_mgr_cancel_resume(struct amdgpu_userq_mgr *userq_mgr); void amdgpu_userq_mgr_fini(struct amdgpu_userq_mgr *userq_mgr); -int amdgpu_userq_create_object(struct amdgpu_userq_mgr *uq_mgr, - struct amdgpu_userq_obj *userq_obj, - int size); - -void amdgpu_userq_destroy_object(struct amdgpu_userq_mgr *uq_mgr, - struct amdgpu_userq_obj *userq_obj); - void amdgpu_userq_evict(struct amdgpu_userq_mgr *uq_mgr); void amdgpu_userq_ensure_ev_fence(struct amdgpu_userq_mgr *userq_mgr, struct amdgpu_eviction_fence_mgr *evf_mgr); -uint64_t amdgpu_userq_get_doorbell_index(struct amdgpu_userq_mgr *uq_mgr, - struct amdgpu_db_info *db_info, - struct drm_file *filp); - u32 amdgpu_userq_get_supported_ip_mask(struct amdgpu_device *adev); bool amdgpu_userq_enabled(struct drm_device *dev); @@ -185,7 +179,8 @@ void amdgpu_userq_process_fence_irq(struct amdgpu_device *adev, u32 doorbell); int amdgpu_userq_input_va_validate(struct amdgpu_device *adev, struct amdgpu_usermode_queue *queue, - u64 addr, u64 expected_size); + u64 addr, u64 expected_size, u64 *va_out); + void amdgpu_userq_gem_va_unmap_validate(struct amdgpu_device *adev, struct amdgpu_bo_va_mapping *mapping, uint64_t saddr); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index fccd758b6699..c9f88ecce1a7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1631,6 +1631,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev, { struct amdgpu_bo_va *bo_va; struct dma_resv *resv; + struct amdgpu_bo *bo; bool clear, unlock; int r; @@ -1650,11 +1651,13 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev, while (!list_empty(&vm->invalidated)) { bo_va = list_first_entry(&vm->invalidated, struct amdgpu_bo_va, base.vm_status); - resv = bo_va->base.bo->tbo.base.resv; + bo = bo_va->base.bo; + resv = bo->tbo.base.resv; spin_unlock(&vm->status_lock); /* Try to reserve the BO to avoid clearing its ptes */ - if (!adev->debug_vm && dma_resv_trylock(resv)) { + if (!adev->debug_vm && !amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) && + dma_resv_trylock(resv)) { clear = false; unlock = true; /* The caller is already holding the reservation lock */ diff --git a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c index 5b4121ddc78c..98aa00eeb2f4 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c @@ -81,7 +81,7 @@ mes_userq_create_wptr_mapping(struct amdgpu_device *adev, ret = amdgpu_ttm_alloc_gart(&wptr_obj->obj->tbo); if (ret) { DRM_ERROR("Failed to bind bo to GART. ret %d\n", ret); - goto fail_map; + goto fail_alloc_gart; } queue->wptr_obj.gpu_addr = amdgpu_bo_gpu_offset(wptr_obj->obj); @@ -89,6 +89,8 @@ mes_userq_create_wptr_mapping(struct amdgpu_device *adev, drm_exec_fini(&exec); return 0; +fail_alloc_gart: + amdgpu_bo_unpin(wptr_obj->obj); fail_map: amdgpu_bo_unref(&wptr_obj->obj); fail_lock: @@ -190,12 +192,16 @@ static int mes_userq_create_ctx_space(struct amdgpu_userq_mgr *uq_mgr, * for the same. */ size = AMDGPU_USERQ_PROC_CTX_SZ + AMDGPU_USERQ_GANG_CTX_SZ; - r = amdgpu_userq_create_object(uq_mgr, ctx, size); + r = amdgpu_bo_create_kernel(uq_mgr->adev, size, 0, + AMDGPU_GEM_DOMAIN_GTT, + &ctx->obj, &ctx->gpu_addr, + &ctx->cpu_ptr); if (r) { DRM_ERROR("Failed to allocate ctx space bo for userqueue, err:%d\n", r); return r; } + memset(ctx->cpu_ptr, 0, size); return 0; } @@ -268,13 +274,19 @@ static int mes_userq_mqd_create(struct amdgpu_usermode_queue *queue, return -ENOMEM; } - r = amdgpu_userq_create_object(uq_mgr, &queue->mqd, - AMDGPU_MQD_SIZE_ALIGN(mqd_hw_default->mqd_size)); + r = amdgpu_bo_create_kernel(adev, + AMDGPU_MQD_SIZE_ALIGN(mqd_hw_default->mqd_size), + 0, AMDGPU_GEM_DOMAIN_GTT, + &queue->mqd.obj, &queue->mqd.gpu_addr, + &queue->mqd.cpu_ptr); if (r) { DRM_ERROR("Failed to create MQD object for userqueue\n"); goto free_props; } + memset(queue->mqd.cpu_ptr, 0, + AMDGPU_MQD_SIZE_ALIGN(mqd_hw_default->mqd_size)); + /* Initialize the MQD BO with user given values */ userq_props->wptr_gpu_addr = mqd_user->wptr_va; userq_props->rptr_gpu_addr = mqd_user->rptr_va; @@ -306,8 +318,9 @@ static int mes_userq_mqd_create(struct amdgpu_usermode_queue *queue, kfree(compute_mqd); goto free_mqd; } - r = amdgpu_userq_input_va_validate(adev, queue, compute_mqd->eop_va, - 2048); + r = amdgpu_userq_input_va_validate(adev, queue, + compute_mqd->eop_va, 2048, + &queue->userq_vas.va.eop); amdgpu_bo_unreserve(queue->vm->root.bo); if (r) { kfree(compute_mqd); @@ -356,7 +369,8 @@ static int mes_userq_mqd_create(struct amdgpu_usermode_queue *queue, goto free_mqd; } r = amdgpu_userq_input_va_validate(adev, queue, mqd_gfx_v11->shadow_va, - shadow_info.shadow_size); + shadow_info.shadow_size, + &queue->userq_vas.va.shadow); if (r) { amdgpu_bo_unreserve(queue->vm->root.bo); kfree(mqd_gfx_v11); @@ -364,7 +378,8 @@ static int mes_userq_mqd_create(struct amdgpu_usermode_queue *queue, } r = amdgpu_userq_input_va_validate(adev, queue, mqd_gfx_v11->csa_va, - shadow_info.csa_size); + shadow_info.csa_size, + &queue->userq_vas.va.csa); amdgpu_bo_unreserve(queue->vm->root.bo); if (r) { kfree(mqd_gfx_v11); @@ -394,7 +409,8 @@ static int mes_userq_mqd_create(struct amdgpu_usermode_queue *queue, goto free_mqd; } r = amdgpu_userq_input_va_validate(adev, queue, mqd_sdma_v11->csa_va, - 32); + 32, + &queue->userq_vas.va.csa); amdgpu_bo_unreserve(queue->vm->root.bo); if (r) { kfree(mqd_sdma_v11); @@ -430,10 +446,12 @@ static int mes_userq_mqd_create(struct amdgpu_usermode_queue *queue, return 0; free_ctx: - amdgpu_userq_destroy_object(uq_mgr, &queue->fw_obj); + amdgpu_bo_free_kernel(&queue->fw_obj.obj, &queue->fw_obj.gpu_addr, + &queue->fw_obj.cpu_ptr); free_mqd: - amdgpu_userq_destroy_object(uq_mgr, &queue->mqd); + amdgpu_bo_free_kernel(&queue->mqd.obj, &queue->mqd.gpu_addr, + &queue->mqd.cpu_ptr); free_props: kfree(userq_props); @@ -443,11 +461,12 @@ static int mes_userq_mqd_create(struct amdgpu_usermode_queue *queue, static void mes_userq_mqd_destroy(struct amdgpu_usermode_queue *queue) { - struct amdgpu_userq_mgr *uq_mgr = queue->userq_mgr; - amdgpu_userq_destroy_object(uq_mgr, &queue->fw_obj); + amdgpu_bo_free_kernel(&queue->fw_obj.obj, &queue->fw_obj.gpu_addr, + &queue->fw_obj.cpu_ptr); kfree(queue->userq_prop); - amdgpu_userq_destroy_object(uq_mgr, &queue->mqd); + amdgpu_bo_free_kernel(&queue->mqd.obj, &queue->mqd.gpu_addr, + &queue->mqd.cpu_ptr); } static int mes_userq_preempt(struct amdgpu_usermode_queue *queue) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 03b266b26738..8785f7810157 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -2300,6 +2300,11 @@ static int criu_restore_devices(struct kfd_process *p, ret = -EINVAL; goto exit; } + + if (pdd->drm_file) { + ret = -EINVAL; + goto exit; + } pdd->user_gpu_id = device_buckets[i].user_gpu_id; drm_file = fget(device_buckets[i].drm_fd); @@ -2310,11 +2315,6 @@ static int criu_restore_devices(struct kfd_process *p, goto exit; } - if (pdd->drm_file) { - ret = -EINVAL; - goto exit; - } - /* create the vm using render nodes for kfd pdd */ if (kfd_process_device_init_vm(pdd, drm_file)) { pr_err("could not init vm for given pdd\n"); 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 e0a31e11f0ff..0d7296c739ed 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -3308,12 +3308,14 @@ static void copy_context_work_handler(struct work_struct *work) static uint32_t *get_queue_ids(uint32_t num_queues, uint32_t *usr_queue_id_array) { - size_t array_size = num_queues * sizeof(uint32_t); - if (!usr_queue_id_array) return NULL; - return memdup_user(usr_queue_id_array, array_size); + if (num_queues > KFD_MAX_NUM_OF_QUEUES_PER_PROCESS) + return ERR_PTR(-EINVAL); + + return memdup_user(usr_queue_id_array, + array_size(num_queues, sizeof(uint32_t))); } int resume_queues(struct kfd_process *p, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 35ec67d9739b..3841943da5ec 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -3732,6 +3732,9 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, svms = &p->svms; + if (!process_info) + return -EINVAL; + mutex_lock(&process_info->lock); svm_range_list_lock_and_flush_work(svms, mm); diff --git a/drivers/gpu/drm/amd/display/dc/dccg/dcn21/dcn21_dccg.c b/drivers/gpu/drm/amd/display/dc/dccg/dcn21/dcn21_dccg.c index c4d4eea140f3..1f23dfccf07a 100644 --- a/drivers/gpu/drm/amd/display/dc/dccg/dcn21/dcn21_dccg.c +++ b/drivers/gpu/drm/amd/display/dc/dccg/dcn21/dcn21_dccg.c @@ -105,15 +105,26 @@ static void dccg21_update_dpp_dto(struct dccg *dccg, int dpp_inst, int req_dppcl * dccg2_init() unconditionally overwrites MICROSECOND_TIME_BASE_DIV to * 0x00120264, destroying the marker before it can be read. * - * Guard the call: if the S0i3 marker is present, skip dccg2_init() so the + * Guard the call: if the S0i3 marker is present, skip init so the * WA can function correctly. bios_golden_init() will handle init in that case. + * + * DCN21 uses 48MHz refclk, not 100MHz, so we must explicitly set the correct + * values (48MHz is taken from rn_clk_mgr_construct()). */ static void dccg21_init(struct dccg *dccg) { + struct dcn_dccg *dccg_dcn = TO_DCN_DCCG(dccg); + if (dccg2_is_s0i3_golden_init_wa_done(dccg)) return; - dccg2_init(dccg); + /* 48MHz refclk from rn_clk_mgr_construct() */ + REG_WRITE(MICROSECOND_TIME_BASE_DIV, 0x00120230); + REG_WRITE(MILLISECOND_TIME_BASE_DIV, 0x0010bb80); + REG_WRITE(DISPCLK_FREQ_CHANGE_CNTL, 0x0e01003c); + + if (REG(REFCLK_CNTL)) + REG_WRITE(REFCLK_CNTL, 0); } static const struct dccg_funcs dccg21_funcs = { diff --git a/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c b/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c index 36942467d4ad..c3aff5d0c53d 100644 --- a/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c +++ b/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c @@ -3076,6 +3076,10 @@ static bool si_dpm_vblank_too_short(void *handle) /* we never hit the non-gddr5 limit so disable it */ u32 switch_limit = adev->gmc.vram_type == AMDGPU_VRAM_TYPE_GDDR5 ? 450 : 0; + /* Disregard vblank time when there are no displays connected */ + if (!adev->pm.pm_display_cfg.num_display) + return false; + /* Consider zero vblank time too short and disable MCLK switching. * Note that the vblank time is set to maximum when no displays are attached, * so we'll still enable MCLK switching in that case.