On Fri, Sep 12, 2025 at 1:55 AM Prike Liang <prike.li...@amd.com> wrote:
>
> Add the userq object virtual address get(),mapped() and put()
> helpers for tracking the userq obj va address usage.
>
> Signed-off-by: Prike Liang <prike.li...@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 73 ++++++++++++++++++++--
> drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h | 17 ++++-
> drivers/gpu/drm/amd/amdgpu/mes_userqueue.c | 13 ++--
> 4 files changed, 90 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> index 87523fcd4386..912c9afaf9e1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> @@ -96,6 +96,7 @@ struct amdgpu_bo_va {
> * if non-zero, cannot unmap from GPU because user queues may still
> access it
> */
> unsigned int queue_refcount;
> + atomic_t userq_va_mapped;
> };
>
> struct amdgpu_bo {
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> index ba7be4db5a01..19a400bd4be1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> @@ -44,7 +44,8 @@ u32 amdgpu_userq_get_supported_ip_mask(struct amdgpu_device
> *adev)
> return userq_ip_mask;
> }
>
> -int amdgpu_userq_input_va_validate(struct amdgpu_vm *vm, u64 addr,
> +int amdgpu_userq_input_va_validate(struct amdgpu_usermode_queue *queue,
> + struct amdgpu_vm *vm, u64 addr,
> u64 expected_size)
> {
> struct amdgpu_bo_va_mapping *va_map;
> @@ -67,6 +68,7 @@ int amdgpu_userq_input_va_validate(struct amdgpu_vm *vm,
> u64 addr,
> /* 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_get(queue, va_map, user_addr);
> amdgpu_bo_unreserve(vm->root.bo);
> return 0;
> }
> @@ -76,6 +78,66 @@ int amdgpu_userq_input_va_validate(struct amdgpu_vm *vm,
> u64 addr,
> return r;
> }
>
> +int amdgpu_userq_buffer_va_get(struct amdgpu_usermode_queue *queue,
> + struct amdgpu_bo_va_mapping *va_map, u64 addr)
I'd rename this to amdgpu_userq_buffer_va_list_add(). We aren't
really treating this like a reference count. we are just adding it to
a list.
> +{
> + struct amdgpu_userq_va_cursor *va_cursor;
> + struct userq_va_list;
> +
> + va_cursor = kzalloc(sizeof(*va_cursor), GFP_KERNEL);
> + if (!va_cursor)
> + return -ENOMEM;
> +
> + INIT_LIST_HEAD(&va_cursor->list);
> + va_cursor->gpu_addr = addr;
> + atomic_set(&va_map->bo_va->userq_va_mapped, 1);
> + list_add(&va_cursor->list, &queue->userq_va_list);
> +
> + return 0;
> +
> +}
> +
> +bool amdgpu_userq_buffer_va_mapped(struct amdgpu_vm *vm, u64 addr)
> +{
> + struct amdgpu_bo_va_mapping *mapping;
> + bool r;
> +
> + if (amdgpu_bo_reserve(vm->root.bo, false))
> + return false;
> +
> + mapping = amdgpu_vm_bo_lookup_mapping(vm, addr);
> + if (!IS_ERR_OR_NULL(mapping) &&
> atomic_read(&mapping->bo_va->userq_va_mapped))
> + r = true;
> + else
> + r = false;
> + amdgpu_bo_unreserve(vm->root.bo);
> +
> + return r;
> +}
> +
> +int amdgpu_userq_buffer_va_put(struct amdgpu_vm *vm,
And here amdgpu_userq_buffer_va_list_del().
Alex
> + u64 addr)
> +{
> + struct amdgpu_bo_va_mapping *mapping;
> + int r;
> +
> + r = amdgpu_bo_reserve(vm->root.bo, false);
> + if (r)
> + return r;
> +
> + mapping = amdgpu_vm_bo_lookup_mapping(vm, addr);
> + if (!mapping)
> + goto out_err;
> +
> + atomic_set(&mapping->bo_va->userq_va_mapped, 0);
> + amdgpu_bo_unreserve(vm->root.bo);
> + return 0;
> +
> +out_err:
> + amdgpu_bo_unreserve(vm->root.bo);
> + return -EINVAL;
> +}
> +
> static int
> amdgpu_userq_unmap_helper(struct amdgpu_userq_mgr *uq_mgr,
> struct amdgpu_usermode_queue *queue)
> @@ -141,6 +203,7 @@ amdgpu_userq_cleanup(struct amdgpu_userq_mgr *uq_mgr,
> uq_funcs->mqd_destroy(uq_mgr, queue);
> amdgpu_userq_fence_driver_free(queue);
> idr_remove(&uq_mgr->userq_idr, queue_id);
> + list_del(&queue->userq_va_list);
> kfree(queue);
> }
>
> @@ -461,10 +524,11 @@ amdgpu_userq_create(struct drm_file *filp, union
> drm_amdgpu_userq *args)
> goto unlock;
> }
>
> + INIT_LIST_HEAD(&queue->userq_va_list);
> /* Validate the userq virtual address.*/
> - if (amdgpu_userq_input_va_validate(&fpriv->vm, args->in.queue_va,
> args->in.queue_size) ||
> - amdgpu_userq_input_va_validate(&fpriv->vm, args->in.rptr_va,
> AMDGPU_GPU_PAGE_SIZE) ||
> - amdgpu_userq_input_va_validate(&fpriv->vm, args->in.wptr_va,
> AMDGPU_GPU_PAGE_SIZE)) {
> + if (amdgpu_userq_input_va_validate(queue, &fpriv->vm,
> args->in.queue_va, args->in.queue_size) ||
> + amdgpu_userq_input_va_validate(queue, &fpriv->vm,
> args->in.rptr_va, AMDGPU_GPU_PAGE_SIZE ) ||
> + amdgpu_userq_input_va_validate(queue, &fpriv->vm,
> args->in.wptr_va, AMDGPU_GPU_PAGE_SIZE)) {
> kfree(queue);
> goto unlock;
> }
> @@ -504,7 +568,6 @@ amdgpu_userq_create(struct drm_file *filp, union
> drm_amdgpu_userq *args)
> goto unlock;
> }
>
> -
> qid = idr_alloc(&uq_mgr->userq_idr, queue, 1, AMDGPU_MAX_USERQ_COUNT,
> GFP_KERNEL);
> if (qid < 0) {
> drm_file_err(uq_mgr->file, "Failed to allocate a queue id\n");
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h
> index 33a582200dda..e4c2d7cabd15 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h
> @@ -48,12 +48,19 @@ struct amdgpu_userq_obj {
> struct amdgpu_bo *obj;
> };
>
> +struct amdgpu_userq_va_cursor {
> + uint64_t gpu_addr;
> + struct list_head list;
> +};
> +
> struct amdgpu_usermode_queue {
> int queue_type;
> enum amdgpu_userq_state state;
> uint64_t doorbell_handle;
> uint64_t doorbell_index;
> +
> uint64_t flags;
> +
> struct amdgpu_mqd_prop *userq_prop;
> struct amdgpu_userq_mgr *userq_mgr;
> struct amdgpu_vm *vm;
> @@ -68,6 +75,8 @@ struct amdgpu_usermode_queue {
> int priority;
> struct dentry *debugfs_queue;
> uint64_t generation;
> +
> + struct list_head userq_va_list;
> };
>
> struct amdgpu_userq_funcs {
> @@ -138,7 +147,11 @@ int amdgpu_userq_stop_sched_for_enforce_isolation(struct
> amdgpu_device *adev,
> u32 idx);
> int amdgpu_userq_start_sched_for_enforce_isolation(struct amdgpu_device
> *adev,
> u32 idx);
> -
> -int amdgpu_userq_input_va_validate(struct amdgpu_vm *vm, u64 addr,
> +int amdgpu_userq_input_va_validate(struct amdgpu_usermode_queue *queue,
> + struct amdgpu_vm *vm, u64 addr,
> u64 expected_size);
> +int amdgpu_userq_buffer_va_get(struct amdgpu_usermode_queue *queue,
> + struct amdgpu_bo_va_mapping *va_map, u64 addr);
> +bool amdgpu_userq_buffer_va_mapped(struct amdgpu_vm *vm, u64 addr);
> +int amdgpu_userq_buffer_va_put(struct amdgpu_vm *vm, u64 addr);
> #endif
> diff --git a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c
> b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c
> index 3a4fd6de08ce..e10e4b6a10d6 100644
> --- a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c
> +++ b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c
> @@ -298,8 +298,8 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr
> *uq_mgr,
> goto free_mqd;
> }
>
> - if (amdgpu_userq_input_va_validate(queue->vm,
> compute_mqd->eop_va,
> - max_t(u32, PAGE_SIZE, AMDGPU_GPU_PAGE_SIZE)))
> + if (amdgpu_userq_input_va_validate(queue, queue->vm,
> compute_mqd->eop_va,
> + max_t(u32, PAGE_SIZE,
> AMDGPU_GPU_PAGE_SIZE)))
> goto free_mqd;
>
> userq_props->eop_gpu_addr = compute_mqd->eop_va;
> @@ -330,8 +330,8 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr
> *uq_mgr,
> userq_props->tmz_queue =
> mqd_user->flags &
> AMDGPU_USERQ_CREATE_FLAGS_QUEUE_SECURE;
>
> - if (amdgpu_userq_input_va_validate(queue->vm,
> mqd_gfx_v11->shadow_va,
> - shadow_info.shadow_size))
> + if (amdgpu_userq_input_va_validate(queue, queue->vm,
> mqd_gfx_v11->shadow_va,
> + shadow_info.shadow_size))
> goto free_mqd;
>
> kfree(mqd_gfx_v11);
> @@ -350,9 +350,8 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr
> *uq_mgr,
> r = -ENOMEM;
> goto free_mqd;
> }
> -
> - if (amdgpu_userq_input_va_validate(queue->vm,
> mqd_sdma_v11->csa_va,
> - shadow_info.csa_size))
> + if (amdgpu_userq_input_va_validate(queue, queue->vm,
> mqd_sdma_v11->csa_va,
> + shadow_info.csa_size))
> goto free_mqd;
>
> userq_props->csa_addr = mqd_sdma_v11->csa_va;
> --
> 2.34.1
>
