From 219be4711a1ba788bc2a9fafc117139d133e5fea Mon Sep 17 00:00:00 2001 From: Prike Liang Date: Wed, 14 May 2025 12:43:57 +0800 Subject: drm/amdgpu: validate userq input args This will help on validating the userq input args, and rejecting for the invalid userq request at the IOCTLs first place. Signed-off-by: Prike Liang Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/mes_userqueue.c | 7 ------- 1 file changed, 7 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/mes_userqueue.c') diff --git a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c index aee26f80bd53..66467f41294c 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c @@ -263,13 +263,6 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr *uq_mgr, return -ENOMEM; } - if (!mqd_user->wptr_va || !mqd_user->rptr_va || - !mqd_user->queue_va || mqd_user->queue_size == 0) { - DRM_ERROR("Invalid MQD parameters for userqueue\n"); - r = -EINVAL; - goto free_props; - } - r = amdgpu_userq_create_object(uq_mgr, &queue->mqd, mqd_hw_default->mqd_size); if (r) { DRM_ERROR("Failed to create MQD object for userqueue\n"); -- cgit v1.2.3 From 9e46b8bb0539d7bc9a9e7b3072fa4f6082490392 Mon Sep 17 00:00:00 2001 From: Prike Liang Date: Mon, 23 Jun 2025 16:29:38 +0800 Subject: drm/amdgpu: validate userq buffer virtual address and size It needs to validate the userq object virtual address to determine whether it is residented in a valid vm mapping. Signed-off-by: Prike Liang Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/mes_userqueue.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/mes_userqueue.c') diff --git a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c index 66467f41294c..3a4fd6de08ce 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c @@ -254,6 +254,7 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr *uq_mgr, struct amdgpu_mqd *mqd_hw_default = &adev->mqds[queue->queue_type]; struct drm_amdgpu_userq_in *mqd_user = args_in; struct amdgpu_mqd_prop *userq_props; + struct amdgpu_gfx_shadow_info shadow_info; int r; /* Structure to initialize MQD for userqueue using generic MQD init function */ @@ -279,6 +280,8 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr *uq_mgr, userq_props->doorbell_index = queue->doorbell_index; userq_props->fence_address = queue->fence_drv->gpu_addr; + if (adev->gfx.funcs->get_gfx_shadow_info) + adev->gfx.funcs->get_gfx_shadow_info(adev, &shadow_info, true); if (queue->queue_type == AMDGPU_HW_IP_COMPUTE) { struct drm_amdgpu_userq_mqd_compute_gfx11 *compute_mqd; @@ -295,6 +298,10 @@ 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))) + goto free_mqd; + userq_props->eop_gpu_addr = compute_mqd->eop_va; userq_props->hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_NORMAL; userq_props->hqd_queue_priority = AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM; @@ -322,6 +329,11 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr *uq_mgr, userq_props->csa_addr = mqd_gfx_v11->csa_va; 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)) + goto free_mqd; + kfree(mqd_gfx_v11); } else if (queue->queue_type == AMDGPU_HW_IP_DMA) { struct drm_amdgpu_userq_mqd_sdma_gfx11 *mqd_sdma_v11; @@ -339,6 +351,10 @@ static int mes_userq_mqd_create(struct amdgpu_userq_mgr *uq_mgr, goto free_mqd; } + if (amdgpu_userq_input_va_validate(queue->vm, mqd_sdma_v11->csa_va, + shadow_info.csa_size)) + goto free_mqd; + userq_props->csa_addr = mqd_sdma_v11->csa_va; kfree(mqd_sdma_v11); } -- cgit v1.2.3 From 5cefcbb306d6b4fa3ac5fba86db94cac81f0b457 Mon Sep 17 00:00:00 2001 From: "Jesse.Zhang" Date: Thu, 4 Sep 2025 16:30:08 +0800 Subject: drm/amdgpu: adjust MES API used for suspend and resume Use the suspend and resume API rather than remove queue and add queue API. The former just preempts the queue while the latter remove it from the scheduler completely. There is no need to do that, we only need preemption in this case. V2: replace queue_active with queue state v3: set the suspend_fence_addr v4: allocate another per queue buffer for the suspend fence, and set the sequence number. also wait for the suspend fence. (Alex) v5: use a wb slot (Alex) v6: Change the timeout period. For MES, the default timeout is 2100000; /* 2100 ms */ (Alex) Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher Signed-off-by: Jesse Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/mes_userqueue.c | 72 ++++++++++++++++++++++++++++++ 1 file changed, 72 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/mes_userqueue.c') diff --git a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c index 3a4fd6de08ce..2db9b2c63693 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_userqueue.c @@ -404,10 +404,82 @@ mes_userq_mqd_destroy(struct amdgpu_userq_mgr *uq_mgr, amdgpu_userq_destroy_object(uq_mgr, &queue->mqd); } +static int mes_userq_preempt(struct amdgpu_userq_mgr *uq_mgr, + struct amdgpu_usermode_queue *queue) +{ + struct amdgpu_device *adev = uq_mgr->adev; + struct mes_suspend_gang_input queue_input; + struct amdgpu_userq_obj *ctx = &queue->fw_obj; + signed long timeout = 2100000; /* 2100 ms */ + u64 fence_gpu_addr; + u32 fence_offset; + u64 *fence_ptr; + int i, r; + + if (queue->state != AMDGPU_USERQ_STATE_MAPPED) + return 0; + r = amdgpu_device_wb_get(adev, &fence_offset); + if (r) + return r; + + fence_gpu_addr = adev->wb.gpu_addr + (fence_offset * 4); + fence_ptr = (u64 *)&adev->wb.wb[fence_offset]; + *fence_ptr = 0; + + memset(&queue_input, 0x0, sizeof(struct mes_suspend_gang_input)); + queue_input.gang_context_addr = ctx->gpu_addr + AMDGPU_USERQ_PROC_CTX_SZ; + queue_input.suspend_fence_addr = fence_gpu_addr; + queue_input.suspend_fence_value = 1; + amdgpu_mes_lock(&adev->mes); + r = adev->mes.funcs->suspend_gang(&adev->mes, &queue_input); + amdgpu_mes_unlock(&adev->mes); + if (r) { + DRM_ERROR("Failed to suspend gang: %d\n", r); + goto out; + } + + for (i = 0; i < timeout; i++) { + if (*fence_ptr == 1) + goto out; + udelay(1); + } + r = -ETIMEDOUT; + +out: + amdgpu_device_wb_free(adev, fence_offset); + return r; +} + +static int mes_userq_restore(struct amdgpu_userq_mgr *uq_mgr, + struct amdgpu_usermode_queue *queue) +{ + struct amdgpu_device *adev = uq_mgr->adev; + struct mes_resume_gang_input queue_input; + struct amdgpu_userq_obj *ctx = &queue->fw_obj; + int r; + + if (queue->state == AMDGPU_USERQ_STATE_HUNG) + return -EINVAL; + if (queue->state != AMDGPU_USERQ_STATE_PREEMPTED) + return 0; + + memset(&queue_input, 0x0, sizeof(struct mes_resume_gang_input)); + queue_input.gang_context_addr = ctx->gpu_addr + AMDGPU_USERQ_PROC_CTX_SZ; + + amdgpu_mes_lock(&adev->mes); + r = adev->mes.funcs->resume_gang(&adev->mes, &queue_input); + amdgpu_mes_unlock(&adev->mes); + if (r) + dev_err(adev->dev, "Failed to resume queue, err (%d)\n", r); + return r; +} + const struct amdgpu_userq_funcs userq_mes_funcs = { .mqd_create = mes_userq_mqd_create, .mqd_destroy = mes_userq_mqd_destroy, .unmap = mes_userq_unmap, .map = mes_userq_map, .detect_and_reset = mes_userq_detect_and_reset, + .preempt = mes_userq_preempt, + .restore = mes_userq_restore, }; -- cgit v1.2.3