radv: Implement vk.check_status

Implement check_status function so the driver can check if the GPU has
been reset by the application, and thus if it's still available.

AMDGPU_CTX_QUERY ioctls work by asking amdgpu if this context was the
cause of a previous GPU reset.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: André Almeida <andrealmeid@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22253>

diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index feaa5ac..51540c8 100644 (file)
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -602,6 +602,28 @@ init_dispatch_tables(struct radv_device *device, struct radv_physical_device *ph
    add_entrypoints(&b, &vk_common_device_entrypoints, RADV_DISPATCH_TABLE_COUNT);
 }
 
+static VkResult
+radv_check_status(struct vk_device *vk_device)
+{
+   struct radv_device *device = container_of(vk_device, struct radv_device, vk);
+   enum radv_reset_status status;
+
+   for (int i = 0; i < RADV_NUM_HW_CTX; i++) {
+      if (device->hw_ctx[i]) {
+         status = device->ws->ctx_query_reset_status(device->hw_ctx[i]);
+
+         if (status == RADV_GUILTY_CONTEXT_RESET)
+            return vk_device_set_lost(&device->vk, "GPU hung detected in this process");
+        if (status == RADV_INNOCENT_CONTEXT_RESET)
+            return vk_device_set_lost(&device->vk, "GPU hung triggered by other process");
+        if (status == RADV_UNKNOWN_CONTEXT_RESET)
+            return vk_device_set_lost(&device->vk, "GPU hung triggered by unknown source");
+      }
+   }
+
+   return VK_SUCCESS;
+}
+
 VKAPI_ATTR VkResult VKAPI_CALL
 radv_CreateDevice(VkPhysicalDevice physicalDevice, const VkDeviceCreateInfo *pCreateInfo,
                   const VkAllocationCallbacks *pAllocator, VkDevice *pDevice)
@@ -750,6 +772,7 @@ radv_CreateDevice(VkPhysicalDevice physicalDevice, const VkDeviceCreateInfo *pCr
    init_dispatch_tables(device, physical_device);
 
    device->vk.command_buffer_ops = &radv_cmd_buffer_ops;
+   device->vk.check_status = radv_check_status;
 
    device->instance = physical_device->instance;
    device->physical_device = physical_device;

diff --git a/src/amd/vulkan/radv_radeon_winsys.h b/src/amd/vulkan/radv_radeon_winsys.h
index 32d8795..28eb6e7 100644 (file)
--- a/src/amd/vulkan/radv_radeon_winsys.h
+++ b/src/amd/vulkan/radv_radeon_winsys.h
@@ -102,6 +102,13 @@ enum radeon_value_id {
    RADEON_CURRENT_MCLK,
 };
 
+enum radv_reset_status {
+   RADV_NO_RESET,
+   RADV_GUILTY_CONTEXT_RESET,
+   RADV_INNOCENT_CONTEXT_RESET,
+   RADV_UNKNOWN_CONTEXT_RESET,
+};
+
 struct radeon_cmdbuf {
    /* These are uint64_t to tell the compiler that buf can't alias them.
     * If they're uint32_t the generated code needs to redundantly
@@ -275,6 +282,8 @@ struct radeon_winsys {
 
    int (*ctx_set_pstate)(struct radeon_winsys_ctx *ctx, uint32_t pstate);
 
+   enum radv_reset_status (*ctx_query_reset_status)(struct radeon_winsys_ctx *rwctx);
+
    enum radeon_bo_domain (*cs_domain)(const struct radeon_winsys *ws);
 
    struct radeon_cmdbuf *(*cs_create)(struct radeon_winsys *ws, enum amd_ip_type amd_ip_type);

diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
index b668113..9e85190 100644 (file)
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
@@ -1560,6 +1560,52 @@ radv_amdgpu_ctx_wait_idle(struct radeon_winsys_ctx *rwctx, enum amd_ip_type ip_t
    return true;
 }
 
+static enum radv_reset_status
+radv_amdgpu_ctx_query_reset_status(struct radeon_winsys_ctx *rwctx)
+{
+   int ret;
+   struct radv_amdgpu_ctx *ctx = (struct radv_amdgpu_ctx *)rwctx;
+
+   if (ctx->ws->info.drm_minor >= 24) {
+      uint64_t flags;
+      ret = amdgpu_cs_query_reset_state2(ctx->ctx, &flags);
+
+      if (ret) {
+         fprintf(stderr, "radv/amdgpu: amdgpu_cs_query_reset_state2 failed. (%i)\n", ret);
+        return RADV_NO_RESET;
+      }
+
+      if (flags & AMDGPU_CTX_QUERY2_FLAGS_RESET) {
+         if (flags & AMDGPU_CTX_QUERY2_FLAGS_GUILTY) {
+            /* Some job from this context once caused a GPU hang */
+            return RADV_GUILTY_CONTEXT_RESET;
+         } else {
+            /* Some job from other context caused a GPU hang */
+            return RADV_INNOCENT_CONTEXT_RESET;
+         }
+      }
+   } else {
+      uint32_t result, hangs;
+
+      ret = amdgpu_cs_query_reset_state(ctx->ctx, &result, &hangs);
+      if (ret) {
+         fprintf(stderr, "radv/amdgpu: amdgpu_cs_query_reset_state failed. (%i)\n", ret);
+         return RADV_NO_RESET;
+      }
+
+      switch (result) {
+      case AMDGPU_CTX_GUILTY_RESET:
+         return RADV_GUILTY_CONTEXT_RESET;
+      case AMDGPU_CTX_INNOCENT_RESET:
+         return RADV_INNOCENT_CONTEXT_RESET;
+      case AMDGPU_CTX_UNKNOWN_RESET:
+         return RADV_UNKNOWN_CONTEXT_RESET;
+      }
+   }
+
+   return RADV_NO_RESET;
+}
+
 static uint32_t
 radv_to_amdgpu_pstate(enum radeon_ctx_pstate radv_pstate)
 {
@@ -1871,6 +1917,7 @@ radv_amdgpu_cs_init_functions(struct radv_amdgpu_winsys *ws)
    ws->base.ctx_destroy = radv_amdgpu_ctx_destroy;
    ws->base.ctx_wait_idle = radv_amdgpu_ctx_wait_idle;
    ws->base.ctx_set_pstate = radv_amdgpu_ctx_set_pstate;
+   ws->base.ctx_query_reset_status = radv_amdgpu_ctx_query_reset_status;
    ws->base.cs_domain = radv_amdgpu_cs_domain;
    ws->base.cs_create = radv_amdgpu_cs_create;
    ws->base.cs_destroy = radv_amdgpu_cs_destroy;