amdgpu_plugin: Refactor code used to implement Checkpoint

Refactor code used to Checkpoint DRM devices. Code is moved
into amdgpu_plugin_drm.c file which hosts various methods to
checkpoint and restore a workload.

Signed-off-by: Ramesh Errabolu <[email protected]>

plugins/amdgpu/amdgpu_plugin.c

@@ -49,6 +49,13 @@ struct vma_metadata {
 
 /************************************ Global Variables ********************************************/
 
+/**
+ * FD of KFD device used to checkpoint. On a multi-process
+ * tree the order of checkpointing goes from parent to child
+ * and so on - so saving the FD will not be overwritten
+ */
+static int kfd_checkpoint_fd;
+
 static LIST_HEAD(update_vma_info_list);
 
 size_t kfd_max_buffer_size;
@@ -990,6 +997,10 @@ static int unpause_process(int fd)
 		goto exit;
 	}
 
+	// Reset the KFD FD
+	kfd_checkpoint_fd = -1;
+	sys_close_drm_render_devices(&src_topology);
+
 exit:
 	pr_info("Process unpaused %s (ret:%d)\n", ret ? "Failed" : "Ok", ret);
 
@@ -1181,44 +1192,25 @@ int amdgpu_plugin_dump_file(int fd, int id)
 		return -1;
 	}
 
+	/* Initialize number of device files that will be checkpointed */
+	init_gpu_count(&src_topology);
+
 	/* Check whether this plugin was called for kfd or render nodes */
 	if (major(st.st_rdev) != major(st_kfd.st_rdev) || minor(st.st_rdev) != 0) {
+
 		/* This is RenderD dumper plugin, for now just save renderD
 		 * minor number to be used during restore. In later phases this
 		 * needs to save more data for video decode etc.
 		 */
-
-		CriuRenderNode rd = CRIU_RENDER_NODE__INIT;
-		struct tp_node *tp_node;
-
-		pr_info("Dumper called for /dev/dri/renderD%d, FD = %d, ID = %d\n", minor(st.st_rdev), fd, id);
-
-		tp_node = sys_get_node_by_render_minor(&src_topology, minor(st.st_rdev));
-		if (!tp_node) {
-			pr_err("Failed to find a device with minor number = %d\n", minor(st.st_rdev));
-
-			return -ENODEV;
-		}
-
-		rd.gpu_id = maps_get_dest_gpu(&checkpoint_maps, tp_node->gpu_id);
-		if (!rd.gpu_id)
-			return -ENODEV;
-
-		len = criu_render_node__get_packed_size(&rd);
-		buf = xmalloc(len);
-		if (!buf)
-			return -ENOMEM;
-
-		criu_render_node__pack(&rd, buf);
-
-		snprintf(img_path, sizeof(img_path), IMG_DRM_FILE, id);
-		ret = write_img_file(img_path, buf, len);
-		if (ret) {
-			xfree(buf);
+		ret = amdgpu_plugin_drm_dump_file(fd, id, &st);
+		if (ret)
 			return ret;
-		}
 
-		xfree(buf);
+		/* Invoke unpause process if needed */
+		decrement_checkpoint_count();
+		if (checkpoint_is_complete()) {
+			ret = unpause_process(kfd_checkpoint_fd);
+		}
 
 		/* Need to return success here so that criu can call plugins for renderD nodes */
 		return ret;
@@ -1315,11 +1307,15 @@ int amdgpu_plugin_dump_file(int fd, int id)
 	ret = write_img_file(img_path, buf, len);
 
 	xfree(buf);
+
 exit:
-	/* Restore all queues */
-	unpause_process(fd);
+	/* Restore all queues if conditions permit */
+	kfd_checkpoint_fd = fd;
+	decrement_checkpoint_count();
+	if (checkpoint_is_complete()) {
+		ret = unpause_process(fd);
+	}
 
-	sys_close_drm_render_devices(&src_topology);
 	xfree((void *)args.devices);
 	xfree((void *)args.bos);
 	xfree((void *)args.priv_data);

plugins/amdgpu/amdgpu_plugin_drm.c

@@ -61,3 +61,41 @@ int amdgpu_plugin_drm_handle_device_vma(int fd, const struct stat *st)
 }
 
 
+int amdgpu_plugin_drm_dump_file(int fd, int id, struct stat *drm)
+{
+	CriuRenderNode rd = CRIU_RENDER_NODE__INIT;
+	struct tp_node *tp_node;
+	char path[PATH_MAX];
+	unsigned char *buf;
+	int minor;
+	int len;
+	int ret;
+
+	/* Get the topology node of the DRM device */
+	minor = minor(drm->st_rdev);
+	tp_node = sys_get_node_by_render_minor(&src_topology, minor);
+	if (!tp_node) {
+		pr_err("Failed to find a device with minor number = %d\n", minor);
+		return -ENODEV;
+	}
+
+	/* Get the GPU_ID of the DRM device */
+	rd.gpu_id = maps_get_dest_gpu(&checkpoint_maps, tp_node->gpu_id);
+	if (!rd.gpu_id) {
+		pr_err("Failed to find valid gpu_id for the device = %d\n", rd.gpu_id);
+		return -ENODEV;
+	}
+
+	len = criu_render_node__get_packed_size(&rd);
+	buf = xmalloc(len);
+	if (!buf)
+		return -ENOMEM;
+
+	criu_render_node__pack(&rd, buf);
+
+	snprintf(path, sizeof(path), IMG_DRM_FILE, id);
+	ret = write_img_file(path, buf, len);
+	xfree(buf);
+	return ret;
+}
+

plugins/amdgpu/amdgpu_plugin_drm.h

@@ -17,6 +17,12 @@
  */
 int amdgpu_plugin_drm_handle_device_vma(int fd, const struct stat *drm);
 
+/**
+ * Serialize meta-data about a particular DRM device, its number of BOs,
+ * etc into a file. The serialized filename has in it the value ID that
+ * is passed in as a parameter
+ */
+int amdgpu_plugin_drm_dump_file(int fd, int id, struct stat *drm);
 
 #endif		/* __AMDGPU_PLUGIN_DRM_H__ */