NVMe: Fix hot cpu notification dead lock

There is a potential dead lock if a cpu event occurs during nvme probe
since it registered with hot cpu notification. This fixes the race by
having the module register with notification outside of probe rather
than have each device register.

The actual work is done in a scheduled work queue instead of in the
notifier since assigning IO queues has the potential to block if the
driver creates additional queues.

Signed-off-by: Keith Busch <[e-mail]>
Signed-off-by: Matthew Wilcox <[e-mail]>

drivers/block/nvme-core.c

···
 static struct task_struct *nvme_thread;
 static struct workqueue_struct *nvme_workq;
 static wait_queue_head_t nvme_kthread_wait;
+static struct notifier_block nvme_nb;
 
 static void nvme_reset_failed_dev(struct work_struct *ws);
 
···
 	return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
 }
 
+static void nvme_cpu_workfn(struct work_struct *work)
+{
+	struct nvme_dev *dev = container_of(work, struct nvme_dev, cpu_work);
+	if (dev->initialized)
+		nvme_assign_io_queues(dev);
+}
+
 static int nvme_cpu_notify(struct notifier_block *self,
 				unsigned long action, void *hcpu)
 {
-	struct nvme_dev *dev = container_of(self, struct nvme_dev, nb);
+	struct nvme_dev *dev;
+
 	switch (action) {
 	case CPU_ONLINE:
 	case CPU_DEAD:
-		nvme_assign_io_queues(dev);
+		spin_lock(&dev_list_lock);
+		list_for_each_entry(dev, &dev_list, node)
+			schedule_work(&dev->cpu_work);
+		spin_unlock(&dev_list_lock);
 		break;
 	}
 	return NOTIFY_OK;
···
 	/* Free previously allocated queues that are no longer usable */
 	nvme_free_queues(dev, nr_io_queues + 1);
 	nvme_assign_io_queues(dev);
-
-	dev->nb.notifier_call = &nvme_cpu_notify;
-	result = register_hotcpu_notifier(&dev->nb);
-	if (result)
-		goto free_queues;
 
 	return 0;
 
···
 	int i;
 
 	dev->initialized = 0;
-	unregister_hotcpu_notifier(&dev->nb);
-
 	nvme_dev_list_remove(dev);
 
 	if (!dev->bar || (dev->bar && readl(&dev->bar->csts) == -1)) {
···
 	INIT_LIST_HEAD(&dev->namespaces);
 	dev->reset_workfn = nvme_reset_failed_dev;
 	INIT_WORK(&dev->reset_work, nvme_reset_workfn);
+	INIT_WORK(&dev->cpu_work, nvme_cpu_workfn);
 	dev->pci_dev = pdev;
 	pci_set_drvdata(pdev, dev);
 	result = nvme_set_instance(dev);
···
 
 	pci_set_drvdata(pdev, NULL);
 	flush_work(&dev->reset_work);
+	flush_work(&dev->cpu_work);
 	misc_deregister(&dev->miscdev);
 	nvme_dev_remove(dev);
 	nvme_dev_shutdown(dev);
···
 	else if (result > 0)
 		nvme_major = result;
 
-	result = pci_register_driver(&nvme_driver);
+	nvme_nb.notifier_call = &nvme_cpu_notify;
+	result = register_hotcpu_notifier(&nvme_nb);
 	if (result)
 		goto unregister_blkdev;
+
+	result = pci_register_driver(&nvme_driver);
+	if (result)
+		goto unregister_hotcpu;
 	return 0;
 
+ unregister_hotcpu:
+	unregister_hotcpu_notifier(&nvme_nb);
  unregister_blkdev:
 	unregister_blkdev(nvme_major, "nvme");
  kill_workq:
···
 static void __exit nvme_exit(void)
 {
 	pci_unregister_driver(&nvme_driver);
+	unregister_hotcpu_notifier(&nvme_nb);
 	unregister_blkdev(nvme_major, "nvme");
 	destroy_workqueue(nvme_workq);
 	BUG_ON(nvme_thread && !IS_ERR(nvme_thread));

include/linux/nvme.h

···
 	struct miscdevice miscdev;
 	work_func_t reset_workfn;
 	struct work_struct reset_work;
-	struct notifier_block nb;
+	struct work_struct cpu_work;
 	char name[12];
 	char serial[20];
 	char model[40];