nvme: remove dead controllers from a work item
authorChristoph Hellwig <hch@lst.de>
Thu, 26 Nov 2015 11:35:49 +0000 (12:35 +0100)
committerJens Axboe <axboe@fb.com>
Tue, 22 Dec 2015 16:38:33 +0000 (09:38 -0700)
Compared to the kthread this gives us multiple call prevention for free.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
drivers/nvme/host/pci.c

index 23cbd93..26e9823 100644 (file)
@@ -80,7 +80,7 @@ struct nvme_iod;
 static int nvme_reset(struct nvme_dev *dev);
 static void nvme_process_cq(struct nvme_queue *nvmeq);
 static void nvme_unmap_data(struct nvme_dev *dev, struct nvme_iod *iod);
-static void nvme_dead_ctrl(struct nvme_dev *dev);
+static void nvme_remove_dead_ctrl(struct nvme_dev *dev);
 static void nvme_dev_shutdown(struct nvme_dev *dev);
 
 struct async_cmd_info {
@@ -113,6 +113,7 @@ struct nvme_dev {
        void __iomem *bar;
        struct work_struct reset_work;
        struct work_struct scan_work;
+       struct work_struct remove_work;
        struct mutex shutdown_lock;
        bool subsystem;
        void __iomem *cmb;
@@ -2218,31 +2219,25 @@ static void nvme_reset_work(struct work_struct *work)
  unmap:
        nvme_dev_unmap(dev);
  out:
-       if (!work_pending(&dev->reset_work))
-               nvme_dead_ctrl(dev);
+       nvme_remove_dead_ctrl(dev);
 }
 
-static int nvme_remove_dead_ctrl(void *arg)
+static void nvme_remove_dead_ctrl_work(struct work_struct *work)
 {
-       struct nvme_dev *dev = (struct nvme_dev *)arg;
+       struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work);
        struct pci_dev *pdev = to_pci_dev(dev->dev);
 
        if (pci_get_drvdata(pdev))
                pci_stop_and_remove_bus_device_locked(pdev);
        nvme_put_ctrl(&dev->ctrl);
-       return 0;
 }
 
-static void nvme_dead_ctrl(struct nvme_dev *dev)
+static void nvme_remove_dead_ctrl(struct nvme_dev *dev)
 {
-       dev_warn(dev->dev, "Device failed to resume\n");
+       dev_warn(dev->dev, "Removing after probe failure\n");
        kref_get(&dev->ctrl.kref);
-       if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d",
-                                               dev->ctrl.instance))) {
-               dev_err(dev->dev,
-                       "Failed to start controller remove task\n");
+       if (!schedule_work(&dev->remove_work))
                nvme_put_ctrl(&dev->ctrl);
-       }
 }
 
 static int nvme_reset(struct nvme_dev *dev)
@@ -2323,6 +2318,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
        INIT_LIST_HEAD(&dev->node);
        INIT_WORK(&dev->scan_work, nvme_dev_scan);
        INIT_WORK(&dev->reset_work, nvme_reset_work);
+       INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work);
        mutex_init(&dev->shutdown_lock);
 
        result = nvme_setup_prp_pools(dev);