nvme: replace the kthread with a per-device watchdog timer
authorChristoph Hellwig <hch@lst.de>
Mon, 29 Feb 2016 14:59:46 +0000 (15:59 +0100)
committerJens Axboe <axboe@fb.com>
Mon, 29 Feb 2016 15:47:16 +0000 (08:47 -0700)
The only work left in the kthread is the periodic health check for each
controller.  There is no need to run this from process context or keep
a thread context around for it, so replace it with a simpler timer.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
drivers/nvme/host/pci.c

index 10839f76179c9ae00c72424b6f968a9ce08a1d9e..a6233605117843fbb97f6b6c1918120c9e73aef2 100644 (file)
@@ -27,7 +27,6 @@
 #include <linux/interrupt.h>
 #include <linux/io.h>
 #include <linux/kdev_t.h>
-#include <linux/kthread.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
 #include <linux/module.h>
@@ -39,6 +38,7 @@
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/t10-pi.h>
+#include <linux/timer.h>
 #include <linux/types.h>
 #include <linux/io-64-nonatomic-lo-hi.h>
 #include <asm/unaligned.h>
@@ -64,11 +64,7 @@ static bool use_cmb_sqes = true;
 module_param(use_cmb_sqes, bool, 0644);
 MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes");
 
-static LIST_HEAD(dev_list);
-static DEFINE_SPINLOCK(dev_list_lock);
-static struct task_struct *nvme_thread;
 static struct workqueue_struct *nvme_workq;
-static wait_queue_head_t nvme_kthread_wait;
 
 struct nvme_dev;
 struct nvme_queue;
@@ -82,7 +78,6 @@ static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown);
  * Represents an NVM Express device.  Each nvme_dev is a PCI function.
  */
 struct nvme_dev {
-       struct list_head node;
        struct nvme_queue **queues;
        struct blk_mq_tag_set tagset;
        struct blk_mq_tag_set admin_tagset;
@@ -101,6 +96,7 @@ struct nvme_dev {
        struct work_struct scan_work;
        struct work_struct remove_work;
        struct work_struct async_work;
+       struct timer_list watchdog_timer;
        struct mutex shutdown_lock;
        bool subsystem;
        void __iomem *cmb;
@@ -1334,36 +1330,26 @@ static int nvme_configure_admin_queue(struct nvme_dev *dev)
        return result;
 }
 
-static int nvme_kthread(void *data)
+static void nvme_watchdog_timer(unsigned long data)
 {
-       struct nvme_dev *dev, *next;
+       struct nvme_dev *dev = (struct nvme_dev *)data;
+       u32 csts = readl(dev->bar + NVME_REG_CSTS);
 
-       while (!kthread_should_stop()) {
-               set_current_state(TASK_INTERRUPTIBLE);
-               spin_lock(&dev_list_lock);
-               list_for_each_entry_safe(dev, next, &dev_list, node) {
-                       u32 csts = readl(dev->bar + NVME_REG_CSTS);
-
-                       /*
-                        * Skip controllers currently under reset.
-                        */
-                       if (work_pending(&dev->reset_work) || work_busy(&dev->reset_work))
-                               continue;
-
-                       if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) ||
-                                                       csts & NVME_CSTS_CFS) {
-                               if (queue_work(nvme_workq, &dev->reset_work)) {
-                                       dev_warn(dev->ctrl.device,
-                                               "Failed status: %x, reset controller\n",
-                                               readl(dev->bar + NVME_REG_CSTS));
-                               }
-                               continue;
-                       }
+       /*
+        * Skip controllers currently under reset.
+        */
+       if (!work_pending(&dev->reset_work) && !work_busy(&dev->reset_work) &&
+           ((csts & NVME_CSTS_CFS) ||
+            (dev->subsystem && (csts & NVME_CSTS_NSSRO)))) {
+               if (queue_work(nvme_workq, &dev->reset_work)) {
+                       dev_warn(dev->dev,
+                               "Failed status: 0x%x, reset controller.\n",
+                               csts);
                }
-               spin_unlock(&dev_list_lock);
-               schedule_timeout(round_jiffies_relative(HZ));
+               return;
        }
-       return 0;
+
+       mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
 }
 
 static int nvme_create_io_queues(struct nvme_dev *dev)
@@ -1777,56 +1763,12 @@ static void nvme_dev_unmap(struct nvme_dev *dev)
        }
 }
 
-static int nvme_dev_list_add(struct nvme_dev *dev)
-{
-       bool start_thread = false;
-
-       spin_lock(&dev_list_lock);
-       if (list_empty(&dev_list) && IS_ERR_OR_NULL(nvme_thread)) {
-               start_thread = true;
-               nvme_thread = NULL;
-       }
-       list_add(&dev->node, &dev_list);
-       spin_unlock(&dev_list_lock);
-
-       if (start_thread) {
-               nvme_thread = kthread_run(nvme_kthread, NULL, "nvme");
-               wake_up_all(&nvme_kthread_wait);
-       } else
-               wait_event_killable(nvme_kthread_wait, nvme_thread);
-
-       if (IS_ERR_OR_NULL(nvme_thread))
-               return nvme_thread ? PTR_ERR(nvme_thread) : -EINTR;
-
-       return 0;
-}
-
-/*
-* Remove the node from the device list and check
-* for whether or not we need to stop the nvme_thread.
-*/
-static void nvme_dev_list_remove(struct nvme_dev *dev)
-{
-       struct task_struct *tmp = NULL;
-
-       spin_lock(&dev_list_lock);
-       list_del_init(&dev->node);
-       if (list_empty(&dev_list) && !IS_ERR_OR_NULL(nvme_thread)) {
-               tmp = nvme_thread;
-               nvme_thread = NULL;
-       }
-       spin_unlock(&dev_list_lock);
-
-       if (tmp)
-               kthread_stop(tmp);
-}
-
 static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown)
 {
        int i;
        u32 csts = -1;
 
-       nvme_dev_list_remove(dev);
+       del_timer_sync(&dev->watchdog_timer);
 
        mutex_lock(&dev->shutdown_lock);
        if (dev->bar) {
@@ -1927,9 +1869,7 @@ static void nvme_reset_work(struct work_struct *work)
        dev->ctrl.event_limit = NVME_NR_AEN_COMMANDS;
        queue_work(nvme_workq, &dev->async_work);
 
-       result = nvme_dev_list_add(dev);
-       if (result)
-               goto remove;
+       mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ));
 
        /*
         * Keep the controller around but remove all namespaces if we don't have
@@ -1946,8 +1886,6 @@ static void nvme_reset_work(struct work_struct *work)
        clear_bit(NVME_CTRL_RESETTING, &dev->flags);
        return;
 
- remove:
-       nvme_dev_list_remove(dev);
  free_tags:
        nvme_dev_remove_admin(dev);
        blk_put_queue(dev->ctrl.admin_q);
@@ -2055,11 +1993,12 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
        dev->dev = get_device(&pdev->dev);
        pci_set_drvdata(pdev, dev);
 
-       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);
        INIT_WORK(&dev->async_work, nvme_async_event_work);
+       setup_timer(&dev->watchdog_timer, nvme_watchdog_timer,
+               (unsigned long)dev);
        mutex_init(&dev->shutdown_lock);
        init_completion(&dev->ioq_wait);
 
@@ -2108,9 +2047,7 @@ static void nvme_remove(struct pci_dev *pdev)
 {
        struct nvme_dev *dev = pci_get_drvdata(pdev);
 
-       spin_lock(&dev_list_lock);
-       list_del_init(&dev->node);
-       spin_unlock(&dev_list_lock);
+       del_timer_sync(&dev->watchdog_timer);
 
        pci_set_drvdata(pdev, NULL);
        flush_work(&dev->async_work);
@@ -2223,8 +2160,6 @@ static int __init nvme_init(void)
 {
        int result;
 
-       init_waitqueue_head(&nvme_kthread_wait);
-
        nvme_workq = alloc_workqueue("nvme", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
        if (!nvme_workq)
                return -ENOMEM;
@@ -2239,7 +2174,6 @@ static void __exit nvme_exit(void)
 {
        pci_unregister_driver(&nvme_driver);
        destroy_workqueue(nvme_workq);
-       BUG_ON(nvme_thread && !IS_ERR(nvme_thread));
        _nvme_check_size();
 }
 
This page took 0.027453 seconds and 5 git commands to generate.