nvme reset and probe updates V2

PATCH New Christoph Hellwig 4 participants 17 messages Oct 3, 2015 – Oct 9, 2015

Subsystems (2)

Block layer NVMe

lore

Share & tools

lore

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #0 lore raw

Hi Jens, hi Keith,

this serie contains a couple small update and cleanups for the reset and probe path in the nvme driver. I've included Keith previous 3 patches dealing with namespaces attachment as I depend on them and this allows the whole series to be applied directly to Jens' tree.

Changes since V1:
- rebased on top of Jens' for-linux tree
- included Keith patches in the series
- dropped two patches

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #1 lore raw

PATCH 1/8 NVMe: Reference count open namespaces

From: Keith Busch <[email protected]>

Dynamic namespace attachment means the namespace may be removed at any time, so the namespace reference count can not be tied to the device reference count. This fixes a NULL dereference if an opened namespace is detached from a controller.

Signed-off-by Keith Busch <keith.busch at intel.com>

Reviewed-by Christoph Hellwig <hch at lst.de>

---
 drivers/block/nvme-core.c | 29 ++++++++++++++++++++---------
 include/linux/nvme.h      |  1 +
 2 files changed, 21 insertions(+), 9 deletions(-)

drivers/block/nvme-core.c +20 -9

index 6f04771..b02ae3d 100644
@@ -1943,6 +1943,18 @@ static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 #define nvme_compat_ioctl	NULL
 #endif
 
+static void nvme_free_ns(struct kref *kref)
+{
+	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
+
+	spin_lock(&dev_list_lock);
+	ns->disk->private_data = NULL;
+	spin_unlock(&dev_list_lock);
+
+	put_disk(ns->disk);
+	kfree(ns);
+}
+
 static int nvme_open(struct block_device *bdev, fmode_t mode)
 {
 	int ret = 0;

@@ -1952,21 +1964,25 @@ static int nvme_open(struct block_device *bdev, fmode_t mode)
 	ns = bdev->bd_disk->private_data;
 	if (!ns)
 		ret = -ENXIO;
-	else if (!kref_get_unless_zero(&ns->dev->kref))
+	else if (!kref_get_unless_zero(&ns->kref))
 		ret = -ENXIO;
+	else if (!kref_get_unless_zero(&ns->dev->kref)) {
+		kref_put(&ns->kref, nvme_free_ns);
+		ret = -ENXIO;
+	}
 	spin_unlock(&dev_list_lock);
 
 	return ret;
 }
 
 static void nvme_free_dev(struct kref *kref);
-
 static void nvme_release(struct gendisk *disk, fmode_t mode)
 {
 	struct nvme_ns *ns = disk->private_data;
 	struct nvme_dev *dev = ns->dev;
 
 	kref_put(&dev->kref, nvme_free_dev);
+	kref_put(&ns->kref, nvme_free_ns);
 }
 
 static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo)

@@ -2126,6 +2142,7 @@ static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid)
 	if (!disk)
 		goto out_free_queue;
 
+	kref_init(&ns->kref);
 	ns->ns_id = nsid;
 	ns->disk = disk;
 	ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */

@@ -2360,13 +2377,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 static void nvme_free_namespace(struct nvme_ns *ns)
 {
 	list_del(&ns->list);
-
-	spin_lock(&dev_list_lock);
-	ns->disk->private_data = NULL;
-	spin_unlock(&dev_list_lock);
-
-	put_disk(ns->disk);
-	kfree(ns);
+	kref_put(&ns->kref, nvme_free_ns);
 }
 
 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)

include/linux/nvme.h +1 -0

index b5812c3..992b9c1 100644
@@ -135,6 +135,7 @@ struct nvme_ns {
 	struct nvme_dev *dev;
 	struct request_queue *queue;
 	struct gendisk *disk;
+	struct kref kref;
 
 	unsigned ns_id;
 	int lba_shift;

Signature · 1 line

1.9.1

Sagi Grimberg

2015-10-08 15:43 UTC #2 lore raw

PATCH 1/8 NVMe: Reference count open namespaces

Quoted 38 lines · Christoph Hellwig

From: Keith Busch <keith.busch at intel.com>
Dynamic namespace attachment means the namespace may be removed at any time, so the namespace reference count can not be tied to the device reference count. This fixes a NULL dereference if an opened namespace is detached from a controller.
Signed-off-by Keith Busch <keith.busch at intel.com>
Reviewed-by Christoph Hellwig <hch at lst.de>
---
  drivers/block/nvme-core.c | 29 ++++++++++++++++++++---------
  include/linux/nvme.h      |  1 +
  2 files changed, 21 insertions(+), 9 deletions(-)
drivers/block/nvme-core.c +12 -0
index 6f04771..b02ae3d 100644
@@ -1943,6 +1943,18 @@ static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
  #define nvme_compat_ioctl	NULL
  #endif
 
+static void nvme_free_ns(struct kref *kref)
+{
+	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
+
+	spin_lock(&dev_list_lock);
+	ns->disk->private_data = NULL;
+	spin_unlock(&dev_list_lock);
+
+	put_disk(ns->disk);
+	kfree(ns);
+}
+
  static int nvme_open(struct block_device *bdev, fmode_t mode)
  {
  	int ret = 0;

drivers/block/nvme-core.c static int nvme_open(struct block_device *bdev, fmode_t mode) · line 1970

Quoted 8 lines

drivers/block/nvme-core.c +3 -1

@@ -1952,21 +1964,25 @@ static int nvme_open(struct block_device *bdev, fmode_t mode)
  	ns = bdev->bd_disk->private_data;
  	if (!ns)
  		ret = -ENXIO;
-	else if (!kref_get_unless_zero(&ns->dev->kref))
+	else if (!kref_get_unless_zero(&ns->kref))
  		ret = -ENXIO;
+	else if (!kref_get_unless_zero(&ns->dev->kref)) {
+		kref_put(&ns->kref, nvme_free_ns);

Is this calling nvme_free_ns with dev_list_lock taken?

Quoted 3 lines

drivers/block/nvme-core.c +2 -0
+		ret = -ENXIO;
+	}
  	spin_unlock(&dev_list_lock);

Keith Busch

2015-10-08 15:47 UTC #3 lore raw

PATCH 1/8 NVMe: Reference count open namespaces

Quoted 5 lines · Sagi Grimberg

Review
Quoted 2 lines · Christoph Hellwig
(quoted diff) +2 -0
+	else if (!kref_get_unless_zero(&ns->dev->kref)) {
+		kref_put(&ns->kref, nvme_free_ns);
Is this calling nvme_free_ns with dev_list_lock taken?

Yeah, patch 2/8 in this series fixes it. I should have squashed those two.

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #4 lore raw

PATCH 2/8 NVMe: Namespace removal simplifications

In reply to Christoph Hellwig

From: Keith Busch <[email protected]>

This liberates namespace removal from the device, allowing gendisk references to be closed independent of the nvme controller reference count.

Signed-off-by Keith Busch <keith.busch at intel.com>

Reviewed-by Christoph Hellwig <hch at lst.de>

---
 drivers/block/nvme-core.c | 42 ++++++++++--------------------------------
 1 file changed, 10 insertions(+), 32 deletions(-)

drivers/block/nvme-core.c +10 ~~-32~~

index b02ae3d..904b54f 100644
@@ -1943,6 +1943,7 @@ static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 #define nvme_compat_ioctl	NULL
 #endif
 
+static void nvme_free_dev(struct kref *kref);
 static void nvme_free_ns(struct kref *kref)
 {
 	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);

@@ -1951,6 +1952,7 @@ static void nvme_free_ns(struct kref *kref)
 	ns->disk->private_data = NULL;
 	spin_unlock(&dev_list_lock);
 
+	kref_put(&ns->dev->kref, nvme_free_dev);
 	put_disk(ns->disk);
 	kfree(ns);
 }

@@ -1966,22 +1968,14 @@ static int nvme_open(struct block_device *bdev, fmode_t mode)
 		ret = -ENXIO;
 	else if (!kref_get_unless_zero(&ns->kref))
 		ret = -ENXIO;
-	else if (!kref_get_unless_zero(&ns->dev->kref)) {
-		kref_put(&ns->kref, nvme_free_ns);
-		ret = -ENXIO;
-	}
 	spin_unlock(&dev_list_lock);
 
 	return ret;
 }
 
-static void nvme_free_dev(struct kref *kref);
 static void nvme_release(struct gendisk *disk, fmode_t mode)
 {
 	struct nvme_ns *ns = disk->private_data;
-	struct nvme_dev *dev = ns->dev;
-
-	kref_put(&dev->kref, nvme_free_dev);
 	kref_put(&ns->kref, nvme_free_ns);
 }

@@ -2179,6 +2173,7 @@ static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid)
 	if (nvme_revalidate_disk(ns->disk))
 		goto out_free_disk;
 
+	kref_get(&dev->kref);
 	add_disk(ns->disk);
 	if (ns->ms) {
 		struct block_device *bd = bdget_disk(ns->disk, 0);

@@ -2374,12 +2369,6 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	return result;
 }
 
-static void nvme_free_namespace(struct nvme_ns *ns)
-{
-	list_del(&ns->list);
-	kref_put(&ns->kref, nvme_free_ns);
-}
-
 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
 {
 	struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);

@@ -2421,7 +2410,9 @@ static void nvme_ns_remove(struct nvme_ns *ns)
 	if (kill || !blk_queue_dying(ns->queue)) {
 		blk_mq_abort_requeue_list(ns->queue);
 		blk_cleanup_queue(ns->queue);
-        }
+	}
+	list_del_init(&ns->list);
+	kref_put(&ns->kref, nvme_free_ns);
 }
 
 static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn)

@@ -2432,18 +2423,14 @@ static void nvme_scan_namespaces(struct nvme_dev *dev, unsigned nn)
 	for (i = 1; i <= nn; i++) {
 		ns = nvme_find_ns(dev, i);
 		if (ns) {
-			if (revalidate_disk(ns->disk)) {
+			if (revalidate_disk(ns->disk))
 				nvme_ns_remove(ns);
-				nvme_free_namespace(ns);
-			}
 		} else
 			nvme_alloc_ns(dev, i);
 	}
 	list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
-		if (ns->ns_id > nn) {
+		if (ns->ns_id > nn)
 			nvme_ns_remove(ns);
-			nvme_free_namespace(ns);
-		}
 	}
 	list_sort(NULL, &dev->namespaces, ns_cmp);
 }

@@ -2833,9 +2820,9 @@ static void nvme_dev_shutdown(struct nvme_dev *dev)
 
 static void nvme_dev_remove(struct nvme_dev *dev)
 {
-	struct nvme_ns *ns;
+	struct nvme_ns *ns, *next;
 
-	list_for_each_entry(ns, &dev->namespaces, list)
+	list_for_each_entry_safe(ns, next, &dev->namespaces, list)
 		nvme_ns_remove(ns);
 }

@@ -2891,21 +2878,12 @@ static void nvme_release_instance(struct nvme_dev *dev)
 	spin_unlock(&dev_list_lock);
 }
 
-static void nvme_free_namespaces(struct nvme_dev *dev)
-{
-	struct nvme_ns *ns, *next;
-
-	list_for_each_entry_safe(ns, next, &dev->namespaces, list)
-		nvme_free_namespace(ns);
-}
-
 static void nvme_free_dev(struct kref *kref)
 {
 	struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
 
 	put_device(dev->dev);
 	put_device(dev->device);
-	nvme_free_namespaces(dev);
 	nvme_release_instance(dev);
 	if (dev->tagset.tags)
 		blk_mq_free_tag_set(&dev->tagset);

Signature · 1 line

1.9.1

Sagi Grimberg

2015-10-08 15:47 UTC #5 lore raw

PATCH 2/8 NVMe: Namespace removal simplifications

Quoted 33 lines · Christoph Hellwig

From: Keith Busch <keith.busch at intel.com>

This liberates namespace removal from the device, allowing gendisk references to be closed independent of the nvme controller reference count.

Signed-off-by Keith Busch <keith.busch at intel.com>

Reviewed-by Christoph Hellwig <hch at lst.de>

---
  drivers/block/nvme-core.c | 42 ++++++++++--------------------------------
  1 file changed, 10 insertions(+), 32 deletions(-)

drivers/block/nvme-core.c +2 -0

index b02ae3d..904b54f 100644
@@ -1943,6 +1943,7 @@ static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
  #define nvme_compat_ioctl	NULL
  #endif
 
+static void nvme_free_dev(struct kref *kref);
  static void nvme_free_ns(struct kref *kref)
  {
  	struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);

@@ -1951,6 +1952,7 @@ static void nvme_free_ns(struct kref *kref)
  	ns->disk->private_data = NULL;
  	spin_unlock(&dev_list_lock);
 
+	kref_put(&ns->dev->kref, nvme_free_dev);
  	put_disk(ns->disk);
  	kfree(ns);
  }

drivers/block/nvme-core.c static int nvme_open(struct block_device *bdev, fmode_t mode) · line 1972

Quoted 7 lines

drivers/block/nvme-core.c +0 -4

@@ -1966,22 +1968,14 @@ static int nvme_open(struct block_device *bdev, fmode_t mode)
  		ret = -ENXIO;
  	else if (!kref_get_unless_zero(&ns->kref))
  		ret = -ENXIO;
-	else if (!kref_get_unless_zero(&ns->dev->kref)) {
-		kref_put(&ns->kref, nvme_free_ns);
-		ret = -ENXIO;
-	}

And this fixes the last patch. I would suggest fixing the former though for bisect-ability.

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #6 lore raw

PATCH 3/8 NVMe: Simplify device resume on io queue failure

In reply to Christoph Hellwig

From: Keith Busch <[email protected]>

Releasing IO queues and disks was done in a work queue outside the controller resume context to delete namespaces if the controller failed after a resume from suspend. This is unnecessary since we can resume a device asynchronously.

This patch makes resume use probe_work so it can directly remove namespaces if the device is manageable but not IO capable. Since the deleting disks was the only reason we had the convoluted "reset_workfn", this patch removes that unnecessary indirection.

Signed-off-by Keith Busch <keith.busch at intel.com>

Reviewed-by Christoph Hellwig <hch at lst.de>

---
 drivers/block/nvme-core.c | 34 ++++++----------------------------
 include/linux/nvme.h      |  1 -
 2 files changed, 6 insertions(+), 29 deletions(-)

drivers/block/nvme-core.c +6 ~~-28~~

index 904b54f..bf35846 100644
@@ -1285,7 +1285,6 @@ static void nvme_abort_req(struct request *req)
 		list_del_init(&dev->node);
 		dev_warn(dev->dev, "I/O %d QID %d timeout, reset controller\n",
 							req->tag, nvmeq->qid);
-		dev->reset_workfn = nvme_reset_failed_dev;
 		queue_work(nvme_workq, &dev->reset_work);
  out:
 		spin_unlock_irqrestore(&dev_list_lock, flags);

@@ -2089,7 +2088,6 @@ static int nvme_kthread(void *data)
 				dev_warn(dev->dev,
 					"Failed status: %x, reset controller\n",
 					readl(&dev->bar->csts));
-				dev->reset_workfn = nvme_reset_failed_dev;
 				queue_work(nvme_workq, &dev->reset_work);
 				continue;
 			}

@@ -3025,14 +3023,6 @@ static int nvme_remove_dead_ctrl(void *arg)
 	return 0;
 }
 
-static void nvme_remove_disks(struct work_struct *ws)
-{
-	struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
-
-	nvme_free_queues(dev, 1);
-	nvme_dev_remove(dev);
-}
-
 static int nvme_dev_resume(struct nvme_dev *dev)
 {
 	int ret;

@@ -3041,10 +3031,9 @@ static int nvme_dev_resume(struct nvme_dev *dev)
 	if (ret)
 		return ret;
 	if (dev->online_queues < 2) {
-		spin_lock(&dev_list_lock);
-		dev->reset_workfn = nvme_remove_disks;
-		queue_work(nvme_workq, &dev->reset_work);
-		spin_unlock(&dev_list_lock);
+		dev_warn(dev->dev, "IO queues not created\n");
+		nvme_free_queues(dev, 1);
+		nvme_dev_remove(dev);
 	} else {
 		nvme_unfreeze_queues(dev);
 		nvme_dev_add(dev);

@@ -3091,12 +3080,6 @@ static void nvme_reset_failed_dev(struct work_struct *ws)
 	nvme_dev_reset(dev);
 }
 
-static void nvme_reset_workfn(struct work_struct *work)
-{
-	struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work);
-	dev->reset_workfn(work);
-}
-
 static int nvme_reset(struct nvme_dev *dev)
 {
 	int ret = -EBUSY;

@@ -3106,7 +3089,6 @@ static int nvme_reset(struct nvme_dev *dev)
 
 	spin_lock(&dev_list_lock);
 	if (!work_pending(&dev->reset_work)) {
-		dev->reset_workfn = nvme_reset_failed_dev;
 		queue_work(nvme_workq, &dev->reset_work);
 		ret = 0;
 	}

@@ -3159,8 +3141,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		goto free;
 
 	INIT_LIST_HEAD(&dev->namespaces);
-	dev->reset_workfn = nvme_reset_failed_dev;
-	INIT_WORK(&dev->reset_work, nvme_reset_workfn);
+	INIT_WORK(&dev->reset_work, nvme_reset_failed_dev);
 	dev->dev = get_device(&pdev->dev);
 	pci_set_drvdata(pdev, dev);
 	result = nvme_set_instance(dev);

@@ -3223,7 +3204,7 @@ static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
 	if (prepare)
 		nvme_dev_shutdown(dev);
 	else
-		nvme_dev_resume(dev);
+		schedule_work(&dev->probe_work);
 }
 
 static void nvme_shutdown(struct pci_dev *pdev)

@@ -3277,10 +3258,7 @@ static int nvme_resume(struct device *dev)
 	struct pci_dev *pdev = to_pci_dev(dev);
 	struct nvme_dev *ndev = pci_get_drvdata(pdev);
 
-	if (nvme_dev_resume(ndev) && !work_busy(&ndev->reset_work)) {
-		ndev->reset_workfn = nvme_reset_failed_dev;
-		queue_work(nvme_workq, &ndev->reset_work);
-	}
+	schedule_work(&ndev->probe_work);
 	return 0;
 }
 #endif

include/linux/nvme.h +0 -1

index 992b9c1..7725b4c 100644
@@ -104,7 +104,6 @@ struct nvme_dev {
 	struct list_head namespaces;
 	struct kref kref;
 	struct device *device;
-	work_func_t reset_workfn;
 	struct work_struct reset_work;
 	struct work_struct probe_work;
 	struct work_struct scan_work;

Signature · 1 line

1.9.1

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #7 lore raw

PATCH 4/8 nvme: delete dev from dev_list in nvme_reset

In reply to Christoph Hellwig

Device resets need to delete the device from the device list before kicking of the reset an re-probe, otherwise we get the device added to the list twice. nvme_reset is the only side missing this deletion at the moment, and this patch adds it.

Signed-off-by Christoph Hellwig <hch at lst.de>

Reviewed-by Keith Busch <keith.busch at intel.com>

---
 drivers/block/nvme-core.c | 1 +
 1 file changed, 1 insertion(+)

drivers/block/nvme-core.c +1 -0

index bf35846..be35b1d 100644
@@ -3089,6 +3089,7 @@ static int nvme_reset(struct nvme_dev *dev)
 
 	spin_lock(&dev_list_lock);
 	if (!work_pending(&dev->reset_work)) {
+		list_del_init(&dev->node);
 		queue_work(nvme_workq, &dev->reset_work);
 		ret = 0;
 	}

Signature · 1 line

1.9.1

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #8 lore raw

PATCH 5/8 nvme: merge nvme_dev_reset into nvme_reset_failed_dev

In reply to Christoph Hellwig

And give the resulting function a more descriptive name.

Signed-off-by Christoph Hellwig <hch at lst.de>

Reviewed-by Keith Busch <keith.busch at intel.com>

---
 drivers/block/nvme-core.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)

drivers/block/nvme-core.c +3 -9

index be35b1d..509ad4b 100644
@@ -84,7 +84,6 @@ static wait_queue_head_t nvme_kthread_wait;
 
 static struct class *nvme_class;
 
-static void nvme_reset_failed_dev(struct work_struct *ws);
 static int nvme_reset(struct nvme_dev *dev);
 static int nvme_process_cq(struct nvme_queue *nvmeq);

@@ -3053,8 +3052,9 @@ static void nvme_dead_ctrl(struct nvme_dev *dev)
 	}
 }
 
-static void nvme_dev_reset(struct nvme_dev *dev)
+static void nvme_reset_work(struct work_struct *ws)
 {
+	struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
 	bool in_probe = work_busy(&dev->probe_work);
 
 	nvme_dev_shutdown(dev);

@@ -3074,12 +3074,6 @@ static void nvme_dev_reset(struct nvme_dev *dev)
 	schedule_work(&dev->probe_work);
 }
 
-static void nvme_reset_failed_dev(struct work_struct *ws)
-{
-	struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
-	nvme_dev_reset(dev);
-}
-
 static int nvme_reset(struct nvme_dev *dev)
 {
 	int ret = -EBUSY;

@@ -3142,7 +3136,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		goto free;
 
 	INIT_LIST_HEAD(&dev->namespaces);
-	INIT_WORK(&dev->reset_work, nvme_reset_failed_dev);
+	INIT_WORK(&dev->reset_work, nvme_reset_work);
 	dev->dev = get_device(&pdev->dev);
 	pci_set_drvdata(pdev, dev);
 	result = nvme_set_instance(dev);

Signature · 1 line

1.9.1

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #9 lore raw

PATCH 6/8 nvme: factor reset code into a common helper

In reply to Christoph Hellwig

Signed-off-by Christoph Hellwig <hch at lst.de>

Reviewed-by Keith Busch <keith.busch at intel.com>

---
 drivers/block/nvme-core.c | 48 +++++++++++++++++++++++------------------------
 1 file changed, 24 insertions(+), 24 deletions(-)

drivers/block/nvme-core.c +24 ~~-24~~

index 509ad4b..e03a95b 100644
@@ -84,6 +84,7 @@ static wait_queue_head_t nvme_kthread_wait;
 
 static struct class *nvme_class;
 
+static int __nvme_reset(struct nvme_dev *dev);
 static int nvme_reset(struct nvme_dev *dev);
 static int nvme_process_cq(struct nvme_queue *nvmeq);

@@ -1276,17 +1277,13 @@ static void nvme_abort_req(struct request *req)
 	struct nvme_command cmd;
 
 	if (!nvmeq->qid || cmd_rq->aborted) {
-		unsigned long flags;
-
-		spin_lock_irqsave(&dev_list_lock, flags);
-		if (work_busy(&dev->reset_work))
-			goto out;
-		list_del_init(&dev->node);
-		dev_warn(dev->dev, "I/O %d QID %d timeout, reset controller\n",
-							req->tag, nvmeq->qid);
-		queue_work(nvme_workq, &dev->reset_work);
- out:
-		spin_unlock_irqrestore(&dev_list_lock, flags);
+		spin_lock(&dev_list_lock);
+		if (!__nvme_reset(dev)) {
+			dev_warn(dev->dev,
+				 "I/O %d QID %d timeout, reset controller\n",
+				 req->tag, nvmeq->qid);
+		}
+		spin_unlock(&dev_list_lock);
 		return;
 	}

@@ -2081,13 +2078,11 @@ static int nvme_kthread(void *data)
 
 			if ((dev->subsystem && (csts & NVME_CSTS_NSSRO)) ||
 							csts & NVME_CSTS_CFS) {
-				if (work_busy(&dev->reset_work))
-					continue;
-				list_del_init(&dev->node);
-				dev_warn(dev->dev,
-					"Failed status: %x, reset controller\n",
-					readl(&dev->bar->csts));
-				queue_work(nvme_workq, &dev->reset_work);
+				if (!__nvme_reset(dev)) {
+					dev_warn(dev->dev,
+						"Failed status: %x, reset controller\n",
+						readl(&dev->bar->csts));
+				}
 				continue;
 			}
 			for (i = 0; i < dev->queue_count; i++) {

@@ -3074,19 +3069,24 @@ static void nvme_reset_work(struct work_struct *ws)
 	schedule_work(&dev->probe_work);
 }
 
+static int __nvme_reset(struct nvme_dev *dev)
+{
+	if (work_pending(&dev->reset_work))
+		return -EBUSY;
+	list_del_init(&dev->node);
+	queue_work(nvme_workq, &dev->reset_work);
+	return 0;
+}
+
 static int nvme_reset(struct nvme_dev *dev)
 {
-	int ret = -EBUSY;
+	int ret;
 
 	if (!dev->admin_q || blk_queue_dying(dev->admin_q))
 		return -ENODEV;
 
 	spin_lock(&dev_list_lock);
-	if (!work_pending(&dev->reset_work)) {
-		list_del_init(&dev->node);
-		queue_work(nvme_workq, &dev->reset_work);
-		ret = 0;
-	}
+	ret = __nvme_reset(dev);
 	spin_unlock(&dev_list_lock);
 
 	if (!ret) {

Signature · 1 line

1.9.1

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #10 lore raw

PATCH 7/8 nvme: merge nvme_dev_start, nvme_dev_resume and nvme_async_probe

In reply to Christoph Hellwig

And give the resulting function a sensible name. This keeps all the error handling in a single place and will allow for further improvements to it.

Signed-off-by Christoph Hellwig <hch at lst.de>

Reviewed-by Keith Busch <keith.busch at intel.com>

---
 drivers/block/nvme-core.c | 53 ++++++++++++++++++-----------------------------
 1 file changed, 20 insertions(+), 33 deletions(-)

drivers/block/nvme-core.c +20 ~~-33~~

index e03a95b..61cfff3 100644
@@ -87,6 +87,7 @@ static struct class *nvme_class;
 static int __nvme_reset(struct nvme_dev *dev);
 static int nvme_reset(struct nvme_dev *dev);
 static int nvme_process_cq(struct nvme_queue *nvmeq);
+static void nvme_dead_ctrl(struct nvme_dev *dev);
 
 struct async_cmd_info {
 	struct kthread_work work;

@@ -2949,14 +2950,15 @@ static const struct file_operations nvme_dev_fops = {
 	.compat_ioctl	= nvme_dev_ioctl,
 };
 
-static int nvme_dev_start(struct nvme_dev *dev)
+static void nvme_probe_work(struct work_struct *work)
 {
-	int result;
+	struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work);
 	bool start_thread = false;
+	int result;
 
 	result = nvme_dev_map(dev);
 	if (result)
-		return result;
+		goto out;
 
 	result = nvme_configure_admin_queue(dev);
 	if (result)

@@ -2991,7 +2993,17 @@ static int nvme_dev_start(struct nvme_dev *dev)
 		goto free_tags;
 
 	dev->event_limit = 1;
-	return result;
+
+	if (dev->online_queues < 2) {
+		dev_warn(dev->dev, "IO queues not created\n");
+		nvme_free_queues(dev, 1);
+		nvme_dev_remove(dev);
+	} else {
+		nvme_unfreeze_queues(dev);
+		nvme_dev_add(dev);
+	}
+
+	return;
 
  free_tags:
 	nvme_dev_remove_admin(dev);

@@ -3003,7 +3015,9 @@ static int nvme_dev_start(struct nvme_dev *dev)
 	nvme_dev_list_remove(dev);
  unmap:
 	nvme_dev_unmap(dev);
-	return result;
+ out:
+	if (!work_busy(&dev->reset_work))
+		nvme_dead_ctrl(dev);
 }
 
 static int nvme_remove_dead_ctrl(void *arg)

@@ -3017,24 +3031,6 @@ static int nvme_remove_dead_ctrl(void *arg)
 	return 0;
 }
 
-static int nvme_dev_resume(struct nvme_dev *dev)
-{
-	int ret;
-
-	ret = nvme_dev_start(dev);
-	if (ret)
-		return ret;
-	if (dev->online_queues < 2) {
-		dev_warn(dev->dev, "IO queues not created\n");
-		nvme_free_queues(dev, 1);
-		nvme_dev_remove(dev);
-	} else {
-		nvme_unfreeze_queues(dev);
-		nvme_dev_add(dev);
-	}
-	return 0;
-}
-
 static void nvme_dead_ctrl(struct nvme_dev *dev)
 {
 	dev_warn(dev->dev, "Device failed to resume\n");

@@ -3113,7 +3109,6 @@ static ssize_t nvme_sysfs_reset(struct device *dev,
 }
 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
 
-static void nvme_async_probe(struct work_struct *work);
 static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	int node, result = -ENOMEM;

@@ -3164,7 +3159,7 @@ 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->probe_work, nvme_async_probe);
+	INIT_WORK(&dev->probe_work, nvme_probe_work);
 	schedule_work(&dev->probe_work);
 	return 0;

@@ -3184,14 +3179,6 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	return result;
 }
 
-static void nvme_async_probe(struct work_struct *work)
-{
-	struct nvme_dev *dev = container_of(work, struct nvme_dev, probe_work);
-
-	if (nvme_dev_resume(dev) && !work_busy(&dev->reset_work))
-		nvme_dead_ctrl(dev);
-}
-
 static void nvme_reset_notify(struct pci_dev *pdev, bool prepare)
 {
 	struct nvme_dev *dev = pci_get_drvdata(pdev);

Signature · 1 line

1.9.1

Christoph Hellwig

author maintainer

2015-10-03 13:39 UTC #11 lore raw

PATCH 8/8 nvme: properly handle partially initialized queues in nvme_create_io_queues

In reply to Christoph Hellwig

This avoids having to clean up later in a seemingly unrelated place.

Signed-off-by Christoph Hellwig <hch at lst.de>

Reviewed-by Christoph Hellwig <hch at lst.de>

---
 drivers/block/nvme-core.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

drivers/block/nvme-core.c +14 -2

index 61cfff3..01a6d1b 100644
@@ -2189,6 +2189,13 @@ static void nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid)
 	kfree(ns);
 }
 
+/*
+ * Create I/O queues.  Failing to create an I/O queue is not an issue,
+ * we can continue with less than the desired amount of queues, and
+ * even a controller without I/O queues an still be used to issue
+ * admin commands.  This might be useful to upgrade a buggy firmware
+ * for example.
+ */
 static void nvme_create_io_queues(struct nvme_dev *dev)
 {
 	unsigned i;

@@ -2198,8 +2205,10 @@ static void nvme_create_io_queues(struct nvme_dev *dev)
 			break;
 
 	for (i = dev->online_queues; i <= dev->queue_count - 1; i++)
-		if (nvme_create_queue(dev->queues[i], i))
+		if (nvme_create_queue(dev->queues[i], i)) {
+			nvme_free_queues(dev, i);
 			break;
+		}
 }
 
 static int set_queue_count(struct nvme_dev *dev, int count)

@@ -2994,9 +3003,12 @@ static void nvme_probe_work(struct work_struct *work)
 
 	dev->event_limit = 1;
 
+	/*
+	 * Keep the controller around but remove all namespaces if we don't have
+	 * any working I/O queue.
+	 */
 	if (dev->online_queues < 2) {
 		dev_warn(dev->dev, "IO queues not created\n");
-		nvme_free_queues(dev, 1);
 		nvme_dev_remove(dev);
 	} else {
 		nvme_unfreeze_queues(dev);

Signature · 1 line

1.9.1

Sagi Grimberg

2015-10-08 15:52 UTC #12 lore raw

PATCH 8/8 nvme: properly handle partially initialized queues in nvme_create_io_queues

Quoted 4 lines · Christoph Hellwig

This avoids having to clean up later in a seemingly unrelated place.
Signed-off-by Christoph Hellwig <hch at lst.de>
Reviewed-by Christoph Hellwig <hch at lst.de>

I assume Keith reviewed this one correct? :)

Christoph Hellwig

author maintainer

2015-10-09 06:15 UTC #13 lore raw

PATCH 8/8 nvme: properly handle partially initialized queues in nvme_create_io_queues

Quoted 7 lines · Sagi Grimberg

Quoted 4 lines · Christoph Hellwig
This avoids having to clean up later in a seemingly unrelated place.
Signed-off-by Christoph Hellwig <hch at lst.de>
Reviewed-by Christoph Hellwig <hch at lst.de>
I assume Keith reviewed this one correct? :)

He did reply with a reviewed-by tag to the previous posting of the series, I think that qualifies.

Keith Busch

2015-10-09 14:20 UTC #14 lore raw

PATCH 8/8 nvme: properly handle partially initialized queues in nvme_create_io_queues

Quoted 11 lines · Christoph Hellwig

Quoted 7 lines · Sagi Grimberg
Quoted 4 lines · Christoph Hellwig
This avoids having to clean up later in a seemingly unrelated place.
Signed-off-by Christoph Hellwig <hch at lst.de>
Reviewed-by Christoph Hellwig <hch at lst.de>
I assume Keith reviewed this one correct? :)
He did reply with a reviewed-by tag to the previous posting of the series, I think that qualifies.

I think Sagi was just making a light of using the wrong name (your's) in the "Reviewed-by". :)

Christoph Hellwig

author maintainer

2015-10-09 14:39 UTC #15 lore raw

PATCH 8/8 nvme: properly handle partially initialized queues in nvme_create_io_queues

Quoted 5 lines · Keith Busch

Quoted 2 lines
He did reply with a reviewed-by tag to the previous posting of the series, I think that qualifies.
I think Sagi was just making a light of using the wrong name (your's) in the "Reviewed-by". :)

Oh, ok. Jens, do you want a resend, or can you fix it up on the fly?

Jens Axboe

2015-10-09 16:16 UTC #16 lore raw

PATCH 8/8 nvme: properly handle partially initialized queues in nvme_create_io_queues

Quoted 8 lines · Christoph Hellwig

Quoted 5 lines · Keith Busch
Quoted 2 lines
He did reply with a reviewed-by tag to the previous posting of the series, I think that qualifies.
I think Sagi was just making a light of using the wrong name (your's) in the "Reviewed-by". :)
Oh, ok. Jens, do you want a resend, or can you fix it up on the fly?

I fixed it up.

Signature · 1 line

Jens Axboe