From b80d5ccca3a012e91ca64a2a0b13049163a6a698 Mon Sep 17 00:00:00 2001
From: Haiyan Hu
Date: Tue, 10 Sep 2013 11:25:37 +0800
Subject: NVMe: Avoid shift operation when writing cq head doorbell

Changes the type of dev->db_stride to unsigned and changes the value
stored there to be 1 << the current value. Then there is less
calculation to be done at completion time.

Signed-off-by: Haiyan Hu <huhaiyan@huawei.com>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 26d03fa0bf26..073aec913c78 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -775,7 +775,7 @@ static int nvme_process_cq(struct nvme_queue *nvmeq)
 	if (head == nvmeq->cq_head && phase == nvmeq->cq_phase)
 		return 0;
 
-	writel(head, nvmeq->q_db + (1 << nvmeq->dev->db_stride));
+	writel(head, nvmeq->q_db + nvmeq->dev->db_stride);
 	nvmeq->cq_head = head;
 	nvmeq->cq_phase = phase;
 
@@ -1113,7 +1113,7 @@ static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
 	init_waitqueue_head(&nvmeq->sq_full);
 	init_waitqueue_entry(&nvmeq->sq_cong_wait, nvme_thread);
 	bio_list_init(&nvmeq->sq_cong);
-	nvmeq->q_db = &dev->dbs[qid << (dev->db_stride + 1)];
+	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
 	nvmeq->q_depth = depth;
 	nvmeq->cq_vector = vector;
 	nvmeq->q_suspended = 1;
@@ -1149,7 +1149,7 @@ static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
 	nvmeq->sq_tail = 0;
 	nvmeq->cq_head = 0;
 	nvmeq->cq_phase = 1;
-	nvmeq->q_db = &dev->dbs[qid << (dev->db_stride + 1)];
+	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
 	memset(nvmeq->cmdid_data, 0, extra);
 	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
 	nvme_cancel_ios(nvmeq, false);
@@ -1741,7 +1741,7 @@ static int set_queue_count(struct nvme_dev *dev, int count)
 
 static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
 {
-	return 4096 + ((nr_io_queues + 1) << (dev->db_stride + 3));
+	return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
 }
 
 static int nvme_setup_io_queues(struct nvme_dev *dev)
@@ -1958,7 +1958,7 @@ static int nvme_dev_map(struct nvme_dev *dev)
 	if (!dev->bar)
 		goto disable;
 
-	dev->db_stride = NVME_CAP_STRIDE(readq(&dev->bar->cap));
+	dev->db_stride = 1 << NVME_CAP_STRIDE(readq(&dev->bar->cap));
 	dev->dbs = ((void __iomem *)dev->bar) + 4096;
 
 	return 0;
-- 
cgit v1.2.3


From 481e5bad84239b01415b888df2040c1860ae3cfd Mon Sep 17 00:00:00 2001
From: Michael Opdenacker
Date: Sat, 12 Oct 2013 06:23:29 +0200
Subject: NVMe: remove deprecated IRQF_DISABLED

This patch proposes to remove the use of the IRQF_DISABLED flag

It's a NOOP since 2.6.35 and it will be removed one day.

Signed-off-by: Michael Opdenacker <michael.opdenacker@free-electrons.com>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 073aec913c78..df3daeb6227f 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1134,11 +1134,10 @@ static int queue_request_irq(struct nvme_dev *dev, struct nvme_queue *nvmeq,
 {
 	if (use_threaded_interrupts)
 		return request_threaded_irq(dev->entry[nvmeq->cq_vector].vector,
-					nvme_irq_check, nvme_irq,
-					IRQF_DISABLED | IRQF_SHARED,
+					nvme_irq_check, nvme_irq, IRQF_SHARED,
 					name, nvmeq);
 	return request_irq(dev->entry[nvmeq->cq_vector].vector, nvme_irq,
-				IRQF_DISABLED | IRQF_SHARED, name, nvmeq);
+				IRQF_SHARED, name, nvmeq);
 }
 
 static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
-- 
cgit v1.2.3


From 320a382746e0ab1304476ea7e986a8d416ab99db Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Wed, 23 Oct 2013 13:07:34 -0600
Subject: NVMe: compat SG_IO ioctl

For 32-bit versions of sg3-utils running on a 64-bit system. This is
mostly a copy from the relevent portions of fs/compat_ioctl.c, with
slight modifications for going through block_device_operations.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Vishal Verma <vishal.l.verma@linux.intel.com>
[fixed up CONFIG_COMPAT=n build problems]
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c |  18 +++++-
 drivers/block/nvme-scsi.c | 147 ++++++++++++++++++++++++++++++++++++++++++++++
 include/linux/nvme.h      |   1 +
 3 files changed, 165 insertions(+), 1 deletion(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index df3daeb6227f..e44350de8a21 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1568,10 +1568,26 @@ static int nvme_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd,
 	}
 }
 
+#ifdef CONFIG_COMPAT
+static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
+					unsigned int cmd, unsigned long arg)
+{
+	struct nvme_ns *ns = bdev->bd_disk->private_data;
+
+	switch (cmd) {
+	case SG_IO:
+		return nvme_sg_io32(ns, arg);
+	}
+	return nvme_ioctl(bdev, mode, cmd, arg);
+}
+#else
+#define nvme_compat_ioctl	NULL
+#endif
+
 static const struct block_device_operations nvme_fops = {
 	.owner		= THIS_MODULE,
 	.ioctl		= nvme_ioctl,
-	.compat_ioctl	= nvme_ioctl,
+	.compat_ioctl	= nvme_compat_ioctl,
 };
 
 static void nvme_resubmit_bios(struct nvme_queue *nvmeq)
diff --git a/drivers/block/nvme-scsi.c b/drivers/block/nvme-scsi.c
index 4a4ff4eb8e23..4a0ceb64e269 100644
--- a/drivers/block/nvme-scsi.c
+++ b/drivers/block/nvme-scsi.c
@@ -25,6 +25,7 @@
 #include <linux/bio.h>
 #include <linux/bitops.h>
 #include <linux/blkdev.h>
+#include <linux/compat.h>
 #include <linux/delay.h>
 #include <linux/errno.h>
 #include <linux/fs.h>
@@ -3038,6 +3039,152 @@ int nvme_sg_io(struct nvme_ns *ns, struct sg_io_hdr __user *u_hdr)
 	return retcode;
 }
 
+#ifdef CONFIG_COMPAT
+typedef struct sg_io_hdr32 {
+	compat_int_t interface_id;	/* [i] 'S' for SCSI generic (required) */
+	compat_int_t dxfer_direction;	/* [i] data transfer direction  */
+	unsigned char cmd_len;		/* [i] SCSI command length ( <= 16 bytes) */
+	unsigned char mx_sb_len;		/* [i] max length to write to sbp */
+	unsigned short iovec_count;	/* [i] 0 implies no scatter gather */
+	compat_uint_t dxfer_len;		/* [i] byte count of data transfer */
+	compat_uint_t dxferp;		/* [i], [*io] points to data transfer memory
+					      or scatter gather list */
+	compat_uptr_t cmdp;		/* [i], [*i] points to command to perform */
+	compat_uptr_t sbp;		/* [i], [*o] points to sense_buffer memory */
+	compat_uint_t timeout;		/* [i] MAX_UINT->no timeout (unit: millisec) */
+	compat_uint_t flags;		/* [i] 0 -> default, see SG_FLAG... */
+	compat_int_t pack_id;		/* [i->o] unused internally (normally) */
+	compat_uptr_t usr_ptr;		/* [i->o] unused internally */
+	unsigned char status;		/* [o] scsi status */
+	unsigned char masked_status;	/* [o] shifted, masked scsi status */
+	unsigned char msg_status;		/* [o] messaging level data (optional) */
+	unsigned char sb_len_wr;		/* [o] byte count actually written to sbp */
+	unsigned short host_status;	/* [o] errors from host adapter */
+	unsigned short driver_status;	/* [o] errors from software driver */
+	compat_int_t resid;		/* [o] dxfer_len - actual_transferred */
+	compat_uint_t duration;		/* [o] time taken by cmd (unit: millisec) */
+	compat_uint_t info;		/* [o] auxiliary information */
+} sg_io_hdr32_t;  /* 64 bytes long (on sparc32) */
+
+typedef struct sg_iovec32 {
+	compat_uint_t iov_base;
+	compat_uint_t iov_len;
+} sg_iovec32_t;
+
+static int sg_build_iovec(sg_io_hdr_t __user *sgio, void __user *dxferp, u16 iovec_count)
+{
+	sg_iovec_t __user *iov = (sg_iovec_t __user *) (sgio + 1);
+	sg_iovec32_t __user *iov32 = dxferp;
+	int i;
+
+	for (i = 0; i < iovec_count; i++) {
+		u32 base, len;
+
+		if (get_user(base, &iov32[i].iov_base) ||
+		    get_user(len, &iov32[i].iov_len) ||
+		    put_user(compat_ptr(base), &iov[i].iov_base) ||
+		    put_user(len, &iov[i].iov_len))
+			return -EFAULT;
+	}
+
+	if (put_user(iov, &sgio->dxferp))
+		return -EFAULT;
+	return 0;
+}
+
+int nvme_sg_io32(struct nvme_ns *ns, unsigned long arg)
+{
+	sg_io_hdr32_t __user *sgio32 = (sg_io_hdr32_t __user *)arg;
+	sg_io_hdr_t __user *sgio;
+	u16 iovec_count;
+	u32 data;
+	void __user *dxferp;
+	int err;
+	int interface_id;
+
+	if (get_user(interface_id, &sgio32->interface_id))
+		return -EFAULT;
+	if (interface_id != 'S')
+		return -EINVAL;
+
+	if (get_user(iovec_count, &sgio32->iovec_count))
+		return -EFAULT;
+
+	{
+		void __user *top = compat_alloc_user_space(0);
+		void __user *new = compat_alloc_user_space(sizeof(sg_io_hdr_t) +
+				       (iovec_count * sizeof(sg_iovec_t)));
+		if (new > top)
+			return -EINVAL;
+
+		sgio = new;
+	}
+
+	/* Ok, now construct.  */
+	if (copy_in_user(&sgio->interface_id, &sgio32->interface_id,
+			 (2 * sizeof(int)) +
+			 (2 * sizeof(unsigned char)) +
+			 (1 * sizeof(unsigned short)) +
+			 (1 * sizeof(unsigned int))))
+		return -EFAULT;
+
+	if (get_user(data, &sgio32->dxferp))
+		return -EFAULT;
+	dxferp = compat_ptr(data);
+	if (iovec_count) {
+		if (sg_build_iovec(sgio, dxferp, iovec_count))
+			return -EFAULT;
+	} else {
+		if (put_user(dxferp, &sgio->dxferp))
+			return -EFAULT;
+	}
+
+	{
+		unsigned char __user *cmdp;
+		unsigned char __user *sbp;
+
+		if (get_user(data, &sgio32->cmdp))
+			return -EFAULT;
+		cmdp = compat_ptr(data);
+
+		if (get_user(data, &sgio32->sbp))
+			return -EFAULT;
+		sbp = compat_ptr(data);
+
+		if (put_user(cmdp, &sgio->cmdp) ||
+		    put_user(sbp, &sgio->sbp))
+			return -EFAULT;
+	}
+
+	if (copy_in_user(&sgio->timeout, &sgio32->timeout,
+			 3 * sizeof(int)))
+		return -EFAULT;
+
+	if (get_user(data, &sgio32->usr_ptr))
+		return -EFAULT;
+	if (put_user(compat_ptr(data), &sgio->usr_ptr))
+		return -EFAULT;
+
+	err = nvme_sg_io(ns, sgio);
+	if (err >= 0) {
+		void __user *datap;
+
+		if (copy_in_user(&sgio32->pack_id, &sgio->pack_id,
+				 sizeof(int)) ||
+		    get_user(datap, &sgio->usr_ptr) ||
+		    put_user((u32)(unsigned long)datap,
+			     &sgio32->usr_ptr) ||
+		    copy_in_user(&sgio32->status, &sgio->status,
+				 (4 * sizeof(unsigned char)) +
+				 (2 * sizeof(unsigned short)) +
+				 (3 * sizeof(int))))
+			err = -EFAULT;
+	}
+
+	return err;
+}
+#endif
+
 int nvme_sg_get_version_num(int __user *ip)
 {
 	return put_user(sg_version_num, ip);
diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index 8119a476cb29..5bc29197d90e 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -165,6 +165,7 @@ int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11,
 struct sg_io_hdr;
 
 int nvme_sg_io(struct nvme_ns *ns, struct sg_io_hdr __user *u_hdr);
+int nvme_sg_io32(struct nvme_ns *ns, unsigned long arg);
 int nvme_sg_get_version_num(int __user *ip);
 
 #endif /* _LINUX_NVME_H */
-- 
cgit v1.2.3


From 0a8d44cb33377969337fa6c1961b631605d5f453 Mon Sep 17 00:00:00 2001
From: Matthew Wilcox
Date: Tue, 15 Oct 2013 15:01:10 -0400
Subject: NVMe: Fix lockdep warnings

During the initialisation path, the queue lock is taken without interrupt
protection.  It's perfectly safe to do so, because the interrupt handler
can't run at this point, but it confuses lockdep.

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index e44350de8a21..0e9c5dcb2bd7 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1172,9 +1172,9 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
 	if (result < 0)
 		goto release_sq;
 
-	spin_lock(&nvmeq->q_lock);
+	spin_lock_irq(&nvmeq->q_lock);
 	nvme_init_queue(nvmeq, qid);
-	spin_unlock(&nvmeq->q_lock);
+	spin_unlock_irq(&nvmeq->q_lock);
 
 	return result;
 
@@ -1290,9 +1290,9 @@ static int nvme_configure_admin_queue(struct nvme_dev *dev)
 	if (result)
 		return result;
 
-	spin_lock(&nvmeq->q_lock);
+	spin_lock_irq(&nvmeq->q_lock);
 	nvme_init_queue(nvmeq, 0);
-	spin_unlock(&nvmeq->q_lock);
+	spin_unlock_irq(&nvmeq->q_lock);
 	return result;
 }
 
@@ -1836,9 +1836,9 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	for (i = dev->queue_count - 1; i > nr_io_queues; i--) {
 		struct nvme_queue *nvmeq = dev->queues[i];
 
-		spin_lock(&nvmeq->q_lock);
+		spin_lock_irq(&nvmeq->q_lock);
 		nvme_cancel_ios(nvmeq, false);
-		spin_unlock(&nvmeq->q_lock);
+		spin_unlock_irq(&nvmeq->q_lock);
 
 		nvme_free_queue(nvmeq);
 		dev->queue_count--;
-- 
cgit v1.2.3


From 68608c268bf265b039daeaafee6c902dfef32024 Mon Sep 17 00:00:00 2001
From: Matthew Wilcox
Date: Fri, 21 Jun 2013 14:36:34 -0400
Subject: NVMe: Cache dev->pci_dev in a local pointer

Helps with line-length issues

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 0e9c5dcb2bd7..300766973d76 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1891,6 +1891,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
  */
 static int nvme_dev_add(struct nvme_dev *dev)
 {
+	struct pci_dev *pdev = dev->pci_dev;
 	int res;
 	unsigned nn, i;
 	struct nvme_ns *ns;
@@ -1900,8 +1901,7 @@ static int nvme_dev_add(struct nvme_dev *dev)
 	dma_addr_t dma_addr;
 	int shift = NVME_CAP_MPSMIN(readq(&dev->bar->cap)) + 12;
 
-	mem = dma_alloc_coherent(&dev->pci_dev->dev, 8192, &dma_addr,
-								GFP_KERNEL);
+	mem = dma_alloc_coherent(&pdev->dev, 8192, &dma_addr, GFP_KERNEL);
 	if (!mem)
 		return -ENOMEM;
 
@@ -1919,8 +1919,8 @@ static int nvme_dev_add(struct nvme_dev *dev)
 	memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr));
 	if (ctrl->mdts)
 		dev->max_hw_sectors = 1 << (ctrl->mdts + shift - 9);
-	if ((dev->pci_dev->vendor == PCI_VENDOR_ID_INTEL) &&
-			(dev->pci_dev->device == 0x0953) && ctrl->vs[3])
+	if ((pdev->vendor == PCI_VENDOR_ID_INTEL) &&
+			(pdev->device == 0x0953) && ctrl->vs[3])
 		dev->stripe_size = 1 << (ctrl->vs[3] + shift);
 
 	id_ns = mem;
-- 
cgit v1.2.3


From 9a6b94584de1a0467d85b435df9c744c5c45a270 Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Tue, 10 Dec 2013 13:10:36 -0700
Subject: NVMe: Device resume error handling

Adds controller error handling on resume power management. If the device
fails to initialize, the device is queued for a reset. If the reset fails,
a thread is spawned to remove the pci device.

If the device resumes as "busy", the device is responding to admin
commands but will not create IO queues. In this case, we need to remove
the gendisks and free the IO queues since they can't be used and may be
holding bios in their lists.

From testing, the dma pools require a pci device so this had to change
the pci driver 'remove' to release the dma resources in line with that
call instead of after all references to the device are released.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 108 +++++++++++++++++++++++++++++++++++++++-------
 include/linux/nvme.h      |   1 +
 2 files changed, 94 insertions(+), 15 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 300766973d76..000bca43c23b 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -58,6 +58,7 @@ module_param(use_threaded_interrupts, int, 0);
 static DEFINE_SPINLOCK(dev_list_lock);
 static LIST_HEAD(dev_list);
 static struct task_struct *nvme_thread;
+static struct workqueue_struct *nvme_workq;
 
 /*
  * An NVM Express queue.  Each device has at least two (one for admin
@@ -1968,7 +1969,6 @@ static int nvme_dev_map(struct nvme_dev *dev)
 	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)))
 		goto disable;
 
-	pci_set_drvdata(pdev, dev);
 	dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
 	if (!dev->bar)
 		goto disable;
@@ -1995,9 +1995,9 @@ static void nvme_dev_unmap(struct nvme_dev *dev)
 	if (dev->bar) {
 		iounmap(dev->bar);
 		dev->bar = NULL;
+		pci_release_regions(dev->pci_dev);
 	}
 
-	pci_release_regions(dev->pci_dev);
 	if (pci_is_enabled(dev->pci_dev))
 		pci_disable_device(dev->pci_dev);
 }
@@ -2085,11 +2085,6 @@ static void nvme_release_instance(struct nvme_dev *dev)
 static void nvme_free_dev(struct kref *kref)
 {
 	struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
-	nvme_dev_remove(dev);
-	nvme_dev_shutdown(dev);
-	nvme_free_queues(dev);
-	nvme_release_instance(dev);
-	nvme_release_prp_pools(dev);
 	kfree(dev->queues);
 	kfree(dev->entry);
 	kfree(dev);
@@ -2161,6 +2156,70 @@ static int nvme_dev_start(struct nvme_dev *dev)
 	return result;
 }
 
+static int nvme_remove_dead_ctrl(void *arg)
+{
+	struct nvme_dev *dev = (struct nvme_dev *)arg;
+	struct pci_dev *pdev = dev->pci_dev;
+
+	if (pci_get_drvdata(pdev))
+		pci_stop_and_remove_bus_device(pdev);
+	kref_put(&dev->kref, nvme_free_dev);
+	return 0;
+}
+
+static void nvme_remove_disks(struct work_struct *ws)
+{
+	int i;
+	struct nvme_dev *dev = container_of(ws, struct nvme_dev, reset_work);
+
+	nvme_dev_remove(dev);
+	spin_lock(&dev_list_lock);
+	for (i = dev->queue_count - 1; i > 0; i--) {
+		BUG_ON(!dev->queues[i] || !dev->queues[i]->q_suspended);
+		nvme_free_queue(dev->queues[i]);
+		dev->queue_count--;
+		dev->queues[i] = NULL;
+	}
+	spin_unlock(&dev_list_lock);
+}
+
+static int nvme_dev_resume(struct nvme_dev *dev)
+{
+	int ret;
+
+	ret = nvme_dev_start(dev);
+	if (ret && ret != -EBUSY)
+		return ret;
+	if (ret == -EBUSY) {
+		spin_lock(&dev_list_lock);
+		INIT_WORK(&dev->reset_work, nvme_remove_disks);
+		queue_work(nvme_workq, &dev->reset_work);
+		spin_unlock(&dev_list_lock);
+	}
+	return 0;
+}
+
+static void nvme_dev_reset(struct nvme_dev *dev)
+{
+	nvme_dev_shutdown(dev);
+	if (nvme_dev_resume(dev)) {
+		dev_err(&dev->pci_dev->dev, "Device failed to resume\n");
+		kref_get(&dev->kref);
+		if (IS_ERR(kthread_run(nvme_remove_dead_ctrl, dev, "nvme%d",
+							dev->instance))) {
+			dev_err(&dev->pci_dev->dev,
+				"Failed to start controller remove task\n");
+			kref_put(&dev->kref, nvme_free_dev);
+		}
+	}
+}
+
+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_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	int result = -ENOMEM;
@@ -2180,7 +2239,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 
 	INIT_LIST_HEAD(&dev->namespaces);
 	dev->pci_dev = pdev;
-
+	pci_set_drvdata(pdev, dev);
 	result = nvme_set_instance(dev);
 	if (result)
 		goto free;
@@ -2232,7 +2291,19 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 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);
+
+	pci_set_drvdata(pdev, NULL);
+	flush_work(&dev->reset_work);
 	misc_deregister(&dev->miscdev);
+	nvme_dev_remove(dev);
+	nvme_dev_shutdown(dev);
+	nvme_free_queues(dev);
+	nvme_release_instance(dev);
+	nvme_release_prp_pools(dev);
 	kref_put(&dev->kref, nvme_free_dev);
 }
 
@@ -2256,13 +2327,12 @@ static int nvme_resume(struct device *dev)
 {
 	struct pci_dev *pdev = to_pci_dev(dev);
 	struct nvme_dev *ndev = pci_get_drvdata(pdev);
-	int ret;
 
-	ret = nvme_dev_start(ndev);
-	/* XXX: should remove gendisks if resume fails */
-	if (ret)
-		nvme_free_queues(ndev);
-	return ret;
+	if (nvme_dev_resume(ndev) && !work_busy(&ndev->reset_work)) {
+		INIT_WORK(&ndev->reset_work, nvme_reset_failed_dev);
+		queue_work(nvme_workq, &ndev->reset_work);
+	}
+	return 0;
 }
 
 static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
@@ -2303,9 +2373,14 @@ static int __init nvme_init(void)
 	if (IS_ERR(nvme_thread))
 		return PTR_ERR(nvme_thread);
 
+	result = -ENOMEM;
+	nvme_workq = create_singlethread_workqueue("nvme");
+	if (!nvme_workq)
+		goto kill_kthread;
+
 	result = register_blkdev(nvme_major, "nvme");
 	if (result < 0)
-		goto kill_kthread;
+		goto kill_workq;
 	else if (result > 0)
 		nvme_major = result;
 
@@ -2316,6 +2391,8 @@ static int __init nvme_init(void)
 
  unregister_blkdev:
 	unregister_blkdev(nvme_major, "nvme");
+ kill_workq:
+	destroy_workqueue(nvme_workq);
  kill_kthread:
 	kthread_stop(nvme_thread);
 	return result;
@@ -2325,6 +2402,7 @@ static void __exit nvme_exit(void)
 {
 	pci_unregister_driver(&nvme_driver);
 	unregister_blkdev(nvme_major, "nvme");
+	destroy_workqueue(nvme_workq);
 	kthread_stop(nvme_thread);
 }
 
diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index 5bc29197d90e..eed81cc56d7e 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -87,6 +87,7 @@ struct nvme_dev {
 	struct list_head namespaces;
 	struct kref kref;
 	struct miscdevice miscdev;
+	struct work_struct reset_work;
 	char name[12];
 	char serial[20];
 	char model[40];
-- 
cgit v1.2.3


From d4b4ff8e28b474fac0fbfa9cfc40f88b9e41e380 Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Tue, 10 Dec 2013 13:10:37 -0700
Subject: NVMe: Schedule reset for failed controllers

Schedules a controller reset when it indicates it has a failed status. If
the device does not become ready after a reset, the pci device will be
scheduled for removal.

Signed-off-by: Keith Busch <keith.busch@intel.com>
[fixed checkpatch issue]
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 21 +++++++++++++++++++--
 include/linux/nvme.h      |  1 +
 2 files changed, 20 insertions(+), 2 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 000bca43c23b..2f5b9f5f5a21 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -60,6 +60,8 @@ static LIST_HEAD(dev_list);
 static struct task_struct *nvme_thread;
 static struct workqueue_struct *nvme_workq;
 
+static void nvme_reset_failed_dev(struct work_struct *ws);
+
 /*
  * An NVM Express queue.  Each device has at least two (one for admin
  * commands and one for I/O commands).
@@ -1612,13 +1614,25 @@ static void nvme_resubmit_bios(struct nvme_queue *nvmeq)
 
 static int nvme_kthread(void *data)
 {
-	struct nvme_dev *dev;
+	struct nvme_dev *dev, *next;
 
 	while (!kthread_should_stop()) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		spin_lock(&dev_list_lock);
-		list_for_each_entry(dev, &dev_list, node) {
+		list_for_each_entry_safe(dev, next, &dev_list, node) {
 			int i;
+			if (readl(&dev->bar->csts) & NVME_CSTS_CFS &&
+							dev->initialized) {
+				if (work_busy(&dev->reset_work))
+					continue;
+				list_del_init(&dev->node);
+				dev_warn(&dev->pci_dev->dev,
+					"Failed status, reset controller\n");
+				INIT_WORK(&dev->reset_work,
+							nvme_reset_failed_dev);
+				queue_work(nvme_workq, &dev->reset_work);
+				continue;
+			}
 			for (i = 0; i < dev->queue_count; i++) {
 				struct nvme_queue *nvmeq = dev->queues[i];
 				if (!nvmeq)
@@ -2006,6 +2020,7 @@ static void nvme_dev_shutdown(struct nvme_dev *dev)
 {
 	int i;
 
+	dev->initialized = 0;
 	for (i = dev->queue_count - 1; i >= 0; i--)
 		nvme_disable_queue(dev, i);
 
@@ -2196,6 +2211,7 @@ static int nvme_dev_resume(struct nvme_dev *dev)
 		queue_work(nvme_workq, &dev->reset_work);
 		spin_unlock(&dev_list_lock);
 	}
+	dev->initialized = 1;
 	return 0;
 }
 
@@ -2269,6 +2285,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (result)
 		goto remove;
 
+	dev->initialized = 1;
 	kref_init(&dev->kref);
 	return 0;
 
diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index eed81cc56d7e..117d877e8be5 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -95,6 +95,7 @@ struct nvme_dev {
 	u32 max_hw_sectors;
 	u32 stripe_size;
 	u16 oncs;
+	u8 initialized;
 };
 
 /*
-- 
cgit v1.2.3


From c30341dc3c436cf43508cd44cdfbb3810c38c195 Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Tue, 10 Dec 2013 13:10:38 -0700
Subject: NVMe: Abort timed out commands

Send nvme abort command to io requests that have timed out on an
initialized device. If the command is not returned after another timeout,
schedule the controller for reset.

Signed-off-by: Keith Busch <keith.busch@intel.com>
[fix endianness issues]
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 68 ++++++++++++++++++++++++++++++++++++++++++++++-
 include/linux/nvme.h      |  1 +
 include/uapi/linux/nvme.h | 11 ++++++++
 3 files changed, 79 insertions(+), 1 deletion(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 2f5b9f5f5a21..75049532df96 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -83,6 +83,7 @@ struct nvme_queue {
 	u16 sq_head;
 	u16 sq_tail;
 	u16 cq_head;
+	u16 qid;
 	u8 cq_phase;
 	u8 cqe_seen;
 	u8 q_suspended;
@@ -100,6 +101,7 @@ static inline void _nvme_check_size(void)
 	BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64);
 	BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
 	BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
+	BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
 	BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
 	BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096);
 	BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096);
@@ -114,6 +116,7 @@ struct nvme_cmd_info {
 	nvme_completion_fn fn;
 	void *ctx;
 	unsigned long timeout;
+	int aborted;
 };
 
 static struct nvme_cmd_info *nvme_cmd_info(struct nvme_queue *nvmeq)
@@ -157,6 +160,7 @@ static int alloc_cmdid(struct nvme_queue *nvmeq, void *ctx,
 	info[cmdid].fn = handler;
 	info[cmdid].ctx = ctx;
 	info[cmdid].timeout = jiffies + timeout;
+	info[cmdid].aborted = 0;
 	return cmdid;
 }
 
@@ -175,6 +179,7 @@ static int alloc_cmdid_killable(struct nvme_queue *nvmeq, void *ctx,
 #define CMD_CTX_COMPLETED	(0x310 + CMD_CTX_BASE)
 #define CMD_CTX_INVALID		(0x314 + CMD_CTX_BASE)
 #define CMD_CTX_FLUSH		(0x318 + CMD_CTX_BASE)
+#define CMD_CTX_ABORT		(0x31C + CMD_CTX_BASE)
 
 static void special_completion(struct nvme_dev *dev, void *ctx,
 						struct nvme_completion *cqe)
@@ -183,6 +188,10 @@ static void special_completion(struct nvme_dev *dev, void *ctx,
 		return;
 	if (ctx == CMD_CTX_FLUSH)
 		return;
+	if (ctx == CMD_CTX_ABORT) {
+		++dev->abort_limit;
+		return;
+	}
 	if (ctx == CMD_CTX_COMPLETED) {
 		dev_warn(&dev->pci_dev->dev,
 				"completed id %d twice on queue %d\n",
@@ -1004,6 +1013,56 @@ int nvme_set_features(struct nvme_dev *dev, unsigned fid, unsigned dword11,
 	return nvme_submit_admin_cmd(dev, &c, result);
 }
 
+/**
+ * nvme_abort_cmd - Attempt aborting a command
+ * @cmdid: Command id of a timed out IO
+ * @queue: The queue with timed out IO
+ *
+ * Schedule controller reset if the command was already aborted once before and
+ * still hasn't been returned to the driver, or if this is the admin queue.
+ */
+static void nvme_abort_cmd(int cmdid, struct nvme_queue *nvmeq)
+{
+	int a_cmdid;
+	struct nvme_command cmd;
+	struct nvme_dev *dev = nvmeq->dev;
+	struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
+
+	if (!nvmeq->qid || info[cmdid].aborted) {
+		if (work_busy(&dev->reset_work))
+			return;
+		list_del_init(&dev->node);
+		dev_warn(&dev->pci_dev->dev,
+			"I/O %d QID %d timeout, reset controller\n", cmdid,
+								nvmeq->qid);
+		INIT_WORK(&dev->reset_work, nvme_reset_failed_dev);
+		queue_work(nvme_workq, &dev->reset_work);
+		return;
+	}
+
+	if (!dev->abort_limit)
+		return;
+
+	a_cmdid = alloc_cmdid(dev->queues[0], CMD_CTX_ABORT, special_completion,
+								ADMIN_TIMEOUT);
+	if (a_cmdid < 0)
+		return;
+
+	memset(&cmd, 0, sizeof(cmd));
+	cmd.abort.opcode = nvme_admin_abort_cmd;
+	cmd.abort.cid = cmdid;
+	cmd.abort.sqid = cpu_to_le16(nvmeq->qid);
+	cmd.abort.command_id = a_cmdid;
+
+	--dev->abort_limit;
+	info[cmdid].aborted = 1;
+	info[cmdid].timeout = jiffies + ADMIN_TIMEOUT;
+
+	dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", cmdid,
+							nvmeq->qid);
+	nvme_submit_cmd(dev->queues[0], &cmd);
+}
+
 /**
  * nvme_cancel_ios - Cancel outstanding I/Os
  * @queue: The queue to cancel I/Os on
@@ -1027,7 +1086,12 @@ static void nvme_cancel_ios(struct nvme_queue *nvmeq, bool timeout)
 			continue;
 		if (info[cmdid].ctx == CMD_CTX_CANCELLED)
 			continue;
-		dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d\n", cmdid);
+		if (timeout && nvmeq->dev->initialized) {
+			nvme_abort_cmd(cmdid, nvmeq);
+			continue;
+		}
+		dev_warn(nvmeq->q_dmadev, "Cancelling I/O %d QID %d\n", cmdid,
+								nvmeq->qid);
 		ctx = cancel_cmdid(nvmeq, cmdid, &fn);
 		fn(nvmeq->dev, ctx, &cqe);
 	}
@@ -1119,6 +1183,7 @@ static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
 	nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride];
 	nvmeq->q_depth = depth;
 	nvmeq->cq_vector = vector;
+	nvmeq->qid = qid;
 	nvmeq->q_suspended = 1;
 	dev->queue_count++;
 
@@ -1929,6 +1994,7 @@ static int nvme_dev_add(struct nvme_dev *dev)
 	ctrl = mem;
 	nn = le32_to_cpup(&ctrl->nn);
 	dev->oncs = le16_to_cpup(&ctrl->oncs);
+	dev->abort_limit = ctrl->acl + 1;
 	memcpy(dev->serial, ctrl->sn, sizeof(ctrl->sn));
 	memcpy(dev->model, ctrl->mn, sizeof(ctrl->mn));
 	memcpy(dev->firmware_rev, ctrl->fr, sizeof(ctrl->fr));
diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index 117d877e8be5..69ae03f6eb15 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -95,6 +95,7 @@ struct nvme_dev {
 	u32 max_hw_sectors;
 	u32 stripe_size;
 	u16 oncs;
+	u16 abort_limit;
 	u8 initialized;
 };
 
diff --git a/include/uapi/linux/nvme.h b/include/uapi/linux/nvme.h
index 989c04e0c563..e5ab62201119 100644
--- a/include/uapi/linux/nvme.h
+++ b/include/uapi/linux/nvme.h
@@ -350,6 +350,16 @@ struct nvme_delete_queue {
 	__u32			rsvd11[5];
 };
 
+struct nvme_abort_cmd {
+	__u8			opcode;
+	__u8			flags;
+	__u16			command_id;
+	__u32			rsvd1[9];
+	__le16			sqid;
+	__u16			cid;
+	__u32			rsvd11[5];
+};
+
 struct nvme_download_firmware {
 	__u8			opcode;
 	__u8			flags;
@@ -384,6 +394,7 @@ struct nvme_command {
 		struct nvme_download_firmware dlfw;
 		struct nvme_format_cmd format;
 		struct nvme_dsm_cmd dsm;
+		struct nvme_abort_cmd abort;
 	};
 };
 
-- 
cgit v1.2.3


From 0e53d18051725da46cbccfb7874a6422d4d4f274 Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Tue, 10 Dec 2013 13:10:39 -0700
Subject: NVMe: Surprise removal handling

This adds checks to see if the nvme pci device was removed. The check
reads the status register for the value of -1, which it should never be
unless the device is no longer present.

If a user performs a surprise removal on an nvme device, the driver will
be notified either by the pci driver remove callback if the platform's
slot is capable of this event, or via reading the device BAR status
register, which will indicate controller failure and trigger a reset.

Either way, the device is not present so all outstanding commands would
timeout. This will not send queue deletion commands to a drive that
isn't present and fail after ioremap, significantly speeding up surprise
removal; previously this took over 2 minutes per IO queue pair created,
but this will complete removing the device within a few seconds.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 75049532df96..a51126129784 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1140,8 +1140,9 @@ static void nvme_disable_queue(struct nvme_dev *dev, int qid)
 	irq_set_affinity_hint(vector, NULL);
 	free_irq(vector, nvmeq);
 
-	/* Don't tell the adapter to delete the admin queue */
-	if (qid) {
+	/* Don't tell the adapter to delete the admin queue.
+	 * Don't tell a removed adapter to delete IO queues. */
+	if (qid && readl(&dev->bar->csts) != -1) {
 		adapter_delete_sq(dev, qid);
 		adapter_delete_cq(dev, qid);
 	}
@@ -2052,12 +2053,18 @@ static int nvme_dev_map(struct nvme_dev *dev)
 	dev->bar = ioremap(pci_resource_start(pdev, 0), 8192);
 	if (!dev->bar)
 		goto disable;
-
+	if (readl(&dev->bar->csts) == -1) {
+		result = -ENODEV;
+		goto unmap;
+	}
 	dev->db_stride = 1 << NVME_CAP_STRIDE(readq(&dev->bar->cap));
 	dev->dbs = ((void __iomem *)dev->bar) + 4096;
 
 	return 0;
 
+ unmap:
+	iounmap(dev->bar);
+	dev->bar = NULL;
  disable:
 	pci_release_regions(pdev);
  disable_pci:
-- 
cgit v1.2.3


From 4d115420707afcabe77d2535e092356df6664b70 Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Tue, 10 Dec 2013 13:10:40 -0700
Subject: NVMe: Async IO queue deletion

This attempts to delete all IO queues at the same time asynchronously on
shutdown. This is necessary for a present device that is not responding;
a shutdown operation previously would take 2 minutes per queue-pair
to timeout before moving on to the next queue, making a device removal
appear to take a very long time or "hung" as reported by users.

In the previous worst case, a removal may be stuck forever until a kill
signal is given if there are more than 32 queue pairs since it would run
out of admin command IDs after over an hour of timed out sync commands
(admin queue depth is 64).

This patch will wait for the admin command timeout for all commands to
complete, so the worst case now for an unresponsive controller is 60
seconds, though that still seems like a long time.

Since this adds another way to take queues offline, some duplicate code
resulted so I moved these into more convienient functions.

Signed-off-by: Keith Busch <keith.busch@intel.com>
[make functions static, correct line length and whitespace issues]
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 229 +++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 217 insertions(+), 12 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index a51126129784..1c8a82fbbc37 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -62,6 +62,14 @@ static struct workqueue_struct *nvme_workq;
 
 static void nvme_reset_failed_dev(struct work_struct *ws);
 
+struct async_cmd_info {
+	struct kthread_work work;
+	struct kthread_worker *worker;
+	u32 result;
+	int status;
+	void *ctx;
+};
+
 /*
  * An NVM Express queue.  Each device has at least two (one for admin
  * commands and one for I/O commands).
@@ -87,6 +95,7 @@ struct nvme_queue {
 	u8 cq_phase;
 	u8 cqe_seen;
 	u8 q_suspended;
+	struct async_cmd_info cmdinfo;
 	unsigned long cmdid_data[];
 };
 
@@ -208,6 +217,15 @@ static void special_completion(struct nvme_dev *dev, void *ctx,
 	dev_warn(&dev->pci_dev->dev, "Unknown special completion %p\n", ctx);
 }
 
+static void async_completion(struct nvme_dev *dev, void *ctx,
+						struct nvme_completion *cqe)
+{
+	struct async_cmd_info *cmdinfo = ctx;
+	cmdinfo->result = le32_to_cpup(&cqe->result);
+	cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
+	queue_kthread_work(cmdinfo->worker, &cmdinfo->work);
+}
+
 /*
  * Called with local interrupts disabled and the q_lock held.  May not sleep.
  */
@@ -898,12 +916,34 @@ int nvme_submit_sync_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd,
 	return cmdinfo.status;
 }
 
+static int nvme_submit_async_cmd(struct nvme_queue *nvmeq,
+			struct nvme_command *cmd,
+			struct async_cmd_info *cmdinfo, unsigned timeout)
+{
+	int cmdid;
+
+	cmdid = alloc_cmdid_killable(nvmeq, cmdinfo, async_completion, timeout);
+	if (cmdid < 0)
+		return cmdid;
+	cmdinfo->status = -EINTR;
+	cmd->common.command_id = cmdid;
+	nvme_submit_cmd(nvmeq, cmd);
+	return 0;
+}
+
 int nvme_submit_admin_cmd(struct nvme_dev *dev, struct nvme_command *cmd,
 								u32 *result)
 {
 	return nvme_submit_sync_cmd(dev->queues[0], cmd, result, ADMIN_TIMEOUT);
 }
 
+static int nvme_submit_admin_cmd_async(struct nvme_dev *dev,
+		struct nvme_command *cmd, struct async_cmd_info *cmdinfo)
+{
+	return nvme_submit_async_cmd(dev->queues[0], cmd, cmdinfo,
+								ADMIN_TIMEOUT);
+}
+
 static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id)
 {
 	int status;
@@ -1124,15 +1164,20 @@ static void nvme_free_queues(struct nvme_dev *dev)
 	}
 }
 
-static void nvme_disable_queue(struct nvme_dev *dev, int qid)
+/**
+ * nvme_suspend_queue - put queue into suspended state
+ * @nvmeq - queue to suspend
+ *
+ * Returns 1 if already suspended, 0 otherwise.
+ */
+static int nvme_suspend_queue(struct nvme_queue *nvmeq)
 {
-	struct nvme_queue *nvmeq = dev->queues[qid];
-	int vector = dev->entry[nvmeq->cq_vector].vector;
+	int vector = nvmeq->dev->entry[nvmeq->cq_vector].vector;
 
 	spin_lock_irq(&nvmeq->q_lock);
 	if (nvmeq->q_suspended) {
 		spin_unlock_irq(&nvmeq->q_lock);
-		return;
+		return 1;
 	}
 	nvmeq->q_suspended = 1;
 	spin_unlock_irq(&nvmeq->q_lock);
@@ -1140,17 +1185,33 @@ static void nvme_disable_queue(struct nvme_dev *dev, int qid)
 	irq_set_affinity_hint(vector, NULL);
 	free_irq(vector, nvmeq);
 
+	return 0;
+}
+
+static void nvme_clear_queue(struct nvme_queue *nvmeq)
+{
+	spin_lock_irq(&nvmeq->q_lock);
+	nvme_process_cq(nvmeq);
+	nvme_cancel_ios(nvmeq, false);
+	spin_unlock_irq(&nvmeq->q_lock);
+}
+
+static void nvme_disable_queue(struct nvme_dev *dev, int qid)
+{
+	struct nvme_queue *nvmeq = dev->queues[qid];
+
+	if (!nvmeq)
+		return;
+	if (nvme_suspend_queue(nvmeq))
+		return;
+
 	/* Don't tell the adapter to delete the admin queue.
 	 * Don't tell a removed adapter to delete IO queues. */
 	if (qid && readl(&dev->bar->csts) != -1) {
 		adapter_delete_sq(dev, qid);
 		adapter_delete_cq(dev, qid);
 	}
-
-	spin_lock_irq(&nvmeq->q_lock);
-	nvme_process_cq(nvmeq);
-	nvme_cancel_ios(nvmeq, false);
-	spin_unlock_irq(&nvmeq->q_lock);
+	nvme_clear_queue(nvmeq);
 }
 
 static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
@@ -2089,20 +2150,164 @@ static void nvme_dev_unmap(struct nvme_dev *dev)
 		pci_disable_device(dev->pci_dev);
 }
 
+struct nvme_delq_ctx {
+	struct task_struct *waiter;
+	struct kthread_worker *worker;
+	atomic_t refcount;
+};
+
+static void nvme_wait_dq(struct nvme_delq_ctx *dq, struct nvme_dev *dev)
+{
+	dq->waiter = current;
+	mb();
+
+	for (;;) {
+		set_current_state(TASK_KILLABLE);
+		if (!atomic_read(&dq->refcount))
+			break;
+		if (!schedule_timeout(ADMIN_TIMEOUT) ||
+					fatal_signal_pending(current)) {
+			set_current_state(TASK_RUNNING);
+
+			nvme_disable_ctrl(dev, readq(&dev->bar->cap));
+			nvme_disable_queue(dev, 0);
+
+			send_sig(SIGKILL, dq->worker->task, 1);
+			flush_kthread_worker(dq->worker);
+			return;
+		}
+	}
+	set_current_state(TASK_RUNNING);
+}
+
+static void nvme_put_dq(struct nvme_delq_ctx *dq)
+{
+	atomic_dec(&dq->refcount);
+	if (dq->waiter)
+		wake_up_process(dq->waiter);
+}
+
+static struct nvme_delq_ctx *nvme_get_dq(struct nvme_delq_ctx *dq)
+{
+	atomic_inc(&dq->refcount);
+	return dq;
+}
+
+static void nvme_del_queue_end(struct nvme_queue *nvmeq)
+{
+	struct nvme_delq_ctx *dq = nvmeq->cmdinfo.ctx;
+
+	nvme_clear_queue(nvmeq);
+	nvme_put_dq(dq);
+}
+
+static int adapter_async_del_queue(struct nvme_queue *nvmeq, u8 opcode,
+						kthread_work_func_t fn)
+{
+	struct nvme_command c;
+
+	memset(&c, 0, sizeof(c));
+	c.delete_queue.opcode = opcode;
+	c.delete_queue.qid = cpu_to_le16(nvmeq->qid);
+
+	init_kthread_work(&nvmeq->cmdinfo.work, fn);
+	return nvme_submit_admin_cmd_async(nvmeq->dev, &c, &nvmeq->cmdinfo);
+}
+
+static void nvme_del_cq_work_handler(struct kthread_work *work)
+{
+	struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
+							cmdinfo.work);
+	nvme_del_queue_end(nvmeq);
+}
+
+static int nvme_delete_cq(struct nvme_queue *nvmeq)
+{
+	return adapter_async_del_queue(nvmeq, nvme_admin_delete_cq,
+						nvme_del_cq_work_handler);
+}
+
+static void nvme_del_sq_work_handler(struct kthread_work *work)
+{
+	struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
+							cmdinfo.work);
+	int status = nvmeq->cmdinfo.status;
+
+	if (!status)
+		status = nvme_delete_cq(nvmeq);
+	if (status)
+		nvme_del_queue_end(nvmeq);
+}
+
+static int nvme_delete_sq(struct nvme_queue *nvmeq)
+{
+	return adapter_async_del_queue(nvmeq, nvme_admin_delete_sq,
+						nvme_del_sq_work_handler);
+}
+
+static void nvme_del_queue_start(struct kthread_work *work)
+{
+	struct nvme_queue *nvmeq = container_of(work, struct nvme_queue,
+							cmdinfo.work);
+	allow_signal(SIGKILL);
+	if (nvme_delete_sq(nvmeq))
+		nvme_del_queue_end(nvmeq);
+}
+
+static void nvme_disable_io_queues(struct nvme_dev *dev)
+{
+	int i;
+	DEFINE_KTHREAD_WORKER_ONSTACK(worker);
+	struct nvme_delq_ctx dq;
+	struct task_struct *kworker_task = kthread_run(kthread_worker_fn,
+					&worker, "nvme%d", dev->instance);
+
+	if (IS_ERR(kworker_task)) {
+		dev_err(&dev->pci_dev->dev,
+			"Failed to create queue del task\n");
+		for (i = dev->queue_count - 1; i > 0; i--)
+			nvme_disable_queue(dev, i);
+		return;
+	}
+
+	dq.waiter = NULL;
+	atomic_set(&dq.refcount, 0);
+	dq.worker = &worker;
+	for (i = dev->queue_count - 1; i > 0; i--) {
+		struct nvme_queue *nvmeq = dev->queues[i];
+
+		if (nvme_suspend_queue(nvmeq))
+			continue;
+		nvmeq->cmdinfo.ctx = nvme_get_dq(&dq);
+		nvmeq->cmdinfo.worker = dq.worker;
+		init_kthread_work(&nvmeq->cmdinfo.work, nvme_del_queue_start);
+		queue_kthread_work(dq.worker, &nvmeq->cmdinfo.work);
+	}
+	nvme_wait_dq(&dq, dev);
+	kthread_stop(kworker_task);
+}
+
 static void nvme_dev_shutdown(struct nvme_dev *dev)
 {
 	int i;
 
 	dev->initialized = 0;
-	for (i = dev->queue_count - 1; i >= 0; i--)
-		nvme_disable_queue(dev, i);
 
 	spin_lock(&dev_list_lock);
 	list_del_init(&dev->node);
 	spin_unlock(&dev_list_lock);
 
-	if (dev->bar)
+	if (!dev->bar || (dev->bar && readl(&dev->bar->csts) == -1)) {
+		for (i = dev->queue_count - 1; i >= 0; i--) {
+			struct nvme_queue *nvmeq = dev->queues[i];
+			nvme_suspend_queue(nvmeq);
+			nvme_clear_queue(nvmeq);
+		}
+	} else {
+		nvme_disable_io_queues(dev);
 		nvme_shutdown_ctrl(dev);
+		nvme_disable_queue(dev, 0);
+	}
 	nvme_dev_unmap(dev);
 }
 
-- 
cgit v1.2.3


From 469071a37afc8a627b6b2ddf29db0a097d864845 Mon Sep 17 00:00:00 2001
From: Matthew Wilcox
Date: Mon, 9 Dec 2013 12:58:46 -0500
Subject: NVMe: Dynamically allocate partition numbers

Some users need more than 64 partitions per device.  Rather than simply
increasing the number of partitions, switch to the dynamic partition
allocation scheme.

This means that minor numbers are not stable across boots, but since major
numbers aren't either, I cannot see this being a significant problem.

Tested-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 36 +++---------------------------------
 1 file changed, 3 insertions(+), 33 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 1c8a82fbbc37..7b615a063e92 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -46,7 +46,6 @@
 #define NVME_Q_DEPTH 1024
 #define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
 #define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
-#define NVME_MINORS 64
 #define ADMIN_TIMEOUT	(60 * HZ)
 
 static int nvme_major;
@@ -1780,33 +1779,6 @@ static int nvme_kthread(void *data)
 	return 0;
 }
 
-static DEFINE_IDA(nvme_index_ida);
-
-static int nvme_get_ns_idx(void)
-{
-	int index, error;
-
-	do {
-		if (!ida_pre_get(&nvme_index_ida, GFP_KERNEL))
-			return -1;
-
-		spin_lock(&dev_list_lock);
-		error = ida_get_new(&nvme_index_ida, &index);
-		spin_unlock(&dev_list_lock);
-	} while (error == -EAGAIN);
-
-	if (error)
-		index = -1;
-	return index;
-}
-
-static void nvme_put_ns_idx(int index)
-{
-	spin_lock(&dev_list_lock);
-	ida_remove(&nvme_index_ida, index);
-	spin_unlock(&dev_list_lock);
-}
-
 static void nvme_config_discard(struct nvme_ns *ns)
 {
 	u32 logical_block_size = queue_logical_block_size(ns->queue);
@@ -1840,7 +1812,7 @@ static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid,
 	ns->dev = dev;
 	ns->queue->queuedata = ns;
 
-	disk = alloc_disk(NVME_MINORS);
+	disk = alloc_disk(0);
 	if (!disk)
 		goto out_free_queue;
 	ns->ns_id = nsid;
@@ -1853,12 +1825,12 @@ static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid,
 		blk_queue_max_hw_sectors(ns->queue, dev->max_hw_sectors);
 
 	disk->major = nvme_major;
-	disk->minors = NVME_MINORS;
-	disk->first_minor = NVME_MINORS * nvme_get_ns_idx();
+	disk->first_minor = 0;
 	disk->fops = &nvme_fops;
 	disk->private_data = ns;
 	disk->queue = ns->queue;
 	disk->driverfs_dev = &dev->pci_dev->dev;
+	disk->flags = GENHD_FL_EXT_DEVT;
 	sprintf(disk->disk_name, "nvme%dn%d", dev->instance, nsid);
 	set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 
@@ -1876,9 +1848,7 @@ static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid,
 
 static void nvme_ns_free(struct nvme_ns *ns)
 {
-	int index = ns->disk->first_minor / NVME_MINORS;
 	put_disk(ns->disk);
-	nvme_put_ns_idx(index);
 	blk_cleanup_queue(ns->queue);
 	kfree(ns);
 }
-- 
cgit v1.2.3


From a1a5ef9998b1379780790b878457b0e0f48b25db Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Mon, 16 Dec 2013 13:50:00 -0500
Subject: NVMe: Disable admin queue on init failure

Disable the admin queue if device fails during initialization so the
queue's irq is freed.

Signed-off-by: Keith Busch <keith.busch@intel.com>
[rewritten to use nvme_free_queues]
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 7b615a063e92..89021d710ec2 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1152,11 +1152,11 @@ static void nvme_free_queue(struct nvme_queue *nvmeq)
 	kfree(nvmeq);
 }
 
-static void nvme_free_queues(struct nvme_dev *dev)
+static void nvme_free_queues(struct nvme_dev *dev, int lowest)
 {
 	int i;
 
-	for (i = dev->queue_count - 1; i >= 0; i--) {
+	for (i = dev->queue_count - 1; i >= lowest; i--) {
 		nvme_free_queue(dev->queues[i]);
 		dev->queue_count--;
 		dev->queues[i] = NULL;
@@ -1991,7 +1991,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	return 0;
 
  free_queues:
-	nvme_free_queues(dev);
+	nvme_free_queues(dev, 1);
 	return result;
 }
 
@@ -2411,6 +2411,7 @@ static int nvme_dev_start(struct nvme_dev *dev)
 	return result;
 
  disable:
+	nvme_disable_queue(dev, 0);
 	spin_lock(&dev_list_lock);
 	list_del_init(&dev->node);
 	spin_unlock(&dev_list_lock);
@@ -2542,7 +2543,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
  shutdown:
 	nvme_dev_shutdown(dev);
  release_pools:
-	nvme_free_queues(dev);
+	nvme_free_queues(dev, 0);
 	nvme_release_prp_pools(dev);
  release:
 	nvme_release_instance(dev);
@@ -2566,7 +2567,7 @@ static void nvme_remove(struct pci_dev *pdev)
 	misc_deregister(&dev->miscdev);
 	nvme_dev_remove(dev);
 	nvme_dev_shutdown(dev);
-	nvme_free_queues(dev);
+	nvme_free_queues(dev, 0);
 	nvme_release_instance(dev);
 	nvme_release_prp_pools(dev);
 	kref_put(&dev->kref, nvme_free_dev);
-- 
cgit v1.2.3


From 09ece1424f52a2d3ad0ab715e96b0fb342af4f03 Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Mon, 27 Jan 2014 11:29:40 -0500
Subject: NVMe: Add a pci_driver shutdown method

We need to shut down the device cleanly when the system is being shut down.
This was in an earlier patch but was inadvertently lost during a rewrite.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 7 +++++++
 1 file changed, 7 insertions(+)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 89021d710ec2..eaace767be49 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -2554,6 +2554,12 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	return result;
 }
 
+static void nvme_shutdown(struct pci_dev *pdev)
+{
+	struct nvme_dev *dev = pci_get_drvdata(pdev);
+	nvme_dev_shutdown(dev);
+}
+
 static void nvme_remove(struct pci_dev *pdev)
 {
 	struct nvme_dev *dev = pci_get_drvdata(pdev);
@@ -2625,6 +2631,7 @@ static struct pci_driver nvme_driver = {
 	.id_table	= nvme_id_table,
 	.probe		= nvme_probe,
 	.remove		= nvme_remove,
+	.shutdown	= nvme_shutdown,
 	.driver		= {
 		.pm	= &nvme_dev_pm_ops,
 	},
-- 
cgit v1.2.3


From 3193f07bb7ae723f33ac8e8b9db317a4f68d7d18 Mon Sep 17 00:00:00 2001
From: Matthew Wilcox
Date: Mon, 27 Jan 2014 15:57:22 -0500
Subject: NVMe: Include device and queue numbers in interrupt name

On larger systems with many drives, it may help debugging to know which
queue is tied to which interrupt, just by looking at /proc/interrupts.

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index eaace767be49..5ffc26937226 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -76,6 +76,7 @@ struct async_cmd_info {
 struct nvme_queue {
 	struct device *q_dmadev;
 	struct nvme_dev *dev;
+	char irqname[24];	/* nvme4294967295-65535\0 */
 	spinlock_t q_lock;
 	struct nvme_command *sq_cmds;
 	volatile struct nvme_completion *cqes;
@@ -1235,6 +1236,8 @@ static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
 
 	nvmeq->q_dmadev = dmadev;
 	nvmeq->dev = dev;
+	snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
+			dev->instance, qid);
 	spin_lock_init(&nvmeq->q_lock);
 	nvmeq->cq_head = 0;
 	nvmeq->cq_phase = 1;
@@ -1297,7 +1300,7 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
 	if (result < 0)
 		goto release_cq;
 
-	result = queue_request_irq(dev, nvmeq, "nvme");
+	result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
 	if (result < 0)
 		goto release_sq;
 
@@ -1415,7 +1418,7 @@ static int nvme_configure_admin_queue(struct nvme_dev *dev)
 	if (result)
 		return result;
 
-	result = queue_request_irq(dev, nvmeq, "nvme admin");
+	result = queue_request_irq(dev, nvmeq, nvmeq->irqname);
 	if (result)
 		return result;
 
@@ -1873,6 +1876,7 @@ static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
 
 static int nvme_setup_io_queues(struct nvme_dev *dev)
 {
+	struct nvme_queue *adminq = dev->queues[0];
 	struct pci_dev *pdev = dev->pci_dev;
 	int result, cpu, i, vecs, nr_io_queues, size, q_depth;
 
@@ -1899,7 +1903,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	}
 
 	/* Deregister the admin queue's interrupt */
-	free_irq(dev->entry[0].vector, dev->queues[0]);
+	free_irq(dev->entry[0].vector, adminq);
 
 	vecs = nr_io_queues;
 	for (i = 0; i < vecs; i++)
@@ -1937,9 +1941,9 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	 */
 	nr_io_queues = vecs;
 
-	result = queue_request_irq(dev, dev->queues[0], "nvme admin");
+	result = queue_request_irq(dev, adminq, adminq->irqname);
 	if (result) {
-		dev->queues[0]->q_suspended = 1;
+		adminq->q_suspended = 1;
 		goto free_queues;
 	}
 
-- 
cgit v1.2.3


From bdfd70fde389412be60f7e8aaed5732dc26fc8ac Mon Sep 17 00:00:00 2001
From: Matthew Wilcox
Date: Wed, 29 Jan 2014 09:33:52 -0500
Subject: NVMe: Correct uses of INIT_WORK

We need to initialise the work_struct when we initialise the rest of the
struct nvme_dev, otherwise we'll hit a lockdep warning when we remove
the device.  Use PREPARE_WORK to change the function pointer instead
of INIT_WORK.

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 5ffc26937226..23728099e36f 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1075,7 +1075,7 @@ static void nvme_abort_cmd(int cmdid, struct nvme_queue *nvmeq)
 		dev_warn(&dev->pci_dev->dev,
 			"I/O %d QID %d timeout, reset controller\n", cmdid,
 								nvmeq->qid);
-		INIT_WORK(&dev->reset_work, nvme_reset_failed_dev);
+		PREPARE_WORK(&dev->reset_work, nvme_reset_failed_dev);
 		queue_work(nvme_workq, &dev->reset_work);
 		return;
 	}
@@ -1757,7 +1757,7 @@ static int nvme_kthread(void *data)
 				list_del_init(&dev->node);
 				dev_warn(&dev->pci_dev->dev,
 					"Failed status, reset controller\n");
-				INIT_WORK(&dev->reset_work,
+				PREPARE_WORK(&dev->reset_work,
 							nvme_reset_failed_dev);
 				queue_work(nvme_workq, &dev->reset_work);
 				continue;
@@ -2460,7 +2460,7 @@ static int nvme_dev_resume(struct nvme_dev *dev)
 		return ret;
 	if (ret == -EBUSY) {
 		spin_lock(&dev_list_lock);
-		INIT_WORK(&dev->reset_work, nvme_remove_disks);
+		PREPARE_WORK(&dev->reset_work, nvme_remove_disks);
 		queue_work(nvme_workq, &dev->reset_work);
 		spin_unlock(&dev_list_lock);
 	}
@@ -2507,6 +2507,7 @@ 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);
 	dev->pci_dev = pdev;
 	pci_set_drvdata(pdev, dev);
 	result = nvme_set_instance(dev);
@@ -2605,7 +2606,7 @@ static int nvme_resume(struct device *dev)
 	struct nvme_dev *ndev = pci_get_drvdata(pdev);
 
 	if (nvme_dev_resume(ndev) && !work_busy(&ndev->reset_work)) {
-		INIT_WORK(&ndev->reset_work, nvme_reset_failed_dev);
+		PREPARE_WORK(&ndev->reset_work, nvme_reset_failed_dev);
 		queue_work(nvme_workq, &ndev->reset_work);
 	}
 	return 0;
-- 
cgit v1.2.3


From 9ac27090f61ea6735a62b0a98c7669c833bcdc09 Mon Sep 17 00:00:00 2001
From: Keith Busch
Date: Fri, 31 Jan 2014 16:53:39 -0700
Subject: NVMe: Namespace use after free on surprise removal

An nvme block device may have open references when the device is
removed. New commands may still be sent on the removed device, so we
need to ref count the opens, return errors for new commands, and not
free the namespace and nvme_dev until all references are closed.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
---
 drivers/block/nvme-core.c | 55 ++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 42 insertions(+), 13 deletions(-)

(limited to 'drivers')

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index 23728099e36f..cd39390710a0 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1716,10 +1716,31 @@ static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 #define nvme_compat_ioctl	NULL
 #endif
 
+static int nvme_open(struct block_device *bdev, fmode_t mode)
+{
+	struct nvme_ns *ns = bdev->bd_disk->private_data;
+	struct nvme_dev *dev = ns->dev;
+
+	kref_get(&dev->kref);
+	return 0;
+}
+
+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);
+}
+
 static const struct block_device_operations nvme_fops = {
 	.owner		= THIS_MODULE,
 	.ioctl		= nvme_ioctl,
 	.compat_ioctl	= nvme_compat_ioctl,
+	.open		= nvme_open,
+	.release	= nvme_release,
 };
 
 static void nvme_resubmit_bios(struct nvme_queue *nvmeq)
@@ -1849,13 +1870,6 @@ static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid,
 	return NULL;
 }
 
-static void nvme_ns_free(struct nvme_ns *ns)
-{
-	put_disk(ns->disk);
-	blk_cleanup_queue(ns->queue);
-	kfree(ns);
-}
-
 static int set_queue_count(struct nvme_dev *dev, int count)
 {
 	int status;
@@ -2287,12 +2301,13 @@ static void nvme_dev_shutdown(struct nvme_dev *dev)
 
 static void nvme_dev_remove(struct nvme_dev *dev)
 {
-	struct nvme_ns *ns, *next;
+	struct nvme_ns *ns;
 
-	list_for_each_entry_safe(ns, next, &dev->namespaces, list) {
-		list_del(&ns->list);
-		del_gendisk(ns->disk);
-		nvme_ns_free(ns);
+	list_for_each_entry(ns, &dev->namespaces, list) {
+		if (ns->disk->flags & GENHD_FL_UP)
+			del_gendisk(ns->disk);
+		if (!blk_queue_dying(ns->queue))
+			blk_cleanup_queue(ns->queue);
 	}
 }
 
@@ -2349,9 +2364,22 @@ 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) {
+		list_del(&ns->list);
+		put_disk(ns->disk);
+		kfree(ns);
+	}
+}
+
 static void nvme_free_dev(struct kref *kref)
 {
 	struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
+
+	nvme_free_namespaces(dev);
 	kfree(dev->queues);
 	kfree(dev->entry);
 	kfree(dev);
@@ -2525,6 +2553,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		goto release_pools;
 	}
 
+	kref_init(&dev->kref);
 	result = nvme_dev_add(dev);
 	if (result)
 		goto shutdown;
@@ -2540,11 +2569,11 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		goto remove;
 
 	dev->initialized = 1;
-	kref_init(&dev->kref);
 	return 0;
 
  remove:
 	nvme_dev_remove(dev);
+	nvme_free_namespaces(dev);
  shutdown:
 	nvme_dev_shutdown(dev);
  release_pools:
-- 
cgit v1.2.3