Merge git://git.infradead.org/users/willy/linux-nvme

Pull NVMe driver updates from Matthew Wilcox:
 "Various updates to the NVMe driver.  The most user-visible change is
  that drive hotplugging now works and CPU hotplug while an NVMe drive
  is installed should also work better"

* git://git.infradead.org/users/willy/linux-nvme:
  NVMe: Retry failed commands with non-fatal errors
  NVMe: Add getgeo to block ops
  NVMe: Start-stop nvme_thread during device add-remove.
  NVMe: Make I/O timeout a module parameter
  NVMe: CPU hot plug notification
  NVMe: per-cpu io queues
  NVMe: Replace DEFINE_PCI_DEVICE_TABLE
  NVMe: Fix divide-by-zero in nvme_trans_io_get_num_cmds
  NVMe: IOCTL path RCU protect queue access
  NVMe: RCU protected access to io queues
  NVMe: Initialize device reference count earlier
  NVMe: Add CONFIG_PM_SLEEP to suspend/resume functions

4 files changed, 504 insertions, 245 deletions

diff --git a/drivers/block/nvme-core.c b/drivers/block/nvme-core.c
index da085ff10d2..7c64fa756cc 100644
--- a/drivers/block/nvme-core.c
+++ b/drivers/block/nvme-core.c
@@ -1,6 +1,6 @@
 /*
  * NVM Express device driver
- * Copyright (c) 2011, Intel Corporation.
+ * Copyright (c) 2011-2014, Intel Corporation.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -20,10 +20,12 @@
 #include <linux/bio.h>
 #include <linux/bitops.h>
 #include <linux/blkdev.h>
+#include <linux/cpu.h>
 #include <linux/delay.h>
 #include <linux/errno.h>
 #include <linux/fs.h>
 #include <linux/genhd.h>
+#include <linux/hdreg.h>
 #include <linux/idr.h>
 #include <linux/init.h>
 #include <linux/interrupt.h>
@@ -35,6 +37,7 @@
 #include <linux/module.h>
 #include <linux/moduleparam.h>
 #include <linux/pci.h>
+#include <linux/percpu.h>
 #include <linux/poison.h>
 #include <linux/ptrace.h>
 #include <linux/sched.h>
@@ -47,6 +50,11 @@
 #define SQ_SIZE(depth)		(depth * sizeof(struct nvme_command))
 #define CQ_SIZE(depth)		(depth * sizeof(struct nvme_completion))
 #define ADMIN_TIMEOUT	(60 * HZ)
+#define IOD_TIMEOUT	(4 * NVME_IO_TIMEOUT)
+
+unsigned char io_timeout = 30;
+module_param(io_timeout, byte, 0644);
+MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
 
 static int nvme_major;
 module_param(nvme_major, int, 0);
@@ -58,6 +66,7 @@ static DEFINE_SPINLOCK(dev_list_lock);
 static LIST_HEAD(dev_list);
 static struct task_struct *nvme_thread;
 static struct workqueue_struct *nvme_workq;
+static wait_queue_head_t nvme_kthread_wait;
 
 static void nvme_reset_failed_dev(struct work_struct *ws);
 
@@ -74,6 +83,7 @@ struct async_cmd_info {
  * commands and one for I/O commands).
  */
 struct nvme_queue {
+	struct rcu_head r_head;
 	struct device *q_dmadev;
 	struct nvme_dev *dev;
 	char irqname[24];	/* nvme4294967295-65535\0 */
@@ -85,6 +95,7 @@ struct nvme_queue {
 	wait_queue_head_t sq_full;
 	wait_queue_t sq_cong_wait;
 	struct bio_list sq_cong;
+	struct list_head iod_bio;
 	u32 __iomem *q_db;
 	u16 q_depth;
 	u16 cq_vector;
@@ -95,6 +106,7 @@ struct nvme_queue {
 	u8 cq_phase;
 	u8 cqe_seen;
 	u8 q_suspended;
+	cpumask_var_t cpu_mask;
 	struct async_cmd_info cmdinfo;
 	unsigned long cmdid_data[];
 };
@@ -118,7 +130,7 @@ static inline void _nvme_check_size(void)
 	BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
 }
 
-typedef void (*nvme_completion_fn)(struct nvme_dev *, void *,
+typedef void (*nvme_completion_fn)(struct nvme_queue *, void *,
 						struct nvme_completion *);
 
 struct nvme_cmd_info {
@@ -190,7 +202,7 @@ static int alloc_cmdid_killable(struct nvme_queue *nvmeq, void *ctx,
 #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,
+static void special_completion(struct nvme_queue *nvmeq, void *ctx,
 						struct nvme_completion *cqe)
 {
 	if (ctx == CMD_CTX_CANCELLED)
@@ -198,26 +210,26 @@ static void special_completion(struct nvme_dev *dev, void *ctx,
 	if (ctx == CMD_CTX_FLUSH)
 		return;
 	if (ctx == CMD_CTX_ABORT) {
-		++dev->abort_limit;
+		++nvmeq->dev->abort_limit;
 		return;
 	}
 	if (ctx == CMD_CTX_COMPLETED) {
-		dev_warn(&dev->pci_dev->dev,
+		dev_warn(nvmeq->q_dmadev,
 				"completed id %d twice on queue %d\n",
 				cqe->command_id, le16_to_cpup(&cqe->sq_id));
 		return;
 	}
 	if (ctx == CMD_CTX_INVALID) {
-		dev_warn(&dev->pci_dev->dev,
+		dev_warn(nvmeq->q_dmadev,
 				"invalid id %d completed on queue %d\n",
 				cqe->command_id, le16_to_cpup(&cqe->sq_id));
 		return;
 	}
 
-	dev_warn(&dev->pci_dev->dev, "Unknown special completion %p\n", ctx);
+	dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx);
 }
 
-static void async_completion(struct nvme_dev *dev, void *ctx,
+static void async_completion(struct nvme_queue *nvmeq, void *ctx,
 						struct nvme_completion *cqe)
 {
 	struct async_cmd_info *cmdinfo = ctx;
@@ -262,14 +274,34 @@ static void *cancel_cmdid(struct nvme_queue *nvmeq, int cmdid,
 	return ctx;
 }
 
-struct nvme_queue *get_nvmeq(struct nvme_dev *dev)
+static struct nvme_queue *raw_nvmeq(struct nvme_dev *dev, int qid)
+{
+	return rcu_dereference_raw(dev->queues[qid]);
+}
+
+static struct nvme_queue *get_nvmeq(struct nvme_dev *dev) __acquires(RCU)
+{
+	unsigned queue_id = get_cpu_var(*dev->io_queue);
+	rcu_read_lock();
+	return rcu_dereference(dev->queues[queue_id]);
+}
+
+static void put_nvmeq(struct nvme_queue *nvmeq) __releases(RCU)
 {
-	return dev->queues[get_cpu() + 1];
+	rcu_read_unlock();
+	put_cpu_var(nvmeq->dev->io_queue);
 }
 
-void put_nvmeq(struct nvme_queue *nvmeq)
+static struct nvme_queue *lock_nvmeq(struct nvme_dev *dev, int q_idx)
+							__acquires(RCU)
 {
-	put_cpu();
+	rcu_read_lock();
+	return rcu_dereference(dev->queues[q_idx]);
+}
+
+static void unlock_nvmeq(struct nvme_queue *nvmeq) __releases(RCU)
+{
+	rcu_read_unlock();
 }
 
 /**
@@ -284,6 +316,10 @@ static int nvme_submit_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd)
 	unsigned long flags;
 	u16 tail;
 	spin_lock_irqsave(&nvmeq->q_lock, flags);
+	if (nvmeq->q_suspended) {
+		spin_unlock_irqrestore(&nvmeq->q_lock, flags);
+		return -EBUSY;
+	}
 	tail = nvmeq->sq_tail;
 	memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd));
 	if (++tail == nvmeq->q_depth)
@@ -323,6 +359,7 @@ nvme_alloc_iod(unsigned nseg, unsigned nbytes, gfp_t gfp)
 		iod->npages = -1;
 		iod->length = nbytes;
 		iod->nents = 0;
+		iod->first_dma = 0ULL;
 		iod->start_time = jiffies;
 	}
 
@@ -371,19 +408,31 @@ static void nvme_end_io_acct(struct bio *bio, unsigned long start_time)
 	part_stat_unlock();
 }
 
-static void bio_completion(struct nvme_dev *dev, void *ctx,
+static void bio_completion(struct nvme_queue *nvmeq, void *ctx,
 						struct nvme_completion *cqe)
 {
 	struct nvme_iod *iod = ctx;
 	struct bio *bio = iod->private;
 	u16 status = le16_to_cpup(&cqe->status) >> 1;
 
+	if (unlikely(status)) {
+		if (!(status & NVME_SC_DNR ||
+				bio->bi_rw & REQ_FAILFAST_MASK) &&
+				(jiffies - iod->start_time) < IOD_TIMEOUT) {
+			if (!waitqueue_active(&nvmeq->sq_full))
+				add_wait_queue(&nvmeq->sq_full,
+							&nvmeq->sq_cong_wait);
+			list_add_tail(&iod->node, &nvmeq->iod_bio);
+			wake_up(&nvmeq->sq_full);
+			return;
+		}
+	}
 	if (iod->nents) {
-		dma_unmap_sg(&dev->pci_dev->dev, iod->sg, iod->nents,
+		dma_unmap_sg(nvmeq->q_dmadev, iod->sg, iod->nents,
 			bio_data_dir(bio) ? DMA_TO_DEVICE : DMA_FROM_DEVICE);
 		nvme_end_io_acct(bio, iod->start_time);
 	}
-	nvme_free_iod(dev, iod);
+	nvme_free_iod(nvmeq->dev, iod);
 	if (status)
 		bio_endio(bio, -EIO);
 	else
@@ -391,8 +440,8 @@ static void bio_completion(struct nvme_dev *dev, void *ctx,
 }
 
 /* length is in bytes.  gfp flags indicates whether we may sleep. */
-int nvme_setup_prps(struct nvme_dev *dev, struct nvme_common_command *cmd,
-			struct nvme_iod *iod, int total_len, gfp_t gfp)
+int nvme_setup_prps(struct nvme_dev *dev, struct nvme_iod *iod, int total_len,
+								gfp_t gfp)
 {
 	struct dma_pool *pool;
 	int length = total_len;
@@ -405,7 +454,6 @@ int nvme_setup_prps(struct nvme_dev *dev, struct nvme_common_command *cmd,
 	dma_addr_t prp_dma;
 	int nprps, i;
 
-	cmd->prp1 = cpu_to_le64(dma_addr);
 	length -= (PAGE_SIZE - offset);
 	if (length <= 0)
 		return total_len;
@@ -420,7 +468,7 @@ int nvme_setup_prps(struct nvme_dev *dev, struct nvme_common_command *cmd,
 	}
 
 	if (length <= PAGE_SIZE) {
-		cmd->prp2 = cpu_to_le64(dma_addr);
+		iod->first_dma = dma_addr;
 		return total_len;
 	}
 
@@ -435,13 +483,12 @@ int nvme_setup_prps(struct nvme_dev *dev, struct nvme_common_command *cmd,
 
 	prp_list = dma_pool_alloc(pool, gfp, &prp_dma);
 	if (!prp_list) {
-		cmd->prp2 = cpu_to_le64(dma_addr);
+		iod->first_dma = dma_addr;
 		iod->npages = -1;
 		return (total_len - length) + PAGE_SIZE;
 	}
 	list[0] = prp_list;
 	iod->first_dma = prp_dma;
-	cmd->prp2 = cpu_to_le64(prp_dma);
 	i = 0;
 	for (;;) {
 		if (i == PAGE_SIZE / 8) {
@@ -480,10 +527,11 @@ static int nvme_split_and_submit(struct bio *bio, struct nvme_queue *nvmeq,
 
 	bio_chain(split, bio);
 
-	if (bio_list_empty(&nvmeq->sq_cong))
+	if (!waitqueue_active(&nvmeq->sq_full))
 		add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait);
 	bio_list_add(&nvmeq->sq_cong, split);
 	bio_list_add(&nvmeq->sq_cong, bio);
+	wake_up(&nvmeq->sq_full);
 
 	return 0;
 }
@@ -536,25 +584,13 @@ static int nvme_map_bio(struct nvme_queue *nvmeq, struct nvme_iod *iod,
 	return length;
 }
 
-/*
- * We reuse the small pool to allocate the 16-byte range here as it is not
- * worth having a special pool for these or additional cases to handle freeing
- * the iod.
- */
 static int nvme_submit_discard(struct nvme_queue *nvmeq, struct nvme_ns *ns,
 		struct bio *bio, struct nvme_iod *iod, int cmdid)
 {
-	struct nvme_dsm_range *range;
+	struct nvme_dsm_range *range =
+				(struct nvme_dsm_range *)iod_list(iod)[0];
 	struct nvme_command *cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail];
 
-	range = dma_pool_alloc(nvmeq->dev->prp_small_pool, GFP_ATOMIC,
-							&iod->first_dma);
-	if (!range)
-		return -ENOMEM;
-
-	iod_list(iod)[0] = (__le64 *)range;
-	iod->npages = 0;
-
 	range->cattr = cpu_to_le32(0);
 	range->nlb = cpu_to_le32(bio->bi_iter.bi_size >> ns->lba_shift);
 	range->slba = cpu_to_le64(nvme_block_nr(ns, bio->bi_iter.bi_sector));
@@ -601,44 +637,22 @@ int nvme_submit_flush_data(struct nvme_queue *nvmeq, struct nvme_ns *ns)
 	return nvme_submit_flush(nvmeq, ns, cmdid);
 }
 
-/*
- * Called with local interrupts disabled and the q_lock held.  May not sleep.
- */
-static int nvme_submit_bio_queue(struct nvme_queue *nvmeq, struct nvme_ns *ns,
-								struct bio *bio)
+static int nvme_submit_iod(struct nvme_queue *nvmeq, struct nvme_iod *iod)
 {
+	struct bio *bio = iod->private;
+	struct nvme_ns *ns = bio->bi_bdev->bd_disk->private_data;
 	struct nvme_command *cmnd;
-	struct nvme_iod *iod;
-	enum dma_data_direction dma_dir;
-	int cmdid, length, result;
+	int cmdid;
 	u16 control;
 	u32 dsmgmt;
-	int psegs = bio_phys_segments(ns->queue, bio);
-
-	if ((bio->bi_rw & REQ_FLUSH) && psegs) {
-		result = nvme_submit_flush_data(nvmeq, ns);
-		if (result)
-			return result;
-	}
 
-	result = -ENOMEM;
-	iod = nvme_alloc_iod(psegs, bio->bi_iter.bi_size, GFP_ATOMIC);
-	if (!iod)
-		goto nomem;
-	iod->private = bio;
-
-	result = -EBUSY;
 	cmdid = alloc_cmdid(nvmeq, iod, bio_completion, NVME_IO_TIMEOUT);
 	if (unlikely(cmdid < 0))
-		goto free_iod;
+		return cmdid;
 
-	if (bio->bi_rw & REQ_DISCARD) {
-		result = nvme_submit_discard(nvmeq, ns, bio, iod, cmdid);
-		if (result)
-			goto free_cmdid;
-		return result;
-	}
-	if ((bio->bi_rw & REQ_FLUSH) && !psegs)
+	if (bio->bi_rw & REQ_DISCARD)
+		return nvme_submit_discard(nvmeq, ns, bio, iod, cmdid);
+	if ((bio->bi_rw & REQ_FLUSH) && !iod->nents)
 		return nvme_submit_flush(nvmeq, ns, cmdid);
 
 	control = 0;
@@ -652,42 +666,85 @@ static int nvme_submit_bio_queue(struct nvme_queue *nvmeq, struct nvme_ns *ns,
 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 
 	cmnd = &nvmeq->sq_cmds[nvmeq->sq_tail];
-
 	memset(cmnd, 0, sizeof(*cmnd));
-	if (bio_data_dir(bio)) {
-		cmnd->rw.opcode = nvme_cmd_write;
-		dma_dir = DMA_TO_DEVICE;
-	} else {
-		cmnd->rw.opcode = nvme_cmd_read;
-		dma_dir = DMA_FROM_DEVICE;
-	}
-
-	result = nvme_map_bio(nvmeq, iod, bio, dma_dir, psegs);
-	if (result <= 0)
-		goto free_cmdid;
-	length = result;
 
+	cmnd->rw.opcode = bio_data_dir(bio) ? nvme_cmd_write : nvme_cmd_read;
 	cmnd->rw.command_id = cmdid;
 	cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
-	length = nvme_setup_prps(nvmeq->dev, &cmnd->common, iod, length,
-								GFP_ATOMIC);
+	cmnd->rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
+	cmnd->rw.prp2 = cpu_to_le64(iod->first_dma);
 	cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, bio->bi_iter.bi_sector));
-	cmnd->rw.length = cpu_to_le16((length >> ns->lba_shift) - 1);
+	cmnd->rw.length =
+		cpu_to_le16((bio->bi_iter.bi_size >> ns->lba_shift) - 1);
 	cmnd->rw.control = cpu_to_le16(control);
 	cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 
-	nvme_start_io_acct(bio);
 	if (++nvmeq->sq_tail == nvmeq->q_depth)
 		nvmeq->sq_tail = 0;
 	writel(nvmeq->sq_tail, nvmeq->q_db);
 
 	return 0;
+}
+
+/*
+ * Called with local interrupts disabled and the q_lock held.  May not sleep.
+ */
+static int nvme_submit_bio_queue(struct nvme_queue *nvmeq, struct nvme_ns *ns,
+								struct bio *bio)
+{
+	struct nvme_iod *iod;
+	int psegs = bio_phys_segments(ns->queue, bio);
+	int result;
+
+	if ((bio->bi_rw & REQ_FLUSH) && psegs) {
+		result = nvme_submit_flush_data(nvmeq, ns);
+		if (result)
+			return result;
+	}
+
+	iod = nvme_alloc_iod(psegs, bio->bi_iter.bi_size, GFP_ATOMIC);
+	if (!iod)
+		return -ENOMEM;
+
+	iod->private = bio;
+	if (bio->bi_rw & REQ_DISCARD) {
+		void *range;
+		/*
+		 * We reuse the small pool to allocate the 16-byte range here
+		 * as it is not worth having a special pool for these or
+		 * additional cases to handle freeing the iod.
+		 */
+		range = dma_pool_alloc(nvmeq->dev->prp_small_pool,
+						GFP_ATOMIC,
+						&iod->first_dma);
+		if (!range) {
+			result = -ENOMEM;
+			goto free_iod;
+		}
+		iod_list(iod)[0] = (__le64 *)range;
+		iod->npages = 0;
+	} else if (psegs) {
+		result = nvme_map_bio(nvmeq, iod, bio,
+			bio_data_dir(bio) ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
+			psegs);
+		if (result <= 0)
+			goto free_iod;
+		if (nvme_setup_prps(nvmeq->dev, iod, result, GFP_ATOMIC) !=
+								result) {
+			result = -ENOMEM;
+			goto free_iod;
+		}
+		nvme_start_io_acct(bio);
+	}
+	if (unlikely(nvme_submit_iod(nvmeq, iod))) {
+		if (!waitqueue_active(&nvmeq->sq_full))
+			add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait);
+		list_add_tail(&iod->node, &nvmeq->iod_bio);
+	}
+	return 0;
 
- free_cmdid:
-	free_cmdid(nvmeq, cmdid, NULL);
  free_iod:
 	nvme_free_iod(nvmeq->dev, iod);
- nomem:
 	return result;
 }
 
@@ -711,7 +768,7 @@ static int nvme_process_cq(struct nvme_queue *nvmeq)
 		}
 
 		ctx = free_cmdid(nvmeq, cqe.command_id, &fn);
-		fn(nvmeq->dev, ctx, &cqe);
+		fn(nvmeq, ctx, &cqe);
 	}
 
 	/* If the controller ignores the cq head doorbell and continuously
@@ -747,7 +804,7 @@ static void nvme_make_request(struct request_queue *q, struct bio *bio)
 	if (!nvmeq->q_suspended && bio_list_empty(&nvmeq->sq_cong))
 		result = nvme_submit_bio_queue(nvmeq, ns, bio);
 	if (unlikely(result)) {
-		if (bio_list_empty(&nvmeq->sq_cong))
+		if (!waitqueue_active(&nvmeq->sq_full))
 			add_wait_queue(&nvmeq->sq_full, &nvmeq->sq_cong_wait);
 		bio_list_add(&nvmeq->sq_cong, bio);
 	}
@@ -791,7 +848,7 @@ struct sync_cmd_info {
 	int status;
 };
 
-static void sync_completion(struct nvme_dev *dev, void *ctx,
+static void sync_completion(struct nvme_queue *nvmeq, void *ctx,
 						struct nvme_completion *cqe)
 {
 	struct sync_cmd_info *cmdinfo = ctx;
@@ -804,27 +861,46 @@ static void sync_completion(struct nvme_dev *dev, void *ctx,
  * Returns 0 on success.  If the result is negative, it's a Linux error code;
  * if the result is positive, it's an NVM Express status code
  */
-int nvme_submit_sync_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd,
+static int nvme_submit_sync_cmd(struct nvme_dev *dev, int q_idx,
+						struct nvme_command *cmd,
 						u32 *result, unsigned timeout)
 {
-	int cmdid;
+	int cmdid, ret;
 	struct sync_cmd_info cmdinfo;
+	struct nvme_queue *nvmeq;
+
+	nvmeq = lock_nvmeq(dev, q_idx);
+	if (!nvmeq) {
+		unlock_nvmeq(nvmeq);
+		return -ENODEV;
+	}
 
 	cmdinfo.task = current;
 	cmdinfo.status = -EINTR;
 
-	cmdid = alloc_cmdid_killable(nvmeq, &cmdinfo, sync_completion,
-								timeout);
-	if (cmdid < 0)
+	cmdid = alloc_cmdid(nvmeq, &cmdinfo, sync_completion, timeout);
+	if (cmdid < 0) {
+		unlock_nvmeq(nvmeq);
 		return cmdid;
+	}
 	cmd->common.command_id = cmdid;
 
 	set_current_state(TASK_KILLABLE);
-	nvme_submit_cmd(nvmeq, cmd);
+	ret = nvme_submit_cmd(nvmeq, cmd);
+	if (ret) {
+		free_cmdid(nvmeq, cmdid, NULL);
+		unlock_nvmeq(nvmeq);
+		set_current_state(TASK_RUNNING);
+		return ret;
+	}
+	unlock_nvmeq(nvmeq);
 	schedule_timeout(timeout);
 
 	if (cmdinfo.status == -EINTR) {
-		nvme_abort_command(nvmeq, cmdid);
+		nvmeq = lock_nvmeq(dev, q_idx);
+		if (nvmeq)
+			nvme_abort_command(nvmeq, cmdid);
+		unlock_nvmeq(nvmeq);
 		return -EINTR;
 	}
 
@@ -845,20 +921,26 @@ static int nvme_submit_async_cmd(struct nvme_queue *nvmeq,
 		return cmdid;
 	cmdinfo->status = -EINTR;
 	cmd->common.command_id = cmdid;
-	nvme_submit_cmd(nvmeq, cmd);
-	return 0;
+	return nvme_submit_cmd(nvmeq, cmd);
 }
 
 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);
+	return nvme_submit_sync_cmd(dev, 0, cmd, result, ADMIN_TIMEOUT);
+}
+
+int nvme_submit_io_cmd(struct nvme_dev *dev, struct nvme_command *cmd,
+								u32 *result)
+{
+	return nvme_submit_sync_cmd(dev, smp_processor_id() + 1, cmd, result,
+							NVME_IO_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,
+	return nvme_submit_async_cmd(raw_nvmeq(dev, 0), cmd, cmdinfo,
 								ADMIN_TIMEOUT);
 }
 
@@ -985,6 +1067,7 @@ static void nvme_abort_cmd(int cmdid, struct nvme_queue *nvmeq)
 	struct nvme_command cmd;
 	struct nvme_dev *dev = nvmeq->dev;
 	struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
+	struct nvme_queue *adminq;
 
 	if (!nvmeq->qid || info[cmdid].aborted) {
 		if (work_busy(&dev->reset_work))
@@ -1001,7 +1084,8 @@ static void nvme_abort_cmd(int cmdid, struct nvme_queue *nvmeq)
 	if (!dev->abort_limit)
 		return;
 
-	a_cmdid = alloc_cmdid(dev->queues[0], CMD_CTX_ABORT, special_completion,
+	adminq = rcu_dereference(dev->queues[0]);
+	a_cmdid = alloc_cmdid(adminq, CMD_CTX_ABORT, special_completion,
 								ADMIN_TIMEOUT);
 	if (a_cmdid < 0)
 		return;
@@ -1018,7 +1102,7 @@ static void nvme_abort_cmd(int cmdid, struct nvme_queue *nvmeq)
 
 	dev_warn(nvmeq->q_dmadev, "Aborting I/O %d QID %d\n", cmdid,
 							nvmeq->qid);
-	nvme_submit_cmd(dev->queues[0], &cmd);
+	nvme_submit_cmd(adminq, &cmd);
 }
 
 /**
@@ -1051,23 +1135,38 @@ static void nvme_cancel_ios(struct nvme_queue *nvmeq, bool timeout)
 		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);
+		fn(nvmeq, ctx, &cqe);
 	}
 }
 
-static void nvme_free_queue(struct nvme_queue *nvmeq)
+static void nvme_free_queue(struct rcu_head *r)
 {
+	struct nvme_queue *nvmeq = container_of(r, struct nvme_queue, r_head);
+
 	spin_lock_irq(&nvmeq->q_lock);
 	while (bio_list_peek(&nvmeq->sq_cong)) {
 		struct bio *bio = bio_list_pop(&nvmeq->sq_cong);
 		bio_endio(bio, -EIO);
 	}
+	while (!list_empty(&nvmeq->iod_bio)) {
+		static struct nvme_completion cqe = {
+			.status = cpu_to_le16(
+				(NVME_SC_ABORT_REQ | NVME_SC_DNR) << 1),
+		};
+		struct nvme_iod *iod = list_first_entry(&nvmeq->iod_bio,
+							struct nvme_iod,
+							node);
+		list_del(&iod->node);
+		bio_completion(nvmeq, iod, &cqe);
+	}
 	spin_unlock_irq(&nvmeq->q_lock);
 
 	dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth),
 				(void *)nvmeq->cqes, nvmeq->cq_dma_addr);
 	dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth),
 					nvmeq->sq_cmds, nvmeq->sq_dma_addr);
+	if (nvmeq->qid)
+		free_cpumask_var(nvmeq->cpu_mask);
 	kfree(nvmeq);
 }
 
@@ -1076,9 +1175,10 @@ static void nvme_free_queues(struct nvme_dev *dev, int lowest)
 	int i;
 
 	for (i = dev->queue_count - 1; i >= lowest; i--) {
-		nvme_free_queue(dev->queues[i]);
+		struct nvme_queue *nvmeq = raw_nvmeq(dev, i);
+		rcu_assign_pointer(dev->queues[i], NULL);
+		call_rcu(&nvmeq->r_head, nvme_free_queue);
 		dev->queue_count--;
-		dev->queues[i] = NULL;
 	}
 }
 
@@ -1098,6 +1198,7 @@ static int nvme_suspend_queue(struct nvme_queue *nvmeq)
 		return 1;
 	}
 	nvmeq->q_suspended = 1;
+	nvmeq->dev->online_queues--;
 	spin_unlock_irq(&nvmeq->q_lock);
 
 	irq_set_affinity_hint(vector, NULL);
@@ -1116,7 +1217,7 @@ static void nvme_clear_queue(struct nvme_queue *nvmeq)
 
 static void nvme_disable_queue(struct nvme_dev *dev, int qid)
 {
-	struct nvme_queue *nvmeq = dev->queues[qid];
+	struct nvme_queue *nvmeq = raw_nvmeq(dev, qid);
 
 	if (!nvmeq)
 		return;
@@ -1152,6 +1253,9 @@ static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid,
 	if (!nvmeq->sq_cmds)
 		goto free_cqdma;
 
+	if (qid && !zalloc_cpumask_var(&nvmeq->cpu_mask, GFP_KERNEL))
+		goto free_sqdma;
+
 	nvmeq->q_dmadev = dmadev;
 	nvmeq->dev = dev;
 	snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d",
@@ -1162,15 +1266,20 @@ 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);
+	INIT_LIST_HEAD(&nvmeq->iod_bio);
 	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++;
+	rcu_assign_pointer(dev->queues[qid], nvmeq);
 
 	return nvmeq;
 
+ free_sqdma:
+	dma_free_coherent(dmadev, SQ_SIZE(depth), (void *)nvmeq->sq_cmds,
+							nvmeq->sq_dma_addr);
  free_cqdma:
 	dma_free_coherent(dmadev, CQ_SIZE(depth), (void *)nvmeq->cqes,
 							nvmeq->cq_dma_addr);
@@ -1203,6 +1312,7 @@ static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid)
 	memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth));
 	nvme_cancel_ios(nvmeq, false);
 	nvmeq->q_suspended = 0;
+	dev->online_queues++;
 }
 
 static int nvme_create_queue(struct nvme_queue *nvmeq, int qid)
@@ -1311,12 +1421,11 @@ static int nvme_configure_admin_queue(struct nvme_dev *dev)
 	if (result < 0)
 		return result;
 
-	nvmeq = dev->queues[0];
+	nvmeq = raw_nvmeq(dev, 0);
 	if (!nvmeq) {
 		nvmeq = nvme_alloc_queue(dev, 0, 64, 0);
 		if (!nvmeq)
 			return -ENOMEM;
-		dev->queues[0] = nvmeq;
 	}
 
 	aqa = nvmeq->q_depth - 1;
@@ -1418,7 +1527,6 @@ void nvme_unmap_user_pages(struct nvme_dev *dev, int write,
 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 {
 	struct nvme_dev *dev = ns->dev;
-	struct nvme_queue *nvmeq;
 	struct nvme_user_io io;
 	struct nvme_command c;
 	unsigned length, meta_len;
@@ -1492,22 +1600,14 @@ static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 		c.rw.metadata = cpu_to_le64(meta_dma_addr);
 	}
 
-	length = nvme_setup_prps(dev, &c.common, iod, length, GFP_KERNEL);
+	length = nvme_setup_prps(dev, iod, length, GFP_KERNEL);
+	c.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
+	c.rw.prp2 = cpu_to_le64(iod->first_dma);
 
-	nvmeq = get_nvmeq(dev);
-	/*
-	 * Since nvme_submit_sync_cmd sleeps, we can't keep preemption
-	 * disabled.  We may be preempted at any point, and be rescheduled
-	 * to a different CPU.  That will cause cacheline bouncing, but no
-	 * additional races since q_lock already protects against other CPUs.
-	 */
-	put_nvmeq(nvmeq);
 	if (length != (io.nblocks + 1) << ns->lba_shift)
 		status = -ENOMEM;
-	else if (!nvmeq || nvmeq->q_suspended)
-		status = -EBUSY;
 	else
-		status = nvme_submit_sync_cmd(nvmeq, &c, NULL, NVME_IO_TIMEOUT);
+		status = nvme_submit_io_cmd(dev, &c, NULL);
 
 	if (meta_len) {
 		if (status == NVME_SC_SUCCESS && !(io.opcode & 1)) {
@@ -1572,8 +1672,9 @@ static int nvme_user_admin_cmd(struct nvme_dev *dev,
 								length);
 		if (IS_ERR(iod))
 			return PTR_ERR(iod);
-		length = nvme_setup_prps(dev, &c.common, iod, length,
-								GFP_KERNEL);
+		length = nvme_setup_prps(dev, iod, length, GFP_KERNEL);
+		c.common.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
+		c.common.prp2 = cpu_to_le64(iod->first_dma);
 	}
 
 	timeout = cmd.timeout_ms ? msecs_to_jiffies(cmd.timeout_ms) :
@@ -1581,8 +1682,7 @@ static int nvme_user_admin_cmd(struct nvme_dev *dev,
 	if (length != cmd.data_len)
 		status = -ENOMEM;
 	else
-		status = nvme_submit_sync_cmd(dev->queues[0], &c, &cmd.result,
-								timeout);
+		status = nvme_submit_sync_cmd(dev, 0, &c, &cmd.result, timeout);
 
 	if (cmd.data_len) {
 		nvme_unmap_user_pages(dev, cmd.opcode & 1, iod);
@@ -1653,25 +1753,51 @@ static void nvme_release(struct gendisk *disk, fmode_t mode)
 	kref_put(&dev->kref, nvme_free_dev);
 }
 
+static int nvme_getgeo(struct block_device *bd, struct hd_geometry *geo)
+{
+	/* some standard values */
+	geo->heads = 1 << 6;
+	geo->sectors = 1 << 5;
+	geo->cylinders = get_capacity(bd->bd_disk) >> 11;
+	return 0;
+}
+
 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,
+	.getgeo		= nvme_getgeo,
 };
 
+static void nvme_resubmit_iods(struct nvme_queue *nvmeq)
+{
+	struct nvme_iod *iod, *next;
+
+	list_for_each_entry_safe(iod, next, &nvmeq->iod_bio, node) {
+		if (unlikely(nvme_submit_iod(nvmeq, iod)))
+			break;
+		list_del(&iod->node);
+		if (bio_list_empty(&nvmeq->sq_cong) &&
+						list_empty(&nvmeq->iod_bio))
+			remove_wait_queue(&nvmeq->sq_full,
+						&nvmeq->sq_cong_wait);
+	}
+}
+
 static void nvme_resubmit_bios(struct nvme_queue *nvmeq)
 {
 	while (bio_list_peek(&nvmeq->sq_cong)) {
 		struct bio *bio = bio_list_pop(&nvmeq->sq_cong);
 		struct nvme_ns *ns = bio->bi_bdev->bd_disk->private_data;
 
-		if (bio_list_empty(&nvmeq->sq_cong))
+		if (bio_list_empty(&nvmeq->sq_cong) &&
+						list_empty(&nvmeq->iod_bio))
 			remove_wait_queue(&nvmeq->sq_full,
 							&nvmeq->sq_cong_wait);
 		if (nvme_submit_bio_queue(nvmeq, ns, bio)) {
-			if (bio_list_empty(&nvmeq->sq_cong))
+			if (!waitqueue_active(&nvmeq->sq_full))
 				add_wait_queue(&nvmeq->sq_full,
 							&nvmeq->sq_cong_wait);
 			bio_list_add_head(&nvmeq->sq_cong, bio);
@@ -1700,8 +1826,10 @@ static int nvme_kthread(void *data)
 				queue_work(nvme_workq, &dev->reset_work);
 				continue;
 			}
+			rcu_read_lock();
 			for (i = 0; i < dev->queue_count; i++) {
-				struct nvme_queue *nvmeq = dev->queues[i];
+				struct nvme_queue *nvmeq =
+						rcu_dereference(dev->queues[i]);
 				if (!nvmeq)
 					continue;
 				spin_lock_irq(&nvmeq->q_lock);
@@ -1710,9 +1838,11 @@ static int nvme_kthread(void *data)
 				nvme_process_cq(nvmeq);
 				nvme_cancel_ios(nvmeq, true);
 				nvme_resubmit_bios(nvmeq);
+				nvme_resubmit_iods(nvmeq);
  unlock:
 				spin_unlock_irq(&nvmeq->q_lock);
 			}
+			rcu_read_unlock();
 		}
 		spin_unlock(&dev_list_lock);
 		schedule_timeout(round_jiffies_relative(HZ));
@@ -1787,6 +1917,143 @@ static struct nvme_ns *nvme_alloc_ns(struct nvme_dev *dev, unsigned nsid,
 	return NULL;
 }
 
+static int nvme_find_closest_node(int node)
+{
+	int n, val, min_val = INT_MAX, best_node = node;
+
+	for_each_online_node(n) {
+		if (n == node)
+			continue;
+		val = node_distance(node, n);
+		if (val < min_val) {
+			min_val = val;
+			best_node = n;
+		}
+	}
+	return best_node;
+}
+
+static void nvme_set_queue_cpus(cpumask_t *qmask, struct nvme_queue *nvmeq,
+								int count)
+{
+	int cpu;
+	for_each_cpu(cpu, qmask) {
+		if (cpumask_weight(nvmeq->cpu_mask) >= count)
+			break;
+		if (!cpumask_test_and_set_cpu(cpu, nvmeq->cpu_mask))
+			*per_cpu_ptr(nvmeq->dev->io_queue, cpu) = nvmeq->qid;
+	}
+}
+
+static void nvme_add_cpus(cpumask_t *mask, const cpumask_t *unassigned_cpus,
+	const cpumask_t *new_mask, struct nvme_queue *nvmeq, int cpus_per_queue)
+{
+	int next_cpu;
+	for_each_cpu(next_cpu, new_mask) {
+		cpumask_or(mask, mask, get_cpu_mask(next_cpu));
+		cpumask_or(mask, mask, topology_thread_cpumask(next_cpu));
+		cpumask_and(mask, mask, unassigned_cpus);
+		nvme_set_queue_cpus(mask, nvmeq, cpus_per_queue);
+	}
+}
+
+static void nvme_create_io_queues(struct nvme_dev *dev)
+{
+	unsigned i, max;
+
+	max = min(dev->max_qid, num_online_cpus());
+	for (i = dev->queue_count; i <= max; i++)
+		if (!nvme_alloc_queue(dev, i, dev->q_depth, i - 1))
+			break;
+
+	max = min(dev->queue_count - 1, num_online_cpus());
+	for (i = dev->online_queues; i <= max; i++)
+		if (nvme_create_queue(raw_nvmeq(dev, i), i))
+			break;
+}
+
+/*
+ * If there are fewer queues than online cpus, this will try to optimally
+ * assign a queue to multiple cpus by grouping cpus that are "close" together:
+ * thread siblings, core, socket, closest node, then whatever else is
+ * available.
+ */
+static void nvme_assign_io_queues(struct nvme_dev *dev)
+{
+	unsigned cpu, cpus_per_queue, queues, remainder, i;
+	cpumask_var_t unassigned_cpus;
+
+	nvme_create_io_queues(dev);
+
+	queues = min(dev->online_queues - 1, num_online_cpus());
+	if (!queues)
+		return;
+
+	cpus_per_queue = num_online_cpus() / queues;
+	remainder = queues - (num_online_cpus() - queues * cpus_per_queue);
+
+	if (!alloc_cpumask_var(&unassigned_cpus, GFP_KERNEL))
+		return;
+
+	cpumask_copy(unassigned_cpus, cpu_online_mask);
+	cpu = cpumask_first(unassigned_cpus);
+	for (i = 1; i <= queues; i++) {
+		struct nvme_queue *nvmeq = lock_nvmeq(dev, i);
+		cpumask_t mask;
+
+		cpumask_clear(nvmeq->cpu_mask);
+		if (!cpumask_weight(unassigned_cpus)) {
+			unlock_nvmeq(nvmeq);
+			break;
+		}
+
+		mask = *get_cpu_mask(cpu);
+		nvme_set_queue_cpus(&mask, nvmeq, cpus_per_queue);
+		if (cpus_weight(mask) < cpus_per_queue)
+			nvme_add_cpus(&mask, unassigned_cpus,
+				topology_thread_cpumask(cpu),
+				nvmeq, cpus_per_queue);
+		if (cpus_weight(mask) < cpus_per_queue)
+			nvme_add_cpus(&mask, unassigned_cpus,
+				topology_core_cpumask(cpu),
+				nvmeq, cpus_per_queue);
+		if (cpus_weight(mask) < cpus_per_queue)
+			nvme_add_cpus(&mask, unassigned_cpus,
+				cpumask_of_node(cpu_to_node(cpu)),
+				nvmeq, cpus_per_queue);
+		if (cpus_weight(mask) < cpus_per_queue)
+			nvme_add_cpus(&mask, unassigned_cpus,
+				cpumask_of_node(
+					nvme_find_closest_node(
+						cpu_to_node(cpu))),
+				nvmeq, cpus_per_queue);
+		if (cpus_weight(mask) < cpus_per_queue)
+			nvme_add_cpus(&mask, unassigned_cpus,
+				unassigned_cpus,
+				nvmeq, cpus_per_queue);
+
+		WARN(cpumask_weight(nvmeq->cpu_mask) != cpus_per_queue,
+			"nvme%d qid:%d mis-matched queue-to-cpu assignment\n",
+			dev->instance, i);
+
+		irq_set_affinity_hint(dev->entry[nvmeq->cq_vector].vector,
+							nvmeq->cpu_mask);
+		cpumask_andnot(unassigned_cpus, unassigned_cpus,
+						nvmeq->cpu_mask);
+		cpu = cpumask_next(cpu, unassigned_cpus);
+		if (remainder && !--remainder)
+			cpus_per_queue++;
+		unlock_nvmeq(nvmeq);
+	}
+	WARN(cpumask_weight(unassigned_cpus), "nvme%d unassigned online cpus\n",
+								dev->instance);
+	i = 0;
+	cpumask_andnot(unassigned_cpus, cpu_possible_mask, cpu_online_mask);
+	for_each_cpu(cpu, unassigned_cpus)
+		*per_cpu_ptr(dev->io_queue, cpu) = (i++ % queues) + 1;
+	free_cpumask_var(unassigned_cpus);
+}
+
 static int set_queue_count(struct nvme_dev *dev, int count)
 {
 	int status;
@@ -1805,13 +2072,26 @@ static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues)
 	return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride);
 }
 
+static int nvme_cpu_notify(struct notifier_block *self,
+				unsigned long action, void *hcpu)
+{
+	struct nvme_dev *dev = container_of(self, struct nvme_dev, nb);
+	switch (action) {
+	case CPU_ONLINE:
+	case CPU_DEAD:
+		nvme_assign_io_queues(dev);
+		break;
+	}
+	return NOTIFY_OK;
+}
+
 static int nvme_setup_io_queues(struct nvme_dev *dev)
 {
-	struct nvme_queue *adminq = dev->queues[0];
+	struct nvme_queue *adminq = raw_nvmeq(dev, 0);
 	struct pci_dev *pdev = dev->pci_dev;
-	int result, cpu, i, vecs, nr_io_queues, size, q_depth;
+	int result, i, vecs, nr_io_queues, size;
 
-	nr_io_queues = num_online_cpus();
+	nr_io_queues = num_possible_cpus();
 	result = set_queue_count(dev, nr_io_queues);
 	if (result < 0)
 		return result;
@@ -1830,7 +2110,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 			size = db_bar_size(dev, nr_io_queues);
 		} while (1);
 		dev->dbs = ((void __iomem *)dev->bar) + 4096;
-		dev->queues[0]->q_db = dev->dbs;
+		adminq->q_db = dev->dbs;
 	}
 
 	/* Deregister the admin queue's interrupt */
@@ -1856,6 +2136,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	 * number of interrupts.
 	 */
 	nr_io_queues = vecs;
+	dev->max_qid = nr_io_queues;
 
 	result = queue_request_irq(dev, adminq, adminq->irqname);
 	if (result) {
@@ -1864,49 +2145,13 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	}
 
 	/* Free previously allocated queues that are no longer usable */
-	spin_lock(&dev_list_lock);
-	for (i = dev->queue_count - 1; i > nr_io_queues; i--) {
-		struct nvme_queue *nvmeq = dev->queues[i];
-
-		spin_lock_irq(&nvmeq->q_lock);
-		nvme_cancel_ios(nvmeq, false);
-		spin_unlock_irq(&nvmeq->q_lock);
-
-		nvme_free_queue(nvmeq);
-		dev->queue_count--;
-		dev->queues[i] = NULL;
-	}
-	spin_unlock(&dev_list_lock);
-
-	cpu = cpumask_first(cpu_online_mask);
-	for (i = 0; i < nr_io_queues; i++) {
-		irq_set_affinity_hint(dev->entry[i].vector, get_cpu_mask(cpu));
-		cpu = cpumask_next(cpu, cpu_online_mask);
-	}
-
-	q_depth = min_t(int, NVME_CAP_MQES(readq(&dev->bar->cap)) + 1,
-								NVME_Q_DEPTH);
-	for (i = dev->queue_count - 1; i < nr_io_queues; i++) {
-		dev->queues[i + 1] = nvme_alloc_queue(dev, i + 1, q_depth, i);
-		if (!dev->queues[i + 1]) {
-			result = -ENOMEM;
-			goto free_queues;
-		}
-	}
-
-	for (; i < num_possible_cpus(); i++) {
-		int target = i % rounddown_pow_of_two(dev->queue_count - 1);
-		dev->queues[i + 1] = dev->queues[target + 1];
-	}
+	nvme_free_queues(dev, nr_io_queues + 1);
+	nvme_assign_io_queues(dev);
 
-	for (i = 1; i < dev->queue_count; i++) {
-		result = nvme_create_queue(dev->queues[i], i);
-		if (result) {
-			for (--i; i > 0; i--)
-				nvme_disable_queue(dev, i);
-			goto free_queues;
-		}
-	}
+	dev->nb.notifier_call = &nvme_cpu_notify;
+	result = register_hotcpu_notifier(&dev->nb);
+	if (result)
+		goto free_queues;
 
 	return 0;
 
@@ -1985,6 +2230,7 @@ static int nvme_dev_add(struct nvme_dev *dev)
 
 static int nvme_dev_map(struct nvme_dev *dev)
 {
+	u64 cap;
 	int bars, result = -ENOMEM;
 	struct pci_dev *pdev = dev->pci_dev;
 
@@ -2008,7 +2254,9 @@ static int nvme_dev_map(struct nvme_dev *dev)
 		result = -ENODEV;
 		goto unmap;
 	}
-	dev->db_stride = 1 << NVME_CAP_STRIDE(readq(&dev->bar->cap));
+	cap = readq(&dev->bar->cap);
+	dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH);
+	dev->db_stride = 1 << NVME_CAP_STRIDE(cap);
 	dev->dbs = ((void __iomem *)dev->bar) + 4096;
 
 	return 0;
@@ -2164,7 +2412,7 @@ static void nvme_disable_io_queues(struct nvme_dev *dev)
 	atomic_set(&dq.refcount, 0);
 	dq.worker = &worker;
 	for (i = dev->queue_count - 1; i > 0; i--) {
-		struct nvme_queue *nvmeq = dev->queues[i];
+		struct nvme_queue *nvmeq = raw_nvmeq(dev, i);
 
 		if (nvme_suspend_queue(nvmeq))
 			continue;
@@ -2177,19 +2425,38 @@ static void nvme_disable_io_queues(struct nvme_dev *dev)
 	kthread_stop(kworker_task);
 }
 
+/*
+* 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_shutdown(struct nvme_dev *dev)
 {
 	int i;
 
 	dev->initialized = 0;
+	unregister_hotcpu_notifier(&dev->nb);
 
-	spin_lock(&dev_list_lock);
-	list_del_init(&dev->node);
-	spin_unlock(&dev_list_lock);
+	nvme_dev_list_remove(dev);
 
 	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];
+			struct nvme_queue *nvmeq = raw_nvmeq(dev, i);
 			nvme_suspend_queue(nvmeq);
 			nvme_clear_queue(nvmeq);
 		}
@@ -2282,6 +2549,7 @@ static void nvme_free_dev(struct kref *kref)
 	struct nvme_dev *dev = container_of(kref, struct nvme_dev, kref);
 
 	nvme_free_namespaces(dev);
+	free_percpu(dev->io_queue);
 	kfree(dev->queues);
 	kfree(dev->entry);
 	kfree(dev);
@@ -2325,6 +2593,7 @@ static const struct file_operations nvme_dev_fops = {
 static int nvme_dev_start(struct nvme_dev *dev)
 {
 	int result;
+	bool start_thread = false;
 
 	result = nvme_dev_map(dev);
 	if (result)
@@ -2335,9 +2604,24 @@ static int nvme_dev_start(struct nvme_dev *dev)
 		goto unmap;
 
 	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(&nvme_kthread_wait);
+	} else
+		wait_event_killable(nvme_kthread_wait, nvme_thread);
+
+	if (IS_ERR_OR_NULL(nvme_thread)) {
+		result = nvme_thread ? PTR_ERR(nvme_thread) : -EINTR;
+		goto disable;
+	}
+
 	result = nvme_setup_io_queues(dev);
 	if (result && result != -EBUSY)
 		goto disable;
@@ -2346,9 +2630,7 @@ static int nvme_dev_start(struct nvme_dev *dev)
 
  disable:
 	nvme_disable_queue(dev, 0);
-	spin_lock(&dev_list_lock);
-	list_del_init(&dev->node);
-	spin_unlock(&dev_list_lock);
+	nvme_dev_list_remove(dev);
  unmap:
 	nvme_dev_unmap(dev);
 	return result;
@@ -2367,18 +2649,10 @@ static int nvme_remove_dead_ctrl(void *arg)
 
 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);
+	nvme_free_queues(dev, 1);
 }
 
 static int nvme_dev_resume(struct nvme_dev *dev)
@@ -2441,6 +2715,9 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 								GFP_KERNEL);
 	if (!dev->queues)
 		goto free;
+	dev->io_queue = alloc_percpu(unsigned short);
+	if (!dev->io_queue)
+		goto free;
 
 	INIT_LIST_HEAD(&dev->namespaces);
 	dev->reset_workfn = nvme_reset_failed_dev;
@@ -2455,6 +2732,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (result)
 		goto release;
 
+	kref_init(&dev->kref);
 	result = nvme_dev_start(dev);
 	if (result) {
 		if (result == -EBUSY)
@@ -2462,7 +2740,6 @@ 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;
@@ -2491,6 +2768,7 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
  release:
 	nvme_release_instance(dev);
  free:
+	free_percpu(dev->io_queue);
 	kfree(dev->queues);
 	kfree(dev->entry);
 	kfree(dev);
@@ -2517,6 +2795,7 @@ static void nvme_remove(struct pci_dev *pdev)
 	nvme_dev_remove(dev);
 	nvme_dev_shutdown(dev);
 	nvme_free_queues(dev, 0);
+	rcu_barrier();
 	nvme_release_instance(dev);
 	nvme_release_prp_pools(dev);
 	kref_put(&dev->kref, nvme_free_dev);
@@ -2529,6 +2808,7 @@ static void nvme_remove(struct pci_dev *pdev)
 #define nvme_slot_reset NULL
 #define nvme_error_resume NULL
 
+#ifdef CONFIG_PM_SLEEP
 static int nvme_suspend(struct device *dev)
 {
 	struct pci_dev *pdev = to_pci_dev(dev);
@@ -2549,6 +2829,7 @@ static int nvme_resume(struct device *dev)
 	}
 	return 0;
 }
+#endif
 
 static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume);
 
@@ -2563,7 +2844,7 @@ static const struct pci_error_handlers nvme_err_handler = {
 /* Move to pci_ids.h later */
 #define PCI_CLASS_STORAGE_EXPRESS	0x010802
 
-static DEFINE_PCI_DEVICE_TABLE(nvme_id_table) = {
+static const struct pci_device_id nvme_id_table[] = {
 	{ PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) },
 	{ 0, }
 };
@@ -2585,14 +2866,11 @@ static int __init nvme_init(void)
 {
 	int result;
 
-	nvme_thread = kthread_run(nvme_kthread, NULL, "nvme");
-	if (IS_ERR(nvme_thread))
-		return PTR_ERR(nvme_thread);
+	init_waitqueue_head(&nvme_kthread_wait);
 
-	result = -ENOMEM;
 	nvme_workq = create_singlethread_workqueue("nvme");
 	if (!nvme_workq)
-		goto kill_kthread;
+		return -ENOMEM;
 
 	result = register_blkdev(nvme_major, "nvme");
 	if (result < 0)
@@ -2609,8 +2887,6 @@ static int __init nvme_init(void)
 	unregister_blkdev(nvme_major, "nvme");
  kill_workq:
 	destroy_workqueue(nvme_workq);
- kill_kthread:
-	kthread_stop(nvme_thread);
 	return result;
 }
 
@@ -2619,11 +2895,11 @@ static void __exit nvme_exit(void)
 	pci_unregister_driver(&nvme_driver);
 	unregister_blkdev(nvme_major, "nvme");
 	destroy_workqueue(nvme_workq);
-	kthread_stop(nvme_thread);
+	BUG_ON(nvme_thread && !IS_ERR(nvme_thread));
 }
 
 MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>");
 MODULE_LICENSE("GPL");
-MODULE_VERSION("0.8");
+MODULE_VERSION("0.9");
 module_init(nvme_init);
 module_exit(nvme_exit);
diff --git a/drivers/block/nvme-scsi.c b/drivers/block/nvme-scsi.c
index 4a0ceb64e26..2c3f5be06da 100644
--- a/drivers/block/nvme-scsi.c
+++ b/drivers/block/nvme-scsi.c
@@ -1562,13 +1562,14 @@ static int nvme_trans_send_fw_cmd(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 			res = PTR_ERR(iod);
 			goto out;
 		}
-		length = nvme_setup_prps(dev, &c.common, iod, tot_len,
-								GFP_KERNEL);
+		length = nvme_setup_prps(dev, iod, tot_len, GFP_KERNEL);
 		if (length != tot_len) {
 			res = -ENOMEM;
 			goto out_unmap;
 		}
 
+		c.dlfw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
+		c.dlfw.prp2 = cpu_to_le64(iod->first_dma);
 		c.dlfw.numd = cpu_to_le32((tot_len/BYTES_TO_DWORDS) - 1);
 		c.dlfw.offset = cpu_to_le32(offset/BYTES_TO_DWORDS);
 	} else if (opcode == nvme_admin_activate_fw) {
@@ -2033,7 +2034,6 @@ static int nvme_trans_do_nvme_io(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 	int res = SNTI_TRANSLATION_SUCCESS;
 	int nvme_sc;
 	struct nvme_dev *dev = ns->dev;
-	struct nvme_queue *nvmeq;
 	u32 num_cmds;
 	struct nvme_iod *iod;
 	u64 unit_len;
@@ -2045,7 +2045,7 @@ static int nvme_trans_do_nvme_io(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 	struct nvme_command c;
 	u8 opcode = (is_write ? nvme_cmd_write : nvme_cmd_read);
 	u16 control;
-	u32 max_blocks = nvme_block_nr(ns, dev->max_hw_sectors);
+	u32 max_blocks = queue_max_hw_sectors(ns->queue);
 
 	num_cmds = nvme_trans_io_get_num_cmds(hdr, cdb_info, max_blocks);
 
@@ -2093,8 +2093,7 @@ static int nvme_trans_do_nvme_io(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 			res = PTR_ERR(iod);
 			goto out;
 		}
-		retcode = nvme_setup_prps(dev, &c.common, iod, unit_len,
-							GFP_KERNEL);
+		retcode = nvme_setup_prps(dev, iod, unit_len, GFP_KERNEL);
 		if (retcode != unit_len) {
 			nvme_unmap_user_pages(dev,
 				(is_write) ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
@@ -2103,21 +2102,12 @@ static int nvme_trans_do_nvme_io(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 			res = -ENOMEM;
 			goto out;
 		}
+		c.rw.prp1 = cpu_to_le64(sg_dma_address(iod->sg));
+		c.rw.prp2 = cpu_to_le64(iod->first_dma);
 
 		nvme_offset += unit_num_blocks;
 
-		nvmeq = get_nvmeq(dev);
-		/*
-		 * Since nvme_submit_sync_cmd sleeps, we can't keep
-		 * preemption disabled.  We may be preempted at any
-		 * point, and be rescheduled to a different CPU.  That
-		 * will cause cacheline bouncing, but no additional
-		 * races since q_lock already protects against other
-		 * CPUs.
-		 */
-		put_nvmeq(nvmeq);
-		nvme_sc = nvme_submit_sync_cmd(nvmeq, &c, NULL,
-						NVME_IO_TIMEOUT);
+		nvme_sc = nvme_submit_io_cmd(dev, &c, NULL);
 		if (nvme_sc != NVME_SC_SUCCESS) {
 			nvme_unmap_user_pages(dev,
 				(is_write) ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
@@ -2644,7 +2634,6 @@ static int nvme_trans_start_stop(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 {
 	int res = SNTI_TRANSLATION_SUCCESS;
 	int nvme_sc;
-	struct nvme_queue *nvmeq;
 	struct nvme_command c;
 	u8 immed, pcmod, pc, no_flush, start;
 
@@ -2671,10 +2660,7 @@ static int nvme_trans_start_stop(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 			c.common.opcode = nvme_cmd_flush;
 			c.common.nsid = cpu_to_le32(ns->ns_id);
 
-			nvmeq = get_nvmeq(ns->dev);
-			put_nvmeq(nvmeq);
-			nvme_sc = nvme_submit_sync_cmd(nvmeq, &c, NULL, NVME_IO_TIMEOUT);
-
+			nvme_sc = nvme_submit_io_cmd(ns->dev, &c, NULL);
 			res = nvme_trans_status_code(hdr, nvme_sc);
 			if (res)
 				goto out;
@@ -2697,15 +2683,12 @@ static int nvme_trans_synchronize_cache(struct nvme_ns *ns,
 	int res = SNTI_TRANSLATION_SUCCESS;
 	int nvme_sc;
 	struct nvme_command c;
-	struct nvme_queue *nvmeq;
 
 	memset(&c, 0, sizeof(c));
 	c.common.opcode = nvme_cmd_flush;
 	c.common.nsid = cpu_to_le32(ns->ns_id);
 
-	nvmeq = get_nvmeq(ns->dev);
-	put_nvmeq(nvmeq);
-	nvme_sc = nvme_submit_sync_cmd(nvmeq, &c, NULL, NVME_IO_TIMEOUT);
+	nvme_sc = nvme_submit_io_cmd(ns->dev, &c, NULL);
 
 	res = nvme_trans_status_code(hdr, nvme_sc);
 	if (res)
@@ -2872,7 +2855,6 @@ static int nvme_trans_unmap(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 	struct nvme_dev *dev = ns->dev;
 	struct scsi_unmap_parm_list *plist;
 	struct nvme_dsm_range *range;
-	struct nvme_queue *nvmeq;
 	struct nvme_command c;
 	int i, nvme_sc, res = -ENOMEM;
 	u16 ndesc, list_len;
@@ -2914,10 +2896,7 @@ static int nvme_trans_unmap(struct nvme_ns *ns, struct sg_io_hdr *hdr,
 	c.dsm.nr = cpu_to_le32(ndesc - 1);
 	c.dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 
-	nvmeq = get_nvmeq(dev);
-	put_nvmeq(nvmeq);
-
-	nvme_sc = nvme_submit_sync_cmd(nvmeq, &c, NULL, NVME_IO_TIMEOUT);
+	nvme_sc = nvme_submit_io_cmd(dev, &c, NULL);
 	res = nvme_trans_status_code(hdr, nvme_sc);
 
 	dma_free_coherent(&dev->pci_dev->dev, ndesc * sizeof(*range),
diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index 6b9aafed225..a50173ca1d7 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -66,20 +66,25 @@ enum {
 
 #define NVME_VS(major, minor)	(major << 16 | minor)
 
-#define NVME_IO_TIMEOUT	(5 * HZ)
+extern unsigned char io_timeout;
+#define NVME_IO_TIMEOUT	(io_timeout * HZ)
 
 /*
  * Represents an NVM Express device.  Each nvme_dev is a PCI function.
  */
 struct nvme_dev {
 	struct list_head node;
-	struct nvme_queue **queues;
+	struct nvme_queue __rcu **queues;
+	unsigned short __percpu *io_queue;
 	u32 __iomem *dbs;
 	struct pci_dev *pci_dev;
 	struct dma_pool *prp_page_pool;
 	struct dma_pool *prp_small_pool;
 	int instance;
-	int queue_count;
+	unsigned queue_count;
+	unsigned online_queues;
+	unsigned max_qid;
+	int q_depth;
 	u32 db_stride;
 	u32 ctrl_config;
 	struct msix_entry *entry;
@@ -89,6 +94,7 @@ struct nvme_dev {
 	struct miscdevice miscdev;
 	work_func_t reset_workfn;
 	struct work_struct reset_work;
+	struct notifier_block nb;
 	char name[12];
 	char serial[20];
 	char model[40];
@@ -131,6 +137,7 @@ struct nvme_iod {
 	int length;		/* Of data, in bytes */
 	unsigned long start_time;
 	dma_addr_t first_dma;
+	struct list_head node;
 	struct scatterlist sg[0];
 };
 
@@ -146,16 +153,12 @@ static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
  */
 void nvme_free_iod(struct nvme_dev *dev, struct nvme_iod *iod);
 
-int nvme_setup_prps(struct nvme_dev *dev, struct nvme_common_command *cmd,
-			struct nvme_iod *iod, int total_len, gfp_t gfp);
+int nvme_setup_prps(struct nvme_dev *, struct nvme_iod *, int , gfp_t);
 struct nvme_iod *nvme_map_user_pages(struct nvme_dev *dev, int write,
 				unsigned long addr, unsigned length);
 void nvme_unmap_user_pages(struct nvme_dev *dev, int write,
 			struct nvme_iod *iod);
-struct nvme_queue *get_nvmeq(struct nvme_dev *dev);
-void put_nvmeq(struct nvme_queue *nvmeq);
-int nvme_submit_sync_cmd(struct nvme_queue *nvmeq, struct nvme_command *cmd,
-						u32 *result, unsigned timeout);
+int nvme_submit_io_cmd(struct nvme_dev *, struct nvme_command *, u32 *);
 int nvme_submit_flush_data(struct nvme_queue *nvmeq, struct nvme_ns *ns);
 int nvme_submit_admin_cmd(struct nvme_dev *, struct nvme_command *,
 							u32 *result);
diff --git a/include/uapi/linux/nvme.h b/include/uapi/linux/nvme.h
index e5ab6220111..096fe1c6f83 100644
--- a/include/uapi/linux/nvme.h
+++ b/include/uapi/linux/nvme.h
@@ -434,6 +434,7 @@ enum {
 	NVME_SC_REFTAG_CHECK		= 0x284,
 	NVME_SC_COMPARE_FAILED		= 0x285,
 	NVME_SC_ACCESS_DENIED		= 0x286,
+	NVME_SC_DNR			= 0x4000,
 };
 
 struct nvme_completion {