Merge tag 'virtio-next-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux

Pull virtio & lguest updates from Rusty Russell:
 "Lots of virtio work which wasn't quite ready for last merge window.

  Plus I dived into lguest again, reworking the pagetable code so we can
  move the switcher page: our fixmaps sometimes take more than 2MB now..."

Ugh.  Annoying conflicts with the tcm_vhost -> vhost_scsi rename.
Hopefully correctly resolved.

* tag 'virtio-next-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux: (57 commits)
  caif_virtio: Remove bouncing email addresses
  lguest: improve code readability in lg_cpu_start.
  virtio-net: fill only rx queues which are being used
  lguest: map Switcher below fixmap.
  lguest: cache last cpu we ran on.
  lguest: map Switcher text whenever we allocate a new pagetable.
  lguest: don't share Switcher PTE pages between guests.
  lguest: expost switcher_pages array (as lg_switcher_pages).
  lguest: extract shadow PTE walking / allocating.
  lguest: make check_gpte et. al return bool.
  lguest: assume Switcher text is a single page.
  lguest: rename switcher_page to switcher_pages.
  lguest: remove RESERVE_MEM constant.
  lguest: check vaddr not pgd for Switcher protection.
  lguest: prepare to make SWITCHER_ADDR a variable.
  virtio: console: replace EMFILE with EBUSY for already-open port
  virtio-scsi: reset virtqueue affinity when doing cpu hotplug
  virtio-scsi: introduce multiqueue support
  virtio-scsi: push vq lock/unlock into virtscsi_vq_done
  virtio-scsi: pass struct virtio_scsi to virtqueue completion function
  ...

22 files changed, 2850 insertions, 677 deletions

diff --git a/drivers/Makefile b/drivers/Makefile
index 33360de6365..8e57688ebd9 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -124,7 +124,7 @@ obj-$(CONFIG_PPC_PS3)		+= ps3/
 obj-$(CONFIG_OF)		+= of/
 obj-$(CONFIG_SSB)		+= ssb/
 obj-$(CONFIG_BCMA)		+= bcma/
-obj-$(CONFIG_VHOST_NET)		+= vhost/
+obj-$(CONFIG_VHOST_RING)	+= vhost/
 obj-$(CONFIG_VLYNQ)		+= vlynq/
 obj-$(CONFIG_STAGING)		+= staging/
 obj-y				+= platform/
diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 8ad21a25bc0..64723953e1c 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -100,96 +100,103 @@ static inline struct virtblk_req *virtblk_alloc_req(struct virtio_blk *vblk,
 	return vbr;
 }
 
-static void virtblk_add_buf_wait(struct virtio_blk *vblk,
-				 struct virtblk_req *vbr,
-				 unsigned long out,
-				 unsigned long in)
+static int __virtblk_add_req(struct virtqueue *vq,
+			     struct virtblk_req *vbr,
+			     struct scatterlist *data_sg,
+			     bool have_data)
 {
-	DEFINE_WAIT(wait);
+	struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
+	unsigned int num_out = 0, num_in = 0;
+	int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT;
 
-	for (;;) {
-		prepare_to_wait_exclusive(&vblk->queue_wait, &wait,
-					  TASK_UNINTERRUPTIBLE);
+	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
+	sgs[num_out++] = &hdr;
 
-		spin_lock_irq(vblk->disk->queue->queue_lock);
-		if (virtqueue_add_buf(vblk->vq, vbr->sg, out, in, vbr,
-				      GFP_ATOMIC) < 0) {
-			spin_unlock_irq(vblk->disk->queue->queue_lock);
-			io_schedule();
-		} else {
-			virtqueue_kick(vblk->vq);
-			spin_unlock_irq(vblk->disk->queue->queue_lock);
-			break;
-		}
+	/*
+	 * If this is a packet command we need a couple of additional headers.
+	 * Behind the normal outhdr we put a segment with the scsi command
+	 * block, and before the normal inhdr we put the sense data and the
+	 * inhdr with additional status information.
+	 */
+	if (type == VIRTIO_BLK_T_SCSI_CMD) {
+		sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
+		sgs[num_out++] = &cmd;
+	}
 
+	if (have_data) {
+		if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT)
+			sgs[num_out++] = data_sg;
+		else
+			sgs[num_out + num_in++] = data_sg;
 	}
 
-	finish_wait(&vblk->queue_wait, &wait);
+	if (type == VIRTIO_BLK_T_SCSI_CMD) {
+		sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
+		sgs[num_out + num_in++] = &sense;
+		sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
+		sgs[num_out + num_in++] = &inhdr;
+	}
+
+	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
+	sgs[num_out + num_in++] = &status;
+
+	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 }
 
-static inline void virtblk_add_req(struct virtblk_req *vbr,
-				   unsigned int out, unsigned int in)
+static void virtblk_add_req(struct virtblk_req *vbr, bool have_data)
 {
 	struct virtio_blk *vblk = vbr->vblk;
+	DEFINE_WAIT(wait);
+	int ret;
 
 	spin_lock_irq(vblk->disk->queue->queue_lock);
-	if (unlikely(virtqueue_add_buf(vblk->vq, vbr->sg, out, in, vbr,
-					GFP_ATOMIC) < 0)) {
+	while (unlikely((ret = __virtblk_add_req(vblk->vq, vbr, vbr->sg,
+						 have_data)) < 0)) {
+		prepare_to_wait_exclusive(&vblk->queue_wait, &wait,
+					  TASK_UNINTERRUPTIBLE);
+
 		spin_unlock_irq(vblk->disk->queue->queue_lock);
-		virtblk_add_buf_wait(vblk, vbr, out, in);
-		return;
+		io_schedule();
+		spin_lock_irq(vblk->disk->queue->queue_lock);
+
+		finish_wait(&vblk->queue_wait, &wait);
 	}
+
 	virtqueue_kick(vblk->vq);
 	spin_unlock_irq(vblk->disk->queue->queue_lock);
 }
 
-static int virtblk_bio_send_flush(struct virtblk_req *vbr)
+static void virtblk_bio_send_flush(struct virtblk_req *vbr)
 {
-	unsigned int out = 0, in = 0;
-
 	vbr->flags |= VBLK_IS_FLUSH;
 	vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 	vbr->out_hdr.sector = 0;
 	vbr->out_hdr.ioprio = 0;
-	sg_set_buf(&vbr->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
-	sg_set_buf(&vbr->sg[out + in++], &vbr->status, sizeof(vbr->status));
-
-	virtblk_add_req(vbr, out, in);
 
-	return 0;
+	virtblk_add_req(vbr, false);
 }
 
-static int virtblk_bio_send_data(struct virtblk_req *vbr)
+static void virtblk_bio_send_data(struct virtblk_req *vbr)
 {
 	struct virtio_blk *vblk = vbr->vblk;
-	unsigned int num, out = 0, in = 0;
 	struct bio *bio = vbr->bio;
+	bool have_data;
 
 	vbr->flags &= ~VBLK_IS_FLUSH;
 	vbr->out_hdr.type = 0;
 	vbr->out_hdr.sector = bio->bi_sector;
 	vbr->out_hdr.ioprio = bio_prio(bio);
 
-	sg_set_buf(&vbr->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
-
-	num = blk_bio_map_sg(vblk->disk->queue, bio, vbr->sg + out);
-
-	sg_set_buf(&vbr->sg[num + out + in++], &vbr->status,
-		   sizeof(vbr->status));
-
-	if (num) {
-		if (bio->bi_rw & REQ_WRITE) {
+	if (blk_bio_map_sg(vblk->disk->queue, bio, vbr->sg)) {
+		have_data = true;
+		if (bio->bi_rw & REQ_WRITE)
 			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
-			out += num;
-		} else {
+		else
 			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
-			in += num;
-		}
-	}
+	} else
+		have_data = false;
 
-	virtblk_add_req(vbr, out, in);
-
-	return 0;
+	virtblk_add_req(vbr, have_data);
 }
 
 static void virtblk_bio_send_data_work(struct work_struct *work)
@@ -298,7 +305,7 @@ static void virtblk_done(struct virtqueue *vq)
 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
 		   struct request *req)
 {
-	unsigned long num, out = 0, in = 0;
+	unsigned int num;
 	struct virtblk_req *vbr;
 
 	vbr = virtblk_alloc_req(vblk, GFP_ATOMIC);
@@ -335,40 +342,15 @@ static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
 		}
 	}
 
-	sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
-
-	/*
-	 * If this is a packet command we need a couple of additional headers.
-	 * Behind the normal outhdr we put a segment with the scsi command
-	 * block, and before the normal inhdr we put the sense data and the
-	 * inhdr with additional status information before the normal inhdr.
-	 */
-	if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
-		sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
-
-	num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
-
-	if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
-		sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
-		sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
-			   sizeof(vbr->in_hdr));
-	}
-
-	sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
-		   sizeof(vbr->status));
-
+	num = blk_rq_map_sg(q, vbr->req, vblk->sg);
 	if (num) {
-		if (rq_data_dir(vbr->req) == WRITE) {
+		if (rq_data_dir(vbr->req) == WRITE)
 			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
-			out += num;
-		} else {
+		else
 			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
-			in += num;
-		}
 	}
 
-	if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr,
-			      GFP_ATOMIC) < 0) {
+	if (__virtblk_add_req(vblk->vq, vbr, vblk->sg, num) < 0) {
 		mempool_free(vbr, vblk->pool);
 		return false;
 	}
@@ -539,6 +521,7 @@ static void virtblk_config_changed_work(struct work_struct *work)
 	struct virtio_device *vdev = vblk->vdev;
 	struct request_queue *q = vblk->disk->queue;
 	char cap_str_2[10], cap_str_10[10];
+	char *envp[] = { "RESIZE=1", NULL };
 	u64 capacity, size;
 
 	mutex_lock(&vblk->config_lock);
@@ -568,6 +551,7 @@ static void virtblk_config_changed_work(struct work_struct *work)
 
 	set_capacity(vblk->disk, capacity);
 	revalidate_disk(vblk->disk);
+	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
 done:
 	mutex_unlock(&vblk->config_lock);
 }
diff --git a/drivers/char/hw_random/virtio-rng.c b/drivers/char/hw_random/virtio-rng.c
index 6bf4d47324e..ef46a9cfd83 100644
--- a/drivers/char/hw_random/virtio-rng.c
+++ b/drivers/char/hw_random/virtio-rng.c
@@ -47,7 +47,7 @@ static void register_buffer(u8 *buf, size_t size)
 	sg_init_one(&sg, buf, size);
 
 	/* There should always be room for one buffer. */
-	if (virtqueue_add_buf(vq, &sg, 0, 1, buf, GFP_KERNEL) < 0)
+	if (virtqueue_add_inbuf(vq, &sg, 1, buf, GFP_KERNEL) < 0)
 		BUG();
 
 	virtqueue_kick(vq);
diff --git a/drivers/char/virtio_console.c b/drivers/char/virtio_console.c
index ce5f3fc25d6..1b456fe9b87 100644
--- a/drivers/char/virtio_console.c
+++ b/drivers/char/virtio_console.c
@@ -78,8 +78,8 @@ struct ports_driver_data {
 };
 static struct ports_driver_data pdrvdata;
 
-DEFINE_SPINLOCK(pdrvdata_lock);
-DECLARE_COMPLETION(early_console_added);
+static DEFINE_SPINLOCK(pdrvdata_lock);
+static DECLARE_COMPLETION(early_console_added);
 
 /* This struct holds information that's relevant only for console ports */
 struct console {
@@ -503,7 +503,7 @@ static int add_inbuf(struct virtqueue *vq, struct port_buffer *buf)
 
 	sg_init_one(sg, buf->buf, buf->size);
 
-	ret = virtqueue_add_buf(vq, sg, 0, 1, buf, GFP_ATOMIC);
+	ret = virtqueue_add_inbuf(vq, sg, 1, buf, GFP_ATOMIC);
 	virtqueue_kick(vq);
 	if (!ret)
 		ret = vq->num_free;
@@ -572,7 +572,7 @@ static ssize_t __send_control_msg(struct ports_device *portdev, u32 port_id,
 	sg_init_one(sg, &cpkt, sizeof(cpkt));
 
 	spin_lock(&portdev->c_ovq_lock);
-	if (virtqueue_add_buf(vq, sg, 1, 0, &cpkt, GFP_ATOMIC) == 0) {
+	if (virtqueue_add_outbuf(vq, sg, 1, &cpkt, GFP_ATOMIC) == 0) {
 		virtqueue_kick(vq);
 		while (!virtqueue_get_buf(vq, &len))
 			cpu_relax();
@@ -622,7 +622,7 @@ static ssize_t __send_to_port(struct port *port, struct scatterlist *sg,
 
 	reclaim_consumed_buffers(port);
 
-	err = virtqueue_add_buf(out_vq, sg, nents, 0, data, GFP_ATOMIC);
+	err = virtqueue_add_outbuf(out_vq, sg, nents, data, GFP_ATOMIC);
 
 	/* Tell Host to go! */
 	virtqueue_kick(out_vq);
@@ -1040,7 +1040,7 @@ static int port_fops_open(struct inode *inode, struct file *filp)
 	spin_lock_irq(&port->inbuf_lock);
 	if (port->guest_connected) {
 		spin_unlock_irq(&port->inbuf_lock);
-		ret = -EMFILE;
+		ret = -EBUSY;
 		goto out;
 	}
 
@@ -1202,7 +1202,7 @@ int __init virtio_cons_early_init(int (*put_chars)(u32, const char *, int))
 	return hvc_instantiate(0, 0, &hv_ops);
 }
 
-int init_port_console(struct port *port)
+static int init_port_console(struct port *port)
 {
 	int ret;
 
diff --git a/drivers/lguest/Kconfig b/drivers/lguest/Kconfig
index 89875ea19ad..ee035ec4526 100644
--- a/drivers/lguest/Kconfig
+++ b/drivers/lguest/Kconfig
@@ -5,10 +5,9 @@ config LGUEST
 	---help---
 	  This is a very simple module which allows you to run
 	  multiple instances of the same Linux kernel, using the
-	  "lguest" command found in the Documentation/virtual/lguest
-	  directory.
+	  "lguest" command found in the tools/lguest directory.
 
 	  Note that "lguest" is pronounced to rhyme with "fell quest",
-	  not "rustyvisor". See Documentation/virtual/lguest/lguest.txt.
+	  not "rustyvisor". See tools/lguest/lguest.txt.
 
 	  If unsure, say N.  If curious, say M.  If masochistic, say Y.
diff --git a/drivers/lguest/core.c b/drivers/lguest/core.c
index a5ebc0083d8..0bf1e4edf04 100644
--- a/drivers/lguest/core.c
+++ b/drivers/lguest/core.c
@@ -20,9 +20,9 @@
 #include <asm/asm-offsets.h>
 #include "lg.h"
 
-
+unsigned long switcher_addr;
+struct page **lg_switcher_pages;
 static struct vm_struct *switcher_vma;
-static struct page **switcher_page;
 
 /* This One Big lock protects all inter-guest data structures. */
 DEFINE_MUTEX(lguest_lock);
@@ -52,13 +52,21 @@ static __init int map_switcher(void)
 	 * easy.
 	 */
 
+	/* We assume Switcher text fits into a single page. */
+	if (end_switcher_text - start_switcher_text > PAGE_SIZE) {
+		printk(KERN_ERR "lguest: switcher text too large (%zu)\n",
+		       end_switcher_text - start_switcher_text);
+		return -EINVAL;
+	}
+
 	/*
 	 * We allocate an array of struct page pointers.  map_vm_area() wants
 	 * this, rather than just an array of pages.
 	 */
-	switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
-				GFP_KERNEL);
-	if (!switcher_page) {
+	lg_switcher_pages = kmalloc(sizeof(lg_switcher_pages[0])
+				    * TOTAL_SWITCHER_PAGES,
+				    GFP_KERNEL);
+	if (!lg_switcher_pages) {
 		err = -ENOMEM;
 		goto out;
 	}
@@ -68,32 +76,29 @@ static __init int map_switcher(void)
 	 * so we make sure they're zeroed.
 	 */
 	for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
-		switcher_page[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
-		if (!switcher_page[i]) {
+		lg_switcher_pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
+		if (!lg_switcher_pages[i]) {
 			err = -ENOMEM;
 			goto free_some_pages;
 		}
 	}
 
 	/*
-	 * First we check that the Switcher won't overlap the fixmap area at
-	 * the top of memory.  It's currently nowhere near, but it could have
-	 * very strange effects if it ever happened.
+	 * We place the Switcher underneath the fixmap area, which is the
+	 * highest virtual address we can get.  This is important, since we
+	 * tell the Guest it can't access this memory, so we want its ceiling
+	 * as high as possible.
 	 */
-	if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
-		err = -ENOMEM;
-		printk("lguest: mapping switcher would thwack fixmap\n");
-		goto free_pages;
-	}
+	switcher_addr = FIXADDR_START - (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE;
 
 	/*
-	 * Now we reserve the "virtual memory area" we want: 0xFFC00000
-	 * (SWITCHER_ADDR).  We might not get it in theory, but in practice
-	 * it's worked so far.  The end address needs +1 because __get_vm_area
-	 * allocates an extra guard page, so we need space for that.
+	 * Now we reserve the "virtual memory area" we want.  We might
+	 * not get it in theory, but in practice it's worked so far.
+	 * The end address needs +1 because __get_vm_area allocates an
+	 * extra guard page, so we need space for that.
 	 */
 	switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
-				     VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR
+				     VM_ALLOC, switcher_addr, switcher_addr
 				     + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
 	if (!switcher_vma) {
 		err = -ENOMEM;
@@ -103,12 +108,12 @@ static __init int map_switcher(void)
 
 	/*
 	 * This code actually sets up the pages we've allocated to appear at
-	 * SWITCHER_ADDR.  map_vm_area() takes the vma we allocated above, the
+	 * switcher_addr.  map_vm_area() takes the vma we allocated above, the
 	 * kind of pages we're mapping (kernel pages), and a pointer to our
 	 * array of struct pages.  It increments that pointer, but we don't
 	 * care.
 	 */
-	pagep = switcher_page;
+	pagep = lg_switcher_pages;
 	err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep);
 	if (err) {
 		printk("lguest: map_vm_area failed: %i\n", err);
@@ -133,8 +138,8 @@ free_pages:
 	i = TOTAL_SWITCHER_PAGES;
 free_some_pages:
 	for (--i; i >= 0; i--)
-		__free_pages(switcher_page[i], 0);
-	kfree(switcher_page);
+		__free_pages(lg_switcher_pages[i], 0);
+	kfree(lg_switcher_pages);
 out:
 	return err;
 }
@@ -149,8 +154,8 @@ static void unmap_switcher(void)
 	vunmap(switcher_vma->addr);
 	/* Now we just need to free the pages we copied the switcher into */
 	for (i = 0; i < TOTAL_SWITCHER_PAGES; i++)
-		__free_pages(switcher_page[i], 0);
-	kfree(switcher_page);
+		__free_pages(lg_switcher_pages[i], 0);
+	kfree(lg_switcher_pages);
 }
 
 /*H:032
@@ -323,15 +328,10 @@ static int __init init(void)
 	if (err)
 		goto out;
 
-	/* Now we set up the pagetable implementation for the Guests. */
-	err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
-	if (err)
-		goto unmap;
-
 	/* We might need to reserve an interrupt vector. */
 	err = init_interrupts();
 	if (err)
-		goto free_pgtables;
+		goto unmap;
 
 	/* /dev/lguest needs to be registered. */
 	err = lguest_device_init();
@@ -346,8 +346,6 @@ static int __init init(void)
 
 free_interrupts:
 	free_interrupts();
-free_pgtables:
-	free_pagetables();
 unmap:
 	unmap_switcher();
 out:
@@ -359,7 +357,6 @@ static void __exit fini(void)
 {
 	lguest_device_remove();
 	free_interrupts();
-	free_pagetables();
 	unmap_switcher();
 
 	lguest_arch_host_fini();
diff --git a/drivers/lguest/lg.h b/drivers/lguest/lg.h
index 295df06e659..2eef40be4c0 100644
--- a/drivers/lguest/lg.h
+++ b/drivers/lguest/lg.h
@@ -14,11 +14,10 @@
 
 #include <asm/lguest.h>
 
-void free_pagetables(void);
-int init_pagetables(struct page **switcher_page, unsigned int pages);
-
 struct pgdir {
 	unsigned long gpgdir;
+	bool switcher_mapped;
+	int last_host_cpu;
 	pgd_t *pgdir;
 };
 
@@ -124,6 +123,7 @@ bool lguest_address_ok(const struct lguest *lg,
 		       unsigned long addr, unsigned long len);
 void __lgread(struct lg_cpu *, void *, unsigned long, unsigned);
 void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
+extern struct page **lg_switcher_pages;
 
 /*H:035
  * Using memory-copy operations like that is usually inconvient, so we
diff --git a/drivers/lguest/lguest_user.c b/drivers/lguest/lguest_user.c
index ff4a0bc9904..4263f4cc8c5 100644
--- a/drivers/lguest/lguest_user.c
+++ b/drivers/lguest/lguest_user.c
@@ -250,13 +250,13 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
  */
 static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
 {
-	/* We have a limited number the number of CPUs in the lguest struct. */
+	/* We have a limited number of CPUs in the lguest struct. */
 	if (id >= ARRAY_SIZE(cpu->lg->cpus))
 		return -EINVAL;
 
 	/* Set up this CPU's id, and pointer back to the lguest struct. */
 	cpu->id = id;
-	cpu->lg = container_of((cpu - id), struct lguest, cpus[0]);
+	cpu->lg = container_of(cpu, struct lguest, cpus[id]);
 	cpu->lg->nr_cpus++;
 
 	/* Each CPU has a timer it can set. */
@@ -270,7 +270,7 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
 	if (!cpu->regs_page)
 		return -ENOMEM;
 
-	/* We actually put the registers at the bottom of the page. */
+	/* We actually put the registers at the end of the page. */
 	cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
 
 	/*
diff --git a/drivers/lguest/page_tables.c b/drivers/lguest/page_tables.c
index 864baabaee2..699187ab380 100644
--- a/drivers/lguest/page_tables.c
+++ b/drivers/lguest/page_tables.c
@@ -7,7 +7,7 @@
  * converted Guest pages when running the Guest.
 :*/
 
-/* Copyright (C) Rusty Russell IBM Corporation 2006.
+/* Copyright (C) Rusty Russell IBM Corporation 2013.
  * GPL v2 and any later version */
 #include <linux/mm.h>
 #include <linux/gfp.h>
@@ -62,22 +62,11 @@
  * will need the last pmd entry of the last pmd page.
  */
 #ifdef CONFIG_X86_PAE
-#define SWITCHER_PMD_INDEX 	(PTRS_PER_PMD - 1)
-#define RESERVE_MEM 		2U
 #define CHECK_GPGD_MASK		_PAGE_PRESENT
 #else
-#define RESERVE_MEM 		4U
 #define CHECK_GPGD_MASK		_PAGE_TABLE
 #endif
 
-/*
- * We actually need a separate PTE page for each CPU.  Remember that after the
- * Switcher code itself comes two pages for each CPU, and we don't want this
- * CPU's guest to see the pages of any other CPU.
- */
-static DEFINE_PER_CPU(pte_t *, switcher_pte_pages);
-#define switcher_pte_page(cpu) per_cpu(switcher_pte_pages, cpu)
-
 /*H:320
  * The page table code is curly enough to need helper functions to keep it
  * clear and clean.  The kernel itself provides many of them; one advantage
@@ -95,13 +84,6 @@ static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr)
 {
 	unsigned int index = pgd_index(vaddr);
 
-#ifndef CONFIG_X86_PAE
-	/* We kill any Guest trying to touch the Switcher addresses. */
-	if (index >= SWITCHER_PGD_INDEX) {
-		kill_guest(cpu, "attempt to access switcher pages");
-		index = 0;
-	}
-#endif
 	/* Return a pointer index'th pgd entry for the i'th page table. */
 	return &cpu->lg->pgdirs[i].pgdir[index];
 }
@@ -117,13 +99,6 @@ static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
 	unsigned int index = pmd_index(vaddr);
 	pmd_t *page;
 
-	/* We kill any Guest trying to touch the Switcher addresses. */
-	if (pgd_index(vaddr) == SWITCHER_PGD_INDEX &&
-					index >= SWITCHER_PMD_INDEX) {
-		kill_guest(cpu, "attempt to access switcher pages");
-		index = 0;
-	}
-
 	/* You should never call this if the PGD entry wasn't valid */
 	BUG_ON(!(pgd_flags(spgd) & _PAGE_PRESENT));
 	page = __va(pgd_pfn(spgd) << PAGE_SHIFT);
@@ -275,122 +250,177 @@ static void release_pte(pte_t pte)
 }
 /*:*/
 
-static void check_gpte(struct lg_cpu *cpu, pte_t gpte)
+static bool check_gpte(struct lg_cpu *cpu, pte_t gpte)
 {
 	if ((pte_flags(gpte) & _PAGE_PSE) ||
-	    pte_pfn(gpte) >= cpu->lg->pfn_limit)
+	    pte_pfn(gpte) >= cpu->lg->pfn_limit) {
 		kill_guest(cpu, "bad page table entry");
+		return false;
+	}
+	return true;
 }
 
-static void check_gpgd(struct lg_cpu *cpu, pgd_t gpgd)
+static bool check_gpgd(struct lg_cpu *cpu, pgd_t gpgd)
 {
 	if ((pgd_flags(gpgd) & ~CHECK_GPGD_MASK) ||
-	   (pgd_pfn(gpgd) >= cpu->lg->pfn_limit))
+	    (pgd_pfn(gpgd) >= cpu->lg->pfn_limit)) {
 		kill_guest(cpu, "bad page directory entry");
+		return false;
+	}
+	return true;
 }
 
 #ifdef CONFIG_X86_PAE
-static void check_gpmd(struct lg_cpu *cpu, pmd_t gpmd)
+static bool check_gpmd(struct lg_cpu *cpu, pmd_t gpmd)
 {
 	if ((pmd_flags(gpmd) & ~_PAGE_TABLE) ||
-	   (pmd_pfn(gpmd) >= cpu->lg->pfn_limit))
+	    (pmd_pfn(gpmd) >= cpu->lg->pfn_limit)) {
 		kill_guest(cpu, "bad page middle directory entry");
+		return false;
+	}
+	return true;
 }
 #endif
 
-/*H:330
- * (i) Looking up a page table entry when the Guest faults.
- *
- * We saw this call in run_guest(): when we see a page fault in the Guest, we
- * come here.  That's because we only set up the shadow page tables lazily as
- * they're needed, so we get page faults all the time and quietly fix them up
- * and return to the Guest without it knowing.
+/*H:331
+ * This is the core routine to walk the shadow page tables and find the page
+ * table entry for a specific address.
  *
- * If we fixed up the fault (ie. we mapped the address), this routine returns
- * true.  Otherwise, it was a real fault and we need to tell the Guest.
+ * If allocate is set, then we allocate any missing levels, setting the flags
+ * on the new page directory and mid-level directories using the arguments
+ * (which are copied from the Guest's page table entries).
  */
-bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
+static pte_t *find_spte(struct lg_cpu *cpu, unsigned long vaddr, bool allocate,
+			int pgd_flags, int pmd_flags)
 {
-	pgd_t gpgd;
 	pgd_t *spgd;
-	unsigned long gpte_ptr;
-	pte_t gpte;
-	pte_t *spte;
-
 	/* Mid level for PAE. */
 #ifdef CONFIG_X86_PAE
 	pmd_t *spmd;
-	pmd_t gpmd;
 #endif
 
-	/* First step: get the top-level Guest page table entry. */
-	if (unlikely(cpu->linear_pages)) {
-		/* Faking up a linear mapping. */
-		gpgd = __pgd(CHECK_GPGD_MASK);
-	} else {
-		gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t);
-		/* Toplevel not present?  We can't map it in. */
-		if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
-			return false;
-	}
-
-	/* Now look at the matching shadow entry. */
+	/* Get top level entry. */
 	spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr);
 	if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) {
 		/* No shadow entry: allocate a new shadow PTE page. */
-		unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
+		unsigned long ptepage;
+
+		/* If they didn't want us to allocate anything, stop. */
+		if (!allocate)
+			return NULL;
+
+		ptepage = get_zeroed_page(GFP_KERNEL);
 		/*
 		 * This is not really the Guest's fault, but killing it is
 		 * simple for this corner case.
 		 */
 		if (!ptepage) {
 			kill_guest(cpu, "out of memory allocating pte page");
-			return false;
+			return NULL;
 		}
-		/* We check that the Guest pgd is OK. */
-		check_gpgd(cpu, gpgd);
 		/*
 		 * And we copy the flags to the shadow PGD entry.  The page
 		 * number in the shadow PGD is the page we just allocated.
 		 */
-		set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags(gpgd)));
+		set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags));
 	}
 
+	/*
+	 * Intel's Physical Address Extension actually uses three levels of
+	 * page tables, so we need to look in the mid-level.
+	 */
 #ifdef CONFIG_X86_PAE
-	if (unlikely(cpu->linear_pages)) {
-		/* Faking up a linear mapping. */
-		gpmd = __pmd(_PAGE_TABLE);
-	} else {
-		gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
-		/* Middle level not present?  We can't map it in. */
-		if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
-			return false;
-	}
-
-	/* Now look at the matching shadow entry. */
+	/* Now look at the mid-level shadow entry. */
 	spmd = spmd_addr(cpu, *spgd, vaddr);
 
 	if (!(pmd_flags(*spmd) & _PAGE_PRESENT)) {
 		/* No shadow entry: allocate a new shadow PTE page. */
-		unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
+		unsigned long ptepage;
+
+		/* If they didn't want us to allocate anything, stop. */
+		if (!allocate)
+			return NULL;
+
+		ptepage = get_zeroed_page(GFP_KERNEL);
 
 		/*
 		 * This is not really the Guest's fault, but killing it is
 		 * simple for this corner case.
 		 */
 		if (!ptepage) {
-			kill_guest(cpu, "out of memory allocating pte page");
-			return false;
+			kill_guest(cpu, "out of memory allocating pmd page");
+			return NULL;
 		}
 
-		/* We check that the Guest pmd is OK. */
-		check_gpmd(cpu, gpmd);
-
 		/*
 		 * And we copy the flags to the shadow PMD entry.  The page
 		 * number in the shadow PMD is the page we just allocated.
 		 */
-		set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd)));
+		set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags));
+	}
+#endif
+
+	/* Get the pointer to the shadow PTE entry we're going to set. */
+	return spte_addr(cpu, *spgd, vaddr);
+}
+
+/*H:330
+ * (i) Looking up a page table entry when the Guest faults.
+ *
+ * We saw this call in run_guest(): when we see a page fault in the Guest, we
+ * come here.  That's because we only set up the shadow page tables lazily as
+ * they're needed, so we get page faults all the time and quietly fix them up
+ * and return to the Guest without it knowing.
+ *
+ * If we fixed up the fault (ie. we mapped the address), this routine returns
+ * true.  Otherwise, it was a real fault and we need to tell the Guest.
+ */
+bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
+{
+	unsigned long gpte_ptr;
+	pte_t gpte;
+	pte_t *spte;
+	pmd_t gpmd;
+	pgd_t gpgd;
+
+	/* We never demand page the Switcher, so trying is a mistake. */
+	if (vaddr >= switcher_addr)
+		return false;
+
+	/* First step: get the top-level Guest page table entry. */
+	if (unlikely(cpu->linear_pages)) {
+		/* Faking up a linear mapping. */
+		gpgd = __pgd(CHECK_GPGD_MASK);
+	} else {
+		gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t);
+		/* Toplevel not present?  We can't map it in. */
+		if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
+			return false;
+
+		/* 
+		 * This kills the Guest if it has weird flags or tries to
+		 * refer to a "physical" address outside the bounds.
+		 */
+		if (!check_gpgd(cpu, gpgd))
+			return false;
+	}
+
+	/* This "mid-level" entry is only used for non-linear, PAE mode. */
+	gpmd = __pmd(_PAGE_TABLE);
+
+#ifdef CONFIG_X86_PAE
+	if (likely(!cpu->linear_pages)) {
+		gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
+		/* Middle level not present?  We can't map it in. */
+		if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
+			return false;
+
+		/* 
+		 * This kills the Guest if it has weird flags or tries to
+		 * refer to a "physical" address outside the bounds.
+		 */
+		if (!check_gpmd(cpu, gpmd))
+			return false;
 	}
 
 	/*
@@ -433,7 +463,8 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
 	 * Check that the Guest PTE flags are OK, and the page number is below
 	 * the pfn_limit (ie. not mapping the Launcher binary).
 	 */
-	check_gpte(cpu, gpte);
+	if (!check_gpte(cpu, gpte))
+		return false;
 
 	/* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */
 	gpte = pte_mkyoung(gpte);
@@ -441,7 +472,9 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
 		gpte = pte_mkdirty(gpte);
 
 	/* Get the pointer to the shadow PTE entry we're going to set. */
-	spte = spte_addr(cpu, *spgd, vaddr);
+	spte = find_spte(cpu, vaddr, true, pgd_flags(gpgd), pmd_flags(gpmd));
+	if (!spte)
+		return false;
 
 	/*
 	 * If there was a valid shadow PTE entry here before, we release it.
@@ -493,29 +526,23 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
  */
 static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr)
 {
-	pgd_t *spgd;
+	pte_t *spte;
 	unsigned long flags;
 
-#ifdef CONFIG_X86_PAE
-	pmd_t *spmd;
-#endif
-	/* Look at the current top level entry: is it present? */
-	spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr);
-	if (!(pgd_flags(*spgd) & _PAGE_PRESENT))
+	/* You can't put your stack in the Switcher! */
+	if (vaddr >= switcher_addr)
 		return false;
 
-#ifdef CONFIG_X86_PAE
-	spmd = spmd_addr(cpu, *spgd, vaddr);
-	if (!(pmd_flags(*spmd) & _PAGE_PRESENT))
+	/* If there's no shadow PTE, it's not writable. */
+	spte = find_spte(cpu, vaddr, false, 0, 0);
+	if (!spte)
 		return false;
-#endif
 
 	/*
 	 * Check the flags on the pte entry itself: it must be present and
 	 * writable.
 	 */
-	flags = pte_flags(*(spte_addr(cpu, *spgd, vaddr)));
-
+	flags = pte_flags(*spte);
 	return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW);
 }
 
@@ -678,9 +705,6 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
 			      int *blank_pgdir)
 {
 	unsigned int next;
-#ifdef CONFIG_X86_PAE
-	pmd_t *pmd_table;
-#endif
 
 	/*
 	 * We pick one entry at random to throw out.  Choosing the Least
@@ -695,29 +719,11 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
 		if (!cpu->lg->pgdirs[next].pgdir)
 			next = cpu->cpu_pgd;
 		else {
-#ifdef CONFIG_X86_PAE
 			/*
-			 * In PAE mode, allocate a pmd page and populate the
-			 * last pgd entry.
+			 * This is a blank page, so there are no kernel
+			 * mappings: caller must map the stack!
 			 */
-			pmd_table = (pmd_t *)get_zeroed_page(GFP_KERNEL);
-			if (!pmd_table) {
-				free_page((long)cpu->lg->pgdirs[next].pgdir);
-				set_pgd(cpu->lg->pgdirs[next].pgdir, __pgd(0));
-				next = cpu->cpu_pgd;
-			} else {
-				set_pgd(cpu->lg->pgdirs[next].pgdir +
-					SWITCHER_PGD_INDEX,
-					__pgd(__pa(pmd_table) | _PAGE_PRESENT));
-				/*
-				 * This is a blank page, so there are no kernel
-				 * mappings: caller must map the stack!
-				 */
-				*blank_pgdir = 1;
-			}
-#else
 			*blank_pgdir = 1;
-#endif
 		}
 	}
 	/* Record which Guest toplevel this shadows. */
@@ -725,9 +731,50 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
 	/* Release all the non-kernel mappings. */
 	flush_user_mappings(cpu->lg, next);
 
+	/* This hasn't run on any CPU at all. */
+	cpu->lg->pgdirs[next].last_host_cpu = -1;
+
 	return next;
 }
 
+/*H:501
+ * We do need the Switcher code mapped at all times, so we allocate that
+ * part of the Guest page table here.  We map the Switcher code immediately,
+ * but defer mapping of the guest register page and IDT/LDT etc page until
+ * just before we run the guest in map_switcher_in_guest().
+ *
+ * We *could* do this setup in map_switcher_in_guest(), but at that point
+ * we've interrupts disabled, and allocating pages like that is fraught: we
+ * can't sleep if we need to free up some memory.
+ */
+static bool allocate_switcher_mapping(struct lg_cpu *cpu)
+{
+	int i;
+
+	for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
+		pte_t *pte = find_spte(cpu, switcher_addr + i * PAGE_SIZE, true,
+				       CHECK_GPGD_MASK, _PAGE_TABLE);
+		if (!pte)
+			return false;
+
+		/*
+		 * Map the switcher page if not already there.  It might
+		 * already be there because we call allocate_switcher_mapping()
+		 * in guest_set_pgd() just in case it did discard our Switcher
+		 * mapping, but it probably didn't.
+		 */
+		if (i == 0 && !(pte_flags(*pte) & _PAGE_PRESENT)) {
+			/* Get a reference to the Switcher page. */
+			get_page(lg_switcher_pages[0]);
+			/* Create a read-only, exectuable, kernel-style PTE */
+			set_pte(pte,
+				mk_pte(lg_switcher_pages[0], PAGE_KERNEL_RX));
+		}
+	}
+	cpu->lg->pgdirs[cpu->cpu_pgd].switcher_mapped = true;
+	return true;
+}
+
 /*H:470
  * Finally, a routine which throws away everything: all PGD entries in all
  * the shadow page tables, including the Guest's kernel mappings.  This is used
@@ -738,28 +785,16 @@ static void release_all_pagetables(struct lguest *lg)
 	unsigned int i, j;
 
 	/* Every shadow pagetable this Guest has */
-	for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++)
-		if (lg->pgdirs[i].pgdir) {
-#ifdef CONFIG_X86_PAE
-			pgd_t *spgd;
-			pmd_t *pmdpage;
-			unsigned int k;
-
-			/* Get the last pmd page. */
-			spgd = lg->pgdirs[i].pgdir + SWITCHER_PGD_INDEX;
-			pmdpage = __va(pgd_pfn(*spgd) << PAGE_SHIFT);
-
-			/*
-			 * And release the pmd entries of that pmd page,
-			 * except for the switcher pmd.
-			 */
-			for (k = 0; k < SWITCHER_PMD_INDEX; k++)
-				release_pmd(&pmdpage[k]);
-#endif
-			/* Every PGD entry except the Switcher at the top */
-			for (j = 0; j < SWITCHER_PGD_INDEX; j++)
-				release_pgd(lg->pgdirs[i].pgdir + j);
-		}
+	for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++) {
+		if (!lg->pgdirs[i].pgdir)
+			continue;
+
+		/* Every PGD entry. */
+		for (j = 0; j < PTRS_PER_PGD; j++)
+			release_pgd(lg->pgdirs[i].pgdir + j);
+		lg->pgdirs[i].switcher_mapped = false;
+		lg->pgdirs[i].last_host_cpu = -1;
+	}
 }
 
 /*
@@ -773,6 +808,9 @@ void guest_pagetable_clear_all(struct lg_cpu *cpu)
 	release_all_pagetables(cpu->lg);
 	/* We need the Guest kernel stack mapped again. */
 	pin_stack_pages(cpu);
+	/* And we need Switcher allocated. */
+	if (!allocate_switcher_mapping(cpu))
+		kill_guest(cpu, "Cannot populate switcher mapping");
 }
 
 /*H:430
@@ -808,9 +846,17 @@ void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable)
 		newpgdir = new_pgdir(cpu, pgtable, &repin);
 	/* Change the current pgd index to the new one. */
 	cpu->cpu_pgd = newpgdir;
-	/* If it was completely blank, we map in the Guest kernel stack */
+	/*
+	 * If it was completely blank, we map in the Guest kernel stack and
+	 * the Switcher.
+	 */
 	if (repin)
 		pin_stack_pages(cpu);
+
+	if (!cpu->lg->pgdirs[cpu->cpu_pgd].switcher_mapped) {
+		if (!allocate_switcher_mapping(cpu))
+			kill_guest(cpu, "Cannot populate switcher mapping");
+	}
 }
 /*:*/
 
@@ -865,7 +911,8 @@ static void do_set_pte(struct lg_cpu *cpu, int idx,
 			 * micro-benchmark.
 			 */
 			if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) {
-				check_gpte(cpu, gpte);
+				if (!check_gpte(cpu, gpte))
+					return;
 				set_pte(spte,
 					gpte_to_spte(cpu, gpte,
 						pte_flags(gpte) & _PAGE_DIRTY));
@@ -897,6 +944,12 @@ static void do_set_pte(struct lg_cpu *cpu, int idx,
 void guest_set_pte(struct lg_cpu *cpu,
 		   unsigned long gpgdir, unsigned long vaddr, pte_t gpte)
 {
+	/* We don't let you remap the Switcher; we need it to get back! */
+	if (vaddr >= switcher_addr) {
+		kill_guest(cpu, "attempt to set pte into Switcher pages");
+		return;
+	}
+
 	/*
 	 * Kernel mappings must be changed on all top levels.  Slow, but doesn't
 	 * happen often.
@@ -933,14 +986,23 @@ void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 idx)
 {
 	int pgdir;
 
-	if (idx >= SWITCHER_PGD_INDEX)
+	if (idx > PTRS_PER_PGD) {
+		kill_guest(&lg->cpus[0], "Attempt to set pgd %u/%u",
+			   idx, PTRS_PER_PGD);
 		return;
+	}
 
 	/* If they're talking about a page table we have a shadow for... */
 	pgdir = find_pgdir(lg, gpgdir);
-	if (pgdir < ARRAY_SIZE(lg->pgdirs))
+	if (pgdir < ARRAY_SIZE(lg->pgdirs)) {
 		/* ... throw it away. */
 		release_pgd(lg->pgdirs[pgdir].pgdir + idx);
+		/* That might have been the Switcher mapping, remap it. */
+		if (!allocate_switcher_mapping(&lg->cpus[0])) {
+			kill_guest(&lg->cpus[0],
+				   "Cannot populate switcher mapping");
+		}
+	}
 }
 
 #ifdef CONFIG_X86_PAE
@@ -958,6 +1020,9 @@ void guest_set_pmd(struct lguest *lg, unsigned long pmdp, u32 idx)
  * we will populate on future faults.  The Guest doesn't have any actual
  * pagetables yet, so we set linear_pages to tell demand_page() to fake it
  * for the moment.
+ *
+ * We do need the Switcher to be mapped at all times, so we allocate that
+ * part of the Guest page table here.
  */
 int init_guest_pagetable(struct lguest *lg)
 {
@@ -971,21 +1036,34 @@ int init_guest_pagetable(struct lguest *lg)
 
 	/* We start with a linear mapping until the initialize. */
 	cpu->linear_pages = true;
+
+	/* Allocate the page tables for the Switcher. */
+	if (!allocate_switcher_mapping(cpu)) {
+		release_all_pagetables(lg);
+		return -ENOMEM;
+	}
+
 	return 0;
 }
 
 /*H:508 When the Guest calls LHCALL_LGUEST_INIT we do more setup. */
 void page_table_guest_data_init(struct lg_cpu *cpu)
 {
+	/*
+	 * We tell the Guest that it can't use the virtual addresses
+	 * used by the Switcher.  This trick is equivalent to 4GB -
+	 * switcher_addr.
+	 */
+	u32 top = ~switcher_addr + 1;
+
 	/* We get the kernel address: above this is all kernel memory. */
 	if (get_user(cpu->lg->kernel_address,
-		&cpu->lg->lguest_data->kernel_address)
+		     &cpu->lg->lguest_data->kernel_address)
 		/*
-		 * We tell the Guest that it can't use the top 2 or 4 MB
-		 * of virtual addresses used by the Switcher.
+		 * We tell the Guest that it can't use the top virtual
+		 * addresses (used by the Switcher).
 		 */
-		|| put_user(RESERVE_MEM * 1024 * 1024,
-			    &cpu->lg->lguest_data->reserve_mem)) {
+	    || put_user(top, &cpu->lg->lguest_data->reserve_mem)) {
 		kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
 		return;
 	}
@@ -995,12 +1073,7 @@ void page_table_guest_data_init(struct lg_cpu *cpu)
 	 * "pgd_index(lg->kernel_address)".  This assumes it won't hit the
 	 * Switcher mappings, so check that now.
 	 */
-#ifdef CONFIG_X86_PAE
-	if (pgd_index(cpu->lg->kernel_address) == SWITCHER_PGD_INDEX &&
-		pmd_index(cpu->lg->kernel_address) == SWITCHER_PMD_INDEX)
-#else
-	if (pgd_index(cpu->lg->kernel_address) >= SWITCHER_PGD_INDEX)
-#endif
+	if (cpu->lg->kernel_address >= switcher_addr)
 		kill_guest(cpu, "bad kernel address %#lx",
 				 cpu->lg->kernel_address);
 }
@@ -1017,102 +1090,96 @@ void free_guest_pagetable(struct lguest *lg)
 		free_page((long)lg->pgdirs[i].pgdir);
 }
 
-/*H:480
- * (vi) Mapping the Switcher when the Guest is about to run.
- *
- * The Switcher and the two pages for this CPU need to be visible in the
- * Guest (and not the pages for other CPUs).  We have the appropriate PTE pages
- * for each CPU already set up, we just need to hook them in now we know which
- * Guest is about to run on this CPU.
+/*H:481
+ * This clears the Switcher mappings for cpu #i.
  */
-void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages)
+static void remove_switcher_percpu_map(struct lg_cpu *cpu, unsigned int i)
 {
-	pte_t *switcher_pte_page = __this_cpu_read(switcher_pte_pages);
-	pte_t regs_pte;
+	unsigned long base = switcher_addr + PAGE_SIZE + i * PAGE_SIZE*2;
+	pte_t *pte;
 
-#ifdef CONFIG_X86_PAE
-	pmd_t switcher_pmd;
-	pmd_t *pmd_table;
-
-	switcher_pmd = pfn_pmd(__pa(switcher_pte_page) >> PAGE_SHIFT,
-			       PAGE_KERNEL_EXEC);
-
-	/* Figure out where the pmd page is, by reading the PGD, and converting
-	 * it to a virtual address. */
-	pmd_table = __va(pgd_pfn(cpu->lg->
-			pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX])
-								<< PAGE_SHIFT);
-	/* Now write it into the shadow page table. */
-	set_pmd(&pmd_table[SWITCHER_PMD_INDEX], switcher_pmd);
-#else
-	pgd_t switcher_pgd;
+	/* Clear the mappings for both pages. */
+	pte = find_spte(cpu, base, false, 0, 0);
+	release_pte(*pte);
+	set_pte(pte, __pte(0));
 
-	/*
-	 * Make the last PGD entry for this Guest point to the Switcher's PTE
-	 * page for this CPU (with appropriate flags).
-	 */
-	switcher_pgd = __pgd(__pa(switcher_pte_page) | __PAGE_KERNEL_EXEC);
-
-	cpu->lg->pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX] = switcher_pgd;
-
-#endif
-	/*
-	 * We also change the Switcher PTE page.  When we're running the Guest,
-	 * we want the Guest's "regs" page to appear where the first Switcher
-	 * page for this CPU is.  This is an optimization: when the Switcher
-	 * saves the Guest registers, it saves them into the first page of this
-	 * CPU's "struct lguest_pages": if we make sure the Guest's register
-	 * page is already mapped there, we don't have to copy them out
-	 * again.
-	 */
-	regs_pte = pfn_pte(__pa(cpu->regs_page) >> PAGE_SHIFT, PAGE_KERNEL);
-	set_pte(&switcher_pte_page[pte_index((unsigned long)pages)], regs_pte);
+	pte = find_spte(cpu, base + PAGE_SIZE, false, 0, 0);
+	release_pte(*pte);
+	set_pte(pte, __pte(0));
 }
-/*:*/
 
-static void free_switcher_pte_pages(void)
-{
-	unsigned int i;
-
-	for_each_possible_cpu(i)
-		free_page((long)switcher_pte_page(i));
-}
-
-/*H:520
- * Setting up the Switcher PTE page for given CPU is fairly easy, given
- * the CPU number and the "struct page"s for the Switcher code itself.
+/*H:480
+ * (vi) Mapping the Switcher when the Guest is about to run.
+ *
+ * The Switcher and the two pages for this CPU need to be visible in the Guest
+ * (and not the pages for other CPUs).
  *
- * Currently the Switcher is less than a page long, so "pages" is always 1.
+ * The pages for the pagetables have all been allocated before: we just need
+ * to make sure the actual PTEs are up-to-date for the CPU we're about to run
+ * on.
  */
-static __init void populate_switcher_pte_page(unsigned int cpu,
-					      struct page *switcher_page[],
-					      unsigned int pages)
+void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages)
 {
-	unsigned int i;
-	pte_t *pte = switcher_pte_page(cpu);
+	unsigned long base;
+	struct page *percpu_switcher_page, *regs_page;
+	pte_t *pte;
+	struct pgdir *pgdir = &cpu->lg->pgdirs[cpu->cpu_pgd];
+
+	/* Switcher page should always be mapped by now! */
+	BUG_ON(!pgdir->switcher_mapped);
+
+	/* 
+	 * Remember that we have two pages for each Host CPU, so we can run a
+	 * Guest on each CPU without them interfering.  We need to make sure
+	 * those pages are mapped correctly in the Guest, but since we usually
+	 * run on the same CPU, we cache that, and only update the mappings
+	 * when we move.
+	 */
+	if (pgdir->last_host_cpu == raw_smp_processor_id())
+		return;
 
-	/* The first entries are easy: they map the Switcher code. */
-	for (i = 0; i < pages; i++) {
-		set_pte(&pte[i], mk_pte(switcher_page[i],
-				__pgprot(_PAGE_PRESENT|_PAGE_ACCESSED)));
+	/* -1 means unknown so we remove everything. */
+	if (pgdir->last_host_cpu == -1) {
+		unsigned int i;
+		for_each_possible_cpu(i)
+			remove_switcher_percpu_map(cpu, i);
+	} else {
+		/* We know exactly what CPU mapping to remove. */
+		remove_switcher_percpu_map(cpu, pgdir->last_host_cpu);
 	}
 
-	/* The only other thing we map is this CPU's pair of pages. */
-	i = pages + cpu*2;
-
-	/* First page (Guest registers) is writable from the Guest */
-	set_pte(&pte[i], pfn_pte(page_to_pfn(switcher_page[i]),
-			 __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW)));
+	/*
+	 * When we're running the Guest, we want the Guest's "regs" page to
+	 * appear where the first Switcher page for this CPU is.  This is an
+	 * optimization: when the Switcher saves the Guest registers, it saves
+	 * them into the first page of this CPU's "struct lguest_pages": if we
+	 * make sure the Guest's register page is already mapped there, we
+	 * don't have to copy them out again.
+	 */
+	/* Find the shadow PTE for this regs page. */
+	base = switcher_addr + PAGE_SIZE
+		+ raw_smp_processor_id() * sizeof(struct lguest_pages);
+	pte = find_spte(cpu, base, false, 0, 0);
+	regs_page = pfn_to_page(__pa(cpu->regs_page) >> PAGE_SHIFT);
+	get_page(regs_page);
+	set_pte(pte, mk_pte(regs_page, __pgprot(__PAGE_KERNEL & ~_PAGE_GLOBAL)));
 
 	/*
-	 * The second page contains the "struct lguest_ro_state", and is
-	 * read-only.
+	 * We map the second page of the struct lguest_pages read-only in
+	 * the Guest: the IDT, GDT and other things it's not supposed to
+	 * change.
 	 */
-	set_pte(&pte[i+1], pfn_pte(page_to_pfn(switcher_page[i+1]),
-			   __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED)));
+	pte = find_spte(cpu, base + PAGE_SIZE, false, 0, 0);
+	percpu_switcher_page
+		= lg_switcher_pages[1 + raw_smp_processor_id()*2 + 1];
+	get_page(percpu_switcher_page);
+	set_pte(pte, mk_pte(percpu_switcher_page,
+			    __pgprot(__PAGE_KERNEL_RO & ~_PAGE_GLOBAL)));
+
+	pgdir->last_host_cpu = raw_smp_processor_id();
 }
 
-/*
+/*H:490
  * We've made it through the page table code.  Perhaps our tired brains are
  * still processing the details, or perhaps we're simply glad it's over.
  *
@@ -1124,29 +1191,3 @@ static __init void populate_switcher_pte_page(unsigned int cpu,
  *
  * There is just one file remaining in the Host.
  */
-
-/*H:510
- * At boot or module load time, init_pagetables() allocates and populates
- * the Switcher PTE page for each CPU.
- */
-__init int init_pagetables(struct page **switcher_page, unsigned int pages)
-{
-	unsigned int i;
-
-	for_each_possible_cpu(i) {
-		switcher_pte_page(i) = (pte_t *)get_zeroed_page(GFP_KERNEL);
-		if (!switcher_pte_page(i)) {
-			free_switcher_pte_pages();
-			return -ENOMEM;
-		}
-		populate_switcher_pte_page(i, switcher_page, pages);
-	}
-	return 0;
-}
-/*:*/
-
-/* Cleaning up simply involves freeing the PTE page for each CPU. */
-void free_pagetables(void)
-{
-	free_switcher_pte_pages();
-}
diff --git a/drivers/lguest/x86/core.c b/drivers/lguest/x86/core.c
index 4af12e1844d..f0a3347b644 100644
--- a/drivers/lguest/x86/core.c
+++ b/drivers/lguest/x86/core.c
@@ -59,14 +59,13 @@ static struct {
 /* Offset from where switcher.S was compiled to where we've copied it */
 static unsigned long switcher_offset(void)
 {
-	return SWITCHER_ADDR - (unsigned long)start_switcher_text;
+	return switcher_addr - (unsigned long)start_switcher_text;
 }
 
-/* This cpu's struct lguest_pages. */
+/* This cpu's struct lguest_pages (after the Switcher text page) */
 static struct lguest_pages *lguest_pages(unsigned int cpu)
 {
-	return &(((struct lguest_pages *)
-		  (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
+	return &(((struct lguest_pages *)(switcher_addr + PAGE_SIZE))[cpu]);
 }
 
 static DEFINE_PER_CPU(struct lg_cpu *, lg_last_cpu);
diff --git a/drivers/net/caif/Kconfig b/drivers/net/caif/Kconfig
index a966128c2a7..7ffc756131a 100644
--- a/drivers/net/caif/Kconfig
+++ b/drivers/net/caif/Kconfig
@@ -40,3 +40,17 @@ config CAIF_HSI
        The caif low level driver for CAIF over HSI.
        Be aware that if you enable this then you also need to
        enable a low-level HSI driver.
+
+config CAIF_VIRTIO
+	tristate "CAIF virtio transport driver"
+	depends on CAIF
+	select VHOST_RING
+	select VIRTIO
+	select GENERIC_ALLOCATOR
+	default n
+	---help---
+	The caif driver for CAIF over Virtio.
+
+if CAIF_VIRTIO
+source "drivers/vhost/Kconfig"
+endif
diff --git a/drivers/net/caif/Makefile b/drivers/net/caif/Makefile
index 15a9d2fc753..9bbd45391f6 100644
--- a/drivers/net/caif/Makefile
+++ b/drivers/net/caif/Makefile
@@ -9,3 +9,6 @@ obj-$(CONFIG_CAIF_SPI_SLAVE) += cfspi_slave.o
 
 # HSI interface
 obj-$(CONFIG_CAIF_HSI) += caif_hsi.o
+
+# Virtio interface
+obj-$(CONFIG_CAIF_VIRTIO) += caif_virtio.o
diff --git a/drivers/net/caif/caif_virtio.c b/drivers/net/caif/caif_virtio.c
new file mode 100644
index 00000000000..b9ed1288ce2
--- /dev/null
+++ b/drivers/net/caif/caif_virtio.c
@@ -0,0 +1,790 @@
+/*
+ * Copyright (C) ST-Ericsson AB 2013
+ * Authors: Vicram Arv
+ *	    Dmitry Tarnyagin <dmitry.tarnyagin@lockless.no>
+ *	    Sjur Brendeland
+ * License terms: GNU General Public License (GPL) version 2
+ */
+#include <linux/module.h>
+#include <linux/if_arp.h>
+#include <linux/virtio.h>
+#include <linux/vringh.h>
+#include <linux/debugfs.h>
+#include <linux/spinlock.h>
+#include <linux/genalloc.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/rtnetlink.h>
+#include <linux/virtio_ids.h>
+#include <linux/virtio_caif.h>
+#include <linux/virtio_ring.h>
+#include <linux/dma-mapping.h>
+#include <net/caif/caif_dev.h>
+#include <linux/virtio_config.h>
+
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Vicram Arv");
+MODULE_AUTHOR("Sjur Brendeland");
+MODULE_DESCRIPTION("Virtio CAIF Driver");
+
+/* NAPI schedule quota */
+#define CFV_DEFAULT_QUOTA 32
+
+/* Defaults used if virtio config space is unavailable */
+#define CFV_DEF_MTU_SIZE 4096
+#define CFV_DEF_HEADROOM 32
+#define CFV_DEF_TAILROOM 32
+
+/* Required IP header alignment */
+#define IP_HDR_ALIGN 4
+
+/* struct cfv_napi_contxt - NAPI context info
+ * @riov: IOV holding data read from the ring. Note that riov may
+ *	  still hold data when cfv_rx_poll() returns.
+ * @head: Last descriptor ID we received from vringh_getdesc_kern.
+ *	  We use this to put descriptor back on the used ring. USHRT_MAX is
+ *	  used to indicate invalid head-id.
+ */
+struct cfv_napi_context {
+	struct vringh_kiov riov;
+	unsigned short head;
+};
+
+/* struct cfv_stats - statistics for debugfs
+ * @rx_napi_complete:	Number of NAPI completions (RX)
+ * @rx_napi_resched:	Number of calls where the full quota was used (RX)
+ * @rx_nomem:		Number of SKB alloc failures (RX)
+ * @rx_kicks:		Number of RX kicks
+ * @tx_full_ring:	Number times TX ring was full
+ * @tx_no_mem:		Number of times TX went out of memory
+ * @tx_flow_on:		Number of flow on (TX)
+ * @tx_kicks:		Number of TX kicks
+ */
+struct cfv_stats {
+	u32 rx_napi_complete;
+	u32 rx_napi_resched;
+	u32 rx_nomem;
+	u32 rx_kicks;
+	u32 tx_full_ring;
+	u32 tx_no_mem;
+	u32 tx_flow_on;
+	u32 tx_kicks;
+};
+
+/* struct cfv_info - Caif Virtio control structure
+ * @cfdev:	caif common header
+ * @vdev:	Associated virtio device
+ * @vr_rx:	rx/downlink host vring
+ * @vq_tx:	tx/uplink virtqueue
+ * @ndev:	CAIF link layer device
+ * @watermark_tx: indicates number of free descriptors we need
+ *		to reopen the tx-queues after overload.
+ * @tx_lock:	protects vq_tx from concurrent use
+ * @tx_release_tasklet: Tasklet for freeing consumed TX buffers
+ * @napi:       Napi context used in cfv_rx_poll()
+ * @ctx:        Context data used in cfv_rx_poll()
+ * @tx_hr:	transmit headroom
+ * @rx_hr:	receive headroom
+ * @tx_tr:	transmit tail room
+ * @rx_tr:	receive tail room
+ * @mtu:	transmit max size
+ * @mru:	receive max size
+ * @allocsz:    size of dma memory reserved for TX buffers
+ * @alloc_addr: virtual address to dma memory for TX buffers
+ * @alloc_dma:  dma address to dma memory for TX buffers
+ * @genpool:    Gen Pool used for allocating TX buffers
+ * @reserved_mem: Pointer to memory reserve allocated from genpool
+ * @reserved_size: Size of memory reserve allocated from genpool
+ * @stats:       Statistics exposed in sysfs
+ * @debugfs:    Debugfs dentry for statistic counters
+ */
+struct cfv_info {
+	struct caif_dev_common cfdev;
+	struct virtio_device *vdev;
+	struct vringh *vr_rx;
+	struct virtqueue *vq_tx;
+	struct net_device *ndev;
+	unsigned int watermark_tx;
+	/* Protect access to vq_tx */
+	spinlock_t tx_lock;
+	struct tasklet_struct tx_release_tasklet;
+	struct napi_struct napi;
+	struct cfv_napi_context ctx;
+	u16 tx_hr;
+	u16 rx_hr;
+	u16 tx_tr;
+	u16 rx_tr;
+	u32 mtu;
+	u32 mru;
+	size_t allocsz;
+	void *alloc_addr;
+	dma_addr_t alloc_dma;
+	struct gen_pool *genpool;
+	unsigned long reserved_mem;
+	size_t reserved_size;
+	struct cfv_stats stats;
+	struct dentry *debugfs;
+};
+
+/* struct buf_info - maintains transmit buffer data handle
+ * @size:	size of transmit buffer
+ * @dma_handle: handle to allocated dma device memory area
+ * @vaddr:	virtual address mapping to allocated memory area
+ */
+struct buf_info {
+	size_t size;
+	u8 *vaddr;
+};
+
+/* Called from virtio device, in IRQ context */
+static void cfv_release_cb(struct virtqueue *vq_tx)
+{
+	struct cfv_info *cfv = vq_tx->vdev->priv;
+
+	++cfv->stats.tx_kicks;
+	tasklet_schedule(&cfv->tx_release_tasklet);
+}
+
+static void free_buf_info(struct cfv_info *cfv, struct buf_info *buf_info)
+{
+	if (!buf_info)
+		return;
+	gen_pool_free(cfv->genpool, (unsigned long) buf_info->vaddr,
+		      buf_info->size);
+	kfree(buf_info);
+}
+
+/* This is invoked whenever the remote processor completed processing
+ * a TX msg we just sent, and the buffer is put back to the used ring.
+ */
+static void cfv_release_used_buf(struct virtqueue *vq_tx)
+{
+	struct cfv_info *cfv = vq_tx->vdev->priv;
+	unsigned long flags;
+
+	BUG_ON(vq_tx != cfv->vq_tx);
+
+	for (;;) {
+		unsigned int len;
+		struct buf_info *buf_info;
+
+		/* Get used buffer from used ring to recycle used descriptors */
+		spin_lock_irqsave(&cfv->tx_lock, flags);
+		buf_info = virtqueue_get_buf(vq_tx, &len);
+		spin_unlock_irqrestore(&cfv->tx_lock, flags);
+
+		/* Stop looping if there are no more buffers to free */
+		if (!buf_info)
+			break;
+
+		free_buf_info(cfv, buf_info);
+
+		/* watermark_tx indicates if we previously stopped the tx
+		 * queues. If we have enough free stots in the virtio ring,
+		 * re-establish memory reserved and open up tx queues.
+		 */
+		if (cfv->vq_tx->num_free <= cfv->watermark_tx)
+			continue;
+
+		/* Re-establish memory reserve */
+		if (cfv->reserved_mem == 0 && cfv->genpool)
+			cfv->reserved_mem =
+				gen_pool_alloc(cfv->genpool,
+					       cfv->reserved_size);
+
+		/* Open up the tx queues */
+		if (cfv->reserved_mem) {
+			cfv->watermark_tx =
+				virtqueue_get_vring_size(cfv->vq_tx);
+			netif_tx_wake_all_queues(cfv->ndev);
+			/* Buffers are recycled in cfv_netdev_tx, so
+			 * disable notifications when queues are opened.
+			 */
+			virtqueue_disable_cb(cfv->vq_tx);
+			++cfv->stats.tx_flow_on;
+		} else {
+			/* if no memory reserve, wait for more free slots */
+			WARN_ON(cfv->watermark_tx >
+			       virtqueue_get_vring_size(cfv->vq_tx));
+			cfv->watermark_tx +=
+				virtqueue_get_vring_size(cfv->vq_tx) / 4;
+		}
+	}
+}
+
+/* Allocate a SKB and copy packet data to it */
+static struct sk_buff *cfv_alloc_and_copy_skb(int *err,
+					      struct cfv_info *cfv,
+					      u8 *frm, u32 frm_len)
+{
+	struct sk_buff *skb;
+	u32 cfpkt_len, pad_len;
+
+	*err = 0;
+	/* Verify that packet size with down-link header and mtu size */
+	if (frm_len > cfv->mru || frm_len <= cfv->rx_hr + cfv->rx_tr) {
+		netdev_err(cfv->ndev,
+			   "Invalid frmlen:%u  mtu:%u hr:%d tr:%d\n",
+			   frm_len, cfv->mru,  cfv->rx_hr,
+			   cfv->rx_tr);
+		*err = -EPROTO;
+		return NULL;
+	}
+
+	cfpkt_len = frm_len - (cfv->rx_hr + cfv->rx_tr);
+	pad_len = (unsigned long)(frm + cfv->rx_hr) & (IP_HDR_ALIGN - 1);
+
+	skb = netdev_alloc_skb(cfv->ndev, frm_len + pad_len);
+	if (!skb) {
+		*err = -ENOMEM;
+		return NULL;
+	}
+
+	skb_reserve(skb, cfv->rx_hr + pad_len);
+
+	memcpy(skb_put(skb, cfpkt_len), frm + cfv->rx_hr, cfpkt_len);
+	return skb;
+}
+
+/* Get packets from the host vring */
+static int cfv_rx_poll(struct napi_struct *napi, int quota)
+{
+	struct cfv_info *cfv = container_of(napi, struct cfv_info, napi);
+	int rxcnt = 0;
+	int err = 0;
+	void *buf;
+	struct sk_buff *skb;
+	struct vringh_kiov *riov = &cfv->ctx.riov;
+	unsigned int skb_len;
+
+again:
+	do {
+		skb = NULL;
+
+		/* Put the previous iovec back on the used ring and
+		 * fetch a new iovec if we have processed all elements.
+		 */
+		if (riov->i == riov->used) {
+			if (cfv->ctx.head != USHRT_MAX) {
+				vringh_complete_kern(cfv->vr_rx,
+						     cfv->ctx.head,
+						     0);
+				cfv->ctx.head = USHRT_MAX;
+			}
+
+			err = vringh_getdesc_kern(
+				cfv->vr_rx,
+				riov,
+				NULL,
+				&cfv->ctx.head,
+				GFP_ATOMIC);
+
+			if (err <= 0)
+				goto exit;
+		}
+
+		buf = phys_to_virt((unsigned long) riov->iov[riov->i].iov_base);
+		/* TODO: Add check on valid buffer address */
+
+		skb = cfv_alloc_and_copy_skb(&err, cfv, buf,
+					     riov->iov[riov->i].iov_len);
+		if (unlikely(err))
+			goto exit;
+
+		/* Push received packet up the stack. */
+		skb_len = skb->len;
+		skb->protocol = htons(ETH_P_CAIF);
+		skb_reset_mac_header(skb);
+		skb->dev = cfv->ndev;
+		err = netif_receive_skb(skb);
+		if (unlikely(err)) {
+			++cfv->ndev->stats.rx_dropped;
+		} else {
+			++cfv->ndev->stats.rx_packets;
+			cfv->ndev->stats.rx_bytes += skb_len;
+		}
+
+		++riov->i;
+		++rxcnt;
+	} while (rxcnt < quota);
+
+	++cfv->stats.rx_napi_resched;
+	goto out;
+
+exit:
+	switch (err) {
+	case 0:
+		++cfv->stats.rx_napi_complete;
+
+		/* Really out of patckets? (stolen from virtio_net)*/
+		napi_complete(napi);
+		if (unlikely(!vringh_notify_enable_kern(cfv->vr_rx)) &&
+		    napi_schedule_prep(napi)) {
+			vringh_notify_disable_kern(cfv->vr_rx);
+			__napi_schedule(napi);
+			goto again;
+		}
+		break;
+
+	case -ENOMEM:
+		++cfv->stats.rx_nomem;
+		dev_kfree_skb(skb);
+		/* Stop NAPI poll on OOM, we hope to be polled later */
+		napi_complete(napi);
+		vringh_notify_enable_kern(cfv->vr_rx);
+		break;
+
+	default:
+		/* We're doomed, any modem fault is fatal */
+		netdev_warn(cfv->ndev, "Bad ring, disable device\n");
+		cfv->ndev->stats.rx_dropped = riov->used - riov->i;
+		napi_complete(napi);
+		vringh_notify_disable_kern(cfv->vr_rx);
+		netif_carrier_off(cfv->ndev);
+		break;
+	}
+out:
+	if (rxcnt && vringh_need_notify_kern(cfv->vr_rx) > 0)
+		vringh_notify(cfv->vr_rx);
+	return rxcnt;
+}
+
+static void cfv_recv(struct virtio_device *vdev, struct vringh *vr_rx)
+{
+	struct cfv_info *cfv = vdev->priv;
+
+	++cfv->stats.rx_kicks;
+	vringh_notify_disable_kern(cfv->vr_rx);
+	napi_schedule(&cfv->napi);
+}
+
+static void cfv_destroy_genpool(struct cfv_info *cfv)
+{
+	if (cfv->alloc_addr)
+		dma_free_coherent(cfv->vdev->dev.parent->parent,
+				  cfv->allocsz, cfv->alloc_addr,
+				  cfv->alloc_dma);
+
+	if (!cfv->genpool)
+		return;
+	gen_pool_free(cfv->genpool,  cfv->reserved_mem,
+		      cfv->reserved_size);
+	gen_pool_destroy(cfv->genpool);
+	cfv->genpool = NULL;
+}
+
+static int cfv_create_genpool(struct cfv_info *cfv)
+{
+	int err;
+
+	/* dma_alloc can only allocate whole pages, and we need a more
+	 * fine graned allocation so we use genpool. We ask for space needed
+	 * by IP and a full ring. If the dma allcoation fails we retry with a
+	 * smaller allocation size.
+	 */
+	err = -ENOMEM;
+	cfv->allocsz = (virtqueue_get_vring_size(cfv->vq_tx) *
+			(ETH_DATA_LEN + cfv->tx_hr + cfv->tx_tr) * 11)/10;
+	if (cfv->allocsz <= (num_possible_cpus() + 1) * cfv->ndev->mtu)
+		return -EINVAL;
+
+	for (;;) {
+		if (cfv->allocsz <= num_possible_cpus() * cfv->ndev->mtu) {
+			netdev_info(cfv->ndev, "Not enough device memory\n");
+			return -ENOMEM;
+		}
+
+		cfv->alloc_addr = dma_alloc_coherent(
+						cfv->vdev->dev.parent->parent,
+						cfv->allocsz, &cfv->alloc_dma,
+						GFP_ATOMIC);
+		if (cfv->alloc_addr)
+			break;
+
+		cfv->allocsz = (cfv->allocsz * 3) >> 2;
+	}
+
+	netdev_dbg(cfv->ndev, "Allocated %zd bytes from dma-memory\n",
+		   cfv->allocsz);
+
+	/* Allocate on 128 bytes boundaries (1 << 7)*/
+	cfv->genpool = gen_pool_create(7, -1);
+	if (!cfv->genpool)
+		goto err;
+
+	err = gen_pool_add_virt(cfv->genpool, (unsigned long)cfv->alloc_addr,
+				(phys_addr_t)virt_to_phys(cfv->alloc_addr),
+				cfv->allocsz, -1);
+	if (err)
+		goto err;
+
+	/* Reserve some memory for low memory situations. If we hit the roof
+	 * in the memory pool, we stop TX flow and release the reserve.
+	 */
+	cfv->reserved_size = num_possible_cpus() * cfv->ndev->mtu;
+	cfv->reserved_mem = gen_pool_alloc(cfv->genpool,
+					   cfv->reserved_size);
+	if (!cfv->reserved_mem) {
+		err = -ENOMEM;
+		goto err;
+	}
+
+	cfv->watermark_tx = virtqueue_get_vring_size(cfv->vq_tx);
+	return 0;
+err:
+	cfv_destroy_genpool(cfv);
+	return err;
+}
+
+/* Enable the CAIF interface and allocate the memory-pool */
+static int cfv_netdev_open(struct net_device *netdev)
+{
+	struct cfv_info *cfv = netdev_priv(netdev);
+
+	if (cfv_create_genpool(cfv))
+		return -ENOMEM;
+
+	netif_carrier_on(netdev);
+	napi_enable(&cfv->napi);
+
+	/* Schedule NAPI to read any pending packets */
+	napi_schedule(&cfv->napi);
+	return 0;
+}
+
+/* Disable the CAIF interface and free the memory-pool */
+static int cfv_netdev_close(struct net_device *netdev)
+{
+	struct cfv_info *cfv = netdev_priv(netdev);
+	unsigned long flags;
+	struct buf_info *buf_info;
+
+	/* Disable interrupts, queues and NAPI polling */
+	netif_carrier_off(netdev);
+	virtqueue_disable_cb(cfv->vq_tx);
+	vringh_notify_disable_kern(cfv->vr_rx);
+	napi_disable(&cfv->napi);
+
+	/* Release any TX buffers on both used and avilable rings */
+	cfv_release_used_buf(cfv->vq_tx);
+	spin_lock_irqsave(&cfv->tx_lock, flags);
+	while ((buf_info = virtqueue_detach_unused_buf(cfv->vq_tx)))
+		free_buf_info(cfv, buf_info);
+	spin_unlock_irqrestore(&cfv->tx_lock, flags);
+
+	/* Release all dma allocated memory and destroy the pool */
+	cfv_destroy_genpool(cfv);
+	return 0;
+}
+
+/* Allocate a buffer in dma-memory and copy skb to it */
+static struct buf_info *cfv_alloc_and_copy_to_shm(struct cfv_info *cfv,
+						       struct sk_buff *skb,
+						       struct scatterlist *sg)
+{
+	struct caif_payload_info *info = (void *)&skb->cb;
+	struct buf_info *buf_info = NULL;
+	u8 pad_len, hdr_ofs;
+
+	if (!cfv->genpool)
+		goto err;
+
+	if (unlikely(cfv->tx_hr + skb->len + cfv->tx_tr > cfv->mtu)) {
+		netdev_warn(cfv->ndev, "Invalid packet len (%d > %d)\n",
+			    cfv->tx_hr + skb->len + cfv->tx_tr, cfv->mtu);
+		goto err;
+	}
+
+	buf_info = kmalloc(sizeof(struct buf_info), GFP_ATOMIC);
+	if (unlikely(!buf_info))
+		goto err;
+
+	/* Make the IP header aligned in tbe buffer */
+	hdr_ofs = cfv->tx_hr + info->hdr_len;
+	pad_len = hdr_ofs & (IP_HDR_ALIGN - 1);
+	buf_info->size = cfv->tx_hr + skb->len + cfv->tx_tr + pad_len;
+
+	/* allocate dma memory buffer */
+	buf_info->vaddr = (void *)gen_pool_alloc(cfv->genpool, buf_info->size);
+	if (unlikely(!buf_info->vaddr))
+		goto err;
+
+	/* copy skbuf contents to send buffer */
+	skb_copy_bits(skb, 0, buf_info->vaddr + cfv->tx_hr + pad_len, skb->len);
+	sg_init_one(sg, buf_info->vaddr + pad_len,
+		    skb->len + cfv->tx_hr + cfv->rx_hr);
+
+	return buf_info;
+err:
+	kfree(buf_info);
+	return NULL;
+}
+
+/* Put the CAIF packet on the virtio ring and kick the receiver */
+static int cfv_netdev_tx(struct sk_buff *skb, struct net_device *netdev)
+{
+	struct cfv_info *cfv = netdev_priv(netdev);
+	struct buf_info *buf_info;
+	struct scatterlist sg;
+	unsigned long flags;
+	bool flow_off = false;
+	int ret;
+
+	/* garbage collect released buffers */
+	cfv_release_used_buf(cfv->vq_tx);
+	spin_lock_irqsave(&cfv->tx_lock, flags);
+
+	/* Flow-off check takes into account number of cpus to make sure
+	 * virtqueue will not be overfilled in any possible smp conditions.
+	 *
+	 * Flow-on is triggered when sufficient buffers are freed
+	 */
+	if (unlikely(cfv->vq_tx->num_free <= num_present_cpus())) {
+		flow_off = true;
+		cfv->stats.tx_full_ring++;
+	}
+
+	/* If we run out of memory, we release the memory reserve and retry
+	 * allocation.
+	 */
+	buf_info = cfv_alloc_and_copy_to_shm(cfv, skb, &sg);
+	if (unlikely(!buf_info)) {
+		cfv->stats.tx_no_mem++;
+		flow_off = true;
+
+		if (cfv->reserved_mem && cfv->genpool) {
+			gen_pool_free(cfv->genpool,  cfv->reserved_mem,
+				      cfv->reserved_size);
+			cfv->reserved_mem = 0;
+			buf_info = cfv_alloc_and_copy_to_shm(cfv, skb, &sg);
+		}
+	}
+
+	if (unlikely(flow_off)) {
+		/* Turn flow on when a 1/4 of the descriptors are released */
+		cfv->watermark_tx = virtqueue_get_vring_size(cfv->vq_tx) / 4;
+		/* Enable notifications of recycled TX buffers */
+		virtqueue_enable_cb(cfv->vq_tx);
+		netif_tx_stop_all_queues(netdev);
+	}
+
+	if (unlikely(!buf_info)) {
+		/* If the memory reserve does it's job, this shouldn't happen */
+		netdev_warn(cfv->ndev, "Out of gen_pool memory\n");
+		goto err;
+	}
+
+	ret = virtqueue_add_outbuf(cfv->vq_tx, &sg, 1, buf_info, GFP_ATOMIC);
+	if (unlikely((ret < 0))) {
+		/* If flow control works, this shouldn't happen */
+		netdev_warn(cfv->ndev, "Failed adding buffer to TX vring:%d\n",
+			    ret);
+		goto err;
+	}
+
+	/* update netdev statistics */
+	cfv->ndev->stats.tx_packets++;
+	cfv->ndev->stats.tx_bytes += skb->len;
+	spin_unlock_irqrestore(&cfv->tx_lock, flags);
+
+	/* tell the remote processor it has a pending message to read */
+	virtqueue_kick(cfv->vq_tx);
+
+	dev_kfree_skb(skb);
+	return NETDEV_TX_OK;
+err:
+	spin_unlock_irqrestore(&cfv->tx_lock, flags);
+	cfv->ndev->stats.tx_dropped++;
+	free_buf_info(cfv, buf_info);
+	dev_kfree_skb(skb);
+	return NETDEV_TX_OK;
+}
+
+static void cfv_tx_release_tasklet(unsigned long drv)
+{
+	struct cfv_info *cfv = (struct cfv_info *)drv;
+	cfv_release_used_buf(cfv->vq_tx);
+}
+
+static const struct net_device_ops cfv_netdev_ops = {
+	.ndo_open = cfv_netdev_open,
+	.ndo_stop = cfv_netdev_close,
+	.ndo_start_xmit = cfv_netdev_tx,
+};
+
+static void cfv_netdev_setup(struct net_device *netdev)
+{
+	netdev->netdev_ops = &cfv_netdev_ops;
+	netdev->type = ARPHRD_CAIF;
+	netdev->tx_queue_len = 100;
+	netdev->flags = IFF_POINTOPOINT | IFF_NOARP;
+	netdev->mtu = CFV_DEF_MTU_SIZE;
+	netdev->destructor = free_netdev;
+}
+
+/* Create debugfs counters for the device */
+static inline void debugfs_init(struct cfv_info *cfv)
+{
+	cfv->debugfs =
+		debugfs_create_dir(netdev_name(cfv->ndev), NULL);
+
+	if (IS_ERR(cfv->debugfs))
+		return;
+
+	debugfs_create_u32("rx-napi-complete", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.rx_napi_complete);
+	debugfs_create_u32("rx-napi-resched", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.rx_napi_resched);
+	debugfs_create_u32("rx-nomem", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.rx_nomem);
+	debugfs_create_u32("rx-kicks", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.rx_kicks);
+	debugfs_create_u32("tx-full-ring", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.tx_full_ring);
+	debugfs_create_u32("tx-no-mem", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.tx_no_mem);
+	debugfs_create_u32("tx-kicks", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.tx_kicks);
+	debugfs_create_u32("tx-flow-on", S_IRUSR, cfv->debugfs,
+			   &cfv->stats.tx_flow_on);
+}
+
+/* Setup CAIF for the a virtio device */
+static int cfv_probe(struct virtio_device *vdev)
+{
+	vq_callback_t *vq_cbs = cfv_release_cb;
+	vrh_callback_t *vrh_cbs = cfv_recv;
+	const char *names =  "output";
+	const char *cfv_netdev_name = "cfvrt";
+	struct net_device *netdev;
+	struct cfv_info *cfv;
+	int err = -EINVAL;
+
+	netdev = alloc_netdev(sizeof(struct cfv_info), cfv_netdev_name,
+			      cfv_netdev_setup);
+	if (!netdev)
+		return -ENOMEM;
+
+	cfv = netdev_priv(netdev);
+	cfv->vdev = vdev;
+	cfv->ndev = netdev;
+
+	spin_lock_init(&cfv->tx_lock);
+
+	/* Get the RX virtio ring. This is a "host side vring". */
+	err = -ENODEV;
+	if (!vdev->vringh_config || !vdev->vringh_config->find_vrhs)
+		goto err;
+
+	err = vdev->vringh_config->find_vrhs(vdev, 1, &cfv->vr_rx, &vrh_cbs);
+	if (err)
+		goto err;
+
+	/* Get the TX virtio ring. This is a "guest side vring". */
+	err = vdev->config->find_vqs(vdev, 1, &cfv->vq_tx, &vq_cbs, &names);
+	if (err)
+		goto err;
+
+	/* Get the CAIF configuration from virtio config space, if available */
+#define GET_VIRTIO_CONFIG_OPS(_v, _var, _f) \
+	((_v)->config->get(_v, offsetof(struct virtio_caif_transf_config, _f), \
+			   &_var, \
+			   FIELD_SIZEOF(struct virtio_caif_transf_config, _f)))
+
+	if (vdev->config->get) {
+		GET_VIRTIO_CONFIG_OPS(vdev, cfv->tx_hr, headroom);
+		GET_VIRTIO_CONFIG_OPS(vdev, cfv->rx_hr, headroom);
+		GET_VIRTIO_CONFIG_OPS(vdev, cfv->tx_tr, tailroom);
+		GET_VIRTIO_CONFIG_OPS(vdev, cfv->rx_tr, tailroom);
+		GET_VIRTIO_CONFIG_OPS(vdev, cfv->mtu, mtu);
+		GET_VIRTIO_CONFIG_OPS(vdev, cfv->mru, mtu);
+	} else {
+		cfv->tx_hr = CFV_DEF_HEADROOM;
+		cfv->rx_hr = CFV_DEF_HEADROOM;
+		cfv->tx_tr = CFV_DEF_TAILROOM;
+		cfv->rx_tr = CFV_DEF_TAILROOM;
+		cfv->mtu = CFV_DEF_MTU_SIZE;
+		cfv->mru = CFV_DEF_MTU_SIZE;
+	}
+
+	netdev->needed_headroom = cfv->tx_hr;
+	netdev->needed_tailroom = cfv->tx_tr;
+
+	/* Disable buffer release interrupts unless we have stopped TX queues */
+	virtqueue_disable_cb(cfv->vq_tx);
+
+	netdev->mtu = cfv->mtu - cfv->tx_tr;
+	vdev->priv = cfv;
+
+	/* Initialize NAPI poll context data */
+	vringh_kiov_init(&cfv->ctx.riov, NULL, 0);
+	cfv->ctx.head = USHRT_MAX;
+	netif_napi_add(netdev, &cfv->napi, cfv_rx_poll, CFV_DEFAULT_QUOTA);
+
+	tasklet_init(&cfv->tx_release_tasklet,
+		     cfv_tx_release_tasklet,
+		     (unsigned long)cfv);
+
+	/* Carrier is off until netdevice is opened */
+	netif_carrier_off(netdev);
+
+	/* register Netdev */
+	err = register_netdev(netdev);
+	if (err) {
+		dev_err(&vdev->dev, "Unable to register netdev (%d)\n", err);
+		goto err;
+	}
+
+	debugfs_init(cfv);
+
+	return 0;
+err:
+	netdev_warn(cfv->ndev, "CAIF Virtio probe failed:%d\n", err);
+
+	if (cfv->vr_rx)
+		vdev->vringh_config->del_vrhs(cfv->vdev);
+	if (cfv->vdev)
+		vdev->config->del_vqs(cfv->vdev);
+	free_netdev(netdev);
+	return err;
+}
+
+static void cfv_remove(struct virtio_device *vdev)
+{
+	struct cfv_info *cfv = vdev->priv;
+
+	rtnl_lock();
+	dev_close(cfv->ndev);
+	rtnl_unlock();
+
+	tasklet_kill(&cfv->tx_release_tasklet);
+	debugfs_remove_recursive(cfv->debugfs);
+
+	vringh_kiov_cleanup(&cfv->ctx.riov);
+	vdev->config->reset(vdev);
+	vdev->vringh_config->del_vrhs(cfv->vdev);
+	cfv->vr_rx = NULL;
+	vdev->config->del_vqs(cfv->vdev);
+	unregister_netdev(cfv->ndev);
+}
+
+static struct virtio_device_id id_table[] = {
+	{ VIRTIO_ID_CAIF, VIRTIO_DEV_ANY_ID },
+	{ 0 },
+};
+
+static unsigned int features[] = {
+};
+
+static struct virtio_driver caif_virtio_driver = {
+	.feature_table		= features,
+	.feature_table_size	= ARRAY_SIZE(features),
+	.driver.name		= KBUILD_MODNAME,
+	.driver.owner		= THIS_MODULE,
+	.id_table		= id_table,
+	.probe			= cfv_probe,
+	.remove			= cfv_remove,
+};
+
+module_virtio_driver(caif_virtio_driver);
+MODULE_DEVICE_TABLE(virtio, id_table);
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 50077753a0e..3c23fdc27bf 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -39,7 +39,6 @@ module_param(gso, bool, 0444);
 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
 #define GOOD_COPY_LEN	128
 
-#define VIRTNET_SEND_COMMAND_SG_MAX    2
 #define VIRTNET_DRIVER_VERSION "1.0.0"
 
 struct virtnet_stats {
@@ -444,7 +443,7 @@ static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
 
 	skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
 
-	err = virtqueue_add_buf(rq->vq, rq->sg, 0, 2, skb, gfp);
+	err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
 	if (err < 0)
 		dev_kfree_skb(skb);
 
@@ -489,8 +488,8 @@ static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
 
 	/* chain first in list head */
 	first->private = (unsigned long)list;
-	err = virtqueue_add_buf(rq->vq, rq->sg, 0, MAX_SKB_FRAGS + 2,
-				first, gfp);
+	err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
+				  first, gfp);
 	if (err < 0)
 		give_pages(rq, first);
 
@@ -508,7 +507,7 @@ static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
 
 	sg_init_one(rq->sg, page_address(page), PAGE_SIZE);
 
-	err = virtqueue_add_buf(rq->vq, rq->sg, 0, 1, page, gfp);
+	err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, page, gfp);
 	if (err < 0)
 		give_pages(rq, page);
 
@@ -582,7 +581,7 @@ static void refill_work(struct work_struct *work)
 	bool still_empty;
 	int i;
 
-	for (i = 0; i < vi->max_queue_pairs; i++) {
+	for (i = 0; i < vi->curr_queue_pairs; i++) {
 		struct receive_queue *rq = &vi->rq[i];
 
 		napi_disable(&rq->napi);
@@ -637,7 +636,7 @@ static int virtnet_open(struct net_device *dev)
 	struct virtnet_info *vi = netdev_priv(dev);
 	int i;
 
-	for (i = 0; i < vi->max_queue_pairs; i++) {
+	for (i = 0; i < vi->curr_queue_pairs; i++) {
 		/* Make sure we have some buffers: if oom use wq. */
 		if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
 			schedule_delayed_work(&vi->refill, 0);
@@ -711,8 +710,7 @@ static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
 		sg_set_buf(sq->sg, &hdr->hdr, sizeof hdr->hdr);
 
 	num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
-	return virtqueue_add_buf(sq->vq, sq->sg, num_sg,
-				 0, skb, GFP_ATOMIC);
+	return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
 }
 
 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
@@ -767,32 +765,35 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
  * never fail unless improperly formated.
  */
 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
-				 struct scatterlist *data, int out, int in)
+				 struct scatterlist *out,
+				 struct scatterlist *in)
 {
-	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
+	struct scatterlist *sgs[4], hdr, stat;
 	struct virtio_net_ctrl_hdr ctrl;
 	virtio_net_ctrl_ack status = ~0;
-	unsigned int tmp;
-	int i;
+	unsigned out_num = 0, in_num = 0, tmp;
 
 	/* Caller should know better */
-	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
-		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
-
-	out++; /* Add header */
-	in++; /* Add return status */
+	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
 
 	ctrl.class = class;
 	ctrl.cmd = cmd;
+	/* Add header */
+	sg_init_one(&hdr, &ctrl, sizeof(ctrl));
+	sgs[out_num++] = &hdr;
 
-	sg_init_table(sg, out + in);
+	if (out)
+		sgs[out_num++] = out;
+	if (in)
+		sgs[out_num + in_num++] = in;
 
-	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
-	for_each_sg(data, s, out + in - 2, i)
-		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
-	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
+	/* Add return status. */
+	sg_init_one(&stat, &status, sizeof(status));
+	sgs[out_num + in_num++] = &stat;
 
-	BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
+	BUG_ON(out_num + in_num > ARRAY_SIZE(sgs));
+	BUG_ON(virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC)
+	       < 0);
 
 	virtqueue_kick(vi->cvq);
 
@@ -821,7 +822,7 @@ static int virtnet_set_mac_address(struct net_device *dev, void *p)
 		sg_init_one(&sg, addr->sa_data, dev->addr_len);
 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 					  VIRTIO_NET_CTRL_MAC_ADDR_SET,
-					  &sg, 1, 0)) {
+					  &sg, NULL)) {
 			dev_warn(&vdev->dev,
 				 "Failed to set mac address by vq command.\n");
 			return -EINVAL;
@@ -889,8 +890,7 @@ static void virtnet_ack_link_announce(struct virtnet_info *vi)
 {
 	rtnl_lock();
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
-				  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
-				  0, 0))
+				  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL, NULL))
 		dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
 	rtnl_unlock();
 }
@@ -900,6 +900,7 @@ static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
 	struct scatterlist sg;
 	struct virtio_net_ctrl_mq s;
 	struct net_device *dev = vi->dev;
+	int i;
 
 	if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
 		return 0;
@@ -908,12 +909,16 @@ static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
 	sg_init_one(&sg, &s, sizeof(s));
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
-				  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, 1, 0)){
+				  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, NULL)) {
 		dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
 			 queue_pairs);
 		return -EINVAL;
-	} else
+	} else {
+		for (i = vi->curr_queue_pairs; i < queue_pairs; i++)
+			if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
+				schedule_delayed_work(&vi->refill, 0);
 		vi->curr_queue_pairs = queue_pairs;
+	}
 
 	return 0;
 }
@@ -955,7 +960,7 @@ static void virtnet_set_rx_mode(struct net_device *dev)
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 				  VIRTIO_NET_CTRL_RX_PROMISC,
-				  sg, 1, 0))
+				  sg, NULL))
 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 			 promisc ? "en" : "dis");
 
@@ -963,7 +968,7 @@ static void virtnet_set_rx_mode(struct net_device *dev)
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
-				  sg, 1, 0))
+				  sg, NULL))
 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 			 allmulti ? "en" : "dis");
 
@@ -1000,7 +1005,7 @@ static void virtnet_set_rx_mode(struct net_device *dev)
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
-				  sg, 2, 0))
+				  sg, NULL))
 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 
 	kfree(buf);
@@ -1015,7 +1020,7 @@ static int virtnet_vlan_rx_add_vid(struct net_device *dev,
 	sg_init_one(&sg, &vid, sizeof(vid));
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
-				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
+				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, NULL))
 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 	return 0;
 }
@@ -1029,7 +1034,7 @@ static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
 	sg_init_one(&sg, &vid, sizeof(vid));
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
-				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
+				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, NULL))
 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 	return 0;
 }
@@ -1570,7 +1575,7 @@ static int virtnet_probe(struct virtio_device *vdev)
 	}
 
 	/* Last of all, set up some receive buffers. */
-	for (i = 0; i < vi->max_queue_pairs; i++) {
+	for (i = 0; i < vi->curr_queue_pairs; i++) {
 		try_fill_recv(&vi->rq[i], GFP_KERNEL);
 
 		/* If we didn't even get one input buffer, we're useless. */
@@ -1694,7 +1699,7 @@ static int virtnet_restore(struct virtio_device *vdev)
 
 	netif_device_attach(vi->dev);
 
-	for (i = 0; i < vi->max_queue_pairs; i++)
+	for (i = 0; i < vi->curr_queue_pairs; i++)
 		if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
 			schedule_delayed_work(&vi->refill, 0);
 
diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c
index 7861f1119b7..56fceafec9e 100644
--- a/drivers/rpmsg/virtio_rpmsg_bus.c
+++ b/drivers/rpmsg/virtio_rpmsg_bus.c
@@ -757,14 +757,14 @@ int rpmsg_send_offchannel_raw(struct rpmsg_channel *rpdev, u32 src, u32 dst,
 	mutex_lock(&vrp->tx_lock);
 
 	/* add message to the remote processor's virtqueue */
-	err = virtqueue_add_buf(vrp->svq, &sg, 1, 0, msg, GFP_KERNEL);
+	err = virtqueue_add_outbuf(vrp->svq, &sg, 1, msg, GFP_KERNEL);
 	if (err) {
 		/*
 		 * need to reclaim the buffer here, otherwise it's lost
 		 * (memory won't leak, but rpmsg won't use it again for TX).
 		 * this will wait for a buffer management overhaul.
 		 */
-		dev_err(dev, "virtqueue_add_buf failed: %d\n", err);
+		dev_err(dev, "virtqueue_add_outbuf failed: %d\n", err);
 		goto out;
 	}
 
@@ -839,7 +839,7 @@ static void rpmsg_recv_done(struct virtqueue *rvq)
 	sg_init_one(&sg, msg, RPMSG_BUF_SIZE);
 
 	/* add the buffer back to the remote processor's virtqueue */
-	err = virtqueue_add_buf(vrp->rvq, &sg, 0, 1, msg, GFP_KERNEL);
+	err = virtqueue_add_inbuf(vrp->rvq, &sg, 1, msg, GFP_KERNEL);
 	if (err < 0) {
 		dev_err(dev, "failed to add a virtqueue buffer: %d\n", err);
 		return;
@@ -972,7 +972,7 @@ static int rpmsg_probe(struct virtio_device *vdev)
 
 		sg_init_one(&sg, cpu_addr, RPMSG_BUF_SIZE);
 
-		err = virtqueue_add_buf(vrp->rvq, &sg, 0, 1, cpu_addr,
+		err = virtqueue_add_inbuf(vrp->rvq, &sg, 1, cpu_addr,
 								GFP_KERNEL);
 		WARN_ON(err); /* sanity check; this can't really happen */
 	}
diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
index 3449a1f8c65..2168258fb2c 100644
--- a/drivers/scsi/virtio_scsi.c
+++ b/drivers/scsi/virtio_scsi.c
@@ -13,6 +13,8 @@
  *
  */
 
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
 #include <linux/module.h>
 #include <linux/slab.h>
 #include <linux/mempool.h>
@@ -20,12 +22,14 @@
 #include <linux/virtio_ids.h>
 #include <linux/virtio_config.h>
 #include <linux/virtio_scsi.h>
+#include <linux/cpu.h>
 #include <scsi/scsi_host.h>
 #include <scsi/scsi_device.h>
 #include <scsi/scsi_cmnd.h>
 
 #define VIRTIO_SCSI_MEMPOOL_SZ 64
 #define VIRTIO_SCSI_EVENT_LEN 8
+#define VIRTIO_SCSI_VQ_BASE 2
 
 /* Command queue element */
 struct virtio_scsi_cmd {
@@ -57,27 +61,61 @@ struct virtio_scsi_vq {
 	struct virtqueue *vq;
 };
 
-/* Per-target queue state */
+/*
+ * Per-target queue state.
+ *
+ * This struct holds the data needed by the queue steering policy.  When a
+ * target is sent multiple requests, we need to drive them to the same queue so
+ * that FIFO processing order is kept.  However, if a target was idle, we can
+ * choose a queue arbitrarily.  In this case the queue is chosen according to
+ * the current VCPU, so the driver expects the number of request queues to be
+ * equal to the number of VCPUs.  This makes it easy and fast to select the
+ * queue, and also lets the driver optimize the IRQ affinity for the virtqueues
+ * (each virtqueue's affinity is set to the CPU that "owns" the queue).
+ *
+ * An interesting effect of this policy is that only writes to req_vq need to
+ * take the tgt_lock.  Read can be done outside the lock because:
+ *
+ * - writes of req_vq only occur when atomic_inc_return(&tgt->reqs) returns 1.
+ *   In that case, no other CPU is reading req_vq: even if they were in
+ *   virtscsi_queuecommand_multi, they would be spinning on tgt_lock.
+ *
+ * - reads of req_vq only occur when the target is not idle (reqs != 0).
+ *   A CPU that enters virtscsi_queuecommand_multi will not modify req_vq.
+ *
+ * Similarly, decrements of reqs are never concurrent with writes of req_vq.
+ * Thus they can happen outside the tgt_lock, provided of course we make reqs
+ * an atomic_t.
+ */
 struct virtio_scsi_target_state {
-	/* Protects sg.  Lock hierarchy is tgt_lock -> vq_lock.  */
+	/* This spinlock never held at the same time as vq_lock. */
 	spinlock_t tgt_lock;
 
-	/* For sglist construction when adding commands to the virtqueue.  */
-	struct scatterlist sg[];
+	/* Count of outstanding requests. */
+	atomic_t reqs;
+
+	/* Currently active virtqueue for requests sent to this target. */
+	struct virtio_scsi_vq *req_vq;
 };
 
 /* Driver instance state */
 struct virtio_scsi {
 	struct virtio_device *vdev;
 
-	struct virtio_scsi_vq ctrl_vq;
-	struct virtio_scsi_vq event_vq;
-	struct virtio_scsi_vq req_vq;
-
 	/* Get some buffers ready for event vq */
 	struct virtio_scsi_event_node event_list[VIRTIO_SCSI_EVENT_LEN];
 
-	struct virtio_scsi_target_state *tgt[];
+	u32 num_queues;
+
+	/* If the affinity hint is set for virtqueues */
+	bool affinity_hint_set;
+
+	/* CPU hotplug notifier */
+	struct notifier_block nb;
+
+	struct virtio_scsi_vq ctrl_vq;
+	struct virtio_scsi_vq event_vq;
+	struct virtio_scsi_vq req_vqs[];
 };
 
 static struct kmem_cache *virtscsi_cmd_cache;
@@ -107,11 +145,13 @@ static void virtscsi_compute_resid(struct scsi_cmnd *sc, u32 resid)
  *
  * Called with vq_lock held.
  */
-static void virtscsi_complete_cmd(void *buf)
+static void virtscsi_complete_cmd(struct virtio_scsi *vscsi, void *buf)
 {
 	struct virtio_scsi_cmd *cmd = buf;
 	struct scsi_cmnd *sc = cmd->sc;
 	struct virtio_scsi_cmd_resp *resp = &cmd->resp.cmd;
+	struct virtio_scsi_target_state *tgt =
+				scsi_target(sc->device)->hostdata;
 
 	dev_dbg(&sc->device->sdev_gendev,
 		"cmd %p response %u status %#02x sense_len %u\n",
@@ -166,32 +206,71 @@ static void virtscsi_complete_cmd(void *buf)
 
 	mempool_free(cmd, virtscsi_cmd_pool);
 	sc->scsi_done(sc);
+
+	atomic_dec(&tgt->reqs);
 }
 
-static void virtscsi_vq_done(struct virtqueue *vq, void (*fn)(void *buf))
+static void virtscsi_vq_done(struct virtio_scsi *vscsi,
+			     struct virtio_scsi_vq *virtscsi_vq,
+			     void (*fn)(struct virtio_scsi *vscsi, void *buf))
 {
 	void *buf;
 	unsigned int len;
+	unsigned long flags;
+	struct virtqueue *vq = virtscsi_vq->vq;
 
+	spin_lock_irqsave(&virtscsi_vq->vq_lock, flags);
 	do {
 		virtqueue_disable_cb(vq);
 		while ((buf = virtqueue_get_buf(vq, &len)) != NULL)
-			fn(buf);
+			fn(vscsi, buf);
 	} while (!virtqueue_enable_cb(vq));
+	spin_unlock_irqrestore(&virtscsi_vq->vq_lock, flags);
 }
 
 static void virtscsi_req_done(struct virtqueue *vq)
 {
 	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
 	struct virtio_scsi *vscsi = shost_priv(sh);
-	unsigned long flags;
+	int index = vq->index - VIRTIO_SCSI_VQ_BASE;
+	struct virtio_scsi_vq *req_vq = &vscsi->req_vqs[index];
 
-	spin_lock_irqsave(&vscsi->req_vq.vq_lock, flags);
-	virtscsi_vq_done(vq, virtscsi_complete_cmd);
-	spin_unlock_irqrestore(&vscsi->req_vq.vq_lock, flags);
+	/*
+	 * Read req_vq before decrementing the reqs field in
+	 * virtscsi_complete_cmd.
+	 *
+	 * With barriers:
+	 *
+	 * 	CPU #0			virtscsi_queuecommand_multi (CPU #1)
+	 * 	------------------------------------------------------------
+	 * 	lock vq_lock
+	 * 	read req_vq
+	 * 	read reqs (reqs = 1)
+	 * 	write reqs (reqs = 0)
+	 * 				increment reqs (reqs = 1)
+	 * 				write req_vq
+	 *
+	 * Possible reordering without barriers:
+	 *
+	 * 	CPU #0			virtscsi_queuecommand_multi (CPU #1)
+	 * 	------------------------------------------------------------
+	 * 	lock vq_lock
+	 * 	read reqs (reqs = 1)
+	 * 	write reqs (reqs = 0)
+	 * 				increment reqs (reqs = 1)
+	 * 				write req_vq
+	 * 	read (wrong) req_vq
+	 *
+	 * We do not need a full smp_rmb, because req_vq is required to get
+	 * to tgt->reqs: tgt is &vscsi->tgt[sc->device->id], where sc is stored
+	 * in the virtqueue as the user token.
+	 */
+	smp_read_barrier_depends();
+
+	virtscsi_vq_done(vscsi, req_vq, virtscsi_complete_cmd);
 };
 
-static void virtscsi_complete_free(void *buf)
+static void virtscsi_complete_free(struct virtio_scsi *vscsi, void *buf)
 {
 	struct virtio_scsi_cmd *cmd = buf;
 
@@ -205,11 +284,8 @@ static void virtscsi_ctrl_done(struct virtqueue *vq)
 {
 	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
 	struct virtio_scsi *vscsi = shost_priv(sh);
-	unsigned long flags;
 
-	spin_lock_irqsave(&vscsi->ctrl_vq.vq_lock, flags);
-	virtscsi_vq_done(vq, virtscsi_complete_free);
-	spin_unlock_irqrestore(&vscsi->ctrl_vq.vq_lock, flags);
+	virtscsi_vq_done(vscsi, &vscsi->ctrl_vq, virtscsi_complete_free);
 };
 
 static int virtscsi_kick_event(struct virtio_scsi *vscsi,
@@ -223,8 +299,8 @@ static int virtscsi_kick_event(struct virtio_scsi *vscsi,
 
 	spin_lock_irqsave(&vscsi->event_vq.vq_lock, flags);
 
-	err = virtqueue_add_buf(vscsi->event_vq.vq, &sg, 0, 1, event_node,
-				GFP_ATOMIC);
+	err = virtqueue_add_inbuf(vscsi->event_vq.vq, &sg, 1, event_node,
+				  GFP_ATOMIC);
 	if (!err)
 		virtqueue_kick(vscsi->event_vq.vq);
 
@@ -254,7 +330,7 @@ static void virtscsi_cancel_event_work(struct virtio_scsi *vscsi)
 }
 
 static void virtscsi_handle_transport_reset(struct virtio_scsi *vscsi,
-						struct virtio_scsi_event *event)
+					    struct virtio_scsi_event *event)
 {
 	struct scsi_device *sdev;
 	struct Scsi_Host *shost = virtio_scsi_host(vscsi->vdev);
@@ -332,7 +408,7 @@ static void virtscsi_handle_event(struct work_struct *work)
 	virtscsi_kick_event(vscsi, event_node);
 }
 
-static void virtscsi_complete_event(void *buf)
+static void virtscsi_complete_event(struct virtio_scsi *vscsi, void *buf)
 {
 	struct virtio_scsi_event_node *event_node = buf;
 
@@ -344,82 +420,65 @@ static void virtscsi_event_done(struct virtqueue *vq)
 {
 	struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
 	struct virtio_scsi *vscsi = shost_priv(sh);
-	unsigned long flags;
 
-	spin_lock_irqsave(&vscsi->event_vq.vq_lock, flags);
-	virtscsi_vq_done(vq, virtscsi_complete_event);
-	spin_unlock_irqrestore(&vscsi->event_vq.vq_lock, flags);
+	virtscsi_vq_done(vscsi, &vscsi->event_vq, virtscsi_complete_event);
 };
 
-static void virtscsi_map_sgl(struct scatterlist *sg, unsigned int *p_idx,
-			     struct scsi_data_buffer *sdb)
-{
-	struct sg_table *table = &sdb->table;
-	struct scatterlist *sg_elem;
-	unsigned int idx = *p_idx;
-	int i;
-
-	for_each_sg(table->sgl, sg_elem, table->nents, i)
-		sg[idx++] = *sg_elem;
-
-	*p_idx = idx;
-}
-
 /**
- * virtscsi_map_cmd - map a scsi_cmd to a virtqueue scatterlist
- * @vscsi	: virtio_scsi state
+ * virtscsi_add_cmd - add a virtio_scsi_cmd to a virtqueue
+ * @vq		: the struct virtqueue we're talking about
  * @cmd		: command structure
- * @out_num	: number of read-only elements
- * @in_num	: number of write-only elements
  * @req_size	: size of the request buffer
  * @resp_size	: size of the response buffer
- *
- * Called with tgt_lock held.
+ * @gfp	: flags to use for memory allocations
  */
-static void virtscsi_map_cmd(struct virtio_scsi_target_state *tgt,
-			     struct virtio_scsi_cmd *cmd,
-			     unsigned *out_num, unsigned *in_num,
-			     size_t req_size, size_t resp_size)
+static int virtscsi_add_cmd(struct virtqueue *vq,
+			    struct virtio_scsi_cmd *cmd,
+			    size_t req_size, size_t resp_size, gfp_t gfp)
 {
 	struct scsi_cmnd *sc = cmd->sc;
-	struct scatterlist *sg = tgt->sg;
-	unsigned int idx = 0;
+	struct scatterlist *sgs[4], req, resp;
+	struct sg_table *out, *in;
+	unsigned out_num = 0, in_num = 0;
+
+	out = in = NULL;
+
+	if (sc && sc->sc_data_direction != DMA_NONE) {
+		if (sc->sc_data_direction != DMA_FROM_DEVICE)
+			out = &scsi_out(sc)->table;
+		if (sc->sc_data_direction != DMA_TO_DEVICE)
+			in = &scsi_in(sc)->table;
+	}
 
 	/* Request header.  */
-	sg_set_buf(&sg[idx++], &cmd->req, req_size);
+	sg_init_one(&req, &cmd->req, req_size);
+	sgs[out_num++] = &req;
 
 	/* Data-out buffer.  */
-	if (sc && sc->sc_data_direction != DMA_FROM_DEVICE)
-		virtscsi_map_sgl(sg, &idx, scsi_out(sc));
-
-	*out_num = idx;
+	if (out)
+		sgs[out_num++] = out->sgl;
 
 	/* Response header.  */
-	sg_set_buf(&sg[idx++], &cmd->resp, resp_size);
+	sg_init_one(&resp, &cmd->resp, resp_size);
+	sgs[out_num + in_num++] = &resp;
 
 	/* Data-in buffer */
-	if (sc && sc->sc_data_direction != DMA_TO_DEVICE)
-		virtscsi_map_sgl(sg, &idx, scsi_in(sc));
+	if (in)
+		sgs[out_num + in_num++] = in->sgl;
 
-	*in_num = idx - *out_num;
+	return virtqueue_add_sgs(vq, sgs, out_num, in_num, cmd, gfp);
 }
 
-static int virtscsi_kick_cmd(struct virtio_scsi_target_state *tgt,
-			     struct virtio_scsi_vq *vq,
+static int virtscsi_kick_cmd(struct virtio_scsi_vq *vq,
 			     struct virtio_scsi_cmd *cmd,
 			     size_t req_size, size_t resp_size, gfp_t gfp)
 {
-	unsigned int out_num, in_num;
 	unsigned long flags;
 	int err;
 	bool needs_kick = false;
 
-	spin_lock_irqsave(&tgt->tgt_lock, flags);
-	virtscsi_map_cmd(tgt, cmd, &out_num, &in_num, req_size, resp_size);
-
-	spin_lock(&vq->vq_lock);
-	err = virtqueue_add_buf(vq->vq, tgt->sg, out_num, in_num, cmd, gfp);
-	spin_unlock(&tgt->tgt_lock);
+	spin_lock_irqsave(&vq->vq_lock, flags);
+	err = virtscsi_add_cmd(vq->vq, cmd, req_size, resp_size, gfp);
 	if (!err)
 		needs_kick = virtqueue_kick_prepare(vq->vq);
 
@@ -430,10 +489,10 @@ static int virtscsi_kick_cmd(struct virtio_scsi_target_state *tgt,
 	return err;
 }
 
-static int virtscsi_queuecommand(struct Scsi_Host *sh, struct scsi_cmnd *sc)
+static int virtscsi_queuecommand(struct virtio_scsi *vscsi,
+				 struct virtio_scsi_vq *req_vq,
+				 struct scsi_cmnd *sc)
 {
-	struct virtio_scsi *vscsi = shost_priv(sh);
-	struct virtio_scsi_target_state *tgt = vscsi->tgt[sc->device->id];
 	struct virtio_scsi_cmd *cmd;
 	int ret;
 
@@ -467,7 +526,7 @@ static int virtscsi_queuecommand(struct Scsi_Host *sh, struct scsi_cmnd *sc)
 	BUG_ON(sc->cmd_len > VIRTIO_SCSI_CDB_SIZE);
 	memcpy(cmd->req.cmd.cdb, sc->cmnd, sc->cmd_len);
 
-	if (virtscsi_kick_cmd(tgt, &vscsi->req_vq, cmd,
+	if (virtscsi_kick_cmd(req_vq, cmd,
 			      sizeof cmd->req.cmd, sizeof cmd->resp.cmd,
 			      GFP_ATOMIC) == 0)
 		ret = 0;
@@ -478,14 +537,62 @@ out:
 	return ret;
 }
 
+static int virtscsi_queuecommand_single(struct Scsi_Host *sh,
+					struct scsi_cmnd *sc)
+{
+	struct virtio_scsi *vscsi = shost_priv(sh);
+	struct virtio_scsi_target_state *tgt =
+				scsi_target(sc->device)->hostdata;
+
+	atomic_inc(&tgt->reqs);
+	return virtscsi_queuecommand(vscsi, &vscsi->req_vqs[0], sc);
+}
+
+static struct virtio_scsi_vq *virtscsi_pick_vq(struct virtio_scsi *vscsi,
+					       struct virtio_scsi_target_state *tgt)
+{
+	struct virtio_scsi_vq *vq;
+	unsigned long flags;
+	u32 queue_num;
+
+	spin_lock_irqsave(&tgt->tgt_lock, flags);
+
+	/*
+	 * The memory barrier after atomic_inc_return matches
+	 * the smp_read_barrier_depends() in virtscsi_req_done.
+	 */
+	if (atomic_inc_return(&tgt->reqs) > 1)
+		vq = ACCESS_ONCE(tgt->req_vq);
+	else {
+		queue_num = smp_processor_id();
+		while (unlikely(queue_num >= vscsi->num_queues))
+			queue_num -= vscsi->num_queues;
+
+		tgt->req_vq = vq = &vscsi->req_vqs[queue_num];
+	}
+
+	spin_unlock_irqrestore(&tgt->tgt_lock, flags);
+	return vq;
+}
+
+static int virtscsi_queuecommand_multi(struct Scsi_Host *sh,
+				       struct scsi_cmnd *sc)
+{
+	struct virtio_scsi *vscsi = shost_priv(sh);
+	struct virtio_scsi_target_state *tgt =
+				scsi_target(sc->device)->hostdata;
+	struct virtio_scsi_vq *req_vq = virtscsi_pick_vq(vscsi, tgt);
+
+	return virtscsi_queuecommand(vscsi, req_vq, sc);
+}
+
 static int virtscsi_tmf(struct virtio_scsi *vscsi, struct virtio_scsi_cmd *cmd)
 {
 	DECLARE_COMPLETION_ONSTACK(comp);
-	struct virtio_scsi_target_state *tgt = vscsi->tgt[cmd->sc->device->id];
 	int ret = FAILED;
 
 	cmd->comp = &comp;
-	if (virtscsi_kick_cmd(tgt, &vscsi->ctrl_vq, cmd,
+	if (virtscsi_kick_cmd(&vscsi->ctrl_vq, cmd,
 			      sizeof cmd->req.tmf, sizeof cmd->resp.tmf,
 			      GFP_NOIO) < 0)
 		goto out;
@@ -547,18 +654,57 @@ static int virtscsi_abort(struct scsi_cmnd *sc)
 	return virtscsi_tmf(vscsi, cmd);
 }
 
-static struct scsi_host_template virtscsi_host_template = {
+static int virtscsi_target_alloc(struct scsi_target *starget)
+{
+	struct virtio_scsi_target_state *tgt =
+				kmalloc(sizeof(*tgt), GFP_KERNEL);
+	if (!tgt)
+		return -ENOMEM;
+
+	spin_lock_init(&tgt->tgt_lock);
+	atomic_set(&tgt->reqs, 0);
+	tgt->req_vq = NULL;
+
+	starget->hostdata = tgt;
+	return 0;
+}
+
+static void virtscsi_target_destroy(struct scsi_target *starget)
+{
+	struct virtio_scsi_target_state *tgt = starget->hostdata;
+	kfree(tgt);
+}
+
+static struct scsi_host_template virtscsi_host_template_single = {
+	.module = THIS_MODULE,
+	.name = "Virtio SCSI HBA",
+	.proc_name = "virtio_scsi",
+	.this_id = -1,
+	.queuecommand = virtscsi_queuecommand_single,
+	.eh_abort_handler = virtscsi_abort,
+	.eh_device_reset_handler = virtscsi_device_reset,
+
+	.can_queue = 1024,
+	.dma_boundary = UINT_MAX,
+	.use_clustering = ENABLE_CLUSTERING,
+	.target_alloc = virtscsi_target_alloc,
+	.target_destroy = virtscsi_target_destroy,
+};
+
+static struct scsi_host_template virtscsi_host_template_multi = {
 	.module = THIS_MODULE,
 	.name = "Virtio SCSI HBA",
 	.proc_name = "virtio_scsi",
-	.queuecommand = virtscsi_queuecommand,
 	.this_id = -1,
+	.queuecommand = virtscsi_queuecommand_multi,
 	.eh_abort_handler = virtscsi_abort,
 	.eh_device_reset_handler = virtscsi_device_reset,
 
 	.can_queue = 1024,
 	.dma_boundary = UINT_MAX,
 	.use_clustering = ENABLE_CLUSTERING,
+	.target_alloc = virtscsi_target_alloc,
+	.target_destroy = virtscsi_target_destroy,
 };
 
 #define virtscsi_config_get(vdev, fld) \
@@ -578,29 +724,69 @@ static struct scsi_host_template virtscsi_host_template = {
 				  &__val, sizeof(__val)); \
 	})
 
-static void virtscsi_init_vq(struct virtio_scsi_vq *virtscsi_vq,
-			     struct virtqueue *vq)
+static void __virtscsi_set_affinity(struct virtio_scsi *vscsi, bool affinity)
 {
-	spin_lock_init(&virtscsi_vq->vq_lock);
-	virtscsi_vq->vq = vq;
+	int i;
+	int cpu;
+
+	/* In multiqueue mode, when the number of cpu is equal
+	 * to the number of request queues, we let the qeueues
+	 * to be private to one cpu by setting the affinity hint
+	 * to eliminate the contention.
+	 */
+	if ((vscsi->num_queues == 1 ||
+	     vscsi->num_queues != num_online_cpus()) && affinity) {
+		if (vscsi->affinity_hint_set)
+			affinity = false;
+		else
+			return;
+	}
+
+	if (affinity) {
+		i = 0;
+		for_each_online_cpu(cpu) {
+			virtqueue_set_affinity(vscsi->req_vqs[i].vq, cpu);
+			i++;
+		}
+
+		vscsi->affinity_hint_set = true;
+	} else {
+		for (i = 0; i < vscsi->num_queues - VIRTIO_SCSI_VQ_BASE; i++)
+			virtqueue_set_affinity(vscsi->req_vqs[i].vq, -1);
+
+		vscsi->affinity_hint_set = false;
+	}
 }
 
-static struct virtio_scsi_target_state *virtscsi_alloc_tgt(
-	struct virtio_device *vdev, int sg_elems)
+static void virtscsi_set_affinity(struct virtio_scsi *vscsi, bool affinity)
 {
-	struct virtio_scsi_target_state *tgt;
-	gfp_t gfp_mask = GFP_KERNEL;
-
-	/* We need extra sg elements at head and tail.  */
-	tgt = kmalloc(sizeof(*tgt) + sizeof(tgt->sg[0]) * (sg_elems + 2),
-		      gfp_mask);
+	get_online_cpus();
+	__virtscsi_set_affinity(vscsi, affinity);
+	put_online_cpus();
+}
 
-	if (!tgt)
-		return NULL;
+static int virtscsi_cpu_callback(struct notifier_block *nfb,
+				 unsigned long action, void *hcpu)
+{
+	struct virtio_scsi *vscsi = container_of(nfb, struct virtio_scsi, nb);
+	switch(action) {
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+	case CPU_DEAD:
+	case CPU_DEAD_FROZEN:
+		__virtscsi_set_affinity(vscsi, true);
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
+}
 
-	spin_lock_init(&tgt->tgt_lock);
-	sg_init_table(tgt->sg, sg_elems + 2);
-	return tgt;
+static void virtscsi_init_vq(struct virtio_scsi_vq *virtscsi_vq,
+			     struct virtqueue *vq)
+{
+	spin_lock_init(&virtscsi_vq->vq_lock);
+	virtscsi_vq->vq = vq;
 }
 
 static void virtscsi_scan(struct virtio_device *vdev)
@@ -614,46 +800,56 @@ static void virtscsi_remove_vqs(struct virtio_device *vdev)
 {
 	struct Scsi_Host *sh = virtio_scsi_host(vdev);
 	struct virtio_scsi *vscsi = shost_priv(sh);
-	u32 i, num_targets;
+
+	virtscsi_set_affinity(vscsi, false);
 
 	/* Stop all the virtqueues. */
 	vdev->config->reset(vdev);
 
-	num_targets = sh->max_id;
-	for (i = 0; i < num_targets; i++) {
-		kfree(vscsi->tgt[i]);
-		vscsi->tgt[i] = NULL;
-	}
-
 	vdev->config->del_vqs(vdev);
 }
 
 static int virtscsi_init(struct virtio_device *vdev,
-			 struct virtio_scsi *vscsi, int num_targets)
+			 struct virtio_scsi *vscsi)
 {
 	int err;
-	struct virtqueue *vqs[3];
-	u32 i, sg_elems;
+	u32 i;
+	u32 num_vqs;
+	vq_callback_t **callbacks;
+	const char **names;
+	struct virtqueue **vqs;
+
+	num_vqs = vscsi->num_queues + VIRTIO_SCSI_VQ_BASE;
+	vqs = kmalloc(num_vqs * sizeof(struct virtqueue *), GFP_KERNEL);
+	callbacks = kmalloc(num_vqs * sizeof(vq_callback_t *), GFP_KERNEL);
+	names = kmalloc(num_vqs * sizeof(char *), GFP_KERNEL);
+
+	if (!callbacks || !vqs || !names) {
+		err = -ENOMEM;
+		goto out;
+	}
 
-	vq_callback_t *callbacks[] = {
-		virtscsi_ctrl_done,
-		virtscsi_event_done,
-		virtscsi_req_done
-	};
-	const char *names[] = {
-		"control",
-		"event",
-		"request"
-	};
+	callbacks[0] = virtscsi_ctrl_done;
+	callbacks[1] = virtscsi_event_done;
+	names[0] = "control";
+	names[1] = "event";
+	for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs; i++) {
+		callbacks[i] = virtscsi_req_done;
+		names[i] = "request";
+	}
 
 	/* Discover virtqueues and write information to configuration.  */
-	err = vdev->config->find_vqs(vdev, 3, vqs, callbacks, names);
+	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
 	if (err)
-		return err;
+		goto out;
 
 	virtscsi_init_vq(&vscsi->ctrl_vq, vqs[0]);
 	virtscsi_init_vq(&vscsi->event_vq, vqs[1]);
-	virtscsi_init_vq(&vscsi->req_vq, vqs[2]);
+	for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs; i++)
+		virtscsi_init_vq(&vscsi->req_vqs[i - VIRTIO_SCSI_VQ_BASE],
+				 vqs[i]);
+
+	virtscsi_set_affinity(vscsi, true);
 
 	virtscsi_config_set(vdev, cdb_size, VIRTIO_SCSI_CDB_SIZE);
 	virtscsi_config_set(vdev, sense_size, VIRTIO_SCSI_SENSE_SIZE);
@@ -661,19 +857,12 @@ static int virtscsi_init(struct virtio_device *vdev,
 	if (virtio_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG))
 		virtscsi_kick_event_all(vscsi);
 
-	/* We need to know how many segments before we allocate.  */
-	sg_elems = virtscsi_config_get(vdev, seg_max) ?: 1;
-
-	for (i = 0; i < num_targets; i++) {
-		vscsi->tgt[i] = virtscsi_alloc_tgt(vdev, sg_elems);
-		if (!vscsi->tgt[i]) {
-			err = -ENOMEM;
-			goto out;
-		}
-	}
 	err = 0;
 
 out:
+	kfree(names);
+	kfree(callbacks);
+	kfree(vqs);
 	if (err)
 		virtscsi_remove_vqs(vdev);
 	return err;
@@ -686,13 +875,21 @@ static int virtscsi_probe(struct virtio_device *vdev)
 	int err;
 	u32 sg_elems, num_targets;
 	u32 cmd_per_lun;
+	u32 num_queues;
+	struct scsi_host_template *hostt;
+
+	/* We need to know how many queues before we allocate. */
+	num_queues = virtscsi_config_get(vdev, num_queues) ? : 1;
 
-	/* Allocate memory and link the structs together.  */
 	num_targets = virtscsi_config_get(vdev, max_target) + 1;
-	shost = scsi_host_alloc(&virtscsi_host_template,
-		sizeof(*vscsi)
-		+ num_targets * sizeof(struct virtio_scsi_target_state));
 
+	if (num_queues == 1)
+		hostt = &virtscsi_host_template_single;
+	else
+		hostt = &virtscsi_host_template_multi;
+
+	shost = scsi_host_alloc(hostt,
+		sizeof(*vscsi) + sizeof(vscsi->req_vqs[0]) * num_queues);
 	if (!shost)
 		return -ENOMEM;
 
@@ -700,12 +897,20 @@ static int virtscsi_probe(struct virtio_device *vdev)
 	shost->sg_tablesize = sg_elems;
 	vscsi = shost_priv(shost);
 	vscsi->vdev = vdev;
+	vscsi->num_queues = num_queues;
 	vdev->priv = shost;
 
-	err = virtscsi_init(vdev, vscsi, num_targets);
+	err = virtscsi_init(vdev, vscsi);
 	if (err)
 		goto virtscsi_init_failed;
 
+	vscsi->nb.notifier_call = &virtscsi_cpu_callback;
+	err = register_hotcpu_notifier(&vscsi->nb);
+	if (err) {
+		pr_err("registering cpu notifier failed\n");
+		goto scsi_add_host_failed;
+	}
+
 	cmd_per_lun = virtscsi_config_get(vdev, cmd_per_lun) ?: 1;
 	shost->cmd_per_lun = min_t(u32, cmd_per_lun, shost->can_queue);
 	shost->max_sectors = virtscsi_config_get(vdev, max_sectors) ?: 0xFFFF;
@@ -743,6 +948,8 @@ static void virtscsi_remove(struct virtio_device *vdev)
 
 	scsi_remove_host(shost);
 
+	unregister_hotcpu_notifier(&vscsi->nb);
+
 	virtscsi_remove_vqs(vdev);
 	scsi_host_put(shost);
 }
@@ -759,7 +966,7 @@ static int virtscsi_restore(struct virtio_device *vdev)
 	struct Scsi_Host *sh = virtio_scsi_host(vdev);
 	struct virtio_scsi *vscsi = shost_priv(sh);
 
-	return virtscsi_init(vdev, vscsi, sh->max_id);
+	return virtscsi_init(vdev, vscsi);
 }
 #endif
 
@@ -794,8 +1001,7 @@ static int __init init(void)
 
 	virtscsi_cmd_cache = KMEM_CACHE(virtio_scsi_cmd, 0);
 	if (!virtscsi_cmd_cache) {
-		printk(KERN_ERR "kmem_cache_create() for "
-				"virtscsi_cmd_cache failed\n");
+		pr_err("kmem_cache_create() for virtscsi_cmd_cache failed\n");
 		goto error;
 	}
 
@@ -804,8 +1010,7 @@ static int __init init(void)
 		mempool_create_slab_pool(VIRTIO_SCSI_MEMPOOL_SZ,
 					 virtscsi_cmd_cache);
 	if (!virtscsi_cmd_pool) {
-		printk(KERN_ERR "mempool_create() for"
-				"virtscsi_cmd_pool failed\n");
+		pr_err("mempool_create() for virtscsi_cmd_pool failed\n");
 		goto error;
 	}
 	ret = register_virtio_driver(&virtio_scsi_driver);
diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig
index 26a64e5b8a5..8b9226da3f5 100644
--- a/drivers/vhost/Kconfig
+++ b/drivers/vhost/Kconfig
@@ -1,6 +1,7 @@
 config VHOST_NET
 	tristate "Host kernel accelerator for virtio net"
 	depends on NET && EVENTFD && (TUN || !TUN) && (MACVTAP || !MACVTAP)
+	select VHOST_RING
 	---help---
 	  This kernel module can be loaded in host kernel to accelerate
 	  guest networking with virtio_net. Not to be confused with virtio_net
@@ -12,7 +13,14 @@ config VHOST_NET
 config VHOST_SCSI
 	tristate "VHOST_SCSI TCM fabric driver"
 	depends on TARGET_CORE && EVENTFD && m
+	select VHOST_RING
 	default n
 	---help---
 	Say M here to enable the vhost_scsi TCM fabric module
 	for use with virtio-scsi guests
+
+config VHOST_RING
+	tristate
+	---help---
+	  This option is selected by any driver which needs to access
+	  the host side of a virtio ring.
diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile
index ef21d5fdfa7..654e9afb11f 100644
--- a/drivers/vhost/Makefile
+++ b/drivers/vhost/Makefile
@@ -3,3 +3,5 @@ vhost_net-y := vhost.o net.o
 
 obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o
 vhost_scsi-y := scsi.o
+
+obj-$(CONFIG_VHOST_RING) += vringh.o
diff --git a/drivers/vhost/test.c b/drivers/vhost/test.c
index be65414d5bb..1ee45bc85f6 100644
--- a/drivers/vhost/test.c
+++ b/drivers/vhost/test.c
@@ -282,7 +282,9 @@ static long vhost_test_ioctl(struct file *f, unsigned int ioctl,
 		return vhost_test_reset_owner(n);
 	default:
 		mutex_lock(&n->dev.mutex);
-		r = vhost_dev_ioctl(&n->dev, ioctl, arg);
+		r = vhost_dev_ioctl(&n->dev, ioctl, argp);
+                if (r == -ENOIOCTLCMD)
+                        r = vhost_vring_ioctl(&n->dev, ioctl, argp);
 		vhost_test_flush(n);
 		mutex_unlock(&n->dev.mutex);
 		return r;
diff --git a/drivers/vhost/vringh.c b/drivers/vhost/vringh.c
new file mode 100644
index 00000000000..bff0775e258
--- /dev/null
+++ b/drivers/vhost/vringh.c
@@ -0,0 +1,1007 @@
+/*
+ * Helpers for the host side of a virtio ring.
+ *
+ * Since these may be in userspace, we use (inline) accessors.
+ */
+#include <linux/vringh.h>
+#include <linux/virtio_ring.h>
+#include <linux/kernel.h>
+#include <linux/ratelimit.h>
+#include <linux/uaccess.h>
+#include <linux/slab.h>
+#include <linux/export.h>
+
+static __printf(1,2) __cold void vringh_bad(const char *fmt, ...)
+{
+	static DEFINE_RATELIMIT_STATE(vringh_rs,
+				      DEFAULT_RATELIMIT_INTERVAL,
+				      DEFAULT_RATELIMIT_BURST);
+	if (__ratelimit(&vringh_rs)) {
+		va_list ap;
+		va_start(ap, fmt);
+		printk(KERN_NOTICE "vringh:");
+		vprintk(fmt, ap);
+		va_end(ap);
+	}
+}
+
+/* Returns vring->num if empty, -ve on error. */
+static inline int __vringh_get_head(const struct vringh *vrh,
+				    int (*getu16)(u16 *val, const u16 *p),
+				    u16 *last_avail_idx)
+{
+	u16 avail_idx, i, head;
+	int err;
+
+	err = getu16(&avail_idx, &vrh->vring.avail->idx);
+	if (err) {
+		vringh_bad("Failed to access avail idx at %p",
+			   &vrh->vring.avail->idx);
+		return err;
+	}
+
+	if (*last_avail_idx == avail_idx)
+		return vrh->vring.num;
+
+	/* Only get avail ring entries after they have been exposed by guest. */
+	virtio_rmb(vrh->weak_barriers);
+
+	i = *last_avail_idx & (vrh->vring.num - 1);
+
+	err = getu16(&head, &vrh->vring.avail->ring[i]);
+	if (err) {
+		vringh_bad("Failed to read head: idx %d address %p",
+			   *last_avail_idx, &vrh->vring.avail->ring[i]);
+		return err;
+	}
+
+	if (head >= vrh->vring.num) {
+		vringh_bad("Guest says index %u > %u is available",
+			   head, vrh->vring.num);
+		return -EINVAL;
+	}
+
+	(*last_avail_idx)++;
+	return head;
+}
+
+/* Copy some bytes to/from the iovec.  Returns num copied. */
+static inline ssize_t vringh_iov_xfer(struct vringh_kiov *iov,
+				      void *ptr, size_t len,
+				      int (*xfer)(void *addr, void *ptr,
+						  size_t len))
+{
+	int err, done = 0;
+
+	while (len && iov->i < iov->used) {
+		size_t partlen;
+
+		partlen = min(iov->iov[iov->i].iov_len, len);
+		err = xfer(iov->iov[iov->i].iov_base, ptr, partlen);
+		if (err)
+			return err;
+		done += partlen;
+		len -= partlen;
+		ptr += partlen;
+		iov->consumed += partlen;
+		iov->iov[iov->i].iov_len -= partlen;
+		iov->iov[iov->i].iov_base += partlen;
+
+		if (!iov->iov[iov->i].iov_len) {
+			/* Fix up old iov element then increment. */
+			iov->iov[iov->i].iov_len = iov->consumed;
+			iov->iov[iov->i].iov_base -= iov->consumed;
+			
+			iov->consumed = 0;
+			iov->i++;
+		}
+	}
+	return done;
+}
+
+/* May reduce *len if range is shorter. */
+static inline bool range_check(struct vringh *vrh, u64 addr, size_t *len,
+			       struct vringh_range *range,
+			       bool (*getrange)(struct vringh *,
+						u64, struct vringh_range *))
+{
+	if (addr < range->start || addr > range->end_incl) {
+		if (!getrange(vrh, addr, range))
+			return false;
+	}
+	BUG_ON(addr < range->start || addr > range->end_incl);
+
+	/* To end of memory? */
+	if (unlikely(addr + *len == 0)) {
+		if (range->end_incl == -1ULL)
+			return true;
+		goto truncate;
+	}
+
+	/* Otherwise, don't wrap. */
+	if (addr + *len < addr) {
+		vringh_bad("Wrapping descriptor %zu@0x%llx",
+			   *len, (unsigned long long)addr);
+		return false;
+	}
+
+	if (unlikely(addr + *len - 1 > range->end_incl))
+		goto truncate;
+	return true;
+
+truncate:
+	*len = range->end_incl + 1 - addr;
+	return true;
+}
+
+static inline bool no_range_check(struct vringh *vrh, u64 addr, size_t *len,
+				  struct vringh_range *range,
+				  bool (*getrange)(struct vringh *,
+						   u64, struct vringh_range *))
+{
+	return true;
+}
+
+/* No reason for this code to be inline. */
+static int move_to_indirect(int *up_next, u16 *i, void *addr,
+			    const struct vring_desc *desc,
+			    struct vring_desc **descs, int *desc_max)
+{
+	/* Indirect tables can't have indirect. */
+	if (*up_next != -1) {
+		vringh_bad("Multilevel indirect %u->%u", *up_next, *i);
+		return -EINVAL;
+	}
+
+	if (unlikely(desc->len % sizeof(struct vring_desc))) {
+		vringh_bad("Strange indirect len %u", desc->len);
+		return -EINVAL;
+	}
+
+	/* We will check this when we follow it! */
+	if (desc->flags & VRING_DESC_F_NEXT)
+		*up_next = desc->next;
+	else
+		*up_next = -2;
+	*descs = addr;
+	*desc_max = desc->len / sizeof(struct vring_desc);
+
+	/* Now, start at the first indirect. */
+	*i = 0;
+	return 0;
+}
+
+static int resize_iovec(struct vringh_kiov *iov, gfp_t gfp)
+{
+	struct kvec *new;
+	unsigned int flag, new_num = (iov->max_num & ~VRINGH_IOV_ALLOCATED) * 2;
+
+	if (new_num < 8)
+		new_num = 8;
+
+	flag = (iov->max_num & VRINGH_IOV_ALLOCATED);
+	if (flag)
+		new = krealloc(iov->iov, new_num * sizeof(struct iovec), gfp);
+	else {
+		new = kmalloc(new_num * sizeof(struct iovec), gfp);
+		if (new) {
+			memcpy(new, iov->iov,
+			       iov->max_num * sizeof(struct iovec));
+			flag = VRINGH_IOV_ALLOCATED;
+		}
+	}
+	if (!new)
+		return -ENOMEM;
+	iov->iov = new;
+	iov->max_num = (new_num | flag);
+	return 0;
+}
+
+static u16 __cold return_from_indirect(const struct vringh *vrh, int *up_next,
+				       struct vring_desc **descs, int *desc_max)
+{
+	u16 i = *up_next;
+
+	*up_next = -1;
+	*descs = vrh->vring.desc;
+	*desc_max = vrh->vring.num;
+	return i;
+}
+
+static int slow_copy(struct vringh *vrh, void *dst, const void *src,
+		     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
+				    struct vringh_range *range,
+				    bool (*getrange)(struct vringh *vrh,
+						     u64,
+						     struct vringh_range *)),
+		     bool (*getrange)(struct vringh *vrh,
+				      u64 addr,
+				      struct vringh_range *r),
+		     struct vringh_range *range,
+		     int (*copy)(void *dst, const void *src, size_t len))
+{
+	size_t part, len = sizeof(struct vring_desc);
+
+	do {
+		u64 addr;
+		int err;
+
+		part = len;
+		addr = (u64)(unsigned long)src - range->offset;
+
+		if (!rcheck(vrh, addr, &part, range, getrange))
+			return -EINVAL;
+
+		err = copy(dst, src, part);
+		if (err)
+			return err;
+
+		dst += part;
+		src += part;
+		len -= part;
+	} while (len);
+	return 0;
+}
+
+static inline int
+__vringh_iov(struct vringh *vrh, u16 i,
+	     struct vringh_kiov *riov,
+	     struct vringh_kiov *wiov,
+	     bool (*rcheck)(struct vringh *vrh, u64 addr, size_t *len,
+			    struct vringh_range *range,
+			    bool (*getrange)(struct vringh *, u64,
+					     struct vringh_range *)),
+	     bool (*getrange)(struct vringh *, u64, struct vringh_range *),
+	     gfp_t gfp,
+	     int (*copy)(void *dst, const void *src, size_t len))
+{
+	int err, count = 0, up_next, desc_max;
+	struct vring_desc desc, *descs;
+	struct vringh_range range = { -1ULL, 0 }, slowrange;
+	bool slow = false;
+
+	/* We start traversing vring's descriptor table. */
+	descs = vrh->vring.desc;
+	desc_max = vrh->vring.num;
+	up_next = -1;
+
+	if (riov)
+		riov->i = riov->used = 0;
+	else if (wiov)
+		wiov->i = wiov->used = 0;
+	else
+		/* You must want something! */
+		BUG();
+
+	for (;;) {
+		void *addr;
+		struct vringh_kiov *iov;
+		size_t len;
+
+		if (unlikely(slow))
+			err = slow_copy(vrh, &desc, &descs[i], rcheck, getrange,
+					&slowrange, copy);
+		else
+			err = copy(&desc, &descs[i], sizeof(desc));
+		if (unlikely(err))
+			goto fail;
+
+		if (unlikely(desc.flags & VRING_DESC_F_INDIRECT)) {
+			/* Make sure it's OK, and get offset. */
+			len = desc.len;
+			if (!rcheck(vrh, desc.addr, &len, &range, getrange)) {
+				err = -EINVAL;
+				goto fail;
+			}
+
+			if (unlikely(len != desc.len)) {
+				slow = true;
+				/* We need to save this range to use offset */
+				slowrange = range;
+			}
+
+			addr = (void *)(long)(desc.addr + range.offset);
+			err = move_to_indirect(&up_next, &i, addr, &desc,
+					       &descs, &desc_max);
+			if (err)
+				goto fail;
+			continue;
+		}
+
+		if (count++ == vrh->vring.num) {
+			vringh_bad("Descriptor loop in %p", descs);
+			err = -ELOOP;
+			goto fail;
+		}
+
+		if (desc.flags & VRING_DESC_F_WRITE)
+			iov = wiov;
+		else {
+			iov = riov;
+			if (unlikely(wiov && wiov->i)) {
+				vringh_bad("Readable desc %p after writable",
+					   &descs[i]);
+				err = -EINVAL;
+				goto fail;
+			}
+		}
+
+		if (!iov) {
+			vringh_bad("Unexpected %s desc",
+				   !wiov ? "writable" : "readable");
+			err = -EPROTO;
+			goto fail;
+		}
+
+	again:
+		/* Make sure it's OK, and get offset. */
+		len = desc.len;
+		if (!rcheck(vrh, desc.addr, &len, &range, getrange)) {
+			err = -EINVAL;
+			goto fail;
+		}
+		addr = (void *)(unsigned long)(desc.addr + range.offset);
+
+		if (unlikely(iov->used == (iov->max_num & ~VRINGH_IOV_ALLOCATED))) {
+			err = resize_iovec(iov, gfp);
+			if (err)
+				goto fail;
+		}
+
+		iov->iov[iov->used].iov_base = addr;
+		iov->iov[iov->used].iov_len = len;
+		iov->used++;
+
+		if (unlikely(len != desc.len)) {
+			desc.len -= len;
+			desc.addr += len;
+			goto again;
+		}
+
+		if (desc.flags & VRING_DESC_F_NEXT) {
+			i = desc.next;
+		} else {
+			/* Just in case we need to finish traversing above. */
+			if (unlikely(up_next > 0)) {
+				i = return_from_indirect(vrh, &up_next,
+							 &descs, &desc_max);
+				slow = false;
+			} else
+				break;
+		}
+
+		if (i >= desc_max) {
+			vringh_bad("Chained index %u > %u", i, desc_max);
+			err = -EINVAL;
+			goto fail;
+		}
+	}
+
+	return 0;
+
+fail:
+	return err;
+}
+
+static inline int __vringh_complete(struct vringh *vrh,
+				    const struct vring_used_elem *used,
+				    unsigned int num_used,
+				    int (*putu16)(u16 *p, u16 val),
+				    int (*putused)(struct vring_used_elem *dst,
+						   const struct vring_used_elem
+						   *src, unsigned num))
+{
+	struct vring_used *used_ring;
+	int err;
+	u16 used_idx, off;
+
+	used_ring = vrh->vring.used;
+	used_idx = vrh->last_used_idx + vrh->completed;
+
+	off = used_idx % vrh->vring.num;
+
+	/* Compiler knows num_used == 1 sometimes, hence extra check */
+	if (num_used > 1 && unlikely(off + num_used >= vrh->vring.num)) {
+		u16 part = vrh->vring.num - off;
+		err = putused(&used_ring->ring[off], used, part);
+		if (!err)
+			err = putused(&used_ring->ring[0], used + part,
+				      num_used - part);
+	} else
+		err = putused(&used_ring->ring[off], used, num_used);
+
+	if (err) {
+		vringh_bad("Failed to write %u used entries %u at %p",
+			   num_used, off, &used_ring->ring[off]);
+		return err;
+	}
+
+	/* Make sure buffer is written before we update index. */
+	virtio_wmb(vrh->weak_barriers);
+
+	err = putu16(&vrh->vring.used->idx, used_idx + num_used);
+	if (err) {
+		vringh_bad("Failed to update used index at %p",
+			   &vrh->vring.used->idx);
+		return err;
+	}
+
+	vrh->completed += num_used;
+	return 0;
+}
+
+
+static inline int __vringh_need_notify(struct vringh *vrh,
+				       int (*getu16)(u16 *val, const u16 *p))
+{
+	bool notify;
+	u16 used_event;
+	int err;
+
+	/* Flush out used index update. This is paired with the
+	 * barrier that the Guest executes when enabling
+	 * interrupts. */
+	virtio_mb(vrh->weak_barriers);
+
+	/* Old-style, without event indices. */
+	if (!vrh->event_indices) {
+		u16 flags;
+		err = getu16(&flags, &vrh->vring.avail->flags);
+		if (err) {
+			vringh_bad("Failed to get flags at %p",
+				   &vrh->vring.avail->flags);
+			return err;
+		}
+		return (!(flags & VRING_AVAIL_F_NO_INTERRUPT));
+	}
+
+	/* Modern: we know when other side wants to know. */
+	err = getu16(&used_event, &vring_used_event(&vrh->vring));
+	if (err) {
+		vringh_bad("Failed to get used event idx at %p",
+			   &vring_used_event(&vrh->vring));
+		return err;
+	}
+
+	/* Just in case we added so many that we wrap. */
+	if (unlikely(vrh->completed > 0xffff))
+		notify = true;
+	else
+		notify = vring_need_event(used_event,
+					  vrh->last_used_idx + vrh->completed,
+					  vrh->last_used_idx);
+
+	vrh->last_used_idx += vrh->completed;
+	vrh->completed = 0;
+	return notify;
+}
+
+static inline bool __vringh_notify_enable(struct vringh *vrh,
+					  int (*getu16)(u16 *val, const u16 *p),
+					  int (*putu16)(u16 *p, u16 val))
+{
+	u16 avail;
+
+	if (!vrh->event_indices) {
+		/* Old-school; update flags. */
+		if (putu16(&vrh->vring.used->flags, 0) != 0) {
+			vringh_bad("Clearing used flags %p",
+				   &vrh->vring.used->flags);
+			return true;
+		}
+	} else {
+		if (putu16(&vring_avail_event(&vrh->vring),
+			   vrh->last_avail_idx) != 0) {
+			vringh_bad("Updating avail event index %p",
+				   &vring_avail_event(&vrh->vring));
+			return true;
+		}
+	}
+
+	/* They could have slipped one in as we were doing that: make
+	 * sure it's written, then check again. */
+	virtio_mb(vrh->weak_barriers);
+
+	if (getu16(&avail, &vrh->vring.avail->idx) != 0) {
+		vringh_bad("Failed to check avail idx at %p",
+			   &vrh->vring.avail->idx);
+		return true;
+	}
+
+	/* This is unlikely, so we just leave notifications enabled
+	 * (if we're using event_indices, we'll only get one
+	 * notification anyway). */
+	return avail == vrh->last_avail_idx;
+}
+
+static inline void __vringh_notify_disable(struct vringh *vrh,
+					   int (*putu16)(u16 *p, u16 val))
+{
+	if (!vrh->event_indices) {
+		/* Old-school; update flags. */
+		if (putu16(&vrh->vring.used->flags, VRING_USED_F_NO_NOTIFY)) {
+			vringh_bad("Setting used flags %p",
+				   &vrh->vring.used->flags);
+		}
+	}
+}
+
+/* Userspace access helpers: in this case, addresses are really userspace. */
+static inline int getu16_user(u16 *val, const u16 *p)
+{
+	return get_user(*val, (__force u16 __user *)p);
+}
+
+static inline int putu16_user(u16 *p, u16 val)
+{
+	return put_user(val, (__force u16 __user *)p);
+}
+
+static inline int copydesc_user(void *dst, const void *src, size_t len)
+{
+	return copy_from_user(dst, (__force void __user *)src, len) ?
+		-EFAULT : 0;
+}
+
+static inline int putused_user(struct vring_used_elem *dst,
+			       const struct vring_used_elem *src,
+			       unsigned int num)
+{
+	return copy_to_user((__force void __user *)dst, src,
+			    sizeof(*dst) * num) ? -EFAULT : 0;
+}
+
+static inline int xfer_from_user(void *src, void *dst, size_t len)
+{
+	return copy_from_user(dst, (__force void __user *)src, len) ?
+		-EFAULT : 0;
+}
+
+static inline int xfer_to_user(void *dst, void *src, size_t len)
+{
+	return copy_to_user((__force void __user *)dst, src, len) ?
+		-EFAULT : 0;
+}
+
+/**
+ * vringh_init_user - initialize a vringh for a userspace vring.
+ * @vrh: the vringh to initialize.
+ * @features: the feature bits for this ring.
+ * @num: the number of elements.
+ * @weak_barriers: true if we only need memory barriers, not I/O.
+ * @desc: the userpace descriptor pointer.
+ * @avail: the userpace avail pointer.
+ * @used: the userpace used pointer.
+ *
+ * Returns an error if num is invalid: you should check pointers
+ * yourself!
+ */
+int vringh_init_user(struct vringh *vrh, u32 features,
+		     unsigned int num, bool weak_barriers,
+		     struct vring_desc __user *desc,
+		     struct vring_avail __user *avail,
+		     struct vring_used __user *used)
+{
+	/* Sane power of 2 please! */
+	if (!num || num > 0xffff || (num & (num - 1))) {
+		vringh_bad("Bad ring size %u", num);
+		return -EINVAL;
+	}
+
+	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
+	vrh->weak_barriers = weak_barriers;
+	vrh->completed = 0;
+	vrh->last_avail_idx = 0;
+	vrh->last_used_idx = 0;
+	vrh->vring.num = num;
+	/* vring expects kernel addresses, but only used via accessors. */
+	vrh->vring.desc = (__force struct vring_desc *)desc;
+	vrh->vring.avail = (__force struct vring_avail *)avail;
+	vrh->vring.used = (__force struct vring_used *)used;
+	return 0;
+}
+EXPORT_SYMBOL(vringh_init_user);
+
+/**
+ * vringh_getdesc_user - get next available descriptor from userspace ring.
+ * @vrh: the userspace vring.
+ * @riov: where to put the readable descriptors (or NULL)
+ * @wiov: where to put the writable descriptors (or NULL)
+ * @getrange: function to call to check ranges.
+ * @head: head index we received, for passing to vringh_complete_user().
+ *
+ * Returns 0 if there was no descriptor, 1 if there was, or -errno.
+ *
+ * Note that on error return, you can tell the difference between an
+ * invalid ring and a single invalid descriptor: in the former case,
+ * *head will be vrh->vring.num.  You may be able to ignore an invalid
+ * descriptor, but there's not much you can do with an invalid ring.
+ *
+ * Note that you may need to clean up riov and wiov, even on error!
+ */
+int vringh_getdesc_user(struct vringh *vrh,
+			struct vringh_iov *riov,
+			struct vringh_iov *wiov,
+			bool (*getrange)(struct vringh *vrh,
+					 u64 addr, struct vringh_range *r),
+			u16 *head)
+{
+	int err;
+
+	*head = vrh->vring.num;
+	err = __vringh_get_head(vrh, getu16_user, &vrh->last_avail_idx);
+	if (err < 0)
+		return err;
+
+	/* Empty... */
+	if (err == vrh->vring.num)
+		return 0;
+
+	/* We need the layouts to be the identical for this to work */
+	BUILD_BUG_ON(sizeof(struct vringh_kiov) != sizeof(struct vringh_iov));
+	BUILD_BUG_ON(offsetof(struct vringh_kiov, iov) !=
+		     offsetof(struct vringh_iov, iov));
+	BUILD_BUG_ON(offsetof(struct vringh_kiov, i) !=
+		     offsetof(struct vringh_iov, i));
+	BUILD_BUG_ON(offsetof(struct vringh_kiov, used) !=
+		     offsetof(struct vringh_iov, used));
+	BUILD_BUG_ON(offsetof(struct vringh_kiov, max_num) !=
+		     offsetof(struct vringh_iov, max_num));
+	BUILD_BUG_ON(sizeof(struct iovec) != sizeof(struct kvec));
+	BUILD_BUG_ON(offsetof(struct iovec, iov_base) !=
+		     offsetof(struct kvec, iov_base));
+	BUILD_BUG_ON(offsetof(struct iovec, iov_len) !=
+		     offsetof(struct kvec, iov_len));
+	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_base)
+		     != sizeof(((struct kvec *)NULL)->iov_base));
+	BUILD_BUG_ON(sizeof(((struct iovec *)NULL)->iov_len)
+		     != sizeof(((struct kvec *)NULL)->iov_len));
+
+	*head = err;
+	err = __vringh_iov(vrh, *head, (struct vringh_kiov *)riov,
+			   (struct vringh_kiov *)wiov,
+			   range_check, getrange, GFP_KERNEL, copydesc_user);
+	if (err)
+		return err;
+
+	return 1;
+}
+EXPORT_SYMBOL(vringh_getdesc_user);
+
+/**
+ * vringh_iov_pull_user - copy bytes from vring_iov.
+ * @riov: the riov as passed to vringh_getdesc_user() (updated as we consume)
+ * @dst: the place to copy.
+ * @len: the maximum length to copy.
+ *
+ * Returns the bytes copied <= len or a negative errno.
+ */
+ssize_t vringh_iov_pull_user(struct vringh_iov *riov, void *dst, size_t len)
+{
+	return vringh_iov_xfer((struct vringh_kiov *)riov,
+			       dst, len, xfer_from_user);
+}
+EXPORT_SYMBOL(vringh_iov_pull_user);
+
+/**
+ * vringh_iov_push_user - copy bytes into vring_iov.
+ * @wiov: the wiov as passed to vringh_getdesc_user() (updated as we consume)
+ * @dst: the place to copy.
+ * @len: the maximum length to copy.
+ *
+ * Returns the bytes copied <= len or a negative errno.
+ */
+ssize_t vringh_iov_push_user(struct vringh_iov *wiov,
+			     const void *src, size_t len)
+{
+	return vringh_iov_xfer((struct vringh_kiov *)wiov,
+			       (void *)src, len, xfer_to_user);
+}
+EXPORT_SYMBOL(vringh_iov_push_user);
+
+/**
+ * vringh_abandon_user - we've decided not to handle the descriptor(s).
+ * @vrh: the vring.
+ * @num: the number of descriptors to put back (ie. num
+ *	 vringh_get_user() to undo).
+ *
+ * The next vringh_get_user() will return the old descriptor(s) again.
+ */
+void vringh_abandon_user(struct vringh *vrh, unsigned int num)
+{
+	/* We only update vring_avail_event(vr) when we want to be notified,
+	 * so we haven't changed that yet. */
+	vrh->last_avail_idx -= num;
+}
+EXPORT_SYMBOL(vringh_abandon_user);
+
+/**
+ * vringh_complete_user - we've finished with descriptor, publish it.
+ * @vrh: the vring.
+ * @head: the head as filled in by vringh_getdesc_user.
+ * @len: the length of data we have written.
+ *
+ * You should check vringh_need_notify_user() after one or more calls
+ * to this function.
+ */
+int vringh_complete_user(struct vringh *vrh, u16 head, u32 len)
+{
+	struct vring_used_elem used;
+
+	used.id = head;
+	used.len = len;
+	return __vringh_complete(vrh, &used, 1, putu16_user, putused_user);
+}
+EXPORT_SYMBOL(vringh_complete_user);
+
+/**
+ * vringh_complete_multi_user - we've finished with many descriptors.
+ * @vrh: the vring.
+ * @used: the head, length pairs.
+ * @num_used: the number of used elements.
+ *
+ * You should check vringh_need_notify_user() after one or more calls
+ * to this function.
+ */
+int vringh_complete_multi_user(struct vringh *vrh,
+			       const struct vring_used_elem used[],
+			       unsigned num_used)
+{
+	return __vringh_complete(vrh, used, num_used,
+				 putu16_user, putused_user);
+}
+EXPORT_SYMBOL(vringh_complete_multi_user);
+
+/**
+ * vringh_notify_enable_user - we want to know if something changes.
+ * @vrh: the vring.
+ *
+ * This always enables notifications, but returns false if there are
+ * now more buffers available in the vring.
+ */
+bool vringh_notify_enable_user(struct vringh *vrh)
+{
+	return __vringh_notify_enable(vrh, getu16_user, putu16_user);
+}
+EXPORT_SYMBOL(vringh_notify_enable_user);
+
+/**
+ * vringh_notify_disable_user - don't tell us if something changes.
+ * @vrh: the vring.
+ *
+ * This is our normal running state: we disable and then only enable when
+ * we're going to sleep.
+ */
+void vringh_notify_disable_user(struct vringh *vrh)
+{
+	__vringh_notify_disable(vrh, putu16_user);
+}
+EXPORT_SYMBOL(vringh_notify_disable_user);
+
+/**
+ * vringh_need_notify_user - must we tell the other side about used buffers?
+ * @vrh: the vring we've called vringh_complete_user() on.
+ *
+ * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
+ */
+int vringh_need_notify_user(struct vringh *vrh)
+{
+	return __vringh_need_notify(vrh, getu16_user);
+}
+EXPORT_SYMBOL(vringh_need_notify_user);
+
+/* Kernelspace access helpers. */
+static inline int getu16_kern(u16 *val, const u16 *p)
+{
+	*val = ACCESS_ONCE(*p);
+	return 0;
+}
+
+static inline int putu16_kern(u16 *p, u16 val)
+{
+	ACCESS_ONCE(*p) = val;
+	return 0;
+}
+
+static inline int copydesc_kern(void *dst, const void *src, size_t len)
+{
+	memcpy(dst, src, len);
+	return 0;
+}
+
+static inline int putused_kern(struct vring_used_elem *dst,
+			       const struct vring_used_elem *src,
+			       unsigned int num)
+{
+	memcpy(dst, src, num * sizeof(*dst));
+	return 0;
+}
+
+static inline int xfer_kern(void *src, void *dst, size_t len)
+{
+	memcpy(dst, src, len);
+	return 0;
+}
+
+/**
+ * vringh_init_kern - initialize a vringh for a kernelspace vring.
+ * @vrh: the vringh to initialize.
+ * @features: the feature bits for this ring.
+ * @num: the number of elements.
+ * @weak_barriers: true if we only need memory barriers, not I/O.
+ * @desc: the userpace descriptor pointer.
+ * @avail: the userpace avail pointer.
+ * @used: the userpace used pointer.
+ *
+ * Returns an error if num is invalid.
+ */
+int vringh_init_kern(struct vringh *vrh, u32 features,
+		     unsigned int num, bool weak_barriers,
+		     struct vring_desc *desc,
+		     struct vring_avail *avail,
+		     struct vring_used *used)
+{
+	/* Sane power of 2 please! */
+	if (!num || num > 0xffff || (num & (num - 1))) {
+		vringh_bad("Bad ring size %u", num);
+		return -EINVAL;
+	}
+
+	vrh->event_indices = (features & (1 << VIRTIO_RING_F_EVENT_IDX));
+	vrh->weak_barriers = weak_barriers;
+	vrh->completed = 0;
+	vrh->last_avail_idx = 0;
+	vrh->last_used_idx = 0;
+	vrh->vring.num = num;
+	vrh->vring.desc = desc;
+	vrh->vring.avail = avail;
+	vrh->vring.used = used;
+	return 0;
+}
+EXPORT_SYMBOL(vringh_init_kern);
+
+/**
+ * vringh_getdesc_kern - get next available descriptor from kernelspace ring.
+ * @vrh: the kernelspace vring.
+ * @riov: where to put the readable descriptors (or NULL)
+ * @wiov: where to put the writable descriptors (or NULL)
+ * @head: head index we received, for passing to vringh_complete_kern().
+ * @gfp: flags for allocating larger riov/wiov.
+ *
+ * Returns 0 if there was no descriptor, 1 if there was, or -errno.
+ *
+ * Note that on error return, you can tell the difference between an
+ * invalid ring and a single invalid descriptor: in the former case,
+ * *head will be vrh->vring.num.  You may be able to ignore an invalid
+ * descriptor, but there's not much you can do with an invalid ring.
+ *
+ * Note that you may need to clean up riov and wiov, even on error!
+ */
+int vringh_getdesc_kern(struct vringh *vrh,
+			struct vringh_kiov *riov,
+			struct vringh_kiov *wiov,
+			u16 *head,
+			gfp_t gfp)
+{
+	int err;
+
+	err = __vringh_get_head(vrh, getu16_kern, &vrh->last_avail_idx);
+	if (err < 0)
+		return err;
+
+	/* Empty... */
+	if (err == vrh->vring.num)
+		return 0;
+
+	*head = err;
+	err = __vringh_iov(vrh, *head, riov, wiov, no_range_check, NULL,
+			   gfp, copydesc_kern);
+	if (err)
+		return err;
+
+	return 1;
+}
+EXPORT_SYMBOL(vringh_getdesc_kern);
+
+/**
+ * vringh_iov_pull_kern - copy bytes from vring_iov.
+ * @riov: the riov as passed to vringh_getdesc_kern() (updated as we consume)
+ * @dst: the place to copy.
+ * @len: the maximum length to copy.
+ *
+ * Returns the bytes copied <= len or a negative errno.
+ */
+ssize_t vringh_iov_pull_kern(struct vringh_kiov *riov, void *dst, size_t len)
+{
+	return vringh_iov_xfer(riov, dst, len, xfer_kern);
+}
+EXPORT_SYMBOL(vringh_iov_pull_kern);
+
+/**
+ * vringh_iov_push_kern - copy bytes into vring_iov.
+ * @wiov: the wiov as passed to vringh_getdesc_kern() (updated as we consume)
+ * @dst: the place to copy.
+ * @len: the maximum length to copy.
+ *
+ * Returns the bytes copied <= len or a negative errno.
+ */
+ssize_t vringh_iov_push_kern(struct vringh_kiov *wiov,
+			     const void *src, size_t len)
+{
+	return vringh_iov_xfer(wiov, (void *)src, len, xfer_kern);
+}
+EXPORT_SYMBOL(vringh_iov_push_kern);
+
+/**
+ * vringh_abandon_kern - we've decided not to handle the descriptor(s).
+ * @vrh: the vring.
+ * @num: the number of descriptors to put back (ie. num
+ *	 vringh_get_kern() to undo).
+ *
+ * The next vringh_get_kern() will return the old descriptor(s) again.
+ */
+void vringh_abandon_kern(struct vringh *vrh, unsigned int num)
+{
+	/* We only update vring_avail_event(vr) when we want to be notified,
+	 * so we haven't changed that yet. */
+	vrh->last_avail_idx -= num;
+}
+EXPORT_SYMBOL(vringh_abandon_kern);
+
+/**
+ * vringh_complete_kern - we've finished with descriptor, publish it.
+ * @vrh: the vring.
+ * @head: the head as filled in by vringh_getdesc_kern.
+ * @len: the length of data we have written.
+ *
+ * You should check vringh_need_notify_kern() after one or more calls
+ * to this function.
+ */
+int vringh_complete_kern(struct vringh *vrh, u16 head, u32 len)
+{
+	struct vring_used_elem used;
+
+	used.id = head;
+	used.len = len;
+
+	return __vringh_complete(vrh, &used, 1, putu16_kern, putused_kern);
+}
+EXPORT_SYMBOL(vringh_complete_kern);
+
+/**
+ * vringh_notify_enable_kern - we want to know if something changes.
+ * @vrh: the vring.
+ *
+ * This always enables notifications, but returns false if there are
+ * now more buffers available in the vring.
+ */
+bool vringh_notify_enable_kern(struct vringh *vrh)
+{
+	return __vringh_notify_enable(vrh, getu16_kern, putu16_kern);
+}
+EXPORT_SYMBOL(vringh_notify_enable_kern);
+
+/**
+ * vringh_notify_disable_kern - don't tell us if something changes.
+ * @vrh: the vring.
+ *
+ * This is our normal running state: we disable and then only enable when
+ * we're going to sleep.
+ */
+void vringh_notify_disable_kern(struct vringh *vrh)
+{
+	__vringh_notify_disable(vrh, putu16_kern);
+}
+EXPORT_SYMBOL(vringh_notify_disable_kern);
+
+/**
+ * vringh_need_notify_kern - must we tell the other side about used buffers?
+ * @vrh: the vring we've called vringh_complete_kern() on.
+ *
+ * Returns -errno or 0 if we don't need to tell the other side, 1 if we do.
+ */
+int vringh_need_notify_kern(struct vringh *vrh)
+{
+	return __vringh_need_notify(vrh, getu16_kern);
+}
+EXPORT_SYMBOL(vringh_need_notify_kern);
diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index 8dab163c5ef..bd3ae324a1a 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -108,7 +108,7 @@ static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
 
 	/* We should always be able to add one buffer to an empty queue. */
-	if (virtqueue_add_buf(vq, &sg, 1, 0, vb, GFP_KERNEL) < 0)
+	if (virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL) < 0)
 		BUG();
 	virtqueue_kick(vq);
 
@@ -256,7 +256,7 @@ static void stats_handle_request(struct virtio_balloon *vb)
 	if (!virtqueue_get_buf(vq, &len))
 		return;
 	sg_init_one(&sg, vb->stats, sizeof(vb->stats));
-	if (virtqueue_add_buf(vq, &sg, 1, 0, vb, GFP_KERNEL) < 0)
+	if (virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL) < 0)
 		BUG();
 	virtqueue_kick(vq);
 }
@@ -341,7 +341,7 @@ static int init_vqs(struct virtio_balloon *vb)
 		 * use it to signal us later.
 		 */
 		sg_init_one(&sg, vb->stats, sizeof vb->stats);
-		if (virtqueue_add_buf(vb->stats_vq, &sg, 1, 0, vb, GFP_KERNEL)
+		if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
 		    < 0)
 			BUG();
 		virtqueue_kick(vb->stats_vq);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index ffd7e7da5d3..5217baf5528 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -24,27 +24,6 @@
 #include <linux/module.h>
 #include <linux/hrtimer.h>
 
-/* virtio guest is communicating with a virtual "device" that actually runs on
- * a host processor.  Memory barriers are used to control SMP effects. */
-#ifdef CONFIG_SMP
-/* Where possible, use SMP barriers which are more lightweight than mandatory
- * barriers, because mandatory barriers control MMIO effects on accesses
- * through relaxed memory I/O windows (which virtio-pci does not use). */
-#define virtio_mb(vq) \
-	do { if ((vq)->weak_barriers) smp_mb(); else mb(); } while(0)
-#define virtio_rmb(vq) \
-	do { if ((vq)->weak_barriers) smp_rmb(); else rmb(); } while(0)
-#define virtio_wmb(vq) \
-	do { if ((vq)->weak_barriers) smp_wmb(); else wmb(); } while(0)
-#else
-/* We must force memory ordering even if guest is UP since host could be
- * running on another CPU, but SMP barriers are defined to barrier() in that
- * configuration. So fall back to mandatory barriers instead. */
-#define virtio_mb(vq) mb()
-#define virtio_rmb(vq) rmb()
-#define virtio_wmb(vq) wmb()
-#endif
-
 #ifdef DEBUG
 /* For development, we want to crash whenever the ring is screwed. */
 #define BAD_RING(_vq, fmt, args...)				\
@@ -119,16 +98,36 @@ struct vring_virtqueue
 
 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
 
+static inline struct scatterlist *sg_next_chained(struct scatterlist *sg,
+						  unsigned int *count)
+{
+	return sg_next(sg);
+}
+
+static inline struct scatterlist *sg_next_arr(struct scatterlist *sg,
+					      unsigned int *count)
+{
+	if (--(*count) == 0)
+		return NULL;
+	return sg + 1;
+}
+
 /* Set up an indirect table of descriptors and add it to the queue. */
-static int vring_add_indirect(struct vring_virtqueue *vq,
-			      struct scatterlist sg[],
-			      unsigned int out,
-			      unsigned int in,
-			      gfp_t gfp)
+static inline int vring_add_indirect(struct vring_virtqueue *vq,
+				     struct scatterlist *sgs[],
+				     struct scatterlist *(*next)
+				       (struct scatterlist *, unsigned int *),
+				     unsigned int total_sg,
+				     unsigned int total_out,
+				     unsigned int total_in,
+				     unsigned int out_sgs,
+				     unsigned int in_sgs,
+				     gfp_t gfp)
 {
 	struct vring_desc *desc;
 	unsigned head;
-	int i;
+	struct scatterlist *sg;
+	int i, n;
 
 	/*
 	 * We require lowmem mappings for the descriptors because
@@ -137,25 +136,31 @@ static int vring_add_indirect(struct vring_virtqueue *vq,
 	 */
 	gfp &= ~(__GFP_HIGHMEM | __GFP_HIGH);
 
-	desc = kmalloc((out + in) * sizeof(struct vring_desc), gfp);
+	desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp);
 	if (!desc)
 		return -ENOMEM;
 
-	/* Transfer entries from the sg list into the indirect page */
-	for (i = 0; i < out; i++) {
-		desc[i].flags = VRING_DESC_F_NEXT;
-		desc[i].addr = sg_phys(sg);
-		desc[i].len = sg->length;
-		desc[i].next = i+1;
-		sg++;
+	/* Transfer entries from the sg lists into the indirect page */
+	i = 0;
+	for (n = 0; n < out_sgs; n++) {
+		for (sg = sgs[n]; sg; sg = next(sg, &total_out)) {
+			desc[i].flags = VRING_DESC_F_NEXT;
+			desc[i].addr = sg_phys(sg);
+			desc[i].len = sg->length;
+			desc[i].next = i+1;
+			i++;
+		}
 	}
-	for (; i < (out + in); i++) {
-		desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
-		desc[i].addr = sg_phys(sg);
-		desc[i].len = sg->length;
-		desc[i].next = i+1;
-		sg++;
+	for (; n < (out_sgs + in_sgs); n++) {
+		for (sg = sgs[n]; sg; sg = next(sg, &total_in)) {
+			desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
+			desc[i].addr = sg_phys(sg);
+			desc[i].len = sg->length;
+			desc[i].next = i+1;
+			i++;
+		}
 	}
+	BUG_ON(i != total_sg);
 
 	/* Last one doesn't continue. */
 	desc[i-1].flags &= ~VRING_DESC_F_NEXT;
@@ -176,29 +181,20 @@ static int vring_add_indirect(struct vring_virtqueue *vq,
 	return head;
 }
 
-/**
- * virtqueue_add_buf - expose buffer to other end
- * @vq: the struct virtqueue we're talking about.
- * @sg: the description of the buffer(s).
- * @out_num: the number of sg readable by other side
- * @in_num: the number of sg which are writable (after readable ones)
- * @data: the token identifying the buffer.
- * @gfp: how to do memory allocations (if necessary).
- *
- * Caller must ensure we don't call this with other virtqueue operations
- * at the same time (except where noted).
- *
- * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
- */
-int virtqueue_add_buf(struct virtqueue *_vq,
-		      struct scatterlist sg[],
-		      unsigned int out,
-		      unsigned int in,
-		      void *data,
-		      gfp_t gfp)
+static inline int virtqueue_add(struct virtqueue *_vq,
+				struct scatterlist *sgs[],
+				struct scatterlist *(*next)
+				  (struct scatterlist *, unsigned int *),
+				unsigned int total_out,
+				unsigned int total_in,
+				unsigned int out_sgs,
+				unsigned int in_sgs,
+				void *data,
+				gfp_t gfp)
 {
 	struct vring_virtqueue *vq = to_vvq(_vq);
-	unsigned int i, avail, uninitialized_var(prev);
+	struct scatterlist *sg;
+	unsigned int i, n, avail, uninitialized_var(prev), total_sg;
 	int head;
 
 	START_USE(vq);
@@ -218,46 +214,54 @@ int virtqueue_add_buf(struct virtqueue *_vq,
 	}
 #endif
 
+	total_sg = total_in + total_out;
+
 	/* If the host supports indirect descriptor tables, and we have multiple
 	 * buffers, then go indirect. FIXME: tune this threshold */
-	if (vq->indirect && (out + in) > 1 && vq->vq.num_free) {
-		head = vring_add_indirect(vq, sg, out, in, gfp);
+	if (vq->indirect && total_sg > 1 && vq->vq.num_free) {
+		head = vring_add_indirect(vq, sgs, next, total_sg, total_out,
+					  total_in,
+					  out_sgs, in_sgs, gfp);
 		if (likely(head >= 0))
 			goto add_head;
 	}
 
-	BUG_ON(out + in > vq->vring.num);
-	BUG_ON(out + in == 0);
+	BUG_ON(total_sg > vq->vring.num);
+	BUG_ON(total_sg == 0);
 
-	if (vq->vq.num_free < out + in) {
+	if (vq->vq.num_free < total_sg) {
 		pr_debug("Can't add buf len %i - avail = %i\n",
-			 out + in, vq->vq.num_free);
+			 total_sg, vq->vq.num_free);
 		/* FIXME: for historical reasons, we force a notify here if
 		 * there are outgoing parts to the buffer.  Presumably the
 		 * host should service the ring ASAP. */
-		if (out)
+		if (out_sgs)
 			vq->notify(&vq->vq);
 		END_USE(vq);
 		return -ENOSPC;
 	}
 
 	/* We're about to use some buffers from the free list. */
-	vq->vq.num_free -= out + in;
-
-	head = vq->free_head;
-	for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
-		vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
-		vq->vring.desc[i].addr = sg_phys(sg);
-		vq->vring.desc[i].len = sg->length;
-		prev = i;
-		sg++;
+	vq->vq.num_free -= total_sg;
+
+	head = i = vq->free_head;
+	for (n = 0; n < out_sgs; n++) {
+		for (sg = sgs[n]; sg; sg = next(sg, &total_out)) {
+			vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
+			vq->vring.desc[i].addr = sg_phys(sg);
+			vq->vring.desc[i].len = sg->length;
+			prev = i;
+			i = vq->vring.desc[i].next;
+		}
 	}
-	for (; in; i = vq->vring.desc[i].next, in--) {
-		vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
-		vq->vring.desc[i].addr = sg_phys(sg);
-		vq->vring.desc[i].len = sg->length;
-		prev = i;
-		sg++;
+	for (; n < (out_sgs + in_sgs); n++) {
+		for (sg = sgs[n]; sg; sg = next(sg, &total_in)) {
+			vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
+			vq->vring.desc[i].addr = sg_phys(sg);
+			vq->vring.desc[i].len = sg->length;
+			prev = i;
+			i = vq->vring.desc[i].next;
+		}
 	}
 	/* Last one doesn't continue. */
 	vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
@@ -276,7 +280,7 @@ add_head:
 
 	/* Descriptors and available array need to be set before we expose the
 	 * new available array entries. */
-	virtio_wmb(vq);
+	virtio_wmb(vq->weak_barriers);
 	vq->vring.avail->idx++;
 	vq->num_added++;
 
@@ -290,9 +294,122 @@ add_head:
 
 	return 0;
 }
+
+/**
+ * virtqueue_add_buf - expose buffer to other end
+ * @vq: the struct virtqueue we're talking about.
+ * @sg: the description of the buffer(s).
+ * @out_num: the number of sg readable by other side
+ * @in_num: the number of sg which are writable (after readable ones)
+ * @data: the token identifying the buffer.
+ * @gfp: how to do memory allocations (if necessary).
+ *
+ * Caller must ensure we don't call this with other virtqueue operations
+ * at the same time (except where noted).
+ *
+ * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
+ */
+int virtqueue_add_buf(struct virtqueue *_vq,
+		      struct scatterlist sg[],
+		      unsigned int out,
+		      unsigned int in,
+		      void *data,
+		      gfp_t gfp)
+{
+	struct scatterlist *sgs[2];
+
+	sgs[0] = sg;
+	sgs[1] = sg + out;
+
+	return virtqueue_add(_vq, sgs, sg_next_arr,
+			     out, in, out ? 1 : 0, in ? 1 : 0, data, gfp);
+}
 EXPORT_SYMBOL_GPL(virtqueue_add_buf);
 
 /**
+ * virtqueue_add_sgs - expose buffers to other end
+ * @vq: the struct virtqueue we're talking about.
+ * @sgs: array of terminated scatterlists.
+ * @out_num: the number of scatterlists readable by other side
+ * @in_num: the number of scatterlists which are writable (after readable ones)
+ * @data: the token identifying the buffer.
+ * @gfp: how to do memory allocations (if necessary).
+ *
+ * Caller must ensure we don't call this with other virtqueue operations
+ * at the same time (except where noted).
+ *
+ * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
+ */
+int virtqueue_add_sgs(struct virtqueue *_vq,
+		      struct scatterlist *sgs[],
+		      unsigned int out_sgs,
+		      unsigned int in_sgs,
+		      void *data,
+		      gfp_t gfp)
+{
+	unsigned int i, total_out, total_in;
+
+	/* Count them first. */
+	for (i = total_out = total_in = 0; i < out_sgs; i++) {
+		struct scatterlist *sg;
+		for (sg = sgs[i]; sg; sg = sg_next(sg))
+			total_out++;
+	}
+	for (; i < out_sgs + in_sgs; i++) {
+		struct scatterlist *sg;
+		for (sg = sgs[i]; sg; sg = sg_next(sg))
+			total_in++;
+	}
+	return virtqueue_add(_vq, sgs, sg_next_chained,
+			     total_out, total_in, out_sgs, in_sgs, data, gfp);
+}
+EXPORT_SYMBOL_GPL(virtqueue_add_sgs);
+
+/**
+ * virtqueue_add_outbuf - expose output buffers to other end
+ * @vq: the struct virtqueue we're talking about.
+ * @sgs: array of scatterlists (need not be terminated!)
+ * @num: the number of scatterlists readable by other side
+ * @data: the token identifying the buffer.
+ * @gfp: how to do memory allocations (if necessary).
+ *
+ * Caller must ensure we don't call this with other virtqueue operations
+ * at the same time (except where noted).
+ *
+ * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
+ */
+int virtqueue_add_outbuf(struct virtqueue *vq,
+			 struct scatterlist sg[], unsigned int num,
+			 void *data,
+			 gfp_t gfp)
+{
+	return virtqueue_add(vq, &sg, sg_next_arr, num, 0, 1, 0, data, gfp);
+}
+EXPORT_SYMBOL_GPL(virtqueue_add_outbuf);
+
+/**
+ * virtqueue_add_inbuf - expose input buffers to other end
+ * @vq: the struct virtqueue we're talking about.
+ * @sgs: array of scatterlists (need not be terminated!)
+ * @num: the number of scatterlists writable by other side
+ * @data: the token identifying the buffer.
+ * @gfp: how to do memory allocations (if necessary).
+ *
+ * Caller must ensure we don't call this with other virtqueue operations
+ * at the same time (except where noted).
+ *
+ * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
+ */
+int virtqueue_add_inbuf(struct virtqueue *vq,
+			struct scatterlist sg[], unsigned int num,
+			void *data,
+			gfp_t gfp)
+{
+	return virtqueue_add(vq, &sg, sg_next_arr, 0, num, 0, 1, data, gfp);
+}
+EXPORT_SYMBOL_GPL(virtqueue_add_inbuf);
+
+/**
  * virtqueue_kick_prepare - first half of split virtqueue_kick call.
  * @vq: the struct virtqueue
  *
@@ -312,7 +429,7 @@ bool virtqueue_kick_prepare(struct virtqueue *_vq)
 	START_USE(vq);
 	/* We need to expose available array entries before checking avail
 	 * event. */
-	virtio_mb(vq);
+	virtio_mb(vq->weak_barriers);
 
 	old = vq->vring.avail->idx - vq->num_added;
 	new = vq->vring.avail->idx;
@@ -436,7 +553,7 @@ void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
 	}
 
 	/* Only get used array entries after they have been exposed by host. */
-	virtio_rmb(vq);
+	virtio_rmb(vq->weak_barriers);
 
 	last_used = (vq->last_used_idx & (vq->vring.num - 1));
 	i = vq->vring.used->ring[last_used].id;
@@ -460,7 +577,7 @@ void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
 	 * the read in the next get_buf call. */
 	if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) {
 		vring_used_event(&vq->vring) = vq->last_used_idx;
-		virtio_mb(vq);
+		virtio_mb(vq->weak_barriers);
 	}
 
 #ifdef DEBUG
@@ -513,7 +630,7 @@ bool virtqueue_enable_cb(struct virtqueue *_vq)
 	 * entry. Always do both to keep code simple. */
 	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 	vring_used_event(&vq->vring) = vq->last_used_idx;
-	virtio_mb(vq);
+	virtio_mb(vq->weak_barriers);
 	if (unlikely(more_used(vq))) {
 		END_USE(vq);
 		return false;
@@ -553,7 +670,7 @@ bool virtqueue_enable_cb_delayed(struct virtqueue *_vq)
 	/* TODO: tune this threshold */
 	bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4;
 	vring_used_event(&vq->vring) = vq->last_used_idx + bufs;
-	virtio_mb(vq);
+	virtio_mb(vq->weak_barriers);
 	if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) {
 		END_USE(vq);
 		return false;