Merge git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml

Pull UML updates from Richard Weinberger:

 - a new and faster epoll based IRQ controller and NIC driver

 - misc fixes and janitorial updates

* git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml:
  Fix vector raw inintialization logic
  Migrate vector timers to new timer API
  um: Compile with modern headers
  um: vector: Fix an error handling path in 'vector_parse()'
  um: vector: Fix a memory allocation check
  um: vector: fix missing unlock on error in vector_net_open()
  um: Add missing EXPORT for free_irq_by_fd()
  High Performance UML Vector Network Driver
  Epoll based IRQ controller
  um: Use POSIX ucontext_t instead of struct ucontext
  um: time: Use timespec64 for persistent clock
  um: Restore symbol versions for __memcpy and memcpy

22 files changed, 3393 insertions, 324 deletions

diff --git a/arch/um/Kconfig.net b/arch/um/Kconfig.net
index e871af24d9cd..c390f3deb0dc 100644
--- a/arch/um/Kconfig.net
+++ b/arch/um/Kconfig.net
@@ -109,6 +109,17 @@ config UML_NET_DAEMON
         more than one without conflict.  If you don't need UML networking,
         say N.
 
+config UML_NET_VECTOR
+	bool "Vector I/O high performance network devices"
+	depends on UML_NET
+	help
+	This User-Mode Linux network driver uses multi-message send
+	and receive functions. The host running the UML guest must have
+	a linux kernel version above 3.0 and a libc version > 2.13.
+	This driver provides tap, raw, gre and l2tpv3 network transports
+	with up to 4 times higher network throughput than the UML network
+	drivers.
+
 config UML_NET_VDE
 	bool "VDE transport"
 	depends on UML_NET
diff --git a/arch/um/drivers/Makefile b/arch/um/drivers/Makefile
index e7582e1d248c..16b3cebddafb 100644
--- a/arch/um/drivers/Makefile
+++ b/arch/um/drivers/Makefile
@@ -9,6 +9,7 @@
 slip-objs := slip_kern.o slip_user.o
 slirp-objs := slirp_kern.o slirp_user.o
 daemon-objs := daemon_kern.o daemon_user.o
+vector-objs := vector_kern.o vector_user.o vector_transports.o
 umcast-objs := umcast_kern.o umcast_user.o
 net-objs := net_kern.o net_user.o
 mconsole-objs := mconsole_kern.o mconsole_user.o
@@ -43,6 +44,7 @@ obj-$(CONFIG_STDERR_CONSOLE) += stderr_console.o
 obj-$(CONFIG_UML_NET_SLIP) += slip.o slip_common.o
 obj-$(CONFIG_UML_NET_SLIRP) += slirp.o slip_common.o
 obj-$(CONFIG_UML_NET_DAEMON) += daemon.o 
+obj-$(CONFIG_UML_NET_VECTOR) += vector.o
 obj-$(CONFIG_UML_NET_VDE) += vde.o
 obj-$(CONFIG_UML_NET_MCAST) += umcast.o
 obj-$(CONFIG_UML_NET_PCAP) += pcap.o
@@ -61,7 +63,7 @@ obj-$(CONFIG_BLK_DEV_COW_COMMON) += cow_user.o
 obj-$(CONFIG_UML_RANDOM) += random.o
 
 # pcap_user.o must be added explicitly.
-USER_OBJS := fd.o null.o pty.o tty.o xterm.o slip_common.o pcap_user.o vde_user.o
+USER_OBJS := fd.o null.o pty.o tty.o xterm.o slip_common.o pcap_user.o vde_user.o vector_user.o
 CFLAGS_null.o = -DDEV_NULL=$(DEV_NULL_PATH)
 
 include arch/um/scripts/Makefile.rules
diff --git a/arch/um/drivers/chan_kern.c b/arch/um/drivers/chan_kern.c
index acbe6c67afba..05588f9466c7 100644
--- a/arch/um/drivers/chan_kern.c
+++ b/arch/um/drivers/chan_kern.c
@@ -171,56 +171,19 @@ int enable_chan(struct line *line)
 	return err;
 }
 
-/* Items are added in IRQ context, when free_irq can't be called, and
- * removed in process context, when it can.
- * This handles interrupt sources which disappear, and which need to
- * be permanently disabled.  This is discovered in IRQ context, but
- * the freeing of the IRQ must be done later.
- */
-static DEFINE_SPINLOCK(irqs_to_free_lock);
-static LIST_HEAD(irqs_to_free);
-
-void free_irqs(void)
-{
-	struct chan *chan;
-	LIST_HEAD(list);
-	struct list_head *ele;
-	unsigned long flags;
-
-	spin_lock_irqsave(&irqs_to_free_lock, flags);
-	list_splice_init(&irqs_to_free, &list);
-	spin_unlock_irqrestore(&irqs_to_free_lock, flags);
-
-	list_for_each(ele, &list) {
-		chan = list_entry(ele, struct chan, free_list);
-
-		if (chan->input && chan->enabled)
-			um_free_irq(chan->line->driver->read_irq, chan);
-		if (chan->output && chan->enabled)
-			um_free_irq(chan->line->driver->write_irq, chan);
-		chan->enabled = 0;
-	}
-}
-
 static void close_one_chan(struct chan *chan, int delay_free_irq)
 {
-	unsigned long flags;
-
 	if (!chan->opened)
 		return;
 
-	if (delay_free_irq) {
-		spin_lock_irqsave(&irqs_to_free_lock, flags);
-		list_add(&chan->free_list, &irqs_to_free);
-		spin_unlock_irqrestore(&irqs_to_free_lock, flags);
-	}
-	else {
-		if (chan->input && chan->enabled)
-			um_free_irq(chan->line->driver->read_irq, chan);
-		if (chan->output && chan->enabled)
-			um_free_irq(chan->line->driver->write_irq, chan);
-		chan->enabled = 0;
-	}
+    /* we can safely call free now - it will be marked
+     *  as free and freed once the IRQ stopped processing
+     */
+	if (chan->input && chan->enabled)
+		um_free_irq(chan->line->driver->read_irq, chan);
+	if (chan->output && chan->enabled)
+		um_free_irq(chan->line->driver->write_irq, chan);
+	chan->enabled = 0;
 	if (chan->ops->close != NULL)
 		(*chan->ops->close)(chan->fd, chan->data);
 
diff --git a/arch/um/drivers/line.c b/arch/um/drivers/line.c
index 366e57f5e8d6..8d80b27502e6 100644
--- a/arch/um/drivers/line.c
+++ b/arch/um/drivers/line.c
@@ -284,7 +284,7 @@ int line_setup_irq(int fd, int input, int output, struct line *line, void *data)
 	if (err)
 		return err;
 	if (output)
-		err = um_request_irq(driver->write_irq, fd, IRQ_WRITE,
+		err = um_request_irq(driver->write_irq, fd, IRQ_NONE,
 				     line_write_interrupt, IRQF_SHARED,
 				     driver->write_irq_name, data);
 	return err;
diff --git a/arch/um/drivers/net_kern.c b/arch/um/drivers/net_kern.c
index b305f8247909..3ef1b48e064a 100644
--- a/arch/um/drivers/net_kern.c
+++ b/arch/um/drivers/net_kern.c
@@ -288,7 +288,7 @@ static void uml_net_user_timer_expire(struct timer_list *t)
 #endif
 }
 
-static void setup_etheraddr(struct net_device *dev, char *str)
+void uml_net_setup_etheraddr(struct net_device *dev, char *str)
 {
 	unsigned char *addr = dev->dev_addr;
 	char *end;
@@ -412,7 +412,7 @@ static void eth_configure(int n, void *init, char *mac,
 	 */
 	snprintf(dev->name, sizeof(dev->name), "eth%d", n);
 
-	setup_etheraddr(dev, mac);
+	uml_net_setup_etheraddr(dev, mac);
 
 	printk(KERN_INFO "Netdevice %d (%pM) : ", n, dev->dev_addr);
 
diff --git a/arch/um/drivers/random.c b/arch/um/drivers/random.c
index 37c51a6be690..778a0e52d5a5 100644
--- a/arch/um/drivers/random.c
+++ b/arch/um/drivers/random.c
@@ -13,6 +13,7 @@
 #include <linux/miscdevice.h>
 #include <linux/delay.h>
 #include <linux/uaccess.h>
+#include <init.h>
 #include <irq_kern.h>
 #include <os.h>
 
@@ -154,7 +155,14 @@ err_out_cleanup_hw:
 /*
  * rng_cleanup - shutdown RNG module
  */
-static void __exit rng_cleanup (void)
+
+static void cleanup(void)
+{
+	free_irq_by_fd(random_fd);
+	os_close_file(random_fd);
+}
+
+static void __exit rng_cleanup(void)
 {
 	os_close_file(random_fd);
 	misc_deregister (&rng_miscdev);
@@ -162,6 +170,7 @@ static void __exit rng_cleanup (void)
 
 module_init (rng_init);
 module_exit (rng_cleanup);
+__uml_exitcall(cleanup);
 
 MODULE_DESCRIPTION("UML Host Random Number Generator (RNG) driver");
 MODULE_LICENSE("GPL");
diff --git a/arch/um/drivers/ubd_kern.c b/arch/um/drivers/ubd_kern.c
index b55fe9bf5d3e..d4e8c497ae86 100644
--- a/arch/um/drivers/ubd_kern.c
+++ b/arch/um/drivers/ubd_kern.c
@@ -1587,11 +1587,11 @@ int io_thread(void *arg)
 
 		do {
 			res = os_write_file(kernel_fd, ((char *) io_req_buffer) + written, n);
-			if (res > 0) {
+			if (res >= 0) {
 				written += res;
 			} else {
 				if (res != -EAGAIN) {
-					printk("io_thread - read failed, fd = %d, "
+					printk("io_thread - write failed, fd = %d, "
 					       "err = %d\n", kernel_fd, -n);
 				}
 			}
diff --git a/arch/um/drivers/vector_kern.c b/arch/um/drivers/vector_kern.c
new file mode 100644
index 000000000000..02168fe25105
--- /dev/null
+++ b/arch/um/drivers/vector_kern.c
@@ -0,0 +1,1633 @@
+/*
+ * Copyright (C) 2017 - Cambridge Greys Limited
+ * Copyright (C) 2011 - 2014 Cisco Systems Inc
+ * Copyright (C) 2001 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
+ * Copyright (C) 2001 Lennert Buytenhek (buytenh@gnu.org) and
+ * James Leu (jleu@mindspring.net).
+ * Copyright (C) 2001 by various other people who didn't put their name here.
+ * Licensed under the GPL.
+ */
+
+#include <linux/version.h>
+#include <linux/bootmem.h>
+#include <linux/etherdevice.h>
+#include <linux/ethtool.h>
+#include <linux/inetdevice.h>
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/netdevice.h>
+#include <linux/platform_device.h>
+#include <linux/rtnetlink.h>
+#include <linux/skbuff.h>
+#include <linux/slab.h>
+#include <linux/interrupt.h>
+#include <init.h>
+#include <irq_kern.h>
+#include <irq_user.h>
+#include <net_kern.h>
+#include <os.h>
+#include "mconsole_kern.h"
+#include "vector_user.h"
+#include "vector_kern.h"
+
+/*
+ * Adapted from network devices with the following major changes:
+ * All transports are static - simplifies the code significantly
+ * Multiple FDs/IRQs per device
+ * Vector IO optionally used for read/write, falling back to legacy
+ * based on configuration and/or availability
+ * Configuration is no longer positional - L2TPv3 and GRE require up to
+ * 10 parameters, passing this as positional is not fit for purpose.
+ * Only socket transports are supported
+ */
+
+
+#define DRIVER_NAME "uml-vector"
+#define DRIVER_VERSION "01"
+struct vector_cmd_line_arg {
+	struct list_head list;
+	int unit;
+	char *arguments;
+};
+
+struct vector_device {
+	struct list_head list;
+	struct net_device *dev;
+	struct platform_device pdev;
+	int unit;
+	int opened;
+};
+
+static LIST_HEAD(vec_cmd_line);
+
+static DEFINE_SPINLOCK(vector_devices_lock);
+static LIST_HEAD(vector_devices);
+
+static int driver_registered;
+
+static void vector_eth_configure(int n, struct arglist *def);
+
+/* Argument accessors to set variables (and/or set default values)
+ * mtu, buffer sizing, default headroom, etc
+ */
+
+#define DEFAULT_HEADROOM 2
+#define SAFETY_MARGIN 32
+#define DEFAULT_VECTOR_SIZE 64
+#define TX_SMALL_PACKET 128
+#define MAX_IOV_SIZE (MAX_SKB_FRAGS + 1)
+
+static const struct {
+	const char string[ETH_GSTRING_LEN];
+} ethtool_stats_keys[] = {
+	{ "rx_queue_max" },
+	{ "rx_queue_running_average" },
+	{ "tx_queue_max" },
+	{ "tx_queue_running_average" },
+	{ "rx_encaps_errors" },
+	{ "tx_timeout_count" },
+	{ "tx_restart_queue" },
+	{ "tx_kicks" },
+	{ "tx_flow_control_xon" },
+	{ "tx_flow_control_xoff" },
+	{ "rx_csum_offload_good" },
+	{ "rx_csum_offload_errors"},
+	{ "sg_ok"},
+	{ "sg_linearized"},
+};
+
+#define VECTOR_NUM_STATS	ARRAY_SIZE(ethtool_stats_keys)
+
+static void vector_reset_stats(struct vector_private *vp)
+{
+	vp->estats.rx_queue_max = 0;
+	vp->estats.rx_queue_running_average = 0;
+	vp->estats.tx_queue_max = 0;
+	vp->estats.tx_queue_running_average = 0;
+	vp->estats.rx_encaps_errors = 0;
+	vp->estats.tx_timeout_count = 0;
+	vp->estats.tx_restart_queue = 0;
+	vp->estats.tx_kicks = 0;
+	vp->estats.tx_flow_control_xon = 0;
+	vp->estats.tx_flow_control_xoff = 0;
+	vp->estats.sg_ok = 0;
+	vp->estats.sg_linearized = 0;
+}
+
+static int get_mtu(struct arglist *def)
+{
+	char *mtu = uml_vector_fetch_arg(def, "mtu");
+	long result;
+
+	if (mtu != NULL) {
+		if (kstrtoul(mtu, 10, &result) == 0)
+			return result;
+	}
+	return ETH_MAX_PACKET;
+}
+
+static int get_depth(struct arglist *def)
+{
+	char *mtu = uml_vector_fetch_arg(def, "depth");
+	long result;
+
+	if (mtu != NULL) {
+		if (kstrtoul(mtu, 10, &result) == 0)
+			return result;
+	}
+	return DEFAULT_VECTOR_SIZE;
+}
+
+static int get_headroom(struct arglist *def)
+{
+	char *mtu = uml_vector_fetch_arg(def, "headroom");
+	long result;
+
+	if (mtu != NULL) {
+		if (kstrtoul(mtu, 10, &result) == 0)
+			return result;
+	}
+	return DEFAULT_HEADROOM;
+}
+
+static int get_req_size(struct arglist *def)
+{
+	char *gro = uml_vector_fetch_arg(def, "gro");
+	long result;
+
+	if (gro != NULL) {
+		if (kstrtoul(gro, 10, &result) == 0) {
+			if (result > 0)
+				return 65536;
+		}
+	}
+	return get_mtu(def) + ETH_HEADER_OTHER +
+		get_headroom(def) + SAFETY_MARGIN;
+}
+
+
+static int get_transport_options(struct arglist *def)
+{
+	char *transport = uml_vector_fetch_arg(def, "transport");
+	char *vector = uml_vector_fetch_arg(def, "vec");
+
+	int vec_rx = VECTOR_RX;
+	int vec_tx = VECTOR_TX;
+	long parsed;
+
+	if (vector != NULL) {
+		if (kstrtoul(vector, 10, &parsed) == 0) {
+			if (parsed == 0) {
+				vec_rx = 0;
+				vec_tx = 0;
+			}
+		}
+	}
+
+
+	if (strncmp(transport, TRANS_TAP, TRANS_TAP_LEN) == 0)
+		return (vec_rx | VECTOR_BPF);
+	if (strncmp(transport, TRANS_RAW, TRANS_RAW_LEN) == 0)
+		return (vec_rx | vec_tx);
+	return (vec_rx | vec_tx);
+}
+
+
+/* A mini-buffer for packet drop read
+ * All of our supported transports are datagram oriented and we always
+ * read using recvmsg or recvmmsg. If we pass a buffer which is smaller
+ * than the packet size it still counts as full packet read and will
+ * clean the incoming stream to keep sigio/epoll happy
+ */
+
+#define DROP_BUFFER_SIZE 32
+
+static char *drop_buffer;
+
+/* Array backed queues optimized for bulk enqueue/dequeue and
+ * 1:N (small values of N) or 1:1 enqueuer/dequeuer ratios.
+ * For more details and full design rationale see
+ * http://foswiki.cambridgegreys.com/Main/EatYourTailAndEnjoyIt
+ */
+
+
+/*
+ * Advance the mmsg queue head by n = advance. Resets the queue to
+ * maximum enqueue/dequeue-at-once capacity if possible. Called by
+ * dequeuers. Caller must hold the head_lock!
+ */
+
+static int vector_advancehead(struct vector_queue *qi, int advance)
+{
+	int queue_depth;
+
+	qi->head =
+		(qi->head + advance)
+			% qi->max_depth;
+
+
+	spin_lock(&qi->tail_lock);
+	qi->queue_depth -= advance;
+
+	/* we are at 0, use this to
+	 * reset head and tail so we can use max size vectors
+	 */
+
+	if (qi->queue_depth == 0) {
+		qi->head = 0;
+		qi->tail = 0;
+	}
+	queue_depth = qi->queue_depth;
+	spin_unlock(&qi->tail_lock);
+	return queue_depth;
+}
+
+/*	Advance the queue tail by n = advance.
+ *	This is called by enqueuers which should hold the
+ *	head lock already
+ */
+
+static int vector_advancetail(struct vector_queue *qi, int advance)
+{
+	int queue_depth;
+
+	qi->tail =
+		(qi->tail + advance)
+			% qi->max_depth;
+	spin_lock(&qi->head_lock);
+	qi->queue_depth += advance;
+	queue_depth = qi->queue_depth;
+	spin_unlock(&qi->head_lock);
+	return queue_depth;
+}
+
+static int prep_msg(struct vector_private *vp,
+	struct sk_buff *skb,
+	struct iovec *iov)
+{
+	int iov_index = 0;
+	int nr_frags, frag;
+	skb_frag_t *skb_frag;
+
+	nr_frags = skb_shinfo(skb)->nr_frags;
+	if (nr_frags > MAX_IOV_SIZE) {
+		if (skb_linearize(skb) != 0)
+			goto drop;
+	}
+	if (vp->header_size > 0) {
+		iov[iov_index].iov_len = vp->header_size;
+		vp->form_header(iov[iov_index].iov_base, skb, vp);
+		iov_index++;
+	}
+	iov[iov_index].iov_base = skb->data;
+	if (nr_frags > 0) {
+		iov[iov_index].iov_len = skb->len - skb->data_len;
+		vp->estats.sg_ok++;
+	} else
+		iov[iov_index].iov_len = skb->len;
+	iov_index++;
+	for (frag = 0; frag < nr_frags; frag++) {
+		skb_frag = &skb_shinfo(skb)->frags[frag];
+		iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
+		iov[iov_index].iov_len = skb_frag_size(skb_frag);
+		iov_index++;
+	}
+	return iov_index;
+drop:
+	return -1;
+}
+/*
+ * Generic vector enqueue with support for forming headers using transport
+ * specific callback. Allows GRE, L2TPv3, RAW and other transports
+ * to use a common enqueue procedure in vector mode
+ */
+
+static int vector_enqueue(struct vector_queue *qi, struct sk_buff *skb)
+{
+	struct vector_private *vp = netdev_priv(qi->dev);
+	int queue_depth;
+	int packet_len;
+	struct mmsghdr *mmsg_vector = qi->mmsg_vector;
+	int iov_count;
+
+	spin_lock(&qi->tail_lock);
+	spin_lock(&qi->head_lock);
+	queue_depth = qi->queue_depth;
+	spin_unlock(&qi->head_lock);
+
+	if (skb)
+		packet_len = skb->len;
+
+	if (queue_depth < qi->max_depth) {
+
+		*(qi->skbuff_vector + qi->tail) = skb;
+		mmsg_vector += qi->tail;
+		iov_count = prep_msg(
+			vp,
+			skb,
+			mmsg_vector->msg_hdr.msg_iov
+		);
+		if (iov_count < 1)
+			goto drop;
+		mmsg_vector->msg_hdr.msg_iovlen = iov_count;
+		mmsg_vector->msg_hdr.msg_name = vp->fds->remote_addr;
+		mmsg_vector->msg_hdr.msg_namelen = vp->fds->remote_addr_size;
+		queue_depth = vector_advancetail(qi, 1);
+	} else
+		goto drop;
+	spin_unlock(&qi->tail_lock);
+	return queue_depth;
+drop:
+	qi->dev->stats.tx_dropped++;
+	if (skb != NULL) {
+		packet_len = skb->len;
+		dev_consume_skb_any(skb);
+		netdev_completed_queue(qi->dev, 1, packet_len);
+	}
+	spin_unlock(&qi->tail_lock);
+	return queue_depth;
+}
+
+static int consume_vector_skbs(struct vector_queue *qi, int count)
+{
+	struct sk_buff *skb;
+	int skb_index;
+	int bytes_compl = 0;
+
+	for (skb_index = qi->head; skb_index < qi->head + count; skb_index++) {
+		skb = *(qi->skbuff_vector + skb_index);
+		/* mark as empty to ensure correct destruction if
+		 * needed
+		 */
+		bytes_compl += skb->len;
+		*(qi->skbuff_vector + skb_index) = NULL;
+		dev_consume_skb_any(skb);
+	}
+	qi->dev->stats.tx_bytes += bytes_compl;
+	qi->dev->stats.tx_packets += count;
+	netdev_completed_queue(qi->dev, count, bytes_compl);
+	return vector_advancehead(qi, count);
+}
+
+/*
+ * Generic vector deque via sendmmsg with support for forming headers
+ * using transport specific callback. Allows GRE, L2TPv3, RAW and
+ * other transports to use a common dequeue procedure in vector mode
+ */
+
+
+static int vector_send(struct vector_queue *qi)
+{
+	struct vector_private *vp = netdev_priv(qi->dev);
+	struct mmsghdr *send_from;
+	int result = 0, send_len, queue_depth = qi->max_depth;
+
+	if (spin_trylock(&qi->head_lock)) {
+		if (spin_trylock(&qi->tail_lock)) {
+			/* update queue_depth to current value */
+			queue_depth = qi->queue_depth;
+			spin_unlock(&qi->tail_lock);
+			while (queue_depth > 0) {
+				/* Calculate the start of the vector */
+				send_len = queue_depth;
+				send_from = qi->mmsg_vector;
+				send_from += qi->head;
+				/* Adjust vector size if wraparound */
+				if (send_len + qi->head > qi->max_depth)
+					send_len = qi->max_depth - qi->head;
+				/* Try to TX as many packets as possible */
+				if (send_len > 0) {
+					result = uml_vector_sendmmsg(
+						 vp->fds->tx_fd,
+						 send_from,
+						 send_len,
+						 0
+					);
+					vp->in_write_poll =
+						(result != send_len);
+				}
+				/* For some of the sendmmsg error scenarios
+				 * we may end being unsure in the TX success
+				 * for all packets. It is safer to declare
+				 * them all TX-ed and blame the network.
+				 */
+				if (result < 0) {
+					if (net_ratelimit())
+						netdev_err(vp->dev, "sendmmsg err=%i\n",
+							result);
+					result = send_len;
+				}
+				if (result > 0) {
+					queue_depth =
+						consume_vector_skbs(qi, result);
+					/* This is equivalent to an TX IRQ.
+					 * Restart the upper layers to feed us
+					 * more packets.
+					 */
+					if (result > vp->estats.tx_queue_max)
+						vp->estats.tx_queue_max = result;
+					vp->estats.tx_queue_running_average =
+						(vp->estats.tx_queue_running_average + result) >> 1;
+				}
+				netif_trans_update(qi->dev);
+				netif_wake_queue(qi->dev);
+				/* if TX is busy, break out of the send loop,
+				 *  poll write IRQ will reschedule xmit for us
+				 */
+				if (result != send_len) {
+					vp->estats.tx_restart_queue++;
+					break;
+				}
+			}
+		}
+		spin_unlock(&qi->head_lock);
+	} else {
+		tasklet_schedule(&vp->tx_poll);
+	}
+	return queue_depth;
+}
+
+/* Queue destructor. Deliberately stateless so we can use
+ * it in queue cleanup if initialization fails.
+ */
+
+static void destroy_queue(struct vector_queue *qi)
+{
+	int i;
+	struct iovec *iov;
+	struct vector_private *vp = netdev_priv(qi->dev);
+	struct mmsghdr *mmsg_vector;
+
+	if (qi == NULL)
+		return;
+	/* deallocate any skbuffs - we rely on any unused to be
+	 * set to NULL.
+	 */
+	if (qi->skbuff_vector != NULL) {
+		for (i = 0; i < qi->max_depth; i++) {
+			if (*(qi->skbuff_vector + i) != NULL)
+				dev_kfree_skb_any(*(qi->skbuff_vector + i));
+		}
+		kfree(qi->skbuff_vector);
+	}
+	/* deallocate matching IOV structures including header buffs */
+	if (qi->mmsg_vector != NULL) {
+		mmsg_vector = qi->mmsg_vector;
+		for (i = 0; i < qi->max_depth; i++) {
+			iov = mmsg_vector->msg_hdr.msg_iov;
+			if (iov != NULL) {
+				if ((vp->header_size > 0) &&
+					(iov->iov_base != NULL))
+					kfree(iov->iov_base);
+				kfree(iov);
+			}
+			mmsg_vector++;
+		}
+		kfree(qi->mmsg_vector);
+	}
+	kfree(qi);
+}
+
+/*
+ * Queue constructor. Create a queue with a given side.
+ */
+static struct vector_queue *create_queue(
+	struct vector_private *vp,
+	int max_size,
+	int header_size,
+	int num_extra_frags)
+{
+	struct vector_queue *result;
+	int i;
+	struct iovec *iov;
+	struct mmsghdr *mmsg_vector;
+
+	result = kmalloc(sizeof(struct vector_queue), GFP_KERNEL);
+	if (result == NULL)
+		goto out_fail;
+	result->max_depth = max_size;
+	result->dev = vp->dev;
+	result->mmsg_vector = kmalloc(
+		(sizeof(struct mmsghdr) * max_size), GFP_KERNEL);
+	result->skbuff_vector = kmalloc(
+		(sizeof(void *) * max_size), GFP_KERNEL);
+	if (result->mmsg_vector == NULL || result->skbuff_vector == NULL)
+		goto out_fail;
+
+	mmsg_vector = result->mmsg_vector;
+	for (i = 0; i < max_size; i++) {
+		/* Clear all pointers - we use non-NULL as marking on
+		 * what to free on destruction
+		 */
+		*(result->skbuff_vector + i) = NULL;
+		mmsg_vector->msg_hdr.msg_iov = NULL;
+		mmsg_vector++;
+	}
+	mmsg_vector = result->mmsg_vector;
+	result->max_iov_frags = num_extra_frags;
+	for (i = 0; i < max_size; i++) {
+		if (vp->header_size > 0)
+			iov = kmalloc(
+				sizeof(struct iovec) * (3 + num_extra_frags),
+				GFP_KERNEL
+			);
+		else
+			iov = kmalloc(
+				sizeof(struct iovec) * (2 + num_extra_frags),
+				GFP_KERNEL
+			);
+		if (iov == NULL)
+			goto out_fail;
+		mmsg_vector->msg_hdr.msg_iov = iov;
+		mmsg_vector->msg_hdr.msg_iovlen = 1;
+		mmsg_vector->msg_hdr.msg_control = NULL;
+		mmsg_vector->msg_hdr.msg_controllen = 0;
+		mmsg_vector->msg_hdr.msg_flags = MSG_DONTWAIT;
+		mmsg_vector->msg_hdr.msg_name = NULL;
+		mmsg_vector->msg_hdr.msg_namelen = 0;
+		if (vp->header_size > 0) {
+			iov->iov_base = kmalloc(header_size, GFP_KERNEL);
+			if (iov->iov_base == NULL)
+				goto out_fail;
+			iov->iov_len = header_size;
+			mmsg_vector->msg_hdr.msg_iovlen = 2;
+			iov++;
+		}
+		iov->iov_base = NULL;
+		iov->iov_len = 0;
+		mmsg_vector++;
+	}
+	spin_lock_init(&result->head_lock);
+	spin_lock_init(&result->tail_lock);
+	result->queue_depth = 0;
+	result->head = 0;
+	result->tail = 0;
+	return result;
+out_fail:
+	destroy_queue(result);
+	return NULL;
+}
+
+/*
+ * We do not use the RX queue as a proper wraparound queue for now
+ * This is not necessary because the consumption via netif_rx()
+ * happens in-line. While we can try using the return code of
+ * netif_rx() for flow control there are no drivers doing this today.
+ * For this RX specific use we ignore the tail/head locks and
+ * just read into a prepared queue filled with skbuffs.
+ */
+
+static struct sk_buff *prep_skb(
+	struct vector_private *vp,
+	struct user_msghdr *msg)
+{
+	int linear = vp->max_packet + vp->headroom + SAFETY_MARGIN;
+	struct sk_buff *result;
+	int iov_index = 0, len;
+	struct iovec *iov = msg->msg_iov;
+	int err, nr_frags, frag;
+	skb_frag_t *skb_frag;
+
+	if (vp->req_size <= linear)
+		len = linear;
+	else
+		len = vp->req_size;
+	result = alloc_skb_with_frags(
+		linear,
+		len - vp->max_packet,
+		3,
+		&err,
+		GFP_ATOMIC
+	);
+	if (vp->header_size > 0)
+		iov_index++;
+	if (result == NULL) {
+		iov[iov_index].iov_base = NULL;
+		iov[iov_index].iov_len = 0;
+		goto done;
+	}
+	skb_reserve(result, vp->headroom);
+	result->dev = vp->dev;
+	skb_put(result, vp->max_packet);
+	result->data_len = len - vp->max_packet;
+	result->len += len - vp->max_packet;
+	skb_reset_mac_header(result);
+	result->ip_summed = CHECKSUM_NONE;
+	iov[iov_index].iov_base = result->data;
+	iov[iov_index].iov_len = vp->max_packet;
+	iov_index++;
+
+	nr_frags = skb_shinfo(result)->nr_frags;
+	for (frag = 0; frag < nr_frags; frag++) {
+		skb_frag = &skb_shinfo(result)->frags[frag];
+		iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
+		if (iov[iov_index].iov_base != NULL)
+			iov[iov_index].iov_len = skb_frag_size(skb_frag);
+		else
+			iov[iov_index].iov_len = 0;
+		iov_index++;
+	}
+done:
+	msg->msg_iovlen = iov_index;
+	return result;
+}
+
+
+/* Prepare queue for recvmmsg one-shot rx - fill with fresh sk_buffs*/
+
+static void prep_queue_for_rx(struct vector_queue *qi)
+{
+	struct vector_private *vp = netdev_priv(qi->dev);
+	struct mmsghdr *mmsg_vector = qi->mmsg_vector;
+	void **skbuff_vector = qi->skbuff_vector;
+	int i;
+
+	if (qi->queue_depth == 0)
+		return;
+	for (i = 0; i < qi->queue_depth; i++) {
+		/* it is OK if allocation fails - recvmmsg with NULL data in
+		 * iov argument still performs an RX, just drops the packet
+		 * This allows us stop faffing around with a "drop buffer"
+		 */
+
+		*skbuff_vector = prep_skb(vp, &mmsg_vector->msg_hdr);
+		skbuff_vector++;
+		mmsg_vector++;
+	}
+	qi->queue_depth = 0;
+}
+
+static struct vector_device *find_device(int n)
+{
+	struct vector_device *device;
+	struct list_head *ele;
+
+	spin_lock(&vector_devices_lock);
+	list_for_each(ele, &vector_devices) {
+		device = list_entry(ele, struct vector_device, list);
+		if (device->unit == n)
+			goto out;
+	}
+	device = NULL;
+ out:
+	spin_unlock(&vector_devices_lock);
+	return device;
+}
+
+static int vector_parse(char *str, int *index_out, char **str_out,
+			char **error_out)
+{
+	int n, len, err;
+	char *start = str;
+
+	len = strlen(str);
+
+	while ((*str != ':') && (strlen(str) > 1))
+		str++;
+	if (*str != ':') {
+		*error_out = "Expected ':' after device number";
+		return -EINVAL;
+	}
+	*str = '\0';
+
+	err = kstrtouint(start, 0, &n);
+	if (err < 0) {
+		*error_out = "Bad device number";
+		return err;
+	}
+
+	str++;
+	if (find_device(n)) {
+		*error_out = "Device already configured";
+		return -EINVAL;
+	}
+
+	*index_out = n;
+	*str_out = str;
+	return 0;
+}
+
+static int vector_config(char *str, char **error_out)
+{
+	int err, n;
+	char *params;
+	struct arglist *parsed;
+
+	err = vector_parse(str, &n, &params, error_out);
+	if (err != 0)
+		return err;
+
+	/* This string is broken up and the pieces used by the underlying
+	 * driver. We should copy it to make sure things do not go wrong
+	 * later.
+	 */
+
+	params = kstrdup(params, GFP_KERNEL);
+	if (params == NULL) {
+		*error_out = "vector_config failed to strdup string";
+		return -ENOMEM;
+	}
+
+	parsed = uml_parse_vector_ifspec(params);
+
+	if (parsed == NULL) {
+		*error_out = "vector_config failed to parse parameters";
+		return -EINVAL;
+	}
+
+	vector_eth_configure(n, parsed);
+	return 0;
+}
+
+static int vector_id(char **str, int *start_out, int *end_out)
+{
+	char *end;
+	int n;
+
+	n = simple_strtoul(*str, &end, 0);
+