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

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

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

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);
 

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;

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);
 

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 @@ static int __init rng_init (void)
 /*
  * 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");

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);
 				}
 			}

arch/um/drivers/vector_kern.h

+/*
+ * Copyright (C) 2002 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
+ * Licensed under the GPL
+ */
+
+#ifndef __UM_VECTOR_KERN_H
+#define __UM_VECTOR_KERN_H
+
+#include <linux/netdevice.h>
+#include <linux/platform_device.h>
+#include <linux/skbuff.h>
+#include <linux/socket.h>
+#include <linux/list.h>
+#include <linux/ctype.h>
+#include <linux/workqueue.h>
+#include <linux/interrupt.h>
+#include "vector_user.h"
+
+/* Queue structure specially adapted for multiple enqueue/dequeue
+ * in a mmsgrecv/mmsgsend context
+ */
+
+/* Dequeue method */
+
+#define QUEUE_SENDMSG 0
+#define QUEUE_SENDMMSG 1
+
+#define VECTOR_RX 1
+#define VECTOR_TX (1 << 1)
+#define VECTOR_BPF (1 << 2)
+#define VECTOR_QDISC_BYPASS (1 << 3)
+
+#define ETH_MAX_PACKET 1500
+#define ETH_HEADER_OTHER 32 /* just in case someone decides to go mad on QnQ */
+
+struct vector_queue {
+	struct mmsghdr *mmsg_vector;
+	void **skbuff_vector;
+	 /* backlink to device which owns us */
+	struct net_device *dev;
+	spinlock_t head_lock;
+	spinlock_t tail_lock;
+	int queue_depth, head, tail, max_depth, max_iov_frags;
+	short options;
+};
+
+struct vector_estats {
+	uint64_t rx_queue_max;
+	uint64_t rx_queue_running_average;
+	uint64_t tx_queue_max;
+	uint64_t tx_queue_running_average;
+	uint64_t rx_encaps_errors;
+	uint64_t tx_timeout_count;
+	uint64_t tx_restart_queue;
+	uint64_t tx_kicks;
+	uint64_t tx_flow_control_xon;
+	uint64_t tx_flow_control_xoff;
+	uint64_t rx_csum_offload_good;
+	uint64_t rx_csum_offload_errors;
+	uint64_t sg_ok;
+	uint64_t sg_linearized;
+};
+
+#define VERIFY_HEADER_NOK -1
+#define VERIFY_HEADER_OK 0
+#define VERIFY_CSUM_OK 1
+
+struct vector_private {
+	struct list_head list;
+	spinlock_t lock;
+	struct net_device *dev;
+
+	int unit;
+
+	/* Timeout timer in TX */
+
+	struct timer_list tl;
+
+	/* Scheduled "remove device" work */
+	struct work_struct reset_tx;
+	struct vector_fds *fds;
+
+	struct vector_queue *rx_queue;
+	struct vector_queue *tx_queue;
+
+	int rx_irq;
+	int tx_irq;
+
+	struct arglist *parsed;
+
+	void *transport_data; /* transport specific params if needed */
+
+	int max_packet;
+	int req_size; /* different from max packet - used for TSO */
+	int headroom;
+
+	int options;
+
+	/* remote address if any - some transports will leave this as null */
+
+	int header_size;
+	int rx_header_size;
+	int coalesce;
+
+	void *header_rxbuffer;
+	void *header_txbuffer;
+
+	int (*form_header)(uint8_t *header,
+		struct sk_buff *skb, struct vector_private *vp);
+	int (*verify_header)(uint8_t *header,
+		struct sk_buff *skb, struct vector_private *vp);
+
+	spinlock_t stats_lock;
+
+	struct tasklet_struct tx_poll;
+	bool rexmit_scheduled;
+	bool opened;
+	bool in_write_poll;
+
+	/* ethtool stats */
+
+	struct vector_estats estats;
+	void *bpf;
+
+	char user[0];
+};
+
+extern int build_transport_data(struct vector_private *vp);
+
+#endif

arch/um/drivers/vector_user.h

+/*
+ * Copyright (C) 2002 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
+ * Licensed under the GPL
+ */
+
+#ifndef __UM_VECTOR_USER_H
+#define __UM_VECTOR_USER_H
+
+#define MAXVARGS	20
+
+#define TOKEN_IFNAME "ifname"
+
+#define TRANS_RAW "raw"
+#define TRANS_RAW_LEN strlen(TRANS_RAW)
+
+#define TRANS_TAP "tap"
+#define TRANS_TAP_LEN strlen(TRANS_TAP)
+
+
+#define TRANS_GRE "gre"
+#define TRANS_GRE_LEN strlen(TRANS_RAW)
+
+#define TRANS_L2TPV3 "l2tpv3"
+#define TRANS_L2TPV3_LEN strlen(TRANS_L2TPV3)
+
+#ifndef IPPROTO_GRE
+#define IPPROTO_GRE 0x2F
+#endif
+
+#define GRE_MODE_CHECKSUM	cpu_to_be16(8 << 12)	/* checksum */
+#define GRE_MODE_RESERVED	cpu_to_be16(4 << 12)	/* unused */
+#define GRE_MODE_KEY		cpu_to_be16(2 << 12)	/* KEY present */
+#define GRE_MODE_SEQUENCE	cpu_to_be16(1 << 12)	/* sequence */
+
+#define GRE_IRB cpu_to_be16(0x6558)
+
+#define L2TPV3_DATA_PACKET 0x30000
+
+/* IANA-assigned IP protocol ID for L2TPv3 */
+
+#ifndef IPPROTO_L2TP
+#define IPPROTO_L2TP 0x73
+#endif
+
+struct arglist {
+	int	numargs;
+	char	*tokens[MAXVARGS];
+	char	*values[MAXVARGS];
+};
+
+/* Separating read and write FDs allows us to have different
+ * rx and tx method. Example - read tap via raw socket using
+ * recvmmsg, write using legacy tap write calls
+ */
+
+struct vector_fds {
+	int rx_fd;
+	int tx_fd;
+	void *remote_addr;
+	int remote_addr_size;
+};
+
+#define VECTOR_READ	1
+#define VECTOR_WRITE	(1 < 1)
+#define VECTOR_HEADERS	(1 < 2)
+
+extern struct arglist *uml_parse_vector_ifspec(char *arg);
+
+extern struct vector_fds *uml_vector_user_open(
+	int unit,
+	struct arglist *parsed
+);
+
+extern char *uml_vector_fetch_arg(
+	struct arglist *ifspec,
+	char *token
+);
+
+extern int uml_vector_recvmsg(int fd, void *hdr, int flags);
+extern int uml_vector_sendmsg(int fd, void *hdr, int flags);
+extern int uml_vector_writev(int fd, void *hdr, int iovcount);
+extern int uml_vector_sendmmsg(
+	int fd, void *msgvec,
+	unsigned int vlen,
+	unsigned int flags
+);
+extern int uml_vector_recvmmsg(
+	int fd,
+	void *msgvec,
+	unsigned int vlen,
+	unsigned int flags
+);
+extern void *uml_vector_default_bpf(int fd, void *mac);
+extern int uml_vector_attach_bpf(int fd, void *bpf, int bpf_len);
+extern bool uml_raw_enable_qdisc_bypass(int fd);
+extern bool uml_raw_enable_vnet_headers(int fd);
+extern bool uml_tap_enable_vnet_headers(int fd);
+
+
+#endif

arch/um/include/asm/asm-prototypes.h

+#include <asm-generic/asm-prototypes.h>

arch/um/include/asm/irq.h

@@ -18,7 +18,19 @@
 #define XTERM_IRQ 		13
 #define RANDOM_IRQ 		14
 
+#ifdef CONFIG_UML_NET_VECTOR
+
+#define VECTOR_BASE_IRQ		15
+#define VECTOR_IRQ_SPACE	8
+
+#define LAST_IRQ (VECTOR_IRQ_SPACE + VECTOR_BASE_IRQ)
+
+#else
+
 #define LAST_IRQ RANDOM_IRQ
+
+#endif
+
 #define NR_IRQS (LAST_IRQ + 1)
 
 #endif

arch/um/include/shared/irq_user.h

@@ -7,6 +7,7 @@
 #define __IRQ_USER_H__
 
 #include <sysdep/ptrace.h>
+#include <stdbool.h>
 
 struct irq_fd {
 	struct irq_fd *next;
@@ -15,10 +16,17 @@ struct irq_fd {
 	int type;
 	int irq;
 	int events;
-	int current_events;
+	bool active;
+	bool pending;
+	bool purge;
 };
 
-enum { IRQ_READ, IRQ_WRITE };
+#define IRQ_READ  0
+#define IRQ_WRITE 1
+#define IRQ_NONE 2
+#define MAX_IRQ_TYPE (IRQ_NONE + 1)
+
+
 
 struct siginfo;
 extern void sigio_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs);

arch/um/include/shared/net_kern.h

@@ -65,5 +65,7 @@ extern int tap_setup_common(char *str, char *type, char **dev_name,
 			    char **mac_out, char **gate_addr);
 extern void register_transport(struct transport *new);
 extern unsigned short eth_protocol(struct sk_buff *skb);
+extern void uml_net_setup_etheraddr(struct net_device *dev, char *str);
+
 
 #endif

arch/um/include/shared/os.h

@@ -290,15 +290,16 @@ extern void halt_skas(void);
 extern void reboot_skas(void);
 
 /* irq.c */
-extern int os_waiting_for_events(struct irq_fd *active_fds);
-extern int os_create_pollfd(int fd, int events, void *tmp_pfd, int size_tmpfds);
-extern void os_free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg,
-		struct irq_fd *active_fds, struct irq_fd ***last_irq_ptr2);
-extern void os_free_irq_later(struct irq_fd *active_fds,
-		int irq, void *dev_id);
-extern int os_get_pollfd(int i);
-extern void os_set_pollfd(int i, int fd);
+extern int os_waiting_for_events_epoll(void);
+extern void *os_epoll_get_data_pointer(int index);
+extern int os_epoll_triggered(int index, int events);
+extern int os_event_mask(int irq_type);
+extern int os_setup_epoll(void);
+extern int os_add_epoll_fd(int events, int fd, void *data);
+extern int os_mod_epoll_fd(int events, int fd, void *data);
+extern int os_del_epoll_fd(int fd);
 extern void os_set_ioignore(void);
+extern void os_close_epoll_fd(void);
 
 /* sigio.c */
 extern int add_sigio_fd(int fd);

arch/um/kernel/time.c

@@ -121,12 +121,12 @@ static void __init um_timer_setup(void)
 	clockevents_register_device(&timer_clockevent);
 }
 
-void read_persistent_clock(struct timespec *ts)
+void read_persistent_clock64(struct timespec64 *ts)
 {
 	long long nsecs = os_persistent_clock_emulation();
 
-	set_normalized_timespec(ts, nsecs / NSEC_PER_SEC,
-				nsecs % NSEC_PER_SEC);
+	set_normalized_timespec64(ts, nsecs / NSEC_PER_SEC,
+				  nsecs % NSEC_PER_SEC);
 }
 
 void __init time_init(void)

arch/um/os-Linux/file.c

@@ -12,6 +12,7 @@
 #include <sys/mount.h>
 #include <sys/socket.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include <sys/un.h>
 #include <sys/types.h>
 #include <os.h>

arch/um/os-Linux/irq.c

 /*
+ * Copyright (C) 2017 - Cambridge Greys Ltd
+ * Copyright (C) 2011 - 2014 Cisco Systems Inc
  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
  * Licensed under the GPL
  */
 
 #include <stdlib.h>
 #include <errno.h>
-#include <poll.h>
+#include <sys/epoll.h>
 #include <signal.h>
 #include <string.h>
 #include <irq_user.h>
 #include <os.h>
 #include <um_malloc.h>
 
+/* Epoll support */
+
+static int epollfd = -1;
+
+#define MAX_EPOLL_EVENTS 64
+
+static struct epoll_event epoll_events[MAX_EPOLL_EVENTS];
+
+/* Helper to return an Epoll data pointer from an epoll event structure.
+ * We need to keep this one on the userspace side to keep includes separate
+ */
+
+void *os_epoll_get_data_pointer(int index)
+{
+	return epoll_events[index].data.ptr;
+}
+
+/* Helper to compare events versus the events in the epoll structure.
+ * Same as above - needs to be on the userspace side
+ */
+
+
+int os_epoll_triggered(int index, int events)
+{
+	return epoll_events[index].events & events;
+}
+/* Helper to set the event mask.
+ * The event mask is opaque to the kernel side, because it does not have
+ * access to the right includes/defines for EPOLL constants.
+ */
+
+int os_event_mask(int irq_type)
+{
+	if (irq_type == IRQ_READ)
+		return EPOLLIN | EPOLLPRI;
+	if (irq_type == IRQ_WRITE)
+		return EPOLLOUT;
+	return 0;
+}
+
 /*
- * Locked by irq_lock in arch/um/kernel/irq.c.  Changed by os_create_pollfd
- * and os_free_irq_by_cb, which are called under irq_lock.
+ * Initial Epoll Setup
  */
-static struct pollfd *pollfds = NULL;
-static int pollfds_num = 0;
-static int pollfds_size = 0;
+int os_setup_epoll(void)
+{
+	epollfd = epoll_create(MAX_EPOLL_EVENTS);
+	return epollfd;
+}
 
-int os_waiting_for_events(struct irq_fd *active_fds)
+/*
+ * Helper to run the actual epoll_wait
+ */
+int os_waiting_for_events_epoll(void)
 {
-	struct irq_fd *irq_fd;
-	int i, n, err;
+	int n, err;
 
-	n = poll(pollfds, pollfds_num, 0);
+	n = epoll_wait(epollfd,
+		(struct epoll_event *) &epoll_events, MAX_EPOLL_EVENTS, 0);
 	if (n < 0) {
 		err = -errno;
 		if (errno != EINTR)
-			printk(UM_KERN_ERR "os_waiting_for_events:"
-			       " poll returned %d, errno = %d\n", n, errno);
+			printk(
+				UM_KERN_ERR "os_waiting_for_events:"
+				" epoll returned %d, error = %s\n", n,
+				strerror(errno)
+			);
 		return err;
 	}
-
-	if (n == 0)
-		return 0;
-
-	irq_fd = active_fds;
-
-	for (i = 0; i < pollfds_num; i++) {
-		if (pollfds[i].revents != 0) {
-			irq_fd->current_events = pollfds[i].revents;
-			pollfds[i].fd = -1;
-		}
-		irq_fd = irq_fd->next;
-	}
 	return n;
 }
 
-int os_create_pollfd(int fd, int events, void *tmp_pfd, int size_tmpfds)
-{
-	if (pollfds_num == pollfds_size) {
-		if (size_tmpfds <= pollfds_size * sizeof(pollfds[0])) {
-			/* return min size needed for new pollfds area */
-			return (pollfds_size + 1) * sizeof(pollfds[0]);
-		}
-
-		if (pollfds != NULL) {
-			memcpy(tmp_pfd, pollfds,
-			       sizeof(pollfds[0]) * pollfds_size);
-			/* remove old pollfds */
-			kfree(pollfds);
-		}
-		pollfds = tmp_pfd;
-		pollfds_size++;
-	} else
-		kfree(tmp_pfd);	/* remove not used tmp_pfd */
-
-	pollfds[pollfds_num] = ((struct pollfd) { .fd		= fd,
-						  .events	= events,
-						  .revents	= 0 });
-	pollfds_num++;
-
-	return 0;
-}
 
-void os_free_irq_by_cb(int (*test)(struct irq_fd *, void *), void *arg,
-		struct irq_fd *active_fds, struct irq_fd ***last_irq_ptr2)
+/*
+ * Helper to add a fd to epoll
+ */
+int os_add_epoll_fd(int events, int fd, void *data)
 {
-	struct irq_fd **prev;
-	int i = 0;
-
-	prev = &active_fds;
-	while (*prev != NULL) {
-		if ((*test)(*prev, arg)) {
-			struct irq_fd *old_fd = *prev;
-			if ((pollfds[i].fd != -1) &&
-			    (pollfds[i].fd != (*prev)->fd)) {
-				printk(UM_KERN_ERR "os_free_irq_by_cb - "
-				       "mismatch between active_fds and "
-				       "pollfds, fd %d vs %d\n",
-				       (*prev)->fd, pollfds[i].fd);
-				goto out;
-			}
-
-			pollfds_num--;
-
-			/*
-			 * This moves the *whole* array after pollfds[i]
-			 * (though it doesn't spot as such)!
-			 */
-			memmove(&pollfds[i], &pollfds[i + 1],
-			       (pollfds_num - i) * sizeof(pollfds[0]));
-			if (*last_irq_ptr2 == &old_fd->next)
-				*last_irq_ptr2 = prev;
-
-			*prev = (*prev)->next;
-			if (old_fd->type == IRQ_WRITE)
-				ignore_sigio_fd(old_fd->fd);
-			kfree(old_fd);
-			continue;
-		}
-		prev = &(*prev)->next;
-		i++;
-	}
- out:
-	return;
+	struct epoll_event event;
+	int result;
+
+	event.data.ptr = data;
+	event.events = events | EPOLLET;
+	result = epoll_ctl(epollfd, EPOLL_CTL_ADD, fd, &event);
+	if ((result) && (errno == EEXIST))
+		result = os_mod_epoll_fd(events, fd, data);
+	if (result)
+		printk("epollctl add err fd %d, %s\n", fd, strerror(errno));
+	return result;
 }
 
-int os_get_pollfd(int i)
+/*
+ * Helper to mod the fd event mask and/or data backreference
+ */
+int os_mod_epoll_fd(int events, int fd, void *data)
 {
-	return pollfds[i].fd;
+	struct epoll_event event;
+	int result;
+
+	event.data.ptr = data;
+	event.events = events;
+	result = epoll_ctl(epollfd, EPOLL_CTL_MOD, fd, &event);
+	if (result)
+		printk(UM_KERN_ERR
+			"epollctl mod err fd %d, %s\n", fd, strerror(errno));
+	return result;
 }
 
-void os_set_pollfd(int i, int fd)
+/*
+ * Helper to delete the epoll fd
+ */
+int os_del_epoll_fd(int fd)
 {
-	pollfds[i].fd = fd;
+	struct epoll_event event;
+	int result;
+	/* This is quiet as we use this as IO ON/OFF - so it is often
+	 * invoked on a non-existent fd
+	 */
+	result = epoll_ctl(epollfd, EPOLL_CTL_DEL, fd, &event);
+	return result;
 }
 
 void os_set_ioignore(void)
 {
 	signal(SIGIO, SIG_IGN);
 }
+
+void os_close_epoll_fd(void)
+{
+	/* Needed so we do not leak an fd when rebooting */
+	os_close_file(epollfd);
+}

arch/um/os-Linux/signal.c

@@ -16,6 +16,7 @@
 #include <os.h>
 #include <sysdep/mcontext.h>
 #include <um_malloc.h>
+#include <sys/ucontext.h>
 
 void (*sig_info[NSIG])(int, struct siginfo *, struct uml_pt_regs *) = {
 	[SIGTRAP]	= relay_signal,
@@ -159,7 +160,7 @@ static void (*handlers[_NSIG])(int sig, struct siginfo *si, mcontext_t *mc) = {
 
 static void hard_handler(int sig, siginfo_t *si, void *p)
 {
-	struct ucontext *uc = p;
+	ucontext_t *uc = p;
 	mcontext_t *mc = &uc->uc_mcontext;
 	unsigned long pending = 1UL << sig;
 

arch/x86/um/stub_segv.c

@@ -6,11 +6,12 @@
 #include <sysdep/stub.h>
 #include <sysdep/faultinfo.h>
 #include <sysdep/mcontext.h>
+#include <sys/ucontext.h>
 
 void __attribute__ ((__section__ (".__syscall_stub")))
 stub_segv_handler(int sig, siginfo_t *info, void *p)
 {
-	struct ucontext *uc = p;
+	ucontext_t *uc = p;
 
 	GET_FAULTINFO_FROM_MC(*((struct faultinfo *) STUB_DATA),
 			      &uc->uc_mcontext);