19 files changed, 907 insertions, 0 deletions

diff --git a/usr/src/cmd/devfsadm/misc_link.c b/usr/src/cmd/devfsadm/misc_link.c
index 70599d6039..9f4c20ac5e 100644
--- a/usr/src/cmd/devfsadm/misc_link.c
+++ b/usr/src/cmd/devfsadm/misc_link.c
@@ -89,6 +89,9 @@ static devfsadm_create_t misc_cbt[] = {
 	{ "pseudo", "ddi_pseudo", "consms",
 	    TYPE_EXACT | DRV_EXACT, ILEVEL_0, consms
 	},
+	{ "pseudo", "ddi_pseudo", "eventfd",
+	    TYPE_EXACT | DRV_EXACT, ILEVEL_0, minor_name
+	},
 	{ "pseudo", "ddi_pseudo", "rsm",
 	    TYPE_EXACT | DRV_EXACT, ILEVEL_0, minor_name
 	},
diff --git a/usr/src/lib/libc/amd64/Makefile b/usr/src/lib/libc/amd64/Makefile
index 9a496b0d7e..12fe1429fb 100644
--- a/usr/src/lib/libc/amd64/Makefile
+++ b/usr/src/lib/libc/amd64/Makefile
@@ -858,6 +858,7 @@ PORTSYS=			\
 	execl.o			\
 	execle.o		\
 	execv.o			\
+	eventfd.o		\
 	fcntl.o			\
 	getpagesizes.o		\
 	getpeerucred.o		\
diff --git a/usr/src/lib/libc/i386/Makefile.com b/usr/src/lib/libc/i386/Makefile.com
index 01f71c334c..b6a9c822c6 100644
--- a/usr/src/lib/libc/i386/Makefile.com
+++ b/usr/src/lib/libc/i386/Makefile.com
@@ -892,6 +892,7 @@ PORTSYS=			\
 	chown.o			\
 	corectl.o		\
 	epoll.o			\
+	eventfd.o		\
 	exacctsys.o		\
 	execl.o			\
 	execle.o		\
diff --git a/usr/src/lib/libc/port/mapfile-vers b/usr/src/lib/libc/port/mapfile-vers
index 70d37d04be..09db2c67e4 100644
--- a/usr/src/lib/libc/port/mapfile-vers
+++ b/usr/src/lib/libc/port/mapfile-vers
@@ -2815,6 +2815,9 @@ $endif
 	epoll_pwait;
 	_errfp;
 	_errxfp;
+	eventfd;
+	eventfd_read;
+	eventfd_write;
 	exportfs;
 	_F_cplx_div;
 	_F_cplx_div_ix;
diff --git a/usr/src/lib/libc/port/sys/eventfd.c b/usr/src/lib/libc/port/sys/eventfd.c
new file mode 100644
index 0000000000..bd2eba4efa
--- /dev/null
+++ b/usr/src/lib/libc/port/sys/eventfd.c
@@ -0,0 +1,65 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014, Joyent, Inc.  All rights reserved.
+ */
+
+#include <sys/eventfd.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <errno.h>
+#include <fcntl.h>
+
+int
+eventfd(unsigned int initval, int flags)
+{
+	int oflags = O_RDWR;
+	uint64_t val = initval;
+	int fd;
+
+	if (flags & ~(EFD_NONBLOCK | EFD_CLOEXEC | EFD_SEMAPHORE))
+		return (EINVAL);
+
+	if (flags & EFD_NONBLOCK)
+		oflags |= O_NONBLOCK;
+
+	if (flags & EFD_CLOEXEC)
+		oflags |= O_CLOEXEC;
+
+	if ((fd = open("/dev/eventfd", oflags)) < 0)
+		return (-1);
+
+	if ((flags & EFD_SEMAPHORE) &&
+	    ioctl(fd, EVENTFDIOC_SEMAPHORE, 0) != 0) {
+		(void) close(fd);
+		return (-1);
+	}
+
+	if (write(fd, &val, sizeof (val)) < sizeof (val)) {
+		(void) close(fd);
+		return (-1);
+	}
+
+	return (fd);
+}
+
+int
+eventfd_read(int fd, eventfd_t *valp)
+{
+	return (read(fd, valp, sizeof (*valp)) < sizeof (*valp) ? -1 : 0);
+}
+
+int
+eventfd_write(int fd, eventfd_t val)
+{
+	return (write(fd, &val, sizeof (val)) < sizeof (val) ? -1 : 0);
+}
diff --git a/usr/src/lib/libc/sparc/Makefile.com b/usr/src/lib/libc/sparc/Makefile.com
index 4c5bbcfc96..11ad0b3996 100644
--- a/usr/src/lib/libc/sparc/Makefile.com
+++ b/usr/src/lib/libc/sparc/Makefile.com
@@ -926,6 +926,7 @@ PORTSYS=			\
 	chown.o			\
 	corectl.o		\
 	epoll.o			\
+	eventfd.o		\
 	exacctsys.o		\
 	execl.o			\
 	execle.o		\
diff --git a/usr/src/man/man3c/Makefile b/usr/src/man/man3c/Makefile
index bb75c1d4b6..55fd0615df 100644
--- a/usr/src/man/man3c/Makefile
+++ b/usr/src/man/man3c/Makefile
@@ -113,6 +113,7 @@ MANFILES= 	__fbufsize.3c					\
 	 	epoll_wait.3c					\
 	 	err.3c						\
 	 	euclen.3c					\
+		eventfd.3c					\
 	 	exit.3c						\
 	 	fattach.3c					\
 	 	fclose.3c					\
diff --git a/usr/src/man/man3c/eventfd.3c b/usr/src/man/man3c/eventfd.3c
new file mode 100644
index 0000000000..6e7d1c6a5b
--- /dev/null
+++ b/usr/src/man/man3c/eventfd.3c
@@ -0,0 +1,161 @@
+'\" te
+.\"  Copyright (c) 2014, Joyent, Inc. All Rights Reserved.
+.\"  This file and its contents are supplied under the terms of the
+.\"  Common Development and Distribution License ("CDDL"), version 1.0.
+.\"  You may only use this file in accordance with the terms of version
+.\"  1.0 of the CDDL.
+.\" 
+.\"  A full copy of the text of the CDDL should have accompanied this
+.\"  source.  A copy of the CDDL is also available via the Internet at
+.\"  http://www.illumos.org/license/CDDL.
+.TH EVENTFD 3C "Dec 3, 2014"
+.SH NAME
+eventfd \- create a file descriptor for event notification
+.SH SYNOPSIS
+
+.LP
+.nf
+#include <sys/eventfd.h>
+
+\fBint\fR \fBeventfd\fR(\fBunsigned int\fR \fIinitval\fR, \fBint\fR \fIflags\fR);
+.fi
+
+.SH DESCRIPTION
+.sp
+.LP
+The \fBeventfd()\fR function creates an \fBeventfd\fR(5) instance that
+has an associated 64-bit unsigned counter. It returns a file descriptor
+that can be operated upon via \fBread\fR(2), \fBwrite\fR(2) and the facilities
+that notify of file descriptor activity (e.g., \fBpoll\fR(2),
+\fBport_get\fR(3C), \fBepoll_wait\fR(3C)). To dispose of the instance,
+\fBclose\fR(2) should be called on the file descriptor.
+
+The \fIinitval\fR argument specifies the
+initial value of the 64-bit counter associated with the instance.  (Note
+that this limits the initial value to be a 32-bit quantity despite the
+fact that the underlying counter is 64-bit.)
+
+The \fIflags\fR argument specifies additional parameters for the
+instance, and can have any of the following values:
+
+.sp
+.ne 2
+.na
+\fBEFD_CLOEXEC\fR
+.ad
+.RS 12n
+Instance will be closed upon an
+\fBexec\fR(2); see \fBopen\fR(2)'s description of \fBO_CLOEXEC\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEFD_NONBLOCK\fR
+.ad
+.RS 12n
+Instance will be set to be non-blocking.  A \fBread\fR(2) on an
+\fBeventfd\fR instance that has been initialized with
+\fBEFD_NONBLOCK\fR will return \fBEAGAIN\fR in lieu of blocking if the
+count associated with the instance is zero.
+.RE
+
+.sp
+.ne 2
+.na
+\fBEFD_SEMAPHORE\fR
+.ad
+.RS 12n
+Provide counting semaphore semantics whereby a \fBread\fR(2) will 
+atomically decrement rather than atomically clear the count when it
+becomes non-zero. See below for details on \fBread\fR(2) semantics.
+.RE
+
+.sp
+The following operations can be performed upon an \fBeventfd\fR instance:
+
+.sp
+.ne 2
+.na
+\fBread\fR(2)
+.ad
+.RS 12n
+Atomically reads and modifies the value of the 64-bit counter associated
+with the instance.  The precise semantics
+of \fBread\fR(2) depend on the disposition of \fBEFD_SEMAPHORE\fR with
+respect to the instance: if \fBEFD_SEMAPHORE\fR was set when the instance
+was created, \fBread\fR(2) will \fIatomically decrement\fR the counter if
+(and when)
+it is non-zero, copying the value 1 to the eight byte buffer passed to
+the system call; if \fBEFD_SEMAPHORE\fR was not set, \fBread\fR(2) will
+\fIatomically clear\fR the counter if (and when) it is non-zero, copying
+the former value of the counter to the eight byte buffer passed to the
+system call.  In either case, \fBread\fR(2) will block if the counter is
+zero (or return \fBEAGAIN\fR if the instance was created with
+\fBEFD_NONBLOCK\fR). If the buffer specified to \fBread\fR(2) is less than
+eight bytes in length, \fBEINVAL\fR will be returned.
+
+.RE
+
+.sp
+.ne 2
+.na
+\fBwrite\fR(2)
+.ad
+.RS 12n
+Atomically adds the 64-bit value pointed to by the buffer to the 64-bit
+counter associated with the instance.  If the resulting value would overflow,
+the \fBwrite\fR(2) will block until the value would not overflow
+(or return \fBEAGAIN\fR if the instance was created with
+\fBEFD_NONBLOCK\fR).  If the buffer specified to \fBwrite\fR(2) is less than
+eight bytes in length, \fBEINVAL\fR will be returned.
+
+.RE
+
+.sp
+.ne 2
+.na
+\fBpoll\fR(2), \fBport_get\fR(3C), \fBepoll_wait\fR(3C)
+.ad
+.RS 12n
+Provide notification when the 64-bit counter associated
+with the instance is ready for reading or writing, as specified.
+If the 64-bit value associated with the instance is non-zero, \fBPOLLIN\fR
+and \fBPOLLRDNORM\fR will be set; if the value 1 can be added the value
+without blocking, \fBPOLLOUT\fR and \fBPOLLWRNORM\fR will be set.
+.RE
+
+.SH RETURN VALUES
+.sp
+.LP
+Upon succesful completion, a file descriptor associated with the instance
+is returned. Otherwise, -1 is returned and errno
+is set to indicate the error.
+.SH ERRORS
+.sp
+.LP
+The \fBeventfd()\fR function will fail if:
+.sp
+.ne 2
+.na
+\fB\fBEINVAL\fR\fR
+.ad
+.RS 10n
+The \fIflags\fR are invalid.
+.RE
+
+.sp
+.ne 2
+.na
+\fB\fBEMFILE\fR\fR
+.ad
+.RS 10n
+There are currently {\fBOPEN_MAX\fR} file descriptors open in the calling
+process.
+.RE
+
+.SH SEE ALSO
+.sp
+.LP
+\fBpoll\fR(2), \fBport_get\fR(3C), \fBepoll_wait\fR(3C), \fBeventfd\fR(5)
+
diff --git a/usr/src/man/man5/Makefile b/usr/src/man/man5/Makefile
index 466d5c0f16..e73c19a10a 100644
--- a/usr/src/man/man5/Makefile
+++ b/usr/src/man/man5/Makefile
@@ -45,6 +45,7 @@ MANFILES=	Intro.5			\
 		epoll.5			\
 		eqn.5			\
 		eqnchar.5		\
+		eventfd.5		\
 		extendedFILE.5		\
 		filesystem.5		\
 		fnmatch.5		\
diff --git a/usr/src/man/man5/eventfd.5 b/usr/src/man/man5/eventfd.5
new file mode 100644
index 0000000000..fc30495070
--- /dev/null
+++ b/usr/src/man/man5/eventfd.5
@@ -0,0 +1,34 @@
+'\" te
+.\"  Copyright (c) 2014, Joyent, Inc. All Rights Reserved.
+.\"  This file and its contents are supplied under the terms of the
+.\"  Common Development and Distribution License ("CDDL"), version 1.0.
+.\"  You may only use this file in accordance with the terms of version
+.\"  1.0 of the CDDL.
+.\" 
+.\"  A full copy of the text of the CDDL should have accompanied this
+.\"  source.  A copy of the CDDL is also available via the Internet at
+.\"  http://www.illumos.org/license/CDDL.
+.TH EVENTFD 5 "Dec 3, 2014"
+.SH NAME
+eventfd \- Linux-compatible user event notification facility
+.SH SYNOPSIS
+
+.LP
+.nf
+#include <sys/eventfd.h>
+.fi
+
+.SH DESCRIPTION
+.sp
+.LP
+
+\fBeventfd\fR is a Linux-borne facility for sending and receiving user
+events via a file descriptor.  While the facility itself is somewhat dubious
+(it can be mimicked in an entirely portable way with a pipe), it is
+small and straightforward and this implementation is entirely compatible
+with its Linux antecedent; see \fBeventfd\fR(3C) for details.
+
+.SH SEE ALSO
+.sp
+.LP
+\fBeventfd\fR(3C)
diff --git a/usr/src/uts/common/Makefile.files b/usr/src/uts/common/Makefile.files
index 1f3166fa85..517a7a6363 100644
--- a/usr/src/uts/common/Makefile.files
+++ b/usr/src/uts/common/Makefile.files
@@ -1005,6 +1005,8 @@ DEVPOLL_OBJS += devpoll.o
 
 DEVPOOL_OBJS += devpool.o
 
+EVENTFD_OBJS +=	eventfd.o
+
 I8042_OBJS +=	i8042.o
 
 INOTIFY_OBJS +=	inotify.o
diff --git a/usr/src/uts/common/io/eventfd.c b/usr/src/uts/common/io/eventfd.c
new file mode 100644
index 0000000000..bd7b6fed5e
--- /dev/null
+++ b/usr/src/uts/common/io/eventfd.c
@@ -0,0 +1,411 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Support for the eventfd facility, a Linux-borne facility for user-generated
+ * file descriptor-based events.
+ */
+
+#include <sys/ddi.h>
+#include <sys/sunddi.h>
+#include <sys/eventfd.h>
+#include <sys/conf.h>
+#include <sys/vmem.h>
+#include <sys/sysmacros.h>
+#include <sys/filio.h>
+#include <sys/stat.h>
+#include <sys/file.h>
+
+struct eventfd_state;
+typedef struct eventfd_state eventfd_state_t;
+
+struct eventfd_state {
+	kmutex_t efd_lock;			/* lock protecting state */
+	boolean_t efd_semaphore;		/* boolean: sema. semantics */
+	kcondvar_t efd_cv;			/* condvar */
+	pollhead_t efd_pollhd;			/* poll head */
+	uint64_t efd_value;			/* value */
+	eventfd_state_t *efd_next;		/* next state on global list */
+};
+
+/*
+ * Internal global variables.
+ */
+static kmutex_t		eventfd_lock;		/* lock protecting state */
+static dev_info_t	*eventfd_devi;		/* device info */
+static vmem_t		*eventfd_minor;		/* minor number arena */
+static void		*eventfd_softstate;	/* softstate pointer */
+static eventfd_state_t	*eventfd_state;		/* global list of state */
+
+/*ARGSUSED*/
+static int
+eventfd_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
+{
+	eventfd_state_t *state;
+	major_t major = getemajor(*devp);
+	minor_t minor = getminor(*devp);
+
+	if (minor != EVENTFDMNRN_INOTIFY)
+		return (ENXIO);
+
+	mutex_enter(&eventfd_lock);
+
+	minor = (minor_t)(uintptr_t)vmem_alloc(eventfd_minor, 1,
+	    VM_BESTFIT | VM_SLEEP);
+
+	if (ddi_soft_state_zalloc(eventfd_softstate, minor) != DDI_SUCCESS) {
+		vmem_free(eventfd_minor, (void *)(uintptr_t)minor, 1);
+		mutex_exit(&eventfd_lock);
+		return (NULL);
+	}
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+	*devp = makedevice(major, minor);
+
+	state->efd_next = eventfd_state;
+	eventfd_state = state;
+
+	mutex_exit(&eventfd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_read(dev_t dev, uio_t *uio, cred_t *cr)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+	uint64_t val, oval;
+	int err;
+
+	if (uio->uio_resid < sizeof (val))
+		return (EINVAL);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	mutex_enter(&state->efd_lock);
+
+	while (state->efd_value == 0) {
+		if (uio->uio_fmode & (FNDELAY|FNONBLOCK)) {
+			mutex_exit(&state->efd_lock);
+			return (EAGAIN);
+		}
+
+		if (!cv_wait_sig_swap(&state->efd_cv, &state->efd_lock)) {
+			mutex_exit(&state->efd_lock);
+			return (EINTR);
+		}
+	}
+
+	/*
+	 * We have a non-zero value and we own the lock; our behavior now
+	 * depends on whether or not EFD_SEMAPHORE was set when the eventfd
+	 * was created.
+	 */
+	val = oval = state->efd_value;
+
+	if (state->efd_semaphore) {
+		state->efd_value--;
+		val = 1;
+	} else {
+		state->efd_value = 0;
+	}
+
+	err = uiomove(&val, sizeof (val), UIO_READ, uio);
+
+	mutex_exit(&state->efd_lock);
+
+	if (oval == EVENTFD_VALMAX) {
+		cv_broadcast(&state->efd_cv);
+		pollwakeup(&state->efd_pollhd, POLLWRNORM | POLLOUT);
+	}
+
+	return (err);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_write(dev_t dev, struct uio *uio, cred_t *credp)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+	uint64_t val, oval;
+	int err;
+
+	if (uio->uio_resid < sizeof (val))
+		return (EINVAL);
+
+	if ((err = uiomove(&val, sizeof (val), UIO_WRITE, uio)) != 0)
+		return (err);
+
+	if (val > EVENTFD_VALMAX)
+		return (EINVAL);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	mutex_enter(&state->efd_lock);
+
+	while (val > EVENTFD_VALMAX - state->efd_value) {
+		if (uio->uio_fmode & (FNDELAY|FNONBLOCK)) {
+			mutex_exit(&state->efd_lock);
+			return (EAGAIN);
+		}
+
+		if (!cv_wait_sig_swap(&state->efd_cv, &state->efd_lock)) {
+			mutex_exit(&state->efd_lock);
+			return (EINTR);
+		}
+	}
+
+	/*
+	 * We now know that we can add the value without overflowing.
+	 */
+	state->efd_value = (oval = state->efd_value) + val;
+
+	mutex_exit(&state->efd_lock);
+
+	if (oval == 0) {
+		cv_broadcast(&state->efd_cv);
+		pollwakeup(&state->efd_pollhd, POLLRDNORM | POLLIN);
+	}
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_poll(dev_t dev, short events, int anyyet, short *reventsp,
+    struct pollhead **phpp)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+	short revents = 0;
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	mutex_enter(&state->efd_lock);
+
+	if (state->efd_value > 0)
+		revents |= POLLRDNORM | POLLIN;
+
+	if (state->efd_value < EVENTFD_VALMAX)
+		revents |= POLLWRNORM | POLLOUT;
+
+	*reventsp = revents & events;
+
+	if (!revents && !anyyet)
+		*phpp = &state->efd_pollhd;
+
+	mutex_exit(&state->efd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv)
+{
+	eventfd_state_t *state;
+	minor_t minor = getminor(dev);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	switch (cmd) {
+	case EVENTFDIOC_SEMAPHORE: {
+		mutex_enter(&state->efd_lock);
+		state->efd_semaphore ^= 1;
+		mutex_exit(&state->efd_lock);
+
+		return (0);
+	}
+
+	default:
+		break;
+	}
+
+	return (ENOTTY);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_close(dev_t dev, int flag, int otyp, cred_t *cred_p)
+{
+	eventfd_state_t *state, **sp;
+	minor_t minor = getminor(dev);
+
+	state = ddi_get_soft_state(eventfd_softstate, minor);
+
+	mutex_enter(&eventfd_lock);
+
+	/*
+	 * Remove our state from our global list.
+	 */
+	for (sp = &eventfd_state; *sp != state; sp = &((*sp)->efd_next))
+		VERIFY(*sp != NULL);
+
+	*sp = (*sp)->efd_next;
+
+	ddi_soft_state_free(eventfd_softstate, minor);
+	vmem_free(eventfd_minor, (void *)(uintptr_t)minor, 1);
+
+	mutex_exit(&eventfd_lock);
+
+	return (0);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
+{
+	mutex_enter(&eventfd_lock);
+
+	if (ddi_soft_state_init(&eventfd_softstate,
+	    sizeof (eventfd_state_t), 0) != 0) {
+		cmn_err(CE_NOTE, "/dev/eventfd failed to create soft state");
+		mutex_exit(&eventfd_lock);
+		return (DDI_FAILURE);
+	}
+
+	if (ddi_create_minor_node(devi, "eventfd", S_IFCHR,
+	    EVENTFDMNRN_INOTIFY, DDI_PSEUDO, NULL) == DDI_FAILURE) {
+		cmn_err(CE_NOTE, "/dev/eventfd couldn't create minor node");
+		ddi_soft_state_fini(&eventfd_softstate);
+		mutex_exit(&eventfd_lock);
+		return (DDI_FAILURE);
+	}
+
+	ddi_report_dev(devi);
+	eventfd_devi = devi;
+
+	eventfd_minor = vmem_create("eventfd_minor", (void *)EVENTFDMNRN_CLONE,
+	    UINT32_MAX - EVENTFDMNRN_CLONE, 1, NULL, NULL, NULL, 0,
+	    VM_SLEEP | VMC_IDENTIFIER);
+
+	mutex_exit(&eventfd_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
+{
+	switch (cmd) {
+	case DDI_DETACH:
+		break;
+
+	case DDI_SUSPEND:
+		return (DDI_SUCCESS);
+
+	default:
+		return (DDI_FAILURE);
+	}
+
+	mutex_enter(&eventfd_lock);
+	vmem_destroy(eventfd_minor);
+
+	ddi_remove_minor_node(eventfd_devi, NULL);
+	eventfd_devi = NULL;
+
+	ddi_soft_state_fini(&eventfd_softstate);
+	mutex_exit(&eventfd_lock);
+
+	return (DDI_SUCCESS);
+}
+
+/*ARGSUSED*/
+static int
+eventfd_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
+{
+	int error;
+
+	switch (infocmd) {
+	case DDI_INFO_DEVT2DEVINFO:
+		*result = (void *)eventfd_devi;
+		error = DDI_SUCCESS;
+		break;
+	case DDI_INFO_DEVT2INSTANCE:
+		*result = (void *)0;
+		error = DDI_SUCCESS;
+		break;
+	default:
+		error = DDI_FAILURE;
+	}
+	return (error);
+}
+
+static struct cb_ops eventfd_cb_ops = {
+	eventfd_open,		/* open */
+	eventfd_close,		/* close */
+	nulldev,		/* strategy */
+	nulldev,		/* print */
+	nodev,			/* dump */
+	eventfd_read,		/* read */
+	eventfd_write,		/* write */
+	eventfd_ioctl,		/* ioctl */
+	nodev,			/* devmap */
+	nodev,			/* mmap */
+	nodev,			/* segmap */
+	eventfd_poll,		/* poll */
+	ddi_prop_op,		/* cb_prop_op */
+	0,			/* streamtab  */
+	D_NEW | D_MP		/* Driver compatibility flag */
+};
+
+static struct dev_ops eventfd_ops = {
+	DEVO_REV,		/* devo_rev */
+	0,			/* refcnt */
+	eventfd_info,		/* get_dev_info */
+	nulldev,		/* identify */
+	nulldev,		/* probe */
+	eventfd_attach,		/* attach */
+	eventfd_detach,		/* detach */
+	nodev,			/* reset */
+	&eventfd_cb_ops,	/* driver operations */
+	NULL,			/* bus operations */
+	nodev,			/* dev power */
+	ddi_quiesce_not_needed,	/* quiesce */
+};
+
+static struct modldrv modldrv = {
+	&mod_driverops,		/* module type (this is a pseudo driver) */
+	"eventfd support",	/* name of module */
+	&eventfd_ops,		/* driver ops */
+};
+
+static struct modlinkage modlinkage = {
+	MODREV_1,
+	(void *)&modldrv,
+	NULL
+};
+
+int
+_init(void)
+{
+	return (mod_install(&modlinkage));
+}
+
+int
+_info(struct modinfo *modinfop)
+{
+	return (mod_info(&modlinkage, modinfop));
+}
+
+int
+_fini(void)
+{
+	return (mod_remove(&modlinkage));
+}
diff --git a/usr/src/uts/common/io/eventfd.conf b/usr/src/uts/common/io/eventfd.conf
new file mode 100644
index 0000000000..f9c6dc11b2
--- /dev/null
+++ b/usr/src/uts/common/io/eventfd.conf
@@ -0,0 +1,16 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+#
+
+name="eventfd" parent="pseudo" instance=0;
diff --git a/usr/src/uts/common/sys/Makefile b/usr/src/uts/common/sys/Makefile
index ce449efd20..b52be94c90 100644
--- a/usr/src/uts/common/sys/Makefile
+++ b/usr/src/uts/common/sys/Makefile
@@ -221,6 +221,7 @@ CHKHDRS=			\
 	ethernet.h		\
 	euc.h			\
 	eucioctl.h		\
+	eventfd.h		\
 	exacct.h		\
 	exacct_catalog.h	\
 	exacct_impl.h		\
diff --git a/usr/src/uts/common/sys/eventfd.h b/usr/src/uts/common/sys/eventfd.h
new file mode 100644
index 0000000000..c7dbaf947f
--- /dev/null
+++ b/usr/src/uts/common/sys/eventfd.h
@@ -0,0 +1,68 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+ */
+
+/*
+ * Header file to support for the eventfd facility.  Note that this facility
+ * is designed to be binary compatible with the Linux eventfd facility; values
+ * for constants here should therefore exactly match those found in Linux, and
+ * this facility shouldn't be extended independently of Linux.
+ */
+
+#ifndef _SYS_EVENTFD_H
+#define	_SYS_EVENTFD_H
+
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+typedef uint64_t eventfd_t;
+
+/*
+ * To assure binary compatibility with Linux, these values are fixed at their
+ * Linux equivalents, not their native ones.
+ */
+#define	EFD_CLOEXEC		02000000		/* LX_O_CLOEXEC */
+#define	EFD_NONBLOCK		04000			/* LX_O_NONBLOCK */
+#define	EFD_SEMAPHORE		1
+
+/*
+ * These ioctl values are specific to the native implementation; applications
+ * shouldn't be using them directly, and they should therefore be safe to
+ * change without breaking apps.
+ */
+#define	EVENTFDIOC		(('e' << 24) | ('f' << 16) | ('d' << 8))
+#define	EVENTFDIOC_SEMAPHORE	(EVENTFDIOC | 1)	/* toggle sem state */
+
+#ifndef _KERNEL
+
+extern int eventfd(unsigned int, int);
+extern int eventfd_read(int, eventfd_t *);
+extern int eventfd_write(int, eventfd_t);
+
+#else
+
+#define	EVENTFDMNRN_INOTIFY	0
+#define	EVENTFDMNRN_CLONE	1
+#define	EVENTFD_VALMAX		(ULLONG_MAX - 1ULL)
+
+#endif /* _KERNEL */
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_EVENTFD_H */
diff --git a/usr/src/uts/intel/Makefile.intel b/usr/src/uts/intel/Makefile.intel
index 4c0194240b..49ff0b8a2e 100644
--- a/usr/src/uts/intel/Makefile.intel
+++ b/usr/src/uts/intel/Makefile.intel
@@ -242,6 +242,7 @@ DRV_KMODS	+= dr_sas
 DRV_KMODS	+= dump
 DRV_KMODS	+= ecpp
 DRV_KMODS	+= emlxs
+DRV_KMODS	+= eventfd
 DRV_KMODS	+= fd
 DRV_KMODS	+= fdc
 DRV_KMODS	+= fm
diff --git a/usr/src/uts/intel/eventfd/Makefile b/usr/src/uts/intel/eventfd/Makefile
new file mode 100644
index 0000000000..6edff0931a
--- /dev/null
+++ b/usr/src/uts/intel/eventfd/Makefile
@@ -0,0 +1,68 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+#
+
+#
+#	Path to the base of the uts directory tree (usually /usr/src/uts).
+#
+UTSBASE	= ../..
+
+#
+#	Define the module and object file sets.
+#
+MODULE		= eventfd
+OBJECTS		= $(EVENTFD_OBJS:%=$(OBJS_DIR)/%)
+LINTS		= $(EVENTFD_OBJS:%.o=$(LINTS_DIR)/%.ln)
+ROOTMODULE	= $(USR_DRV_DIR)/$(MODULE)
+CONF_SRCDIR	= $(UTSBASE)/common/io
+
+#
+#	Include common rules.
+#
+include $(UTSBASE)/intel/Makefile.intel
+
+CERRWARN	+= -_gcc=-Wno-parentheses
+
+#
+#	Define targets
+#
+ALL_TARGET	= $(BINARY) $(SRC_CONFILE)
+LINT_TARGET	= $(MODULE).lint
+INSTALL_TARGET	= $(BINARY) $(ROOTMODULE) $(ROOT_CONFFILE)
+
+#
+#	Default build targets.
+#
+.KEEP_STATE:
+
+def:		$(DEF_DEPS)
+
+all:		$(ALL_DEPS)
+
+clean:		$(CLEAN_DEPS)
+
+clobber:	$(CLOBBER_DEPS)
+
+lint:		$(LINT_DEPS)
+
+modlintlib:	$(MODLINTLIB_DEPS)
+
+clean.lint:	$(CLEAN_LINT_DEPS)
+
+install:	$(INSTALL_DEPS)
+
+#
+#	Include common targets.
+#
+include $(UTSBASE)/intel/Makefile.targ
diff --git a/usr/src/uts/sparc/Makefile.sparc b/usr/src/uts/sparc/Makefile.sparc
index b989364998..cd7206a353 100644
--- a/usr/src/uts/sparc/Makefile.sparc
+++ b/usr/src/uts/sparc/Makefile.sparc
@@ -238,6 +238,7 @@ DRV_KMODS	+= bridge trill
 DRV_KMODS	+= bpf
 DRV_KMODS	+= dca
 DRV_KMODS	+= inotify
+DRV_KMODS	+= eventfd
 
 #
 #       Hardware Drivers in common space
diff --git a/usr/src/uts/sparc/eventfd/Makefile b/usr/src/uts/sparc/eventfd/Makefile
new file mode 100644
index 0000000000..063aa29b78
--- /dev/null
+++ b/usr/src/uts/sparc/eventfd/Makefile
@@ -0,0 +1,68 @@
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2014 Joyent, Inc.  All rights reserved.
+#
+
+#
+#	Path to the base of the uts directory tree (usually /usr/src/uts).
+#
+UTSBASE	= ../..
+
+#
+#	Define the module and object file sets.
+#
+MODULE		= eventfd
+OBJECTS		= $(EVENTFD_OBJS:%=$(OBJS_DIR)/%)
+LINTS		= $(EVENTFD_OBJS:%.o=$(LINTS_DIR)/%.ln)
+ROOTMODULE	= $(USR_DRV_DIR)/$(MODULE)
+CONF_SRCDIR	= $(UTSBASE)/common/io
+
+#
+#	Include common rules.
+#
+include $(UTSBASE)/sparc/Makefile.sparc
+
+CERRWARN	+= -_gcc=-Wno-parentheses
+
+#
+#	Define targets
+#
+ALL_TARGET	= $(BINARY) $(SRC_CONFILE)
+LINT_TARGET	= $(MODULE).lint
+INSTALL_TARGET	= $(BINARY) $(ROOTMODULE) $(ROOT_CONFFILE)
+
+#
+#	Default build targets.
+#
+.KEEP_STATE:
+
+def:		$(DEF_DEPS)
+
+all:		$(ALL_DEPS)
+
+clean:		$(CLEAN_DEPS)
+
+clobber:	$(CLOBBER_DEPS)
+
+lint:		$(LINT_DEPS)
+
+modlintlib:	$(MODLINTLIB_DEPS)
+
+clean.lint:	$(CLEAN_LINT_DEPS)
+
+install:	$(INSTALL_DEPS)
+
+#
+#	Include common targets.
+#
+include $(UTSBASE)/sparc/Makefile.targ