xen: add /dev/xen/evtchn driver
Ian Campbell [Sat, 7 Feb 2009 03:21:19 +0000 (19:21 -0800)]
This driver is used by application which wish to receive notifications
from the hypervisor or other guests via Xen's event channel
mechanism. In particular it is used by the xenstore daemon in domain
0.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>

drivers/xen/Kconfig
drivers/xen/Makefile
drivers/xen/evtchn.c [new file with mode: 0644]
include/xen/evtchn.h [new file with mode: 0644]

index 526187c..1bbb910 100644 (file)
@@ -18,6 +18,16 @@ config XEN_SCRUB_PAGES
          secure, but slightly less efficient.
          If in doubt, say yes.
 
+config XEN_DEV_EVTCHN
+       tristate "Xen /dev/xen/evtchn device"
+       depends on XEN
+       default y
+       help
+         The evtchn driver allows a userspace process to triger event
+         channels and to receive notification of an event channel
+         firing.
+         If in doubt, say yes.
+
 config XENFS
        tristate "Xen filesystem"
        depends on XEN
index ff8accc..1567639 100644 (file)
@@ -4,4 +4,5 @@ obj-y   += xenbus/
 obj-$(CONFIG_HOTPLUG_CPU)      += cpu_hotplug.o
 obj-$(CONFIG_XEN_XENCOMM)      += xencomm.o
 obj-$(CONFIG_XEN_BALLOON)      += balloon.o
-obj-$(CONFIG_XENFS)            += xenfs/
\ No newline at end of file
+obj-$(CONFIG_XEN_DEV_EVTCHN)   += evtchn.o
+obj-$(CONFIG_XENFS)            += xenfs/
diff --git a/drivers/xen/evtchn.c b/drivers/xen/evtchn.c
new file mode 100644 (file)
index 0000000..517b9ee
--- /dev/null
@@ -0,0 +1,494 @@
+/******************************************************************************
+ * evtchn.c
+ *
+ * Driver for receiving and demuxing event-channel signals.
+ *
+ * Copyright (c) 2004-2005, K A Fraser
+ * Multi-process extensions Copyright (c) 2004, Steven Smith
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version 2
+ * as published by the Free Software Foundation; or, when distributed
+ * separately from the Linux kernel or incorporated into other
+ * software packages, subject to the following license:
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this source file (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use, copy, modify,
+ * merge, publish, distribute, sublicense, and/or sell copies of the Software,
+ * and to permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/fs.h>
+#include <linux/errno.h>
+#include <linux/miscdevice.h>
+#include <linux/major.h>
+#include <linux/proc_fs.h>
+#include <linux/stat.h>
+#include <linux/poll.h>
+#include <linux/irq.h>
+#include <linux/init.h>
+#include <linux/gfp.h>
+#include <linux/mutex.h>
+#include <linux/cpu.h>
+#include <xen/events.h>
+#include <xen/evtchn.h>
+#include <asm/xen/hypervisor.h>
+
+struct per_user_data {
+       /* Notification ring, accessed via /dev/xen/evtchn. */
+#define EVTCHN_RING_SIZE     (PAGE_SIZE / sizeof(evtchn_port_t))
+#define EVTCHN_RING_MASK(_i) ((_i)&(EVTCHN_RING_SIZE-1))
+       evtchn_port_t *ring;
+       unsigned int ring_cons, ring_prod, ring_overflow;
+       struct mutex ring_cons_mutex; /* protect against concurrent readers */
+
+       /* Processes wait on this queue when ring is empty. */
+       wait_queue_head_t evtchn_wait;
+       struct fasync_struct *evtchn_async_queue;
+       const char *name;
+};
+
+/* Who's bound to each port? */
+static struct per_user_data *port_user[NR_EVENT_CHANNELS];
+static DEFINE_SPINLOCK(port_user_lock);
+
+irqreturn_t evtchn_interrupt(int irq, void *data)
+{
+       unsigned int port = (unsigned long)data;
+       struct per_user_data *u;
+
+       spin_lock(&port_user_lock);
+
+       u = port_user[port];
+
+       disable_irq_nosync(irq);
+
+       if ((u->ring_prod - u->ring_cons) < EVTCHN_RING_SIZE) {
+               u->ring[EVTCHN_RING_MASK(u->ring_prod)] = port;
+               wmb(); /* Ensure ring contents visible */
+               if (u->ring_cons == u->ring_prod++) {
+                       wake_up_interruptible(&u->evtchn_wait);
+                       kill_fasync(&u->evtchn_async_queue,
+                                   SIGIO, POLL_IN);
+               }
+       } else {
+               u->ring_overflow = 1;
+       }
+
+       spin_unlock(&port_user_lock);
+
+       return IRQ_HANDLED;
+}
+
+static ssize_t evtchn_read(struct file *file, char __user *buf,
+                          size_t count, loff_t *ppos)
+{
+       int rc;
+       unsigned int c, p, bytes1 = 0, bytes2 = 0;
+       struct per_user_data *u = file->private_data;
+
+       /* Whole number of ports. */
+       count &= ~(sizeof(evtchn_port_t)-1);
+
+       if (count == 0)
+               return 0;
+
+       if (count > PAGE_SIZE)
+               count = PAGE_SIZE;
+
+       for (;;) {
+               mutex_lock(&u->ring_cons_mutex);
+
+               rc = -EFBIG;
+               if (u->ring_overflow)
+                       goto unlock_out;
+
+               c = u->ring_cons;
+               p = u->ring_prod;
+               if (c != p)
+                       break;
+
+               mutex_unlock(&u->ring_cons_mutex);
+
+               if (file->f_flags & O_NONBLOCK)
+                       return -EAGAIN;
+
+               rc = wait_event_interruptible(u->evtchn_wait,
+                                             u->ring_cons != u->ring_prod);
+               if (rc)
+                       return rc;
+       }
+
+       /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
+       if (((c ^ p) & EVTCHN_RING_SIZE) != 0) {
+               bytes1 = (EVTCHN_RING_SIZE - EVTCHN_RING_MASK(c)) *
+                       sizeof(evtchn_port_t);
+               bytes2 = EVTCHN_RING_MASK(p) * sizeof(evtchn_port_t);
+       } else {
+               bytes1 = (p - c) * sizeof(evtchn_port_t);
+               bytes2 = 0;
+       }
+
+       /* Truncate chunks according to caller's maximum byte count. */
+       if (bytes1 > count) {
+               bytes1 = count;
+               bytes2 = 0;
+       } else if ((bytes1 + bytes2) > count) {
+               bytes2 = count - bytes1;
+       }
+
+       rc = -EFAULT;
+       rmb(); /* Ensure that we see the port before we copy it. */
+       if (copy_to_user(buf, &u->ring[EVTCHN_RING_MASK(c)], bytes1) ||
+           ((bytes2 != 0) &&
+            copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
+               goto unlock_out;
+
+       u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t);
+       rc = bytes1 + bytes2;
+
+ unlock_out:
+       mutex_unlock(&u->ring_cons_mutex);
+       return rc;
+}
+
+static ssize_t evtchn_write(struct file *file, const char __user *buf,
+                           size_t count, loff_t *ppos)
+{
+       int rc, i;
+       evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
+       struct per_user_data *u = file->private_data;
+
+       if (kbuf == NULL)
+               return -ENOMEM;
+
+       /* Whole number of ports. */
+       count &= ~(sizeof(evtchn_port_t)-1);
+
+       rc = 0;
+       if (count == 0)
+               goto out;
+
+       if (count > PAGE_SIZE)
+               count = PAGE_SIZE;
+
+       rc = -EFAULT;
+       if (copy_from_user(kbuf, buf, count) != 0)
+               goto out;
+
+       spin_lock_irq(&port_user_lock);
+       for (i = 0; i < (count/sizeof(evtchn_port_t)); i++)
+               if ((kbuf[i] < NR_EVENT_CHANNELS) && (port_user[kbuf[i]] == u))
+                       enable_irq(irq_from_evtchn(kbuf[i]));
+       spin_unlock_irq(&port_user_lock);
+
+       rc = count;
+
+ out:
+       free_page((unsigned long)kbuf);
+       return rc;
+}
+
+static int evtchn_bind_to_user(struct per_user_data *u, int port)
+{
+       int irq;
+       int rc = 0;
+
+       spin_lock_irq(&port_user_lock);
+
+       BUG_ON(port_user[port] != NULL);
+
+       irq = bind_evtchn_to_irqhandler(port, evtchn_interrupt, IRQF_DISABLED,
+                                       u->name, (void *)(unsigned long)port);
+       if (rc < 0)
+               goto fail;
+
+       port_user[port] = u;
+
+fail:
+       spin_unlock_irq(&port_user_lock);
+       return rc;
+}
+
+static void evtchn_unbind_from_user(struct per_user_data *u, int port)
+{
+       int irq = irq_from_evtchn(port);
+
+       unbind_from_irqhandler(irq, (void *)(unsigned long)port);
+       port_user[port] = NULL;
+}
+
+static long evtchn_ioctl(struct file *file,
+                        unsigned int cmd, unsigned long arg)
+{
+       int rc;
+       struct per_user_data *u = file->private_data;
+       void __user *uarg = (void __user *) arg;
+
+       switch (cmd) {
+       case IOCTL_EVTCHN_BIND_VIRQ: {
+               struct ioctl_evtchn_bind_virq bind;
+               struct evtchn_bind_virq bind_virq;
+
+               rc = -EFAULT;
+               if (copy_from_user(&bind, uarg, sizeof(bind)))
+                       break;
+
+               bind_virq.virq = bind.virq;
+               bind_virq.vcpu = 0;
+               rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
+                                                &bind_virq);
+               if (rc != 0)
+                       break;
+
+               rc = evtchn_bind_to_user(u, bind_virq.port);
+               if (rc == 0)
+                       rc = bind_virq.port;
+               break;
+       }
+
+       case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
+               struct ioctl_evtchn_bind_interdomain bind;
+               struct evtchn_bind_interdomain bind_interdomain;
+
+               rc = -EFAULT;
+               if (copy_from_user(&bind, uarg, sizeof(bind)))
+                       break;
+
+               bind_interdomain.remote_dom  = bind.remote_domain;
+               bind_interdomain.remote_port = bind.remote_port;
+               rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
+                                                &bind_interdomain);
+               if (rc != 0)
+                       break;
+
+               rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
+               if (rc == 0)
+                       rc = bind_interdomain.local_port;
+               break;
+       }
+
+       case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
+               struct ioctl_evtchn_bind_unbound_port bind;
+               struct evtchn_alloc_unbound alloc_unbound;
+
+               rc = -EFAULT;
+               if (copy_from_user(&bind, uarg, sizeof(bind)))
+                       break;
+
+               alloc_unbound.dom        = DOMID_SELF;
+               alloc_unbound.remote_dom = bind.remote_domain;
+               rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
+                                                &alloc_unbound);
+               if (rc != 0)
+                       break;
+
+               rc = evtchn_bind_to_user(u, alloc_unbound.port);
+               if (rc == 0)
+                       rc = alloc_unbound.port;
+               break;
+       }
+
+       case IOCTL_EVTCHN_UNBIND: {
+               struct ioctl_evtchn_unbind unbind;
+
+               rc = -EFAULT;
+               if (copy_from_user(&unbind, uarg, sizeof(unbind)))
+                       break;
+
+               rc = -EINVAL;
+               if (unbind.port >= NR_EVENT_CHANNELS)
+                       break;
+
+               spin_lock_irq(&port_user_lock);
+
+               rc = -ENOTCONN;
+               if (port_user[unbind.port] != u) {
+                       spin_unlock_irq(&port_user_lock);
+                       break;
+               }
+
+               evtchn_unbind_from_user(u, unbind.port);
+
+               spin_unlock_irq(&port_user_lock);
+
+               rc = 0;
+               break;
+       }
+
+       case IOCTL_EVTCHN_NOTIFY: {
+               struct ioctl_evtchn_notify notify;
+
+               rc = -EFAULT;
+               if (copy_from_user(&notify, uarg, sizeof(notify)))
+                       break;
+
+               if (notify.port >= NR_EVENT_CHANNELS) {
+                       rc = -EINVAL;
+               } else if (port_user[notify.port] != u) {
+                       rc = -ENOTCONN;
+               } else {
+                       notify_remote_via_evtchn(notify.port);
+                       rc = 0;
+               }
+               break;
+       }
+
+       case IOCTL_EVTCHN_RESET: {
+               /* Initialise the ring to empty. Clear errors. */
+               mutex_lock(&u->ring_cons_mutex);
+               spin_lock_irq(&port_user_lock);
+               u->ring_cons = u->ring_prod = u->ring_overflow = 0;
+               spin_unlock_irq(&port_user_lock);
+               mutex_unlock(&u->ring_cons_mutex);
+               rc = 0;
+               break;
+       }
+
+       default:
+               rc = -ENOSYS;
+               break;
+       }
+
+       return rc;
+}
+
+static unsigned int evtchn_poll(struct file *file, poll_table *wait)
+{
+       unsigned int mask = POLLOUT | POLLWRNORM;
+       struct per_user_data *u = file->private_data;
+
+       poll_wait(file, &u->evtchn_wait, wait);
+       if (u->ring_cons != u->ring_prod)
+               mask |= POLLIN | POLLRDNORM;
+       if (u->ring_overflow)
+               mask = POLLERR;
+       return mask;
+}
+
+static int evtchn_fasync(int fd, struct file *filp, int on)
+{
+       struct per_user_data *u = filp->private_data;
+       return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
+}
+
+static int evtchn_open(struct inode *inode, struct file *filp)
+{
+       struct per_user_data *u;
+
+       u = kzalloc(sizeof(*u), GFP_KERNEL);
+       if (u == NULL)
+               return -ENOMEM;
+
+       u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
+       if (u->name == NULL) {
+               kfree(u);
+               return -ENOMEM;
+       }
+
+       init_waitqueue_head(&u->evtchn_wait);
+
+       u->ring = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
+       if (u->ring == NULL) {
+               kfree(u->name);
+               kfree(u);
+               return -ENOMEM;
+       }
+
+       mutex_init(&u->ring_cons_mutex);
+
+       filp->private_data = u;
+
+       return 0;
+}
+
+static int evtchn_release(struct inode *inode, struct file *filp)
+{
+       int i;
+       struct per_user_data *u = filp->private_data;
+
+       spin_lock_irq(&port_user_lock);
+
+       free_page((unsigned long)u->ring);
+
+       for (i = 0; i < NR_EVENT_CHANNELS; i++) {
+               if (port_user[i] != u)
+                       continue;
+
+               evtchn_unbind_from_user(port_user[i], i);
+       }
+
+       spin_unlock_irq(&port_user_lock);
+
+       kfree(u->name);
+       kfree(u);
+
+       return 0;
+}
+
+static const struct file_operations evtchn_fops = {
+       .owner   = THIS_MODULE,
+       .read    = evtchn_read,
+       .write   = evtchn_write,
+       .unlocked_ioctl = evtchn_ioctl,
+       .poll    = evtchn_poll,
+       .fasync  = evtchn_fasync,
+       .open    = evtchn_open,
+       .release = evtchn_release,
+};
+
+static struct miscdevice evtchn_miscdev = {
+       .minor        = MISC_DYNAMIC_MINOR,
+       .name         = "evtchn",
+       .fops         = &evtchn_fops,
+};
+static int __init evtchn_init(void)
+{
+       int err;
+
+       if (!xen_domain())
+               return -ENODEV;
+
+       spin_lock_init(&port_user_lock);
+       memset(port_user, 0, sizeof(port_user));
+
+       /* Create '/dev/misc/evtchn'. */
+       err = misc_register(&evtchn_miscdev);
+       if (err != 0) {
+               printk(KERN_ALERT "Could not register /dev/misc/evtchn\n");
+               return err;
+       }
+
+       printk(KERN_INFO "Event-channel device installed.\n");
+
+       return 0;
+}
+
+static void __exit evtchn_cleanup(void)
+{
+       misc_deregister(&evtchn_miscdev);
+}
+
+module_init(evtchn_init);
+module_exit(evtchn_cleanup);
+
+MODULE_LICENSE("GPL");
diff --git a/include/xen/evtchn.h b/include/xen/evtchn.h
new file mode 100644 (file)
index 0000000..14e833e
--- /dev/null
@@ -0,0 +1,88 @@
+/******************************************************************************
+ * evtchn.h
+ *
+ * Interface to /dev/xen/evtchn.
+ *
+ * Copyright (c) 2003-2005, K A Fraser
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version 2
+ * as published by the Free Software Foundation; or, when distributed
+ * separately from the Linux kernel or incorporated into other
+ * software packages, subject to the following license:
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this source file (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use, copy, modify,
+ * merge, publish, distribute, sublicense, and/or sell copies of the Software,
+ * and to permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#ifndef __LINUX_PUBLIC_EVTCHN_H__
+#define __LINUX_PUBLIC_EVTCHN_H__
+
+/*
+ * Bind a fresh port to VIRQ @virq.
+ * Return allocated port.
+ */
+#define IOCTL_EVTCHN_BIND_VIRQ                         \
+       _IOC(_IOC_NONE, 'E', 0, sizeof(struct ioctl_evtchn_bind_virq))
+struct ioctl_evtchn_bind_virq {
+       unsigned int virq;
+};
+
+/*
+ * Bind a fresh port to remote <@remote_domain, @remote_port>.
+ * Return allocated port.
+ */
+#define IOCTL_EVTCHN_BIND_INTERDOMAIN                  \
+       _IOC(_IOC_NONE, 'E', 1, sizeof(struct ioctl_evtchn_bind_interdomain))
+struct ioctl_evtchn_bind_interdomain {
+       unsigned int remote_domain, remote_port;
+};
+
+/*
+ * Allocate a fresh port for binding to @remote_domain.
+ * Return allocated port.
+ */
+#define IOCTL_EVTCHN_BIND_UNBOUND_PORT                 \
+       _IOC(_IOC_NONE, 'E', 2, sizeof(struct ioctl_evtchn_bind_unbound_port))
+struct ioctl_evtchn_bind_unbound_port {
+       unsigned int remote_domain;
+};
+
+/*
+ * Unbind previously allocated @port.
+ */
+#define IOCTL_EVTCHN_UNBIND                            \
+       _IOC(_IOC_NONE, 'E', 3, sizeof(struct ioctl_evtchn_unbind))
+struct ioctl_evtchn_unbind {
+       unsigned int port;
+};
+
+/*
+ * Unbind previously allocated @port.
+ */
+#define IOCTL_EVTCHN_NOTIFY                            \
+       _IOC(_IOC_NONE, 'E', 4, sizeof(struct ioctl_evtchn_notify))
+struct ioctl_evtchn_notify {
+       unsigned int port;
+};
+
+/* Clear and reinitialise the event buffer. Clear error condition. */
+#define IOCTL_EVTCHN_RESET                             \
+       _IOC(_IOC_NONE, 'E', 5, 0)
+
+#endif /* __LINUX_PUBLIC_EVTCHN_H__ */