1 /******************************************************************************
4 * Driver for receiving and demuxing event-channel signals.
6 * Copyright (c) 2004-2005, K A Fraser
7 * Multi-process extensions Copyright (c) 2004, Steven Smith
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License version 2
11 * as published by the Free Software Foundation; or, when distributed
12 * separately from the Linux kernel or incorporated into other
13 * software packages, subject to the following license:
15 * Permission is hereby granted, free of charge, to any person obtaining a copy
16 * of this source file (the "Software"), to deal in the Software without
17 * restriction, including without limitation the rights to use, copy, modify,
18 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
19 * and to permit persons to whom the Software is furnished to do so, subject to
20 * the following conditions:
22 * The above copyright notice and this permission notice shall be included in
23 * all copies or substantial portions of the Software.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
26 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
27 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
28 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
29 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
30 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
34 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/sched.h>
39 #include <linux/slab.h>
40 #include <linux/string.h>
41 #include <linux/errno.h>
43 #include <linux/miscdevice.h>
44 #include <linux/major.h>
45 #include <linux/proc_fs.h>
46 #include <linux/stat.h>
47 #include <linux/poll.h>
48 #include <linux/irq.h>
49 #include <linux/init.h>
50 #include <linux/mutex.h>
51 #include <linux/cpu.h>
53 #include <linux/vmalloc.h>
56 #include <xen/events.h>
57 #include <xen/evtchn.h>
58 #include <xen/xen-ops.h>
59 #include <asm/xen/hypervisor.h>
61 struct per_user_data {
62 struct mutex bind_mutex; /* serialize bind/unbind operations */
63 struct rb_root evtchns;
64 unsigned int nr_evtchns;
66 /* Notification ring, accessed via /dev/xen/evtchn. */
67 unsigned int ring_size;
69 unsigned int ring_cons, ring_prod, ring_overflow;
70 struct mutex ring_cons_mutex; /* protect against concurrent readers */
71 spinlock_t ring_prod_lock; /* product against concurrent interrupts */
73 /* Processes wait on this queue when ring is empty. */
74 wait_queue_head_t evtchn_wait;
75 struct fasync_struct *evtchn_async_queue;
78 domid_t restrict_domid;
81 #define UNRESTRICTED_DOMID ((domid_t)-1)
85 struct per_user_data *user;
90 static void evtchn_free_ring(evtchn_port_t *ring)
95 static unsigned int evtchn_ring_offset(struct per_user_data *u,
98 return idx & (u->ring_size - 1);
101 static evtchn_port_t *evtchn_ring_entry(struct per_user_data *u,
104 return u->ring + evtchn_ring_offset(u, idx);
107 static int add_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
109 struct rb_node **new = &(u->evtchns.rb_node), *parent = NULL;
114 struct user_evtchn *this;
116 this = rb_entry(*new, struct user_evtchn, node);
119 if (this->port < evtchn->port)
120 new = &((*new)->rb_left);
121 else if (this->port > evtchn->port)
122 new = &((*new)->rb_right);
127 /* Add new node and rebalance tree. */
128 rb_link_node(&evtchn->node, parent, new);
129 rb_insert_color(&evtchn->node, &u->evtchns);
134 static void del_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
137 rb_erase(&evtchn->node, &u->evtchns);
141 static struct user_evtchn *find_evtchn(struct per_user_data *u,
144 struct rb_node *node = u->evtchns.rb_node;
147 struct user_evtchn *evtchn;
149 evtchn = rb_entry(node, struct user_evtchn, node);
151 if (evtchn->port < port)
152 node = node->rb_left;
153 else if (evtchn->port > port)
154 node = node->rb_right;
161 static irqreturn_t evtchn_interrupt(int irq, void *data)
163 struct user_evtchn *evtchn = data;
164 struct per_user_data *u = evtchn->user;
165 unsigned int prod, cons;
167 WARN(!evtchn->enabled,
168 "Interrupt for port %u, but apparently not enabled; per-user %p\n",
171 evtchn->enabled = false;
173 spin_lock(&u->ring_prod_lock);
175 prod = READ_ONCE(u->ring_prod);
176 cons = READ_ONCE(u->ring_cons);
178 if ((prod - cons) < u->ring_size) {
179 *evtchn_ring_entry(u, prod) = evtchn->port;
180 smp_wmb(); /* Ensure ring contents visible */
181 WRITE_ONCE(u->ring_prod, prod + 1);
183 wake_up_interruptible(&u->evtchn_wait);
184 kill_fasync(&u->evtchn_async_queue,
188 u->ring_overflow = 1;
190 spin_unlock(&u->ring_prod_lock);
195 static ssize_t evtchn_read(struct file *file, char __user *buf,
196 size_t count, loff_t *ppos)
199 unsigned int c, p, bytes1 = 0, bytes2 = 0;
200 struct per_user_data *u = file->private_data;
202 /* Whole number of ports. */
203 count &= ~(sizeof(evtchn_port_t)-1);
208 if (count > PAGE_SIZE)
212 mutex_lock(&u->ring_cons_mutex);
215 if (u->ring_overflow)
218 c = READ_ONCE(u->ring_cons);
219 p = READ_ONCE(u->ring_prod);
223 mutex_unlock(&u->ring_cons_mutex);
225 if (file->f_flags & O_NONBLOCK)
228 rc = wait_event_interruptible(u->evtchn_wait,
229 READ_ONCE(u->ring_cons) != READ_ONCE(u->ring_prod));
234 /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
235 if (((c ^ p) & u->ring_size) != 0) {
236 bytes1 = (u->ring_size - evtchn_ring_offset(u, c)) *
237 sizeof(evtchn_port_t);
238 bytes2 = evtchn_ring_offset(u, p) * sizeof(evtchn_port_t);
240 bytes1 = (p - c) * sizeof(evtchn_port_t);
244 /* Truncate chunks according to caller's maximum byte count. */
245 if (bytes1 > count) {
248 } else if ((bytes1 + bytes2) > count) {
249 bytes2 = count - bytes1;
253 smp_rmb(); /* Ensure that we see the port before we copy it. */
254 if (copy_to_user(buf, evtchn_ring_entry(u, c), bytes1) ||
256 copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
259 WRITE_ONCE(u->ring_cons, c + (bytes1 + bytes2) / sizeof(evtchn_port_t));
260 rc = bytes1 + bytes2;
263 mutex_unlock(&u->ring_cons_mutex);
267 static ssize_t evtchn_write(struct file *file, const char __user *buf,
268 size_t count, loff_t *ppos)
271 evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
272 struct per_user_data *u = file->private_data;
277 /* Whole number of ports. */
278 count &= ~(sizeof(evtchn_port_t)-1);
284 if (count > PAGE_SIZE)
288 if (copy_from_user(kbuf, buf, count) != 0)
291 mutex_lock(&u->bind_mutex);
293 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
294 evtchn_port_t port = kbuf[i];
295 struct user_evtchn *evtchn;
297 evtchn = find_evtchn(u, port);
298 if (evtchn && !evtchn->enabled) {
299 evtchn->enabled = true;
300 xen_irq_lateeoi(irq_from_evtchn(port), 0);
304 mutex_unlock(&u->bind_mutex);
309 free_page((unsigned long)kbuf);
313 static int evtchn_resize_ring(struct per_user_data *u)
315 unsigned int new_size;
316 evtchn_port_t *new_ring, *old_ring;
319 * Ensure the ring is large enough to capture all possible
320 * events. i.e., one free slot for each bound event.
322 if (u->nr_evtchns <= u->ring_size)
325 if (u->ring_size == 0)
328 new_size = 2 * u->ring_size;
330 new_ring = kvmalloc_array(new_size, sizeof(*new_ring), GFP_KERNEL);
337 * Access to the ring contents is serialized by either the
338 * prod /or/ cons lock so take both when resizing.
340 mutex_lock(&u->ring_cons_mutex);
341 spin_lock_irq(&u->ring_prod_lock);
344 * Copy the old ring contents to the new ring.
346 * To take care of wrapping, a full ring, and the new index
347 * pointing into the second half, simply copy the old contents
350 * +---------+ +------------------+
351 * |34567 12| -> |34567 1234567 12|
352 * +-----p-c-+ +-------c------p---+
354 memcpy(new_ring, old_ring, u->ring_size * sizeof(*u->ring));
355 memcpy(new_ring + u->ring_size, old_ring,
356 u->ring_size * sizeof(*u->ring));
359 u->ring_size = new_size;
361 spin_unlock_irq(&u->ring_prod_lock);
362 mutex_unlock(&u->ring_cons_mutex);
364 evtchn_free_ring(old_ring);
369 static int evtchn_bind_to_user(struct per_user_data *u, evtchn_port_t port)
371 struct user_evtchn *evtchn;
372 struct evtchn_close close;
376 * Ports are never reused, so every caller should pass in a
379 * (Locking not necessary because we haven't registered the
380 * interrupt handler yet, and our caller has already
381 * serialized bind operations.)
384 evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
390 evtchn->enabled = true; /* start enabled */
392 rc = add_evtchn(u, evtchn);
396 rc = evtchn_resize_ring(u);
400 rc = bind_evtchn_to_irqhandler_lateeoi(port, evtchn_interrupt, 0,
405 rc = evtchn_make_refcounted(port);
409 /* bind failed, should close the port now */
411 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
413 del_evtchn(u, evtchn);
417 static void evtchn_unbind_from_user(struct per_user_data *u,
418 struct user_evtchn *evtchn)
420 int irq = irq_from_evtchn(evtchn->port);
424 unbind_from_irqhandler(irq, evtchn);
426 del_evtchn(u, evtchn);
429 static long evtchn_ioctl(struct file *file,
430 unsigned int cmd, unsigned long arg)
433 struct per_user_data *u = file->private_data;
434 void __user *uarg = (void __user *) arg;
436 /* Prevent bind from racing with unbind */
437 mutex_lock(&u->bind_mutex);
440 case IOCTL_EVTCHN_BIND_VIRQ: {
441 struct ioctl_evtchn_bind_virq bind;
442 struct evtchn_bind_virq bind_virq;
445 if (u->restrict_domid != UNRESTRICTED_DOMID)
449 if (copy_from_user(&bind, uarg, sizeof(bind)))
452 bind_virq.virq = bind.virq;
453 bind_virq.vcpu = xen_vcpu_nr(0);
454 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
459 rc = evtchn_bind_to_user(u, bind_virq.port);
465 case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
466 struct ioctl_evtchn_bind_interdomain bind;
467 struct evtchn_bind_interdomain bind_interdomain;
470 if (copy_from_user(&bind, uarg, sizeof(bind)))
474 if (u->restrict_domid != UNRESTRICTED_DOMID &&
475 u->restrict_domid != bind.remote_domain)
478 bind_interdomain.remote_dom = bind.remote_domain;
479 bind_interdomain.remote_port = bind.remote_port;
480 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
485 rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
487 rc = bind_interdomain.local_port;
491 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
492 struct ioctl_evtchn_bind_unbound_port bind;
493 struct evtchn_alloc_unbound alloc_unbound;
496 if (u->restrict_domid != UNRESTRICTED_DOMID)
500 if (copy_from_user(&bind, uarg, sizeof(bind)))
503 alloc_unbound.dom = DOMID_SELF;
504 alloc_unbound.remote_dom = bind.remote_domain;
505 rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
510 rc = evtchn_bind_to_user(u, alloc_unbound.port);
512 rc = alloc_unbound.port;
516 case IOCTL_EVTCHN_UNBIND: {
517 struct ioctl_evtchn_unbind unbind;
518 struct user_evtchn *evtchn;
521 if (copy_from_user(&unbind, uarg, sizeof(unbind)))
525 if (unbind.port >= xen_evtchn_nr_channels())
529 evtchn = find_evtchn(u, unbind.port);
533 disable_irq(irq_from_evtchn(unbind.port));
534 evtchn_unbind_from_user(u, evtchn);
539 case IOCTL_EVTCHN_NOTIFY: {
540 struct ioctl_evtchn_notify notify;
541 struct user_evtchn *evtchn;
544 if (copy_from_user(¬ify, uarg, sizeof(notify)))
548 evtchn = find_evtchn(u, notify.port);
550 notify_remote_via_evtchn(notify.port);
556 case IOCTL_EVTCHN_RESET: {
557 /* Initialise the ring to empty. Clear errors. */
558 mutex_lock(&u->ring_cons_mutex);
559 spin_lock_irq(&u->ring_prod_lock);
560 WRITE_ONCE(u->ring_cons, 0);
561 WRITE_ONCE(u->ring_prod, 0);
562 u->ring_overflow = 0;
563 spin_unlock_irq(&u->ring_prod_lock);
564 mutex_unlock(&u->ring_cons_mutex);
569 case IOCTL_EVTCHN_RESTRICT_DOMID: {
570 struct ioctl_evtchn_restrict_domid ierd;
573 if (u->restrict_domid != UNRESTRICTED_DOMID)
577 if (copy_from_user(&ierd, uarg, sizeof(ierd)))
581 if (ierd.domid == 0 || ierd.domid >= DOMID_FIRST_RESERVED)
584 u->restrict_domid = ierd.domid;
594 mutex_unlock(&u->bind_mutex);
599 static __poll_t evtchn_poll(struct file *file, poll_table *wait)
601 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
602 struct per_user_data *u = file->private_data;
604 poll_wait(file, &u->evtchn_wait, wait);
605 if (READ_ONCE(u->ring_cons) != READ_ONCE(u->ring_prod))
606 mask |= EPOLLIN | EPOLLRDNORM;
607 if (u->ring_overflow)
612 static int evtchn_fasync(int fd, struct file *filp, int on)
614 struct per_user_data *u = filp->private_data;
615 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
618 static int evtchn_open(struct inode *inode, struct file *filp)
620 struct per_user_data *u;
622 u = kzalloc(sizeof(*u), GFP_KERNEL);
626 u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
627 if (u->name == NULL) {
632 init_waitqueue_head(&u->evtchn_wait);
634 mutex_init(&u->bind_mutex);
635 mutex_init(&u->ring_cons_mutex);
636 spin_lock_init(&u->ring_prod_lock);
638 u->restrict_domid = UNRESTRICTED_DOMID;
640 filp->private_data = u;
642 return stream_open(inode, filp);
645 static int evtchn_release(struct inode *inode, struct file *filp)
647 struct per_user_data *u = filp->private_data;
648 struct rb_node *node;
650 while ((node = u->evtchns.rb_node)) {
651 struct user_evtchn *evtchn;
653 evtchn = rb_entry(node, struct user_evtchn, node);
654 disable_irq(irq_from_evtchn(evtchn->port));
655 evtchn_unbind_from_user(u, evtchn);
658 evtchn_free_ring(u->ring);
665 static const struct file_operations evtchn_fops = {
666 .owner = THIS_MODULE,
668 .write = evtchn_write,
669 .unlocked_ioctl = evtchn_ioctl,
671 .fasync = evtchn_fasync,
673 .release = evtchn_release,
677 static struct miscdevice evtchn_miscdev = {
678 .minor = MISC_DYNAMIC_MINOR,
679 .name = "xen/evtchn",
680 .fops = &evtchn_fops,
682 static int __init evtchn_init(void)
689 /* Create '/dev/xen/evtchn'. */
690 err = misc_register(&evtchn_miscdev);
692 pr_err("Could not register /dev/xen/evtchn\n");
696 pr_info("Event-channel device installed\n");
701 static void __exit evtchn_cleanup(void)
703 misc_deregister(&evtchn_miscdev);
706 module_init(evtchn_init);
707 module_exit(evtchn_cleanup);
709 MODULE_LICENSE("GPL");