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 <asm/xen/hypervisor.h>
60 struct per_user_data {
61 struct mutex bind_mutex; /* serialize bind/unbind operations */
62 struct rb_root evtchns;
63 unsigned int nr_evtchns;
65 /* Notification ring, accessed via /dev/xen/evtchn. */
66 unsigned int ring_size;
68 unsigned int ring_cons, ring_prod, ring_overflow;
69 struct mutex ring_cons_mutex; /* protect against concurrent readers */
70 spinlock_t ring_prod_lock; /* product against concurrent interrupts */
72 /* Processes wait on this queue when ring is empty. */
73 wait_queue_head_t evtchn_wait;
74 struct fasync_struct *evtchn_async_queue;
80 struct per_user_data *user;
85 static evtchn_port_t *evtchn_alloc_ring(unsigned int size)
88 size_t s = size * sizeof(*ring);
90 ring = kmalloc(s, GFP_KERNEL);
97 static void evtchn_free_ring(evtchn_port_t *ring)
102 static unsigned int evtchn_ring_offset(struct per_user_data *u,
105 return idx & (u->ring_size - 1);
108 static evtchn_port_t *evtchn_ring_entry(struct per_user_data *u,
111 return u->ring + evtchn_ring_offset(u, idx);
114 static int add_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
116 struct rb_node **new = &(u->evtchns.rb_node), *parent = NULL;
121 struct user_evtchn *this;
123 this = container_of(*new, struct user_evtchn, node);
126 if (this->port < evtchn->port)
127 new = &((*new)->rb_left);
128 else if (this->port > evtchn->port)
129 new = &((*new)->rb_right);
134 /* Add new node and rebalance tree. */
135 rb_link_node(&evtchn->node, parent, new);
136 rb_insert_color(&evtchn->node, &u->evtchns);
141 static void del_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
144 rb_erase(&evtchn->node, &u->evtchns);
148 static struct user_evtchn *find_evtchn(struct per_user_data *u, unsigned port)
150 struct rb_node *node = u->evtchns.rb_node;
153 struct user_evtchn *evtchn;
155 evtchn = container_of(node, struct user_evtchn, node);
157 if (evtchn->port < port)
158 node = node->rb_left;
159 else if (evtchn->port > port)
160 node = node->rb_right;
167 static irqreturn_t evtchn_interrupt(int irq, void *data)
169 struct user_evtchn *evtchn = data;
170 struct per_user_data *u = evtchn->user;
172 WARN(!evtchn->enabled,
173 "Interrupt for port %d, but apparently not enabled; per-user %p\n",
176 disable_irq_nosync(irq);
177 evtchn->enabled = false;
179 spin_lock(&u->ring_prod_lock);
181 if ((u->ring_prod - u->ring_cons) < u->ring_size) {
182 *evtchn_ring_entry(u, u->ring_prod) = evtchn->port;
183 wmb(); /* Ensure ring contents visible */
184 if (u->ring_cons == u->ring_prod++) {
185 wake_up_interruptible(&u->evtchn_wait);
186 kill_fasync(&u->evtchn_async_queue,
190 u->ring_overflow = 1;
192 spin_unlock(&u->ring_prod_lock);
197 static ssize_t evtchn_read(struct file *file, char __user *buf,
198 size_t count, loff_t *ppos)
201 unsigned int c, p, bytes1 = 0, bytes2 = 0;
202 struct per_user_data *u = file->private_data;
204 /* Whole number of ports. */
205 count &= ~(sizeof(evtchn_port_t)-1);
210 if (count > PAGE_SIZE)
214 mutex_lock(&u->ring_cons_mutex);
217 if (u->ring_overflow)
225 mutex_unlock(&u->ring_cons_mutex);
227 if (file->f_flags & O_NONBLOCK)
230 rc = wait_event_interruptible(u->evtchn_wait,
231 u->ring_cons != u->ring_prod);
236 /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
237 if (((c ^ p) & u->ring_size) != 0) {
238 bytes1 = (u->ring_size - evtchn_ring_offset(u, c)) *
239 sizeof(evtchn_port_t);
240 bytes2 = evtchn_ring_offset(u, p) * sizeof(evtchn_port_t);
242 bytes1 = (p - c) * sizeof(evtchn_port_t);
246 /* Truncate chunks according to caller's maximum byte count. */
247 if (bytes1 > count) {
250 } else if ((bytes1 + bytes2) > count) {
251 bytes2 = count - bytes1;
255 rmb(); /* Ensure that we see the port before we copy it. */
256 if (copy_to_user(buf, evtchn_ring_entry(u, c), bytes1) ||
258 copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
261 u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t);
262 rc = bytes1 + bytes2;
265 mutex_unlock(&u->ring_cons_mutex);
269 static ssize_t evtchn_write(struct file *file, const char __user *buf,
270 size_t count, loff_t *ppos)
273 evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
274 struct per_user_data *u = file->private_data;
279 /* Whole number of ports. */
280 count &= ~(sizeof(evtchn_port_t)-1);
286 if (count > PAGE_SIZE)
290 if (copy_from_user(kbuf, buf, count) != 0)
293 mutex_lock(&u->bind_mutex);
295 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
296 unsigned port = kbuf[i];
297 struct user_evtchn *evtchn;
299 evtchn = find_evtchn(u, port);
300 if (evtchn && !evtchn->enabled) {
301 evtchn->enabled = true;
302 enable_irq(irq_from_evtchn(port));
306 mutex_unlock(&u->bind_mutex);
311 free_page((unsigned long)kbuf);
315 static int evtchn_resize_ring(struct per_user_data *u)
317 unsigned int new_size;
318 evtchn_port_t *new_ring, *old_ring;
322 * Ensure the ring is large enough to capture all possible
323 * events. i.e., one free slot for each bound event.
325 if (u->nr_evtchns <= u->ring_size)
328 if (u->ring_size == 0)
331 new_size = 2 * u->ring_size;
333 new_ring = evtchn_alloc_ring(new_size);
340 * Access to the ring contents is serialized by either the
341 * prod /or/ cons lock so take both when resizing.
343 mutex_lock(&u->ring_cons_mutex);
344 spin_lock_irq(&u->ring_prod_lock);
347 * Copy the old ring contents to the new ring.
349 * If the ring contents crosses the end of the current ring,
350 * it needs to be copied in two chunks.
352 * +---------+ +------------------+
353 * |34567 12| -> | 1234567 |
354 * +-----p-c-+ +------------------+
356 p = evtchn_ring_offset(u, u->ring_prod);
357 c = evtchn_ring_offset(u, u->ring_cons);
359 memcpy(new_ring + c, u->ring + c, (u->ring_size - c) * sizeof(*u->ring));
360 memcpy(new_ring + u->ring_size, u->ring, p * sizeof(*u->ring));
362 memcpy(new_ring + c, u->ring + c, (p - c) * sizeof(*u->ring));
365 u->ring_size = new_size;
367 spin_unlock_irq(&u->ring_prod_lock);
368 mutex_unlock(&u->ring_cons_mutex);
370 evtchn_free_ring(old_ring);
375 static int evtchn_bind_to_user(struct per_user_data *u, int port)
377 struct user_evtchn *evtchn;
378 struct evtchn_close close;
382 * Ports are never reused, so every caller should pass in a
385 * (Locking not necessary because we haven't registered the
386 * interrupt handler yet, and our caller has already
387 * serialized bind operations.)
390 evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
396 evtchn->enabled = true; /* start enabled */
398 rc = add_evtchn(u, evtchn);
402 rc = evtchn_resize_ring(u);
406 rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, 0,
411 rc = evtchn_make_refcounted(port);
415 /* bind failed, should close the port now */
417 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
419 del_evtchn(u, evtchn);
423 static void evtchn_unbind_from_user(struct per_user_data *u,
424 struct user_evtchn *evtchn)
426 int irq = irq_from_evtchn(evtchn->port);
430 unbind_from_irqhandler(irq, evtchn);
432 del_evtchn(u, evtchn);
435 static long evtchn_ioctl(struct file *file,
436 unsigned int cmd, unsigned long arg)
439 struct per_user_data *u = file->private_data;
440 void __user *uarg = (void __user *) arg;
442 /* Prevent bind from racing with unbind */
443 mutex_lock(&u->bind_mutex);
446 case IOCTL_EVTCHN_BIND_VIRQ: {
447 struct ioctl_evtchn_bind_virq bind;
448 struct evtchn_bind_virq bind_virq;
451 if (copy_from_user(&bind, uarg, sizeof(bind)))
454 bind_virq.virq = bind.virq;
456 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
461 rc = evtchn_bind_to_user(u, bind_virq.port);
467 case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
468 struct ioctl_evtchn_bind_interdomain bind;
469 struct evtchn_bind_interdomain bind_interdomain;
472 if (copy_from_user(&bind, uarg, sizeof(bind)))
475 bind_interdomain.remote_dom = bind.remote_domain;
476 bind_interdomain.remote_port = bind.remote_port;
477 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
482 rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
484 rc = bind_interdomain.local_port;
488 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
489 struct ioctl_evtchn_bind_unbound_port bind;
490 struct evtchn_alloc_unbound alloc_unbound;
493 if (copy_from_user(&bind, uarg, sizeof(bind)))
496 alloc_unbound.dom = DOMID_SELF;
497 alloc_unbound.remote_dom = bind.remote_domain;
498 rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
503 rc = evtchn_bind_to_user(u, alloc_unbound.port);
505 rc = alloc_unbound.port;
509 case IOCTL_EVTCHN_UNBIND: {
510 struct ioctl_evtchn_unbind unbind;
511 struct user_evtchn *evtchn;
514 if (copy_from_user(&unbind, uarg, sizeof(unbind)))
518 if (unbind.port >= xen_evtchn_nr_channels())
522 evtchn = find_evtchn(u, unbind.port);
526 disable_irq(irq_from_evtchn(unbind.port));
527 evtchn_unbind_from_user(u, evtchn);
532 case IOCTL_EVTCHN_NOTIFY: {
533 struct ioctl_evtchn_notify notify;
534 struct user_evtchn *evtchn;
537 if (copy_from_user(¬ify, uarg, sizeof(notify)))
541 evtchn = find_evtchn(u, notify.port);
543 notify_remote_via_evtchn(notify.port);
549 case IOCTL_EVTCHN_RESET: {
550 /* Initialise the ring to empty. Clear errors. */
551 mutex_lock(&u->ring_cons_mutex);
552 spin_lock_irq(&u->ring_prod_lock);
553 u->ring_cons = u->ring_prod = u->ring_overflow = 0;
554 spin_unlock_irq(&u->ring_prod_lock);
555 mutex_unlock(&u->ring_cons_mutex);
564 mutex_unlock(&u->bind_mutex);
569 static unsigned int evtchn_poll(struct file *file, poll_table *wait)
571 unsigned int mask = POLLOUT | POLLWRNORM;
572 struct per_user_data *u = file->private_data;
574 poll_wait(file, &u->evtchn_wait, wait);
575 if (u->ring_cons != u->ring_prod)
576 mask |= POLLIN | POLLRDNORM;
577 if (u->ring_overflow)
582 static int evtchn_fasync(int fd, struct file *filp, int on)
584 struct per_user_data *u = filp->private_data;
585 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
588 static int evtchn_open(struct inode *inode, struct file *filp)
590 struct per_user_data *u;
592 u = kzalloc(sizeof(*u), GFP_KERNEL);
596 u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
597 if (u->name == NULL) {
602 init_waitqueue_head(&u->evtchn_wait);
604 mutex_init(&u->bind_mutex);
605 mutex_init(&u->ring_cons_mutex);
606 spin_lock_init(&u->ring_prod_lock);
608 filp->private_data = u;
610 return nonseekable_open(inode, filp);
613 static int evtchn_release(struct inode *inode, struct file *filp)
615 struct per_user_data *u = filp->private_data;
616 struct rb_node *node;
618 while ((node = u->evtchns.rb_node)) {
619 struct user_evtchn *evtchn;
621 evtchn = rb_entry(node, struct user_evtchn, node);
622 disable_irq(irq_from_evtchn(evtchn->port));
623 evtchn_unbind_from_user(u, evtchn);
626 evtchn_free_ring(u->ring);
633 static const struct file_operations evtchn_fops = {
634 .owner = THIS_MODULE,
636 .write = evtchn_write,
637 .unlocked_ioctl = evtchn_ioctl,
639 .fasync = evtchn_fasync,
641 .release = evtchn_release,
645 static struct miscdevice evtchn_miscdev = {
646 .minor = MISC_DYNAMIC_MINOR,
647 .name = "xen/evtchn",
648 .fops = &evtchn_fops,
650 static int __init evtchn_init(void)
657 /* Create '/dev/xen/evtchn'. */
658 err = misc_register(&evtchn_miscdev);
660 pr_err("Could not register /dev/xen/evtchn\n");
664 pr_info("Event-channel device installed\n");
669 static void __exit evtchn_cleanup(void)
671 misc_deregister(&evtchn_miscdev);
674 module_init(evtchn_init);
675 module_exit(evtchn_cleanup);
677 MODULE_LICENSE("GPL");