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, unsigned port)
143 struct rb_node *node = u->evtchns.rb_node;
146 struct user_evtchn *evtchn;
148 evtchn = rb_entry(node, struct user_evtchn, node);
150 if (evtchn->port < port)
151 node = node->rb_left;
152 else if (evtchn->port > port)
153 node = node->rb_right;
160 static irqreturn_t evtchn_interrupt(int irq, void *data)
162 struct user_evtchn *evtchn = data;
163 struct per_user_data *u = evtchn->user;
165 WARN(!evtchn->enabled,
166 "Interrupt for port %d, but apparently not enabled; per-user %p\n",
169 evtchn->enabled = false;
171 spin_lock(&u->ring_prod_lock);
173 if ((u->ring_prod - u->ring_cons) < u->ring_size) {
174 *evtchn_ring_entry(u, u->ring_prod) = evtchn->port;
175 wmb(); /* Ensure ring contents visible */
176 if (u->ring_cons == u->ring_prod++) {
177 wake_up_interruptible(&u->evtchn_wait);
178 kill_fasync(&u->evtchn_async_queue,
182 u->ring_overflow = 1;
184 spin_unlock(&u->ring_prod_lock);
189 static ssize_t evtchn_read(struct file *file, char __user *buf,
190 size_t count, loff_t *ppos)
193 unsigned int c, p, bytes1 = 0, bytes2 = 0;
194 struct per_user_data *u = file->private_data;
196 /* Whole number of ports. */
197 count &= ~(sizeof(evtchn_port_t)-1);
202 if (count > PAGE_SIZE)
206 mutex_lock(&u->ring_cons_mutex);
209 if (u->ring_overflow)
217 mutex_unlock(&u->ring_cons_mutex);
219 if (file->f_flags & O_NONBLOCK)
222 rc = wait_event_interruptible(u->evtchn_wait,
223 u->ring_cons != u->ring_prod);
228 /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
229 if (((c ^ p) & u->ring_size) != 0) {
230 bytes1 = (u->ring_size - evtchn_ring_offset(u, c)) *
231 sizeof(evtchn_port_t);
232 bytes2 = evtchn_ring_offset(u, p) * sizeof(evtchn_port_t);
234 bytes1 = (p - c) * sizeof(evtchn_port_t);
238 /* Truncate chunks according to caller's maximum byte count. */
239 if (bytes1 > count) {
242 } else if ((bytes1 + bytes2) > count) {
243 bytes2 = count - bytes1;
247 rmb(); /* Ensure that we see the port before we copy it. */
248 if (copy_to_user(buf, evtchn_ring_entry(u, c), bytes1) ||
250 copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
253 u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t);
254 rc = bytes1 + bytes2;
257 mutex_unlock(&u->ring_cons_mutex);
261 static ssize_t evtchn_write(struct file *file, const char __user *buf,
262 size_t count, loff_t *ppos)
265 evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
266 struct per_user_data *u = file->private_data;
271 /* Whole number of ports. */
272 count &= ~(sizeof(evtchn_port_t)-1);
278 if (count > PAGE_SIZE)
282 if (copy_from_user(kbuf, buf, count) != 0)
285 mutex_lock(&u->bind_mutex);
287 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
288 unsigned port = kbuf[i];
289 struct user_evtchn *evtchn;
291 evtchn = find_evtchn(u, port);
292 if (evtchn && !evtchn->enabled) {
293 evtchn->enabled = true;
294 xen_irq_lateeoi(irq_from_evtchn(port), 0);
298 mutex_unlock(&u->bind_mutex);
303 free_page((unsigned long)kbuf);
307 static int evtchn_resize_ring(struct per_user_data *u)
309 unsigned int new_size;
310 evtchn_port_t *new_ring, *old_ring;
313 * Ensure the ring is large enough to capture all possible
314 * events. i.e., one free slot for each bound event.
316 if (u->nr_evtchns <= u->ring_size)
319 if (u->ring_size == 0)
322 new_size = 2 * u->ring_size;
324 new_ring = kvmalloc(new_size * sizeof(*new_ring), GFP_KERNEL);
331 * Access to the ring contents is serialized by either the
332 * prod /or/ cons lock so take both when resizing.
334 mutex_lock(&u->ring_cons_mutex);
335 spin_lock_irq(&u->ring_prod_lock);
338 * Copy the old ring contents to the new ring.
340 * To take care of wrapping, a full ring, and the new index
341 * pointing into the second half, simply copy the old contents
344 * +---------+ +------------------+
345 * |34567 12| -> |34567 1234567 12|
346 * +-----p-c-+ +-------c------p---+
348 memcpy(new_ring, old_ring, u->ring_size * sizeof(*u->ring));
349 memcpy(new_ring + u->ring_size, old_ring,
350 u->ring_size * sizeof(*u->ring));
353 u->ring_size = new_size;
355 spin_unlock_irq(&u->ring_prod_lock);
356 mutex_unlock(&u->ring_cons_mutex);
358 evtchn_free_ring(old_ring);
363 static int evtchn_bind_to_user(struct per_user_data *u, int port)
365 struct user_evtchn *evtchn;
366 struct evtchn_close close;
370 * Ports are never reused, so every caller should pass in a
373 * (Locking not necessary because we haven't registered the
374 * interrupt handler yet, and our caller has already
375 * serialized bind operations.)
378 evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
384 evtchn->enabled = true; /* start enabled */
386 rc = add_evtchn(u, evtchn);
390 rc = evtchn_resize_ring(u);
394 rc = bind_evtchn_to_irqhandler_lateeoi(port, evtchn_interrupt, 0,
399 rc = evtchn_make_refcounted(port);
403 /* bind failed, should close the port now */
405 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
407 del_evtchn(u, evtchn);
411 static void evtchn_unbind_from_user(struct per_user_data *u,
412 struct user_evtchn *evtchn)
414 int irq = irq_from_evtchn(evtchn->port);
418 unbind_from_irqhandler(irq, evtchn);
420 del_evtchn(u, evtchn);
423 static DEFINE_PER_CPU(int, bind_last_selected_cpu);
425 static void evtchn_bind_interdom_next_vcpu(int evtchn)
427 unsigned int selected_cpu, irq;
428 struct irq_desc *desc;
431 irq = irq_from_evtchn(evtchn);
432 desc = irq_to_desc(irq);
437 raw_spin_lock_irqsave(&desc->lock, flags);
438 selected_cpu = this_cpu_read(bind_last_selected_cpu);
439 selected_cpu = cpumask_next_and(selected_cpu,
440 desc->irq_common_data.affinity, cpu_online_mask);
442 if (unlikely(selected_cpu >= nr_cpu_ids))
443 selected_cpu = cpumask_first_and(desc->irq_common_data.affinity,
446 this_cpu_write(bind_last_selected_cpu, selected_cpu);
448 /* unmask expects irqs to be disabled */
449 xen_set_affinity_evtchn(desc, selected_cpu);
450 raw_spin_unlock_irqrestore(&desc->lock, flags);
453 static long evtchn_ioctl(struct file *file,
454 unsigned int cmd, unsigned long arg)
457 struct per_user_data *u = file->private_data;
458 void __user *uarg = (void __user *) arg;
460 /* Prevent bind from racing with unbind */
461 mutex_lock(&u->bind_mutex);
464 case IOCTL_EVTCHN_BIND_VIRQ: {
465 struct ioctl_evtchn_bind_virq bind;
466 struct evtchn_bind_virq bind_virq;
469 if (u->restrict_domid != UNRESTRICTED_DOMID)
473 if (copy_from_user(&bind, uarg, sizeof(bind)))
476 bind_virq.virq = bind.virq;
477 bind_virq.vcpu = xen_vcpu_nr(0);
478 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
483 rc = evtchn_bind_to_user(u, bind_virq.port);
489 case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
490 struct ioctl_evtchn_bind_interdomain bind;
491 struct evtchn_bind_interdomain bind_interdomain;
494 if (copy_from_user(&bind, uarg, sizeof(bind)))
498 if (u->restrict_domid != UNRESTRICTED_DOMID &&
499 u->restrict_domid != bind.remote_domain)
502 bind_interdomain.remote_dom = bind.remote_domain;
503 bind_interdomain.remote_port = bind.remote_port;
504 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
509 rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
511 rc = bind_interdomain.local_port;
512 evtchn_bind_interdom_next_vcpu(rc);
517 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
518 struct ioctl_evtchn_bind_unbound_port bind;
519 struct evtchn_alloc_unbound alloc_unbound;
522 if (u->restrict_domid != UNRESTRICTED_DOMID)
526 if (copy_from_user(&bind, uarg, sizeof(bind)))
529 alloc_unbound.dom = DOMID_SELF;
530 alloc_unbound.remote_dom = bind.remote_domain;
531 rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
536 rc = evtchn_bind_to_user(u, alloc_unbound.port);
538 rc = alloc_unbound.port;
542 case IOCTL_EVTCHN_UNBIND: {
543 struct ioctl_evtchn_unbind unbind;
544 struct user_evtchn *evtchn;
547 if (copy_from_user(&unbind, uarg, sizeof(unbind)))
551 if (unbind.port >= xen_evtchn_nr_channels())
555 evtchn = find_evtchn(u, unbind.port);
559 disable_irq(irq_from_evtchn(unbind.port));
560 evtchn_unbind_from_user(u, evtchn);
565 case IOCTL_EVTCHN_NOTIFY: {
566 struct ioctl_evtchn_notify notify;
567 struct user_evtchn *evtchn;
570 if (copy_from_user(¬ify, uarg, sizeof(notify)))
574 evtchn = find_evtchn(u, notify.port);
576 notify_remote_via_evtchn(notify.port);
582 case IOCTL_EVTCHN_RESET: {
583 /* Initialise the ring to empty. Clear errors. */
584 mutex_lock(&u->ring_cons_mutex);
585 spin_lock_irq(&u->ring_prod_lock);
586 u->ring_cons = u->ring_prod = u->ring_overflow = 0;
587 spin_unlock_irq(&u->ring_prod_lock);
588 mutex_unlock(&u->ring_cons_mutex);
593 case IOCTL_EVTCHN_RESTRICT_DOMID: {
594 struct ioctl_evtchn_restrict_domid ierd;
597 if (u->restrict_domid != UNRESTRICTED_DOMID)
601 if (copy_from_user(&ierd, uarg, sizeof(ierd)))
605 if (ierd.domid == 0 || ierd.domid >= DOMID_FIRST_RESERVED)
608 u->restrict_domid = ierd.domid;
618 mutex_unlock(&u->bind_mutex);
623 static unsigned int evtchn_poll(struct file *file, poll_table *wait)
625 unsigned int mask = POLLOUT | POLLWRNORM;
626 struct per_user_data *u = file->private_data;
628 poll_wait(file, &u->evtchn_wait, wait);
629 if (u->ring_cons != u->ring_prod)
630 mask |= POLLIN | POLLRDNORM;
631 if (u->ring_overflow)
636 static int evtchn_fasync(int fd, struct file *filp, int on)
638 struct per_user_data *u = filp->private_data;
639 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
642 static int evtchn_open(struct inode *inode, struct file *filp)
644 struct per_user_data *u;
646 u = kzalloc(sizeof(*u), GFP_KERNEL);
650 u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
651 if (u->name == NULL) {
656 init_waitqueue_head(&u->evtchn_wait);
658 mutex_init(&u->bind_mutex);
659 mutex_init(&u->ring_cons_mutex);
660 spin_lock_init(&u->ring_prod_lock);
662 u->restrict_domid = UNRESTRICTED_DOMID;
664 filp->private_data = u;
666 return nonseekable_open(inode, filp);
669 static int evtchn_release(struct inode *inode, struct file *filp)
671 struct per_user_data *u = filp->private_data;
672 struct rb_node *node;
674 while ((node = u->evtchns.rb_node)) {
675 struct user_evtchn *evtchn;
677 evtchn = rb_entry(node, struct user_evtchn, node);
678 disable_irq(irq_from_evtchn(evtchn->port));
679 evtchn_unbind_from_user(u, evtchn);
682 evtchn_free_ring(u->ring);
689 static const struct file_operations evtchn_fops = {
690 .owner = THIS_MODULE,
692 .write = evtchn_write,
693 .unlocked_ioctl = evtchn_ioctl,
695 .fasync = evtchn_fasync,
697 .release = evtchn_release,
701 static struct miscdevice evtchn_miscdev = {
702 .minor = MISC_DYNAMIC_MINOR,
703 .name = "xen/evtchn",
704 .fops = &evtchn_fops,
706 static int __init evtchn_init(void)
713 /* Create '/dev/xen/evtchn'. */
714 err = misc_register(&evtchn_miscdev);
716 pr_err("Could not register /dev/xen/evtchn\n");
720 pr_info("Event-channel device installed\n");
725 static void __exit evtchn_cleanup(void)
727 misc_deregister(&evtchn_miscdev);
730 module_init(evtchn_init);
731 module_exit(evtchn_cleanup);
733 MODULE_LICENSE("GPL");