Loading...
1/******************************************************************************
2 * evtchn.c
3 *
4 * Driver for receiving and demuxing event-channel signals.
5 *
6 * Copyright (c) 2004-2005, K A Fraser
7 * Multi-process extensions Copyright (c) 2004, Steven Smith
8 *
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:
14 *
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:
21 *
22 * The above copyright notice and this permission notice shall be included in
23 * all copies or substantial portions of the Software.
24 *
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
31 * IN THE SOFTWARE.
32 */
33
34#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
35
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>
42#include <linux/fs.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>
52#include <linux/mm.h>
53#include <linux/vmalloc.h>
54
55#include <xen/xen.h>
56#include <xen/events.h>
57#include <xen/evtchn.h>
58#include <xen/xen-ops.h>
59#include <asm/xen/hypervisor.h>
60
61struct per_user_data {
62 struct mutex bind_mutex; /* serialize bind/unbind operations */
63 struct rb_root evtchns;
64 unsigned int nr_evtchns;
65
66 /* Notification ring, accessed via /dev/xen/evtchn. */
67 unsigned int ring_size;
68 evtchn_port_t *ring;
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 */
72
73 /* Processes wait on this queue when ring is empty. */
74 wait_queue_head_t evtchn_wait;
75 struct fasync_struct *evtchn_async_queue;
76 const char *name;
77
78 domid_t restrict_domid;
79};
80
81#define UNRESTRICTED_DOMID ((domid_t)-1)
82
83struct user_evtchn {
84 struct rb_node node;
85 struct per_user_data *user;
86 evtchn_port_t port;
87 bool enabled;
88};
89
90static void evtchn_free_ring(evtchn_port_t *ring)
91{
92 kvfree(ring);
93}
94
95static unsigned int evtchn_ring_offset(struct per_user_data *u,
96 unsigned int idx)
97{
98 return idx & (u->ring_size - 1);
99}
100
101static evtchn_port_t *evtchn_ring_entry(struct per_user_data *u,
102 unsigned int idx)
103{
104 return u->ring + evtchn_ring_offset(u, idx);
105}
106
107static int add_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
108{
109 struct rb_node **new = &(u->evtchns.rb_node), *parent = NULL;
110
111 u->nr_evtchns++;
112
113 while (*new) {
114 struct user_evtchn *this;
115
116 this = rb_entry(*new, struct user_evtchn, node);
117
118 parent = *new;
119 if (this->port < evtchn->port)
120 new = &((*new)->rb_left);
121 else if (this->port > evtchn->port)
122 new = &((*new)->rb_right);
123 else
124 return -EEXIST;
125 }
126
127 /* Add new node and rebalance tree. */
128 rb_link_node(&evtchn->node, parent, new);
129 rb_insert_color(&evtchn->node, &u->evtchns);
130
131 return 0;
132}
133
134static void del_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
135{
136 u->nr_evtchns--;
137 rb_erase(&evtchn->node, &u->evtchns);
138 kfree(evtchn);
139}
140
141static struct user_evtchn *find_evtchn(struct per_user_data *u,
142 evtchn_port_t port)
143{
144 struct rb_node *node = u->evtchns.rb_node;
145
146 while (node) {
147 struct user_evtchn *evtchn;
148
149 evtchn = rb_entry(node, struct user_evtchn, node);
150
151 if (evtchn->port < port)
152 node = node->rb_left;
153 else if (evtchn->port > port)
154 node = node->rb_right;
155 else
156 return evtchn;
157 }
158 return NULL;
159}
160
161static irqreturn_t evtchn_interrupt(int irq, void *data)
162{
163 struct user_evtchn *evtchn = data;
164 struct per_user_data *u = evtchn->user;
165 unsigned int prod, cons;
166
167 WARN(!evtchn->enabled,
168 "Interrupt for port %u, but apparently not enabled; per-user %p\n",
169 evtchn->port, u);
170
171 evtchn->enabled = false;
172
173 spin_lock(&u->ring_prod_lock);
174
175 prod = READ_ONCE(u->ring_prod);
176 cons = READ_ONCE(u->ring_cons);
177
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);
182 if (cons == prod) {
183 wake_up_interruptible(&u->evtchn_wait);
184 kill_fasync(&u->evtchn_async_queue,
185 SIGIO, POLL_IN);
186 }
187 } else
188 u->ring_overflow = 1;
189
190 spin_unlock(&u->ring_prod_lock);
191
192 return IRQ_HANDLED;
193}
194
195static ssize_t evtchn_read(struct file *file, char __user *buf,
196 size_t count, loff_t *ppos)
197{
198 int rc;
199 unsigned int c, p, bytes1 = 0, bytes2 = 0;
200 struct per_user_data *u = file->private_data;
201
202 /* Whole number of ports. */
203 count &= ~(sizeof(evtchn_port_t)-1);
204
205 if (count == 0)
206 return 0;
207
208 if (count > PAGE_SIZE)
209 count = PAGE_SIZE;
210
211 for (;;) {
212 mutex_lock(&u->ring_cons_mutex);
213
214 rc = -EFBIG;
215 if (u->ring_overflow)
216 goto unlock_out;
217
218 c = READ_ONCE(u->ring_cons);
219 p = READ_ONCE(u->ring_prod);
220 if (c != p)
221 break;
222
223 mutex_unlock(&u->ring_cons_mutex);
224
225 if (file->f_flags & O_NONBLOCK)
226 return -EAGAIN;
227
228 rc = wait_event_interruptible(u->evtchn_wait,
229 READ_ONCE(u->ring_cons) != READ_ONCE(u->ring_prod));
230 if (rc)
231 return rc;
232 }
233
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);
239 } else {
240 bytes1 = (p - c) * sizeof(evtchn_port_t);
241 bytes2 = 0;
242 }
243
244 /* Truncate chunks according to caller's maximum byte count. */
245 if (bytes1 > count) {
246 bytes1 = count;
247 bytes2 = 0;
248 } else if ((bytes1 + bytes2) > count) {
249 bytes2 = count - bytes1;
250 }
251
252 rc = -EFAULT;
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) ||
255 ((bytes2 != 0) &&
256 copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
257 goto unlock_out;
258
259 WRITE_ONCE(u->ring_cons, c + (bytes1 + bytes2) / sizeof(evtchn_port_t));
260 rc = bytes1 + bytes2;
261
262 unlock_out:
263 mutex_unlock(&u->ring_cons_mutex);
264 return rc;
265}
266
267static ssize_t evtchn_write(struct file *file, const char __user *buf,
268 size_t count, loff_t *ppos)
269{
270 int rc, i;
271 evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
272 struct per_user_data *u = file->private_data;
273
274 if (kbuf == NULL)
275 return -ENOMEM;
276
277 /* Whole number of ports. */
278 count &= ~(sizeof(evtchn_port_t)-1);
279
280 rc = 0;
281 if (count == 0)
282 goto out;
283
284 if (count > PAGE_SIZE)
285 count = PAGE_SIZE;
286
287 rc = -EFAULT;
288 if (copy_from_user(kbuf, buf, count) != 0)
289 goto out;
290
291 mutex_lock(&u->bind_mutex);
292
293 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
294 evtchn_port_t port = kbuf[i];
295 struct user_evtchn *evtchn;
296
297 evtchn = find_evtchn(u, port);
298 if (evtchn && !evtchn->enabled) {
299 evtchn->enabled = true;
300 xen_irq_lateeoi(irq_from_evtchn(port), 0);
301 }
302 }
303
304 mutex_unlock(&u->bind_mutex);
305
306 rc = count;
307
308 out:
309 free_page((unsigned long)kbuf);
310 return rc;
311}
312
313static int evtchn_resize_ring(struct per_user_data *u)
314{
315 unsigned int new_size;
316 evtchn_port_t *new_ring, *old_ring;
317
318 /*
319 * Ensure the ring is large enough to capture all possible
320 * events. i.e., one free slot for each bound event.
321 */
322 if (u->nr_evtchns <= u->ring_size)
323 return 0;
324
325 if (u->ring_size == 0)
326 new_size = 64;
327 else
328 new_size = 2 * u->ring_size;
329
330 new_ring = kvmalloc_array(new_size, sizeof(*new_ring), GFP_KERNEL);
331 if (!new_ring)
332 return -ENOMEM;
333
334 old_ring = u->ring;
335
336 /*
337 * Access to the ring contents is serialized by either the
338 * prod /or/ cons lock so take both when resizing.
339 */
340 mutex_lock(&u->ring_cons_mutex);
341 spin_lock_irq(&u->ring_prod_lock);
342
343 /*
344 * Copy the old ring contents to the new ring.
345 *
346 * To take care of wrapping, a full ring, and the new index
347 * pointing into the second half, simply copy the old contents
348 * twice.
349 *
350 * +---------+ +------------------+
351 * |34567 12| -> |34567 1234567 12|
352 * +-----p-c-+ +-------c------p---+
353 */
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));
357
358 u->ring = new_ring;
359 u->ring_size = new_size;
360
361 spin_unlock_irq(&u->ring_prod_lock);
362 mutex_unlock(&u->ring_cons_mutex);
363
364 evtchn_free_ring(old_ring);
365
366 return 0;
367}
368
369static int evtchn_bind_to_user(struct per_user_data *u, evtchn_port_t port)
370{
371 struct user_evtchn *evtchn;
372 struct evtchn_close close;
373 int rc = 0;
374
375 /*
376 * Ports are never reused, so every caller should pass in a
377 * unique port.
378 *
379 * (Locking not necessary because we haven't registered the
380 * interrupt handler yet, and our caller has already
381 * serialized bind operations.)
382 */
383
384 evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
385 if (!evtchn)
386 return -ENOMEM;
387
388 evtchn->user = u;
389 evtchn->port = port;
390 evtchn->enabled = true; /* start enabled */
391
392 rc = add_evtchn(u, evtchn);
393 if (rc < 0)
394 goto err;
395
396 rc = evtchn_resize_ring(u);
397 if (rc < 0)
398 goto err;
399
400 rc = bind_evtchn_to_irqhandler_lateeoi(port, evtchn_interrupt, 0,
401 u->name, evtchn);
402 if (rc < 0)
403 goto err;
404
405 rc = evtchn_make_refcounted(port);
406 return rc;
407
408err:
409 /* bind failed, should close the port now */
410 close.port = port;
411 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
412 BUG();
413 del_evtchn(u, evtchn);
414 return rc;
415}
416
417static void evtchn_unbind_from_user(struct per_user_data *u,
418 struct user_evtchn *evtchn)
419{
420 int irq = irq_from_evtchn(evtchn->port);
421
422 BUG_ON(irq < 0);
423
424 unbind_from_irqhandler(irq, evtchn);
425
426 del_evtchn(u, evtchn);
427}
428
429static long evtchn_ioctl(struct file *file,
430 unsigned int cmd, unsigned long arg)
431{
432 int rc;
433 struct per_user_data *u = file->private_data;
434 void __user *uarg = (void __user *) arg;
435
436 /* Prevent bind from racing with unbind */
437 mutex_lock(&u->bind_mutex);
438
439 switch (cmd) {
440 case IOCTL_EVTCHN_BIND_VIRQ: {
441 struct ioctl_evtchn_bind_virq bind;
442 struct evtchn_bind_virq bind_virq;
443
444 rc = -EACCES;
445 if (u->restrict_domid != UNRESTRICTED_DOMID)
446 break;
447
448 rc = -EFAULT;
449 if (copy_from_user(&bind, uarg, sizeof(bind)))
450 break;
451
452 bind_virq.virq = bind.virq;
453 bind_virq.vcpu = xen_vcpu_nr(0);
454 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
455 &bind_virq);
456 if (rc != 0)
457 break;
458
459 rc = evtchn_bind_to_user(u, bind_virq.port);
460 if (rc == 0)
461 rc = bind_virq.port;
462 break;
463 }
464
465 case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
466 struct ioctl_evtchn_bind_interdomain bind;
467 struct evtchn_bind_interdomain bind_interdomain;
468
469 rc = -EFAULT;
470 if (copy_from_user(&bind, uarg, sizeof(bind)))
471 break;
472
473 rc = -EACCES;
474 if (u->restrict_domid != UNRESTRICTED_DOMID &&
475 u->restrict_domid != bind.remote_domain)
476 break;
477
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,
481 &bind_interdomain);
482 if (rc != 0)
483 break;
484
485 rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
486 if (rc == 0)
487 rc = bind_interdomain.local_port;
488 break;
489 }
490
491 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
492 struct ioctl_evtchn_bind_unbound_port bind;
493 struct evtchn_alloc_unbound alloc_unbound;
494
495 rc = -EACCES;
496 if (u->restrict_domid != UNRESTRICTED_DOMID)
497 break;
498
499 rc = -EFAULT;
500 if (copy_from_user(&bind, uarg, sizeof(bind)))
501 break;
502
503 alloc_unbound.dom = DOMID_SELF;
504 alloc_unbound.remote_dom = bind.remote_domain;
505 rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
506 &alloc_unbound);
507 if (rc != 0)
508 break;
509
510 rc = evtchn_bind_to_user(u, alloc_unbound.port);
511 if (rc == 0)
512 rc = alloc_unbound.port;
513 break;
514 }
515
516 case IOCTL_EVTCHN_UNBIND: {
517 struct ioctl_evtchn_unbind unbind;
518 struct user_evtchn *evtchn;
519
520 rc = -EFAULT;
521 if (copy_from_user(&unbind, uarg, sizeof(unbind)))
522 break;
523
524 rc = -EINVAL;
525 if (unbind.port >= xen_evtchn_nr_channels())
526 break;
527
528 rc = -ENOTCONN;
529 evtchn = find_evtchn(u, unbind.port);
530 if (!evtchn)
531 break;
532
533 disable_irq(irq_from_evtchn(unbind.port));
534 evtchn_unbind_from_user(u, evtchn);
535 rc = 0;
536 break;
537 }
538
539 case IOCTL_EVTCHN_NOTIFY: {
540 struct ioctl_evtchn_notify notify;
541 struct user_evtchn *evtchn;
542
543 rc = -EFAULT;
544 if (copy_from_user(¬ify, uarg, sizeof(notify)))
545 break;
546
547 rc = -ENOTCONN;
548 evtchn = find_evtchn(u, notify.port);
549 if (evtchn) {
550 notify_remote_via_evtchn(notify.port);
551 rc = 0;
552 }
553 break;
554 }
555
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);
565 rc = 0;
566 break;
567 }
568
569 case IOCTL_EVTCHN_RESTRICT_DOMID: {
570 struct ioctl_evtchn_restrict_domid ierd;
571
572 rc = -EACCES;
573 if (u->restrict_domid != UNRESTRICTED_DOMID)
574 break;
575
576 rc = -EFAULT;
577 if (copy_from_user(&ierd, uarg, sizeof(ierd)))
578 break;
579
580 rc = -EINVAL;
581 if (ierd.domid == 0 || ierd.domid >= DOMID_FIRST_RESERVED)
582 break;
583
584 u->restrict_domid = ierd.domid;
585 rc = 0;
586
587 break;
588 }
589
590 default:
591 rc = -ENOSYS;
592 break;
593 }
594 mutex_unlock(&u->bind_mutex);
595
596 return rc;
597}
598
599static __poll_t evtchn_poll(struct file *file, poll_table *wait)
600{
601 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
602 struct per_user_data *u = file->private_data;
603
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)
608 mask = EPOLLERR;
609 return mask;
610}
611
612static int evtchn_fasync(int fd, struct file *filp, int on)
613{
614 struct per_user_data *u = filp->private_data;
615 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
616}
617
618static int evtchn_open(struct inode *inode, struct file *filp)
619{
620 struct per_user_data *u;
621
622 u = kzalloc(sizeof(*u), GFP_KERNEL);
623 if (u == NULL)
624 return -ENOMEM;
625
626 u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
627 if (u->name == NULL) {
628 kfree(u);
629 return -ENOMEM;
630 }
631
632 init_waitqueue_head(&u->evtchn_wait);
633
634 mutex_init(&u->bind_mutex);
635 mutex_init(&u->ring_cons_mutex);
636 spin_lock_init(&u->ring_prod_lock);
637
638 u->restrict_domid = UNRESTRICTED_DOMID;
639
640 filp->private_data = u;
641
642 return stream_open(inode, filp);
643}
644
645static int evtchn_release(struct inode *inode, struct file *filp)
646{
647 struct per_user_data *u = filp->private_data;
648 struct rb_node *node;
649
650 while ((node = u->evtchns.rb_node)) {
651 struct user_evtchn *evtchn;
652
653 evtchn = rb_entry(node, struct user_evtchn, node);
654 disable_irq(irq_from_evtchn(evtchn->port));
655 evtchn_unbind_from_user(u, evtchn);
656 }
657
658 evtchn_free_ring(u->ring);
659 kfree(u->name);
660 kfree(u);
661
662 return 0;
663}
664
665static const struct file_operations evtchn_fops = {
666 .owner = THIS_MODULE,
667 .read = evtchn_read,
668 .write = evtchn_write,
669 .unlocked_ioctl = evtchn_ioctl,
670 .poll = evtchn_poll,
671 .fasync = evtchn_fasync,
672 .open = evtchn_open,
673 .release = evtchn_release,
674 .llseek = no_llseek,
675};
676
677static struct miscdevice evtchn_miscdev = {
678 .minor = MISC_DYNAMIC_MINOR,
679 .name = "xen/evtchn",
680 .fops = &evtchn_fops,
681};
682static int __init evtchn_init(void)
683{
684 int err;
685
686 if (!xen_domain())
687 return -ENODEV;
688
689 /* Create '/dev/xen/evtchn'. */
690 err = misc_register(&evtchn_miscdev);
691 if (err != 0) {
692 pr_err("Could not register /dev/xen/evtchn\n");
693 return err;
694 }
695
696 pr_info("Event-channel device installed\n");
697
698 return 0;
699}
700
701static void __exit evtchn_cleanup(void)
702{
703 misc_deregister(&evtchn_miscdev);
704}
705
706module_init(evtchn_init);
707module_exit(evtchn_cleanup);
708
709MODULE_LICENSE("GPL");
1/******************************************************************************
2 * evtchn.c
3 *
4 * Driver for receiving and demuxing event-channel signals.
5 *
6 * Copyright (c) 2004-2005, K A Fraser
7 * Multi-process extensions Copyright (c) 2004, Steven Smith
8 *
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:
14 *
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:
21 *
22 * The above copyright notice and this permission notice shall be included in
23 * all copies or substantial portions of the Software.
24 *
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
31 * IN THE SOFTWARE.
32 */
33
34#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
35
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>
42#include <linux/fs.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>
52#include <linux/mm.h>
53#include <linux/vmalloc.h>
54
55#include <xen/xen.h>
56#include <xen/events.h>
57#include <xen/evtchn.h>
58#include <xen/xen-ops.h>
59#include <asm/xen/hypervisor.h>
60
61struct per_user_data {
62 struct mutex bind_mutex; /* serialize bind/unbind operations */
63 struct rb_root evtchns;
64 unsigned int nr_evtchns;
65
66 /* Notification ring, accessed via /dev/xen/evtchn. */
67 unsigned int ring_size;
68 evtchn_port_t *ring;
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 */
72
73 /* Processes wait on this queue when ring is empty. */
74 wait_queue_head_t evtchn_wait;
75 struct fasync_struct *evtchn_async_queue;
76 const char *name;
77
78 domid_t restrict_domid;
79};
80
81#define UNRESTRICTED_DOMID ((domid_t)-1)
82
83struct user_evtchn {
84 struct rb_node node;
85 struct per_user_data *user;
86 unsigned port;
87 bool enabled;
88};
89
90static evtchn_port_t *evtchn_alloc_ring(unsigned int size)
91{
92 evtchn_port_t *ring;
93 size_t s = size * sizeof(*ring);
94
95 ring = kmalloc(s, GFP_KERNEL);
96 if (!ring)
97 ring = vmalloc(s);
98
99 return ring;
100}
101
102static void evtchn_free_ring(evtchn_port_t *ring)
103{
104 kvfree(ring);
105}
106
107static unsigned int evtchn_ring_offset(struct per_user_data *u,
108 unsigned int idx)
109{
110 return idx & (u->ring_size - 1);
111}
112
113static evtchn_port_t *evtchn_ring_entry(struct per_user_data *u,
114 unsigned int idx)
115{
116 return u->ring + evtchn_ring_offset(u, idx);
117}
118
119static int add_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
120{
121 struct rb_node **new = &(u->evtchns.rb_node), *parent = NULL;
122
123 u->nr_evtchns++;
124
125 while (*new) {
126 struct user_evtchn *this;
127
128 this = rb_entry(*new, struct user_evtchn, node);
129
130 parent = *new;
131 if (this->port < evtchn->port)
132 new = &((*new)->rb_left);
133 else if (this->port > evtchn->port)
134 new = &((*new)->rb_right);
135 else
136 return -EEXIST;
137 }
138
139 /* Add new node and rebalance tree. */
140 rb_link_node(&evtchn->node, parent, new);
141 rb_insert_color(&evtchn->node, &u->evtchns);
142
143 return 0;
144}
145
146static void del_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
147{
148 u->nr_evtchns--;
149 rb_erase(&evtchn->node, &u->evtchns);
150 kfree(evtchn);
151}
152
153static struct user_evtchn *find_evtchn(struct per_user_data *u, unsigned port)
154{
155 struct rb_node *node = u->evtchns.rb_node;
156
157 while (node) {
158 struct user_evtchn *evtchn;
159
160 evtchn = rb_entry(node, struct user_evtchn, node);
161
162 if (evtchn->port < port)
163 node = node->rb_left;
164 else if (evtchn->port > port)
165 node = node->rb_right;
166 else
167 return evtchn;
168 }
169 return NULL;
170}
171
172static irqreturn_t evtchn_interrupt(int irq, void *data)
173{
174 struct user_evtchn *evtchn = data;
175 struct per_user_data *u = evtchn->user;
176
177 WARN(!evtchn->enabled,
178 "Interrupt for port %d, but apparently not enabled; per-user %p\n",
179 evtchn->port, u);
180
181 disable_irq_nosync(irq);
182 evtchn->enabled = false;
183
184 spin_lock(&u->ring_prod_lock);
185
186 if ((u->ring_prod - u->ring_cons) < u->ring_size) {
187 *evtchn_ring_entry(u, u->ring_prod) = evtchn->port;
188 wmb(); /* Ensure ring contents visible */
189 if (u->ring_cons == u->ring_prod++) {
190 wake_up_interruptible(&u->evtchn_wait);
191 kill_fasync(&u->evtchn_async_queue,
192 SIGIO, POLL_IN);
193 }
194 } else
195 u->ring_overflow = 1;
196
197 spin_unlock(&u->ring_prod_lock);
198
199 return IRQ_HANDLED;
200}
201
202static ssize_t evtchn_read(struct file *file, char __user *buf,
203 size_t count, loff_t *ppos)
204{
205 int rc;
206 unsigned int c, p, bytes1 = 0, bytes2 = 0;
207 struct per_user_data *u = file->private_data;
208
209 /* Whole number of ports. */
210 count &= ~(sizeof(evtchn_port_t)-1);
211
212 if (count == 0)
213 return 0;
214
215 if (count > PAGE_SIZE)
216 count = PAGE_SIZE;
217
218 for (;;) {
219 mutex_lock(&u->ring_cons_mutex);
220
221 rc = -EFBIG;
222 if (u->ring_overflow)
223 goto unlock_out;
224
225 c = u->ring_cons;
226 p = u->ring_prod;
227 if (c != p)
228 break;
229
230 mutex_unlock(&u->ring_cons_mutex);
231
232 if (file->f_flags & O_NONBLOCK)
233 return -EAGAIN;
234
235 rc = wait_event_interruptible(u->evtchn_wait,
236 u->ring_cons != u->ring_prod);
237 if (rc)
238 return rc;
239 }
240
241 /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
242 if (((c ^ p) & u->ring_size) != 0) {
243 bytes1 = (u->ring_size - evtchn_ring_offset(u, c)) *
244 sizeof(evtchn_port_t);
245 bytes2 = evtchn_ring_offset(u, p) * sizeof(evtchn_port_t);
246 } else {
247 bytes1 = (p - c) * sizeof(evtchn_port_t);
248 bytes2 = 0;
249 }
250
251 /* Truncate chunks according to caller's maximum byte count. */
252 if (bytes1 > count) {
253 bytes1 = count;
254 bytes2 = 0;
255 } else if ((bytes1 + bytes2) > count) {
256 bytes2 = count - bytes1;
257 }
258
259 rc = -EFAULT;
260 rmb(); /* Ensure that we see the port before we copy it. */
261 if (copy_to_user(buf, evtchn_ring_entry(u, c), bytes1) ||
262 ((bytes2 != 0) &&
263 copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
264 goto unlock_out;
265
266 u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t);
267 rc = bytes1 + bytes2;
268
269 unlock_out:
270 mutex_unlock(&u->ring_cons_mutex);
271 return rc;
272}
273
274static ssize_t evtchn_write(struct file *file, const char __user *buf,
275 size_t count, loff_t *ppos)
276{
277 int rc, i;
278 evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
279 struct per_user_data *u = file->private_data;
280
281 if (kbuf == NULL)
282 return -ENOMEM;
283
284 /* Whole number of ports. */
285 count &= ~(sizeof(evtchn_port_t)-1);
286
287 rc = 0;
288 if (count == 0)
289 goto out;
290
291 if (count > PAGE_SIZE)
292 count = PAGE_SIZE;
293
294 rc = -EFAULT;
295 if (copy_from_user(kbuf, buf, count) != 0)
296 goto out;
297
298 mutex_lock(&u->bind_mutex);
299
300 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
301 unsigned port = kbuf[i];
302 struct user_evtchn *evtchn;
303
304 evtchn = find_evtchn(u, port);
305 if (evtchn && !evtchn->enabled) {
306 evtchn->enabled = true;
307 enable_irq(irq_from_evtchn(port));
308 }
309 }
310
311 mutex_unlock(&u->bind_mutex);
312
313 rc = count;
314
315 out:
316 free_page((unsigned long)kbuf);
317 return rc;
318}
319
320static int evtchn_resize_ring(struct per_user_data *u)
321{
322 unsigned int new_size;
323 evtchn_port_t *new_ring, *old_ring;
324
325 /*
326 * Ensure the ring is large enough to capture all possible
327 * events. i.e., one free slot for each bound event.
328 */
329 if (u->nr_evtchns <= u->ring_size)
330 return 0;
331
332 if (u->ring_size == 0)
333 new_size = 64;
334 else
335 new_size = 2 * u->ring_size;
336
337 new_ring = evtchn_alloc_ring(new_size);
338 if (!new_ring)
339 return -ENOMEM;
340
341 old_ring = u->ring;
342
343 /*
344 * Access to the ring contents is serialized by either the
345 * prod /or/ cons lock so take both when resizing.
346 */
347 mutex_lock(&u->ring_cons_mutex);
348 spin_lock_irq(&u->ring_prod_lock);
349
350 /*
351 * Copy the old ring contents to the new ring.
352 *
353 * To take care of wrapping, a full ring, and the new index
354 * pointing into the second half, simply copy the old contents
355 * twice.
356 *
357 * +---------+ +------------------+
358 * |34567 12| -> |34567 1234567 12|
359 * +-----p-c-+ +-------c------p---+
360 */
361 memcpy(new_ring, old_ring, u->ring_size * sizeof(*u->ring));
362 memcpy(new_ring + u->ring_size, old_ring,
363 u->ring_size * sizeof(*u->ring));
364
365 u->ring = new_ring;
366 u->ring_size = new_size;
367
368 spin_unlock_irq(&u->ring_prod_lock);
369 mutex_unlock(&u->ring_cons_mutex);
370
371 evtchn_free_ring(old_ring);
372
373 return 0;
374}
375
376static int evtchn_bind_to_user(struct per_user_data *u, int port)
377{
378 struct user_evtchn *evtchn;
379 struct evtchn_close close;
380 int rc = 0;
381
382 /*
383 * Ports are never reused, so every caller should pass in a
384 * unique port.
385 *
386 * (Locking not necessary because we haven't registered the
387 * interrupt handler yet, and our caller has already
388 * serialized bind operations.)
389 */
390
391 evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
392 if (!evtchn)
393 return -ENOMEM;
394
395 evtchn->user = u;
396 evtchn->port = port;
397 evtchn->enabled = true; /* start enabled */
398
399 rc = add_evtchn(u, evtchn);
400 if (rc < 0)
401 goto err;
402
403 rc = evtchn_resize_ring(u);
404 if (rc < 0)
405 goto err;
406
407 rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, 0,
408 u->name, evtchn);
409 if (rc < 0)
410 goto err;
411
412 rc = evtchn_make_refcounted(port);
413 return rc;
414
415err:
416 /* bind failed, should close the port now */
417 close.port = port;
418 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
419 BUG();
420 del_evtchn(u, evtchn);
421 return rc;
422}
423
424static void evtchn_unbind_from_user(struct per_user_data *u,
425 struct user_evtchn *evtchn)
426{
427 int irq = irq_from_evtchn(evtchn->port);
428
429 BUG_ON(irq < 0);
430
431 unbind_from_irqhandler(irq, evtchn);
432
433 del_evtchn(u, evtchn);
434}
435
436static long evtchn_ioctl(struct file *file,
437 unsigned int cmd, unsigned long arg)
438{
439 int rc;
440 struct per_user_data *u = file->private_data;
441 void __user *uarg = (void __user *) arg;
442
443 /* Prevent bind from racing with unbind */
444 mutex_lock(&u->bind_mutex);
445
446 switch (cmd) {
447 case IOCTL_EVTCHN_BIND_VIRQ: {
448 struct ioctl_evtchn_bind_virq bind;
449 struct evtchn_bind_virq bind_virq;
450
451 rc = -EACCES;
452 if (u->restrict_domid != UNRESTRICTED_DOMID)
453 break;
454
455 rc = -EFAULT;
456 if (copy_from_user(&bind, uarg, sizeof(bind)))
457 break;
458
459 bind_virq.virq = bind.virq;
460 bind_virq.vcpu = xen_vcpu_nr(0);
461 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
462 &bind_virq);
463 if (rc != 0)
464 break;
465
466 rc = evtchn_bind_to_user(u, bind_virq.port);
467 if (rc == 0)
468 rc = bind_virq.port;
469 break;
470 }
471
472 case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
473 struct ioctl_evtchn_bind_interdomain bind;
474 struct evtchn_bind_interdomain bind_interdomain;
475
476 rc = -EFAULT;
477 if (copy_from_user(&bind, uarg, sizeof(bind)))
478 break;
479
480 rc = -EACCES;
481 if (u->restrict_domid != UNRESTRICTED_DOMID &&
482 u->restrict_domid != bind.remote_domain)
483 break;
484
485 bind_interdomain.remote_dom = bind.remote_domain;
486 bind_interdomain.remote_port = bind.remote_port;
487 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
488 &bind_interdomain);
489 if (rc != 0)
490 break;
491
492 rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
493 if (rc == 0)
494 rc = bind_interdomain.local_port;
495 break;
496 }
497
498 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
499 struct ioctl_evtchn_bind_unbound_port bind;
500 struct evtchn_alloc_unbound alloc_unbound;
501
502 rc = -EACCES;
503 if (u->restrict_domid != UNRESTRICTED_DOMID)
504 break;
505
506 rc = -EFAULT;
507 if (copy_from_user(&bind, uarg, sizeof(bind)))
508 break;
509
510 alloc_unbound.dom = DOMID_SELF;
511 alloc_unbound.remote_dom = bind.remote_domain;
512 rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
513 &alloc_unbound);
514 if (rc != 0)
515 break;
516
517 rc = evtchn_bind_to_user(u, alloc_unbound.port);
518 if (rc == 0)
519 rc = alloc_unbound.port;
520 break;
521 }
522
523 case IOCTL_EVTCHN_UNBIND: {
524 struct ioctl_evtchn_unbind unbind;
525 struct user_evtchn *evtchn;
526
527 rc = -EFAULT;
528 if (copy_from_user(&unbind, uarg, sizeof(unbind)))
529 break;
530
531 rc = -EINVAL;
532 if (unbind.port >= xen_evtchn_nr_channels())
533 break;
534
535 rc = -ENOTCONN;
536 evtchn = find_evtchn(u, unbind.port);
537 if (!evtchn)
538 break;
539
540 disable_irq(irq_from_evtchn(unbind.port));
541 evtchn_unbind_from_user(u, evtchn);
542 rc = 0;
543 break;
544 }
545
546 case IOCTL_EVTCHN_NOTIFY: {
547 struct ioctl_evtchn_notify notify;
548 struct user_evtchn *evtchn;
549
550 rc = -EFAULT;
551 if (copy_from_user(¬ify, uarg, sizeof(notify)))
552 break;
553
554 rc = -ENOTCONN;
555 evtchn = find_evtchn(u, notify.port);
556 if (evtchn) {
557 notify_remote_via_evtchn(notify.port);
558 rc = 0;
559 }
560 break;
561 }
562
563 case IOCTL_EVTCHN_RESET: {
564 /* Initialise the ring to empty. Clear errors. */
565 mutex_lock(&u->ring_cons_mutex);
566 spin_lock_irq(&u->ring_prod_lock);
567 u->ring_cons = u->ring_prod = u->ring_overflow = 0;
568 spin_unlock_irq(&u->ring_prod_lock);
569 mutex_unlock(&u->ring_cons_mutex);
570 rc = 0;
571 break;
572 }
573
574 case IOCTL_EVTCHN_RESTRICT_DOMID: {
575 struct ioctl_evtchn_restrict_domid ierd;
576
577 rc = -EACCES;
578 if (u->restrict_domid != UNRESTRICTED_DOMID)
579 break;
580
581 rc = -EFAULT;
582 if (copy_from_user(&ierd, uarg, sizeof(ierd)))
583 break;
584
585 rc = -EINVAL;
586 if (ierd.domid == 0 || ierd.domid >= DOMID_FIRST_RESERVED)
587 break;
588
589 u->restrict_domid = ierd.domid;
590 rc = 0;
591
592 break;
593 }
594
595 default:
596 rc = -ENOSYS;
597 break;
598 }
599 mutex_unlock(&u->bind_mutex);
600
601 return rc;
602}
603
604static unsigned int evtchn_poll(struct file *file, poll_table *wait)
605{
606 unsigned int mask = POLLOUT | POLLWRNORM;
607 struct per_user_data *u = file->private_data;
608
609 poll_wait(file, &u->evtchn_wait, wait);
610 if (u->ring_cons != u->ring_prod)
611 mask |= POLLIN | POLLRDNORM;
612 if (u->ring_overflow)
613 mask = POLLERR;
614 return mask;
615}
616
617static int evtchn_fasync(int fd, struct file *filp, int on)
618{
619 struct per_user_data *u = filp->private_data;
620 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
621}
622
623static int evtchn_open(struct inode *inode, struct file *filp)
624{
625 struct per_user_data *u;
626
627 u = kzalloc(sizeof(*u), GFP_KERNEL);
628 if (u == NULL)
629 return -ENOMEM;
630
631 u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
632 if (u->name == NULL) {
633 kfree(u);
634 return -ENOMEM;
635 }
636
637 init_waitqueue_head(&u->evtchn_wait);
638
639 mutex_init(&u->bind_mutex);
640 mutex_init(&u->ring_cons_mutex);
641 spin_lock_init(&u->ring_prod_lock);
642
643 u->restrict_domid = UNRESTRICTED_DOMID;
644
645 filp->private_data = u;
646
647 return nonseekable_open(inode, filp);
648}
649
650static int evtchn_release(struct inode *inode, struct file *filp)
651{
652 struct per_user_data *u = filp->private_data;
653 struct rb_node *node;
654
655 while ((node = u->evtchns.rb_node)) {
656 struct user_evtchn *evtchn;
657
658 evtchn = rb_entry(node, struct user_evtchn, node);
659 disable_irq(irq_from_evtchn(evtchn->port));
660 evtchn_unbind_from_user(u, evtchn);
661 }
662
663 evtchn_free_ring(u->ring);
664 kfree(u->name);
665 kfree(u);
666
667 return 0;
668}
669
670static const struct file_operations evtchn_fops = {
671 .owner = THIS_MODULE,
672 .read = evtchn_read,
673 .write = evtchn_write,
674 .unlocked_ioctl = evtchn_ioctl,
675 .poll = evtchn_poll,
676 .fasync = evtchn_fasync,
677 .open = evtchn_open,
678 .release = evtchn_release,
679 .llseek = no_llseek,
680};
681
682static struct miscdevice evtchn_miscdev = {
683 .minor = MISC_DYNAMIC_MINOR,
684 .name = "xen/evtchn",
685 .fops = &evtchn_fops,
686};
687static int __init evtchn_init(void)
688{
689 int err;
690
691 if (!xen_domain())
692 return -ENODEV;
693
694 /* Create '/dev/xen/evtchn'. */
695 err = misc_register(&evtchn_miscdev);
696 if (err != 0) {
697 pr_err("Could not register /dev/xen/evtchn\n");
698 return err;
699 }
700
701 pr_info("Event-channel device installed\n");
702
703 return 0;
704}
705
706static void __exit evtchn_cleanup(void)
707{
708 misc_deregister(&evtchn_miscdev);
709}
710
711module_init(evtchn_init);
712module_exit(evtchn_cleanup);
713
714MODULE_LICENSE("GPL");