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 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, unsigned port)
142{
143 struct rb_node *node = u->evtchns.rb_node;
144
145 while (node) {
146 struct user_evtchn *evtchn;
147
148 evtchn = rb_entry(node, struct user_evtchn, node);
149
150 if (evtchn->port < port)
151 node = node->rb_left;
152 else if (evtchn->port > port)
153 node = node->rb_right;
154 else
155 return evtchn;
156 }
157 return NULL;
158}
159
160static irqreturn_t evtchn_interrupt(int irq, void *data)
161{
162 struct user_evtchn *evtchn = data;
163 struct per_user_data *u = evtchn->user;
164
165 WARN(!evtchn->enabled,
166 "Interrupt for port %d, but apparently not enabled; per-user %p\n",
167 evtchn->port, u);
168
169 disable_irq_nosync(irq);
170 evtchn->enabled = false;
171
172 spin_lock(&u->ring_prod_lock);
173
174 if ((u->ring_prod - u->ring_cons) < u->ring_size) {
175 *evtchn_ring_entry(u, u->ring_prod) = evtchn->port;
176 wmb(); /* Ensure ring contents visible */
177 if (u->ring_cons == u->ring_prod++) {
178 wake_up_interruptible(&u->evtchn_wait);
179 kill_fasync(&u->evtchn_async_queue,
180 SIGIO, POLL_IN);
181 }
182 } else
183 u->ring_overflow = 1;
184
185 spin_unlock(&u->ring_prod_lock);
186
187 return IRQ_HANDLED;
188}
189
190static ssize_t evtchn_read(struct file *file, char __user *buf,
191 size_t count, loff_t *ppos)
192{
193 int rc;
194 unsigned int c, p, bytes1 = 0, bytes2 = 0;
195 struct per_user_data *u = file->private_data;
196
197 /* Whole number of ports. */
198 count &= ~(sizeof(evtchn_port_t)-1);
199
200 if (count == 0)
201 return 0;
202
203 if (count > PAGE_SIZE)
204 count = PAGE_SIZE;
205
206 for (;;) {
207 mutex_lock(&u->ring_cons_mutex);
208
209 rc = -EFBIG;
210 if (u->ring_overflow)
211 goto unlock_out;
212
213 c = u->ring_cons;
214 p = u->ring_prod;
215 if (c != p)
216 break;
217
218 mutex_unlock(&u->ring_cons_mutex);
219
220 if (file->f_flags & O_NONBLOCK)
221 return -EAGAIN;
222
223 rc = wait_event_interruptible(u->evtchn_wait,
224 u->ring_cons != u->ring_prod);
225 if (rc)
226 return rc;
227 }
228
229 /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
230 if (((c ^ p) & u->ring_size) != 0) {
231 bytes1 = (u->ring_size - evtchn_ring_offset(u, c)) *
232 sizeof(evtchn_port_t);
233 bytes2 = evtchn_ring_offset(u, p) * sizeof(evtchn_port_t);
234 } else {
235 bytes1 = (p - c) * sizeof(evtchn_port_t);
236 bytes2 = 0;
237 }
238
239 /* Truncate chunks according to caller's maximum byte count. */
240 if (bytes1 > count) {
241 bytes1 = count;
242 bytes2 = 0;
243 } else if ((bytes1 + bytes2) > count) {
244 bytes2 = count - bytes1;
245 }
246
247 rc = -EFAULT;
248 rmb(); /* Ensure that we see the port before we copy it. */
249 if (copy_to_user(buf, evtchn_ring_entry(u, c), bytes1) ||
250 ((bytes2 != 0) &&
251 copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
252 goto unlock_out;
253
254 u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t);
255 rc = bytes1 + bytes2;
256
257 unlock_out:
258 mutex_unlock(&u->ring_cons_mutex);
259 return rc;
260}
261
262static ssize_t evtchn_write(struct file *file, const char __user *buf,
263 size_t count, loff_t *ppos)
264{
265 int rc, i;
266 evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
267 struct per_user_data *u = file->private_data;
268
269 if (kbuf == NULL)
270 return -ENOMEM;
271
272 /* Whole number of ports. */
273 count &= ~(sizeof(evtchn_port_t)-1);
274
275 rc = 0;
276 if (count == 0)
277 goto out;
278
279 if (count > PAGE_SIZE)
280 count = PAGE_SIZE;
281
282 rc = -EFAULT;
283 if (copy_from_user(kbuf, buf, count) != 0)
284 goto out;
285
286 mutex_lock(&u->bind_mutex);
287
288 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
289 unsigned port = kbuf[i];
290 struct user_evtchn *evtchn;
291
292 evtchn = find_evtchn(u, port);
293 if (evtchn && !evtchn->enabled) {
294 evtchn->enabled = true;
295 enable_irq(irq_from_evtchn(port));
296 }
297 }
298
299 mutex_unlock(&u->bind_mutex);
300
301 rc = count;
302
303 out:
304 free_page((unsigned long)kbuf);
305 return rc;
306}
307
308static int evtchn_resize_ring(struct per_user_data *u)
309{
310 unsigned int new_size;
311 evtchn_port_t *new_ring, *old_ring;
312
313 /*
314 * Ensure the ring is large enough to capture all possible
315 * events. i.e., one free slot for each bound event.
316 */
317 if (u->nr_evtchns <= u->ring_size)
318 return 0;
319
320 if (u->ring_size == 0)
321 new_size = 64;
322 else
323 new_size = 2 * u->ring_size;
324
325 new_ring = kvmalloc_array(new_size, sizeof(*new_ring), GFP_KERNEL);
326 if (!new_ring)
327 return -ENOMEM;
328
329 old_ring = u->ring;
330
331 /*
332 * Access to the ring contents is serialized by either the
333 * prod /or/ cons lock so take both when resizing.
334 */
335 mutex_lock(&u->ring_cons_mutex);
336 spin_lock_irq(&u->ring_prod_lock);
337
338 /*
339 * Copy the old ring contents to the new ring.
340 *
341 * To take care of wrapping, a full ring, and the new index
342 * pointing into the second half, simply copy the old contents
343 * twice.
344 *
345 * +---------+ +------------------+
346 * |34567 12| -> |34567 1234567 12|
347 * +-----p-c-+ +-------c------p---+
348 */
349 memcpy(new_ring, old_ring, u->ring_size * sizeof(*u->ring));
350 memcpy(new_ring + u->ring_size, old_ring,
351 u->ring_size * sizeof(*u->ring));
352
353 u->ring = new_ring;
354 u->ring_size = new_size;
355
356 spin_unlock_irq(&u->ring_prod_lock);
357 mutex_unlock(&u->ring_cons_mutex);
358
359 evtchn_free_ring(old_ring);
360
361 return 0;
362}
363
364static int evtchn_bind_to_user(struct per_user_data *u, int port)
365{
366 struct user_evtchn *evtchn;
367 struct evtchn_close close;
368 int rc = 0;
369
370 /*
371 * Ports are never reused, so every caller should pass in a
372 * unique port.
373 *
374 * (Locking not necessary because we haven't registered the
375 * interrupt handler yet, and our caller has already
376 * serialized bind operations.)
377 */
378
379 evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
380 if (!evtchn)
381 return -ENOMEM;
382
383 evtchn->user = u;
384 evtchn->port = port;
385 evtchn->enabled = true; /* start enabled */
386
387 rc = add_evtchn(u, evtchn);
388 if (rc < 0)
389 goto err;
390
391 rc = evtchn_resize_ring(u);
392 if (rc < 0)
393 goto err;
394
395 rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, 0,
396 u->name, evtchn);
397 if (rc < 0)
398 goto err;
399
400 rc = evtchn_make_refcounted(port);
401 return rc;
402
403err:
404 /* bind failed, should close the port now */
405 close.port = port;
406 if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
407 BUG();
408 del_evtchn(u, evtchn);
409 return rc;
410}
411
412static void evtchn_unbind_from_user(struct per_user_data *u,
413 struct user_evtchn *evtchn)
414{
415 int irq = irq_from_evtchn(evtchn->port);
416
417 BUG_ON(irq < 0);
418
419 unbind_from_irqhandler(irq, evtchn);
420
421 del_evtchn(u, evtchn);
422}
423
424static DEFINE_PER_CPU(int, bind_last_selected_cpu);
425
426static void evtchn_bind_interdom_next_vcpu(int evtchn)
427{
428 unsigned int selected_cpu, irq;
429 struct irq_desc *desc;
430 unsigned long flags;
431
432 irq = irq_from_evtchn(evtchn);
433 desc = irq_to_desc(irq);
434
435 if (!desc)
436 return;
437
438 raw_spin_lock_irqsave(&desc->lock, flags);
439 selected_cpu = this_cpu_read(bind_last_selected_cpu);
440 selected_cpu = cpumask_next_and(selected_cpu,
441 desc->irq_common_data.affinity, cpu_online_mask);
442
443 if (unlikely(selected_cpu >= nr_cpu_ids))
444 selected_cpu = cpumask_first_and(desc->irq_common_data.affinity,
445 cpu_online_mask);
446
447 this_cpu_write(bind_last_selected_cpu, selected_cpu);
448
449 /* unmask expects irqs to be disabled */
450 xen_set_affinity_evtchn(desc, selected_cpu);
451 raw_spin_unlock_irqrestore(&desc->lock, flags);
452}
453
454static long evtchn_ioctl(struct file *file,
455 unsigned int cmd, unsigned long arg)
456{
457 int rc;
458 struct per_user_data *u = file->private_data;
459 void __user *uarg = (void __user *) arg;
460
461 /* Prevent bind from racing with unbind */
462 mutex_lock(&u->bind_mutex);
463
464 switch (cmd) {
465 case IOCTL_EVTCHN_BIND_VIRQ: {
466 struct ioctl_evtchn_bind_virq bind;
467 struct evtchn_bind_virq bind_virq;
468
469 rc = -EACCES;
470 if (u->restrict_domid != UNRESTRICTED_DOMID)
471 break;
472
473 rc = -EFAULT;
474 if (copy_from_user(&bind, uarg, sizeof(bind)))
475 break;
476
477 bind_virq.virq = bind.virq;
478 bind_virq.vcpu = xen_vcpu_nr(0);
479 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
480 &bind_virq);
481 if (rc != 0)
482 break;
483
484 rc = evtchn_bind_to_user(u, bind_virq.port);
485 if (rc == 0)
486 rc = bind_virq.port;
487 break;
488 }
489
490 case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
491 struct ioctl_evtchn_bind_interdomain bind;
492 struct evtchn_bind_interdomain bind_interdomain;
493
494 rc = -EFAULT;
495 if (copy_from_user(&bind, uarg, sizeof(bind)))
496 break;
497
498 rc = -EACCES;
499 if (u->restrict_domid != UNRESTRICTED_DOMID &&
500 u->restrict_domid != bind.remote_domain)
501 break;
502
503 bind_interdomain.remote_dom = bind.remote_domain;
504 bind_interdomain.remote_port = bind.remote_port;
505 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
506 &bind_interdomain);
507 if (rc != 0)
508 break;
509
510 rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
511 if (rc == 0) {
512 rc = bind_interdomain.local_port;
513 evtchn_bind_interdom_next_vcpu(rc);
514 }
515 break;
516 }
517
518 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
519 struct ioctl_evtchn_bind_unbound_port bind;
520 struct evtchn_alloc_unbound alloc_unbound;
521
522 rc = -EACCES;
523 if (u->restrict_domid != UNRESTRICTED_DOMID)
524 break;
525
526 rc = -EFAULT;
527 if (copy_from_user(&bind, uarg, sizeof(bind)))
528 break;
529
530 alloc_unbound.dom = DOMID_SELF;
531 alloc_unbound.remote_dom = bind.remote_domain;
532 rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
533 &alloc_unbound);
534 if (rc != 0)
535 break;
536
537 rc = evtchn_bind_to_user(u, alloc_unbound.port);
538 if (rc == 0)
539 rc = alloc_unbound.port;
540 break;
541 }
542
543 case IOCTL_EVTCHN_UNBIND: {
544 struct ioctl_evtchn_unbind unbind;
545 struct user_evtchn *evtchn;
546
547 rc = -EFAULT;
548 if (copy_from_user(&unbind, uarg, sizeof(unbind)))
549 break;
550
551 rc = -EINVAL;
552 if (unbind.port >= xen_evtchn_nr_channels())
553 break;
554
555 rc = -ENOTCONN;
556 evtchn = find_evtchn(u, unbind.port);
557 if (!evtchn)
558 break;
559
560 disable_irq(irq_from_evtchn(unbind.port));
561 evtchn_unbind_from_user(u, evtchn);
562 rc = 0;
563 break;
564 }
565
566 case IOCTL_EVTCHN_NOTIFY: {
567 struct ioctl_evtchn_notify notify;
568 struct user_evtchn *evtchn;
569
570 rc = -EFAULT;
571 if (copy_from_user(¬ify, uarg, sizeof(notify)))
572 break;
573
574 rc = -ENOTCONN;
575 evtchn = find_evtchn(u, notify.port);
576 if (evtchn) {
577 notify_remote_via_evtchn(notify.port);
578 rc = 0;
579 }
580 break;
581 }
582
583 case IOCTL_EVTCHN_RESET: {
584 /* Initialise the ring to empty. Clear errors. */
585 mutex_lock(&u->ring_cons_mutex);
586 spin_lock_irq(&u->ring_prod_lock);
587 u->ring_cons = u->ring_prod = u->ring_overflow = 0;
588 spin_unlock_irq(&u->ring_prod_lock);
589 mutex_unlock(&u->ring_cons_mutex);
590 rc = 0;
591 break;
592 }
593
594 case IOCTL_EVTCHN_RESTRICT_DOMID: {
595 struct ioctl_evtchn_restrict_domid ierd;
596
597 rc = -EACCES;
598 if (u->restrict_domid != UNRESTRICTED_DOMID)
599 break;
600
601 rc = -EFAULT;
602 if (copy_from_user(&ierd, uarg, sizeof(ierd)))
603 break;
604
605 rc = -EINVAL;
606 if (ierd.domid == 0 || ierd.domid >= DOMID_FIRST_RESERVED)
607 break;
608
609 u->restrict_domid = ierd.domid;
610 rc = 0;
611
612 break;
613 }
614
615 default:
616 rc = -ENOSYS;
617 break;
618 }
619 mutex_unlock(&u->bind_mutex);
620
621 return rc;
622}
623
624static __poll_t evtchn_poll(struct file *file, poll_table *wait)
625{
626 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
627 struct per_user_data *u = file->private_data;
628
629 poll_wait(file, &u->evtchn_wait, wait);
630 if (u->ring_cons != u->ring_prod)
631 mask |= EPOLLIN | EPOLLRDNORM;
632 if (u->ring_overflow)
633 mask = EPOLLERR;
634 return mask;
635}
636
637static int evtchn_fasync(int fd, struct file *filp, int on)
638{
639 struct per_user_data *u = filp->private_data;
640 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
641}
642
643static int evtchn_open(struct inode *inode, struct file *filp)
644{
645 struct per_user_data *u;
646
647 u = kzalloc(sizeof(*u), GFP_KERNEL);
648 if (u == NULL)
649 return -ENOMEM;
650
651 u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
652 if (u->name == NULL) {
653 kfree(u);
654 return -ENOMEM;
655 }
656
657 init_waitqueue_head(&u->evtchn_wait);
658
659 mutex_init(&u->bind_mutex);
660 mutex_init(&u->ring_cons_mutex);
661 spin_lock_init(&u->ring_prod_lock);
662
663 u->restrict_domid = UNRESTRICTED_DOMID;
664
665 filp->private_data = u;
666
667 return stream_open(inode, filp);
668}
669
670static int evtchn_release(struct inode *inode, struct file *filp)
671{
672 struct per_user_data *u = filp->private_data;
673 struct rb_node *node;
674
675 while ((node = u->evtchns.rb_node)) {
676 struct user_evtchn *evtchn;
677
678 evtchn = rb_entry(node, struct user_evtchn, node);
679 disable_irq(irq_from_evtchn(evtchn->port));
680 evtchn_unbind_from_user(u, evtchn);
681 }
682
683 evtchn_free_ring(u->ring);
684 kfree(u->name);
685 kfree(u);
686
687 return 0;
688}
689
690static const struct file_operations evtchn_fops = {
691 .owner = THIS_MODULE,
692 .read = evtchn_read,
693 .write = evtchn_write,
694 .unlocked_ioctl = evtchn_ioctl,
695 .poll = evtchn_poll,
696 .fasync = evtchn_fasync,
697 .open = evtchn_open,
698 .release = evtchn_release,
699 .llseek = no_llseek,
700};
701
702static struct miscdevice evtchn_miscdev = {
703 .minor = MISC_DYNAMIC_MINOR,
704 .name = "xen/evtchn",
705 .fops = &evtchn_fops,
706};
707static int __init evtchn_init(void)
708{
709 int err;
710
711 if (!xen_domain())
712 return -ENODEV;
713
714 /* Create '/dev/xen/evtchn'. */
715 err = misc_register(&evtchn_miscdev);
716 if (err != 0) {
717 pr_err("Could not register /dev/xen/evtchn\n");
718 return err;
719 }
720
721 pr_info("Event-channel device installed\n");
722
723 return 0;
724}
725
726static void __exit evtchn_cleanup(void)
727{
728 misc_deregister(&evtchn_miscdev);
729}
730
731module_init(evtchn_init);
732module_exit(evtchn_cleanup);
733
734MODULE_LICENSE("GPL");