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(addr: 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(node: &evtchn->node, parent, rb_link: 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(objp: 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(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, idx: 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(lock: &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(lock: &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, idx: c)) *
237 sizeof(evtchn_port_t);
238 bytes2 = evtchn_ring_offset(u, idx: 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(to: buf, from: evtchn_ring_entry(u, idx: c), n: bytes1) ||
255 ((bytes2 != 0) &&
256 copy_to_user(to: &buf[bytes1], from: &u->ring[0], n: 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(lock: &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(to: kbuf, from: buf, n: 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: irq_from_evtchn(evtchn: port), eoi_flags: 0);
301 }
302 }
303
304 mutex_unlock(lock: &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(n: new_size, 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(lock: &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(lock: &u->ring_prod_lock);
362 mutex_unlock(lock: &u->ring_cons_mutex);
363
364 evtchn_free_ring(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 bool is_static)
371{
372 struct user_evtchn *evtchn;
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(size: 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(evtchn: port, handler: evtchn_interrupt, IRQF_SHARED,
401 devname: u->name, dev_id: evtchn);
402 if (rc < 0)
403 goto err;
404
405 rc = evtchn_make_refcounted(evtchn: port, is_static);
406 return rc;
407
408err:
409 /* bind failed, should close the port now */
410 if (!is_static)
411 xen_evtchn_close(port);
412
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: evtchn->port);
421
422 BUG_ON(irq < 0);
423
424 unbind_from_irqhandler(irq, dev_id: 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(to: &bind, from: uarg, n: sizeof(bind)))
450 break;
451
452 bind_virq.virq = bind.virq;
453 bind_virq.vcpu = xen_vcpu_nr(cpu: 0);
454 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
455 arg: &bind_virq);
456 if (rc != 0)
457 break;
458
459 rc = evtchn_bind_to_user(u, port: bind_virq.port, is_static: false);
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(to: &bind, from: uarg, n: 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 arg: &bind_interdomain);
482 if (rc != 0)
483 break;
484
485 rc = evtchn_bind_to_user(u, port: bind_interdomain.local_port, is_static: false);
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(to: &bind, from: uarg, n: 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 arg: &alloc_unbound);
507 if (rc != 0)
508 break;
509
510 rc = evtchn_bind_to_user(u, port: alloc_unbound.port, is_static: false);
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(to: &unbind, from: uarg, n: 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, port: unbind.port);
530 if (!evtchn)
531 break;
532
533 disable_irq(irq: irq_from_evtchn(evtchn: unbind.port));
534 evtchn_unbind_from_user(u, evtchn);
535 rc = 0;
536 break;
537 }
538
539 case IOCTL_EVTCHN_BIND_STATIC: {
540 struct ioctl_evtchn_bind bind;
541 struct user_evtchn *evtchn;
542
543 rc = -EFAULT;
544 if (copy_from_user(to: &bind, from: uarg, n: sizeof(bind)))
545 break;
546
547 rc = -EISCONN;
548 evtchn = find_evtchn(u, port: bind.port);
549 if (evtchn)
550 break;
551
552 rc = evtchn_bind_to_user(u, port: bind.port, is_static: true);
553 break;
554 }
555
556 case IOCTL_EVTCHN_NOTIFY: {
557 struct ioctl_evtchn_notify notify;
558 struct user_evtchn *evtchn;
559
560 rc = -EFAULT;
561 if (copy_from_user(to: &notify, from: uarg, n: sizeof(notify)))
562 break;
563
564 rc = -ENOTCONN;
565 evtchn = find_evtchn(u, port: notify.port);
566 if (evtchn) {
567 notify_remote_via_evtchn(port: notify.port);
568 rc = 0;
569 }
570 break;
571 }
572
573 case IOCTL_EVTCHN_RESET: {
574 /* Initialise the ring to empty. Clear errors. */
575 mutex_lock(&u->ring_cons_mutex);
576 spin_lock_irq(lock: &u->ring_prod_lock);
577 WRITE_ONCE(u->ring_cons, 0);
578 WRITE_ONCE(u->ring_prod, 0);
579 u->ring_overflow = 0;
580 spin_unlock_irq(lock: &u->ring_prod_lock);
581 mutex_unlock(lock: &u->ring_cons_mutex);
582 rc = 0;
583 break;
584 }
585
586 case IOCTL_EVTCHN_RESTRICT_DOMID: {
587 struct ioctl_evtchn_restrict_domid ierd;
588
589 rc = -EACCES;
590 if (u->restrict_domid != UNRESTRICTED_DOMID)
591 break;
592
593 rc = -EFAULT;
594 if (copy_from_user(to: &ierd, from: uarg, n: sizeof(ierd)))
595 break;
596
597 rc = -EINVAL;
598 if (ierd.domid == 0 || ierd.domid >= DOMID_FIRST_RESERVED)
599 break;
600
601 u->restrict_domid = ierd.domid;
602 rc = 0;
603
604 break;
605 }
606
607 default:
608 rc = -ENOSYS;
609 break;
610 }
611 mutex_unlock(lock: &u->bind_mutex);
612
613 return rc;
614}
615
616static __poll_t evtchn_poll(struct file *file, poll_table *wait)
617{
618 __poll_t mask = EPOLLOUT | EPOLLWRNORM;
619 struct per_user_data *u = file->private_data;
620
621 poll_wait(filp: file, wait_address: &u->evtchn_wait, p: wait);
622 if (READ_ONCE(u->ring_cons) != READ_ONCE(u->ring_prod))
623 mask |= EPOLLIN | EPOLLRDNORM;
624 if (u->ring_overflow)
625 mask = EPOLLERR;
626 return mask;
627}
628
629static int evtchn_fasync(int fd, struct file *filp, int on)
630{
631 struct per_user_data *u = filp->private_data;
632 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
633}
634
635static int evtchn_open(struct inode *inode, struct file *filp)
636{
637 struct per_user_data *u;
638
639 u = kzalloc(size: sizeof(*u), GFP_KERNEL);
640 if (u == NULL)
641 return -ENOMEM;
642
643 u->name = kasprintf(GFP_KERNEL, fmt: "evtchn:%s", current->comm);
644 if (u->name == NULL) {
645 kfree(objp: u);
646 return -ENOMEM;
647 }
648
649 init_waitqueue_head(&u->evtchn_wait);
650
651 mutex_init(&u->bind_mutex);
652 mutex_init(&u->ring_cons_mutex);
653 spin_lock_init(&u->ring_prod_lock);
654
655 u->restrict_domid = UNRESTRICTED_DOMID;
656
657 filp->private_data = u;
658
659 return stream_open(inode, filp);
660}
661
662static int evtchn_release(struct inode *inode, struct file *filp)
663{
664 struct per_user_data *u = filp->private_data;
665 struct rb_node *node;
666
667 while ((node = u->evtchns.rb_node)) {
668 struct user_evtchn *evtchn;
669
670 evtchn = rb_entry(node, struct user_evtchn, node);
671 disable_irq(irq: irq_from_evtchn(evtchn: evtchn->port));
672 evtchn_unbind_from_user(u, evtchn);
673 }
674
675 evtchn_free_ring(ring: u->ring);
676 kfree(objp: u->name);
677 kfree(objp: u);
678
679 return 0;
680}
681
682static const struct file_operations evtchn_fops = {
683 .owner = THIS_MODULE,
684 .read = evtchn_read,
685 .write = evtchn_write,
686 .unlocked_ioctl = evtchn_ioctl,
687 .poll = evtchn_poll,
688 .fasync = evtchn_fasync,
689 .open = evtchn_open,
690 .release = evtchn_release,
691 .llseek = no_llseek,
692};
693
694static struct miscdevice evtchn_miscdev = {
695 .minor = MISC_DYNAMIC_MINOR,
696 .name = "xen/evtchn",
697 .fops = &evtchn_fops,
698};
699static int __init evtchn_init(void)
700{
701 int err;
702
703 if (!xen_domain())
704 return -ENODEV;
705
706 /* Create '/dev/xen/evtchn'. */
707 err = misc_register(misc: &evtchn_miscdev);
708 if (err != 0) {
709 pr_err("Could not register /dev/xen/evtchn\n");
710 return err;
711 }
712
713 pr_info("Event-channel device installed\n");
714
715 return 0;
716}
717
718static void __exit evtchn_cleanup(void)
719{
720 misc_deregister(misc: &evtchn_miscdev);
721}
722
723module_init(evtchn_init);
724module_exit(evtchn_cleanup);
725
726MODULE_LICENSE("GPL");
727

source code of linux/drivers/xen/evtchn.c