Commit | Line | Data |
---|---|---|
f7116284 IC |
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 | #include <linux/module.h> | |
35 | #include <linux/kernel.h> | |
36 | #include <linux/sched.h> | |
37 | #include <linux/slab.h> | |
38 | #include <linux/string.h> | |
39 | #include <linux/errno.h> | |
40 | #include <linux/fs.h> | |
f7116284 IC |
41 | #include <linux/miscdevice.h> |
42 | #include <linux/major.h> | |
43 | #include <linux/proc_fs.h> | |
44 | #include <linux/stat.h> | |
45 | #include <linux/poll.h> | |
46 | #include <linux/irq.h> | |
47 | #include <linux/init.h> | |
48 | #include <linux/gfp.h> | |
49 | #include <linux/mutex.h> | |
50 | #include <linux/cpu.h> | |
1ccbf534 JF |
51 | |
52 | #include <xen/xen.h> | |
f7116284 IC |
53 | #include <xen/events.h> |
54 | #include <xen/evtchn.h> | |
55 | #include <asm/xen/hypervisor.h> | |
56 | ||
57 | struct per_user_data { | |
0a4666b5 JF |
58 | struct mutex bind_mutex; /* serialize bind/unbind operations */ |
59 | ||
f7116284 IC |
60 | /* Notification ring, accessed via /dev/xen/evtchn. */ |
61 | #define EVTCHN_RING_SIZE (PAGE_SIZE / sizeof(evtchn_port_t)) | |
62 | #define EVTCHN_RING_MASK(_i) ((_i)&(EVTCHN_RING_SIZE-1)) | |
63 | evtchn_port_t *ring; | |
64 | unsigned int ring_cons, ring_prod, ring_overflow; | |
65 | struct mutex ring_cons_mutex; /* protect against concurrent readers */ | |
66 | ||
67 | /* Processes wait on this queue when ring is empty. */ | |
68 | wait_queue_head_t evtchn_wait; | |
69 | struct fasync_struct *evtchn_async_queue; | |
70 | const char *name; | |
71 | }; | |
72 | ||
73 | /* Who's bound to each port? */ | |
74 | static struct per_user_data *port_user[NR_EVENT_CHANNELS]; | |
0a4666b5 | 75 | static DEFINE_SPINLOCK(port_user_lock); /* protects port_user[] and ring_prod */ |
f7116284 IC |
76 | |
77 | irqreturn_t evtchn_interrupt(int irq, void *data) | |
78 | { | |
79 | unsigned int port = (unsigned long)data; | |
80 | struct per_user_data *u; | |
81 | ||
82 | spin_lock(&port_user_lock); | |
83 | ||
84 | u = port_user[port]; | |
85 | ||
86 | disable_irq_nosync(irq); | |
87 | ||
88 | if ((u->ring_prod - u->ring_cons) < EVTCHN_RING_SIZE) { | |
89 | u->ring[EVTCHN_RING_MASK(u->ring_prod)] = port; | |
90 | wmb(); /* Ensure ring contents visible */ | |
91 | if (u->ring_cons == u->ring_prod++) { | |
92 | wake_up_interruptible(&u->evtchn_wait); | |
93 | kill_fasync(&u->evtchn_async_queue, | |
94 | SIGIO, POLL_IN); | |
95 | } | |
96 | } else { | |
97 | u->ring_overflow = 1; | |
98 | } | |
99 | ||
100 | spin_unlock(&port_user_lock); | |
101 | ||
102 | return IRQ_HANDLED; | |
103 | } | |
104 | ||
105 | static ssize_t evtchn_read(struct file *file, char __user *buf, | |
106 | size_t count, loff_t *ppos) | |
107 | { | |
108 | int rc; | |
109 | unsigned int c, p, bytes1 = 0, bytes2 = 0; | |
110 | struct per_user_data *u = file->private_data; | |
111 | ||
112 | /* Whole number of ports. */ | |
113 | count &= ~(sizeof(evtchn_port_t)-1); | |
114 | ||
115 | if (count == 0) | |
116 | return 0; | |
117 | ||
118 | if (count > PAGE_SIZE) | |
119 | count = PAGE_SIZE; | |
120 | ||
121 | for (;;) { | |
122 | mutex_lock(&u->ring_cons_mutex); | |
123 | ||
124 | rc = -EFBIG; | |
125 | if (u->ring_overflow) | |
126 | goto unlock_out; | |
127 | ||
128 | c = u->ring_cons; | |
129 | p = u->ring_prod; | |
130 | if (c != p) | |
131 | break; | |
132 | ||
133 | mutex_unlock(&u->ring_cons_mutex); | |
134 | ||
135 | if (file->f_flags & O_NONBLOCK) | |
136 | return -EAGAIN; | |
137 | ||
138 | rc = wait_event_interruptible(u->evtchn_wait, | |
139 | u->ring_cons != u->ring_prod); | |
140 | if (rc) | |
141 | return rc; | |
142 | } | |
143 | ||
144 | /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */ | |
145 | if (((c ^ p) & EVTCHN_RING_SIZE) != 0) { | |
146 | bytes1 = (EVTCHN_RING_SIZE - EVTCHN_RING_MASK(c)) * | |
147 | sizeof(evtchn_port_t); | |
148 | bytes2 = EVTCHN_RING_MASK(p) * sizeof(evtchn_port_t); | |
149 | } else { | |
150 | bytes1 = (p - c) * sizeof(evtchn_port_t); | |
151 | bytes2 = 0; | |
152 | } | |
153 | ||
154 | /* Truncate chunks according to caller's maximum byte count. */ | |
155 | if (bytes1 > count) { | |
156 | bytes1 = count; | |
157 | bytes2 = 0; | |
158 | } else if ((bytes1 + bytes2) > count) { | |
159 | bytes2 = count - bytes1; | |
160 | } | |
161 | ||
162 | rc = -EFAULT; | |
163 | rmb(); /* Ensure that we see the port before we copy it. */ | |
164 | if (copy_to_user(buf, &u->ring[EVTCHN_RING_MASK(c)], bytes1) || | |
165 | ((bytes2 != 0) && | |
166 | copy_to_user(&buf[bytes1], &u->ring[0], bytes2))) | |
167 | goto unlock_out; | |
168 | ||
169 | u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t); | |
170 | rc = bytes1 + bytes2; | |
171 | ||
172 | unlock_out: | |
173 | mutex_unlock(&u->ring_cons_mutex); | |
174 | return rc; | |
175 | } | |
176 | ||
177 | static ssize_t evtchn_write(struct file *file, const char __user *buf, | |
178 | size_t count, loff_t *ppos) | |
179 | { | |
180 | int rc, i; | |
181 | evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL); | |
182 | struct per_user_data *u = file->private_data; | |
183 | ||
184 | if (kbuf == NULL) | |
185 | return -ENOMEM; | |
186 | ||
187 | /* Whole number of ports. */ | |
188 | count &= ~(sizeof(evtchn_port_t)-1); | |
189 | ||
190 | rc = 0; | |
191 | if (count == 0) | |
192 | goto out; | |
193 | ||
194 | if (count > PAGE_SIZE) | |
195 | count = PAGE_SIZE; | |
196 | ||
197 | rc = -EFAULT; | |
198 | if (copy_from_user(kbuf, buf, count) != 0) | |
199 | goto out; | |
200 | ||
201 | spin_lock_irq(&port_user_lock); | |
202 | for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) | |
203 | if ((kbuf[i] < NR_EVENT_CHANNELS) && (port_user[kbuf[i]] == u)) | |
204 | enable_irq(irq_from_evtchn(kbuf[i])); | |
205 | spin_unlock_irq(&port_user_lock); | |
206 | ||
207 | rc = count; | |
208 | ||
209 | out: | |
210 | free_page((unsigned long)kbuf); | |
211 | return rc; | |
212 | } | |
213 | ||
214 | static int evtchn_bind_to_user(struct per_user_data *u, int port) | |
215 | { | |
f7116284 IC |
216 | int rc = 0; |
217 | ||
0a4666b5 JF |
218 | /* |
219 | * Ports are never reused, so every caller should pass in a | |
220 | * unique port. | |
221 | * | |
222 | * (Locking not necessary because we haven't registered the | |
223 | * interrupt handler yet, and our caller has already | |
224 | * serialized bind operations.) | |
225 | */ | |
f7116284 | 226 | BUG_ON(port_user[port] != NULL); |
f7116284 IC |
227 | port_user[port] = u; |
228 | ||
0a4666b5 JF |
229 | rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, IRQF_DISABLED, |
230 | u->name, (void *)(unsigned long)port); | |
231 | if (rc >= 0) | |
232 | rc = 0; | |
233 | ||
f7116284 IC |
234 | return rc; |
235 | } | |
236 | ||
237 | static void evtchn_unbind_from_user(struct per_user_data *u, int port) | |
238 | { | |
239 | int irq = irq_from_evtchn(port); | |
240 | ||
241 | unbind_from_irqhandler(irq, (void *)(unsigned long)port); | |
0a4666b5 JF |
242 | |
243 | /* make sure we unbind the irq handler before clearing the port */ | |
244 | barrier(); | |
245 | ||
f7116284 IC |
246 | port_user[port] = NULL; |
247 | } | |
248 | ||
249 | static long evtchn_ioctl(struct file *file, | |
250 | unsigned int cmd, unsigned long arg) | |
251 | { | |
252 | int rc; | |
253 | struct per_user_data *u = file->private_data; | |
254 | void __user *uarg = (void __user *) arg; | |
255 | ||
0a4666b5 JF |
256 | /* Prevent bind from racing with unbind */ |
257 | mutex_lock(&u->bind_mutex); | |
258 | ||
f7116284 IC |
259 | switch (cmd) { |
260 | case IOCTL_EVTCHN_BIND_VIRQ: { | |
261 | struct ioctl_evtchn_bind_virq bind; | |
262 | struct evtchn_bind_virq bind_virq; | |
263 | ||
264 | rc = -EFAULT; | |
265 | if (copy_from_user(&bind, uarg, sizeof(bind))) | |
266 | break; | |
267 | ||
268 | bind_virq.virq = bind.virq; | |
269 | bind_virq.vcpu = 0; | |
270 | rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq, | |
271 | &bind_virq); | |
272 | if (rc != 0) | |
273 | break; | |
274 | ||
275 | rc = evtchn_bind_to_user(u, bind_virq.port); | |
276 | if (rc == 0) | |
277 | rc = bind_virq.port; | |
278 | break; | |
279 | } | |
280 | ||
281 | case IOCTL_EVTCHN_BIND_INTERDOMAIN: { | |
282 | struct ioctl_evtchn_bind_interdomain bind; | |
283 | struct evtchn_bind_interdomain bind_interdomain; | |
284 | ||
285 | rc = -EFAULT; | |
286 | if (copy_from_user(&bind, uarg, sizeof(bind))) | |
287 | break; | |
288 | ||
289 | bind_interdomain.remote_dom = bind.remote_domain; | |
290 | bind_interdomain.remote_port = bind.remote_port; | |
291 | rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain, | |
292 | &bind_interdomain); | |
293 | if (rc != 0) | |
294 | break; | |
295 | ||
296 | rc = evtchn_bind_to_user(u, bind_interdomain.local_port); | |
297 | if (rc == 0) | |
298 | rc = bind_interdomain.local_port; | |
299 | break; | |
300 | } | |
301 | ||
302 | case IOCTL_EVTCHN_BIND_UNBOUND_PORT: { | |
303 | struct ioctl_evtchn_bind_unbound_port bind; | |
304 | struct evtchn_alloc_unbound alloc_unbound; | |
305 | ||
306 | rc = -EFAULT; | |
307 | if (copy_from_user(&bind, uarg, sizeof(bind))) | |
308 | break; | |
309 | ||
310 | alloc_unbound.dom = DOMID_SELF; | |
311 | alloc_unbound.remote_dom = bind.remote_domain; | |
312 | rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound, | |
313 | &alloc_unbound); | |
314 | if (rc != 0) | |
315 | break; | |
316 | ||
317 | rc = evtchn_bind_to_user(u, alloc_unbound.port); | |
318 | if (rc == 0) | |
319 | rc = alloc_unbound.port; | |
320 | break; | |
321 | } | |
322 | ||
323 | case IOCTL_EVTCHN_UNBIND: { | |
324 | struct ioctl_evtchn_unbind unbind; | |
325 | ||
326 | rc = -EFAULT; | |
327 | if (copy_from_user(&unbind, uarg, sizeof(unbind))) | |
328 | break; | |
329 | ||
330 | rc = -EINVAL; | |
331 | if (unbind.port >= NR_EVENT_CHANNELS) | |
332 | break; | |
333 | ||
334 | spin_lock_irq(&port_user_lock); | |
335 | ||
336 | rc = -ENOTCONN; | |
337 | if (port_user[unbind.port] != u) { | |
338 | spin_unlock_irq(&port_user_lock); | |
339 | break; | |
340 | } | |
341 | ||
342 | evtchn_unbind_from_user(u, unbind.port); | |
343 | ||
344 | spin_unlock_irq(&port_user_lock); | |
345 | ||
346 | rc = 0; | |
347 | break; | |
348 | } | |
349 | ||
350 | case IOCTL_EVTCHN_NOTIFY: { | |
351 | struct ioctl_evtchn_notify notify; | |
352 | ||
353 | rc = -EFAULT; | |
354 | if (copy_from_user(¬ify, uarg, sizeof(notify))) | |
355 | break; | |
356 | ||
357 | if (notify.port >= NR_EVENT_CHANNELS) { | |
358 | rc = -EINVAL; | |
359 | } else if (port_user[notify.port] != u) { | |
360 | rc = -ENOTCONN; | |
361 | } else { | |
362 | notify_remote_via_evtchn(notify.port); | |
363 | rc = 0; | |
364 | } | |
365 | break; | |
366 | } | |
367 | ||
368 | case IOCTL_EVTCHN_RESET: { | |
369 | /* Initialise the ring to empty. Clear errors. */ | |
370 | mutex_lock(&u->ring_cons_mutex); | |
371 | spin_lock_irq(&port_user_lock); | |
372 | u->ring_cons = u->ring_prod = u->ring_overflow = 0; | |
373 | spin_unlock_irq(&port_user_lock); | |
374 | mutex_unlock(&u->ring_cons_mutex); | |
375 | rc = 0; | |
376 | break; | |
377 | } | |
378 | ||
379 | default: | |
380 | rc = -ENOSYS; | |
381 | break; | |
382 | } | |
0a4666b5 | 383 | mutex_unlock(&u->bind_mutex); |
f7116284 IC |
384 | |
385 | return rc; | |
386 | } | |
387 | ||
388 | static unsigned int evtchn_poll(struct file *file, poll_table *wait) | |
389 | { | |
390 | unsigned int mask = POLLOUT | POLLWRNORM; | |
391 | struct per_user_data *u = file->private_data; | |
392 | ||
393 | poll_wait(file, &u->evtchn_wait, wait); | |
394 | if (u->ring_cons != u->ring_prod) | |
395 | mask |= POLLIN | POLLRDNORM; | |
396 | if (u->ring_overflow) | |
397 | mask = POLLERR; | |
398 | return mask; | |
399 | } | |
400 | ||
401 | static int evtchn_fasync(int fd, struct file *filp, int on) | |
402 | { | |
403 | struct per_user_data *u = filp->private_data; | |
404 | return fasync_helper(fd, filp, on, &u->evtchn_async_queue); | |
405 | } | |
406 | ||
407 | static int evtchn_open(struct inode *inode, struct file *filp) | |
408 | { | |
409 | struct per_user_data *u; | |
410 | ||
411 | u = kzalloc(sizeof(*u), GFP_KERNEL); | |
412 | if (u == NULL) | |
413 | return -ENOMEM; | |
414 | ||
415 | u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm); | |
416 | if (u->name == NULL) { | |
417 | kfree(u); | |
418 | return -ENOMEM; | |
419 | } | |
420 | ||
421 | init_waitqueue_head(&u->evtchn_wait); | |
422 | ||
423 | u->ring = (evtchn_port_t *)__get_free_page(GFP_KERNEL); | |
424 | if (u->ring == NULL) { | |
425 | kfree(u->name); | |
426 | kfree(u); | |
427 | return -ENOMEM; | |
428 | } | |
429 | ||
0a4666b5 | 430 | mutex_init(&u->bind_mutex); |
f7116284 IC |
431 | mutex_init(&u->ring_cons_mutex); |
432 | ||
433 | filp->private_data = u; | |
434 | ||
435 | return 0; | |
436 | } | |
437 | ||
438 | static int evtchn_release(struct inode *inode, struct file *filp) | |
439 | { | |
440 | int i; | |
441 | struct per_user_data *u = filp->private_data; | |
442 | ||
443 | spin_lock_irq(&port_user_lock); | |
444 | ||
445 | free_page((unsigned long)u->ring); | |
446 | ||
447 | for (i = 0; i < NR_EVENT_CHANNELS; i++) { | |
448 | if (port_user[i] != u) | |
449 | continue; | |
450 | ||
451 | evtchn_unbind_from_user(port_user[i], i); | |
452 | } | |
453 | ||
454 | spin_unlock_irq(&port_user_lock); | |
455 | ||
456 | kfree(u->name); | |
457 | kfree(u); | |
458 | ||
459 | return 0; | |
460 | } | |
461 | ||
462 | static const struct file_operations evtchn_fops = { | |
463 | .owner = THIS_MODULE, | |
464 | .read = evtchn_read, | |
465 | .write = evtchn_write, | |
466 | .unlocked_ioctl = evtchn_ioctl, | |
467 | .poll = evtchn_poll, | |
468 | .fasync = evtchn_fasync, | |
469 | .open = evtchn_open, | |
470 | .release = evtchn_release, | |
471 | }; | |
472 | ||
473 | static struct miscdevice evtchn_miscdev = { | |
474 | .minor = MISC_DYNAMIC_MINOR, | |
475 | .name = "evtchn", | |
476 | .fops = &evtchn_fops, | |
477 | }; | |
478 | static int __init evtchn_init(void) | |
479 | { | |
480 | int err; | |
481 | ||
482 | if (!xen_domain()) | |
483 | return -ENODEV; | |
484 | ||
485 | spin_lock_init(&port_user_lock); | |
486 | memset(port_user, 0, sizeof(port_user)); | |
487 | ||
488 | /* Create '/dev/misc/evtchn'. */ | |
489 | err = misc_register(&evtchn_miscdev); | |
490 | if (err != 0) { | |
491 | printk(KERN_ALERT "Could not register /dev/misc/evtchn\n"); | |
492 | return err; | |
493 | } | |
494 | ||
495 | printk(KERN_INFO "Event-channel device installed.\n"); | |
496 | ||
497 | return 0; | |
498 | } | |
499 | ||
500 | static void __exit evtchn_cleanup(void) | |
501 | { | |
502 | misc_deregister(&evtchn_miscdev); | |
503 | } | |
504 | ||
505 | module_init(evtchn_init); | |
506 | module_exit(evtchn_cleanup); | |
507 | ||
508 | MODULE_LICENSE("GPL"); |