xref: /linux/drivers/xen/evtchn.c (revision f884ab15afdc5514e88105c92a4e2e1e6539869a) 
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>
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/mutex.h>
49 #include <linux/cpu.h>
50 
51 #include <xen/xen.h>
52 #include <xen/events.h>
53 #include <xen/evtchn.h>
54 #include <asm/xen/hypervisor.h>
55 
56 struct per_user_data {
57 	struct mutex bind_mutex; /* serialize bind/unbind operations */
58 
59 	/* Notification ring, accessed via /dev/xen/evtchn. */
60 #define EVTCHN_RING_SIZE     (PAGE_SIZE / sizeof(evtchn_port_t))
61 #define EVTCHN_RING_MASK(_i) ((_i)&(EVTCHN_RING_SIZE-1))
62 	evtchn_port_t *ring;
63 	unsigned int ring_cons, ring_prod, ring_overflow;
64 	struct mutex ring_cons_mutex; /* protect against concurrent readers */
65 
66 	/* Processes wait on this queue when ring is empty. */
67 	wait_queue_head_t evtchn_wait;
68 	struct fasync_struct *evtchn_async_queue;
69 	const char *name;
70 };
71 
72 /*
73  * Who's bound to each port?  This is logically an array of struct
74  * per_user_data *, but we encode the current enabled-state in bit 0.
75  */
76 static unsigned long *port_user;
77 static DEFINE_SPINLOCK(port_user_lock); /* protects port_user[] and ring_prod */
78 
79 static inline struct per_user_data *get_port_user(unsigned port)
80 {
81 	return (struct per_user_data *)(port_user[port] & ~1);
82 }
83 
84 static inline void set_port_user(unsigned port, struct per_user_data *u)
85 {
86 	port_user[port] = (unsigned long)u;
87 }
88 
89 static inline bool get_port_enabled(unsigned port)
90 {
91 	return port_user[port] & 1;
92 }
93 
94 static inline void set_port_enabled(unsigned port, bool enabled)
95 {
96 	if (enabled)
97 		port_user[port] |= 1;
98 	else
99 		port_user[port] &= ~1;
100 }
101 
102 static irqreturn_t evtchn_interrupt(int irq, void *data)
103 {
104 	unsigned int port = (unsigned long)data;
105 	struct per_user_data *u;
106 
107 	spin_lock(&port_user_lock);
108 
109 	u = get_port_user(port);
110 
111 	WARN(!get_port_enabled(port),
112 	     "Interrupt for port %d, but apparently not enabled; per-user %p\n",
113 	     port, u);
114 
115 	disable_irq_nosync(irq);
116 	set_port_enabled(port, false);
117 
118 	if ((u->ring_prod - u->ring_cons) < EVTCHN_RING_SIZE) {
119 		u->ring[EVTCHN_RING_MASK(u->ring_prod)] = port;
120 		wmb(); /* Ensure ring contents visible */
121 		if (u->ring_cons == u->ring_prod++) {
122 			wake_up_interruptible(&u->evtchn_wait);
123 			kill_fasync(&u->evtchn_async_queue,
124 				    SIGIO, POLL_IN);
125 		}
126 	} else
127 		u->ring_overflow = 1;
128 
129 	spin_unlock(&port_user_lock);
130 
131 	return IRQ_HANDLED;
132 }
133 
134 static ssize_t evtchn_read(struct file *file, char __user *buf,
135 			   size_t count, loff_t *ppos)
136 {
137 	int rc;
138 	unsigned int c, p, bytes1 = 0, bytes2 = 0;
139 	struct per_user_data *u = file->private_data;
140 
141 	/* Whole number of ports. */
142 	count &= ~(sizeof(evtchn_port_t)-1);
143 
144 	if (count == 0)
145 		return 0;
146 
147 	if (count > PAGE_SIZE)
148 		count = PAGE_SIZE;
149 
150 	for (;;) {
151 		mutex_lock(&u->ring_cons_mutex);
152 
153 		rc = -EFBIG;
154 		if (u->ring_overflow)
155 			goto unlock_out;
156 
157 		c = u->ring_cons;
158 		p = u->ring_prod;
159 		if (c != p)
160 			break;
161 
162 		mutex_unlock(&u->ring_cons_mutex);
163 
164 		if (file->f_flags & O_NONBLOCK)
165 			return -EAGAIN;
166 
167 		rc = wait_event_interruptible(u->evtchn_wait,
168 					      u->ring_cons != u->ring_prod);
169 		if (rc)
170 			return rc;
171 	}
172 
173 	/* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
174 	if (((c ^ p) & EVTCHN_RING_SIZE) != 0) {
175 		bytes1 = (EVTCHN_RING_SIZE - EVTCHN_RING_MASK(c)) *
176 			sizeof(evtchn_port_t);
177 		bytes2 = EVTCHN_RING_MASK(p) * sizeof(evtchn_port_t);
178 	} else {
179 		bytes1 = (p - c) * sizeof(evtchn_port_t);
180 		bytes2 = 0;
181 	}
182 
183 	/* Truncate chunks according to caller's maximum byte count. */
184 	if (bytes1 > count) {
185 		bytes1 = count;
186 		bytes2 = 0;
187 	} else if ((bytes1 + bytes2) > count) {
188 		bytes2 = count - bytes1;
189 	}
190 
191 	rc = -EFAULT;
192 	rmb(); /* Ensure that we see the port before we copy it. */
193 	if (copy_to_user(buf, &u->ring[EVTCHN_RING_MASK(c)], bytes1) ||
194 	    ((bytes2 != 0) &&
195 	     copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
196 		goto unlock_out;
197 
198 	u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t);
199 	rc = bytes1 + bytes2;
200 
201  unlock_out:
202 	mutex_unlock(&u->ring_cons_mutex);
203 	return rc;
204 }
205 
206 static ssize_t evtchn_write(struct file *file, const char __user *buf,
207 			    size_t count, loff_t *ppos)
208 {
209 	int rc, i;
210 	evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
211 	struct per_user_data *u = file->private_data;
212 
213 	if (kbuf == NULL)
214 		return -ENOMEM;
215 
216 	/* Whole number of ports. */
217 	count &= ~(sizeof(evtchn_port_t)-1);
218 
219 	rc = 0;
220 	if (count == 0)
221 		goto out;
222 
223 	if (count > PAGE_SIZE)
224 		count = PAGE_SIZE;
225 
226 	rc = -EFAULT;
227 	if (copy_from_user(kbuf, buf, count) != 0)
228 		goto out;
229 
230 	spin_lock_irq(&port_user_lock);
231 
232 	for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
233 		unsigned port = kbuf[i];
234 
235 		if (port < NR_EVENT_CHANNELS &&
236 		    get_port_user(port) == u &&
237 		    !get_port_enabled(port)) {
238 			set_port_enabled(port, true);
239 			enable_irq(irq_from_evtchn(port));
240 		}
241 	}
242 
243 	spin_unlock_irq(&port_user_lock);
244 
245 	rc = count;
246 
247  out:
248 	free_page((unsigned long)kbuf);
249 	return rc;
250 }
251 
252 static int evtchn_bind_to_user(struct per_user_data *u, int port)
253 {
254 	int rc = 0;
255 
256 	/*
257 	 * Ports are never reused, so every caller should pass in a
258 	 * unique port.
259 	 *
260 	 * (Locking not necessary because we haven't registered the
261 	 * interrupt handler yet, and our caller has already
262 	 * serialized bind operations.)
263 	 */
264 	BUG_ON(get_port_user(port) != NULL);
265 	set_port_user(port, u);
266 	set_port_enabled(port, true); /* start enabled */
267 
268 	rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, IRQF_DISABLED,
269 				       u->name, (void *)(unsigned long)port);
270 	if (rc >= 0)
271 		rc = evtchn_make_refcounted(port);
272 	else {
273 		/* bind failed, should close the port now */
274 		struct evtchn_close close;
275 		close.port = port;
276 		if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
277 			BUG();
278 		set_port_user(port, NULL);
279 	}
280 
281 	return rc;
282 }
283 
284 static void evtchn_unbind_from_user(struct per_user_data *u, int port)
285 {
286 	int irq = irq_from_evtchn(port);
287 
288 	BUG_ON(irq < 0);
289 
290 	unbind_from_irqhandler(irq, (void *)(unsigned long)port);
291 
292 	set_port_user(port, NULL);
293 }
294 
295 static long evtchn_ioctl(struct file *file,
296 			 unsigned int cmd, unsigned long arg)
297 {
298 	int rc;
299 	struct per_user_data *u = file->private_data;
300 	void __user *uarg = (void __user *) arg;
301 
302 	/* Prevent bind from racing with unbind */
303 	mutex_lock(&u->bind_mutex);
304 
305 	switch (cmd) {
306 	case IOCTL_EVTCHN_BIND_VIRQ: {
307 		struct ioctl_evtchn_bind_virq bind;
308 		struct evtchn_bind_virq bind_virq;
309 
310 		rc = -EFAULT;
311 		if (copy_from_user(&bind, uarg, sizeof(bind)))
312 			break;
313 
314 		bind_virq.virq = bind.virq;
315 		bind_virq.vcpu = 0;
316 		rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
317 						 &bind_virq);
318 		if (rc != 0)
319 			break;
320 
321 		rc = evtchn_bind_to_user(u, bind_virq.port);
322 		if (rc == 0)
323 			rc = bind_virq.port;
324 		break;
325 	}
326 
327 	case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
328 		struct ioctl_evtchn_bind_interdomain bind;
329 		struct evtchn_bind_interdomain bind_interdomain;
330 
331 		rc = -EFAULT;
332 		if (copy_from_user(&bind, uarg, sizeof(bind)))
333 			break;
334 
335 		bind_interdomain.remote_dom  = bind.remote_domain;
336 		bind_interdomain.remote_port = bind.remote_port;
337 		rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
338 						 &bind_interdomain);
339 		if (rc != 0)
340 			break;
341 
342 		rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
343 		if (rc == 0)
344 			rc = bind_interdomain.local_port;
345 		break;
346 	}
347 
348 	case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
349 		struct ioctl_evtchn_bind_unbound_port bind;
350 		struct evtchn_alloc_unbound alloc_unbound;
351 
352 		rc = -EFAULT;
353 		if (copy_from_user(&bind, uarg, sizeof(bind)))
354 			break;
355 
356 		alloc_unbound.dom        = DOMID_SELF;
357 		alloc_unbound.remote_dom = bind.remote_domain;
358 		rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
359 						 &alloc_unbound);
360 		if (rc != 0)
361 			break;
362 
363 		rc = evtchn_bind_to_user(u, alloc_unbound.port);
364 		if (rc == 0)
365 			rc = alloc_unbound.port;
366 		break;
367 	}
368 
369 	case IOCTL_EVTCHN_UNBIND: {
370 		struct ioctl_evtchn_unbind unbind;
371 
372 		rc = -EFAULT;
373 		if (copy_from_user(&unbind, uarg, sizeof(unbind)))
374 			break;
375 
376 		rc = -EINVAL;
377 		if (unbind.port >= NR_EVENT_CHANNELS)
378 			break;
379 
380 		spin_lock_irq(&port_user_lock);
381 
382 		rc = -ENOTCONN;
383 		if (get_port_user(unbind.port) != u) {
384 			spin_unlock_irq(&port_user_lock);
385 			break;
386 		}
387 
388 		disable_irq(irq_from_evtchn(unbind.port));
389 
390 		spin_unlock_irq(&port_user_lock);
391 
392 		evtchn_unbind_from_user(u, unbind.port);
393 
394 		rc = 0;
395 		break;
396 	}
397 
398 	case IOCTL_EVTCHN_NOTIFY: {
399 		struct ioctl_evtchn_notify notify;
400 
401 		rc = -EFAULT;
402 		if (copy_from_user(&notify, uarg, sizeof(notify)))
403 			break;
404 
405 		if (notify.port >= NR_EVENT_CHANNELS) {
406 			rc = -EINVAL;
407 		} else if (get_port_user(notify.port) != u) {
408 			rc = -ENOTCONN;
409 		} else {
410 			notify_remote_via_evtchn(notify.port);
411 			rc = 0;
412 		}
413 		break;
414 	}
415 
416 	case IOCTL_EVTCHN_RESET: {
417 		/* Initialise the ring to empty. Clear errors. */
418 		mutex_lock(&u->ring_cons_mutex);
419 		spin_lock_irq(&port_user_lock);
420 		u->ring_cons = u->ring_prod = u->ring_overflow = 0;
421 		spin_unlock_irq(&port_user_lock);
422 		mutex_unlock(&u->ring_cons_mutex);
423 		rc = 0;
424 		break;
425 	}
426 
427 	default:
428 		rc = -ENOSYS;
429 		break;
430 	}
431 	mutex_unlock(&u->bind_mutex);
432 
433 	return rc;
434 }
435 
436 static unsigned int evtchn_poll(struct file *file, poll_table *wait)
437 {
438 	unsigned int mask = POLLOUT | POLLWRNORM;
439 	struct per_user_data *u = file->private_data;
440 
441 	poll_wait(file, &u->evtchn_wait, wait);
442 	if (u->ring_cons != u->ring_prod)
443 		mask |= POLLIN | POLLRDNORM;
444 	if (u->ring_overflow)
445 		mask = POLLERR;
446 	return mask;
447 }
448 
449 static int evtchn_fasync(int fd, struct file *filp, int on)
450 {
451 	struct per_user_data *u = filp->private_data;
452 	return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
453 }
454 
455 static int evtchn_open(struct inode *inode, struct file *filp)
456 {
457 	struct per_user_data *u;
458 
459 	u = kzalloc(sizeof(*u), GFP_KERNEL);
460 	if (u == NULL)
461 		return -ENOMEM;
462 
463 	u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
464 	if (u->name == NULL) {
465 		kfree(u);
466 		return -ENOMEM;
467 	}
468 
469 	init_waitqueue_head(&u->evtchn_wait);
470 
471 	u->ring = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
472 	if (u->ring == NULL) {
473 		kfree(u->name);
474 		kfree(u);
475 		return -ENOMEM;
476 	}
477 
478 	mutex_init(&u->bind_mutex);
479 	mutex_init(&u->ring_cons_mutex);
480 
481 	filp->private_data = u;
482 
483 	return nonseekable_open(inode, filp);
484 }
485 
486 static int evtchn_release(struct inode *inode, struct file *filp)
487 {
488 	int i;
489 	struct per_user_data *u = filp->private_data;
490 
491 	spin_lock_irq(&port_user_lock);
492 
493 	free_page((unsigned long)u->ring);
494 
495 	for (i = 0; i < NR_EVENT_CHANNELS; i++) {
496 		if (get_port_user(i) != u)
497 			continue;
498 
499 		disable_irq(irq_from_evtchn(i));
500 	}
501 
502 	spin_unlock_irq(&port_user_lock);
503 
504 	for (i = 0; i < NR_EVENT_CHANNELS; i++) {
505 		if (get_port_user(i) != u)
506 			continue;
507 
508 		evtchn_unbind_from_user(get_port_user(i), i);
509 	}
510 
511 	kfree(u->name);
512 	kfree(u);
513 
514 	return 0;
515 }
516 
517 static const struct file_operations evtchn_fops = {
518 	.owner   = THIS_MODULE,
519 	.read    = evtchn_read,
520 	.write   = evtchn_write,
521 	.unlocked_ioctl = evtchn_ioctl,
522 	.poll    = evtchn_poll,
523 	.fasync  = evtchn_fasync,
524 	.open    = evtchn_open,
525 	.release = evtchn_release,
526 	.llseek	 = no_llseek,
527 };
528 
529 static struct miscdevice evtchn_miscdev = {
530 	.minor        = MISC_DYNAMIC_MINOR,
531 	.name         = "xen/evtchn",
532 	.fops         = &evtchn_fops,
533 };
534 static int __init evtchn_init(void)
535 {
536 	int err;
537 
538 	if (!xen_domain())
539 		return -ENODEV;
540 
541 	port_user = kcalloc(NR_EVENT_CHANNELS, sizeof(*port_user), GFP_KERNEL);
542 	if (port_user == NULL)
543 		return -ENOMEM;
544 
545 	spin_lock_init(&port_user_lock);
546 
547 	/* Create '/dev/xen/evtchn'. */
548 	err = misc_register(&evtchn_miscdev);
549 	if (err != 0) {
550 		printk(KERN_ERR "Could not register /dev/xen/evtchn\n");
551 		return err;
552 	}
553 
554 	printk(KERN_INFO "Event-channel device installed.\n");
555 
556 	return 0;
557 }
558 
559 static void __exit evtchn_cleanup(void)
560 {
561 	kfree(port_user);
562 	port_user = NULL;
563 
564 	misc_deregister(&evtchn_miscdev);
565 }
566 
567 module_init(evtchn_init);
568 module_exit(evtchn_cleanup);
569 
570 MODULE_LICENSE("GPL");
571