xref: /openbmc/linux/drivers/xen/evtchn.c (revision 56a0eccd) 
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 <asm/xen/hypervisor.h>
59 
60 struct per_user_data {
61 	struct mutex bind_mutex; /* serialize bind/unbind operations */
62 	struct rb_root evtchns;
63 	unsigned int nr_evtchns;
64 
65 	/* Notification ring, accessed via /dev/xen/evtchn. */
66 	unsigned int ring_size;
67 	evtchn_port_t *ring;
68 	unsigned int ring_cons, ring_prod, ring_overflow;
69 	struct mutex ring_cons_mutex; /* protect against concurrent readers */
70 	spinlock_t ring_prod_lock; /* product against concurrent interrupts */
71 
72 	/* Processes wait on this queue when ring is empty. */
73 	wait_queue_head_t evtchn_wait;
74 	struct fasync_struct *evtchn_async_queue;
75 	const char *name;
76 };
77 
78 struct user_evtchn {
79 	struct rb_node node;
80 	struct per_user_data *user;
81 	unsigned port;
82 	bool enabled;
83 };
84 
85 static evtchn_port_t *evtchn_alloc_ring(unsigned int size)
86 {
87 	evtchn_port_t *ring;
88 	size_t s = size * sizeof(*ring);
89 
90 	ring = kmalloc(s, GFP_KERNEL);
91 	if (!ring)
92 		ring = vmalloc(s);
93 
94 	return ring;
95 }
96 
97 static void evtchn_free_ring(evtchn_port_t *ring)
98 {
99 	kvfree(ring);
100 }
101 
102 static unsigned int evtchn_ring_offset(struct per_user_data *u,
103 				       unsigned int idx)
104 {
105 	return idx & (u->ring_size - 1);
106 }
107 
108 static evtchn_port_t *evtchn_ring_entry(struct per_user_data *u,
109 					unsigned int idx)
110 {
111 	return u->ring + evtchn_ring_offset(u, idx);
112 }
113 
114 static int add_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
115 {
116 	struct rb_node **new = &(u->evtchns.rb_node), *parent = NULL;
117 
118 	u->nr_evtchns++;
119 
120 	while (*new) {
121 		struct user_evtchn *this;
122 
123 		this = container_of(*new, struct user_evtchn, node);
124 
125 		parent = *new;
126 		if (this->port < evtchn->port)
127 			new = &((*new)->rb_left);
128 		else if (this->port > evtchn->port)
129 			new = &((*new)->rb_right);
130 		else
131 			return -EEXIST;
132 	}
133 
134 	/* Add new node and rebalance tree. */
135 	rb_link_node(&evtchn->node, parent, new);
136 	rb_insert_color(&evtchn->node, &u->evtchns);
137 
138 	return 0;
139 }
140 
141 static void del_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)
142 {
143 	u->nr_evtchns--;
144 	rb_erase(&evtchn->node, &u->evtchns);
145 	kfree(evtchn);
146 }
147 
148 static struct user_evtchn *find_evtchn(struct per_user_data *u, unsigned port)
149 {
150 	struct rb_node *node = u->evtchns.rb_node;
151 
152 	while (node) {
153 		struct user_evtchn *evtchn;
154 
155 		evtchn = container_of(node, struct user_evtchn, node);
156 
157 		if (evtchn->port < port)
158 			node = node->rb_left;
159 		else if (evtchn->port > port)
160 			node = node->rb_right;
161 		else
162 			return evtchn;
163 	}
164 	return NULL;
165 }
166 
167 static irqreturn_t evtchn_interrupt(int irq, void *data)
168 {
169 	struct user_evtchn *evtchn = data;
170 	struct per_user_data *u = evtchn->user;
171 
172 	WARN(!evtchn->enabled,
173 	     "Interrupt for port %d, but apparently not enabled; per-user %p\n",
174 	     evtchn->port, u);
175 
176 	disable_irq_nosync(irq);
177 	evtchn->enabled = false;
178 
179 	spin_lock(&u->ring_prod_lock);
180 
181 	if ((u->ring_prod - u->ring_cons) < u->ring_size) {
182 		*evtchn_ring_entry(u, u->ring_prod) = evtchn->port;
183 		wmb(); /* Ensure ring contents visible */
184 		if (u->ring_cons == u->ring_prod++) {
185 			wake_up_interruptible(&u->evtchn_wait);
186 			kill_fasync(&u->evtchn_async_queue,
187 				    SIGIO, POLL_IN);
188 		}
189 	} else
190 		u->ring_overflow = 1;
191 
192 	spin_unlock(&u->ring_prod_lock);
193 
194 	return IRQ_HANDLED;
195 }
196 
197 static ssize_t evtchn_read(struct file *file, char __user *buf,
198 			   size_t count, loff_t *ppos)
199 {
200 	int rc;
201 	unsigned int c, p, bytes1 = 0, bytes2 = 0;
202 	struct per_user_data *u = file->private_data;
203 
204 	/* Whole number of ports. */
205 	count &= ~(sizeof(evtchn_port_t)-1);
206 
207 	if (count == 0)
208 		return 0;
209 
210 	if (count > PAGE_SIZE)
211 		count = PAGE_SIZE;
212 
213 	for (;;) {
214 		mutex_lock(&u->ring_cons_mutex);
215 
216 		rc = -EFBIG;
217 		if (u->ring_overflow)
218 			goto unlock_out;
219 
220 		c = u->ring_cons;
221 		p = u->ring_prod;
222 		if (c != p)
223 			break;
224 
225 		mutex_unlock(&u->ring_cons_mutex);
226 
227 		if (file->f_flags & O_NONBLOCK)
228 			return -EAGAIN;
229 
230 		rc = wait_event_interruptible(u->evtchn_wait,
231 					      u->ring_cons != u->ring_prod);
232 		if (rc)
233 			return rc;
234 	}
235 
236 	/* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */
237 	if (((c ^ p) & u->ring_size) != 0) {
238 		bytes1 = (u->ring_size - evtchn_ring_offset(u, c)) *
239 			sizeof(evtchn_port_t);
240 		bytes2 = evtchn_ring_offset(u, p) * sizeof(evtchn_port_t);
241 	} else {
242 		bytes1 = (p - c) * sizeof(evtchn_port_t);
243 		bytes2 = 0;
244 	}
245 
246 	/* Truncate chunks according to caller's maximum byte count. */
247 	if (bytes1 > count) {
248 		bytes1 = count;
249 		bytes2 = 0;
250 	} else if ((bytes1 + bytes2) > count) {
251 		bytes2 = count - bytes1;
252 	}
253 
254 	rc = -EFAULT;
255 	rmb(); /* Ensure that we see the port before we copy it. */
256 	if (copy_to_user(buf, evtchn_ring_entry(u, c), bytes1) ||
257 	    ((bytes2 != 0) &&
258 	     copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))
259 		goto unlock_out;
260 
261 	u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t);
262 	rc = bytes1 + bytes2;
263 
264  unlock_out:
265 	mutex_unlock(&u->ring_cons_mutex);
266 	return rc;
267 }
268 
269 static ssize_t evtchn_write(struct file *file, const char __user *buf,
270 			    size_t count, loff_t *ppos)
271 {
272 	int rc, i;
273 	evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);
274 	struct per_user_data *u = file->private_data;
275 
276 	if (kbuf == NULL)
277 		return -ENOMEM;
278 
279 	/* Whole number of ports. */
280 	count &= ~(sizeof(evtchn_port_t)-1);
281 
282 	rc = 0;
283 	if (count == 0)
284 		goto out;
285 
286 	if (count > PAGE_SIZE)
287 		count = PAGE_SIZE;
288 
289 	rc = -EFAULT;
290 	if (copy_from_user(kbuf, buf, count) != 0)
291 		goto out;
292 
293 	mutex_lock(&u->bind_mutex);
294 
295 	for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {
296 		unsigned port = kbuf[i];
297 		struct user_evtchn *evtchn;
298 
299 		evtchn = find_evtchn(u, port);
300 		if (evtchn && !evtchn->enabled) {
301 			evtchn->enabled = true;
302 			enable_irq(irq_from_evtchn(port));
303 		}
304 	}
305 
306 	mutex_unlock(&u->bind_mutex);
307 
308 	rc = count;
309 
310  out:
311 	free_page((unsigned long)kbuf);
312 	return rc;
313 }
314 
315 static int evtchn_resize_ring(struct per_user_data *u)
316 {
317 	unsigned int new_size;
318 	evtchn_port_t *new_ring, *old_ring;
319 	unsigned int p, c;
320 
321 	/*
322 	 * Ensure the ring is large enough to capture all possible
323 	 * events. i.e., one free slot for each bound event.
324 	 */
325 	if (u->nr_evtchns <= u->ring_size)
326 		return 0;
327 
328 	if (u->ring_size == 0)
329 		new_size = 64;
330 	else
331 		new_size = 2 * u->ring_size;
332 
333 	new_ring = evtchn_alloc_ring(new_size);
334 	if (!new_ring)
335 		return -ENOMEM;
336 
337 	old_ring = u->ring;
338 
339 	/*
340 	 * Access to the ring contents is serialized by either the
341 	 * prod /or/ cons lock so take both when resizing.
342 	 */
343 	mutex_lock(&u->ring_cons_mutex);
344 	spin_lock_irq(&u->ring_prod_lock);
345 
346 	/*
347 	 * Copy the old ring contents to the new ring.
348 	 *
349 	 * If the ring contents crosses the end of the current ring,
350 	 * it needs to be copied in two chunks.
351 	 *
352 	 * +---------+    +------------------+
353 	 * |34567  12| -> |       1234567    |
354 	 * +-----p-c-+    +------------------+
355 	 */
356 	p = evtchn_ring_offset(u, u->ring_prod);
357 	c = evtchn_ring_offset(u, u->ring_cons);
358 	if (p < c) {
359 		memcpy(new_ring + c, u->ring + c, (u->ring_size - c) * sizeof(*u->ring));
360 		memcpy(new_ring + u->ring_size, u->ring, p * sizeof(*u->ring));
361 	} else
362 		memcpy(new_ring + c, u->ring + c, (p - c) * sizeof(*u->ring));
363 
364 	u->ring = new_ring;
365 	u->ring_size = new_size;
366 
367 	spin_unlock_irq(&u->ring_prod_lock);
368 	mutex_unlock(&u->ring_cons_mutex);
369 
370 	evtchn_free_ring(old_ring);
371 
372 	return 0;
373 }
374 
375 static int evtchn_bind_to_user(struct per_user_data *u, int port)
376 {
377 	struct user_evtchn *evtchn;
378 	struct evtchn_close close;
379 	int rc = 0;
380 
381 	/*
382 	 * Ports are never reused, so every caller should pass in a
383 	 * unique port.
384 	 *
385 	 * (Locking not necessary because we haven't registered the
386 	 * interrupt handler yet, and our caller has already
387 	 * serialized bind operations.)
388 	 */
389 
390 	evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
391 	if (!evtchn)
392 		return -ENOMEM;
393 
394 	evtchn->user = u;
395 	evtchn->port = port;
396 	evtchn->enabled = true; /* start enabled */
397 
398 	rc = add_evtchn(u, evtchn);
399 	if (rc < 0)
400 		goto err;
401 
402 	rc = evtchn_resize_ring(u);
403 	if (rc < 0)
404 		goto err;
405 
406 	rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, 0,
407 				       u->name, evtchn);
408 	if (rc < 0)
409 		goto err;
410 
411 	rc = evtchn_make_refcounted(port);
412 	return rc;
413 
414 err:
415 	/* bind failed, should close the port now */
416 	close.port = port;
417 	if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
418 		BUG();
419 	del_evtchn(u, evtchn);
420 	return rc;
421 }
422 
423 static void evtchn_unbind_from_user(struct per_user_data *u,
424 				    struct user_evtchn *evtchn)
425 {
426 	int irq = irq_from_evtchn(evtchn->port);
427 
428 	BUG_ON(irq < 0);
429 
430 	unbind_from_irqhandler(irq, evtchn);
431 
432 	del_evtchn(u, evtchn);
433 }
434 
435 static long evtchn_ioctl(struct file *file,
436 			 unsigned int cmd, unsigned long arg)
437 {
438 	int rc;
439 	struct per_user_data *u = file->private_data;
440 	void __user *uarg = (void __user *) arg;
441 
442 	/* Prevent bind from racing with unbind */
443 	mutex_lock(&u->bind_mutex);
444 
445 	switch (cmd) {
446 	case IOCTL_EVTCHN_BIND_VIRQ: {
447 		struct ioctl_evtchn_bind_virq bind;
448 		struct evtchn_bind_virq bind_virq;
449 
450 		rc = -EFAULT;
451 		if (copy_from_user(&bind, uarg, sizeof(bind)))
452 			break;
453 
454 		bind_virq.virq = bind.virq;
455 		bind_virq.vcpu = 0;
456 		rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
457 						 &bind_virq);
458 		if (rc != 0)
459 			break;
460 
461 		rc = evtchn_bind_to_user(u, bind_virq.port);
462 		if (rc == 0)
463 			rc = bind_virq.port;
464 		break;
465 	}
466 
467 	case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
468 		struct ioctl_evtchn_bind_interdomain bind;
469 		struct evtchn_bind_interdomain bind_interdomain;
470 
471 		rc = -EFAULT;
472 		if (copy_from_user(&bind, uarg, sizeof(bind)))
473 			break;
474 
475 		bind_interdomain.remote_dom  = bind.remote_domain;
476 		bind_interdomain.remote_port = bind.remote_port;
477 		rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
478 						 &bind_interdomain);
479 		if (rc != 0)
480 			break;
481 
482 		rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
483 		if (rc == 0)
484 			rc = bind_interdomain.local_port;
485 		break;
486 	}
487 
488 	case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
489 		struct ioctl_evtchn_bind_unbound_port bind;
490 		struct evtchn_alloc_unbound alloc_unbound;
491 
492 		rc = -EFAULT;
493 		if (copy_from_user(&bind, uarg, sizeof(bind)))
494 			break;
495 
496 		alloc_unbound.dom        = DOMID_SELF;
497 		alloc_unbound.remote_dom = bind.remote_domain;
498 		rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
499 						 &alloc_unbound);
500 		if (rc != 0)
501 			break;
502 
503 		rc = evtchn_bind_to_user(u, alloc_unbound.port);
504 		if (rc == 0)
505 			rc = alloc_unbound.port;
506 		break;
507 	}
508 
509 	case IOCTL_EVTCHN_UNBIND: {
510 		struct ioctl_evtchn_unbind unbind;
511 		struct user_evtchn *evtchn;
512 
513 		rc = -EFAULT;
514 		if (copy_from_user(&unbind, uarg, sizeof(unbind)))
515 			break;
516 
517 		rc = -EINVAL;
518 		if (unbind.port >= xen_evtchn_nr_channels())
519 			break;
520 
521 		rc = -ENOTCONN;
522 		evtchn = find_evtchn(u, unbind.port);
523 		if (!evtchn)
524 			break;
525 
526 		disable_irq(irq_from_evtchn(unbind.port));
527 		evtchn_unbind_from_user(u, evtchn);
528 		rc = 0;
529 		break;
530 	}
531 
532 	case IOCTL_EVTCHN_NOTIFY: {
533 		struct ioctl_evtchn_notify notify;
534 		struct user_evtchn *evtchn;
535 
536 		rc = -EFAULT;
537 		if (copy_from_user(&notify, uarg, sizeof(notify)))
538 			break;
539 
540 		rc = -ENOTCONN;
541 		evtchn = find_evtchn(u, notify.port);
542 		if (evtchn) {
543 			notify_remote_via_evtchn(notify.port);
544 			rc = 0;
545 		}
546 		break;
547 	}
548 
549 	case IOCTL_EVTCHN_RESET: {
550 		/* Initialise the ring to empty. Clear errors. */
551 		mutex_lock(&u->ring_cons_mutex);
552 		spin_lock_irq(&u->ring_prod_lock);
553 		u->ring_cons = u->ring_prod = u->ring_overflow = 0;
554 		spin_unlock_irq(&u->ring_prod_lock);
555 		mutex_unlock(&u->ring_cons_mutex);
556 		rc = 0;
557 		break;
558 	}
559 
560 	default:
561 		rc = -ENOSYS;
562 		break;
563 	}
564 	mutex_unlock(&u->bind_mutex);
565 
566 	return rc;
567 }
568 
569 static unsigned int evtchn_poll(struct file *file, poll_table *wait)
570 {
571 	unsigned int mask = POLLOUT | POLLWRNORM;
572 	struct per_user_data *u = file->private_data;
573 
574 	poll_wait(file, &u->evtchn_wait, wait);
575 	if (u->ring_cons != u->ring_prod)
576 		mask |= POLLIN | POLLRDNORM;
577 	if (u->ring_overflow)
578 		mask = POLLERR;
579 	return mask;
580 }
581 
582 static int evtchn_fasync(int fd, struct file *filp, int on)
583 {
584 	struct per_user_data *u = filp->private_data;
585 	return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
586 }
587 
588 static int evtchn_open(struct inode *inode, struct file *filp)
589 {
590 	struct per_user_data *u;
591 
592 	u = kzalloc(sizeof(*u), GFP_KERNEL);
593 	if (u == NULL)
594 		return -ENOMEM;
595 
596 	u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
597 	if (u->name == NULL) {
598 		kfree(u);
599 		return -ENOMEM;
600 	}
601 
602 	init_waitqueue_head(&u->evtchn_wait);
603 
604 	mutex_init(&u->bind_mutex);
605 	mutex_init(&u->ring_cons_mutex);
606 	spin_lock_init(&u->ring_prod_lock);
607 
608 	filp->private_data = u;
609 
610 	return nonseekable_open(inode, filp);
611 }
612 
613 static int evtchn_release(struct inode *inode, struct file *filp)
614 {
615 	struct per_user_data *u = filp->private_data;
616 	struct rb_node *node;
617 
618 	while ((node = u->evtchns.rb_node)) {
619 		struct user_evtchn *evtchn;
620 
621 		evtchn = rb_entry(node, struct user_evtchn, node);
622 		disable_irq(irq_from_evtchn(evtchn->port));
623 		evtchn_unbind_from_user(u, evtchn);
624 	}
625 
626 	evtchn_free_ring(u->ring);
627 	kfree(u->name);
628 	kfree(u);
629 
630 	return 0;
631 }
632 
633 static const struct file_operations evtchn_fops = {
634 	.owner   = THIS_MODULE,
635 	.read    = evtchn_read,
636 	.write   = evtchn_write,
637 	.unlocked_ioctl = evtchn_ioctl,
638 	.poll    = evtchn_poll,
639 	.fasync  = evtchn_fasync,
640 	.open    = evtchn_open,
641 	.release = evtchn_release,
642 	.llseek	 = no_llseek,
643 };
644 
645 static struct miscdevice evtchn_miscdev = {
646 	.minor        = MISC_DYNAMIC_MINOR,
647 	.name         = "xen/evtchn",
648 	.fops         = &evtchn_fops,
649 };
650 static int __init evtchn_init(void)
651 {
652 	int err;
653 
654 	if (!xen_domain())
655 		return -ENODEV;
656 
657 	/* Create '/dev/xen/evtchn'. */
658 	err = misc_register(&evtchn_miscdev);
659 	if (err != 0) {
660 		pr_err("Could not register /dev/xen/evtchn\n");
661 		return err;
662 	}
663 
664 	pr_info("Event-channel device installed\n");
665 
666 	return 0;
667 }
668 
669 static void __exit evtchn_cleanup(void)
670 {
671 	misc_deregister(&evtchn_miscdev);
672 }
673 
674 module_init(evtchn_init);
675 module_exit(evtchn_cleanup);
676 
677 MODULE_LICENSE("GPL");
678