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
60struct 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
78struct user_evtchn {
79	struct rb_node node;
80	struct per_user_data *user;
81	unsigned port;
82	bool enabled;
83};
84
85static 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
97static void evtchn_free_ring(evtchn_port_t *ring)
98{
99	kvfree(ring);
100}
101
102static unsigned int evtchn_ring_offset(struct per_user_data *u,
103				       unsigned int idx)
104{
105	return idx & (u->ring_size - 1);
106}
107
108static 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
114static 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
141static 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
148static 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
167static 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
197static 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
269static 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
315static int evtchn_resize_ring(struct per_user_data *u)
316{
317	unsigned int new_size;
318	evtchn_port_t *new_ring, *old_ring;
319
320	/*
321	 * Ensure the ring is large enough to capture all possible
322	 * events. i.e., one free slot for each bound event.
323	 */
324	if (u->nr_evtchns <= u->ring_size)
325		return 0;
326
327	if (u->ring_size == 0)
328		new_size = 64;
329	else
330		new_size = 2 * u->ring_size;
331
332	new_ring = evtchn_alloc_ring(new_size);
333	if (!new_ring)
334		return -ENOMEM;
335
336	old_ring = u->ring;
337
338	/*
339	 * Access to the ring contents is serialized by either the
340	 * prod /or/ cons lock so take both when resizing.
341	 */
342	mutex_lock(&u->ring_cons_mutex);
343	spin_lock_irq(&u->ring_prod_lock);
344
345	/*
346	 * Copy the old ring contents to the new ring.
347	 *
348	 * To take care of wrapping, a full ring, and the new index
349	 * pointing into the second half, simply copy the old contents
350	 * twice.
351	 *
352	 * +---------+    +------------------+
353	 * |34567  12| -> |34567  1234567  12|
354	 * +-----p-c-+    +-------c------p---+
355	 */
356	memcpy(new_ring, old_ring, u->ring_size * sizeof(*u->ring));
357	memcpy(new_ring + u->ring_size, old_ring,
358	       u->ring_size * sizeof(*u->ring));
359
360	u->ring = new_ring;
361	u->ring_size = new_size;
362
363	spin_unlock_irq(&u->ring_prod_lock);
364	mutex_unlock(&u->ring_cons_mutex);
365
366	evtchn_free_ring(old_ring);
367
368	return 0;
369}
370
371static int evtchn_bind_to_user(struct per_user_data *u, int port)
372{
373	struct user_evtchn *evtchn;
374	struct evtchn_close close;
375	int rc = 0;
376
377	/*
378	 * Ports are never reused, so every caller should pass in a
379	 * unique port.
380	 *
381	 * (Locking not necessary because we haven't registered the
382	 * interrupt handler yet, and our caller has already
383	 * serialized bind operations.)
384	 */
385
386	evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);
387	if (!evtchn)
388		return -ENOMEM;
389
390	evtchn->user = u;
391	evtchn->port = port;
392	evtchn->enabled = true; /* start enabled */
393
394	rc = add_evtchn(u, evtchn);
395	if (rc < 0)
396		goto err;
397
398	rc = evtchn_resize_ring(u);
399	if (rc < 0)
400		goto err;
401
402	rc = bind_evtchn_to_irqhandler(port, evtchn_interrupt, 0,
403				       u->name, evtchn);
404	if (rc < 0)
405		goto err;
406
407	rc = evtchn_make_refcounted(port);
408	return rc;
409
410err:
411	/* bind failed, should close the port now */
412	close.port = port;
413	if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close) != 0)
414		BUG();
415	del_evtchn(u, evtchn);
416	return rc;
417}
418
419static void evtchn_unbind_from_user(struct per_user_data *u,
420				    struct user_evtchn *evtchn)
421{
422	int irq = irq_from_evtchn(evtchn->port);
423
424	BUG_ON(irq < 0);
425
426	unbind_from_irqhandler(irq, evtchn);
427
428	del_evtchn(u, evtchn);
429}
430
431static long evtchn_ioctl(struct file *file,
432			 unsigned int cmd, unsigned long arg)
433{
434	int rc;
435	struct per_user_data *u = file->private_data;
436	void __user *uarg = (void __user *) arg;
437
438	/* Prevent bind from racing with unbind */
439	mutex_lock(&u->bind_mutex);
440
441	switch (cmd) {
442	case IOCTL_EVTCHN_BIND_VIRQ: {
443		struct ioctl_evtchn_bind_virq bind;
444		struct evtchn_bind_virq bind_virq;
445
446		rc = -EFAULT;
447		if (copy_from_user(&bind, uarg, sizeof(bind)))
448			break;
449
450		bind_virq.virq = bind.virq;
451		bind_virq.vcpu = 0;
452		rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
453						 &bind_virq);
454		if (rc != 0)
455			break;
456
457		rc = evtchn_bind_to_user(u, bind_virq.port);
458		if (rc == 0)
459			rc = bind_virq.port;
460		break;
461	}
462
463	case IOCTL_EVTCHN_BIND_INTERDOMAIN: {
464		struct ioctl_evtchn_bind_interdomain bind;
465		struct evtchn_bind_interdomain bind_interdomain;
466
467		rc = -EFAULT;
468		if (copy_from_user(&bind, uarg, sizeof(bind)))
469			break;
470
471		bind_interdomain.remote_dom  = bind.remote_domain;
472		bind_interdomain.remote_port = bind.remote_port;
473		rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
474						 &bind_interdomain);
475		if (rc != 0)
476			break;
477
478		rc = evtchn_bind_to_user(u, bind_interdomain.local_port);
479		if (rc == 0)
480			rc = bind_interdomain.local_port;
481		break;
482	}
483
484	case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {
485		struct ioctl_evtchn_bind_unbound_port bind;
486		struct evtchn_alloc_unbound alloc_unbound;
487
488		rc = -EFAULT;
489		if (copy_from_user(&bind, uarg, sizeof(bind)))
490			break;
491
492		alloc_unbound.dom        = DOMID_SELF;
493		alloc_unbound.remote_dom = bind.remote_domain;
494		rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
495						 &alloc_unbound);
496		if (rc != 0)
497			break;
498
499		rc = evtchn_bind_to_user(u, alloc_unbound.port);
500		if (rc == 0)
501			rc = alloc_unbound.port;
502		break;
503	}
504
505	case IOCTL_EVTCHN_UNBIND: {
506		struct ioctl_evtchn_unbind unbind;
507		struct user_evtchn *evtchn;
508
509		rc = -EFAULT;
510		if (copy_from_user(&unbind, uarg, sizeof(unbind)))
511			break;
512
513		rc = -EINVAL;
514		if (unbind.port >= xen_evtchn_nr_channels())
515			break;
516
517		rc = -ENOTCONN;
518		evtchn = find_evtchn(u, unbind.port);
519		if (!evtchn)
520			break;
521
522		disable_irq(irq_from_evtchn(unbind.port));
523		evtchn_unbind_from_user(u, evtchn);
524		rc = 0;
525		break;
526	}
527
528	case IOCTL_EVTCHN_NOTIFY: {
529		struct ioctl_evtchn_notify notify;
530		struct user_evtchn *evtchn;
531
532		rc = -EFAULT;
533		if (copy_from_user(&notify, uarg, sizeof(notify)))
534			break;
535
536		rc = -ENOTCONN;
537		evtchn = find_evtchn(u, notify.port);
538		if (evtchn) {
539			notify_remote_via_evtchn(notify.port);
540			rc = 0;
541		}
542		break;
543	}
544
545	case IOCTL_EVTCHN_RESET: {
546		/* Initialise the ring to empty. Clear errors. */
547		mutex_lock(&u->ring_cons_mutex);
548		spin_lock_irq(&u->ring_prod_lock);
549		u->ring_cons = u->ring_prod = u->ring_overflow = 0;
550		spin_unlock_irq(&u->ring_prod_lock);
551		mutex_unlock(&u->ring_cons_mutex);
552		rc = 0;
553		break;
554	}
555
556	default:
557		rc = -ENOSYS;
558		break;
559	}
560	mutex_unlock(&u->bind_mutex);
561
562	return rc;
563}
564
565static unsigned int evtchn_poll(struct file *file, poll_table *wait)
566{
567	unsigned int mask = POLLOUT | POLLWRNORM;
568	struct per_user_data *u = file->private_data;
569
570	poll_wait(file, &u->evtchn_wait, wait);
571	if (u->ring_cons != u->ring_prod)
572		mask |= POLLIN | POLLRDNORM;
573	if (u->ring_overflow)
574		mask = POLLERR;
575	return mask;
576}
577
578static int evtchn_fasync(int fd, struct file *filp, int on)
579{
580	struct per_user_data *u = filp->private_data;
581	return fasync_helper(fd, filp, on, &u->evtchn_async_queue);
582}
583
584static int evtchn_open(struct inode *inode, struct file *filp)
585{
586	struct per_user_data *u;
587
588	u = kzalloc(sizeof(*u), GFP_KERNEL);
589	if (u == NULL)
590		return -ENOMEM;
591
592	u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);
593	if (u->name == NULL) {
594		kfree(u);
595		return -ENOMEM;
596	}
597
598	init_waitqueue_head(&u->evtchn_wait);
599
600	mutex_init(&u->bind_mutex);
601	mutex_init(&u->ring_cons_mutex);
602	spin_lock_init(&u->ring_prod_lock);
603
604	filp->private_data = u;
605
606	return nonseekable_open(inode, filp);
607}
608
609static int evtchn_release(struct inode *inode, struct file *filp)
610{
611	struct per_user_data *u = filp->private_data;
612	struct rb_node *node;
613
614	while ((node = u->evtchns.rb_node)) {
615		struct user_evtchn *evtchn;
616
617		evtchn = rb_entry(node, struct user_evtchn, node);
618		disable_irq(irq_from_evtchn(evtchn->port));
619		evtchn_unbind_from_user(u, evtchn);
620	}
621
622	evtchn_free_ring(u->ring);
623	kfree(u->name);
624	kfree(u);
625
626	return 0;
627}
628
629static const struct file_operations evtchn_fops = {
630	.owner   = THIS_MODULE,
631	.read    = evtchn_read,
632	.write   = evtchn_write,
633	.unlocked_ioctl = evtchn_ioctl,
634	.poll    = evtchn_poll,
635	.fasync  = evtchn_fasync,
636	.open    = evtchn_open,
637	.release = evtchn_release,
638	.llseek	 = no_llseek,
639};
640
641static struct miscdevice evtchn_miscdev = {
642	.minor        = MISC_DYNAMIC_MINOR,
643	.name         = "xen/evtchn",
644	.fops         = &evtchn_fops,
645};
646static int __init evtchn_init(void)
647{
648	int err;
649
650	if (!xen_domain())
651		return -ENODEV;
652
653	/* Create '/dev/xen/evtchn'. */
654	err = misc_register(&evtchn_miscdev);
655	if (err != 0) {
656		pr_err("Could not register /dev/xen/evtchn\n");
657		return err;
658	}
659
660	pr_info("Event-channel device installed\n");
661
662	return 0;
663}
664
665static void __exit evtchn_cleanup(void)
666{
667	misc_deregister(&evtchn_miscdev);
668}
669
670module_init(evtchn_init);
671module_exit(evtchn_cleanup);
672
673MODULE_LICENSE("GPL");
674