733 lines · c
1/******************************************************************************2 * evtchn.c3 *4 * Driver for receiving and demuxing event-channel signals.5 *6 * Copyright (c) 2004-2005, K A Fraser7 * Multi-process extensions Copyright (c) 2004, Steven Smith8 *9 * This program is free software; you can redistribute it and/or10 * modify it under the terms of the GNU General Public License version 211 * as published by the Free Software Foundation; or, when distributed12 * separately from the Linux kernel or incorporated into other13 * software packages, subject to the following license:14 *15 * Permission is hereby granted, free of charge, to any person obtaining a copy16 * of this source file (the "Software"), to deal in the Software without17 * 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 to20 * the following conditions:21 *22 * The above copyright notice and this permission notice shall be included in23 * all copies or substantial portions of the Software.24 *25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR26 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,27 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE28 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER29 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING30 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS31 * IN THE SOFTWARE.32 */33 34#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt35 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 bool unbinding;89};90 91static void evtchn_free_ring(evtchn_port_t *ring)92{93 kvfree(ring);94}95 96static unsigned int evtchn_ring_offset(struct per_user_data *u,97 unsigned int idx)98{99 return idx & (u->ring_size - 1);100}101 102static evtchn_port_t *evtchn_ring_entry(struct per_user_data *u,103 unsigned int idx)104{105 return u->ring + evtchn_ring_offset(u, idx);106}107 108static int add_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)109{110 struct rb_node **new = &(u->evtchns.rb_node), *parent = NULL;111 112 u->nr_evtchns++;113 114 while (*new) {115 struct user_evtchn *this;116 117 this = rb_entry(*new, struct user_evtchn, node);118 119 parent = *new;120 if (this->port < evtchn->port)121 new = &((*new)->rb_left);122 else if (this->port > evtchn->port)123 new = &((*new)->rb_right);124 else125 return -EEXIST;126 }127 128 /* Add new node and rebalance tree. */129 rb_link_node(&evtchn->node, parent, new);130 rb_insert_color(&evtchn->node, &u->evtchns);131 132 return 0;133}134 135static void del_evtchn(struct per_user_data *u, struct user_evtchn *evtchn)136{137 u->nr_evtchns--;138 rb_erase(&evtchn->node, &u->evtchns);139 kfree(evtchn);140}141 142static struct user_evtchn *find_evtchn(struct per_user_data *u,143 evtchn_port_t port)144{145 struct rb_node *node = u->evtchns.rb_node;146 147 while (node) {148 struct user_evtchn *evtchn;149 150 evtchn = rb_entry(node, struct user_evtchn, node);151 152 if (evtchn->port < port)153 node = node->rb_left;154 else if (evtchn->port > port)155 node = node->rb_right;156 else157 return evtchn;158 }159 return NULL;160}161 162static irqreturn_t evtchn_interrupt(int irq, void *data)163{164 struct user_evtchn *evtchn = data;165 struct per_user_data *u = evtchn->user;166 unsigned int prod, cons;167 168 /* Handler might be called when tearing down the IRQ. */169 if (evtchn->unbinding)170 return IRQ_HANDLED;171 172 WARN(!evtchn->enabled,173 "Interrupt for port %u, but apparently not enabled; per-user %p\n",174 evtchn->port, u);175 176 evtchn->enabled = false;177 178 spin_lock(&u->ring_prod_lock);179 180 prod = READ_ONCE(u->ring_prod);181 cons = READ_ONCE(u->ring_cons);182 183 if ((prod - cons) < u->ring_size) {184 *evtchn_ring_entry(u, prod) = evtchn->port;185 smp_wmb(); /* Ensure ring contents visible */186 WRITE_ONCE(u->ring_prod, prod + 1);187 if (cons == prod) {188 wake_up_interruptible(&u->evtchn_wait);189 kill_fasync(&u->evtchn_async_queue,190 SIGIO, POLL_IN);191 }192 } else193 u->ring_overflow = 1;194 195 spin_unlock(&u->ring_prod_lock);196 197 return IRQ_HANDLED;198}199 200static ssize_t evtchn_read(struct file *file, char __user *buf,201 size_t count, loff_t *ppos)202{203 int rc;204 unsigned int c, p, bytes1 = 0, bytes2 = 0;205 struct per_user_data *u = file->private_data;206 207 /* Whole number of ports. */208 count &= ~(sizeof(evtchn_port_t)-1);209 210 if (count == 0)211 return 0;212 213 if (count > PAGE_SIZE)214 count = PAGE_SIZE;215 216 for (;;) {217 mutex_lock(&u->ring_cons_mutex);218 219 rc = -EFBIG;220 if (u->ring_overflow)221 goto unlock_out;222 223 c = READ_ONCE(u->ring_cons);224 p = READ_ONCE(u->ring_prod);225 if (c != p)226 break;227 228 mutex_unlock(&u->ring_cons_mutex);229 230 if (file->f_flags & O_NONBLOCK)231 return -EAGAIN;232 233 rc = wait_event_interruptible(u->evtchn_wait,234 READ_ONCE(u->ring_cons) != READ_ONCE(u->ring_prod));235 if (rc)236 return rc;237 }238 239 /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */240 if (((c ^ p) & u->ring_size) != 0) {241 bytes1 = (u->ring_size - evtchn_ring_offset(u, c)) *242 sizeof(evtchn_port_t);243 bytes2 = evtchn_ring_offset(u, p) * sizeof(evtchn_port_t);244 } else {245 bytes1 = (p - c) * sizeof(evtchn_port_t);246 bytes2 = 0;247 }248 249 /* Truncate chunks according to caller's maximum byte count. */250 if (bytes1 > count) {251 bytes1 = count;252 bytes2 = 0;253 } else if ((bytes1 + bytes2) > count) {254 bytes2 = count - bytes1;255 }256 257 rc = -EFAULT;258 smp_rmb(); /* Ensure that we see the port before we copy it. */259 if (copy_to_user(buf, evtchn_ring_entry(u, c), bytes1) ||260 ((bytes2 != 0) &&261 copy_to_user(&buf[bytes1], &u->ring[0], bytes2)))262 goto unlock_out;263 264 WRITE_ONCE(u->ring_cons, c + (bytes1 + bytes2) / sizeof(evtchn_port_t));265 rc = bytes1 + bytes2;266 267 unlock_out:268 mutex_unlock(&u->ring_cons_mutex);269 return rc;270}271 272static ssize_t evtchn_write(struct file *file, const char __user *buf,273 size_t count, loff_t *ppos)274{275 int rc, i;276 evtchn_port_t *kbuf = (evtchn_port_t *)__get_free_page(GFP_KERNEL);277 struct per_user_data *u = file->private_data;278 279 if (kbuf == NULL)280 return -ENOMEM;281 282 /* Whole number of ports. */283 count &= ~(sizeof(evtchn_port_t)-1);284 285 rc = 0;286 if (count == 0)287 goto out;288 289 if (count > PAGE_SIZE)290 count = PAGE_SIZE;291 292 rc = -EFAULT;293 if (copy_from_user(kbuf, buf, count) != 0)294 goto out;295 296 mutex_lock(&u->bind_mutex);297 298 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) {299 evtchn_port_t port = kbuf[i];300 struct user_evtchn *evtchn;301 302 evtchn = find_evtchn(u, port);303 if (evtchn && !evtchn->enabled) {304 evtchn->enabled = true;305 xen_irq_lateeoi(irq_from_evtchn(port), 0);306 }307 }308 309 mutex_unlock(&u->bind_mutex);310 311 rc = count;312 313 out:314 free_page((unsigned long)kbuf);315 return rc;316}317 318static int evtchn_resize_ring(struct per_user_data *u)319{320 unsigned int new_size;321 evtchn_port_t *new_ring, *old_ring;322 323 /*324 * Ensure the ring is large enough to capture all possible325 * events. i.e., one free slot for each bound event.326 */327 if (u->nr_evtchns <= u->ring_size)328 return 0;329 330 if (u->ring_size == 0)331 new_size = 64;332 else333 new_size = 2 * u->ring_size;334 335 new_ring = kvmalloc_array(new_size, sizeof(*new_ring), GFP_KERNEL);336 if (!new_ring)337 return -ENOMEM;338 339 old_ring = u->ring;340 341 /*342 * Access to the ring contents is serialized by either the343 * prod /or/ cons lock so take both when resizing.344 */345 mutex_lock(&u->ring_cons_mutex);346 spin_lock_irq(&u->ring_prod_lock);347 348 /*349 * Copy the old ring contents to the new ring.350 *351 * To take care of wrapping, a full ring, and the new index352 * pointing into the second half, simply copy the old contents353 * twice.354 *355 * +---------+ +------------------+356 * |34567 12| -> |34567 1234567 12|357 * +-----p-c-+ +-------c------p---+358 */359 memcpy(new_ring, old_ring, u->ring_size * sizeof(*u->ring));360 memcpy(new_ring + u->ring_size, old_ring,361 u->ring_size * sizeof(*u->ring));362 363 u->ring = new_ring;364 u->ring_size = new_size;365 366 spin_unlock_irq(&u->ring_prod_lock);367 mutex_unlock(&u->ring_cons_mutex);368 369 evtchn_free_ring(old_ring);370 371 return 0;372}373 374static int evtchn_bind_to_user(struct per_user_data *u, evtchn_port_t port,375 bool is_static)376{377 struct user_evtchn *evtchn;378 int rc = 0;379 380 /*381 * Ports are never reused, so every caller should pass in a382 * unique port.383 *384 * (Locking not necessary because we haven't registered the385 * interrupt handler yet, and our caller has already386 * serialized bind operations.)387 */388 389 evtchn = kzalloc(sizeof(*evtchn), GFP_KERNEL);390 if (!evtchn)391 return -ENOMEM;392 393 evtchn->user = u;394 evtchn->port = port;395 evtchn->enabled = true; /* start enabled */396 397 rc = add_evtchn(u, evtchn);398 if (rc < 0)399 goto err;400 401 rc = evtchn_resize_ring(u);402 if (rc < 0)403 goto err;404 405 rc = bind_evtchn_to_irqhandler_lateeoi(port, evtchn_interrupt, IRQF_SHARED,406 u->name, evtchn);407 if (rc < 0)408 goto err;409 410 rc = evtchn_make_refcounted(port, is_static);411 return rc;412 413err:414 /* bind failed, should close the port now */415 if (!is_static)416 xen_evtchn_close(port);417 418 del_evtchn(u, evtchn);419 return rc;420}421 422static void evtchn_unbind_from_user(struct per_user_data *u,423 struct user_evtchn *evtchn)424{425 int irq = irq_from_evtchn(evtchn->port);426 427 BUG_ON(irq < 0);428 429 evtchn->unbinding = true;430 unbind_from_irqhandler(irq, evtchn);431 432 del_evtchn(u, evtchn);433}434 435static 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 = -EACCES;451 if (u->restrict_domid != UNRESTRICTED_DOMID)452 break;453 454 rc = -EFAULT;455 if (copy_from_user(&bind, uarg, sizeof(bind)))456 break;457 458 bind_virq.virq = bind.virq;459 bind_virq.vcpu = xen_vcpu_nr(0);460 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,461 &bind_virq);462 if (rc != 0)463 break;464 465 rc = evtchn_bind_to_user(u, bind_virq.port, false);466 if (rc == 0)467 rc = bind_virq.port;468 break;469 }470 471 case IOCTL_EVTCHN_BIND_INTERDOMAIN: {472 struct ioctl_evtchn_bind_interdomain bind;473 struct evtchn_bind_interdomain bind_interdomain;474 475 rc = -EFAULT;476 if (copy_from_user(&bind, uarg, sizeof(bind)))477 break;478 479 rc = -EACCES;480 if (u->restrict_domid != UNRESTRICTED_DOMID &&481 u->restrict_domid != bind.remote_domain)482 break;483 484 bind_interdomain.remote_dom = bind.remote_domain;485 bind_interdomain.remote_port = bind.remote_port;486 rc = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,487 &bind_interdomain);488 if (rc != 0)489 break;490 491 rc = evtchn_bind_to_user(u, bind_interdomain.local_port, false);492 if (rc == 0)493 rc = bind_interdomain.local_port;494 break;495 }496 497 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: {498 struct ioctl_evtchn_bind_unbound_port bind;499 struct evtchn_alloc_unbound alloc_unbound;500 501 rc = -EACCES;502 if (u->restrict_domid != UNRESTRICTED_DOMID)503 break;504 505 rc = -EFAULT;506 if (copy_from_user(&bind, uarg, sizeof(bind)))507 break;508 509 alloc_unbound.dom = DOMID_SELF;510 alloc_unbound.remote_dom = bind.remote_domain;511 rc = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,512 &alloc_unbound);513 if (rc != 0)514 break;515 516 rc = evtchn_bind_to_user(u, alloc_unbound.port, false);517 if (rc == 0)518 rc = alloc_unbound.port;519 break;520 }521 522 case IOCTL_EVTCHN_UNBIND: {523 struct ioctl_evtchn_unbind unbind;524 struct user_evtchn *evtchn;525 526 rc = -EFAULT;527 if (copy_from_user(&unbind, uarg, sizeof(unbind)))528 break;529 530 rc = -EINVAL;531 if (unbind.port >= xen_evtchn_nr_channels())532 break;533 534 rc = -ENOTCONN;535 evtchn = find_evtchn(u, unbind.port);536 if (!evtchn)537 break;538 539 disable_irq(irq_from_evtchn(unbind.port));540 evtchn_unbind_from_user(u, evtchn);541 rc = 0;542 break;543 }544 545 case IOCTL_EVTCHN_BIND_STATIC: {546 struct ioctl_evtchn_bind bind;547 struct user_evtchn *evtchn;548 549 rc = -EFAULT;550 if (copy_from_user(&bind, uarg, sizeof(bind)))551 break;552 553 rc = -EISCONN;554 evtchn = find_evtchn(u, bind.port);555 if (evtchn)556 break;557 558 rc = evtchn_bind_to_user(u, bind.port, true);559 break;560 }561 562 case IOCTL_EVTCHN_NOTIFY: {563 struct ioctl_evtchn_notify notify;564 struct user_evtchn *evtchn;565 566 rc = -EFAULT;567 if (copy_from_user(¬ify, uarg, sizeof(notify)))568 break;569 570 rc = -ENOTCONN;571 evtchn = find_evtchn(u, notify.port);572 if (evtchn) {573 notify_remote_via_evtchn(notify.port);574 rc = 0;575 }576 break;577 }578 579 case IOCTL_EVTCHN_RESET: {580 /* Initialise the ring to empty. Clear errors. */581 mutex_lock(&u->ring_cons_mutex);582 spin_lock_irq(&u->ring_prod_lock);583 WRITE_ONCE(u->ring_cons, 0);584 WRITE_ONCE(u->ring_prod, 0);585 u->ring_overflow = 0;586 spin_unlock_irq(&u->ring_prod_lock);587 mutex_unlock(&u->ring_cons_mutex);588 rc = 0;589 break;590 }591 592 case IOCTL_EVTCHN_RESTRICT_DOMID: {593 struct ioctl_evtchn_restrict_domid ierd;594 595 rc = -EACCES;596 if (u->restrict_domid != UNRESTRICTED_DOMID)597 break;598 599 rc = -EFAULT;600 if (copy_from_user(&ierd, uarg, sizeof(ierd)))601 break;602 603 rc = -EINVAL;604 if (ierd.domid == 0 || ierd.domid >= DOMID_FIRST_RESERVED)605 break;606 607 u->restrict_domid = ierd.domid;608 rc = 0;609 610 break;611 }612 613 default:614 rc = -ENOSYS;615 break;616 }617 mutex_unlock(&u->bind_mutex);618 619 return rc;620}621 622static __poll_t evtchn_poll(struct file *file, poll_table *wait)623{624 __poll_t mask = EPOLLOUT | EPOLLWRNORM;625 struct per_user_data *u = file->private_data;626 627 poll_wait(file, &u->evtchn_wait, wait);628 if (READ_ONCE(u->ring_cons) != READ_ONCE(u->ring_prod))629 mask |= EPOLLIN | EPOLLRDNORM;630 if (u->ring_overflow)631 mask = EPOLLERR;632 return mask;633}634 635static int evtchn_fasync(int fd, struct file *filp, int on)636{637 struct per_user_data *u = filp->private_data;638 return fasync_helper(fd, filp, on, &u->evtchn_async_queue);639}640 641static int evtchn_open(struct inode *inode, struct file *filp)642{643 struct per_user_data *u;644 645 u = kzalloc(sizeof(*u), GFP_KERNEL);646 if (u == NULL)647 return -ENOMEM;648 649 u->name = kasprintf(GFP_KERNEL, "evtchn:%s", current->comm);650 if (u->name == NULL) {651 kfree(u);652 return -ENOMEM;653 }654 655 init_waitqueue_head(&u->evtchn_wait);656 657 mutex_init(&u->bind_mutex);658 mutex_init(&u->ring_cons_mutex);659 spin_lock_init(&u->ring_prod_lock);660 661 u->restrict_domid = UNRESTRICTED_DOMID;662 663 filp->private_data = u;664 665 return stream_open(inode, filp);666}667 668static int evtchn_release(struct inode *inode, struct file *filp)669{670 struct per_user_data *u = filp->private_data;671 struct rb_node *node;672 673 while ((node = u->evtchns.rb_node)) {674 struct user_evtchn *evtchn;675 676 evtchn = rb_entry(node, struct user_evtchn, node);677 disable_irq(irq_from_evtchn(evtchn->port));678 evtchn_unbind_from_user(u, evtchn);679 }680 681 evtchn_free_ring(u->ring);682 kfree(u->name);683 kfree(u);684 685 return 0;686}687 688static const struct file_operations evtchn_fops = {689 .owner = THIS_MODULE,690 .read = evtchn_read,691 .write = evtchn_write,692 .unlocked_ioctl = evtchn_ioctl,693 .poll = evtchn_poll,694 .fasync = evtchn_fasync,695 .open = evtchn_open,696 .release = evtchn_release,697};698 699static struct miscdevice evtchn_miscdev = {700 .minor = MISC_DYNAMIC_MINOR,701 .name = "xen/evtchn",702 .fops = &evtchn_fops,703};704static int __init evtchn_init(void)705{706 int err;707 708 if (!xen_domain())709 return -ENODEV;710 711 /* Create '/dev/xen/evtchn'. */712 err = misc_register(&evtchn_miscdev);713 if (err != 0) {714 pr_err("Could not register /dev/xen/evtchn\n");715 return err;716 }717 718 pr_info("Event-channel device installed\n");719 720 return 0;721}722 723static void __exit evtchn_cleanup(void)724{725 misc_deregister(&evtchn_miscdev);726}727 728module_init(evtchn_init);729module_exit(evtchn_cleanup);730 731MODULE_DESCRIPTION("Xen /dev/xen/evtchn device driver");732MODULE_LICENSE("GPL");733