drivers/xen/evtchn.c

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