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 * FreeBSD port Copyright (c) 2014, Roger Pau Monné 9 * Fetched from git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git 10 * File: drivers/xen/evtchn.c 11 * Git commit: 0dc0064add422bc0ef5165ebe9ece3052bbd457d 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License version 2 15 * as published by the Free Software Foundation; or, when distributed 16 * separately from the Linux kernel or incorporated into other 17 * software packages, subject to the following license: 18 * 19 * Permission is hereby granted, free of charge, to any person obtaining a copy 20 * of this source file (the "Software"), to deal in the Software without 21 * restriction, including without limitation the rights to use, copy, modify, 22 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 23 * and to permit persons to whom the Software is furnished to do so, subject to 24 * the following conditions: 25 * 26 * The above copyright notice and this permission notice shall be included in 27 * all copies or substantial portions of the Software. 28 * 29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 35 * IN THE SOFTWARE. 36 */ 37 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/uio.h> 44 #include <sys/bus.h> 45 #include <sys/malloc.h> 46 #include <sys/kernel.h> 47 #include <sys/lock.h> 48 #include <sys/mutex.h> 49 #include <sys/sx.h> 50 #include <sys/selinfo.h> 51 #include <sys/poll.h> 52 #include <sys/conf.h> 53 #include <sys/fcntl.h> 54 #include <sys/ioccom.h> 55 #include <sys/rman.h> 56 #include <sys/tree.h> 57 #include <sys/module.h> 58 #include <sys/filio.h> 59 #include <sys/vnode.h> 60 61 #include <machine/intr_machdep.h> 62 #include <machine/xen/synch_bitops.h> 63 64 #include <xen/xen-os.h> 65 #include <xen/evtchn.h> 66 #include <xen/xen_intr.h> 67 68 #include <xen/evtchn/evtchnvar.h> 69 70 MALLOC_DEFINE(M_EVTCHN, "evtchn_dev", "Xen event channel user-space device"); 71 72 struct user_evtchn; 73 74 static int evtchn_cmp(struct user_evtchn *u1, struct user_evtchn *u2); 75 76 RB_HEAD(evtchn_tree, user_evtchn); 77 78 struct per_user_data { 79 struct mtx bind_mutex; /* serialize bind/unbind operations */ 80 struct evtchn_tree evtchns; 81 82 /* Notification ring, accessed via /dev/xen/evtchn. */ 83 #define EVTCHN_RING_SIZE (PAGE_SIZE / sizeof(evtchn_port_t)) 84 #define EVTCHN_RING_MASK(_i) ((_i)&(EVTCHN_RING_SIZE-1)) 85 evtchn_port_t *ring; 86 unsigned int ring_cons, ring_prod, ring_overflow; 87 struct sx ring_cons_mutex; /* protect against concurrent readers */ 88 struct mtx ring_prod_mutex; /* product against concurrent interrupts */ 89 struct selinfo ev_rsel; 90 }; 91 92 struct user_evtchn { 93 RB_ENTRY(user_evtchn) node; 94 struct per_user_data *user; 95 evtchn_port_t port; 96 xen_intr_handle_t handle; 97 bool enabled; 98 }; 99 100 RB_GENERATE_STATIC(evtchn_tree, user_evtchn, node, evtchn_cmp); 101 102 static device_t evtchn_dev; 103 104 static d_read_t evtchn_read; 105 static d_write_t evtchn_write; 106 static d_ioctl_t evtchn_ioctl; 107 static d_poll_t evtchn_poll; 108 static d_open_t evtchn_open; 109 110 static void evtchn_release(void *arg); 111 112 static struct cdevsw evtchn_devsw = { 113 .d_version = D_VERSION, 114 .d_open = evtchn_open, 115 .d_read = evtchn_read, 116 .d_write = evtchn_write, 117 .d_ioctl = evtchn_ioctl, 118 .d_poll = evtchn_poll, 119 .d_name = "evtchn", 120 }; 121 122 /*------------------------- Red-black tree helpers ---------------------------*/ 123 static int 124 evtchn_cmp(struct user_evtchn *u1, struct user_evtchn *u2) 125 { 126 127 return (u1->port - u2->port); 128 } 129 130 static struct user_evtchn * 131 find_evtchn(struct per_user_data *u, evtchn_port_t port) 132 { 133 struct user_evtchn tmp = { 134 .port = port, 135 }; 136 137 return (RB_FIND(evtchn_tree, &u->evtchns, &tmp)); 138 } 139 140 /*--------------------------- Interrupt handlers -----------------------------*/ 141 static int 142 evtchn_filter(void *arg) 143 { 144 struct user_evtchn *evtchn; 145 146 evtchn = arg; 147 148 if (!evtchn->enabled && bootverbose) { 149 device_printf(evtchn_dev, 150 "Received upcall for disabled event channel %d\n", 151 evtchn->port); 152 } 153 154 evtchn_mask_port(evtchn->port); 155 evtchn->enabled = false; 156 157 return (FILTER_SCHEDULE_THREAD); 158 } 159 160 static void 161 evtchn_interrupt(void *arg) 162 { 163 struct user_evtchn *evtchn; 164 struct per_user_data *u; 165 166 evtchn = arg; 167 u = evtchn->user; 168 169 /* 170 * Protect against concurrent events using this handler 171 * on different CPUs. 172 */ 173 mtx_lock(&u->ring_prod_mutex); 174 if ((u->ring_prod - u->ring_cons) < EVTCHN_RING_SIZE) { 175 u->ring[EVTCHN_RING_MASK(u->ring_prod)] = evtchn->port; 176 wmb(); /* Ensure ring contents visible */ 177 if (u->ring_cons == u->ring_prod++) { 178 wakeup(u); 179 selwakeup(&u->ev_rsel); 180 } 181 } else 182 u->ring_overflow = 1; 183 mtx_unlock(&u->ring_prod_mutex); 184 } 185 186 /*------------------------- Character device methods -------------------------*/ 187 static int 188 evtchn_open(struct cdev *dev, int flag, int otyp, struct thread *td) 189 { 190 struct per_user_data *u; 191 int error; 192 193 u = malloc(sizeof(*u), M_EVTCHN, M_WAITOK | M_ZERO); 194 u->ring = malloc(PAGE_SIZE, M_EVTCHN, M_WAITOK | M_ZERO); 195 196 /* Initialize locks */ 197 mtx_init(&u->bind_mutex, "evtchn_bind_mutex", NULL, MTX_DEF); 198 sx_init(&u->ring_cons_mutex, "evtchn_ringc_sx"); 199 mtx_init(&u->ring_prod_mutex, "evtchn_ringp_mutex", NULL, MTX_DEF); 200 201 /* Initialize red-black tree. */ 202 RB_INIT(&u->evtchns); 203 204 /* Assign the allocated per_user_data to this open instance. */ 205 error = devfs_set_cdevpriv(u, evtchn_release); 206 if (error != 0) { 207 mtx_destroy(&u->bind_mutex); 208 mtx_destroy(&u->ring_prod_mutex); 209 sx_destroy(&u->ring_cons_mutex); 210 free(u->ring, M_EVTCHN); 211 free(u, M_EVTCHN); 212 } 213 214 return (error); 215 } 216 217 static void 218 evtchn_release(void *arg) 219 { 220 struct per_user_data *u; 221 struct user_evtchn *evtchn, *tmp; 222 223 u = arg; 224 225 seldrain(&u->ev_rsel); 226 227 RB_FOREACH_SAFE(evtchn, evtchn_tree, &u->evtchns, tmp) { 228 xen_intr_unbind(&evtchn->handle); 229 230 RB_REMOVE(evtchn_tree, &u->evtchns, evtchn); 231 free(evtchn, M_EVTCHN); 232 } 233 234 mtx_destroy(&u->bind_mutex); 235 mtx_destroy(&u->ring_prod_mutex); 236 sx_destroy(&u->ring_cons_mutex); 237 free(u->ring, M_EVTCHN); 238 free(u, M_EVTCHN); 239 } 240 241 static int 242 evtchn_read(struct cdev *dev, struct uio *uio, int ioflag) 243 { 244 int error, count; 245 unsigned int c, p, bytes1 = 0, bytes2 = 0; 246 struct per_user_data *u; 247 248 error = devfs_get_cdevpriv((void **)&u); 249 if (error != 0) 250 return (EINVAL); 251 252 /* Whole number of ports. */ 253 count = uio->uio_resid; 254 count &= ~(sizeof(evtchn_port_t)-1); 255 256 if (count == 0) 257 return (0); 258 259 if (count > PAGE_SIZE) 260 count = PAGE_SIZE; 261 262 sx_xlock(&u->ring_cons_mutex); 263 for (;;) { 264 error = EFBIG; 265 if (u->ring_overflow) 266 goto unlock_out; 267 268 c = u->ring_cons; 269 p = u->ring_prod; 270 if (c != p) 271 break; 272 273 if (ioflag & IO_NDELAY) { 274 sx_xunlock(&u->ring_cons_mutex); 275 return (EWOULDBLOCK); 276 } 277 278 error = sx_sleep(u, &u->ring_cons_mutex, PCATCH, "evtchw", 0); 279 if ((error != 0) && (error != EWOULDBLOCK)) 280 return (error); 281 } 282 283 /* Byte lengths of two chunks. Chunk split (if any) is at ring wrap. */ 284 if (((c ^ p) & EVTCHN_RING_SIZE) != 0) { 285 bytes1 = (EVTCHN_RING_SIZE - EVTCHN_RING_MASK(c)) * 286 sizeof(evtchn_port_t); 287 bytes2 = EVTCHN_RING_MASK(p) * sizeof(evtchn_port_t); 288 } else { 289 bytes1 = (p - c) * sizeof(evtchn_port_t); 290 bytes2 = 0; 291 } 292 293 /* Truncate chunks according to caller's maximum byte count. */ 294 if (bytes1 > count) { 295 bytes1 = count; 296 bytes2 = 0; 297 } else if ((bytes1 + bytes2) > count) { 298 bytes2 = count - bytes1; 299 } 300 301 error = EFAULT; 302 rmb(); /* Ensure that we see the port before we copy it. */ 303 304 if (uiomove(&u->ring[EVTCHN_RING_MASK(c)], bytes1, uio) || 305 ((bytes2 != 0) && uiomove(&u->ring[0], bytes2, uio))) 306 goto unlock_out; 307 308 u->ring_cons += (bytes1 + bytes2) / sizeof(evtchn_port_t); 309 error = 0; 310 311 unlock_out: 312 sx_xunlock(&u->ring_cons_mutex); 313 return (error); 314 } 315 316 static int 317 evtchn_write(struct cdev *dev, struct uio *uio, int ioflag) 318 { 319 int error, i, count; 320 evtchn_port_t *kbuf; 321 struct per_user_data *u; 322 323 error = devfs_get_cdevpriv((void **)&u); 324 if (error != 0) 325 return (EINVAL); 326 327 kbuf = malloc(PAGE_SIZE, M_EVTCHN, M_WAITOK); 328 if (kbuf == NULL) 329 return (ENOMEM); 330 331 count = uio->uio_resid; 332 /* Whole number of ports. */ 333 count &= ~(sizeof(evtchn_port_t)-1); 334 335 error = 0; 336 if (count == 0) 337 goto out; 338 339 if (count > PAGE_SIZE) 340 count = PAGE_SIZE; 341 342 error = uiomove(kbuf, count, uio); 343 if (error != 0) 344 goto out; 345 346 mtx_lock(&u->bind_mutex); 347 348 for (i = 0; i < (count/sizeof(evtchn_port_t)); i++) { 349 evtchn_port_t port = kbuf[i]; 350 struct user_evtchn *evtchn; 351 352 evtchn = find_evtchn(u, port); 353 if (evtchn && !evtchn->enabled) { 354 evtchn->enabled = true; 355 evtchn_unmask_port(evtchn->port); 356 } 357 } 358 359 mtx_unlock(&u->bind_mutex); 360 error = 0; 361 362 out: 363 free(kbuf, M_EVTCHN); 364 return (error); 365 } 366 367 static inline int 368 evtchn_bind_user_port(struct per_user_data *u, struct user_evtchn *evtchn) 369 { 370 int error; 371 372 evtchn->port = xen_intr_port(evtchn->handle); 373 evtchn->user = u; 374 evtchn->enabled = true; 375 mtx_lock(&u->bind_mutex); 376 RB_INSERT(evtchn_tree, &u->evtchns, evtchn); 377 mtx_unlock(&u->bind_mutex); 378 error = xen_intr_add_handler(evtchn_dev, evtchn_filter, 379 evtchn_interrupt, evtchn, INTR_TYPE_MISC | INTR_MPSAFE, 380 evtchn->handle); 381 if (error != 0) { 382 xen_intr_unbind(&evtchn->handle); 383 mtx_lock(&u->bind_mutex); 384 RB_REMOVE(evtchn_tree, &u->evtchns, evtchn); 385 mtx_unlock(&u->bind_mutex); 386 free(evtchn, M_EVTCHN); 387 } 388 return (error); 389 } 390 391 static int 392 evtchn_ioctl(struct cdev *dev, unsigned long cmd, caddr_t arg, 393 int mode, struct thread *td __unused) 394 { 395 struct per_user_data *u; 396 int error; 397 398 error = devfs_get_cdevpriv((void **)&u); 399 if (error != 0) 400 return (EINVAL); 401 402 switch (cmd) { 403 case IOCTL_EVTCHN_BIND_VIRQ: { 404 struct ioctl_evtchn_bind_virq *bind; 405 struct user_evtchn *evtchn; 406 407 evtchn = malloc(sizeof(*evtchn), M_EVTCHN, M_WAITOK | M_ZERO); 408 409 bind = (struct ioctl_evtchn_bind_virq *)arg; 410 411 error = xen_intr_bind_virq(evtchn_dev, bind->virq, 0, 412 NULL, NULL, NULL, 0, &evtchn->handle); 413 if (error != 0) { 414 free(evtchn, M_EVTCHN); 415 break; 416 } 417 error = evtchn_bind_user_port(u, evtchn); 418 if (error != 0) 419 break; 420 bind->port = evtchn->port; 421 break; 422 } 423 424 case IOCTL_EVTCHN_BIND_INTERDOMAIN: { 425 struct ioctl_evtchn_bind_interdomain *bind; 426 struct user_evtchn *evtchn; 427 428 evtchn = malloc(sizeof(*evtchn), M_EVTCHN, M_WAITOK | M_ZERO); 429 430 bind = (struct ioctl_evtchn_bind_interdomain *)arg; 431 432 error = xen_intr_bind_remote_port(evtchn_dev, 433 bind->remote_domain, bind->remote_port, NULL, 434 NULL, NULL, 0, &evtchn->handle); 435 if (error != 0) { 436 free(evtchn, M_EVTCHN); 437 break; 438 } 439 error = evtchn_bind_user_port(u, evtchn); 440 if (error != 0) 441 break; 442 bind->port = evtchn->port; 443 break; 444 } 445 446 case IOCTL_EVTCHN_BIND_UNBOUND_PORT: { 447 struct ioctl_evtchn_bind_unbound_port *bind; 448 struct user_evtchn *evtchn; 449 450 evtchn = malloc(sizeof(*evtchn), M_EVTCHN, M_WAITOK | M_ZERO); 451 452 bind = (struct ioctl_evtchn_bind_unbound_port *)arg; 453 454 error = xen_intr_alloc_and_bind_local_port(evtchn_dev, 455 bind->remote_domain, NULL, NULL, NULL, 0, &evtchn->handle); 456 if (error != 0) { 457 free(evtchn, M_EVTCHN); 458 break; 459 } 460 error = evtchn_bind_user_port(u, evtchn); 461 if (error != 0) 462 break; 463 bind->port = evtchn->port; 464 break; 465 } 466 467 case IOCTL_EVTCHN_UNBIND: { 468 struct ioctl_evtchn_unbind *unbind; 469 struct user_evtchn *evtchn; 470 471 unbind = (struct ioctl_evtchn_unbind *)arg; 472 473 mtx_lock(&u->bind_mutex); 474 evtchn = find_evtchn(u, unbind->port); 475 if (evtchn == NULL) { 476 error = ENOTCONN; 477 break; 478 } 479 480 xen_intr_unbind(&evtchn->handle); 481 RB_REMOVE(evtchn_tree, &u->evtchns, evtchn); 482 mtx_unlock(&u->bind_mutex); 483 free(evtchn, M_EVTCHN); 484 error = 0; 485 break; 486 } 487 488 case IOCTL_EVTCHN_NOTIFY: { 489 struct ioctl_evtchn_notify *notify; 490 struct user_evtchn *evtchn; 491 492 notify = (struct ioctl_evtchn_notify *)arg; 493 494 mtx_lock(&u->bind_mutex); 495 evtchn = find_evtchn(u, notify->port); 496 if (evtchn == NULL) { 497 error = ENOTCONN; 498 break; 499 } 500 501 xen_intr_signal(evtchn->handle); 502 mtx_unlock(&u->bind_mutex); 503 error = 0; 504 break; 505 } 506 507 case IOCTL_EVTCHN_RESET: { 508 /* Initialise the ring to empty. Clear errors. */ 509 sx_xlock(&u->ring_cons_mutex); 510 mtx_lock(&u->ring_prod_mutex); 511 u->ring_cons = u->ring_prod = u->ring_overflow = 0; 512 mtx_unlock(&u->ring_prod_mutex); 513 sx_xunlock(&u->ring_cons_mutex); 514 error = 0; 515 break; 516 } 517 518 case FIONBIO: 519 case FIOASYNC: 520 /* Handled in an upper layer */ 521 error = 0; 522 break; 523 524 default: 525 error = ENOTTY; 526 break; 527 } 528 529 return (error); 530 } 531 532 static int 533 evtchn_poll(struct cdev *dev, int events, struct thread *td) 534 { 535 struct per_user_data *u; 536 int error, mask; 537 538 error = devfs_get_cdevpriv((void **)&u); 539 if (error != 0) 540 return (POLLERR); 541 542 /* we can always write */ 543 mask = events & (POLLOUT | POLLWRNORM); 544 545 mtx_lock(&u->ring_prod_mutex); 546 if (events & (POLLIN | POLLRDNORM)) { 547 if (u->ring_cons != u->ring_prod) { 548 mask |= events & (POLLIN | POLLRDNORM); 549 } else { 550 /* Record that someone is waiting */ 551 selrecord(td, &u->ev_rsel); 552 } 553 } 554 mtx_unlock(&u->ring_prod_mutex); 555 556 return (mask); 557 } 558 559 /*------------------ Private Device Attachment Functions --------------------*/ 560 static void 561 evtchn_identify(driver_t *driver, device_t parent) 562 { 563 564 KASSERT((xen_domain()), 565 ("Trying to attach evtchn device on non Xen domain")); 566 567 evtchn_dev = BUS_ADD_CHILD(parent, 0, "evtchn", 0); 568 if (evtchn_dev == NULL) 569 panic("unable to attach evtchn user-space device"); 570 } 571 572 static int 573 evtchn_probe(device_t dev) 574 { 575 576 device_set_desc(dev, "Xen event channel user-space device"); 577 return (BUS_PROBE_NOWILDCARD); 578 } 579 580 static int 581 evtchn_attach(device_t dev) 582 { 583 584 make_dev_credf(MAKEDEV_ETERNAL, &evtchn_devsw, 0, NULL, UID_ROOT, 585 GID_WHEEL, 0600, "xen/evtchn"); 586 return (0); 587 } 588 589 /*-------------------- Private Device Attachment Data -----------------------*/ 590 static device_method_t evtchn_methods[] = { 591 DEVMETHOD(device_identify, evtchn_identify), 592 DEVMETHOD(device_probe, evtchn_probe), 593 DEVMETHOD(device_attach, evtchn_attach), 594 595 DEVMETHOD_END 596 }; 597 598 static driver_t evtchn_driver = { 599 "evtchn", 600 evtchn_methods, 601 0, 602 }; 603 604 devclass_t evtchn_devclass; 605 606 DRIVER_MODULE(evtchn, xenpv, evtchn_driver, evtchn_devclass, 0, 0); 607 MODULE_DEPEND(evtchn, xenpv, 1, 1, 1); 608