1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * ipmi_watchdog.c 4 * 5 * A watchdog timer based upon the IPMI interface. 6 * 7 * Author: MontaVista Software, Inc. 8 * Corey Minyard <minyard@mvista.com> 9 * source@mvista.com 10 * 11 * Copyright 2002 MontaVista Software Inc. 12 */ 13 14 #define pr_fmt(fmt) "IPMI Watchdog: " fmt 15 16 #include <linux/module.h> 17 #include <linux/moduleparam.h> 18 #include <linux/ipmi.h> 19 #include <linux/ipmi_smi.h> 20 #include <linux/mutex.h> 21 #include <linux/watchdog.h> 22 #include <linux/miscdevice.h> 23 #include <linux/init.h> 24 #include <linux/completion.h> 25 #include <linux/kdebug.h> 26 #include <linux/rwsem.h> 27 #include <linux/errno.h> 28 #include <linux/uaccess.h> 29 #include <linux/notifier.h> 30 #include <linux/nmi.h> 31 #include <linux/reboot.h> 32 #include <linux/wait.h> 33 #include <linux/poll.h> 34 #include <linux/string.h> 35 #include <linux/ctype.h> 36 #include <linux/delay.h> 37 #include <linux/atomic.h> 38 #include <linux/sched/signal.h> 39 40 #ifdef CONFIG_X86 41 /* 42 * This is ugly, but I've determined that x86 is the only architecture 43 * that can reasonably support the IPMI NMI watchdog timeout at this 44 * time. If another architecture adds this capability somehow, it 45 * will have to be a somewhat different mechanism and I have no idea 46 * how it will work. So in the unlikely event that another 47 * architecture supports this, we can figure out a good generic 48 * mechanism for it at that time. 49 */ 50 #include <asm/kdebug.h> 51 #include <asm/nmi.h> 52 #define HAVE_DIE_NMI 53 #endif 54 55 /* 56 * The IPMI command/response information for the watchdog timer. 57 */ 58 59 /* values for byte 1 of the set command, byte 2 of the get response. */ 60 #define WDOG_DONT_LOG (1 << 7) 61 #define WDOG_DONT_STOP_ON_SET (1 << 6) 62 #define WDOG_SET_TIMER_USE(byte, use) \ 63 byte = ((byte) & 0xf8) | ((use) & 0x7) 64 #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) 65 #define WDOG_TIMER_USE_BIOS_FRB2 1 66 #define WDOG_TIMER_USE_BIOS_POST 2 67 #define WDOG_TIMER_USE_OS_LOAD 3 68 #define WDOG_TIMER_USE_SMS_OS 4 69 #define WDOG_TIMER_USE_OEM 5 70 71 /* values for byte 2 of the set command, byte 3 of the get response. */ 72 #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ 73 byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) 74 #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) 75 #define WDOG_PRETIMEOUT_NONE 0 76 #define WDOG_PRETIMEOUT_SMI 1 77 #define WDOG_PRETIMEOUT_NMI 2 78 #define WDOG_PRETIMEOUT_MSG_INT 3 79 80 /* Operations that can be performed on a pretimout. */ 81 #define WDOG_PREOP_NONE 0 82 #define WDOG_PREOP_PANIC 1 83 /* Cause data to be available to read. Doesn't work in NMI mode. */ 84 #define WDOG_PREOP_GIVE_DATA 2 85 86 /* Actions to perform on a full timeout. */ 87 #define WDOG_SET_TIMEOUT_ACT(byte, use) \ 88 byte = ((byte) & 0xf8) | ((use) & 0x7) 89 #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) 90 #define WDOG_TIMEOUT_NONE 0 91 #define WDOG_TIMEOUT_RESET 1 92 #define WDOG_TIMEOUT_POWER_DOWN 2 93 #define WDOG_TIMEOUT_POWER_CYCLE 3 94 95 /* 96 * Byte 3 of the get command, byte 4 of the get response is the 97 * pre-timeout in seconds. 98 */ 99 100 /* Bits for setting byte 4 of the set command, byte 5 of the get response. */ 101 #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1) 102 #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2) 103 #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3) 104 #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4) 105 #define WDOG_EXPIRE_CLEAR_OEM (1 << 5) 106 107 /* 108 * Setting/getting the watchdog timer value. This is for bytes 5 and 109 * 6 (the timeout time) of the set command, and bytes 6 and 7 (the 110 * timeout time) and 8 and 9 (the current countdown value) of the 111 * response. The timeout value is given in seconds (in the command it 112 * is 100ms intervals). 113 */ 114 #define WDOG_SET_TIMEOUT(byte1, byte2, val) \ 115 (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8) 116 #define WDOG_GET_TIMEOUT(byte1, byte2) \ 117 (((byte1) | ((byte2) << 8)) / 10) 118 119 #define IPMI_WDOG_RESET_TIMER 0x22 120 #define IPMI_WDOG_SET_TIMER 0x24 121 #define IPMI_WDOG_GET_TIMER 0x25 122 123 #define IPMI_WDOG_TIMER_NOT_INIT_RESP 0x80 124 125 static DEFINE_MUTEX(ipmi_watchdog_mutex); 126 static bool nowayout = WATCHDOG_NOWAYOUT; 127 128 static struct ipmi_user *watchdog_user; 129 static int watchdog_ifnum; 130 131 /* Default the timeout to 10 seconds. */ 132 static int timeout = 10; 133 134 /* The pre-timeout is disabled by default. */ 135 static int pretimeout; 136 137 /* Default timeout to set on panic */ 138 static int panic_wdt_timeout = 255; 139 140 /* Default action is to reset the board on a timeout. */ 141 static unsigned char action_val = WDOG_TIMEOUT_RESET; 142 143 static char action[16] = "reset"; 144 145 static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; 146 147 static char preaction[16] = "pre_none"; 148 149 static unsigned char preop_val = WDOG_PREOP_NONE; 150 151 static char preop[16] = "preop_none"; 152 static DEFINE_SPINLOCK(ipmi_read_lock); 153 static char data_to_read; 154 static DECLARE_WAIT_QUEUE_HEAD(read_q); 155 static struct fasync_struct *fasync_q; 156 static atomic_t pretimeout_since_last_heartbeat; 157 static char expect_close; 158 159 static int ifnum_to_use = -1; 160 161 /* Parameters to ipmi_set_timeout */ 162 #define IPMI_SET_TIMEOUT_NO_HB 0 163 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 164 #define IPMI_SET_TIMEOUT_FORCE_HB 2 165 166 static int ipmi_set_timeout(int do_heartbeat); 167 static void ipmi_register_watchdog(int ipmi_intf); 168 static void ipmi_unregister_watchdog(int ipmi_intf); 169 170 /* 171 * If true, the driver will start running as soon as it is configured 172 * and ready. 173 */ 174 static int start_now; 175 176 static int set_param_timeout(const char *val, const struct kernel_param *kp) 177 { 178 char *endp; 179 int l; 180 int rv = 0; 181 182 if (!val) 183 return -EINVAL; 184 l = simple_strtoul(val, &endp, 0); 185 if (endp == val) 186 return -EINVAL; 187 188 *((int *)kp->arg) = l; 189 if (watchdog_user) 190 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 191 192 return rv; 193 } 194 195 static const struct kernel_param_ops param_ops_timeout = { 196 .set = set_param_timeout, 197 .get = param_get_int, 198 }; 199 #define param_check_timeout param_check_int 200 201 typedef int (*action_fn)(const char *intval, char *outval); 202 203 static int action_op(const char *inval, char *outval); 204 static int preaction_op(const char *inval, char *outval); 205 static int preop_op(const char *inval, char *outval); 206 static void check_parms(void); 207 208 static int set_param_str(const char *val, const struct kernel_param *kp) 209 { 210 action_fn fn = (action_fn) kp->arg; 211 int rv = 0; 212 char valcp[16]; 213 char *s; 214 215 strncpy(valcp, val, 15); 216 valcp[15] = '\0'; 217 218 s = strstrip(valcp); 219 220 rv = fn(s, NULL); 221 if (rv) 222 goto out; 223 224 check_parms(); 225 if (watchdog_user) 226 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 227 228 out: 229 return rv; 230 } 231 232 static int get_param_str(char *buffer, const struct kernel_param *kp) 233 { 234 action_fn fn = (action_fn) kp->arg; 235 int rv; 236 237 rv = fn(NULL, buffer); 238 if (rv) 239 return rv; 240 return strlen(buffer); 241 } 242 243 244 static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp) 245 { 246 int rv = param_set_int(val, kp); 247 if (rv) 248 return rv; 249 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum)) 250 return 0; 251 252 ipmi_unregister_watchdog(watchdog_ifnum); 253 ipmi_register_watchdog(ifnum_to_use); 254 return 0; 255 } 256 257 static const struct kernel_param_ops param_ops_wdog_ifnum = { 258 .set = set_param_wdog_ifnum, 259 .get = param_get_int, 260 }; 261 262 #define param_check_wdog_ifnum param_check_int 263 264 static const struct kernel_param_ops param_ops_str = { 265 .set = set_param_str, 266 .get = get_param_str, 267 }; 268 269 module_param(ifnum_to_use, wdog_ifnum, 0644); 270 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog " 271 "timer. Setting to -1 defaults to the first registered " 272 "interface"); 273 274 module_param(timeout, timeout, 0644); 275 MODULE_PARM_DESC(timeout, "Timeout value in seconds."); 276 277 module_param(pretimeout, timeout, 0644); 278 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); 279 280 module_param(panic_wdt_timeout, timeout, 0644); 281 MODULE_PARM_DESC(panic_wdt_timeout, "Timeout value on kernel panic in seconds."); 282 283 module_param_cb(action, ¶m_ops_str, action_op, 0644); 284 MODULE_PARM_DESC(action, "Timeout action. One of: " 285 "reset, none, power_cycle, power_off."); 286 287 module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644); 288 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " 289 "pre_none, pre_smi, pre_nmi, pre_int."); 290 291 module_param_cb(preop, ¶m_ops_str, preop_op, 0644); 292 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " 293 "preop_none, preop_panic, preop_give_data."); 294 295 module_param(start_now, int, 0444); 296 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" 297 "soon as the driver is loaded."); 298 299 module_param(nowayout, bool, 0644); 300 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started " 301 "(default=CONFIG_WATCHDOG_NOWAYOUT)"); 302 303 /* Default state of the timer. */ 304 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 305 306 /* Is someone using the watchdog? Only one user is allowed. */ 307 static unsigned long ipmi_wdog_open; 308 309 /* 310 * If set to 1, the heartbeat command will set the state to reset and 311 * start the timer. The timer doesn't normally run when the driver is 312 * first opened until the heartbeat is set the first time, this 313 * variable is used to accomplish this. 314 */ 315 static int ipmi_start_timer_on_heartbeat; 316 317 /* IPMI version of the BMC. */ 318 static unsigned char ipmi_version_major; 319 static unsigned char ipmi_version_minor; 320 321 /* If a pretimeout occurs, this is used to allow only one panic to happen. */ 322 static atomic_t preop_panic_excl = ATOMIC_INIT(-1); 323 324 #ifdef HAVE_DIE_NMI 325 static int testing_nmi; 326 static int nmi_handler_registered; 327 #endif 328 329 static int __ipmi_heartbeat(void); 330 331 /* 332 * We use a mutex to make sure that only one thing can send a set a 333 * message at one time. The mutex is claimed when a message is sent 334 * and freed when both the send and receive messages are free. 335 */ 336 static atomic_t msg_tofree = ATOMIC_INIT(0); 337 static DECLARE_COMPLETION(msg_wait); 338 static void msg_free_smi(struct ipmi_smi_msg *msg) 339 { 340 if (atomic_dec_and_test(&msg_tofree)) 341 complete(&msg_wait); 342 } 343 static void msg_free_recv(struct ipmi_recv_msg *msg) 344 { 345 if (atomic_dec_and_test(&msg_tofree)) 346 complete(&msg_wait); 347 } 348 static struct ipmi_smi_msg smi_msg = { 349 .done = msg_free_smi 350 }; 351 static struct ipmi_recv_msg recv_msg = { 352 .done = msg_free_recv 353 }; 354 355 static int __ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, 356 struct ipmi_recv_msg *recv_msg, 357 int *send_heartbeat_now) 358 { 359 struct kernel_ipmi_msg msg; 360 unsigned char data[6]; 361 int rv; 362 struct ipmi_system_interface_addr addr; 363 int hbnow = 0; 364 365 366 data[0] = 0; 367 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); 368 369 if ((ipmi_version_major > 1) 370 || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) { 371 /* This is an IPMI 1.5-only feature. */ 372 data[0] |= WDOG_DONT_STOP_ON_SET; 373 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { 374 /* 375 * In ipmi 1.0, setting the timer stops the watchdog, we 376 * need to start it back up again. 377 */ 378 hbnow = 1; 379 } 380 381 data[1] = 0; 382 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); 383 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { 384 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); 385 data[2] = pretimeout; 386 } else { 387 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); 388 data[2] = 0; /* No pretimeout. */ 389 } 390 data[3] = 0; 391 WDOG_SET_TIMEOUT(data[4], data[5], timeout); 392 393 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 394 addr.channel = IPMI_BMC_CHANNEL; 395 addr.lun = 0; 396 397 msg.netfn = 0x06; 398 msg.cmd = IPMI_WDOG_SET_TIMER; 399 msg.data = data; 400 msg.data_len = sizeof(data); 401 rv = ipmi_request_supply_msgs(watchdog_user, 402 (struct ipmi_addr *) &addr, 403 0, 404 &msg, 405 NULL, 406 smi_msg, 407 recv_msg, 408 1); 409 if (rv) 410 pr_warn("set timeout error: %d\n", rv); 411 else if (send_heartbeat_now) 412 *send_heartbeat_now = hbnow; 413 414 return rv; 415 } 416 417 static int _ipmi_set_timeout(int do_heartbeat) 418 { 419 int send_heartbeat_now; 420 int rv; 421 422 if (!watchdog_user) 423 return -ENODEV; 424 425 atomic_set(&msg_tofree, 2); 426 427 rv = __ipmi_set_timeout(&smi_msg, 428 &recv_msg, 429 &send_heartbeat_now); 430 if (rv) 431 return rv; 432 433 wait_for_completion(&msg_wait); 434 435 if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) 436 || ((send_heartbeat_now) 437 && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) 438 rv = __ipmi_heartbeat(); 439 440 return rv; 441 } 442 443 static int ipmi_set_timeout(int do_heartbeat) 444 { 445 int rv; 446 447 mutex_lock(&ipmi_watchdog_mutex); 448 rv = _ipmi_set_timeout(do_heartbeat); 449 mutex_unlock(&ipmi_watchdog_mutex); 450 451 return rv; 452 } 453 454 static atomic_t panic_done_count = ATOMIC_INIT(0); 455 456 static void panic_smi_free(struct ipmi_smi_msg *msg) 457 { 458 atomic_dec(&panic_done_count); 459 } 460 static void panic_recv_free(struct ipmi_recv_msg *msg) 461 { 462 atomic_dec(&panic_done_count); 463 } 464 465 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = { 466 .done = panic_smi_free 467 }; 468 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = { 469 .done = panic_recv_free 470 }; 471 472 static void panic_halt_ipmi_heartbeat(void) 473 { 474 struct kernel_ipmi_msg msg; 475 struct ipmi_system_interface_addr addr; 476 int rv; 477 478 /* 479 * Don't reset the timer if we have the timer turned off, that 480 * re-enables the watchdog. 481 */ 482 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) 483 return; 484 485 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 486 addr.channel = IPMI_BMC_CHANNEL; 487 addr.lun = 0; 488 489 msg.netfn = 0x06; 490 msg.cmd = IPMI_WDOG_RESET_TIMER; 491 msg.data = NULL; 492 msg.data_len = 0; 493 atomic_add(1, &panic_done_count); 494 rv = ipmi_request_supply_msgs(watchdog_user, 495 (struct ipmi_addr *) &addr, 496 0, 497 &msg, 498 NULL, 499 &panic_halt_heartbeat_smi_msg, 500 &panic_halt_heartbeat_recv_msg, 501 1); 502 if (rv) 503 atomic_sub(1, &panic_done_count); 504 } 505 506 static struct ipmi_smi_msg panic_halt_smi_msg = { 507 .done = panic_smi_free 508 }; 509 static struct ipmi_recv_msg panic_halt_recv_msg = { 510 .done = panic_recv_free 511 }; 512 513 /* 514 * Special call, doesn't claim any locks. This is only to be called 515 * at panic or halt time, in run-to-completion mode, when the caller 516 * is the only CPU and the only thing that will be going is these IPMI 517 * calls. 518 */ 519 static void panic_halt_ipmi_set_timeout(void) 520 { 521 int send_heartbeat_now; 522 int rv; 523 524 /* Wait for the messages to be free. */ 525 while (atomic_read(&panic_done_count) != 0) 526 ipmi_poll_interface(watchdog_user); 527 atomic_add(1, &panic_done_count); 528 rv = __ipmi_set_timeout(&panic_halt_smi_msg, 529 &panic_halt_recv_msg, 530 &send_heartbeat_now); 531 if (rv) { 532 atomic_sub(1, &panic_done_count); 533 pr_warn("Unable to extend the watchdog timeout\n"); 534 } else { 535 if (send_heartbeat_now) 536 panic_halt_ipmi_heartbeat(); 537 } 538 while (atomic_read(&panic_done_count) != 0) 539 ipmi_poll_interface(watchdog_user); 540 } 541 542 static int __ipmi_heartbeat(void) 543 { 544 struct kernel_ipmi_msg msg; 545 int rv; 546 struct ipmi_system_interface_addr addr; 547 int timeout_retries = 0; 548 549 restart: 550 /* 551 * Don't reset the timer if we have the timer turned off, that 552 * re-enables the watchdog. 553 */ 554 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) 555 return 0; 556 557 atomic_set(&msg_tofree, 2); 558 559 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 560 addr.channel = IPMI_BMC_CHANNEL; 561 addr.lun = 0; 562 563 msg.netfn = 0x06; 564 msg.cmd = IPMI_WDOG_RESET_TIMER; 565 msg.data = NULL; 566 msg.data_len = 0; 567 rv = ipmi_request_supply_msgs(watchdog_user, 568 (struct ipmi_addr *) &addr, 569 0, 570 &msg, 571 NULL, 572 &smi_msg, 573 &recv_msg, 574 1); 575 if (rv) { 576 pr_warn("heartbeat send failure: %d\n", rv); 577 return rv; 578 } 579 580 /* Wait for the heartbeat to be sent. */ 581 wait_for_completion(&msg_wait); 582 583 if (recv_msg.msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) { 584 timeout_retries++; 585 if (timeout_retries > 3) { 586 pr_err("Unable to restore the IPMI watchdog's settings, giving up\n"); 587 rv = -EIO; 588 goto out; 589 } 590 591 /* 592 * The timer was not initialized, that means the BMC was 593 * probably reset and lost the watchdog information. Attempt 594 * to restore the timer's info. Note that we still hold 595 * the heartbeat lock, to keep a heartbeat from happening 596 * in this process, so must say no heartbeat to avoid a 597 * deadlock on this mutex 598 */ 599 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 600 if (rv) { 601 pr_err("Unable to send the command to set the watchdog's settings, giving up\n"); 602 goto out; 603 } 604 605 /* Might need a heartbeat send, go ahead and do it. */ 606 goto restart; 607 } else if (recv_msg.msg.data[0] != 0) { 608 /* 609 * Got an error in the heartbeat response. It was already 610 * reported in ipmi_wdog_msg_handler, but we should return 611 * an error here. 612 */ 613 rv = -EINVAL; 614 } 615 616 out: 617 return rv; 618 } 619 620 static int _ipmi_heartbeat(void) 621 { 622 int rv; 623 624 if (!watchdog_user) 625 return -ENODEV; 626 627 if (ipmi_start_timer_on_heartbeat) { 628 ipmi_start_timer_on_heartbeat = 0; 629 ipmi_watchdog_state = action_val; 630 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 631 } else if (atomic_cmpxchg(&pretimeout_since_last_heartbeat, 1, 0)) { 632 /* 633 * A pretimeout occurred, make sure we set the timeout. 634 * We don't want to set the action, though, we want to 635 * leave that alone (thus it can't be combined with the 636 * above operation. 637 */ 638 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 639 } else { 640 rv = __ipmi_heartbeat(); 641 } 642 643 return rv; 644 } 645 646 static int ipmi_heartbeat(void) 647 { 648 int rv; 649 650 mutex_lock(&ipmi_watchdog_mutex); 651 rv = _ipmi_heartbeat(); 652 mutex_unlock(&ipmi_watchdog_mutex); 653 654 return rv; 655 } 656 657 static struct watchdog_info ident = { 658 .options = 0, /* WDIOF_SETTIMEOUT, */ 659 .firmware_version = 1, 660 .identity = "IPMI" 661 }; 662 663 static int ipmi_ioctl(struct file *file, 664 unsigned int cmd, unsigned long arg) 665 { 666 void __user *argp = (void __user *)arg; 667 int i; 668 int val; 669 670 switch (cmd) { 671 case WDIOC_GETSUPPORT: 672 i = copy_to_user(argp, &ident, sizeof(ident)); 673 return i ? -EFAULT : 0; 674 675 case WDIOC_SETTIMEOUT: 676 i = copy_from_user(&val, argp, sizeof(int)); 677 if (i) 678 return -EFAULT; 679 timeout = val; 680 return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 681 682 case WDIOC_GETTIMEOUT: 683 i = copy_to_user(argp, &timeout, sizeof(timeout)); 684 if (i) 685 return -EFAULT; 686 return 0; 687 688 case WDIOC_SETPRETIMEOUT: 689 i = copy_from_user(&val, argp, sizeof(int)); 690 if (i) 691 return -EFAULT; 692 pretimeout = val; 693 return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 694 695 case WDIOC_GETPRETIMEOUT: 696 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); 697 if (i) 698 return -EFAULT; 699 return 0; 700 701 case WDIOC_KEEPALIVE: 702 return _ipmi_heartbeat(); 703 704 case WDIOC_SETOPTIONS: 705 i = copy_from_user(&val, argp, sizeof(int)); 706 if (i) 707 return -EFAULT; 708 if (val & WDIOS_DISABLECARD) { 709 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 710 _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 711 ipmi_start_timer_on_heartbeat = 0; 712 } 713 714 if (val & WDIOS_ENABLECARD) { 715 ipmi_watchdog_state = action_val; 716 _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 717 } 718 return 0; 719 720 case WDIOC_GETSTATUS: 721 val = 0; 722 i = copy_to_user(argp, &val, sizeof(val)); 723 if (i) 724 return -EFAULT; 725 return 0; 726 727 default: 728 return -ENOIOCTLCMD; 729 } 730 } 731 732 static long ipmi_unlocked_ioctl(struct file *file, 733 unsigned int cmd, 734 unsigned long arg) 735 { 736 int ret; 737 738 mutex_lock(&ipmi_watchdog_mutex); 739 ret = ipmi_ioctl(file, cmd, arg); 740 mutex_unlock(&ipmi_watchdog_mutex); 741 742 return ret; 743 } 744 745 static ssize_t ipmi_write(struct file *file, 746 const char __user *buf, 747 size_t len, 748 loff_t *ppos) 749 { 750 int rv; 751 752 if (len) { 753 if (!nowayout) { 754 size_t i; 755 756 /* In case it was set long ago */ 757 expect_close = 0; 758 759 for (i = 0; i != len; i++) { 760 char c; 761 762 if (get_user(c, buf + i)) 763 return -EFAULT; 764 if (c == 'V') 765 expect_close = 42; 766 } 767 } 768 rv = ipmi_heartbeat(); 769 if (rv) 770 return rv; 771 } 772 return len; 773 } 774 775 static ssize_t ipmi_read(struct file *file, 776 char __user *buf, 777 size_t count, 778 loff_t *ppos) 779 { 780 int rv = 0; 781 wait_queue_entry_t wait; 782 783 if (count <= 0) 784 return 0; 785 786 /* 787 * Reading returns if the pretimeout has gone off, and it only does 788 * it once per pretimeout. 789 */ 790 spin_lock_irq(&ipmi_read_lock); 791 if (!data_to_read) { 792 if (file->f_flags & O_NONBLOCK) { 793 rv = -EAGAIN; 794 goto out; 795 } 796 797 init_waitqueue_entry(&wait, current); 798 add_wait_queue(&read_q, &wait); 799 while (!data_to_read) { 800 set_current_state(TASK_INTERRUPTIBLE); 801 spin_unlock_irq(&ipmi_read_lock); 802 schedule(); 803 spin_lock_irq(&ipmi_read_lock); 804 } 805 remove_wait_queue(&read_q, &wait); 806 807 if (signal_pending(current)) { 808 rv = -ERESTARTSYS; 809 goto out; 810 } 811 } 812 data_to_read = 0; 813 814 out: 815 spin_unlock_irq(&ipmi_read_lock); 816 817 if (rv == 0) { 818 if (copy_to_user(buf, &data_to_read, 1)) 819 rv = -EFAULT; 820 else 821 rv = 1; 822 } 823 824 return rv; 825 } 826 827 static int ipmi_open(struct inode *ino, struct file *filep) 828 { 829 switch (iminor(ino)) { 830 case WATCHDOG_MINOR: 831 if (test_and_set_bit(0, &ipmi_wdog_open)) 832 return -EBUSY; 833 834 835 /* 836 * Don't start the timer now, let it start on the 837 * first heartbeat. 838 */ 839 ipmi_start_timer_on_heartbeat = 1; 840 return stream_open(ino, filep); 841 842 default: 843 return (-ENODEV); 844 } 845 } 846 847 static __poll_t ipmi_poll(struct file *file, poll_table *wait) 848 { 849 __poll_t mask = 0; 850 851 poll_wait(file, &read_q, wait); 852 853 spin_lock_irq(&ipmi_read_lock); 854 if (data_to_read) 855 mask |= (EPOLLIN | EPOLLRDNORM); 856 spin_unlock_irq(&ipmi_read_lock); 857 858 return mask; 859 } 860 861 static int ipmi_fasync(int fd, struct file *file, int on) 862 { 863 int result; 864 865 result = fasync_helper(fd, file, on, &fasync_q); 866 867 return (result); 868 } 869 870 static int ipmi_close(struct inode *ino, struct file *filep) 871 { 872 if (iminor(ino) == WATCHDOG_MINOR) { 873 if (expect_close == 42) { 874 mutex_lock(&ipmi_watchdog_mutex); 875 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 876 _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 877 mutex_unlock(&ipmi_watchdog_mutex); 878 } else { 879 pr_crit("Unexpected close, not stopping watchdog!\n"); 880 ipmi_heartbeat(); 881 } 882 clear_bit(0, &ipmi_wdog_open); 883 } 884 885 expect_close = 0; 886 887 return 0; 888 } 889 890 static const struct file_operations ipmi_wdog_fops = { 891 .owner = THIS_MODULE, 892 .read = ipmi_read, 893 .poll = ipmi_poll, 894 .write = ipmi_write, 895 .unlocked_ioctl = ipmi_unlocked_ioctl, 896 .open = ipmi_open, 897 .release = ipmi_close, 898 .fasync = ipmi_fasync, 899 .llseek = no_llseek, 900 }; 901 902 static struct miscdevice ipmi_wdog_miscdev = { 903 .minor = WATCHDOG_MINOR, 904 .name = "watchdog", 905 .fops = &ipmi_wdog_fops 906 }; 907 908 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, 909 void *handler_data) 910 { 911 if (msg->msg.cmd == IPMI_WDOG_RESET_TIMER && 912 msg->msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) 913 pr_info("response: The IPMI controller appears to have been reset, will attempt to reinitialize the watchdog timer\n"); 914 else if (msg->msg.data[0] != 0) 915 pr_err("response: Error %x on cmd %x\n", 916 msg->msg.data[0], 917 msg->msg.cmd); 918 919 ipmi_free_recv_msg(msg); 920 } 921 922 static void ipmi_wdog_pretimeout_handler(void *handler_data) 923 { 924 if (preaction_val != WDOG_PRETIMEOUT_NONE) { 925 if (preop_val == WDOG_PREOP_PANIC) { 926 if (atomic_inc_and_test(&preop_panic_excl)) 927 panic("Watchdog pre-timeout"); 928 } else if (preop_val == WDOG_PREOP_GIVE_DATA) { 929 unsigned long flags; 930 931 spin_lock_irqsave(&ipmi_read_lock, flags); 932 data_to_read = 1; 933 wake_up_interruptible(&read_q); 934 kill_fasync(&fasync_q, SIGIO, POLL_IN); 935 spin_unlock_irqrestore(&ipmi_read_lock, flags); 936 } 937 } 938 939 /* 940 * On some machines, the heartbeat will give an error and not 941 * work unless we re-enable the timer. So do so. 942 */ 943 atomic_set(&pretimeout_since_last_heartbeat, 1); 944 } 945 946 static void ipmi_wdog_panic_handler(void *user_data) 947 { 948 static int panic_event_handled; 949 950 /* 951 * On a panic, if we have a panic timeout, make sure to extend 952 * the watchdog timer to a reasonable value to complete the 953 * panic, if the watchdog timer is running. Plus the 954 * pretimeout is meaningless at panic time. 955 */ 956 if (watchdog_user && !panic_event_handled && 957 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { 958 /* Make sure we do this only once. */ 959 panic_event_handled = 1; 960 961 timeout = panic_wdt_timeout; 962 pretimeout = 0; 963 panic_halt_ipmi_set_timeout(); 964 } 965 } 966 967 static const struct ipmi_user_hndl ipmi_hndlrs = { 968 .ipmi_recv_hndl = ipmi_wdog_msg_handler, 969 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler, 970 .ipmi_panic_handler = ipmi_wdog_panic_handler 971 }; 972 973 static void ipmi_register_watchdog(int ipmi_intf) 974 { 975 int rv = -EBUSY; 976 977 if (watchdog_user) 978 goto out; 979 980 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf)) 981 goto out; 982 983 watchdog_ifnum = ipmi_intf; 984 985 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); 986 if (rv < 0) { 987 pr_crit("Unable to register with ipmi\n"); 988 goto out; 989 } 990 991 rv = ipmi_get_version(watchdog_user, 992 &ipmi_version_major, 993 &ipmi_version_minor); 994 if (rv) { 995 pr_warn("Unable to get IPMI version, assuming 1.0\n"); 996 ipmi_version_major = 1; 997 ipmi_version_minor = 0; 998 } 999 1000 rv = misc_register(&ipmi_wdog_miscdev); 1001 if (rv < 0) { 1002 ipmi_destroy_user(watchdog_user); 1003 watchdog_user = NULL; 1004 pr_crit("Unable to register misc device\n"); 1005 } 1006 1007 #ifdef HAVE_DIE_NMI 1008 if (nmi_handler_registered) { 1009 int old_pretimeout = pretimeout; 1010 int old_timeout = timeout; 1011 int old_preop_val = preop_val; 1012 1013 /* 1014 * Set the pretimeout to go off in a second and give 1015 * ourselves plenty of time to stop the timer. 1016 */ 1017 ipmi_watchdog_state = WDOG_TIMEOUT_RESET; 1018 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */ 1019 pretimeout = 99; 1020 timeout = 100; 1021 1022 testing_nmi = 1; 1023 1024 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 1025 if (rv) { 1026 pr_warn("Error starting timer to test NMI: 0x%x. The NMI pretimeout will likely not work\n", 1027 rv); 1028 rv = 0; 1029 goto out_restore; 1030 } 1031 1032 msleep(1500); 1033 1034 if (testing_nmi != 2) { 1035 pr_warn("IPMI NMI didn't seem to occur. The NMI pretimeout will likely not work\n"); 1036 } 1037 out_restore: 1038 testing_nmi = 0; 1039 preop_val = old_preop_val; 1040 pretimeout = old_pretimeout; 1041 timeout = old_timeout; 1042 } 1043 #endif 1044 1045 out: 1046 if ((start_now) && (rv == 0)) { 1047 /* Run from startup, so start the timer now. */ 1048 start_now = 0; /* Disable this function after first startup. */ 1049 ipmi_watchdog_state = action_val; 1050 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 1051 pr_info("Starting now!\n"); 1052 } else { 1053 /* Stop the timer now. */ 1054 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 1055 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 1056 } 1057 } 1058 1059 static void ipmi_unregister_watchdog(int ipmi_intf) 1060 { 1061 int rv; 1062 struct ipmi_user *loc_user = watchdog_user; 1063 1064 if (!loc_user) 1065 return; 1066 1067 if (watchdog_ifnum != ipmi_intf) 1068 return; 1069 1070 /* Make sure no one can call us any more. */ 1071 misc_deregister(&ipmi_wdog_miscdev); 1072 1073 watchdog_user = NULL; 1074 1075 /* 1076 * Wait to make sure the message makes it out. The lower layer has 1077 * pointers to our buffers, we want to make sure they are done before 1078 * we release our memory. 1079 */ 1080 while (atomic_read(&msg_tofree)) 1081 msg_free_smi(NULL); 1082 1083 mutex_lock(&ipmi_watchdog_mutex); 1084 1085 /* Disconnect from IPMI. */ 1086 rv = ipmi_destroy_user(loc_user); 1087 if (rv) 1088 pr_warn("error unlinking from IPMI: %d\n", rv); 1089 1090 /* If it comes back, restart it properly. */ 1091 ipmi_start_timer_on_heartbeat = 1; 1092 1093 mutex_unlock(&ipmi_watchdog_mutex); 1094 } 1095 1096 #ifdef HAVE_DIE_NMI 1097 static int 1098 ipmi_nmi(unsigned int val, struct pt_regs *regs) 1099 { 1100 /* 1101 * If we get here, it's an NMI that's not a memory or I/O 1102 * error. We can't truly tell if it's from IPMI or not 1103 * without sending a message, and sending a message is almost 1104 * impossible because of locking. 1105 */ 1106 1107 if (testing_nmi) { 1108 testing_nmi = 2; 1109 return NMI_HANDLED; 1110 } 1111 1112 /* If we are not expecting a timeout, ignore it. */ 1113 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) 1114 return NMI_DONE; 1115 1116 if (preaction_val != WDOG_PRETIMEOUT_NMI) 1117 return NMI_DONE; 1118 1119 /* 1120 * If no one else handled the NMI, we assume it was the IPMI 1121 * watchdog. 1122 */ 1123 if (preop_val == WDOG_PREOP_PANIC) { 1124 /* On some machines, the heartbeat will give 1125 an error and not work unless we re-enable 1126 the timer. So do so. */ 1127 atomic_set(&pretimeout_since_last_heartbeat, 1); 1128 if (atomic_inc_and_test(&preop_panic_excl)) 1129 nmi_panic(regs, "pre-timeout"); 1130 } 1131 1132 return NMI_HANDLED; 1133 } 1134 #endif 1135 1136 static int wdog_reboot_handler(struct notifier_block *this, 1137 unsigned long code, 1138 void *unused) 1139 { 1140 static int reboot_event_handled; 1141 1142 if ((watchdog_user) && (!reboot_event_handled)) { 1143 /* Make sure we only do this once. */ 1144 reboot_event_handled = 1; 1145 1146 if (code == SYS_POWER_OFF || code == SYS_HALT) { 1147 /* Disable the WDT if we are shutting down. */ 1148 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 1149 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 1150 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { 1151 /* Set a long timer to let the reboot happen or 1152 reset if it hangs, but only if the watchdog 1153 timer was already running. */ 1154 if (timeout < 120) 1155 timeout = 120; 1156 pretimeout = 0; 1157 ipmi_watchdog_state = WDOG_TIMEOUT_RESET; 1158 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 1159 } 1160 } 1161 return NOTIFY_OK; 1162 } 1163 1164 static struct notifier_block wdog_reboot_notifier = { 1165 .notifier_call = wdog_reboot_handler, 1166 .next = NULL, 1167 .priority = 0 1168 }; 1169 1170 static void ipmi_new_smi(int if_num, struct device *device) 1171 { 1172 ipmi_register_watchdog(if_num); 1173 } 1174 1175 static void ipmi_smi_gone(int if_num) 1176 { 1177 ipmi_unregister_watchdog(if_num); 1178 } 1179 1180 static struct ipmi_smi_watcher smi_watcher = { 1181 .owner = THIS_MODULE, 1182 .new_smi = ipmi_new_smi, 1183 .smi_gone = ipmi_smi_gone 1184 }; 1185 1186 static int action_op(const char *inval, char *outval) 1187 { 1188 if (outval) 1189 strcpy(outval, action); 1190 1191 if (!inval) 1192 return 0; 1193 1194 if (strcmp(inval, "reset") == 0) 1195 action_val = WDOG_TIMEOUT_RESET; 1196 else if (strcmp(inval, "none") == 0) 1197 action_val = WDOG_TIMEOUT_NONE; 1198 else if (strcmp(inval, "power_cycle") == 0) 1199 action_val = WDOG_TIMEOUT_POWER_CYCLE; 1200 else if (strcmp(inval, "power_off") == 0) 1201 action_val = WDOG_TIMEOUT_POWER_DOWN; 1202 else 1203 return -EINVAL; 1204 strcpy(action, inval); 1205 return 0; 1206 } 1207 1208 static int preaction_op(const char *inval, char *outval) 1209 { 1210 if (outval) 1211 strcpy(outval, preaction); 1212 1213 if (!inval) 1214 return 0; 1215 1216 if (strcmp(inval, "pre_none") == 0) 1217 preaction_val = WDOG_PRETIMEOUT_NONE; 1218 else if (strcmp(inval, "pre_smi") == 0) 1219 preaction_val = WDOG_PRETIMEOUT_SMI; 1220 #ifdef HAVE_DIE_NMI 1221 else if (strcmp(inval, "pre_nmi") == 0) 1222 preaction_val = WDOG_PRETIMEOUT_NMI; 1223 #endif 1224 else if (strcmp(inval, "pre_int") == 0) 1225 preaction_val = WDOG_PRETIMEOUT_MSG_INT; 1226 else 1227 return -EINVAL; 1228 strcpy(preaction, inval); 1229 return 0; 1230 } 1231 1232 static int preop_op(const char *inval, char *outval) 1233 { 1234 if (outval) 1235 strcpy(outval, preop); 1236 1237 if (!inval) 1238 return 0; 1239 1240 if (strcmp(inval, "preop_none") == 0) 1241 preop_val = WDOG_PREOP_NONE; 1242 else if (strcmp(inval, "preop_panic") == 0) 1243 preop_val = WDOG_PREOP_PANIC; 1244 else if (strcmp(inval, "preop_give_data") == 0) 1245 preop_val = WDOG_PREOP_GIVE_DATA; 1246 else 1247 return -EINVAL; 1248 strcpy(preop, inval); 1249 return 0; 1250 } 1251 1252 static void check_parms(void) 1253 { 1254 #ifdef HAVE_DIE_NMI 1255 int do_nmi = 0; 1256 int rv; 1257 1258 if (preaction_val == WDOG_PRETIMEOUT_NMI) { 1259 do_nmi = 1; 1260 if (preop_val == WDOG_PREOP_GIVE_DATA) { 1261 pr_warn("Pretimeout op is to give data but NMI pretimeout is enabled, setting pretimeout op to none\n"); 1262 preop_op("preop_none", NULL); 1263 do_nmi = 0; 1264 } 1265 } 1266 if (do_nmi && !nmi_handler_registered) { 1267 rv = register_nmi_handler(NMI_UNKNOWN, ipmi_nmi, 0, 1268 "ipmi"); 1269 if (rv) { 1270 pr_warn("Can't register nmi handler\n"); 1271 return; 1272 } else 1273 nmi_handler_registered = 1; 1274 } else if (!do_nmi && nmi_handler_registered) { 1275 unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); 1276 nmi_handler_registered = 0; 1277 } 1278 #endif 1279 } 1280 1281 static int __init ipmi_wdog_init(void) 1282 { 1283 int rv; 1284 1285 if (action_op(action, NULL)) { 1286 action_op("reset", NULL); 1287 pr_info("Unknown action '%s', defaulting to reset\n", action); 1288 } 1289 1290 if (preaction_op(preaction, NULL)) { 1291 preaction_op("pre_none", NULL); 1292 pr_info("Unknown preaction '%s', defaulting to none\n", 1293 preaction); 1294 } 1295 1296 if (preop_op(preop, NULL)) { 1297 preop_op("preop_none", NULL); 1298 pr_info("Unknown preop '%s', defaulting to none\n", preop); 1299 } 1300 1301 check_parms(); 1302 1303 register_reboot_notifier(&wdog_reboot_notifier); 1304 1305 rv = ipmi_smi_watcher_register(&smi_watcher); 1306 if (rv) { 1307 #ifdef HAVE_DIE_NMI 1308 if (nmi_handler_registered) 1309 unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); 1310 #endif 1311 unregister_reboot_notifier(&wdog_reboot_notifier); 1312 pr_warn("can't register smi watcher\n"); 1313 return rv; 1314 } 1315 1316 pr_info("driver initialized\n"); 1317 1318 return 0; 1319 } 1320 1321 static void __exit ipmi_wdog_exit(void) 1322 { 1323 ipmi_smi_watcher_unregister(&smi_watcher); 1324 ipmi_unregister_watchdog(watchdog_ifnum); 1325 1326 #ifdef HAVE_DIE_NMI 1327 if (nmi_handler_registered) 1328 unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); 1329 #endif 1330 1331 unregister_reboot_notifier(&wdog_reboot_notifier); 1332 } 1333 module_exit(ipmi_wdog_exit); 1334 module_init(ipmi_wdog_init); 1335 MODULE_LICENSE("GPL"); 1336 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); 1337 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); 1338