1 /* 2 * ipmi_watchdog.c 3 * 4 * A watchdog timer based upon the IPMI interface. 5 * 6 * Author: MontaVista Software, Inc. 7 * Corey Minyard <minyard@mvista.com> 8 * source@mvista.com 9 * 10 * Copyright 2002 MontaVista Software Inc. 11 * 12 * This program is free software; you can redistribute it and/or modify it 13 * under the terms of the GNU General Public License as published by the 14 * Free Software Foundation; either version 2 of the License, or (at your 15 * option) any later version. 16 * 17 * 18 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED 19 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 20 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 23 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS 24 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 25 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR 26 * TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE 27 * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 * 29 * You should have received a copy of the GNU General Public License along 30 * with this program; if not, write to the Free Software Foundation, Inc., 31 * 675 Mass Ave, Cambridge, MA 02139, USA. 32 */ 33 34 #include <linux/module.h> 35 #include <linux/moduleparam.h> 36 #include <linux/ipmi.h> 37 #include <linux/ipmi_smi.h> 38 #include <linux/smp_lock.h> 39 #include <linux/watchdog.h> 40 #include <linux/miscdevice.h> 41 #include <linux/init.h> 42 #include <linux/completion.h> 43 #include <linux/kdebug.h> 44 #include <linux/rwsem.h> 45 #include <linux/errno.h> 46 #include <asm/uaccess.h> 47 #include <linux/notifier.h> 48 #include <linux/nmi.h> 49 #include <linux/reboot.h> 50 #include <linux/wait.h> 51 #include <linux/poll.h> 52 #include <linux/string.h> 53 #include <linux/ctype.h> 54 #include <linux/delay.h> 55 #include <asm/atomic.h> 56 57 #ifdef CONFIG_X86 58 /* 59 * This is ugly, but I've determined that x86 is the only architecture 60 * that can reasonably support the IPMI NMI watchdog timeout at this 61 * time. If another architecture adds this capability somehow, it 62 * will have to be a somewhat different mechanism and I have no idea 63 * how it will work. So in the unlikely event that another 64 * architecture supports this, we can figure out a good generic 65 * mechanism for it at that time. 66 */ 67 #include <asm/kdebug.h> 68 #define HAVE_DIE_NMI 69 #endif 70 71 #define PFX "IPMI Watchdog: " 72 73 /* 74 * The IPMI command/response information for the watchdog timer. 75 */ 76 77 /* values for byte 1 of the set command, byte 2 of the get response. */ 78 #define WDOG_DONT_LOG (1 << 7) 79 #define WDOG_DONT_STOP_ON_SET (1 << 6) 80 #define WDOG_SET_TIMER_USE(byte, use) \ 81 byte = ((byte) & 0xf8) | ((use) & 0x7) 82 #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) 83 #define WDOG_TIMER_USE_BIOS_FRB2 1 84 #define WDOG_TIMER_USE_BIOS_POST 2 85 #define WDOG_TIMER_USE_OS_LOAD 3 86 #define WDOG_TIMER_USE_SMS_OS 4 87 #define WDOG_TIMER_USE_OEM 5 88 89 /* values for byte 2 of the set command, byte 3 of the get response. */ 90 #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ 91 byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) 92 #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) 93 #define WDOG_PRETIMEOUT_NONE 0 94 #define WDOG_PRETIMEOUT_SMI 1 95 #define WDOG_PRETIMEOUT_NMI 2 96 #define WDOG_PRETIMEOUT_MSG_INT 3 97 98 /* Operations that can be performed on a pretimout. */ 99 #define WDOG_PREOP_NONE 0 100 #define WDOG_PREOP_PANIC 1 101 /* Cause data to be available to read. Doesn't work in NMI mode. */ 102 #define WDOG_PREOP_GIVE_DATA 2 103 104 /* Actions to perform on a full timeout. */ 105 #define WDOG_SET_TIMEOUT_ACT(byte, use) \ 106 byte = ((byte) & 0xf8) | ((use) & 0x7) 107 #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) 108 #define WDOG_TIMEOUT_NONE 0 109 #define WDOG_TIMEOUT_RESET 1 110 #define WDOG_TIMEOUT_POWER_DOWN 2 111 #define WDOG_TIMEOUT_POWER_CYCLE 3 112 113 /* 114 * Byte 3 of the get command, byte 4 of the get response is the 115 * pre-timeout in seconds. 116 */ 117 118 /* Bits for setting byte 4 of the set command, byte 5 of the get response. */ 119 #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1) 120 #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2) 121 #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3) 122 #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4) 123 #define WDOG_EXPIRE_CLEAR_OEM (1 << 5) 124 125 /* 126 * Setting/getting the watchdog timer value. This is for bytes 5 and 127 * 6 (the timeout time) of the set command, and bytes 6 and 7 (the 128 * timeout time) and 8 and 9 (the current countdown value) of the 129 * response. The timeout value is given in seconds (in the command it 130 * is 100ms intervals). 131 */ 132 #define WDOG_SET_TIMEOUT(byte1, byte2, val) \ 133 (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8) 134 #define WDOG_GET_TIMEOUT(byte1, byte2) \ 135 (((byte1) | ((byte2) << 8)) / 10) 136 137 #define IPMI_WDOG_RESET_TIMER 0x22 138 #define IPMI_WDOG_SET_TIMER 0x24 139 #define IPMI_WDOG_GET_TIMER 0x25 140 141 /* These are here until the real ones get into the watchdog.h interface. */ 142 #ifndef WDIOC_GETTIMEOUT 143 #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int) 144 #endif 145 #ifndef WDIOC_SET_PRETIMEOUT 146 #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int) 147 #endif 148 #ifndef WDIOC_GET_PRETIMEOUT 149 #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int) 150 #endif 151 152 static int nowayout = WATCHDOG_NOWAYOUT; 153 154 static ipmi_user_t watchdog_user; 155 static int watchdog_ifnum; 156 157 /* Default the timeout to 10 seconds. */ 158 static int timeout = 10; 159 160 /* The pre-timeout is disabled by default. */ 161 static int pretimeout; 162 163 /* Default action is to reset the board on a timeout. */ 164 static unsigned char action_val = WDOG_TIMEOUT_RESET; 165 166 static char action[16] = "reset"; 167 168 static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; 169 170 static char preaction[16] = "pre_none"; 171 172 static unsigned char preop_val = WDOG_PREOP_NONE; 173 174 static char preop[16] = "preop_none"; 175 static DEFINE_SPINLOCK(ipmi_read_lock); 176 static char data_to_read; 177 static DECLARE_WAIT_QUEUE_HEAD(read_q); 178 static struct fasync_struct *fasync_q; 179 static char pretimeout_since_last_heartbeat; 180 static char expect_close; 181 182 static int ifnum_to_use = -1; 183 184 /* Parameters to ipmi_set_timeout */ 185 #define IPMI_SET_TIMEOUT_NO_HB 0 186 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 187 #define IPMI_SET_TIMEOUT_FORCE_HB 2 188 189 static int ipmi_set_timeout(int do_heartbeat); 190 static void ipmi_register_watchdog(int ipmi_intf); 191 static void ipmi_unregister_watchdog(int ipmi_intf); 192 193 /* 194 * If true, the driver will start running as soon as it is configured 195 * and ready. 196 */ 197 static int start_now; 198 199 static int set_param_int(const char *val, struct kernel_param *kp) 200 { 201 char *endp; 202 int l; 203 int rv = 0; 204 205 if (!val) 206 return -EINVAL; 207 l = simple_strtoul(val, &endp, 0); 208 if (endp == val) 209 return -EINVAL; 210 211 *((int *)kp->arg) = l; 212 if (watchdog_user) 213 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 214 215 return rv; 216 } 217 218 static int get_param_int(char *buffer, struct kernel_param *kp) 219 { 220 return sprintf(buffer, "%i", *((int *)kp->arg)); 221 } 222 223 typedef int (*action_fn)(const char *intval, char *outval); 224 225 static int action_op(const char *inval, char *outval); 226 static int preaction_op(const char *inval, char *outval); 227 static int preop_op(const char *inval, char *outval); 228 static void check_parms(void); 229 230 static int set_param_str(const char *val, struct kernel_param *kp) 231 { 232 action_fn fn = (action_fn) kp->arg; 233 int rv = 0; 234 char valcp[16]; 235 char *s; 236 237 strncpy(valcp, val, 16); 238 valcp[15] = '\0'; 239 240 s = strstrip(valcp); 241 242 rv = fn(s, NULL); 243 if (rv) 244 goto out; 245 246 check_parms(); 247 if (watchdog_user) 248 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 249 250 out: 251 return rv; 252 } 253 254 static int get_param_str(char *buffer, struct kernel_param *kp) 255 { 256 action_fn fn = (action_fn) kp->arg; 257 int rv; 258 259 rv = fn(NULL, buffer); 260 if (rv) 261 return rv; 262 return strlen(buffer); 263 } 264 265 266 static int set_param_wdog_ifnum(const char *val, struct kernel_param *kp) 267 { 268 int rv = param_set_int(val, kp); 269 if (rv) 270 return rv; 271 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum)) 272 return 0; 273 274 ipmi_unregister_watchdog(watchdog_ifnum); 275 ipmi_register_watchdog(ifnum_to_use); 276 return 0; 277 } 278 279 module_param_call(ifnum_to_use, set_param_wdog_ifnum, get_param_int, 280 &ifnum_to_use, 0644); 281 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog " 282 "timer. Setting to -1 defaults to the first registered " 283 "interface"); 284 285 module_param_call(timeout, set_param_int, get_param_int, &timeout, 0644); 286 MODULE_PARM_DESC(timeout, "Timeout value in seconds."); 287 288 module_param_call(pretimeout, set_param_int, get_param_int, &pretimeout, 0644); 289 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); 290 291 module_param_call(action, set_param_str, get_param_str, action_op, 0644); 292 MODULE_PARM_DESC(action, "Timeout action. One of: " 293 "reset, none, power_cycle, power_off."); 294 295 module_param_call(preaction, set_param_str, get_param_str, preaction_op, 0644); 296 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " 297 "pre_none, pre_smi, pre_nmi, pre_int."); 298 299 module_param_call(preop, set_param_str, get_param_str, preop_op, 0644); 300 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " 301 "preop_none, preop_panic, preop_give_data."); 302 303 module_param(start_now, int, 0444); 304 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" 305 "soon as the driver is loaded."); 306 307 module_param(nowayout, int, 0644); 308 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started " 309 "(default=CONFIG_WATCHDOG_NOWAYOUT)"); 310 311 /* Default state of the timer. */ 312 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 313 314 /* If shutting down via IPMI, we ignore the heartbeat. */ 315 static int ipmi_ignore_heartbeat; 316 317 /* Is someone using the watchdog? Only one user is allowed. */ 318 static unsigned long ipmi_wdog_open; 319 320 /* 321 * If set to 1, the heartbeat command will set the state to reset and 322 * start the timer. The timer doesn't normally run when the driver is 323 * first opened until the heartbeat is set the first time, this 324 * variable is used to accomplish this. 325 */ 326 static int ipmi_start_timer_on_heartbeat; 327 328 /* IPMI version of the BMC. */ 329 static unsigned char ipmi_version_major; 330 static unsigned char ipmi_version_minor; 331 332 /* If a pretimeout occurs, this is used to allow only one panic to happen. */ 333 static atomic_t preop_panic_excl = ATOMIC_INIT(-1); 334 335 #ifdef HAVE_DIE_NMI 336 static int testing_nmi; 337 static int nmi_handler_registered; 338 #endif 339 340 static int ipmi_heartbeat(void); 341 342 /* 343 * We use a mutex to make sure that only one thing can send a set 344 * timeout at one time, because we only have one copy of the data. 345 * The mutex is claimed when the set_timeout is sent and freed 346 * when both messages are free. 347 */ 348 static atomic_t set_timeout_tofree = ATOMIC_INIT(0); 349 static DEFINE_MUTEX(set_timeout_lock); 350 static DECLARE_COMPLETION(set_timeout_wait); 351 static void set_timeout_free_smi(struct ipmi_smi_msg *msg) 352 { 353 if (atomic_dec_and_test(&set_timeout_tofree)) 354 complete(&set_timeout_wait); 355 } 356 static void set_timeout_free_recv(struct ipmi_recv_msg *msg) 357 { 358 if (atomic_dec_and_test(&set_timeout_tofree)) 359 complete(&set_timeout_wait); 360 } 361 static struct ipmi_smi_msg set_timeout_smi_msg = { 362 .done = set_timeout_free_smi 363 }; 364 static struct ipmi_recv_msg set_timeout_recv_msg = { 365 .done = set_timeout_free_recv 366 }; 367 368 static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, 369 struct ipmi_recv_msg *recv_msg, 370 int *send_heartbeat_now) 371 { 372 struct kernel_ipmi_msg msg; 373 unsigned char data[6]; 374 int rv; 375 struct ipmi_system_interface_addr addr; 376 int hbnow = 0; 377 378 379 /* These can be cleared as we are setting the timeout. */ 380 pretimeout_since_last_heartbeat = 0; 381 382 data[0] = 0; 383 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); 384 385 if ((ipmi_version_major > 1) 386 || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) { 387 /* This is an IPMI 1.5-only feature. */ 388 data[0] |= WDOG_DONT_STOP_ON_SET; 389 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { 390 /* 391 * In ipmi 1.0, setting the timer stops the watchdog, we 392 * need to start it back up again. 393 */ 394 hbnow = 1; 395 } 396 397 data[1] = 0; 398 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); 399 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { 400 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); 401 data[2] = pretimeout; 402 } else { 403 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); 404 data[2] = 0; /* No pretimeout. */ 405 } 406 data[3] = 0; 407 WDOG_SET_TIMEOUT(data[4], data[5], timeout); 408 409 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 410 addr.channel = IPMI_BMC_CHANNEL; 411 addr.lun = 0; 412 413 msg.netfn = 0x06; 414 msg.cmd = IPMI_WDOG_SET_TIMER; 415 msg.data = data; 416 msg.data_len = sizeof(data); 417 rv = ipmi_request_supply_msgs(watchdog_user, 418 (struct ipmi_addr *) &addr, 419 0, 420 &msg, 421 NULL, 422 smi_msg, 423 recv_msg, 424 1); 425 if (rv) { 426 printk(KERN_WARNING PFX "set timeout error: %d\n", 427 rv); 428 } 429 430 if (send_heartbeat_now) 431 *send_heartbeat_now = hbnow; 432 433 return rv; 434 } 435 436 static int ipmi_set_timeout(int do_heartbeat) 437 { 438 int send_heartbeat_now; 439 int rv; 440 441 442 /* We can only send one of these at a time. */ 443 mutex_lock(&set_timeout_lock); 444 445 atomic_set(&set_timeout_tofree, 2); 446 447 rv = i_ipmi_set_timeout(&set_timeout_smi_msg, 448 &set_timeout_recv_msg, 449 &send_heartbeat_now); 450 if (rv) { 451 mutex_unlock(&set_timeout_lock); 452 goto out; 453 } 454 455 wait_for_completion(&set_timeout_wait); 456 457 mutex_unlock(&set_timeout_lock); 458 459 if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) 460 || ((send_heartbeat_now) 461 && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) 462 rv = ipmi_heartbeat(); 463 464 out: 465 return rv; 466 } 467 468 static atomic_t panic_done_count = ATOMIC_INIT(0); 469 470 static void panic_smi_free(struct ipmi_smi_msg *msg) 471 { 472 atomic_dec(&panic_done_count); 473 } 474 static void panic_recv_free(struct ipmi_recv_msg *msg) 475 { 476 atomic_dec(&panic_done_count); 477 } 478 479 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = { 480 .done = panic_smi_free 481 }; 482 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = { 483 .done = panic_recv_free 484 }; 485 486 static void panic_halt_ipmi_heartbeat(void) 487 { 488 struct kernel_ipmi_msg msg; 489 struct ipmi_system_interface_addr addr; 490 int rv; 491 492 /* 493 * Don't reset the timer if we have the timer turned off, that 494 * re-enables the watchdog. 495 */ 496 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) 497 return; 498 499 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 500 addr.channel = IPMI_BMC_CHANNEL; 501 addr.lun = 0; 502 503 msg.netfn = 0x06; 504 msg.cmd = IPMI_WDOG_RESET_TIMER; 505 msg.data = NULL; 506 msg.data_len = 0; 507 rv = ipmi_request_supply_msgs(watchdog_user, 508 (struct ipmi_addr *) &addr, 509 0, 510 &msg, 511 NULL, 512 &panic_halt_heartbeat_smi_msg, 513 &panic_halt_heartbeat_recv_msg, 514 1); 515 if (!rv) 516 atomic_add(2, &panic_done_count); 517 } 518 519 static struct ipmi_smi_msg panic_halt_smi_msg = { 520 .done = panic_smi_free 521 }; 522 static struct ipmi_recv_msg panic_halt_recv_msg = { 523 .done = panic_recv_free 524 }; 525 526 /* 527 * Special call, doesn't claim any locks. This is only to be called 528 * at panic or halt time, in run-to-completion mode, when the caller 529 * is the only CPU and the only thing that will be going is these IPMI 530 * calls. 531 */ 532 static void panic_halt_ipmi_set_timeout(void) 533 { 534 int send_heartbeat_now; 535 int rv; 536 537 /* Wait for the messages to be free. */ 538 while (atomic_read(&panic_done_count) != 0) 539 ipmi_poll_interface(watchdog_user); 540 rv = i_ipmi_set_timeout(&panic_halt_smi_msg, 541 &panic_halt_recv_msg, 542 &send_heartbeat_now); 543 if (!rv) { 544 atomic_add(2, &panic_done_count); 545 if (send_heartbeat_now) 546 panic_halt_ipmi_heartbeat(); 547 } else 548 printk(KERN_WARNING PFX 549 "Unable to extend the watchdog timeout."); 550 while (atomic_read(&panic_done_count) != 0) 551 ipmi_poll_interface(watchdog_user); 552 } 553 554 /* 555 * We use a mutex to make sure that only one thing can send a 556 * heartbeat at one time, because we only have one copy of the data. 557 * The semaphore is claimed when the set_timeout is sent and freed 558 * when both messages are free. 559 */ 560 static atomic_t heartbeat_tofree = ATOMIC_INIT(0); 561 static DEFINE_MUTEX(heartbeat_lock); 562 static DECLARE_COMPLETION(heartbeat_wait); 563 static void heartbeat_free_smi(struct ipmi_smi_msg *msg) 564 { 565 if (atomic_dec_and_test(&heartbeat_tofree)) 566 complete(&heartbeat_wait); 567 } 568 static void heartbeat_free_recv(struct ipmi_recv_msg *msg) 569 { 570 if (atomic_dec_and_test(&heartbeat_tofree)) 571 complete(&heartbeat_wait); 572 } 573 static struct ipmi_smi_msg heartbeat_smi_msg = { 574 .done = heartbeat_free_smi 575 }; 576 static struct ipmi_recv_msg heartbeat_recv_msg = { 577 .done = heartbeat_free_recv 578 }; 579 580 static int ipmi_heartbeat(void) 581 { 582 struct kernel_ipmi_msg msg; 583 int rv; 584 struct ipmi_system_interface_addr addr; 585 586 if (ipmi_ignore_heartbeat) 587 return 0; 588 589 if (ipmi_start_timer_on_heartbeat) { 590 ipmi_start_timer_on_heartbeat = 0; 591 ipmi_watchdog_state = action_val; 592 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 593 } else if (pretimeout_since_last_heartbeat) { 594 /* 595 * A pretimeout occurred, make sure we set the timeout. 596 * We don't want to set the action, though, we want to 597 * leave that alone (thus it can't be combined with the 598 * above operation. 599 */ 600 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 601 } 602 603 mutex_lock(&heartbeat_lock); 604 605 atomic_set(&heartbeat_tofree, 2); 606 607 /* 608 * Don't reset the timer if we have the timer turned off, that 609 * re-enables the watchdog. 610 */ 611 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) { 612 mutex_unlock(&heartbeat_lock); 613 return 0; 614 } 615 616 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 617 addr.channel = IPMI_BMC_CHANNEL; 618 addr.lun = 0; 619 620 msg.netfn = 0x06; 621 msg.cmd = IPMI_WDOG_RESET_TIMER; 622 msg.data = NULL; 623 msg.data_len = 0; 624 rv = ipmi_request_supply_msgs(watchdog_user, 625 (struct ipmi_addr *) &addr, 626 0, 627 &msg, 628 NULL, 629 &heartbeat_smi_msg, 630 &heartbeat_recv_msg, 631 1); 632 if (rv) { 633 mutex_unlock(&heartbeat_lock); 634 printk(KERN_WARNING PFX "heartbeat failure: %d\n", 635 rv); 636 return rv; 637 } 638 639 /* Wait for the heartbeat to be sent. */ 640 wait_for_completion(&heartbeat_wait); 641 642 if (heartbeat_recv_msg.msg.data[0] != 0) { 643 /* 644 * Got an error in the heartbeat response. It was already 645 * reported in ipmi_wdog_msg_handler, but we should return 646 * an error here. 647 */ 648 rv = -EINVAL; 649 } 650 651 mutex_unlock(&heartbeat_lock); 652 653 return rv; 654 } 655 656 static struct watchdog_info ident = { 657 .options = 0, /* WDIOF_SETTIMEOUT, */ 658 .firmware_version = 1, 659 .identity = "IPMI" 660 }; 661 662 static int ipmi_ioctl(struct file *file, 663 unsigned int cmd, unsigned long arg) 664 { 665 void __user *argp = (void __user *)arg; 666 int i; 667 int val; 668 669 switch (cmd) { 670 case WDIOC_GETSUPPORT: 671 i = copy_to_user(argp, &ident, sizeof(ident)); 672 return i ? -EFAULT : 0; 673 674 case WDIOC_SETTIMEOUT: 675 i = copy_from_user(&val, argp, sizeof(int)); 676 if (i) 677 return -EFAULT; 678 timeout = val; 679 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 680 681 case WDIOC_GETTIMEOUT: 682 i = copy_to_user(argp, &timeout, sizeof(timeout)); 683 if (i) 684 return -EFAULT; 685 return 0; 686 687 case WDIOC_SET_PRETIMEOUT: 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_GET_PRETIMEOUT: 696 case WDIOC_GETPRETIMEOUT: 697 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); 698 if (i) 699 return -EFAULT; 700 return 0; 701 702 case WDIOC_KEEPALIVE: 703 return ipmi_heartbeat(); 704 705 case WDIOC_SETOPTIONS: 706 i = copy_from_user(&val, argp, sizeof(int)); 707 if (i) 708 return -EFAULT; 709 if (val & WDIOS_DISABLECARD) { 710 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 711 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 712 ipmi_start_timer_on_heartbeat = 0; 713 } 714 715 if (val & WDIOS_ENABLECARD) { 716 ipmi_watchdog_state = action_val; 717 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 718 } 719 return 0; 720 721 case WDIOC_GETSTATUS: 722 val = 0; 723 i = copy_to_user(argp, &val, sizeof(val)); 724 if (i) 725 return -EFAULT; 726 return 0; 727 728 default: 729 return -ENOIOCTLCMD; 730 } 731 } 732 733 static long ipmi_unlocked_ioctl(struct file *file, 734 unsigned int cmd, 735 unsigned long arg) 736 { 737 int ret; 738 739 lock_kernel(); 740 ret = ipmi_ioctl(file, cmd, arg); 741 unlock_kernel(); 742 743 return ret; 744 } 745 746 static ssize_t ipmi_write(struct file *file, 747 const char __user *buf, 748 size_t len, 749 loff_t *ppos) 750 { 751 int rv; 752 753 if (len) { 754 if (!nowayout) { 755 size_t i; 756 757 /* In case it was set long ago */ 758 expect_close = 0; 759 760 for (i = 0; i != len; i++) { 761 char c; 762 763 if (get_user(c, buf + i)) 764 return -EFAULT; 765 if (c == 'V') 766 expect_close = 42; 767 } 768 } 769 rv = ipmi_heartbeat(); 770 if (rv) 771 return rv; 772 } 773 return len; 774 } 775 776 static ssize_t ipmi_read(struct file *file, 777 char __user *buf, 778 size_t count, 779 loff_t *ppos) 780 { 781 int rv = 0; 782 wait_queue_t wait; 783 784 if (count <= 0) 785 return 0; 786 787 /* 788 * Reading returns if the pretimeout has gone off, and it only does 789 * it once per pretimeout. 790 */ 791 spin_lock(&ipmi_read_lock); 792 if (!data_to_read) { 793 if (file->f_flags & O_NONBLOCK) { 794 rv = -EAGAIN; 795 goto out; 796 } 797 798 init_waitqueue_entry(&wait, current); 799 add_wait_queue(&read_q, &wait); 800 while (!data_to_read) { 801 set_current_state(TASK_INTERRUPTIBLE); 802 spin_unlock(&ipmi_read_lock); 803 schedule(); 804 spin_lock(&ipmi_read_lock); 805 } 806 remove_wait_queue(&read_q, &wait); 807 808 if (signal_pending(current)) { 809 rv = -ERESTARTSYS; 810 goto out; 811 } 812 } 813 data_to_read = 0; 814 815 out: 816 spin_unlock(&ipmi_read_lock); 817 818 if (rv == 0) { 819 if (copy_to_user(buf, &data_to_read, 1)) 820 rv = -EFAULT; 821 else 822 rv = 1; 823 } 824 825 return rv; 826 } 827 828 static int ipmi_open(struct inode *ino, struct file *filep) 829 { 830 switch (iminor(ino)) { 831 case WATCHDOG_MINOR: 832 if (test_and_set_bit(0, &ipmi_wdog_open)) 833 return -EBUSY; 834 835 cycle_kernel_lock(); 836 837 /* 838 * Don't start the timer now, let it start on the 839 * first heartbeat. 840 */ 841 ipmi_start_timer_on_heartbeat = 1; 842 return nonseekable_open(ino, filep); 843 844 default: 845 return (-ENODEV); 846 } 847 } 848 849 static unsigned int ipmi_poll(struct file *file, poll_table *wait) 850 { 851 unsigned int mask = 0; 852 853 poll_wait(file, &read_q, wait); 854 855 spin_lock(&ipmi_read_lock); 856 if (data_to_read) 857 mask |= (POLLIN | POLLRDNORM); 858 spin_unlock(&ipmi_read_lock); 859 860 return mask; 861 } 862 863 static int ipmi_fasync(int fd, struct file *file, int on) 864 { 865 int result; 866 867 result = fasync_helper(fd, file, on, &fasync_q); 868 869 return (result); 870 } 871 872 static int ipmi_close(struct inode *ino, struct file *filep) 873 { 874 if (iminor(ino) == WATCHDOG_MINOR) { 875 if (expect_close == 42) { 876 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 877 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 878 } else { 879 printk(KERN_CRIT PFX 880 "Unexpected close, not stopping watchdog!\n"); 881 ipmi_heartbeat(); 882 } 883 clear_bit(0, &ipmi_wdog_open); 884 } 885 886 expect_close = 0; 887 888 return 0; 889 } 890 891 static const struct file_operations ipmi_wdog_fops = { 892 .owner = THIS_MODULE, 893 .read = ipmi_read, 894 .poll = ipmi_poll, 895 .write = ipmi_write, 896 .unlocked_ioctl = ipmi_unlocked_ioctl, 897 .open = ipmi_open, 898 .release = ipmi_close, 899 .fasync = ipmi_fasync, 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.data[0] != 0) { 912 printk(KERN_ERR PFX "response: Error %x on cmd %x\n", 913 msg->msg.data[0], 914 msg->msg.cmd); 915 } 916 917 ipmi_free_recv_msg(msg); 918 } 919 920 static void ipmi_wdog_pretimeout_handler(void *handler_data) 921 { 922 if (preaction_val != WDOG_PRETIMEOUT_NONE) { 923 if (preop_val == WDOG_PREOP_PANIC) { 924 if (atomic_inc_and_test(&preop_panic_excl)) 925 panic("Watchdog pre-timeout"); 926 } else if (preop_val == WDOG_PREOP_GIVE_DATA) { 927 spin_lock(&ipmi_read_lock); 928 data_to_read = 1; 929 wake_up_interruptible(&read_q); 930 kill_fasync(&fasync_q, SIGIO, POLL_IN); 931 932 spin_unlock(&ipmi_read_lock); 933 } 934 } 935 936 /* 937 * On some machines, the heartbeat will give an error and not 938 * work unless we re-enable the timer. So do so. 939 */ 940 pretimeout_since_last_heartbeat = 1; 941 } 942 943 static struct ipmi_user_hndl ipmi_hndlrs = { 944 .ipmi_recv_hndl = ipmi_wdog_msg_handler, 945 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler 946 }; 947 948 static void ipmi_register_watchdog(int ipmi_intf) 949 { 950 int rv = -EBUSY; 951 952 if (watchdog_user) 953 goto out; 954 955 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf)) 956 goto out; 957 958 watchdog_ifnum = ipmi_intf; 959 960 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); 961 if (rv < 0) { 962 printk(KERN_CRIT PFX "Unable to register with ipmi\n"); 963 goto out; 964 } 965 966 ipmi_get_version(watchdog_user, 967 &ipmi_version_major, 968 &ipmi_version_minor); 969 970 rv = misc_register(&ipmi_wdog_miscdev); 971 if (rv < 0) { 972 ipmi_destroy_user(watchdog_user); 973 watchdog_user = NULL; 974 printk(KERN_CRIT PFX "Unable to register misc device\n"); 975 } 976 977 #ifdef HAVE_DIE_NMI 978 if (nmi_handler_registered) { 979 int old_pretimeout = pretimeout; 980 int old_timeout = timeout; 981 int old_preop_val = preop_val; 982 983 /* 984 * Set the pretimeout to go off in a second and give 985 * ourselves plenty of time to stop the timer. 986 */ 987 ipmi_watchdog_state = WDOG_TIMEOUT_RESET; 988 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */ 989 pretimeout = 99; 990 timeout = 100; 991 992 testing_nmi = 1; 993 994 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 995 if (rv) { 996 printk(KERN_WARNING PFX "Error starting timer to" 997 " test NMI: 0x%x. The NMI pretimeout will" 998 " likely not work\n", rv); 999 rv = 0; 1000 goto out_restore; 1001 } 1002 1003 msleep(1500); 1004 1005 if (testing_nmi != 2) { 1006 printk(KERN_WARNING PFX "IPMI NMI didn't seem to" 1007 " occur. The NMI pretimeout will" 1008 " likely not work\n"); 1009 } 1010 out_restore: 1011 testing_nmi = 0; 1012 preop_val = old_preop_val; 1013 pretimeout = old_pretimeout; 1014 timeout = old_timeout; 1015 } 1016 #endif 1017 1018 out: 1019 if ((start_now) && (rv == 0)) { 1020 /* Run from startup, so start the timer now. */ 1021 start_now = 0; /* Disable this function after first startup. */ 1022 ipmi_watchdog_state = action_val; 1023 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 1024 printk(KERN_INFO PFX "Starting now!\n"); 1025 } else { 1026 /* Stop the timer now. */ 1027 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 1028 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 1029 } 1030 } 1031 1032 static void ipmi_unregister_watchdog(int ipmi_intf) 1033 { 1034 int rv; 1035 1036 if (!watchdog_user) 1037 goto out; 1038 1039 if (watchdog_ifnum != ipmi_intf) 1040 goto out; 1041 1042 /* Make sure no one can call us any more. */ 1043 misc_deregister(&ipmi_wdog_miscdev); 1044 1045 /* 1046 * Wait to make sure the message makes it out. The lower layer has 1047 * pointers to our buffers, we want to make sure they are done before 1048 * we release our memory. 1049 */ 1050 while (atomic_read(&set_timeout_tofree)) 1051 schedule_timeout_uninterruptible(1); 1052 1053 /* Disconnect from IPMI. */ 1054 rv = ipmi_destroy_user(watchdog_user); 1055 if (rv) { 1056 printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n", 1057 rv); 1058 } 1059 watchdog_user = NULL; 1060 1061 out: 1062 return; 1063 } 1064 1065 #ifdef HAVE_DIE_NMI 1066 static int 1067 ipmi_nmi(struct notifier_block *self, unsigned long val, void *data) 1068 { 1069 struct die_args *args = data; 1070 1071 if (val != DIE_NMI) 1072 return NOTIFY_OK; 1073 1074 /* Hack, if it's a memory or I/O error, ignore it. */ 1075 if (args->err & 0xc0) 1076 return NOTIFY_OK; 1077 1078 /* 1079 * If we get here, it's an NMI that's not a memory or I/O 1080 * error. We can't truly tell if it's from IPMI or not 1081 * without sending a message, and sending a message is almost 1082 * impossible because of locking. 1083 */ 1084 1085 if (testing_nmi) { 1086 testing_nmi = 2; 1087 return NOTIFY_STOP; 1088 } 1089 1090 /* If we are not expecting a timeout, ignore it. */ 1091 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) 1092 return NOTIFY_OK; 1093 1094 if (preaction_val != WDOG_PRETIMEOUT_NMI) 1095 return NOTIFY_OK; 1096 1097 /* 1098 * If no one else handled the NMI, we assume it was the IPMI 1099 * watchdog. 1100 */ 1101 if (preop_val == WDOG_PREOP_PANIC) { 1102 /* On some machines, the heartbeat will give 1103 an error and not work unless we re-enable 1104 the timer. So do so. */ 1105 pretimeout_since_last_heartbeat = 1; 1106 if (atomic_inc_and_test(&preop_panic_excl)) 1107 panic(PFX "pre-timeout"); 1108 } 1109 1110 return NOTIFY_STOP; 1111 } 1112 1113 static struct notifier_block ipmi_nmi_handler = { 1114 .notifier_call = ipmi_nmi 1115 }; 1116 #endif 1117 1118 static int wdog_reboot_handler(struct notifier_block *this, 1119 unsigned long code, 1120 void *unused) 1121 { 1122 static int reboot_event_handled; 1123 1124 if ((watchdog_user) && (!reboot_event_handled)) { 1125 /* Make sure we only do this once. */ 1126 reboot_event_handled = 1; 1127 1128 if (code == SYS_POWER_OFF || code == SYS_HALT) { 1129 /* Disable the WDT if we are shutting down. */ 1130 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 1131 panic_halt_ipmi_set_timeout(); 1132 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { 1133 /* Set a long timer to let the reboot happens, but 1134 reboot if it hangs, but only if the watchdog 1135 timer was already running. */ 1136 timeout = 120; 1137 pretimeout = 0; 1138 ipmi_watchdog_state = WDOG_TIMEOUT_RESET; 1139 panic_halt_ipmi_set_timeout(); 1140 } 1141 } 1142 return NOTIFY_OK; 1143 } 1144 1145 static struct notifier_block wdog_reboot_notifier = { 1146 .notifier_call = wdog_reboot_handler, 1147 .next = NULL, 1148 .priority = 0 1149 }; 1150 1151 static int wdog_panic_handler(struct notifier_block *this, 1152 unsigned long event, 1153 void *unused) 1154 { 1155 static int panic_event_handled; 1156 1157 /* On a panic, if we have a panic timeout, make sure to extend 1158 the watchdog timer to a reasonable value to complete the 1159 panic, if the watchdog timer is running. Plus the 1160 pretimeout is meaningless at panic time. */ 1161 if (watchdog_user && !panic_event_handled && 1162 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { 1163 /* Make sure we do this only once. */ 1164 panic_event_handled = 1; 1165 1166 timeout = 255; 1167 pretimeout = 0; 1168 panic_halt_ipmi_set_timeout(); 1169 } 1170 1171 return NOTIFY_OK; 1172 } 1173 1174 static struct notifier_block wdog_panic_notifier = { 1175 .notifier_call = wdog_panic_handler, 1176 .next = NULL, 1177 .priority = 150 /* priority: INT_MAX >= x >= 0 */ 1178 }; 1179 1180 1181 static void ipmi_new_smi(int if_num, struct device *device) 1182 { 1183 ipmi_register_watchdog(if_num); 1184 } 1185 1186 static void ipmi_smi_gone(int if_num) 1187 { 1188 ipmi_unregister_watchdog(if_num); 1189 } 1190 1191 static struct ipmi_smi_watcher smi_watcher = { 1192 .owner = THIS_MODULE, 1193 .new_smi = ipmi_new_smi, 1194 .smi_gone = ipmi_smi_gone 1195 }; 1196 1197 static int action_op(const char *inval, char *outval) 1198 { 1199 if (outval) 1200 strcpy(outval, action); 1201 1202 if (!inval) 1203 return 0; 1204 1205 if (strcmp(inval, "reset") == 0) 1206 action_val = WDOG_TIMEOUT_RESET; 1207 else if (strcmp(inval, "none") == 0) 1208 action_val = WDOG_TIMEOUT_NONE; 1209 else if (strcmp(inval, "power_cycle") == 0) 1210 action_val = WDOG_TIMEOUT_POWER_CYCLE; 1211 else if (strcmp(inval, "power_off") == 0) 1212 action_val = WDOG_TIMEOUT_POWER_DOWN; 1213 else 1214 return -EINVAL; 1215 strcpy(action, inval); 1216 return 0; 1217 } 1218 1219 static int preaction_op(const char *inval, char *outval) 1220 { 1221 if (outval) 1222 strcpy(outval, preaction); 1223 1224 if (!inval) 1225 return 0; 1226 1227 if (strcmp(inval, "pre_none") == 0) 1228 preaction_val = WDOG_PRETIMEOUT_NONE; 1229 else if (strcmp(inval, "pre_smi") == 0) 1230 preaction_val = WDOG_PRETIMEOUT_SMI; 1231 #ifdef HAVE_DIE_NMI 1232 else if (strcmp(inval, "pre_nmi") == 0) 1233 preaction_val = WDOG_PRETIMEOUT_NMI; 1234 #endif 1235 else if (strcmp(inval, "pre_int") == 0) 1236 preaction_val = WDOG_PRETIMEOUT_MSG_INT; 1237 else 1238 return -EINVAL; 1239 strcpy(preaction, inval); 1240 return 0; 1241 } 1242 1243 static int preop_op(const char *inval, char *outval) 1244 { 1245 if (outval) 1246 strcpy(outval, preop); 1247 1248 if (!inval) 1249 return 0; 1250 1251 if (strcmp(inval, "preop_none") == 0) 1252 preop_val = WDOG_PREOP_NONE; 1253 else if (strcmp(inval, "preop_panic") == 0) 1254 preop_val = WDOG_PREOP_PANIC; 1255 else if (strcmp(inval, "preop_give_data") == 0) 1256 preop_val = WDOG_PREOP_GIVE_DATA; 1257 else 1258 return -EINVAL; 1259 strcpy(preop, inval); 1260 return 0; 1261 } 1262 1263 static void check_parms(void) 1264 { 1265 #ifdef HAVE_DIE_NMI 1266 int do_nmi = 0; 1267 int rv; 1268 1269 if (preaction_val == WDOG_PRETIMEOUT_NMI) { 1270 do_nmi = 1; 1271 if (preop_val == WDOG_PREOP_GIVE_DATA) { 1272 printk(KERN_WARNING PFX "Pretimeout op is to give data" 1273 " but NMI pretimeout is enabled, setting" 1274 " pretimeout op to none\n"); 1275 preop_op("preop_none", NULL); 1276 do_nmi = 0; 1277 } 1278 } 1279 if (do_nmi && !nmi_handler_registered) { 1280 rv = register_die_notifier(&ipmi_nmi_handler); 1281 if (rv) { 1282 printk(KERN_WARNING PFX 1283 "Can't register nmi handler\n"); 1284 return; 1285 } else 1286 nmi_handler_registered = 1; 1287 } else if (!do_nmi && nmi_handler_registered) { 1288 unregister_die_notifier(&ipmi_nmi_handler); 1289 nmi_handler_registered = 0; 1290 } 1291 #endif 1292 } 1293 1294 static int __init ipmi_wdog_init(void) 1295 { 1296 int rv; 1297 1298 if (action_op(action, NULL)) { 1299 action_op("reset", NULL); 1300 printk(KERN_INFO PFX "Unknown action '%s', defaulting to" 1301 " reset\n", action); 1302 } 1303 1304 if (preaction_op(preaction, NULL)) { 1305 preaction_op("pre_none", NULL); 1306 printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to" 1307 " none\n", preaction); 1308 } 1309 1310 if (preop_op(preop, NULL)) { 1311 preop_op("preop_none", NULL); 1312 printk(KERN_INFO PFX "Unknown preop '%s', defaulting to" 1313 " none\n", preop); 1314 } 1315 1316 check_parms(); 1317 1318 register_reboot_notifier(&wdog_reboot_notifier); 1319 atomic_notifier_chain_register(&panic_notifier_list, 1320 &wdog_panic_notifier); 1321 1322 rv = ipmi_smi_watcher_register(&smi_watcher); 1323 if (rv) { 1324 #ifdef HAVE_DIE_NMI 1325 if (nmi_handler_registered) 1326 unregister_die_notifier(&ipmi_nmi_handler); 1327 #endif 1328 atomic_notifier_chain_unregister(&panic_notifier_list, 1329 &wdog_panic_notifier); 1330 unregister_reboot_notifier(&wdog_reboot_notifier); 1331 printk(KERN_WARNING PFX "can't register smi watcher\n"); 1332 return rv; 1333 } 1334 1335 printk(KERN_INFO PFX "driver initialized\n"); 1336 1337 return 0; 1338 } 1339 1340 static void __exit ipmi_wdog_exit(void) 1341 { 1342 ipmi_smi_watcher_unregister(&smi_watcher); 1343 ipmi_unregister_watchdog(watchdog_ifnum); 1344 1345 #ifdef HAVE_DIE_NMI 1346 if (nmi_handler_registered) 1347 unregister_die_notifier(&ipmi_nmi_handler); 1348 #endif 1349 1350 atomic_notifier_chain_unregister(&panic_notifier_list, 1351 &wdog_panic_notifier); 1352 unregister_reboot_notifier(&wdog_reboot_notifier); 1353 } 1354 module_exit(ipmi_wdog_exit); 1355 module_init(ipmi_wdog_init); 1356 MODULE_LICENSE("GPL"); 1357 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); 1358 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); 1359