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/config.h> 35 #include <linux/module.h> 36 #include <linux/moduleparam.h> 37 #include <linux/ipmi.h> 38 #include <linux/ipmi_smi.h> 39 #include <linux/watchdog.h> 40 #include <linux/miscdevice.h> 41 #include <linux/init.h> 42 #include <linux/rwsem.h> 43 #include <linux/errno.h> 44 #include <asm/uaccess.h> 45 #include <linux/notifier.h> 46 #include <linux/nmi.h> 47 #include <linux/reboot.h> 48 #include <linux/wait.h> 49 #include <linux/poll.h> 50 #ifdef CONFIG_X86_LOCAL_APIC 51 #include <asm/apic.h> 52 #endif 53 54 #define PFX "IPMI Watchdog: " 55 56 /* 57 * The IPMI command/response information for the watchdog timer. 58 */ 59 60 /* values for byte 1 of the set command, byte 2 of the get response. */ 61 #define WDOG_DONT_LOG (1 << 7) 62 #define WDOG_DONT_STOP_ON_SET (1 << 6) 63 #define WDOG_SET_TIMER_USE(byte, use) \ 64 byte = ((byte) & 0xf8) | ((use) & 0x7) 65 #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) 66 #define WDOG_TIMER_USE_BIOS_FRB2 1 67 #define WDOG_TIMER_USE_BIOS_POST 2 68 #define WDOG_TIMER_USE_OS_LOAD 3 69 #define WDOG_TIMER_USE_SMS_OS 4 70 #define WDOG_TIMER_USE_OEM 5 71 72 /* values for byte 2 of the set command, byte 3 of the get response. */ 73 #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ 74 byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) 75 #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) 76 #define WDOG_PRETIMEOUT_NONE 0 77 #define WDOG_PRETIMEOUT_SMI 1 78 #define WDOG_PRETIMEOUT_NMI 2 79 #define WDOG_PRETIMEOUT_MSG_INT 3 80 81 /* Operations that can be performed on a pretimout. */ 82 #define WDOG_PREOP_NONE 0 83 #define WDOG_PREOP_PANIC 1 84 #define WDOG_PREOP_GIVE_DATA 2 /* Cause data to be available to 85 read. Doesn't work in NMI 86 mode. */ 87 88 /* Actions to perform on a full timeout. */ 89 #define WDOG_SET_TIMEOUT_ACT(byte, use) \ 90 byte = ((byte) & 0xf8) | ((use) & 0x7) 91 #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) 92 #define WDOG_TIMEOUT_NONE 0 93 #define WDOG_TIMEOUT_RESET 1 94 #define WDOG_TIMEOUT_POWER_DOWN 2 95 #define WDOG_TIMEOUT_POWER_CYCLE 3 96 97 /* Byte 3 of the get command, byte 4 of the get response is the 98 pre-timeout in seconds. */ 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 /* Setting/getting the watchdog timer value. This is for bytes 5 and 108 6 (the timeout time) of the set command, and bytes 6 and 7 (the 109 timeout time) and 8 and 9 (the current countdown value) of the 110 response. The timeout value is given in seconds (in the command it 111 is 100ms intervals). */ 112 #define WDOG_SET_TIMEOUT(byte1, byte2, val) \ 113 (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8) 114 #define WDOG_GET_TIMEOUT(byte1, byte2) \ 115 (((byte1) | ((byte2) << 8)) / 10) 116 117 #define IPMI_WDOG_RESET_TIMER 0x22 118 #define IPMI_WDOG_SET_TIMER 0x24 119 #define IPMI_WDOG_GET_TIMER 0x25 120 121 /* These are here until the real ones get into the watchdog.h interface. */ 122 #ifndef WDIOC_GETTIMEOUT 123 #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int) 124 #endif 125 #ifndef WDIOC_SET_PRETIMEOUT 126 #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int) 127 #endif 128 #ifndef WDIOC_GET_PRETIMEOUT 129 #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int) 130 #endif 131 132 static int nowayout = WATCHDOG_NOWAYOUT; 133 134 static ipmi_user_t watchdog_user = NULL; 135 136 /* Default the timeout to 10 seconds. */ 137 static int timeout = 10; 138 139 /* The pre-timeout is disabled by default. */ 140 static int pretimeout = 0; 141 142 /* Default action is to reset the board on a timeout. */ 143 static unsigned char action_val = WDOG_TIMEOUT_RESET; 144 145 static char action[16] = "reset"; 146 147 static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; 148 149 static char preaction[16] = "pre_none"; 150 151 static unsigned char preop_val = WDOG_PREOP_NONE; 152 153 static char preop[16] = "preop_none"; 154 static DEFINE_SPINLOCK(ipmi_read_lock); 155 static char data_to_read = 0; 156 static DECLARE_WAIT_QUEUE_HEAD(read_q); 157 static struct fasync_struct *fasync_q = NULL; 158 static char pretimeout_since_last_heartbeat = 0; 159 static char expect_close; 160 161 /* If true, the driver will start running as soon as it is configured 162 and ready. */ 163 static int start_now = 0; 164 165 module_param(timeout, int, 0); 166 MODULE_PARM_DESC(timeout, "Timeout value in seconds."); 167 module_param(pretimeout, int, 0); 168 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); 169 module_param_string(action, action, sizeof(action), 0); 170 MODULE_PARM_DESC(action, "Timeout action. One of: " 171 "reset, none, power_cycle, power_off."); 172 module_param_string(preaction, preaction, sizeof(preaction), 0); 173 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " 174 "pre_none, pre_smi, pre_nmi, pre_int."); 175 module_param_string(preop, preop, sizeof(preop), 0); 176 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " 177 "preop_none, preop_panic, preop_give_data."); 178 module_param(start_now, int, 0); 179 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" 180 "soon as the driver is loaded."); 181 module_param(nowayout, int, 0); 182 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default=CONFIG_WATCHDOG_NOWAYOUT)"); 183 184 /* Default state of the timer. */ 185 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 186 187 /* If shutting down via IPMI, we ignore the heartbeat. */ 188 static int ipmi_ignore_heartbeat = 0; 189 190 /* Is someone using the watchdog? Only one user is allowed. */ 191 static unsigned long ipmi_wdog_open = 0; 192 193 /* If set to 1, the heartbeat command will set the state to reset and 194 start the timer. The timer doesn't normally run when the driver is 195 first opened until the heartbeat is set the first time, this 196 variable is used to accomplish this. */ 197 static int ipmi_start_timer_on_heartbeat = 0; 198 199 /* IPMI version of the BMC. */ 200 static unsigned char ipmi_version_major; 201 static unsigned char ipmi_version_minor; 202 203 204 static int ipmi_heartbeat(void); 205 static void panic_halt_ipmi_heartbeat(void); 206 207 208 /* We use a semaphore to make sure that only one thing can send a set 209 timeout at one time, because we only have one copy of the data. 210 The semaphore is claimed when the set_timeout is sent and freed 211 when both messages are free. */ 212 static atomic_t set_timeout_tofree = ATOMIC_INIT(0); 213 static DECLARE_MUTEX(set_timeout_lock); 214 static void set_timeout_free_smi(struct ipmi_smi_msg *msg) 215 { 216 if (atomic_dec_and_test(&set_timeout_tofree)) 217 up(&set_timeout_lock); 218 } 219 static void set_timeout_free_recv(struct ipmi_recv_msg *msg) 220 { 221 if (atomic_dec_and_test(&set_timeout_tofree)) 222 up(&set_timeout_lock); 223 } 224 static struct ipmi_smi_msg set_timeout_smi_msg = 225 { 226 .done = set_timeout_free_smi 227 }; 228 static struct ipmi_recv_msg set_timeout_recv_msg = 229 { 230 .done = set_timeout_free_recv 231 }; 232 233 static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, 234 struct ipmi_recv_msg *recv_msg, 235 int *send_heartbeat_now) 236 { 237 struct kernel_ipmi_msg msg; 238 unsigned char data[6]; 239 int rv; 240 struct ipmi_system_interface_addr addr; 241 int hbnow = 0; 242 243 244 data[0] = 0; 245 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); 246 247 if ((ipmi_version_major > 1) 248 || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) 249 { 250 /* This is an IPMI 1.5-only feature. */ 251 data[0] |= WDOG_DONT_STOP_ON_SET; 252 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { 253 /* In ipmi 1.0, setting the timer stops the watchdog, we 254 need to start it back up again. */ 255 hbnow = 1; 256 } 257 258 data[1] = 0; 259 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); 260 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { 261 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); 262 data[2] = pretimeout; 263 } else { 264 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); 265 data[2] = 0; /* No pretimeout. */ 266 } 267 data[3] = 0; 268 WDOG_SET_TIMEOUT(data[4], data[5], timeout); 269 270 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 271 addr.channel = IPMI_BMC_CHANNEL; 272 addr.lun = 0; 273 274 msg.netfn = 0x06; 275 msg.cmd = IPMI_WDOG_SET_TIMER; 276 msg.data = data; 277 msg.data_len = sizeof(data); 278 rv = ipmi_request_supply_msgs(watchdog_user, 279 (struct ipmi_addr *) &addr, 280 0, 281 &msg, 282 NULL, 283 smi_msg, 284 recv_msg, 285 1); 286 if (rv) { 287 printk(KERN_WARNING PFX "set timeout error: %d\n", 288 rv); 289 } 290 291 if (send_heartbeat_now) 292 *send_heartbeat_now = hbnow; 293 294 return rv; 295 } 296 297 /* Parameters to ipmi_set_timeout */ 298 #define IPMI_SET_TIMEOUT_NO_HB 0 299 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 300 #define IPMI_SET_TIMEOUT_FORCE_HB 2 301 302 static int ipmi_set_timeout(int do_heartbeat) 303 { 304 int send_heartbeat_now; 305 int rv; 306 307 308 /* We can only send one of these at a time. */ 309 down(&set_timeout_lock); 310 311 atomic_set(&set_timeout_tofree, 2); 312 313 rv = i_ipmi_set_timeout(&set_timeout_smi_msg, 314 &set_timeout_recv_msg, 315 &send_heartbeat_now); 316 if (rv) { 317 up(&set_timeout_lock); 318 } else { 319 if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) 320 || ((send_heartbeat_now) 321 && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) 322 { 323 rv = ipmi_heartbeat(); 324 } 325 } 326 327 return rv; 328 } 329 330 static void dummy_smi_free(struct ipmi_smi_msg *msg) 331 { 332 } 333 static void dummy_recv_free(struct ipmi_recv_msg *msg) 334 { 335 } 336 static struct ipmi_smi_msg panic_halt_smi_msg = 337 { 338 .done = dummy_smi_free 339 }; 340 static struct ipmi_recv_msg panic_halt_recv_msg = 341 { 342 .done = dummy_recv_free 343 }; 344 345 /* Special call, doesn't claim any locks. This is only to be called 346 at panic or halt time, in run-to-completion mode, when the caller 347 is the only CPU and the only thing that will be going is these IPMI 348 calls. */ 349 static void panic_halt_ipmi_set_timeout(void) 350 { 351 int send_heartbeat_now; 352 int rv; 353 354 rv = i_ipmi_set_timeout(&panic_halt_smi_msg, 355 &panic_halt_recv_msg, 356 &send_heartbeat_now); 357 if (!rv) { 358 if (send_heartbeat_now) 359 panic_halt_ipmi_heartbeat(); 360 } 361 } 362 363 /* We use a semaphore to make sure that only one thing can send a 364 heartbeat at one time, because we only have one copy of the data. 365 The semaphore is claimed when the set_timeout is sent and freed 366 when both messages are free. */ 367 static atomic_t heartbeat_tofree = ATOMIC_INIT(0); 368 static DECLARE_MUTEX(heartbeat_lock); 369 static DECLARE_MUTEX_LOCKED(heartbeat_wait_lock); 370 static void heartbeat_free_smi(struct ipmi_smi_msg *msg) 371 { 372 if (atomic_dec_and_test(&heartbeat_tofree)) 373 up(&heartbeat_wait_lock); 374 } 375 static void heartbeat_free_recv(struct ipmi_recv_msg *msg) 376 { 377 if (atomic_dec_and_test(&heartbeat_tofree)) 378 up(&heartbeat_wait_lock); 379 } 380 static struct ipmi_smi_msg heartbeat_smi_msg = 381 { 382 .done = heartbeat_free_smi 383 }; 384 static struct ipmi_recv_msg heartbeat_recv_msg = 385 { 386 .done = heartbeat_free_recv 387 }; 388 389 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = 390 { 391 .done = dummy_smi_free 392 }; 393 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = 394 { 395 .done = dummy_recv_free 396 }; 397 398 static int ipmi_heartbeat(void) 399 { 400 struct kernel_ipmi_msg msg; 401 int rv; 402 struct ipmi_system_interface_addr addr; 403 404 if (ipmi_ignore_heartbeat) { 405 return 0; 406 } 407 408 if (ipmi_start_timer_on_heartbeat) { 409 ipmi_start_timer_on_heartbeat = 0; 410 ipmi_watchdog_state = action_val; 411 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 412 } else if (pretimeout_since_last_heartbeat) { 413 /* A pretimeout occurred, make sure we set the timeout. 414 We don't want to set the action, though, we want to 415 leave that alone (thus it can't be combined with the 416 above operation. */ 417 pretimeout_since_last_heartbeat = 0; 418 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 419 } 420 421 down(&heartbeat_lock); 422 423 atomic_set(&heartbeat_tofree, 2); 424 425 /* Don't reset the timer if we have the timer turned off, that 426 re-enables the watchdog. */ 427 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) { 428 up(&heartbeat_lock); 429 return 0; 430 } 431 432 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 433 addr.channel = IPMI_BMC_CHANNEL; 434 addr.lun = 0; 435 436 msg.netfn = 0x06; 437 msg.cmd = IPMI_WDOG_RESET_TIMER; 438 msg.data = NULL; 439 msg.data_len = 0; 440 rv = ipmi_request_supply_msgs(watchdog_user, 441 (struct ipmi_addr *) &addr, 442 0, 443 &msg, 444 NULL, 445 &heartbeat_smi_msg, 446 &heartbeat_recv_msg, 447 1); 448 if (rv) { 449 up(&heartbeat_lock); 450 printk(KERN_WARNING PFX "heartbeat failure: %d\n", 451 rv); 452 return rv; 453 } 454 455 /* Wait for the heartbeat to be sent. */ 456 down(&heartbeat_wait_lock); 457 458 if (heartbeat_recv_msg.msg.data[0] != 0) { 459 /* Got an error in the heartbeat response. It was already 460 reported in ipmi_wdog_msg_handler, but we should return 461 an error here. */ 462 rv = -EINVAL; 463 } 464 465 up(&heartbeat_lock); 466 467 return rv; 468 } 469 470 static void panic_halt_ipmi_heartbeat(void) 471 { 472 struct kernel_ipmi_msg msg; 473 struct ipmi_system_interface_addr addr; 474 475 476 /* Don't reset the timer if we have the timer turned off, that 477 re-enables the watchdog. */ 478 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) 479 return; 480 481 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; 482 addr.channel = IPMI_BMC_CHANNEL; 483 addr.lun = 0; 484 485 msg.netfn = 0x06; 486 msg.cmd = IPMI_WDOG_RESET_TIMER; 487 msg.data = NULL; 488 msg.data_len = 0; 489 ipmi_request_supply_msgs(watchdog_user, 490 (struct ipmi_addr *) &addr, 491 0, 492 &msg, 493 NULL, 494 &panic_halt_heartbeat_smi_msg, 495 &panic_halt_heartbeat_recv_msg, 496 1); 497 } 498 499 static struct watchdog_info ident= 500 { 501 .options = 0, /* WDIOF_SETTIMEOUT, */ 502 .firmware_version = 1, 503 .identity = "IPMI" 504 }; 505 506 static int ipmi_ioctl(struct inode *inode, struct file *file, 507 unsigned int cmd, unsigned long arg) 508 { 509 void __user *argp = (void __user *)arg; 510 int i; 511 int val; 512 513 switch(cmd) { 514 case WDIOC_GETSUPPORT: 515 i = copy_to_user(argp, &ident, sizeof(ident)); 516 return i ? -EFAULT : 0; 517 518 case WDIOC_SETTIMEOUT: 519 i = copy_from_user(&val, argp, sizeof(int)); 520 if (i) 521 return -EFAULT; 522 timeout = val; 523 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 524 525 case WDIOC_GETTIMEOUT: 526 i = copy_to_user(argp, &timeout, sizeof(timeout)); 527 if (i) 528 return -EFAULT; 529 return 0; 530 531 case WDIOC_SET_PRETIMEOUT: 532 i = copy_from_user(&val, argp, sizeof(int)); 533 if (i) 534 return -EFAULT; 535 pretimeout = val; 536 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); 537 538 case WDIOC_GET_PRETIMEOUT: 539 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); 540 if (i) 541 return -EFAULT; 542 return 0; 543 544 case WDIOC_KEEPALIVE: 545 return ipmi_heartbeat(); 546 547 case WDIOC_SETOPTIONS: 548 i = copy_from_user(&val, argp, sizeof(int)); 549 if (i) 550 return -EFAULT; 551 if (val & WDIOS_DISABLECARD) 552 { 553 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 554 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 555 ipmi_start_timer_on_heartbeat = 0; 556 } 557 558 if (val & WDIOS_ENABLECARD) 559 { 560 ipmi_watchdog_state = action_val; 561 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 562 } 563 return 0; 564 565 case WDIOC_GETSTATUS: 566 val = 0; 567 i = copy_to_user(argp, &val, sizeof(val)); 568 if (i) 569 return -EFAULT; 570 return 0; 571 572 default: 573 return -ENOIOCTLCMD; 574 } 575 } 576 577 static ssize_t ipmi_write(struct file *file, 578 const char __user *buf, 579 size_t len, 580 loff_t *ppos) 581 { 582 int rv; 583 584 if (len) { 585 if (!nowayout) { 586 size_t i; 587 588 /* In case it was set long ago */ 589 expect_close = 0; 590 591 for (i = 0; i != len; i++) { 592 char c; 593 594 if (get_user(c, buf + i)) 595 return -EFAULT; 596 if (c == 'V') 597 expect_close = 42; 598 } 599 } 600 rv = ipmi_heartbeat(); 601 if (rv) 602 return rv; 603 return 1; 604 } 605 return 0; 606 } 607 608 static ssize_t ipmi_read(struct file *file, 609 char __user *buf, 610 size_t count, 611 loff_t *ppos) 612 { 613 int rv = 0; 614 wait_queue_t wait; 615 616 if (count <= 0) 617 return 0; 618 619 /* Reading returns if the pretimeout has gone off, and it only does 620 it once per pretimeout. */ 621 spin_lock(&ipmi_read_lock); 622 if (!data_to_read) { 623 if (file->f_flags & O_NONBLOCK) { 624 rv = -EAGAIN; 625 goto out; 626 } 627 628 init_waitqueue_entry(&wait, current); 629 add_wait_queue(&read_q, &wait); 630 while (!data_to_read) { 631 set_current_state(TASK_INTERRUPTIBLE); 632 spin_unlock(&ipmi_read_lock); 633 schedule(); 634 spin_lock(&ipmi_read_lock); 635 } 636 remove_wait_queue(&read_q, &wait); 637 638 if (signal_pending(current)) { 639 rv = -ERESTARTSYS; 640 goto out; 641 } 642 } 643 data_to_read = 0; 644 645 out: 646 spin_unlock(&ipmi_read_lock); 647 648 if (rv == 0) { 649 if (copy_to_user(buf, &data_to_read, 1)) 650 rv = -EFAULT; 651 else 652 rv = 1; 653 } 654 655 return rv; 656 } 657 658 static int ipmi_open(struct inode *ino, struct file *filep) 659 { 660 switch (iminor(ino)) { 661 case WATCHDOG_MINOR: 662 if (test_and_set_bit(0, &ipmi_wdog_open)) 663 return -EBUSY; 664 665 /* Don't start the timer now, let it start on the 666 first heartbeat. */ 667 ipmi_start_timer_on_heartbeat = 1; 668 return nonseekable_open(ino, filep); 669 670 default: 671 return (-ENODEV); 672 } 673 } 674 675 static unsigned int ipmi_poll(struct file *file, poll_table *wait) 676 { 677 unsigned int mask = 0; 678 679 poll_wait(file, &read_q, wait); 680 681 spin_lock(&ipmi_read_lock); 682 if (data_to_read) 683 mask |= (POLLIN | POLLRDNORM); 684 spin_unlock(&ipmi_read_lock); 685 686 return mask; 687 } 688 689 static int ipmi_fasync(int fd, struct file *file, int on) 690 { 691 int result; 692 693 result = fasync_helper(fd, file, on, &fasync_q); 694 695 return (result); 696 } 697 698 static int ipmi_close(struct inode *ino, struct file *filep) 699 { 700 if (iminor(ino)==WATCHDOG_MINOR) 701 { 702 if (expect_close == 42) { 703 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 704 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); 705 } else { 706 printk(KERN_CRIT PFX "Unexpected close, not stopping watchdog!\n"); 707 ipmi_heartbeat(); 708 } 709 clear_bit(0, &ipmi_wdog_open); 710 } 711 712 ipmi_fasync (-1, filep, 0); 713 expect_close = 0; 714 715 return 0; 716 } 717 718 static struct file_operations ipmi_wdog_fops = { 719 .owner = THIS_MODULE, 720 .read = ipmi_read, 721 .poll = ipmi_poll, 722 .write = ipmi_write, 723 .ioctl = ipmi_ioctl, 724 .open = ipmi_open, 725 .release = ipmi_close, 726 .fasync = ipmi_fasync, 727 }; 728 729 static struct miscdevice ipmi_wdog_miscdev = { 730 .minor = WATCHDOG_MINOR, 731 .name = "watchdog", 732 .fops = &ipmi_wdog_fops 733 }; 734 735 static DECLARE_RWSEM(register_sem); 736 737 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, 738 void *handler_data) 739 { 740 if (msg->msg.data[0] != 0) { 741 printk(KERN_ERR PFX "response: Error %x on cmd %x\n", 742 msg->msg.data[0], 743 msg->msg.cmd); 744 } 745 746 ipmi_free_recv_msg(msg); 747 } 748 749 static void ipmi_wdog_pretimeout_handler(void *handler_data) 750 { 751 if (preaction_val != WDOG_PRETIMEOUT_NONE) { 752 if (preop_val == WDOG_PREOP_PANIC) 753 panic("Watchdog pre-timeout"); 754 else if (preop_val == WDOG_PREOP_GIVE_DATA) { 755 spin_lock(&ipmi_read_lock); 756 data_to_read = 1; 757 wake_up_interruptible(&read_q); 758 kill_fasync(&fasync_q, SIGIO, POLL_IN); 759 760 spin_unlock(&ipmi_read_lock); 761 } 762 } 763 764 /* On some machines, the heartbeat will give 765 an error and not work unless we re-enable 766 the timer. So do so. */ 767 pretimeout_since_last_heartbeat = 1; 768 } 769 770 static struct ipmi_user_hndl ipmi_hndlrs = 771 { 772 .ipmi_recv_hndl = ipmi_wdog_msg_handler, 773 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler 774 }; 775 776 static void ipmi_register_watchdog(int ipmi_intf) 777 { 778 int rv = -EBUSY; 779 780 down_write(®ister_sem); 781 if (watchdog_user) 782 goto out; 783 784 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); 785 if (rv < 0) { 786 printk(KERN_CRIT PFX "Unable to register with ipmi\n"); 787 goto out; 788 } 789 790 ipmi_get_version(watchdog_user, 791 &ipmi_version_major, 792 &ipmi_version_minor); 793 794 rv = misc_register(&ipmi_wdog_miscdev); 795 if (rv < 0) { 796 ipmi_destroy_user(watchdog_user); 797 watchdog_user = NULL; 798 printk(KERN_CRIT PFX "Unable to register misc device\n"); 799 } 800 801 out: 802 up_write(®ister_sem); 803 804 if ((start_now) && (rv == 0)) { 805 /* Run from startup, so start the timer now. */ 806 start_now = 0; /* Disable this function after first startup. */ 807 ipmi_watchdog_state = action_val; 808 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); 809 printk(KERN_INFO PFX "Starting now!\n"); 810 } 811 } 812 813 #ifdef HAVE_NMI_HANDLER 814 static int 815 ipmi_nmi(void *dev_id, struct pt_regs *regs, int cpu, int handled) 816 { 817 /* If we are not expecting a timeout, ignore it. */ 818 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) 819 return NOTIFY_DONE; 820 821 /* If no one else handled the NMI, we assume it was the IPMI 822 watchdog. */ 823 if ((!handled) && (preop_val == WDOG_PREOP_PANIC)) { 824 /* On some machines, the heartbeat will give 825 an error and not work unless we re-enable 826 the timer. So do so. */ 827 pretimeout_since_last_heartbeat = 1; 828 panic(PFX "pre-timeout"); 829 } 830 831 return NOTIFY_DONE; 832 } 833 834 static struct nmi_handler ipmi_nmi_handler = 835 { 836 .link = LIST_HEAD_INIT(ipmi_nmi_handler.link), 837 .dev_name = "ipmi_watchdog", 838 .dev_id = NULL, 839 .handler = ipmi_nmi, 840 .priority = 0, /* Call us last. */ 841 }; 842 #endif 843 844 static int wdog_reboot_handler(struct notifier_block *this, 845 unsigned long code, 846 void *unused) 847 { 848 static int reboot_event_handled = 0; 849 850 if ((watchdog_user) && (!reboot_event_handled)) { 851 /* Make sure we only do this once. */ 852 reboot_event_handled = 1; 853 854 if (code == SYS_DOWN || code == SYS_HALT) { 855 /* Disable the WDT if we are shutting down. */ 856 ipmi_watchdog_state = WDOG_TIMEOUT_NONE; 857 panic_halt_ipmi_set_timeout(); 858 } else { 859 /* Set a long timer to let the reboot happens, but 860 reboot if it hangs. */ 861 timeout = 120; 862 pretimeout = 0; 863 ipmi_watchdog_state = WDOG_TIMEOUT_RESET; 864 panic_halt_ipmi_set_timeout(); 865 } 866 } 867 return NOTIFY_OK; 868 } 869 870 static struct notifier_block wdog_reboot_notifier = { 871 .notifier_call = wdog_reboot_handler, 872 .next = NULL, 873 .priority = 0 874 }; 875 876 static int wdog_panic_handler(struct notifier_block *this, 877 unsigned long event, 878 void *unused) 879 { 880 static int panic_event_handled = 0; 881 882 /* On a panic, if we have a panic timeout, make sure that the thing 883 reboots, even if it hangs during that panic. */ 884 if (watchdog_user && !panic_event_handled) { 885 /* Make sure the panic doesn't hang, and make sure we 886 do this only once. */ 887 panic_event_handled = 1; 888 889 timeout = 255; 890 pretimeout = 0; 891 ipmi_watchdog_state = WDOG_TIMEOUT_RESET; 892 panic_halt_ipmi_set_timeout(); 893 } 894 895 return NOTIFY_OK; 896 } 897 898 static struct notifier_block wdog_panic_notifier = { 899 .notifier_call = wdog_panic_handler, 900 .next = NULL, 901 .priority = 150 /* priority: INT_MAX >= x >= 0 */ 902 }; 903 904 905 static void ipmi_new_smi(int if_num) 906 { 907 ipmi_register_watchdog(if_num); 908 } 909 910 static void ipmi_smi_gone(int if_num) 911 { 912 /* This can never be called, because once the watchdog is 913 registered, the interface can't go away until the watchdog 914 is unregistered. */ 915 } 916 917 static struct ipmi_smi_watcher smi_watcher = 918 { 919 .owner = THIS_MODULE, 920 .new_smi = ipmi_new_smi, 921 .smi_gone = ipmi_smi_gone 922 }; 923 924 static int __init ipmi_wdog_init(void) 925 { 926 int rv; 927 928 if (strcmp(action, "reset") == 0) { 929 action_val = WDOG_TIMEOUT_RESET; 930 } else if (strcmp(action, "none") == 0) { 931 action_val = WDOG_TIMEOUT_NONE; 932 } else if (strcmp(action, "power_cycle") == 0) { 933 action_val = WDOG_TIMEOUT_POWER_CYCLE; 934 } else if (strcmp(action, "power_off") == 0) { 935 action_val = WDOG_TIMEOUT_POWER_DOWN; 936 } else { 937 action_val = WDOG_TIMEOUT_RESET; 938 printk(KERN_INFO PFX "Unknown action '%s', defaulting to" 939 " reset\n", action); 940 } 941 942 if (strcmp(preaction, "pre_none") == 0) { 943 preaction_val = WDOG_PRETIMEOUT_NONE; 944 } else if (strcmp(preaction, "pre_smi") == 0) { 945 preaction_val = WDOG_PRETIMEOUT_SMI; 946 #ifdef HAVE_NMI_HANDLER 947 } else if (strcmp(preaction, "pre_nmi") == 0) { 948 preaction_val = WDOG_PRETIMEOUT_NMI; 949 #endif 950 } else if (strcmp(preaction, "pre_int") == 0) { 951 preaction_val = WDOG_PRETIMEOUT_MSG_INT; 952 } else { 953 preaction_val = WDOG_PRETIMEOUT_NONE; 954 printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to" 955 " none\n", preaction); 956 } 957 958 if (strcmp(preop, "preop_none") == 0) { 959 preop_val = WDOG_PREOP_NONE; 960 } else if (strcmp(preop, "preop_panic") == 0) { 961 preop_val = WDOG_PREOP_PANIC; 962 } else if (strcmp(preop, "preop_give_data") == 0) { 963 preop_val = WDOG_PREOP_GIVE_DATA; 964 } else { 965 preop_val = WDOG_PREOP_NONE; 966 printk(KERN_INFO PFX "Unknown preop '%s', defaulting to" 967 " none\n", preop); 968 } 969 970 #ifdef HAVE_NMI_HANDLER 971 if (preaction_val == WDOG_PRETIMEOUT_NMI) { 972 if (preop_val == WDOG_PREOP_GIVE_DATA) { 973 printk(KERN_WARNING PFX "Pretimeout op is to give data" 974 " but NMI pretimeout is enabled, setting" 975 " pretimeout op to none\n"); 976 preop_val = WDOG_PREOP_NONE; 977 } 978 #ifdef CONFIG_X86_LOCAL_APIC 979 if (nmi_watchdog == NMI_IO_APIC) { 980 printk(KERN_WARNING PFX "nmi_watchdog is set to IO APIC" 981 " mode (value is %d), that is incompatible" 982 " with using NMI in the IPMI watchdog." 983 " Disabling IPMI nmi pretimeout.\n", 984 nmi_watchdog); 985 preaction_val = WDOG_PRETIMEOUT_NONE; 986 } else { 987 #endif 988 rv = request_nmi(&ipmi_nmi_handler); 989 if (rv) { 990 printk(KERN_WARNING PFX "Can't register nmi handler\n"); 991 return rv; 992 } 993 #ifdef CONFIG_X86_LOCAL_APIC 994 } 995 #endif 996 } 997 #endif 998 999 rv = ipmi_smi_watcher_register(&smi_watcher); 1000 if (rv) { 1001 #ifdef HAVE_NMI_HANDLER 1002 if (preaction_val == WDOG_PRETIMEOUT_NMI) 1003 release_nmi(&ipmi_nmi_handler); 1004 #endif 1005 printk(KERN_WARNING PFX "can't register smi watcher\n"); 1006 return rv; 1007 } 1008 1009 register_reboot_notifier(&wdog_reboot_notifier); 1010 notifier_chain_register(&panic_notifier_list, &wdog_panic_notifier); 1011 1012 printk(KERN_INFO PFX "driver initialized\n"); 1013 1014 return 0; 1015 } 1016 1017 static __exit void ipmi_unregister_watchdog(void) 1018 { 1019 int rv; 1020 1021 down_write(®ister_sem); 1022 1023 #ifdef HAVE_NMI_HANDLER 1024 if (preaction_val == WDOG_PRETIMEOUT_NMI) 1025 release_nmi(&ipmi_nmi_handler); 1026 #endif 1027 1028 notifier_chain_unregister(&panic_notifier_list, &wdog_panic_notifier); 1029 unregister_reboot_notifier(&wdog_reboot_notifier); 1030 1031 if (! watchdog_user) 1032 goto out; 1033 1034 /* Make sure no one can call us any more. */ 1035 misc_deregister(&ipmi_wdog_miscdev); 1036 1037 /* Wait to make sure the message makes it out. The lower layer has 1038 pointers to our buffers, we want to make sure they are done before 1039 we release our memory. */ 1040 while (atomic_read(&set_timeout_tofree)) 1041 schedule_timeout_uninterruptible(1); 1042 1043 /* Disconnect from IPMI. */ 1044 rv = ipmi_destroy_user(watchdog_user); 1045 if (rv) { 1046 printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n", 1047 rv); 1048 } 1049 watchdog_user = NULL; 1050 1051 out: 1052 up_write(®ister_sem); 1053 } 1054 1055 static void __exit ipmi_wdog_exit(void) 1056 { 1057 ipmi_smi_watcher_unregister(&smi_watcher); 1058 ipmi_unregister_watchdog(); 1059 } 1060 module_exit(ipmi_wdog_exit); 1061 module_init(ipmi_wdog_init); 1062 MODULE_LICENSE("GPL"); 1063 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); 1064 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); 1065