Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 50 | #include <linux/string.h> |
| 51 | #include <linux/ctype.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | #ifdef CONFIG_X86_LOCAL_APIC |
| 53 | #include <asm/apic.h> |
| 54 | #endif |
| 55 | |
| 56 | #define PFX "IPMI Watchdog: " |
| 57 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | /* |
| 59 | * The IPMI command/response information for the watchdog timer. |
| 60 | */ |
| 61 | |
| 62 | /* values for byte 1 of the set command, byte 2 of the get response. */ |
| 63 | #define WDOG_DONT_LOG (1 << 7) |
| 64 | #define WDOG_DONT_STOP_ON_SET (1 << 6) |
| 65 | #define WDOG_SET_TIMER_USE(byte, use) \ |
| 66 | byte = ((byte) & 0xf8) | ((use) & 0x7) |
| 67 | #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) |
| 68 | #define WDOG_TIMER_USE_BIOS_FRB2 1 |
| 69 | #define WDOG_TIMER_USE_BIOS_POST 2 |
| 70 | #define WDOG_TIMER_USE_OS_LOAD 3 |
| 71 | #define WDOG_TIMER_USE_SMS_OS 4 |
| 72 | #define WDOG_TIMER_USE_OEM 5 |
| 73 | |
| 74 | /* values for byte 2 of the set command, byte 3 of the get response. */ |
| 75 | #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ |
| 76 | byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) |
| 77 | #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) |
| 78 | #define WDOG_PRETIMEOUT_NONE 0 |
| 79 | #define WDOG_PRETIMEOUT_SMI 1 |
| 80 | #define WDOG_PRETIMEOUT_NMI 2 |
| 81 | #define WDOG_PRETIMEOUT_MSG_INT 3 |
| 82 | |
| 83 | /* Operations that can be performed on a pretimout. */ |
| 84 | #define WDOG_PREOP_NONE 0 |
| 85 | #define WDOG_PREOP_PANIC 1 |
| 86 | #define WDOG_PREOP_GIVE_DATA 2 /* Cause data to be available to |
| 87 | read. Doesn't work in NMI |
| 88 | mode. */ |
| 89 | |
| 90 | /* Actions to perform on a full timeout. */ |
| 91 | #define WDOG_SET_TIMEOUT_ACT(byte, use) \ |
| 92 | byte = ((byte) & 0xf8) | ((use) & 0x7) |
| 93 | #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) |
| 94 | #define WDOG_TIMEOUT_NONE 0 |
| 95 | #define WDOG_TIMEOUT_RESET 1 |
| 96 | #define WDOG_TIMEOUT_POWER_DOWN 2 |
| 97 | #define WDOG_TIMEOUT_POWER_CYCLE 3 |
| 98 | |
| 99 | /* Byte 3 of the get command, byte 4 of the get response is the |
| 100 | pre-timeout in seconds. */ |
| 101 | |
| 102 | /* Bits for setting byte 4 of the set command, byte 5 of the get response. */ |
| 103 | #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1) |
| 104 | #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2) |
| 105 | #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3) |
| 106 | #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4) |
| 107 | #define WDOG_EXPIRE_CLEAR_OEM (1 << 5) |
| 108 | |
| 109 | /* Setting/getting the watchdog timer value. This is for bytes 5 and |
| 110 | 6 (the timeout time) of the set command, and bytes 6 and 7 (the |
| 111 | timeout time) and 8 and 9 (the current countdown value) of the |
| 112 | response. The timeout value is given in seconds (in the command it |
| 113 | is 100ms intervals). */ |
| 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 | /* These are here until the real ones get into the watchdog.h interface. */ |
| 124 | #ifndef WDIOC_GETTIMEOUT |
| 125 | #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int) |
| 126 | #endif |
| 127 | #ifndef WDIOC_SET_PRETIMEOUT |
| 128 | #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int) |
| 129 | #endif |
| 130 | #ifndef WDIOC_GET_PRETIMEOUT |
| 131 | #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int) |
| 132 | #endif |
| 133 | |
Andrey Panin | 4bfdf37 | 2005-07-27 11:43:58 -0700 | [diff] [blame] | 134 | static int nowayout = WATCHDOG_NOWAYOUT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | |
| 136 | static ipmi_user_t watchdog_user = NULL; |
| 137 | |
| 138 | /* Default the timeout to 10 seconds. */ |
| 139 | static int timeout = 10; |
| 140 | |
| 141 | /* The pre-timeout is disabled by default. */ |
| 142 | static int pretimeout = 0; |
| 143 | |
| 144 | /* Default action is to reset the board on a timeout. */ |
| 145 | static unsigned char action_val = WDOG_TIMEOUT_RESET; |
| 146 | |
| 147 | static char action[16] = "reset"; |
| 148 | |
| 149 | static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; |
| 150 | |
| 151 | static char preaction[16] = "pre_none"; |
| 152 | |
| 153 | static unsigned char preop_val = WDOG_PREOP_NONE; |
| 154 | |
| 155 | static char preop[16] = "preop_none"; |
| 156 | static DEFINE_SPINLOCK(ipmi_read_lock); |
| 157 | static char data_to_read = 0; |
| 158 | static DECLARE_WAIT_QUEUE_HEAD(read_q); |
| 159 | static struct fasync_struct *fasync_q = NULL; |
| 160 | static char pretimeout_since_last_heartbeat = 0; |
| 161 | static char expect_close; |
| 162 | |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 163 | static DECLARE_RWSEM(register_sem); |
| 164 | |
| 165 | /* Parameters to ipmi_set_timeout */ |
| 166 | #define IPMI_SET_TIMEOUT_NO_HB 0 |
| 167 | #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 |
| 168 | #define IPMI_SET_TIMEOUT_FORCE_HB 2 |
| 169 | |
| 170 | static int ipmi_set_timeout(int do_heartbeat); |
| 171 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | /* If true, the driver will start running as soon as it is configured |
| 173 | and ready. */ |
| 174 | static int start_now = 0; |
| 175 | |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 176 | static int set_param_int(const char *val, 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 | down_read(®ister_sem); |
| 189 | *((int *)kp->arg) = l; |
| 190 | if (watchdog_user) |
| 191 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
| 192 | up_read(®ister_sem); |
| 193 | |
| 194 | return rv; |
| 195 | } |
| 196 | |
| 197 | static int get_param_int(char *buffer, struct kernel_param *kp) |
| 198 | { |
| 199 | return sprintf(buffer, "%i", *((int *)kp->arg)); |
| 200 | } |
| 201 | |
| 202 | typedef int (*action_fn)(const char *intval, char *outval); |
| 203 | |
| 204 | static int action_op(const char *inval, char *outval); |
| 205 | static int preaction_op(const char *inval, char *outval); |
| 206 | static int preop_op(const char *inval, char *outval); |
| 207 | static void check_parms(void); |
| 208 | |
| 209 | static int set_param_str(const char *val, struct kernel_param *kp) |
| 210 | { |
| 211 | action_fn fn = (action_fn) kp->arg; |
| 212 | int rv = 0; |
| 213 | const char *end; |
| 214 | char valcp[16]; |
| 215 | int len; |
| 216 | |
| 217 | /* Truncate leading and trailing spaces. */ |
| 218 | while (isspace(*val)) |
| 219 | val++; |
| 220 | end = val + strlen(val) - 1; |
| 221 | while ((end >= val) && isspace(*end)) |
| 222 | end--; |
| 223 | len = end - val + 1; |
| 224 | if (len > sizeof(valcp) - 1) |
| 225 | return -EINVAL; |
| 226 | memcpy(valcp, val, len); |
| 227 | valcp[len] = '\0'; |
| 228 | |
| 229 | down_read(®ister_sem); |
| 230 | rv = fn(valcp, NULL); |
| 231 | if (rv) |
| 232 | goto out_unlock; |
| 233 | |
| 234 | check_parms(); |
| 235 | if (watchdog_user) |
| 236 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
| 237 | |
| 238 | out_unlock: |
| 239 | up_read(®ister_sem); |
| 240 | return rv; |
| 241 | } |
| 242 | |
| 243 | static int get_param_str(char *buffer, struct kernel_param *kp) |
| 244 | { |
| 245 | action_fn fn = (action_fn) kp->arg; |
| 246 | int rv; |
| 247 | |
| 248 | rv = fn(NULL, buffer); |
| 249 | if (rv) |
| 250 | return rv; |
| 251 | return strlen(buffer); |
| 252 | } |
| 253 | |
| 254 | module_param_call(timeout, set_param_int, get_param_int, &timeout, 0644); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | MODULE_PARM_DESC(timeout, "Timeout value in seconds."); |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 256 | |
| 257 | module_param_call(pretimeout, set_param_int, get_param_int, &pretimeout, 0644); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 259 | |
| 260 | module_param_call(action, set_param_str, get_param_str, action_op, 0644); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 261 | MODULE_PARM_DESC(action, "Timeout action. One of: " |
| 262 | "reset, none, power_cycle, power_off."); |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 263 | |
| 264 | module_param_call(preaction, set_param_str, get_param_str, preaction_op, 0644); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " |
| 266 | "pre_none, pre_smi, pre_nmi, pre_int."); |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 267 | |
| 268 | module_param_call(preop, set_param_str, get_param_str, preop_op, 0644); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " |
| 270 | "preop_none, preop_panic, preop_give_data."); |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 271 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | module_param(start_now, int, 0); |
| 273 | MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" |
| 274 | "soon as the driver is loaded."); |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 275 | |
| 276 | module_param(nowayout, int, 0644); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default=CONFIG_WATCHDOG_NOWAYOUT)"); |
| 278 | |
| 279 | /* Default state of the timer. */ |
| 280 | static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
| 281 | |
| 282 | /* If shutting down via IPMI, we ignore the heartbeat. */ |
| 283 | static int ipmi_ignore_heartbeat = 0; |
| 284 | |
| 285 | /* Is someone using the watchdog? Only one user is allowed. */ |
| 286 | static unsigned long ipmi_wdog_open = 0; |
| 287 | |
| 288 | /* If set to 1, the heartbeat command will set the state to reset and |
| 289 | start the timer. The timer doesn't normally run when the driver is |
| 290 | first opened until the heartbeat is set the first time, this |
| 291 | variable is used to accomplish this. */ |
| 292 | static int ipmi_start_timer_on_heartbeat = 0; |
| 293 | |
| 294 | /* IPMI version of the BMC. */ |
| 295 | static unsigned char ipmi_version_major; |
| 296 | static unsigned char ipmi_version_minor; |
| 297 | |
| 298 | |
| 299 | static int ipmi_heartbeat(void); |
| 300 | static void panic_halt_ipmi_heartbeat(void); |
| 301 | |
| 302 | |
| 303 | /* We use a semaphore to make sure that only one thing can send a set |
| 304 | timeout at one time, because we only have one copy of the data. |
| 305 | The semaphore is claimed when the set_timeout is sent and freed |
| 306 | when both messages are free. */ |
| 307 | static atomic_t set_timeout_tofree = ATOMIC_INIT(0); |
| 308 | static DECLARE_MUTEX(set_timeout_lock); |
| 309 | static void set_timeout_free_smi(struct ipmi_smi_msg *msg) |
| 310 | { |
| 311 | if (atomic_dec_and_test(&set_timeout_tofree)) |
| 312 | up(&set_timeout_lock); |
| 313 | } |
| 314 | static void set_timeout_free_recv(struct ipmi_recv_msg *msg) |
| 315 | { |
| 316 | if (atomic_dec_and_test(&set_timeout_tofree)) |
| 317 | up(&set_timeout_lock); |
| 318 | } |
| 319 | static struct ipmi_smi_msg set_timeout_smi_msg = |
| 320 | { |
| 321 | .done = set_timeout_free_smi |
| 322 | }; |
| 323 | static struct ipmi_recv_msg set_timeout_recv_msg = |
| 324 | { |
| 325 | .done = set_timeout_free_recv |
| 326 | }; |
| 327 | |
| 328 | static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, |
| 329 | struct ipmi_recv_msg *recv_msg, |
| 330 | int *send_heartbeat_now) |
| 331 | { |
| 332 | struct kernel_ipmi_msg msg; |
| 333 | unsigned char data[6]; |
| 334 | int rv; |
| 335 | struct ipmi_system_interface_addr addr; |
| 336 | int hbnow = 0; |
| 337 | |
| 338 | |
| 339 | data[0] = 0; |
| 340 | WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); |
| 341 | |
| 342 | if ((ipmi_version_major > 1) |
| 343 | || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) |
| 344 | { |
| 345 | /* This is an IPMI 1.5-only feature. */ |
| 346 | data[0] |= WDOG_DONT_STOP_ON_SET; |
| 347 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
| 348 | /* In ipmi 1.0, setting the timer stops the watchdog, we |
| 349 | need to start it back up again. */ |
| 350 | hbnow = 1; |
| 351 | } |
| 352 | |
| 353 | data[1] = 0; |
| 354 | WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); |
Corey Minyard | 8f05ee9 | 2005-09-06 15:18:39 -0700 | [diff] [blame] | 355 | if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); |
| 357 | data[2] = pretimeout; |
| 358 | } else { |
| 359 | WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); |
| 360 | data[2] = 0; /* No pretimeout. */ |
| 361 | } |
| 362 | data[3] = 0; |
| 363 | WDOG_SET_TIMEOUT(data[4], data[5], timeout); |
| 364 | |
| 365 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; |
| 366 | addr.channel = IPMI_BMC_CHANNEL; |
| 367 | addr.lun = 0; |
| 368 | |
| 369 | msg.netfn = 0x06; |
| 370 | msg.cmd = IPMI_WDOG_SET_TIMER; |
| 371 | msg.data = data; |
| 372 | msg.data_len = sizeof(data); |
| 373 | rv = ipmi_request_supply_msgs(watchdog_user, |
| 374 | (struct ipmi_addr *) &addr, |
| 375 | 0, |
| 376 | &msg, |
| 377 | NULL, |
| 378 | smi_msg, |
| 379 | recv_msg, |
| 380 | 1); |
| 381 | if (rv) { |
| 382 | printk(KERN_WARNING PFX "set timeout error: %d\n", |
| 383 | rv); |
| 384 | } |
| 385 | |
| 386 | if (send_heartbeat_now) |
| 387 | *send_heartbeat_now = hbnow; |
| 388 | |
| 389 | return rv; |
| 390 | } |
| 391 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 392 | static int ipmi_set_timeout(int do_heartbeat) |
| 393 | { |
| 394 | int send_heartbeat_now; |
| 395 | int rv; |
| 396 | |
| 397 | |
| 398 | /* We can only send one of these at a time. */ |
| 399 | down(&set_timeout_lock); |
| 400 | |
| 401 | atomic_set(&set_timeout_tofree, 2); |
| 402 | |
| 403 | rv = i_ipmi_set_timeout(&set_timeout_smi_msg, |
| 404 | &set_timeout_recv_msg, |
| 405 | &send_heartbeat_now); |
| 406 | if (rv) { |
| 407 | up(&set_timeout_lock); |
| 408 | } else { |
| 409 | if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) |
| 410 | || ((send_heartbeat_now) |
| 411 | && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) |
| 412 | { |
| 413 | rv = ipmi_heartbeat(); |
| 414 | } |
| 415 | } |
| 416 | |
| 417 | return rv; |
| 418 | } |
| 419 | |
| 420 | static void dummy_smi_free(struct ipmi_smi_msg *msg) |
| 421 | { |
| 422 | } |
| 423 | static void dummy_recv_free(struct ipmi_recv_msg *msg) |
| 424 | { |
| 425 | } |
| 426 | static struct ipmi_smi_msg panic_halt_smi_msg = |
| 427 | { |
| 428 | .done = dummy_smi_free |
| 429 | }; |
| 430 | static struct ipmi_recv_msg panic_halt_recv_msg = |
| 431 | { |
| 432 | .done = dummy_recv_free |
| 433 | }; |
| 434 | |
| 435 | /* Special call, doesn't claim any locks. This is only to be called |
| 436 | at panic or halt time, in run-to-completion mode, when the caller |
| 437 | is the only CPU and the only thing that will be going is these IPMI |
| 438 | calls. */ |
| 439 | static void panic_halt_ipmi_set_timeout(void) |
| 440 | { |
| 441 | int send_heartbeat_now; |
| 442 | int rv; |
| 443 | |
| 444 | rv = i_ipmi_set_timeout(&panic_halt_smi_msg, |
| 445 | &panic_halt_recv_msg, |
| 446 | &send_heartbeat_now); |
| 447 | if (!rv) { |
| 448 | if (send_heartbeat_now) |
| 449 | panic_halt_ipmi_heartbeat(); |
| 450 | } |
| 451 | } |
| 452 | |
| 453 | /* We use a semaphore to make sure that only one thing can send a |
| 454 | heartbeat at one time, because we only have one copy of the data. |
| 455 | The semaphore is claimed when the set_timeout is sent and freed |
| 456 | when both messages are free. */ |
| 457 | static atomic_t heartbeat_tofree = ATOMIC_INIT(0); |
| 458 | static DECLARE_MUTEX(heartbeat_lock); |
| 459 | static DECLARE_MUTEX_LOCKED(heartbeat_wait_lock); |
| 460 | static void heartbeat_free_smi(struct ipmi_smi_msg *msg) |
| 461 | { |
| 462 | if (atomic_dec_and_test(&heartbeat_tofree)) |
| 463 | up(&heartbeat_wait_lock); |
| 464 | } |
| 465 | static void heartbeat_free_recv(struct ipmi_recv_msg *msg) |
| 466 | { |
| 467 | if (atomic_dec_and_test(&heartbeat_tofree)) |
| 468 | up(&heartbeat_wait_lock); |
| 469 | } |
| 470 | static struct ipmi_smi_msg heartbeat_smi_msg = |
| 471 | { |
| 472 | .done = heartbeat_free_smi |
| 473 | }; |
| 474 | static struct ipmi_recv_msg heartbeat_recv_msg = |
| 475 | { |
| 476 | .done = heartbeat_free_recv |
| 477 | }; |
| 478 | |
| 479 | static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = |
| 480 | { |
| 481 | .done = dummy_smi_free |
| 482 | }; |
| 483 | static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = |
| 484 | { |
| 485 | .done = dummy_recv_free |
| 486 | }; |
| 487 | |
| 488 | static int ipmi_heartbeat(void) |
| 489 | { |
| 490 | struct kernel_ipmi_msg msg; |
| 491 | int rv; |
| 492 | struct ipmi_system_interface_addr addr; |
| 493 | |
| 494 | if (ipmi_ignore_heartbeat) { |
| 495 | return 0; |
| 496 | } |
| 497 | |
| 498 | if (ipmi_start_timer_on_heartbeat) { |
| 499 | ipmi_start_timer_on_heartbeat = 0; |
| 500 | ipmi_watchdog_state = action_val; |
| 501 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
| 502 | } else if (pretimeout_since_last_heartbeat) { |
| 503 | /* A pretimeout occurred, make sure we set the timeout. |
| 504 | We don't want to set the action, though, we want to |
| 505 | leave that alone (thus it can't be combined with the |
| 506 | above operation. */ |
| 507 | pretimeout_since_last_heartbeat = 0; |
| 508 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
| 509 | } |
| 510 | |
| 511 | down(&heartbeat_lock); |
| 512 | |
| 513 | atomic_set(&heartbeat_tofree, 2); |
| 514 | |
| 515 | /* Don't reset the timer if we have the timer turned off, that |
| 516 | re-enables the watchdog. */ |
| 517 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) { |
| 518 | up(&heartbeat_lock); |
| 519 | return 0; |
| 520 | } |
| 521 | |
| 522 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; |
| 523 | addr.channel = IPMI_BMC_CHANNEL; |
| 524 | addr.lun = 0; |
| 525 | |
| 526 | msg.netfn = 0x06; |
| 527 | msg.cmd = IPMI_WDOG_RESET_TIMER; |
| 528 | msg.data = NULL; |
| 529 | msg.data_len = 0; |
| 530 | rv = ipmi_request_supply_msgs(watchdog_user, |
| 531 | (struct ipmi_addr *) &addr, |
| 532 | 0, |
| 533 | &msg, |
| 534 | NULL, |
| 535 | &heartbeat_smi_msg, |
| 536 | &heartbeat_recv_msg, |
| 537 | 1); |
| 538 | if (rv) { |
| 539 | up(&heartbeat_lock); |
| 540 | printk(KERN_WARNING PFX "heartbeat failure: %d\n", |
| 541 | rv); |
| 542 | return rv; |
| 543 | } |
| 544 | |
| 545 | /* Wait for the heartbeat to be sent. */ |
| 546 | down(&heartbeat_wait_lock); |
| 547 | |
| 548 | if (heartbeat_recv_msg.msg.data[0] != 0) { |
| 549 | /* Got an error in the heartbeat response. It was already |
| 550 | reported in ipmi_wdog_msg_handler, but we should return |
| 551 | an error here. */ |
| 552 | rv = -EINVAL; |
| 553 | } |
| 554 | |
| 555 | up(&heartbeat_lock); |
| 556 | |
| 557 | return rv; |
| 558 | } |
| 559 | |
| 560 | static void panic_halt_ipmi_heartbeat(void) |
| 561 | { |
| 562 | struct kernel_ipmi_msg msg; |
| 563 | struct ipmi_system_interface_addr addr; |
| 564 | |
| 565 | |
| 566 | /* Don't reset the timer if we have the timer turned off, that |
| 567 | re-enables the watchdog. */ |
| 568 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
| 569 | return; |
| 570 | |
| 571 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; |
| 572 | addr.channel = IPMI_BMC_CHANNEL; |
| 573 | addr.lun = 0; |
| 574 | |
| 575 | msg.netfn = 0x06; |
| 576 | msg.cmd = IPMI_WDOG_RESET_TIMER; |
| 577 | msg.data = NULL; |
| 578 | msg.data_len = 0; |
| 579 | ipmi_request_supply_msgs(watchdog_user, |
| 580 | (struct ipmi_addr *) &addr, |
| 581 | 0, |
| 582 | &msg, |
| 583 | NULL, |
| 584 | &panic_halt_heartbeat_smi_msg, |
| 585 | &panic_halt_heartbeat_recv_msg, |
| 586 | 1); |
| 587 | } |
| 588 | |
| 589 | static struct watchdog_info ident= |
| 590 | { |
| 591 | .options = 0, /* WDIOF_SETTIMEOUT, */ |
| 592 | .firmware_version = 1, |
| 593 | .identity = "IPMI" |
| 594 | }; |
| 595 | |
| 596 | static int ipmi_ioctl(struct inode *inode, struct file *file, |
| 597 | unsigned int cmd, unsigned long arg) |
| 598 | { |
| 599 | void __user *argp = (void __user *)arg; |
| 600 | int i; |
| 601 | int val; |
| 602 | |
| 603 | switch(cmd) { |
| 604 | case WDIOC_GETSUPPORT: |
| 605 | i = copy_to_user(argp, &ident, sizeof(ident)); |
| 606 | return i ? -EFAULT : 0; |
| 607 | |
| 608 | case WDIOC_SETTIMEOUT: |
| 609 | i = copy_from_user(&val, argp, sizeof(int)); |
| 610 | if (i) |
| 611 | return -EFAULT; |
| 612 | timeout = val; |
| 613 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
| 614 | |
| 615 | case WDIOC_GETTIMEOUT: |
| 616 | i = copy_to_user(argp, &timeout, sizeof(timeout)); |
| 617 | if (i) |
| 618 | return -EFAULT; |
| 619 | return 0; |
| 620 | |
| 621 | case WDIOC_SET_PRETIMEOUT: |
| 622 | i = copy_from_user(&val, argp, sizeof(int)); |
| 623 | if (i) |
| 624 | return -EFAULT; |
| 625 | pretimeout = val; |
| 626 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
| 627 | |
| 628 | case WDIOC_GET_PRETIMEOUT: |
| 629 | i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); |
| 630 | if (i) |
| 631 | return -EFAULT; |
| 632 | return 0; |
| 633 | |
| 634 | case WDIOC_KEEPALIVE: |
| 635 | return ipmi_heartbeat(); |
| 636 | |
| 637 | case WDIOC_SETOPTIONS: |
| 638 | i = copy_from_user(&val, argp, sizeof(int)); |
| 639 | if (i) |
| 640 | return -EFAULT; |
| 641 | if (val & WDIOS_DISABLECARD) |
| 642 | { |
| 643 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
| 644 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
| 645 | ipmi_start_timer_on_heartbeat = 0; |
| 646 | } |
| 647 | |
| 648 | if (val & WDIOS_ENABLECARD) |
| 649 | { |
| 650 | ipmi_watchdog_state = action_val; |
| 651 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
| 652 | } |
| 653 | return 0; |
| 654 | |
| 655 | case WDIOC_GETSTATUS: |
| 656 | val = 0; |
| 657 | i = copy_to_user(argp, &val, sizeof(val)); |
| 658 | if (i) |
| 659 | return -EFAULT; |
| 660 | return 0; |
| 661 | |
| 662 | default: |
| 663 | return -ENOIOCTLCMD; |
| 664 | } |
| 665 | } |
| 666 | |
| 667 | static ssize_t ipmi_write(struct file *file, |
| 668 | const char __user *buf, |
| 669 | size_t len, |
| 670 | loff_t *ppos) |
| 671 | { |
| 672 | int rv; |
| 673 | |
| 674 | if (len) { |
| 675 | if (!nowayout) { |
| 676 | size_t i; |
| 677 | |
| 678 | /* In case it was set long ago */ |
| 679 | expect_close = 0; |
| 680 | |
| 681 | for (i = 0; i != len; i++) { |
| 682 | char c; |
| 683 | |
| 684 | if (get_user(c, buf + i)) |
| 685 | return -EFAULT; |
| 686 | if (c == 'V') |
| 687 | expect_close = 42; |
| 688 | } |
| 689 | } |
| 690 | rv = ipmi_heartbeat(); |
| 691 | if (rv) |
| 692 | return rv; |
| 693 | return 1; |
| 694 | } |
| 695 | return 0; |
| 696 | } |
| 697 | |
| 698 | static ssize_t ipmi_read(struct file *file, |
| 699 | char __user *buf, |
| 700 | size_t count, |
| 701 | loff_t *ppos) |
| 702 | { |
| 703 | int rv = 0; |
| 704 | wait_queue_t wait; |
| 705 | |
| 706 | if (count <= 0) |
| 707 | return 0; |
| 708 | |
| 709 | /* Reading returns if the pretimeout has gone off, and it only does |
| 710 | it once per pretimeout. */ |
| 711 | spin_lock(&ipmi_read_lock); |
| 712 | if (!data_to_read) { |
| 713 | if (file->f_flags & O_NONBLOCK) { |
| 714 | rv = -EAGAIN; |
| 715 | goto out; |
| 716 | } |
| 717 | |
| 718 | init_waitqueue_entry(&wait, current); |
| 719 | add_wait_queue(&read_q, &wait); |
| 720 | while (!data_to_read) { |
| 721 | set_current_state(TASK_INTERRUPTIBLE); |
| 722 | spin_unlock(&ipmi_read_lock); |
| 723 | schedule(); |
| 724 | spin_lock(&ipmi_read_lock); |
| 725 | } |
| 726 | remove_wait_queue(&read_q, &wait); |
| 727 | |
| 728 | if (signal_pending(current)) { |
| 729 | rv = -ERESTARTSYS; |
| 730 | goto out; |
| 731 | } |
| 732 | } |
| 733 | data_to_read = 0; |
| 734 | |
| 735 | out: |
| 736 | spin_unlock(&ipmi_read_lock); |
| 737 | |
| 738 | if (rv == 0) { |
| 739 | if (copy_to_user(buf, &data_to_read, 1)) |
| 740 | rv = -EFAULT; |
| 741 | else |
| 742 | rv = 1; |
| 743 | } |
| 744 | |
| 745 | return rv; |
| 746 | } |
| 747 | |
| 748 | static int ipmi_open(struct inode *ino, struct file *filep) |
| 749 | { |
Corey Minyard | e8b3361 | 2005-09-06 15:18:45 -0700 | [diff] [blame] | 750 | switch (iminor(ino)) { |
| 751 | case WATCHDOG_MINOR: |
| 752 | if (test_and_set_bit(0, &ipmi_wdog_open)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 753 | return -EBUSY; |
| 754 | |
Corey Minyard | e8b3361 | 2005-09-06 15:18:45 -0700 | [diff] [blame] | 755 | /* Don't start the timer now, let it start on the |
| 756 | first heartbeat. */ |
| 757 | ipmi_start_timer_on_heartbeat = 1; |
| 758 | return nonseekable_open(ino, filep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | |
Corey Minyard | e8b3361 | 2005-09-06 15:18:45 -0700 | [diff] [blame] | 760 | default: |
| 761 | return (-ENODEV); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | } |
| 763 | } |
| 764 | |
| 765 | static unsigned int ipmi_poll(struct file *file, poll_table *wait) |
| 766 | { |
| 767 | unsigned int mask = 0; |
| 768 | |
| 769 | poll_wait(file, &read_q, wait); |
| 770 | |
| 771 | spin_lock(&ipmi_read_lock); |
| 772 | if (data_to_read) |
| 773 | mask |= (POLLIN | POLLRDNORM); |
| 774 | spin_unlock(&ipmi_read_lock); |
| 775 | |
| 776 | return mask; |
| 777 | } |
| 778 | |
| 779 | static int ipmi_fasync(int fd, struct file *file, int on) |
| 780 | { |
| 781 | int result; |
| 782 | |
| 783 | result = fasync_helper(fd, file, on, &fasync_q); |
| 784 | |
| 785 | return (result); |
| 786 | } |
| 787 | |
| 788 | static int ipmi_close(struct inode *ino, struct file *filep) |
| 789 | { |
| 790 | if (iminor(ino)==WATCHDOG_MINOR) |
| 791 | { |
| 792 | if (expect_close == 42) { |
| 793 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
| 794 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | } else { |
| 796 | printk(KERN_CRIT PFX "Unexpected close, not stopping watchdog!\n"); |
| 797 | ipmi_heartbeat(); |
| 798 | } |
Corey Minyard | ec26d79 | 2005-05-01 08:59:11 -0700 | [diff] [blame] | 799 | clear_bit(0, &ipmi_wdog_open); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | } |
| 801 | |
| 802 | ipmi_fasync (-1, filep, 0); |
| 803 | expect_close = 0; |
| 804 | |
| 805 | return 0; |
| 806 | } |
| 807 | |
| 808 | static struct file_operations ipmi_wdog_fops = { |
| 809 | .owner = THIS_MODULE, |
| 810 | .read = ipmi_read, |
| 811 | .poll = ipmi_poll, |
| 812 | .write = ipmi_write, |
| 813 | .ioctl = ipmi_ioctl, |
| 814 | .open = ipmi_open, |
| 815 | .release = ipmi_close, |
| 816 | .fasync = ipmi_fasync, |
| 817 | }; |
| 818 | |
| 819 | static struct miscdevice ipmi_wdog_miscdev = { |
| 820 | .minor = WATCHDOG_MINOR, |
| 821 | .name = "watchdog", |
| 822 | .fops = &ipmi_wdog_fops |
| 823 | }; |
| 824 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, |
| 826 | void *handler_data) |
| 827 | { |
| 828 | if (msg->msg.data[0] != 0) { |
| 829 | printk(KERN_ERR PFX "response: Error %x on cmd %x\n", |
| 830 | msg->msg.data[0], |
| 831 | msg->msg.cmd); |
| 832 | } |
| 833 | |
| 834 | ipmi_free_recv_msg(msg); |
| 835 | } |
| 836 | |
| 837 | static void ipmi_wdog_pretimeout_handler(void *handler_data) |
| 838 | { |
| 839 | if (preaction_val != WDOG_PRETIMEOUT_NONE) { |
| 840 | if (preop_val == WDOG_PREOP_PANIC) |
| 841 | panic("Watchdog pre-timeout"); |
| 842 | else if (preop_val == WDOG_PREOP_GIVE_DATA) { |
| 843 | spin_lock(&ipmi_read_lock); |
| 844 | data_to_read = 1; |
| 845 | wake_up_interruptible(&read_q); |
| 846 | kill_fasync(&fasync_q, SIGIO, POLL_IN); |
| 847 | |
| 848 | spin_unlock(&ipmi_read_lock); |
| 849 | } |
| 850 | } |
| 851 | |
| 852 | /* On some machines, the heartbeat will give |
| 853 | an error and not work unless we re-enable |
| 854 | the timer. So do so. */ |
| 855 | pretimeout_since_last_heartbeat = 1; |
| 856 | } |
| 857 | |
| 858 | static struct ipmi_user_hndl ipmi_hndlrs = |
| 859 | { |
| 860 | .ipmi_recv_hndl = ipmi_wdog_msg_handler, |
| 861 | .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler |
| 862 | }; |
| 863 | |
| 864 | static void ipmi_register_watchdog(int ipmi_intf) |
| 865 | { |
| 866 | int rv = -EBUSY; |
| 867 | |
| 868 | down_write(®ister_sem); |
| 869 | if (watchdog_user) |
| 870 | goto out; |
| 871 | |
| 872 | rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); |
| 873 | if (rv < 0) { |
| 874 | printk(KERN_CRIT PFX "Unable to register with ipmi\n"); |
| 875 | goto out; |
| 876 | } |
| 877 | |
| 878 | ipmi_get_version(watchdog_user, |
| 879 | &ipmi_version_major, |
| 880 | &ipmi_version_minor); |
| 881 | |
| 882 | rv = misc_register(&ipmi_wdog_miscdev); |
| 883 | if (rv < 0) { |
| 884 | ipmi_destroy_user(watchdog_user); |
| 885 | watchdog_user = NULL; |
| 886 | printk(KERN_CRIT PFX "Unable to register misc device\n"); |
| 887 | } |
| 888 | |
| 889 | out: |
| 890 | up_write(®ister_sem); |
| 891 | |
| 892 | if ((start_now) && (rv == 0)) { |
| 893 | /* Run from startup, so start the timer now. */ |
| 894 | start_now = 0; /* Disable this function after first startup. */ |
| 895 | ipmi_watchdog_state = action_val; |
| 896 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
| 897 | printk(KERN_INFO PFX "Starting now!\n"); |
| 898 | } |
| 899 | } |
| 900 | |
| 901 | #ifdef HAVE_NMI_HANDLER |
| 902 | static int |
| 903 | ipmi_nmi(void *dev_id, struct pt_regs *regs, int cpu, int handled) |
| 904 | { |
Corey Minyard | 8f05ee9 | 2005-09-06 15:18:39 -0700 | [diff] [blame] | 905 | /* If we are not expecting a timeout, ignore it. */ |
| 906 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
| 907 | return NOTIFY_DONE; |
| 908 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 909 | /* If no one else handled the NMI, we assume it was the IPMI |
| 910 | watchdog. */ |
Corey Minyard | 8f05ee9 | 2005-09-06 15:18:39 -0700 | [diff] [blame] | 911 | if ((!handled) && (preop_val == WDOG_PREOP_PANIC)) { |
| 912 | /* On some machines, the heartbeat will give |
| 913 | an error and not work unless we re-enable |
| 914 | the timer. So do so. */ |
| 915 | pretimeout_since_last_heartbeat = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | panic(PFX "pre-timeout"); |
Corey Minyard | 8f05ee9 | 2005-09-06 15:18:39 -0700 | [diff] [blame] | 917 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | |
| 919 | return NOTIFY_DONE; |
| 920 | } |
| 921 | |
| 922 | static struct nmi_handler ipmi_nmi_handler = |
| 923 | { |
| 924 | .link = LIST_HEAD_INIT(ipmi_nmi_handler.link), |
| 925 | .dev_name = "ipmi_watchdog", |
| 926 | .dev_id = NULL, |
| 927 | .handler = ipmi_nmi, |
| 928 | .priority = 0, /* Call us last. */ |
| 929 | }; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 930 | int nmi_handler_registered; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 931 | #endif |
| 932 | |
| 933 | static int wdog_reboot_handler(struct notifier_block *this, |
| 934 | unsigned long code, |
| 935 | void *unused) |
| 936 | { |
| 937 | static int reboot_event_handled = 0; |
| 938 | |
| 939 | if ((watchdog_user) && (!reboot_event_handled)) { |
| 940 | /* Make sure we only do this once. */ |
| 941 | reboot_event_handled = 1; |
| 942 | |
| 943 | if (code == SYS_DOWN || code == SYS_HALT) { |
| 944 | /* Disable the WDT if we are shutting down. */ |
| 945 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
| 946 | panic_halt_ipmi_set_timeout(); |
| 947 | } else { |
| 948 | /* Set a long timer to let the reboot happens, but |
| 949 | reboot if it hangs. */ |
| 950 | timeout = 120; |
| 951 | pretimeout = 0; |
| 952 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; |
| 953 | panic_halt_ipmi_set_timeout(); |
| 954 | } |
| 955 | } |
| 956 | return NOTIFY_OK; |
| 957 | } |
| 958 | |
| 959 | static struct notifier_block wdog_reboot_notifier = { |
| 960 | .notifier_call = wdog_reboot_handler, |
| 961 | .next = NULL, |
| 962 | .priority = 0 |
| 963 | }; |
| 964 | |
| 965 | static int wdog_panic_handler(struct notifier_block *this, |
| 966 | unsigned long event, |
| 967 | void *unused) |
| 968 | { |
| 969 | static int panic_event_handled = 0; |
| 970 | |
| 971 | /* On a panic, if we have a panic timeout, make sure that the thing |
| 972 | reboots, even if it hangs during that panic. */ |
| 973 | if (watchdog_user && !panic_event_handled) { |
| 974 | /* Make sure the panic doesn't hang, and make sure we |
| 975 | do this only once. */ |
| 976 | panic_event_handled = 1; |
| 977 | |
| 978 | timeout = 255; |
| 979 | pretimeout = 0; |
| 980 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; |
| 981 | panic_halt_ipmi_set_timeout(); |
| 982 | } |
| 983 | |
| 984 | return NOTIFY_OK; |
| 985 | } |
| 986 | |
| 987 | static struct notifier_block wdog_panic_notifier = { |
| 988 | .notifier_call = wdog_panic_handler, |
| 989 | .next = NULL, |
| 990 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ |
| 991 | }; |
| 992 | |
| 993 | |
| 994 | static void ipmi_new_smi(int if_num) |
| 995 | { |
| 996 | ipmi_register_watchdog(if_num); |
| 997 | } |
| 998 | |
| 999 | static void ipmi_smi_gone(int if_num) |
| 1000 | { |
| 1001 | /* This can never be called, because once the watchdog is |
| 1002 | registered, the interface can't go away until the watchdog |
| 1003 | is unregistered. */ |
| 1004 | } |
| 1005 | |
| 1006 | static struct ipmi_smi_watcher smi_watcher = |
| 1007 | { |
| 1008 | .owner = THIS_MODULE, |
| 1009 | .new_smi = ipmi_new_smi, |
| 1010 | .smi_gone = ipmi_smi_gone |
| 1011 | }; |
| 1012 | |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1013 | static int action_op(const char *inval, char *outval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1014 | { |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1015 | if (outval) |
| 1016 | strcpy(outval, action); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1018 | if (!inval) |
| 1019 | return 0; |
| 1020 | |
| 1021 | if (strcmp(inval, "reset") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | action_val = WDOG_TIMEOUT_RESET; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1023 | else if (strcmp(inval, "none") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | action_val = WDOG_TIMEOUT_NONE; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1025 | else if (strcmp(inval, "power_cycle") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | action_val = WDOG_TIMEOUT_POWER_CYCLE; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1027 | else if (strcmp(inval, "power_off") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1028 | action_val = WDOG_TIMEOUT_POWER_DOWN; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1029 | else |
| 1030 | return -EINVAL; |
| 1031 | strcpy(action, inval); |
| 1032 | return 0; |
| 1033 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1035 | static int preaction_op(const char *inval, char *outval) |
| 1036 | { |
| 1037 | if (outval) |
| 1038 | strcpy(outval, preaction); |
| 1039 | |
| 1040 | if (!inval) |
| 1041 | return 0; |
| 1042 | |
| 1043 | if (strcmp(inval, "pre_none") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1044 | preaction_val = WDOG_PRETIMEOUT_NONE; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1045 | else if (strcmp(inval, "pre_smi") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | preaction_val = WDOG_PRETIMEOUT_SMI; |
| 1047 | #ifdef HAVE_NMI_HANDLER |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1048 | else if (strcmp(inval, "pre_nmi") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | preaction_val = WDOG_PRETIMEOUT_NMI; |
| 1050 | #endif |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1051 | else if (strcmp(inval, "pre_int") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1052 | preaction_val = WDOG_PRETIMEOUT_MSG_INT; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1053 | else |
| 1054 | return -EINVAL; |
| 1055 | strcpy(preaction, inval); |
| 1056 | return 0; |
| 1057 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1059 | static int preop_op(const char *inval, char *outval) |
| 1060 | { |
| 1061 | if (outval) |
| 1062 | strcpy(outval, preop); |
| 1063 | |
| 1064 | if (!inval) |
| 1065 | return 0; |
| 1066 | |
| 1067 | if (strcmp(inval, "preop_none") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | preop_val = WDOG_PREOP_NONE; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1069 | else if (strcmp(inval, "preop_panic") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1070 | preop_val = WDOG_PREOP_PANIC; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1071 | else if (strcmp(inval, "preop_give_data") == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | preop_val = WDOG_PREOP_GIVE_DATA; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1073 | else |
| 1074 | return -EINVAL; |
| 1075 | strcpy(preop, inval); |
| 1076 | return 0; |
| 1077 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1078 | |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1079 | static void check_parms(void) |
| 1080 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1081 | #ifdef HAVE_NMI_HANDLER |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1082 | int do_nmi = 0; |
| 1083 | int rv; |
| 1084 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | if (preaction_val == WDOG_PRETIMEOUT_NMI) { |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1086 | do_nmi = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1087 | if (preop_val == WDOG_PREOP_GIVE_DATA) { |
| 1088 | printk(KERN_WARNING PFX "Pretimeout op is to give data" |
| 1089 | " but NMI pretimeout is enabled, setting" |
| 1090 | " pretimeout op to none\n"); |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1091 | preop_op("preop_none", NULL); |
| 1092 | do_nmi = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1093 | } |
| 1094 | #ifdef CONFIG_X86_LOCAL_APIC |
| 1095 | if (nmi_watchdog == NMI_IO_APIC) { |
| 1096 | printk(KERN_WARNING PFX "nmi_watchdog is set to IO APIC" |
| 1097 | " mode (value is %d), that is incompatible" |
| 1098 | " with using NMI in the IPMI watchdog." |
| 1099 | " Disabling IPMI nmi pretimeout.\n", |
| 1100 | nmi_watchdog); |
| 1101 | preaction_val = WDOG_PRETIMEOUT_NONE; |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1102 | do_nmi = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1103 | } |
| 1104 | #endif |
| 1105 | } |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1106 | if (do_nmi && !nmi_handler_registered) { |
| 1107 | rv = request_nmi(&ipmi_nmi_handler); |
| 1108 | if (rv) { |
| 1109 | printk(KERN_WARNING PFX |
| 1110 | "Can't register nmi handler\n"); |
| 1111 | return; |
| 1112 | } else |
| 1113 | nmi_handler_registered = 1; |
| 1114 | } else if (!do_nmi && nmi_handler_registered) { |
| 1115 | release_nmi(&ipmi_nmi_handler); |
| 1116 | nmi_handler_registered = 0; |
| 1117 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | #endif |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1119 | } |
| 1120 | |
| 1121 | static int __init ipmi_wdog_init(void) |
| 1122 | { |
| 1123 | int rv; |
| 1124 | |
| 1125 | if (action_op(action, NULL)) { |
| 1126 | action_op("reset", NULL); |
| 1127 | printk(KERN_INFO PFX "Unknown action '%s', defaulting to" |
| 1128 | " reset\n", action); |
| 1129 | } |
| 1130 | |
| 1131 | if (preaction_op(preaction, NULL)) { |
| 1132 | preaction_op("pre_none", NULL); |
| 1133 | printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to" |
| 1134 | " none\n", preaction); |
| 1135 | } |
| 1136 | |
| 1137 | if (preop_op(preop, NULL)) { |
| 1138 | preop_op("preop_none", NULL); |
| 1139 | printk(KERN_INFO PFX "Unknown preop '%s', defaulting to" |
| 1140 | " none\n", preop); |
| 1141 | } |
| 1142 | |
| 1143 | check_parms(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1144 | |
| 1145 | rv = ipmi_smi_watcher_register(&smi_watcher); |
| 1146 | if (rv) { |
| 1147 | #ifdef HAVE_NMI_HANDLER |
| 1148 | if (preaction_val == WDOG_PRETIMEOUT_NMI) |
| 1149 | release_nmi(&ipmi_nmi_handler); |
| 1150 | #endif |
| 1151 | printk(KERN_WARNING PFX "can't register smi watcher\n"); |
| 1152 | return rv; |
| 1153 | } |
| 1154 | |
| 1155 | register_reboot_notifier(&wdog_reboot_notifier); |
| 1156 | notifier_chain_register(&panic_notifier_list, &wdog_panic_notifier); |
| 1157 | |
Corey Minyard | 1fdd75b | 2005-09-06 15:18:42 -0700 | [diff] [blame] | 1158 | printk(KERN_INFO PFX "driver initialized\n"); |
| 1159 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1160 | return 0; |
| 1161 | } |
| 1162 | |
| 1163 | static __exit void ipmi_unregister_watchdog(void) |
| 1164 | { |
| 1165 | int rv; |
| 1166 | |
| 1167 | down_write(®ister_sem); |
| 1168 | |
| 1169 | #ifdef HAVE_NMI_HANDLER |
Corey Minyard | cc4673e | 2005-11-07 00:59:57 -0800 | [diff] [blame^] | 1170 | if (nmi_handler_registered) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1171 | release_nmi(&ipmi_nmi_handler); |
| 1172 | #endif |
| 1173 | |
| 1174 | notifier_chain_unregister(&panic_notifier_list, &wdog_panic_notifier); |
| 1175 | unregister_reboot_notifier(&wdog_reboot_notifier); |
| 1176 | |
| 1177 | if (! watchdog_user) |
| 1178 | goto out; |
| 1179 | |
| 1180 | /* Make sure no one can call us any more. */ |
| 1181 | misc_deregister(&ipmi_wdog_miscdev); |
| 1182 | |
| 1183 | /* Wait to make sure the message makes it out. The lower layer has |
| 1184 | pointers to our buffers, we want to make sure they are done before |
| 1185 | we release our memory. */ |
Nishanth Aravamudan | da4cd8d | 2005-09-10 00:27:30 -0700 | [diff] [blame] | 1186 | while (atomic_read(&set_timeout_tofree)) |
| 1187 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1188 | |
| 1189 | /* Disconnect from IPMI. */ |
| 1190 | rv = ipmi_destroy_user(watchdog_user); |
| 1191 | if (rv) { |
| 1192 | printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n", |
| 1193 | rv); |
| 1194 | } |
| 1195 | watchdog_user = NULL; |
| 1196 | |
| 1197 | out: |
| 1198 | up_write(®ister_sem); |
| 1199 | } |
| 1200 | |
| 1201 | static void __exit ipmi_wdog_exit(void) |
| 1202 | { |
| 1203 | ipmi_smi_watcher_unregister(&smi_watcher); |
| 1204 | ipmi_unregister_watchdog(); |
| 1205 | } |
| 1206 | module_exit(ipmi_wdog_exit); |
| 1207 | module_init(ipmi_wdog_init); |
| 1208 | MODULE_LICENSE("GPL"); |
Corey Minyard | 1fdd75b | 2005-09-06 15:18:42 -0700 | [diff] [blame] | 1209 | MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); |
| 1210 | MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); |