1 | // SPDX-License-Identifier: GPL-2.0+ |
2 | /* |
3 | * ipmi_watchdog.c |
4 | * |
5 | * A watchdog timer based upon the IPMI interface. |
6 | * |
7 | * Author: MontaVista Software, Inc. |
8 | * Corey Minyard <minyard@mvista.com> |
9 | * source@mvista.com |
10 | * |
11 | * Copyright 2002 MontaVista Software Inc. |
12 | */ |
13 | |
14 | #define pr_fmt(fmt) "IPMI Watchdog: " fmt |
15 | |
16 | #include <linux/module.h> |
17 | #include <linux/moduleparam.h> |
18 | #include <linux/ipmi.h> |
19 | #include <linux/ipmi_smi.h> |
20 | #include <linux/mutex.h> |
21 | #include <linux/watchdog.h> |
22 | #include <linux/miscdevice.h> |
23 | #include <linux/init.h> |
24 | #include <linux/completion.h> |
25 | #include <linux/kdebug.h> |
26 | #include <linux/kstrtox.h> |
27 | #include <linux/rwsem.h> |
28 | #include <linux/errno.h> |
29 | #include <linux/uaccess.h> |
30 | #include <linux/notifier.h> |
31 | #include <linux/nmi.h> |
32 | #include <linux/reboot.h> |
33 | #include <linux/wait.h> |
34 | #include <linux/poll.h> |
35 | #include <linux/string.h> |
36 | #include <linux/ctype.h> |
37 | #include <linux/delay.h> |
38 | #include <linux/atomic.h> |
39 | #include <linux/sched/signal.h> |
40 | |
41 | #ifdef CONFIG_X86 |
42 | /* |
43 | * This is ugly, but I've determined that x86 is the only architecture |
44 | * that can reasonably support the IPMI NMI watchdog timeout at this |
45 | * time. If another architecture adds this capability somehow, it |
46 | * will have to be a somewhat different mechanism and I have no idea |
47 | * how it will work. So in the unlikely event that another |
48 | * architecture supports this, we can figure out a good generic |
49 | * mechanism for it at that time. |
50 | */ |
51 | #include <asm/kdebug.h> |
52 | #include <asm/nmi.h> |
53 | #define HAVE_DIE_NMI |
54 | #endif |
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 | /* Cause data to be available to read. Doesn't work in NMI mode. */ |
85 | #define WDOG_PREOP_GIVE_DATA 2 |
86 | |
87 | /* Actions to perform on a full timeout. */ |
88 | #define WDOG_SET_TIMEOUT_ACT(byte, use) \ |
89 | byte = ((byte) & 0xf8) | ((use) & 0x7) |
90 | #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) |
91 | #define WDOG_TIMEOUT_NONE 0 |
92 | #define WDOG_TIMEOUT_RESET 1 |
93 | #define WDOG_TIMEOUT_POWER_DOWN 2 |
94 | #define WDOG_TIMEOUT_POWER_CYCLE 3 |
95 | |
96 | /* |
97 | * Byte 3 of the get command, byte 4 of the get response is the |
98 | * pre-timeout in seconds. |
99 | */ |
100 | |
101 | /* Bits for setting byte 4 of the set command, byte 5 of the get response. */ |
102 | #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1) |
103 | #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2) |
104 | #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3) |
105 | #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4) |
106 | #define WDOG_EXPIRE_CLEAR_OEM (1 << 5) |
107 | |
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 | */ |
115 | #define WDOG_SET_TIMEOUT(byte1, byte2, val) \ |
116 | (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8) |
117 | #define WDOG_GET_TIMEOUT(byte1, byte2) \ |
118 | (((byte1) | ((byte2) << 8)) / 10) |
119 | |
120 | #define IPMI_WDOG_RESET_TIMER 0x22 |
121 | #define IPMI_WDOG_SET_TIMER 0x24 |
122 | #define IPMI_WDOG_GET_TIMER 0x25 |
123 | |
124 | #define IPMI_WDOG_TIMER_NOT_INIT_RESP 0x80 |
125 | |
126 | static DEFINE_MUTEX(ipmi_watchdog_mutex); |
127 | static bool nowayout = WATCHDOG_NOWAYOUT; |
128 | |
129 | static struct ipmi_user *watchdog_user; |
130 | static int watchdog_ifnum; |
131 | |
132 | /* Default the timeout to 10 seconds. */ |
133 | static int timeout = 10; |
134 | |
135 | /* The pre-timeout is disabled by default. */ |
136 | static int pretimeout; |
137 | |
138 | /* Default timeout to set on panic */ |
139 | static int panic_wdt_timeout = 255; |
140 | |
141 | /* Default action is to reset the board on a timeout. */ |
142 | static unsigned char action_val = WDOG_TIMEOUT_RESET; |
143 | |
144 | static char action[16] = "reset" ; |
145 | |
146 | static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; |
147 | |
148 | static char preaction[16] = "pre_none" ; |
149 | |
150 | static unsigned char preop_val = WDOG_PREOP_NONE; |
151 | |
152 | static char preop[16] = "preop_none" ; |
153 | static DEFINE_SPINLOCK(ipmi_read_lock); |
154 | static char data_to_read; |
155 | static DECLARE_WAIT_QUEUE_HEAD(read_q); |
156 | static struct fasync_struct *fasync_q; |
157 | static atomic_t pretimeout_since_last_heartbeat; |
158 | static char expect_close; |
159 | |
160 | static int ifnum_to_use = -1; |
161 | |
162 | /* Parameters to ipmi_set_timeout */ |
163 | #define IPMI_SET_TIMEOUT_NO_HB 0 |
164 | #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 |
165 | #define IPMI_SET_TIMEOUT_FORCE_HB 2 |
166 | |
167 | static int ipmi_set_timeout(int do_heartbeat); |
168 | static void ipmi_register_watchdog(int ipmi_intf); |
169 | static void ipmi_unregister_watchdog(int ipmi_intf); |
170 | |
171 | /* |
172 | * If true, the driver will start running as soon as it is configured |
173 | * and ready. |
174 | */ |
175 | static int start_now; |
176 | |
177 | static int set_param_timeout(const char *val, const struct kernel_param *kp) |
178 | { |
179 | char *endp; |
180 | int l; |
181 | int rv = 0; |
182 | |
183 | if (!val) |
184 | return -EINVAL; |
185 | l = simple_strtoul(val, &endp, 0); |
186 | if (endp == val) |
187 | return -EINVAL; |
188 | |
189 | *((int *)kp->arg) = l; |
190 | if (watchdog_user) |
191 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
192 | |
193 | return rv; |
194 | } |
195 | |
196 | static const struct kernel_param_ops param_ops_timeout = { |
197 | .set = set_param_timeout, |
198 | .get = param_get_int, |
199 | }; |
200 | #define param_check_timeout param_check_int |
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, const struct kernel_param *kp) |
210 | { |
211 | action_fn fn = (action_fn) kp->arg; |
212 | int rv = 0; |
213 | char valcp[16]; |
214 | char *s; |
215 | |
216 | strscpy(p: valcp, q: val, size: 16); |
217 | |
218 | s = strstrip(str: valcp); |
219 | |
220 | rv = fn(s, NULL); |
221 | if (rv) |
222 | goto out; |
223 | |
224 | check_parms(); |
225 | if (watchdog_user) |
226 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
227 | |
228 | out: |
229 | return rv; |
230 | } |
231 | |
232 | static int get_param_str(char *buffer, const struct kernel_param *kp) |
233 | { |
234 | action_fn fn = (action_fn) kp->arg; |
235 | int rv, len; |
236 | |
237 | rv = fn(NULL, buffer); |
238 | if (rv) |
239 | return rv; |
240 | |
241 | len = strlen(buffer); |
242 | buffer[len++] = '\n'; |
243 | buffer[len] = 0; |
244 | |
245 | return len; |
246 | } |
247 | |
248 | |
249 | static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp) |
250 | { |
251 | int rv = param_set_int(val, kp); |
252 | if (rv) |
253 | return rv; |
254 | if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum)) |
255 | return 0; |
256 | |
257 | ipmi_unregister_watchdog(ipmi_intf: watchdog_ifnum); |
258 | ipmi_register_watchdog(ipmi_intf: ifnum_to_use); |
259 | return 0; |
260 | } |
261 | |
262 | static const struct kernel_param_ops param_ops_wdog_ifnum = { |
263 | .set = set_param_wdog_ifnum, |
264 | .get = param_get_int, |
265 | }; |
266 | |
267 | #define param_check_wdog_ifnum param_check_int |
268 | |
269 | static const struct kernel_param_ops param_ops_str = { |
270 | .set = set_param_str, |
271 | .get = get_param_str, |
272 | }; |
273 | |
274 | module_param(ifnum_to_use, wdog_ifnum, 0644); |
275 | MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog " |
276 | "timer. Setting to -1 defaults to the first registered " |
277 | "interface" ); |
278 | |
279 | module_param(timeout, timeout, 0644); |
280 | MODULE_PARM_DESC(timeout, "Timeout value in seconds." ); |
281 | |
282 | module_param(pretimeout, timeout, 0644); |
283 | MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds." ); |
284 | |
285 | module_param(panic_wdt_timeout, timeout, 0644); |
286 | MODULE_PARM_DESC(panic_wdt_timeout, "Timeout value on kernel panic in seconds." ); |
287 | |
288 | module_param_cb(action, ¶m_ops_str, action_op, 0644); |
289 | MODULE_PARM_DESC(action, "Timeout action. One of: " |
290 | "reset, none, power_cycle, power_off." ); |
291 | |
292 | module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644); |
293 | MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " |
294 | "pre_none, pre_smi, pre_nmi, pre_int." ); |
295 | |
296 | module_param_cb(preop, ¶m_ops_str, preop_op, 0644); |
297 | MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " |
298 | "preop_none, preop_panic, preop_give_data." ); |
299 | |
300 | module_param(start_now, int, 0444); |
301 | MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" |
302 | "soon as the driver is loaded." ); |
303 | |
304 | module_param(nowayout, bool, 0644); |
305 | MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started " |
306 | "(default=CONFIG_WATCHDOG_NOWAYOUT)" ); |
307 | |
308 | /* Default state of the timer. */ |
309 | static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
310 | |
311 | /* Is someone using the watchdog? Only one user is allowed. */ |
312 | static unsigned long ipmi_wdog_open; |
313 | |
314 | /* |
315 | * If set to 1, the heartbeat command will set the state to reset and |
316 | * start the timer. The timer doesn't normally run when the driver is |
317 | * first opened until the heartbeat is set the first time, this |
318 | * variable is used to accomplish this. |
319 | */ |
320 | static int ipmi_start_timer_on_heartbeat; |
321 | |
322 | /* IPMI version of the BMC. */ |
323 | static unsigned char ipmi_version_major; |
324 | static unsigned char ipmi_version_minor; |
325 | |
326 | /* If a pretimeout occurs, this is used to allow only one panic to happen. */ |
327 | static atomic_t preop_panic_excl = ATOMIC_INIT(-1); |
328 | |
329 | #ifdef HAVE_DIE_NMI |
330 | static int testing_nmi; |
331 | static int nmi_handler_registered; |
332 | #endif |
333 | |
334 | static int __ipmi_heartbeat(void); |
335 | |
336 | /* |
337 | * We use a mutex to make sure that only one thing can send a set a |
338 | * message at one time. The mutex is claimed when a message is sent |
339 | * and freed when both the send and receive messages are free. |
340 | */ |
341 | static atomic_t msg_tofree = ATOMIC_INIT(0); |
342 | static DECLARE_COMPLETION(msg_wait); |
343 | static void msg_free_smi(struct ipmi_smi_msg *msg) |
344 | { |
345 | if (atomic_dec_and_test(v: &msg_tofree)) { |
346 | if (!oops_in_progress) |
347 | complete(&msg_wait); |
348 | } |
349 | } |
350 | static void msg_free_recv(struct ipmi_recv_msg *msg) |
351 | { |
352 | if (atomic_dec_and_test(v: &msg_tofree)) { |
353 | if (!oops_in_progress) |
354 | complete(&msg_wait); |
355 | } |
356 | } |
357 | static struct ipmi_smi_msg smi_msg = INIT_IPMI_SMI_MSG(msg_free_smi); |
358 | static struct ipmi_recv_msg recv_msg = INIT_IPMI_RECV_MSG(msg_free_recv); |
359 | |
360 | static int __ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, |
361 | struct ipmi_recv_msg *recv_msg, |
362 | int *send_heartbeat_now) |
363 | { |
364 | struct kernel_ipmi_msg msg; |
365 | unsigned char data[6]; |
366 | int rv; |
367 | struct ipmi_system_interface_addr addr; |
368 | int hbnow = 0; |
369 | |
370 | |
371 | data[0] = 0; |
372 | WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); |
373 | |
374 | if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
375 | if ((ipmi_version_major > 1) || |
376 | ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) { |
377 | /* This is an IPMI 1.5-only feature. */ |
378 | data[0] |= WDOG_DONT_STOP_ON_SET; |
379 | } else { |
380 | /* |
381 | * In ipmi 1.0, setting the timer stops the watchdog, we |
382 | * need to start it back up again. |
383 | */ |
384 | hbnow = 1; |
385 | } |
386 | } |
387 | |
388 | data[1] = 0; |
389 | WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); |
390 | if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { |
391 | WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); |
392 | data[2] = pretimeout; |
393 | } else { |
394 | WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); |
395 | data[2] = 0; /* No pretimeout. */ |
396 | } |
397 | data[3] = 0; |
398 | WDOG_SET_TIMEOUT(data[4], data[5], timeout); |
399 | |
400 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; |
401 | addr.channel = IPMI_BMC_CHANNEL; |
402 | addr.lun = 0; |
403 | |
404 | msg.netfn = 0x06; |
405 | msg.cmd = IPMI_WDOG_SET_TIMER; |
406 | msg.data = data; |
407 | msg.data_len = sizeof(data); |
408 | rv = ipmi_request_supply_msgs(user: watchdog_user, |
409 | addr: (struct ipmi_addr *) &addr, |
410 | msgid: 0, |
411 | msg: &msg, |
412 | NULL, |
413 | supplied_smi: smi_msg, |
414 | supplied_recv: recv_msg, |
415 | priority: 1); |
416 | if (rv) |
417 | pr_warn("set timeout error: %d\n" , rv); |
418 | else if (send_heartbeat_now) |
419 | *send_heartbeat_now = hbnow; |
420 | |
421 | return rv; |
422 | } |
423 | |
424 | static int _ipmi_set_timeout(int do_heartbeat) |
425 | { |
426 | int send_heartbeat_now; |
427 | int rv; |
428 | |
429 | if (!watchdog_user) |
430 | return -ENODEV; |
431 | |
432 | atomic_set(v: &msg_tofree, i: 2); |
433 | |
434 | rv = __ipmi_set_timeout(smi_msg: &smi_msg, |
435 | recv_msg: &recv_msg, |
436 | send_heartbeat_now: &send_heartbeat_now); |
437 | if (rv) { |
438 | atomic_set(v: &msg_tofree, i: 0); |
439 | return rv; |
440 | } |
441 | |
442 | wait_for_completion(&msg_wait); |
443 | |
444 | if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) |
445 | || ((send_heartbeat_now) |
446 | && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) |
447 | rv = __ipmi_heartbeat(); |
448 | |
449 | return rv; |
450 | } |
451 | |
452 | static int ipmi_set_timeout(int do_heartbeat) |
453 | { |
454 | int rv; |
455 | |
456 | mutex_lock(&ipmi_watchdog_mutex); |
457 | rv = _ipmi_set_timeout(do_heartbeat); |
458 | mutex_unlock(lock: &ipmi_watchdog_mutex); |
459 | |
460 | return rv; |
461 | } |
462 | |
463 | static atomic_t panic_done_count = ATOMIC_INIT(0); |
464 | |
465 | static void panic_smi_free(struct ipmi_smi_msg *msg) |
466 | { |
467 | atomic_dec(v: &panic_done_count); |
468 | } |
469 | static void panic_recv_free(struct ipmi_recv_msg *msg) |
470 | { |
471 | atomic_dec(v: &panic_done_count); |
472 | } |
473 | |
474 | static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = |
475 | INIT_IPMI_SMI_MSG(panic_smi_free); |
476 | static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = |
477 | INIT_IPMI_RECV_MSG(panic_recv_free); |
478 | |
479 | static void panic_halt_ipmi_heartbeat(void) |
480 | { |
481 | struct kernel_ipmi_msg msg; |
482 | struct ipmi_system_interface_addr addr; |
483 | int rv; |
484 | |
485 | /* |
486 | * Don't reset the timer if we have the timer turned off, that |
487 | * re-enables the watchdog. |
488 | */ |
489 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
490 | return; |
491 | |
492 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; |
493 | addr.channel = IPMI_BMC_CHANNEL; |
494 | addr.lun = 0; |
495 | |
496 | msg.netfn = 0x06; |
497 | msg.cmd = IPMI_WDOG_RESET_TIMER; |
498 | msg.data = NULL; |
499 | msg.data_len = 0; |
500 | atomic_add(i: 2, v: &panic_done_count); |
501 | rv = ipmi_request_supply_msgs(user: watchdog_user, |
502 | addr: (struct ipmi_addr *) &addr, |
503 | msgid: 0, |
504 | msg: &msg, |
505 | NULL, |
506 | supplied_smi: &panic_halt_heartbeat_smi_msg, |
507 | supplied_recv: &panic_halt_heartbeat_recv_msg, |
508 | priority: 1); |
509 | if (rv) |
510 | atomic_sub(i: 2, v: &panic_done_count); |
511 | } |
512 | |
513 | static struct ipmi_smi_msg panic_halt_smi_msg = |
514 | INIT_IPMI_SMI_MSG(panic_smi_free); |
515 | static struct ipmi_recv_msg panic_halt_recv_msg = |
516 | INIT_IPMI_RECV_MSG(panic_recv_free); |
517 | |
518 | /* |
519 | * Special call, doesn't claim any locks. This is only to be called |
520 | * at panic or halt time, in run-to-completion mode, when the caller |
521 | * is the only CPU and the only thing that will be going is these IPMI |
522 | * calls. |
523 | */ |
524 | static void panic_halt_ipmi_set_timeout(void) |
525 | { |
526 | int send_heartbeat_now; |
527 | int rv; |
528 | |
529 | /* Wait for the messages to be free. */ |
530 | while (atomic_read(v: &panic_done_count) != 0) |
531 | ipmi_poll_interface(user: watchdog_user); |
532 | atomic_add(i: 2, v: &panic_done_count); |
533 | rv = __ipmi_set_timeout(smi_msg: &panic_halt_smi_msg, |
534 | recv_msg: &panic_halt_recv_msg, |
535 | send_heartbeat_now: &send_heartbeat_now); |
536 | if (rv) { |
537 | atomic_sub(i: 2, v: &panic_done_count); |
538 | pr_warn("Unable to extend the watchdog timeout\n" ); |
539 | } else { |
540 | if (send_heartbeat_now) |
541 | panic_halt_ipmi_heartbeat(); |
542 | } |
543 | while (atomic_read(v: &panic_done_count) != 0) |
544 | ipmi_poll_interface(user: watchdog_user); |
545 | } |
546 | |
547 | static int __ipmi_heartbeat(void) |
548 | { |
549 | struct kernel_ipmi_msg msg; |
550 | int rv; |
551 | struct ipmi_system_interface_addr addr; |
552 | int timeout_retries = 0; |
553 | |
554 | restart: |
555 | /* |
556 | * Don't reset the timer if we have the timer turned off, that |
557 | * re-enables the watchdog. |
558 | */ |
559 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
560 | return 0; |
561 | |
562 | atomic_set(v: &msg_tofree, i: 2); |
563 | |
564 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; |
565 | addr.channel = IPMI_BMC_CHANNEL; |
566 | addr.lun = 0; |
567 | |
568 | msg.netfn = 0x06; |
569 | msg.cmd = IPMI_WDOG_RESET_TIMER; |
570 | msg.data = NULL; |
571 | msg.data_len = 0; |
572 | rv = ipmi_request_supply_msgs(user: watchdog_user, |
573 | addr: (struct ipmi_addr *) &addr, |
574 | msgid: 0, |
575 | msg: &msg, |
576 | NULL, |
577 | supplied_smi: &smi_msg, |
578 | supplied_recv: &recv_msg, |
579 | priority: 1); |
580 | if (rv) { |
581 | atomic_set(v: &msg_tofree, i: 0); |
582 | pr_warn("heartbeat send failure: %d\n" , rv); |
583 | return rv; |
584 | } |
585 | |
586 | /* Wait for the heartbeat to be sent. */ |
587 | wait_for_completion(&msg_wait); |
588 | |
589 | if (recv_msg.msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) { |
590 | timeout_retries++; |
591 | if (timeout_retries > 3) { |
592 | pr_err("Unable to restore the IPMI watchdog's settings, giving up\n" ); |
593 | rv = -EIO; |
594 | goto out; |
595 | } |
596 | |
597 | /* |
598 | * The timer was not initialized, that means the BMC was |
599 | * probably reset and lost the watchdog information. Attempt |
600 | * to restore the timer's info. Note that we still hold |
601 | * the heartbeat lock, to keep a heartbeat from happening |
602 | * in this process, so must say no heartbeat to avoid a |
603 | * deadlock on this mutex |
604 | */ |
605 | rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
606 | if (rv) { |
607 | pr_err("Unable to send the command to set the watchdog's settings, giving up\n" ); |
608 | goto out; |
609 | } |
610 | |
611 | /* Might need a heartbeat send, go ahead and do it. */ |
612 | goto restart; |
613 | } else if (recv_msg.msg.data[0] != 0) { |
614 | /* |
615 | * Got an error in the heartbeat response. It was already |
616 | * reported in ipmi_wdog_msg_handler, but we should return |
617 | * an error here. |
618 | */ |
619 | rv = -EINVAL; |
620 | } |
621 | |
622 | out: |
623 | return rv; |
624 | } |
625 | |
626 | static int _ipmi_heartbeat(void) |
627 | { |
628 | int rv; |
629 | |
630 | if (!watchdog_user) |
631 | return -ENODEV; |
632 | |
633 | if (ipmi_start_timer_on_heartbeat) { |
634 | ipmi_start_timer_on_heartbeat = 0; |
635 | ipmi_watchdog_state = action_val; |
636 | rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
637 | } else if (atomic_cmpxchg(v: &pretimeout_since_last_heartbeat, old: 1, new: 0)) { |
638 | /* |
639 | * A pretimeout occurred, make sure we set the timeout. |
640 | * We don't want to set the action, though, we want to |
641 | * leave that alone (thus it can't be combined with the |
642 | * above operation. |
643 | */ |
644 | rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
645 | } else { |
646 | rv = __ipmi_heartbeat(); |
647 | } |
648 | |
649 | return rv; |
650 | } |
651 | |
652 | static int ipmi_heartbeat(void) |
653 | { |
654 | int rv; |
655 | |
656 | mutex_lock(&ipmi_watchdog_mutex); |
657 | rv = _ipmi_heartbeat(); |
658 | mutex_unlock(lock: &ipmi_watchdog_mutex); |
659 | |
660 | return rv; |
661 | } |
662 | |
663 | static const struct watchdog_info ident = { |
664 | .options = 0, /* WDIOF_SETTIMEOUT, */ |
665 | .firmware_version = 1, |
666 | .identity = "IPMI" |
667 | }; |
668 | |
669 | static int ipmi_ioctl(struct file *file, |
670 | unsigned int cmd, unsigned long arg) |
671 | { |
672 | void __user *argp = (void __user *)arg; |
673 | int i; |
674 | int val; |
675 | |
676 | switch (cmd) { |
677 | case WDIOC_GETSUPPORT: |
678 | i = copy_to_user(to: argp, from: &ident, n: sizeof(ident)); |
679 | return i ? -EFAULT : 0; |
680 | |
681 | case WDIOC_SETTIMEOUT: |
682 | i = copy_from_user(to: &val, from: argp, n: sizeof(int)); |
683 | if (i) |
684 | return -EFAULT; |
685 | timeout = val; |
686 | return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
687 | |
688 | case WDIOC_GETTIMEOUT: |
689 | i = copy_to_user(to: argp, from: &timeout, n: sizeof(timeout)); |
690 | if (i) |
691 | return -EFAULT; |
692 | return 0; |
693 | |
694 | case WDIOC_SETPRETIMEOUT: |
695 | i = copy_from_user(to: &val, from: argp, n: sizeof(int)); |
696 | if (i) |
697 | return -EFAULT; |
698 | pretimeout = val; |
699 | return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
700 | |
701 | case WDIOC_GETPRETIMEOUT: |
702 | i = copy_to_user(to: argp, from: &pretimeout, n: sizeof(pretimeout)); |
703 | if (i) |
704 | return -EFAULT; |
705 | return 0; |
706 | |
707 | case WDIOC_KEEPALIVE: |
708 | return _ipmi_heartbeat(); |
709 | |
710 | case WDIOC_SETOPTIONS: |
711 | i = copy_from_user(to: &val, from: argp, n: sizeof(int)); |
712 | if (i) |
713 | return -EFAULT; |
714 | if (val & WDIOS_DISABLECARD) { |
715 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
716 | _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
717 | ipmi_start_timer_on_heartbeat = 0; |
718 | } |
719 | |
720 | if (val & WDIOS_ENABLECARD) { |
721 | ipmi_watchdog_state = action_val; |
722 | _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
723 | } |
724 | return 0; |
725 | |
726 | case WDIOC_GETSTATUS: |
727 | val = 0; |
728 | i = copy_to_user(to: argp, from: &val, n: sizeof(val)); |
729 | if (i) |
730 | return -EFAULT; |
731 | return 0; |
732 | |
733 | default: |
734 | return -ENOIOCTLCMD; |
735 | } |
736 | } |
737 | |
738 | static long ipmi_unlocked_ioctl(struct file *file, |
739 | unsigned int cmd, |
740 | unsigned long arg) |
741 | { |
742 | int ret; |
743 | |
744 | mutex_lock(&ipmi_watchdog_mutex); |
745 | ret = ipmi_ioctl(file, cmd, arg); |
746 | mutex_unlock(lock: &ipmi_watchdog_mutex); |
747 | |
748 | return ret; |
749 | } |
750 | |
751 | static ssize_t ipmi_write(struct file *file, |
752 | const char __user *buf, |
753 | size_t len, |
754 | loff_t *ppos) |
755 | { |
756 | int rv; |
757 | |
758 | if (len) { |
759 | if (!nowayout) { |
760 | size_t i; |
761 | |
762 | /* In case it was set long ago */ |
763 | expect_close = 0; |
764 | |
765 | for (i = 0; i != len; i++) { |
766 | char c; |
767 | |
768 | if (get_user(c, buf + i)) |
769 | return -EFAULT; |
770 | if (c == 'V') |
771 | expect_close = 42; |
772 | } |
773 | } |
774 | rv = ipmi_heartbeat(); |
775 | if (rv) |
776 | return rv; |
777 | } |
778 | return len; |
779 | } |
780 | |
781 | static ssize_t ipmi_read(struct file *file, |
782 | char __user *buf, |
783 | size_t count, |
784 | loff_t *ppos) |
785 | { |
786 | int rv = 0; |
787 | wait_queue_entry_t wait; |
788 | |
789 | if (count <= 0) |
790 | return 0; |
791 | |
792 | /* |
793 | * Reading returns if the pretimeout has gone off, and it only does |
794 | * it once per pretimeout. |
795 | */ |
796 | spin_lock_irq(lock: &ipmi_read_lock); |
797 | if (!data_to_read) { |
798 | if (file->f_flags & O_NONBLOCK) { |
799 | rv = -EAGAIN; |
800 | goto out; |
801 | } |
802 | |
803 | init_waitqueue_entry(wq_entry: &wait, current); |
804 | add_wait_queue(wq_head: &read_q, wq_entry: &wait); |
805 | while (!data_to_read && !signal_pending(current)) { |
806 | set_current_state(TASK_INTERRUPTIBLE); |
807 | spin_unlock_irq(lock: &ipmi_read_lock); |
808 | schedule(); |
809 | spin_lock_irq(lock: &ipmi_read_lock); |
810 | } |
811 | remove_wait_queue(wq_head: &read_q, wq_entry: &wait); |
812 | |
813 | if (signal_pending(current)) { |
814 | rv = -ERESTARTSYS; |
815 | goto out; |
816 | } |
817 | } |
818 | data_to_read = 0; |
819 | |
820 | out: |
821 | spin_unlock_irq(lock: &ipmi_read_lock); |
822 | |
823 | if (rv == 0) { |
824 | if (copy_to_user(to: buf, from: &data_to_read, n: 1)) |
825 | rv = -EFAULT; |
826 | else |
827 | rv = 1; |
828 | } |
829 | |
830 | return rv; |
831 | } |
832 | |
833 | static int ipmi_open(struct inode *ino, struct file *filep) |
834 | { |
835 | switch (iminor(inode: ino)) { |
836 | case WATCHDOG_MINOR: |
837 | if (test_and_set_bit(nr: 0, addr: &ipmi_wdog_open)) |
838 | return -EBUSY; |
839 | |
840 | |
841 | /* |
842 | * Don't start the timer now, let it start on the |
843 | * first heartbeat. |
844 | */ |
845 | ipmi_start_timer_on_heartbeat = 1; |
846 | return stream_open(inode: ino, filp: filep); |
847 | |
848 | default: |
849 | return (-ENODEV); |
850 | } |
851 | } |
852 | |
853 | static __poll_t ipmi_poll(struct file *file, poll_table *wait) |
854 | { |
855 | __poll_t mask = 0; |
856 | |
857 | poll_wait(filp: file, wait_address: &read_q, p: wait); |
858 | |
859 | spin_lock_irq(lock: &ipmi_read_lock); |
860 | if (data_to_read) |
861 | mask |= (EPOLLIN | EPOLLRDNORM); |
862 | spin_unlock_irq(lock: &ipmi_read_lock); |
863 | |
864 | return mask; |
865 | } |
866 | |
867 | static int ipmi_fasync(int fd, struct file *file, int on) |
868 | { |
869 | int result; |
870 | |
871 | result = fasync_helper(fd, file, on, &fasync_q); |
872 | |
873 | return (result); |
874 | } |
875 | |
876 | static int ipmi_close(struct inode *ino, struct file *filep) |
877 | { |
878 | if (iminor(inode: ino) == WATCHDOG_MINOR) { |
879 | if (expect_close == 42) { |
880 | mutex_lock(&ipmi_watchdog_mutex); |
881 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
882 | _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
883 | mutex_unlock(lock: &ipmi_watchdog_mutex); |
884 | } else { |
885 | pr_crit("Unexpected close, not stopping watchdog!\n" ); |
886 | ipmi_heartbeat(); |
887 | } |
888 | clear_bit(nr: 0, addr: &ipmi_wdog_open); |
889 | } |
890 | |
891 | expect_close = 0; |
892 | |
893 | return 0; |
894 | } |
895 | |
896 | static const struct file_operations ipmi_wdog_fops = { |
897 | .owner = THIS_MODULE, |
898 | .read = ipmi_read, |
899 | .poll = ipmi_poll, |
900 | .write = ipmi_write, |
901 | .unlocked_ioctl = ipmi_unlocked_ioctl, |
902 | .compat_ioctl = compat_ptr_ioctl, |
903 | .open = ipmi_open, |
904 | .release = ipmi_close, |
905 | .fasync = ipmi_fasync, |
906 | .llseek = no_llseek, |
907 | }; |
908 | |
909 | static struct miscdevice ipmi_wdog_miscdev = { |
910 | .minor = WATCHDOG_MINOR, |
911 | .name = "watchdog" , |
912 | .fops = &ipmi_wdog_fops |
913 | }; |
914 | |
915 | static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, |
916 | void *handler_data) |
917 | { |
918 | if (msg->msg.cmd == IPMI_WDOG_RESET_TIMER && |
919 | msg->msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) |
920 | pr_info("response: The IPMI controller appears to have been reset, will attempt to reinitialize the watchdog timer\n" ); |
921 | else if (msg->msg.data[0] != 0) |
922 | pr_err("response: Error %x on cmd %x\n" , |
923 | msg->msg.data[0], |
924 | msg->msg.cmd); |
925 | |
926 | ipmi_free_recv_msg(msg); |
927 | } |
928 | |
929 | static void ipmi_wdog_pretimeout_handler(void *handler_data) |
930 | { |
931 | if (preaction_val != WDOG_PRETIMEOUT_NONE) { |
932 | if (preop_val == WDOG_PREOP_PANIC) { |
933 | if (atomic_inc_and_test(v: &preop_panic_excl)) |
934 | panic(fmt: "Watchdog pre-timeout" ); |
935 | } else if (preop_val == WDOG_PREOP_GIVE_DATA) { |
936 | unsigned long flags; |
937 | |
938 | spin_lock_irqsave(&ipmi_read_lock, flags); |
939 | data_to_read = 1; |
940 | wake_up_interruptible(&read_q); |
941 | kill_fasync(&fasync_q, SIGIO, POLL_IN); |
942 | spin_unlock_irqrestore(lock: &ipmi_read_lock, flags); |
943 | } |
944 | } |
945 | |
946 | /* |
947 | * On some machines, the heartbeat will give an error and not |
948 | * work unless we re-enable the timer. So do so. |
949 | */ |
950 | atomic_set(v: &pretimeout_since_last_heartbeat, i: 1); |
951 | } |
952 | |
953 | static void ipmi_wdog_panic_handler(void *user_data) |
954 | { |
955 | static int panic_event_handled; |
956 | |
957 | /* |
958 | * On a panic, if we have a panic timeout, make sure to extend |
959 | * the watchdog timer to a reasonable value to complete the |
960 | * panic, if the watchdog timer is running. Plus the |
961 | * pretimeout is meaningless at panic time. |
962 | */ |
963 | if (watchdog_user && !panic_event_handled && |
964 | ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
965 | /* Make sure we do this only once. */ |
966 | panic_event_handled = 1; |
967 | |
968 | timeout = panic_wdt_timeout; |
969 | pretimeout = 0; |
970 | panic_halt_ipmi_set_timeout(); |
971 | } |
972 | } |
973 | |
974 | static const struct ipmi_user_hndl ipmi_hndlrs = { |
975 | .ipmi_recv_hndl = ipmi_wdog_msg_handler, |
976 | .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler, |
977 | .ipmi_panic_handler = ipmi_wdog_panic_handler |
978 | }; |
979 | |
980 | static void ipmi_register_watchdog(int ipmi_intf) |
981 | { |
982 | int rv = -EBUSY; |
983 | |
984 | if (watchdog_user) |
985 | goto out; |
986 | |
987 | if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf)) |
988 | goto out; |
989 | |
990 | watchdog_ifnum = ipmi_intf; |
991 | |
992 | rv = ipmi_create_user(if_num: ipmi_intf, handler: &ipmi_hndlrs, NULL, user: &watchdog_user); |
993 | if (rv < 0) { |
994 | pr_crit("Unable to register with ipmi\n" ); |
995 | goto out; |
996 | } |
997 | |
998 | rv = ipmi_get_version(user: watchdog_user, |
999 | major: &ipmi_version_major, |
1000 | minor: &ipmi_version_minor); |
1001 | if (rv) { |
1002 | pr_warn("Unable to get IPMI version, assuming 1.0\n" ); |
1003 | ipmi_version_major = 1; |
1004 | ipmi_version_minor = 0; |
1005 | } |
1006 | |
1007 | rv = misc_register(misc: &ipmi_wdog_miscdev); |
1008 | if (rv < 0) { |
1009 | ipmi_destroy_user(user: watchdog_user); |
1010 | watchdog_user = NULL; |
1011 | pr_crit("Unable to register misc device\n" ); |
1012 | } |
1013 | |
1014 | #ifdef HAVE_DIE_NMI |
1015 | if (nmi_handler_registered) { |
1016 | int old_pretimeout = pretimeout; |
1017 | int old_timeout = timeout; |
1018 | int old_preop_val = preop_val; |
1019 | |
1020 | /* |
1021 | * Set the pretimeout to go off in a second and give |
1022 | * ourselves plenty of time to stop the timer. |
1023 | */ |
1024 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; |
1025 | preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */ |
1026 | pretimeout = 99; |
1027 | timeout = 100; |
1028 | |
1029 | testing_nmi = 1; |
1030 | |
1031 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
1032 | if (rv) { |
1033 | pr_warn("Error starting timer to test NMI: 0x%x. The NMI pretimeout will likely not work\n" , |
1034 | rv); |
1035 | rv = 0; |
1036 | goto out_restore; |
1037 | } |
1038 | |
1039 | msleep(msecs: 1500); |
1040 | |
1041 | if (testing_nmi != 2) { |
1042 | pr_warn("IPMI NMI didn't seem to occur. The NMI pretimeout will likely not work\n" ); |
1043 | } |
1044 | out_restore: |
1045 | testing_nmi = 0; |
1046 | preop_val = old_preop_val; |
1047 | pretimeout = old_pretimeout; |
1048 | timeout = old_timeout; |
1049 | } |
1050 | #endif |
1051 | |
1052 | out: |
1053 | if ((start_now) && (rv == 0)) { |
1054 | /* Run from startup, so start the timer now. */ |
1055 | start_now = 0; /* Disable this function after first startup. */ |
1056 | ipmi_watchdog_state = action_val; |
1057 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
1058 | pr_info("Starting now!\n" ); |
1059 | } else { |
1060 | /* Stop the timer now. */ |
1061 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
1062 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
1063 | } |
1064 | } |
1065 | |
1066 | static void ipmi_unregister_watchdog(int ipmi_intf) |
1067 | { |
1068 | int rv; |
1069 | struct ipmi_user *loc_user = watchdog_user; |
1070 | |
1071 | if (!loc_user) |
1072 | return; |
1073 | |
1074 | if (watchdog_ifnum != ipmi_intf) |
1075 | return; |
1076 | |
1077 | /* Make sure no one can call us any more. */ |
1078 | misc_deregister(misc: &ipmi_wdog_miscdev); |
1079 | |
1080 | watchdog_user = NULL; |
1081 | |
1082 | /* |
1083 | * Wait to make sure the message makes it out. The lower layer has |
1084 | * pointers to our buffers, we want to make sure they are done before |
1085 | * we release our memory. |
1086 | */ |
1087 | while (atomic_read(v: &msg_tofree)) |
1088 | msg_free_smi(NULL); |
1089 | |
1090 | mutex_lock(&ipmi_watchdog_mutex); |
1091 | |
1092 | /* Disconnect from IPMI. */ |
1093 | rv = ipmi_destroy_user(user: loc_user); |
1094 | if (rv) |
1095 | pr_warn("error unlinking from IPMI: %d\n" , rv); |
1096 | |
1097 | /* If it comes back, restart it properly. */ |
1098 | ipmi_start_timer_on_heartbeat = 1; |
1099 | |
1100 | mutex_unlock(lock: &ipmi_watchdog_mutex); |
1101 | } |
1102 | |
1103 | #ifdef HAVE_DIE_NMI |
1104 | static int |
1105 | ipmi_nmi(unsigned int val, struct pt_regs *regs) |
1106 | { |
1107 | /* |
1108 | * If we get here, it's an NMI that's not a memory or I/O |
1109 | * error. We can't truly tell if it's from IPMI or not |
1110 | * without sending a message, and sending a message is almost |
1111 | * impossible because of locking. |
1112 | */ |
1113 | |
1114 | if (testing_nmi) { |
1115 | testing_nmi = 2; |
1116 | return NMI_HANDLED; |
1117 | } |
1118 | |
1119 | /* If we are not expecting a timeout, ignore it. */ |
1120 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
1121 | return NMI_DONE; |
1122 | |
1123 | if (preaction_val != WDOG_PRETIMEOUT_NMI) |
1124 | return NMI_DONE; |
1125 | |
1126 | /* |
1127 | * If no one else handled the NMI, we assume it was the IPMI |
1128 | * watchdog. |
1129 | */ |
1130 | if (preop_val == WDOG_PREOP_PANIC) { |
1131 | /* On some machines, the heartbeat will give |
1132 | an error and not work unless we re-enable |
1133 | the timer. So do so. */ |
1134 | atomic_set(v: &pretimeout_since_last_heartbeat, i: 1); |
1135 | if (atomic_inc_and_test(v: &preop_panic_excl)) |
1136 | nmi_panic(regs, msg: "pre-timeout" ); |
1137 | } |
1138 | |
1139 | return NMI_HANDLED; |
1140 | } |
1141 | #endif |
1142 | |
1143 | static int wdog_reboot_handler(struct notifier_block *this, |
1144 | unsigned long code, |
1145 | void *unused) |
1146 | { |
1147 | static int reboot_event_handled; |
1148 | |
1149 | if ((watchdog_user) && (!reboot_event_handled)) { |
1150 | /* Make sure we only do this once. */ |
1151 | reboot_event_handled = 1; |
1152 | |
1153 | if (code == SYS_POWER_OFF || code == SYS_HALT) { |
1154 | /* Disable the WDT if we are shutting down. */ |
1155 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
1156 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
1157 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
1158 | /* Set a long timer to let the reboot happen or |
1159 | reset if it hangs, but only if the watchdog |
1160 | timer was already running. */ |
1161 | if (timeout < 120) |
1162 | timeout = 120; |
1163 | pretimeout = 0; |
1164 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; |
1165 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
1166 | } |
1167 | } |
1168 | return NOTIFY_OK; |
1169 | } |
1170 | |
1171 | static struct notifier_block wdog_reboot_notifier = { |
1172 | .notifier_call = wdog_reboot_handler, |
1173 | .next = NULL, |
1174 | .priority = 0 |
1175 | }; |
1176 | |
1177 | static void ipmi_new_smi(int if_num, struct device *device) |
1178 | { |
1179 | ipmi_register_watchdog(ipmi_intf: if_num); |
1180 | } |
1181 | |
1182 | static void ipmi_smi_gone(int if_num) |
1183 | { |
1184 | ipmi_unregister_watchdog(ipmi_intf: if_num); |
1185 | } |
1186 | |
1187 | static struct ipmi_smi_watcher smi_watcher = { |
1188 | .owner = THIS_MODULE, |
1189 | .new_smi = ipmi_new_smi, |
1190 | .smi_gone = ipmi_smi_gone |
1191 | }; |
1192 | |
1193 | static int action_op(const char *inval, char *outval) |
1194 | { |
1195 | if (outval) |
1196 | strcpy(p: outval, q: action); |
1197 | |
1198 | if (!inval) |
1199 | return 0; |
1200 | |
1201 | if (strcmp(inval, "reset" ) == 0) |
1202 | action_val = WDOG_TIMEOUT_RESET; |
1203 | else if (strcmp(inval, "none" ) == 0) |
1204 | action_val = WDOG_TIMEOUT_NONE; |
1205 | else if (strcmp(inval, "power_cycle" ) == 0) |
1206 | action_val = WDOG_TIMEOUT_POWER_CYCLE; |
1207 | else if (strcmp(inval, "power_off" ) == 0) |
1208 | action_val = WDOG_TIMEOUT_POWER_DOWN; |
1209 | else |
1210 | return -EINVAL; |
1211 | strcpy(p: action, q: inval); |
1212 | return 0; |
1213 | } |
1214 | |
1215 | static int preaction_op(const char *inval, char *outval) |
1216 | { |
1217 | if (outval) |
1218 | strcpy(p: outval, q: preaction); |
1219 | |
1220 | if (!inval) |
1221 | return 0; |
1222 | |
1223 | if (strcmp(inval, "pre_none" ) == 0) |
1224 | preaction_val = WDOG_PRETIMEOUT_NONE; |
1225 | else if (strcmp(inval, "pre_smi" ) == 0) |
1226 | preaction_val = WDOG_PRETIMEOUT_SMI; |
1227 | #ifdef HAVE_DIE_NMI |
1228 | else if (strcmp(inval, "pre_nmi" ) == 0) |
1229 | preaction_val = WDOG_PRETIMEOUT_NMI; |
1230 | #endif |
1231 | else if (strcmp(inval, "pre_int" ) == 0) |
1232 | preaction_val = WDOG_PRETIMEOUT_MSG_INT; |
1233 | else |
1234 | return -EINVAL; |
1235 | strcpy(p: preaction, q: inval); |
1236 | return 0; |
1237 | } |
1238 | |
1239 | static int preop_op(const char *inval, char *outval) |
1240 | { |
1241 | if (outval) |
1242 | strcpy(p: outval, q: preop); |
1243 | |
1244 | if (!inval) |
1245 | return 0; |
1246 | |
1247 | if (strcmp(inval, "preop_none" ) == 0) |
1248 | preop_val = WDOG_PREOP_NONE; |
1249 | else if (strcmp(inval, "preop_panic" ) == 0) |
1250 | preop_val = WDOG_PREOP_PANIC; |
1251 | else if (strcmp(inval, "preop_give_data" ) == 0) |
1252 | preop_val = WDOG_PREOP_GIVE_DATA; |
1253 | else |
1254 | return -EINVAL; |
1255 | strcpy(p: preop, q: inval); |
1256 | return 0; |
1257 | } |
1258 | |
1259 | static void check_parms(void) |
1260 | { |
1261 | #ifdef HAVE_DIE_NMI |
1262 | int do_nmi = 0; |
1263 | int rv; |
1264 | |
1265 | if (preaction_val == WDOG_PRETIMEOUT_NMI) { |
1266 | do_nmi = 1; |
1267 | if (preop_val == WDOG_PREOP_GIVE_DATA) { |
1268 | pr_warn("Pretimeout op is to give data but NMI pretimeout is enabled, setting pretimeout op to none\n" ); |
1269 | preop_op(inval: "preop_none" , NULL); |
1270 | do_nmi = 0; |
1271 | } |
1272 | } |
1273 | if (do_nmi && !nmi_handler_registered) { |
1274 | rv = register_nmi_handler(NMI_UNKNOWN, ipmi_nmi, 0, |
1275 | "ipmi" ); |
1276 | if (rv) { |
1277 | pr_warn("Can't register nmi handler\n" ); |
1278 | return; |
1279 | } else |
1280 | nmi_handler_registered = 1; |
1281 | } else if (!do_nmi && nmi_handler_registered) { |
1282 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi" ); |
1283 | nmi_handler_registered = 0; |
1284 | } |
1285 | #endif |
1286 | } |
1287 | |
1288 | static int __init ipmi_wdog_init(void) |
1289 | { |
1290 | int rv; |
1291 | |
1292 | if (action_op(inval: action, NULL)) { |
1293 | action_op(inval: "reset" , NULL); |
1294 | pr_info("Unknown action '%s', defaulting to reset\n" , action); |
1295 | } |
1296 | |
1297 | if (preaction_op(inval: preaction, NULL)) { |
1298 | preaction_op(inval: "pre_none" , NULL); |
1299 | pr_info("Unknown preaction '%s', defaulting to none\n" , |
1300 | preaction); |
1301 | } |
1302 | |
1303 | if (preop_op(inval: preop, NULL)) { |
1304 | preop_op(inval: "preop_none" , NULL); |
1305 | pr_info("Unknown preop '%s', defaulting to none\n" , preop); |
1306 | } |
1307 | |
1308 | check_parms(); |
1309 | |
1310 | register_reboot_notifier(&wdog_reboot_notifier); |
1311 | |
1312 | rv = ipmi_smi_watcher_register(watcher: &smi_watcher); |
1313 | if (rv) { |
1314 | #ifdef HAVE_DIE_NMI |
1315 | if (nmi_handler_registered) |
1316 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi" ); |
1317 | #endif |
1318 | unregister_reboot_notifier(&wdog_reboot_notifier); |
1319 | pr_warn("can't register smi watcher\n" ); |
1320 | return rv; |
1321 | } |
1322 | |
1323 | pr_info("driver initialized\n" ); |
1324 | |
1325 | return 0; |
1326 | } |
1327 | |
1328 | static void __exit ipmi_wdog_exit(void) |
1329 | { |
1330 | ipmi_smi_watcher_unregister(watcher: &smi_watcher); |
1331 | ipmi_unregister_watchdog(ipmi_intf: watchdog_ifnum); |
1332 | |
1333 | #ifdef HAVE_DIE_NMI |
1334 | if (nmi_handler_registered) |
1335 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi" ); |
1336 | #endif |
1337 | |
1338 | unregister_reboot_notifier(&wdog_reboot_notifier); |
1339 | } |
1340 | module_exit(ipmi_wdog_exit); |
1341 | module_init(ipmi_wdog_init); |
1342 | MODULE_LICENSE("GPL" ); |
1343 | MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>" ); |
1344 | MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface." ); |
1345 | |