]>
Commit | Line | Data |
---|---|---|
243ac210 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
1da177e4 LT |
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. | |
1da177e4 LT |
12 | */ |
13 | ||
1da177e4 LT |
14 | #include <linux/module.h> |
15 | #include <linux/moduleparam.h> | |
16 | #include <linux/ipmi.h> | |
17 | #include <linux/ipmi_smi.h> | |
609146fd | 18 | #include <linux/mutex.h> |
1da177e4 LT |
19 | #include <linux/watchdog.h> |
20 | #include <linux/miscdevice.h> | |
21 | #include <linux/init.h> | |
d6dfd131 | 22 | #include <linux/completion.h> |
1eeb66a1 | 23 | #include <linux/kdebug.h> |
1da177e4 LT |
24 | #include <linux/rwsem.h> |
25 | #include <linux/errno.h> | |
7c0f6ba6 | 26 | #include <linux/uaccess.h> |
1da177e4 LT |
27 | #include <linux/notifier.h> |
28 | #include <linux/nmi.h> | |
29 | #include <linux/reboot.h> | |
30 | #include <linux/wait.h> | |
31 | #include <linux/poll.h> | |
cc4673ee CM |
32 | #include <linux/string.h> |
33 | #include <linux/ctype.h> | |
612b5a8d | 34 | #include <linux/delay.h> |
60063497 | 35 | #include <linux/atomic.h> |
174cd4b1 | 36 | #include <linux/sched/signal.h> |
f64da958 | 37 | |
612b5a8d | 38 | #ifdef CONFIG_X86 |
36c7dc44 CM |
39 | /* |
40 | * This is ugly, but I've determined that x86 is the only architecture | |
41 | * that can reasonably support the IPMI NMI watchdog timeout at this | |
42 | * time. If another architecture adds this capability somehow, it | |
43 | * will have to be a somewhat different mechanism and I have no idea | |
44 | * how it will work. So in the unlikely event that another | |
45 | * architecture supports this, we can figure out a good generic | |
46 | * mechanism for it at that time. | |
47 | */ | |
612b5a8d | 48 | #include <asm/kdebug.h> |
9c48f1c6 | 49 | #include <asm/nmi.h> |
612b5a8d | 50 | #define HAVE_DIE_NMI |
1da177e4 LT |
51 | #endif |
52 | ||
53 | #define PFX "IPMI Watchdog: " | |
54 | ||
1da177e4 LT |
55 | /* |
56 | * The IPMI command/response information for the watchdog timer. | |
57 | */ | |
58 | ||
59 | /* values for byte 1 of the set command, byte 2 of the get response. */ | |
60 | #define WDOG_DONT_LOG (1 << 7) | |
61 | #define WDOG_DONT_STOP_ON_SET (1 << 6) | |
62 | #define WDOG_SET_TIMER_USE(byte, use) \ | |
63 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
64 | #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) | |
65 | #define WDOG_TIMER_USE_BIOS_FRB2 1 | |
66 | #define WDOG_TIMER_USE_BIOS_POST 2 | |
67 | #define WDOG_TIMER_USE_OS_LOAD 3 | |
68 | #define WDOG_TIMER_USE_SMS_OS 4 | |
69 | #define WDOG_TIMER_USE_OEM 5 | |
70 | ||
71 | /* values for byte 2 of the set command, byte 3 of the get response. */ | |
72 | #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ | |
73 | byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) | |
74 | #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) | |
75 | #define WDOG_PRETIMEOUT_NONE 0 | |
76 | #define WDOG_PRETIMEOUT_SMI 1 | |
77 | #define WDOG_PRETIMEOUT_NMI 2 | |
78 | #define WDOG_PRETIMEOUT_MSG_INT 3 | |
79 | ||
80 | /* Operations that can be performed on a pretimout. */ | |
81 | #define WDOG_PREOP_NONE 0 | |
82 | #define WDOG_PREOP_PANIC 1 | |
36c7dc44 CM |
83 | /* Cause data to be available to read. Doesn't work in NMI mode. */ |
84 | #define WDOG_PREOP_GIVE_DATA 2 | |
1da177e4 LT |
85 | |
86 | /* Actions to perform on a full timeout. */ | |
87 | #define WDOG_SET_TIMEOUT_ACT(byte, use) \ | |
88 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
89 | #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) | |
90 | #define WDOG_TIMEOUT_NONE 0 | |
91 | #define WDOG_TIMEOUT_RESET 1 | |
92 | #define WDOG_TIMEOUT_POWER_DOWN 2 | |
93 | #define WDOG_TIMEOUT_POWER_CYCLE 3 | |
94 | ||
36c7dc44 CM |
95 | /* |
96 | * Byte 3 of the get command, byte 4 of the get response is the | |
97 | * pre-timeout in seconds. | |
98 | */ | |
1da177e4 LT |
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 | ||
36c7dc44 CM |
107 | /* |
108 | * Setting/getting the watchdog timer value. This is for bytes 5 and | |
109 | * 6 (the timeout time) of the set command, and bytes 6 and 7 (the | |
110 | * timeout time) and 8 and 9 (the current countdown value) of the | |
111 | * response. The timeout value is given in seconds (in the command it | |
112 | * is 100ms intervals). | |
113 | */ | |
1da177e4 LT |
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 | ||
b75d91f7 CM |
123 | #define IPMI_WDOG_TIMER_NOT_INIT_RESP 0x80 |
124 | ||
609146fd | 125 | static DEFINE_MUTEX(ipmi_watchdog_mutex); |
86a1e189 | 126 | static bool nowayout = WATCHDOG_NOWAYOUT; |
1da177e4 | 127 | |
2911c988 | 128 | static struct ipmi_user *watchdog_user; |
b2c03941 | 129 | static int watchdog_ifnum; |
1da177e4 LT |
130 | |
131 | /* Default the timeout to 10 seconds. */ | |
132 | static int timeout = 10; | |
133 | ||
134 | /* The pre-timeout is disabled by default. */ | |
0c8204b3 | 135 | static int pretimeout; |
1da177e4 | 136 | |
c7f42c63 JYF |
137 | /* Default timeout to set on panic */ |
138 | static int panic_wdt_timeout = 255; | |
139 | ||
1da177e4 LT |
140 | /* Default action is to reset the board on a timeout. */ |
141 | static unsigned char action_val = WDOG_TIMEOUT_RESET; | |
142 | ||
143 | static char action[16] = "reset"; | |
144 | ||
145 | static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; | |
146 | ||
147 | static char preaction[16] = "pre_none"; | |
148 | ||
149 | static unsigned char preop_val = WDOG_PREOP_NONE; | |
150 | ||
151 | static char preop[16] = "preop_none"; | |
152 | static DEFINE_SPINLOCK(ipmi_read_lock); | |
0c8204b3 | 153 | static char data_to_read; |
1da177e4 | 154 | static DECLARE_WAIT_QUEUE_HEAD(read_q); |
0c8204b3 | 155 | static struct fasync_struct *fasync_q; |
d1b29b97 | 156 | static atomic_t pretimeout_since_last_heartbeat; |
1da177e4 LT |
157 | static char expect_close; |
158 | ||
b2c03941 CM |
159 | static int ifnum_to_use = -1; |
160 | ||
cc4673ee CM |
161 | /* Parameters to ipmi_set_timeout */ |
162 | #define IPMI_SET_TIMEOUT_NO_HB 0 | |
163 | #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 | |
164 | #define IPMI_SET_TIMEOUT_FORCE_HB 2 | |
165 | ||
166 | static int ipmi_set_timeout(int do_heartbeat); | |
b2c03941 CM |
167 | static void ipmi_register_watchdog(int ipmi_intf); |
168 | static void ipmi_unregister_watchdog(int ipmi_intf); | |
cc4673ee | 169 | |
36c7dc44 CM |
170 | /* |
171 | * If true, the driver will start running as soon as it is configured | |
172 | * and ready. | |
173 | */ | |
0c8204b3 | 174 | static int start_now; |
1da177e4 | 175 | |
c8ba6c52 | 176 | static int set_param_timeout(const char *val, const struct kernel_param *kp) |
cc4673ee CM |
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 | ||
cc4673ee CM |
188 | *((int *)kp->arg) = l; |
189 | if (watchdog_user) | |
190 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
cc4673ee CM |
191 | |
192 | return rv; | |
193 | } | |
194 | ||
9c27847d | 195 | static const struct kernel_param_ops param_ops_timeout = { |
c8ba6c52 RR |
196 | .set = set_param_timeout, |
197 | .get = param_get_int, | |
198 | }; | |
199 | #define param_check_timeout param_check_int | |
cc4673ee CM |
200 | |
201 | typedef int (*action_fn)(const char *intval, char *outval); | |
202 | ||
203 | static int action_op(const char *inval, char *outval); | |
204 | static int preaction_op(const char *inval, char *outval); | |
205 | static int preop_op(const char *inval, char *outval); | |
206 | static void check_parms(void); | |
207 | ||
c8ba6c52 | 208 | static int set_param_str(const char *val, const struct kernel_param *kp) |
cc4673ee CM |
209 | { |
210 | action_fn fn = (action_fn) kp->arg; | |
211 | int rv = 0; | |
43cdff92 SD |
212 | char valcp[16]; |
213 | char *s; | |
66f969d0 | 214 | |
1b4254ce | 215 | strncpy(valcp, val, 15); |
43cdff92 | 216 | valcp[15] = '\0'; |
66f969d0 | 217 | |
43cdff92 | 218 | s = strstrip(valcp); |
cc4673ee | 219 | |
66f969d0 | 220 | rv = fn(s, NULL); |
cc4673ee | 221 | if (rv) |
f8fbcd3b | 222 | goto out; |
cc4673ee CM |
223 | |
224 | check_parms(); | |
225 | if (watchdog_user) | |
226 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
227 | ||
f8fbcd3b | 228 | out: |
cc4673ee CM |
229 | return rv; |
230 | } | |
231 | ||
c8ba6c52 | 232 | static int get_param_str(char *buffer, const struct kernel_param *kp) |
cc4673ee CM |
233 | { |
234 | action_fn fn = (action_fn) kp->arg; | |
235 | int rv; | |
236 | ||
237 | rv = fn(NULL, buffer); | |
238 | if (rv) | |
239 | return rv; | |
240 | return strlen(buffer); | |
241 | } | |
242 | ||
b2c03941 | 243 | |
c8ba6c52 | 244 | static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp) |
b2c03941 CM |
245 | { |
246 | int rv = param_set_int(val, kp); | |
247 | if (rv) | |
248 | return rv; | |
249 | if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum)) | |
250 | return 0; | |
251 | ||
252 | ipmi_unregister_watchdog(watchdog_ifnum); | |
253 | ipmi_register_watchdog(ifnum_to_use); | |
254 | return 0; | |
255 | } | |
256 | ||
9c27847d | 257 | static const struct kernel_param_ops param_ops_wdog_ifnum = { |
c8ba6c52 RR |
258 | .set = set_param_wdog_ifnum, |
259 | .get = param_get_int, | |
260 | }; | |
261 | ||
262 | #define param_check_wdog_ifnum param_check_int | |
263 | ||
9c27847d | 264 | static const struct kernel_param_ops param_ops_str = { |
c8ba6c52 RR |
265 | .set = set_param_str, |
266 | .get = get_param_str, | |
267 | }; | |
268 | ||
269 | module_param(ifnum_to_use, wdog_ifnum, 0644); | |
b2c03941 CM |
270 | MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog " |
271 | "timer. Setting to -1 defaults to the first registered " | |
272 | "interface"); | |
273 | ||
c8ba6c52 | 274 | module_param(timeout, timeout, 0644); |
1da177e4 | 275 | MODULE_PARM_DESC(timeout, "Timeout value in seconds."); |
cc4673ee | 276 | |
c8ba6c52 | 277 | module_param(pretimeout, timeout, 0644); |
1da177e4 | 278 | MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); |
cc4673ee | 279 | |
c7f42c63 | 280 | module_param(panic_wdt_timeout, timeout, 0644); |
7195a5a6 | 281 | MODULE_PARM_DESC(panic_wdt_timeout, "Timeout value on kernel panic in seconds."); |
c7f42c63 | 282 | |
c8ba6c52 | 283 | module_param_cb(action, ¶m_ops_str, action_op, 0644); |
1da177e4 LT |
284 | MODULE_PARM_DESC(action, "Timeout action. One of: " |
285 | "reset, none, power_cycle, power_off."); | |
cc4673ee | 286 | |
c8ba6c52 | 287 | module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644); |
1da177e4 LT |
288 | MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " |
289 | "pre_none, pre_smi, pre_nmi, pre_int."); | |
cc4673ee | 290 | |
c8ba6c52 | 291 | module_param_cb(preop, ¶m_ops_str, preop_op, 0644); |
1da177e4 LT |
292 | MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " |
293 | "preop_none, preop_panic, preop_give_data."); | |
cc4673ee | 294 | |
b2c03941 | 295 | module_param(start_now, int, 0444); |
1da177e4 LT |
296 | MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" |
297 | "soon as the driver is loaded."); | |
cc4673ee | 298 | |
86a1e189 | 299 | module_param(nowayout, bool, 0644); |
b2c03941 CM |
300 | MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started " |
301 | "(default=CONFIG_WATCHDOG_NOWAYOUT)"); | |
1da177e4 LT |
302 | |
303 | /* Default state of the timer. */ | |
304 | static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
305 | ||
1da177e4 | 306 | /* Is someone using the watchdog? Only one user is allowed. */ |
0c8204b3 | 307 | static unsigned long ipmi_wdog_open; |
1da177e4 | 308 | |
36c7dc44 CM |
309 | /* |
310 | * If set to 1, the heartbeat command will set the state to reset and | |
311 | * start the timer. The timer doesn't normally run when the driver is | |
312 | * first opened until the heartbeat is set the first time, this | |
313 | * variable is used to accomplish this. | |
314 | */ | |
0c8204b3 | 315 | static int ipmi_start_timer_on_heartbeat; |
1da177e4 LT |
316 | |
317 | /* IPMI version of the BMC. */ | |
318 | static unsigned char ipmi_version_major; | |
319 | static unsigned char ipmi_version_minor; | |
320 | ||
b385676b CM |
321 | /* If a pretimeout occurs, this is used to allow only one panic to happen. */ |
322 | static atomic_t preop_panic_excl = ATOMIC_INIT(-1); | |
1da177e4 | 323 | |
612b5a8d CM |
324 | #ifdef HAVE_DIE_NMI |
325 | static int testing_nmi; | |
326 | static int nmi_handler_registered; | |
327 | #endif | |
328 | ||
d1b29b97 | 329 | static int __ipmi_heartbeat(void); |
1da177e4 | 330 | |
36c7dc44 | 331 | /* |
d1b29b97 CM |
332 | * We use a mutex to make sure that only one thing can send a set a |
333 | * message at one time. The mutex is claimed when a message is sent | |
334 | * and freed when both the send and receive messages are free. | |
36c7dc44 | 335 | */ |
d1b29b97 CM |
336 | static atomic_t msg_tofree = ATOMIC_INIT(0); |
337 | static DECLARE_COMPLETION(msg_wait); | |
338 | static void msg_free_smi(struct ipmi_smi_msg *msg) | |
1da177e4 | 339 | { |
d1b29b97 CM |
340 | if (atomic_dec_and_test(&msg_tofree)) |
341 | complete(&msg_wait); | |
1da177e4 | 342 | } |
d1b29b97 | 343 | static void msg_free_recv(struct ipmi_recv_msg *msg) |
1da177e4 | 344 | { |
d1b29b97 CM |
345 | if (atomic_dec_and_test(&msg_tofree)) |
346 | complete(&msg_wait); | |
1da177e4 | 347 | } |
d1b29b97 CM |
348 | static struct ipmi_smi_msg smi_msg = { |
349 | .done = msg_free_smi | |
1da177e4 | 350 | }; |
d1b29b97 CM |
351 | static struct ipmi_recv_msg recv_msg = { |
352 | .done = msg_free_recv | |
1da177e4 | 353 | }; |
36c7dc44 | 354 | |
d1b29b97 | 355 | static int __ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, |
1da177e4 LT |
356 | struct ipmi_recv_msg *recv_msg, |
357 | int *send_heartbeat_now) | |
358 | { | |
359 | struct kernel_ipmi_msg msg; | |
360 | unsigned char data[6]; | |
361 | int rv; | |
362 | struct ipmi_system_interface_addr addr; | |
363 | int hbnow = 0; | |
364 | ||
365 | ||
366 | data[0] = 0; | |
367 | WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); | |
368 | ||
369 | if ((ipmi_version_major > 1) | |
36c7dc44 | 370 | || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) { |
1da177e4 LT |
371 | /* This is an IPMI 1.5-only feature. */ |
372 | data[0] |= WDOG_DONT_STOP_ON_SET; | |
373 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
36c7dc44 CM |
374 | /* |
375 | * In ipmi 1.0, setting the timer stops the watchdog, we | |
376 | * need to start it back up again. | |
377 | */ | |
1da177e4 LT |
378 | hbnow = 1; |
379 | } | |
380 | ||
381 | data[1] = 0; | |
382 | WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); | |
8f05ee9a | 383 | if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { |
1da177e4 LT |
384 | WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); |
385 | data[2] = pretimeout; | |
386 | } else { | |
387 | WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); | |
388 | data[2] = 0; /* No pretimeout. */ | |
389 | } | |
390 | data[3] = 0; | |
391 | WDOG_SET_TIMEOUT(data[4], data[5], timeout); | |
392 | ||
393 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
394 | addr.channel = IPMI_BMC_CHANNEL; | |
395 | addr.lun = 0; | |
396 | ||
397 | msg.netfn = 0x06; | |
398 | msg.cmd = IPMI_WDOG_SET_TIMER; | |
399 | msg.data = data; | |
400 | msg.data_len = sizeof(data); | |
401 | rv = ipmi_request_supply_msgs(watchdog_user, | |
402 | (struct ipmi_addr *) &addr, | |
403 | 0, | |
404 | &msg, | |
405 | NULL, | |
406 | smi_msg, | |
407 | recv_msg, | |
408 | 1); | |
d1b29b97 CM |
409 | if (rv) |
410 | pr_warn(PFX "set timeout error: %d\n", rv); | |
411 | else if (send_heartbeat_now) | |
412 | *send_heartbeat_now = hbnow; | |
1da177e4 LT |
413 | |
414 | return rv; | |
415 | } | |
416 | ||
d1b29b97 | 417 | static int _ipmi_set_timeout(int do_heartbeat) |
1da177e4 LT |
418 | { |
419 | int send_heartbeat_now; | |
420 | int rv; | |
421 | ||
d1b29b97 CM |
422 | if (!watchdog_user) |
423 | return -ENODEV; | |
1da177e4 | 424 | |
d1b29b97 | 425 | atomic_set(&msg_tofree, 2); |
1da177e4 | 426 | |
d1b29b97 CM |
427 | rv = __ipmi_set_timeout(&smi_msg, |
428 | &recv_msg, | |
1da177e4 | 429 | &send_heartbeat_now); |
d1b29b97 CM |
430 | if (rv) |
431 | return rv; | |
d6dfd131 | 432 | |
d1b29b97 | 433 | wait_for_completion(&msg_wait); |
612b5a8d | 434 | |
d6dfd131 | 435 | if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) |
d1b29b97 CM |
436 | || ((send_heartbeat_now) |
437 | && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) | |
438 | rv = __ipmi_heartbeat(); | |
439 | ||
440 | return rv; | |
441 | } | |
442 | ||
443 | static int ipmi_set_timeout(int do_heartbeat) | |
444 | { | |
445 | int rv; | |
446 | ||
447 | mutex_lock(&ipmi_watchdog_mutex); | |
448 | rv = _ipmi_set_timeout(do_heartbeat); | |
449 | mutex_unlock(&ipmi_watchdog_mutex); | |
1da177e4 LT |
450 | |
451 | return rv; | |
452 | } | |
453 | ||
fcfa4724 CM |
454 | static atomic_t panic_done_count = ATOMIC_INIT(0); |
455 | ||
456 | static void panic_smi_free(struct ipmi_smi_msg *msg) | |
1da177e4 | 457 | { |
fcfa4724 | 458 | atomic_dec(&panic_done_count); |
1da177e4 | 459 | } |
fcfa4724 CM |
460 | static void panic_recv_free(struct ipmi_recv_msg *msg) |
461 | { | |
462 | atomic_dec(&panic_done_count); | |
463 | } | |
464 | ||
36c7dc44 | 465 | static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = { |
fcfa4724 CM |
466 | .done = panic_smi_free |
467 | }; | |
36c7dc44 | 468 | static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = { |
fcfa4724 CM |
469 | .done = panic_recv_free |
470 | }; | |
471 | ||
472 | static void panic_halt_ipmi_heartbeat(void) | |
473 | { | |
474 | struct kernel_ipmi_msg msg; | |
475 | struct ipmi_system_interface_addr addr; | |
476 | int rv; | |
477 | ||
36c7dc44 CM |
478 | /* |
479 | * Don't reset the timer if we have the timer turned off, that | |
480 | * re-enables the watchdog. | |
481 | */ | |
fcfa4724 CM |
482 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
483 | return; | |
484 | ||
485 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
486 | addr.channel = IPMI_BMC_CHANNEL; | |
487 | addr.lun = 0; | |
488 | ||
489 | msg.netfn = 0x06; | |
490 | msg.cmd = IPMI_WDOG_RESET_TIMER; | |
491 | msg.data = NULL; | |
492 | msg.data_len = 0; | |
2c1175c2 | 493 | atomic_add(1, &panic_done_count); |
fcfa4724 CM |
494 | rv = ipmi_request_supply_msgs(watchdog_user, |
495 | (struct ipmi_addr *) &addr, | |
496 | 0, | |
497 | &msg, | |
498 | NULL, | |
499 | &panic_halt_heartbeat_smi_msg, | |
500 | &panic_halt_heartbeat_recv_msg, | |
501 | 1); | |
895dcfd1 | 502 | if (rv) |
2c1175c2 | 503 | atomic_sub(1, &panic_done_count); |
1da177e4 | 504 | } |
fcfa4724 | 505 | |
36c7dc44 | 506 | static struct ipmi_smi_msg panic_halt_smi_msg = { |
fcfa4724 | 507 | .done = panic_smi_free |
1da177e4 | 508 | }; |
36c7dc44 | 509 | static struct ipmi_recv_msg panic_halt_recv_msg = { |
fcfa4724 | 510 | .done = panic_recv_free |
1da177e4 LT |
511 | }; |
512 | ||
36c7dc44 CM |
513 | /* |
514 | * Special call, doesn't claim any locks. This is only to be called | |
515 | * at panic or halt time, in run-to-completion mode, when the caller | |
516 | * is the only CPU and the only thing that will be going is these IPMI | |
517 | * calls. | |
518 | */ | |
1da177e4 LT |
519 | static void panic_halt_ipmi_set_timeout(void) |
520 | { | |
521 | int send_heartbeat_now; | |
522 | int rv; | |
523 | ||
fcfa4724 CM |
524 | /* Wait for the messages to be free. */ |
525 | while (atomic_read(&panic_done_count) != 0) | |
526 | ipmi_poll_interface(watchdog_user); | |
2c1175c2 | 527 | atomic_add(1, &panic_done_count); |
d1b29b97 | 528 | rv = __ipmi_set_timeout(&panic_halt_smi_msg, |
1da177e4 LT |
529 | &panic_halt_recv_msg, |
530 | &send_heartbeat_now); | |
895dcfd1 | 531 | if (rv) { |
2c1175c2 | 532 | atomic_sub(1, &panic_done_count); |
e2384917 | 533 | pr_warn(PFX "Unable to extend the watchdog timeout."); |
895dcfd1 CM |
534 | } else { |
535 | if (send_heartbeat_now) | |
536 | panic_halt_ipmi_heartbeat(); | |
537 | } | |
fcfa4724 CM |
538 | while (atomic_read(&panic_done_count) != 0) |
539 | ipmi_poll_interface(watchdog_user); | |
1da177e4 LT |
540 | } |
541 | ||
d1b29b97 | 542 | static int __ipmi_heartbeat(void) |
1da177e4 | 543 | { |
d1b29b97 CM |
544 | struct kernel_ipmi_msg msg; |
545 | int rv; | |
1da177e4 | 546 | struct ipmi_system_interface_addr addr; |
d1b29b97 | 547 | int timeout_retries = 0; |
1da177e4 | 548 | |
b75d91f7 | 549 | restart: |
36c7dc44 CM |
550 | /* |
551 | * Don't reset the timer if we have the timer turned off, that | |
552 | * re-enables the watchdog. | |
553 | */ | |
d1b29b97 | 554 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
1da177e4 | 555 | return 0; |
d1b29b97 CM |
556 | |
557 | atomic_set(&msg_tofree, 2); | |
1da177e4 LT |
558 | |
559 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
560 | addr.channel = IPMI_BMC_CHANNEL; | |
561 | addr.lun = 0; | |
562 | ||
563 | msg.netfn = 0x06; | |
564 | msg.cmd = IPMI_WDOG_RESET_TIMER; | |
565 | msg.data = NULL; | |
566 | msg.data_len = 0; | |
567 | rv = ipmi_request_supply_msgs(watchdog_user, | |
568 | (struct ipmi_addr *) &addr, | |
569 | 0, | |
570 | &msg, | |
571 | NULL, | |
d1b29b97 CM |
572 | &smi_msg, |
573 | &recv_msg, | |
1da177e4 LT |
574 | 1); |
575 | if (rv) { | |
d1b29b97 | 576 | pr_warn(PFX "heartbeat send failure: %d\n", rv); |
1da177e4 LT |
577 | return rv; |
578 | } | |
579 | ||
580 | /* Wait for the heartbeat to be sent. */ | |
d1b29b97 | 581 | wait_for_completion(&msg_wait); |
1da177e4 | 582 | |
d1b29b97 | 583 | if (recv_msg.msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) { |
b75d91f7 CM |
584 | timeout_retries++; |
585 | if (timeout_retries > 3) { | |
d1b29b97 | 586 | pr_err(PFX ": Unable to restore the IPMI watchdog's settings, giving up.\n"); |
b75d91f7 | 587 | rv = -EIO; |
d1b29b97 | 588 | goto out; |
b75d91f7 CM |
589 | } |
590 | ||
591 | /* | |
592 | * The timer was not initialized, that means the BMC was | |
593 | * probably reset and lost the watchdog information. Attempt | |
594 | * to restore the timer's info. Note that we still hold | |
595 | * the heartbeat lock, to keep a heartbeat from happening | |
596 | * in this process, so must say no heartbeat to avoid a | |
d1b29b97 | 597 | * deadlock on this mutex |
b75d91f7 | 598 | */ |
d1b29b97 | 599 | rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
b75d91f7 | 600 | if (rv) { |
d1b29b97 CM |
601 | pr_err(PFX ": Unable to send the command to set the watchdog's settings, giving up.\n"); |
602 | goto out; | |
b75d91f7 CM |
603 | } |
604 | ||
d1b29b97 | 605 | /* Might need a heartbeat send, go ahead and do it. */ |
b75d91f7 | 606 | goto restart; |
d1b29b97 | 607 | } else if (recv_msg.msg.data[0] != 0) { |
36c7dc44 CM |
608 | /* |
609 | * Got an error in the heartbeat response. It was already | |
610 | * reported in ipmi_wdog_msg_handler, but we should return | |
611 | * an error here. | |
612 | */ | |
613 | rv = -EINVAL; | |
1da177e4 LT |
614 | } |
615 | ||
d1b29b97 CM |
616 | out: |
617 | return rv; | |
618 | } | |
619 | ||
620 | static int _ipmi_heartbeat(void) | |
621 | { | |
622 | int rv; | |
623 | ||
624 | if (!watchdog_user) | |
625 | return -ENODEV; | |
626 | ||
627 | if (ipmi_start_timer_on_heartbeat) { | |
628 | ipmi_start_timer_on_heartbeat = 0; | |
629 | ipmi_watchdog_state = action_val; | |
630 | rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
631 | } else if (atomic_cmpxchg(&pretimeout_since_last_heartbeat, 1, 0)) { | |
632 | /* | |
633 | * A pretimeout occurred, make sure we set the timeout. | |
634 | * We don't want to set the action, though, we want to | |
635 | * leave that alone (thus it can't be combined with the | |
636 | * above operation. | |
637 | */ | |
638 | rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
639 | } else { | |
640 | rv = __ipmi_heartbeat(); | |
641 | } | |
642 | ||
643 | return rv; | |
644 | } | |
645 | ||
646 | static int ipmi_heartbeat(void) | |
647 | { | |
648 | int rv; | |
649 | ||
650 | mutex_lock(&ipmi_watchdog_mutex); | |
651 | rv = _ipmi_heartbeat(); | |
652 | mutex_unlock(&ipmi_watchdog_mutex); | |
1da177e4 LT |
653 | |
654 | return rv; | |
655 | } | |
656 | ||
36c7dc44 | 657 | static struct watchdog_info ident = { |
1da177e4 LT |
658 | .options = 0, /* WDIOF_SETTIMEOUT, */ |
659 | .firmware_version = 1, | |
660 | .identity = "IPMI" | |
661 | }; | |
662 | ||
55929332 | 663 | static int ipmi_ioctl(struct file *file, |
1da177e4 LT |
664 | unsigned int cmd, unsigned long arg) |
665 | { | |
666 | void __user *argp = (void __user *)arg; | |
667 | int i; | |
668 | int val; | |
669 | ||
36c7dc44 | 670 | switch (cmd) { |
1da177e4 LT |
671 | case WDIOC_GETSUPPORT: |
672 | i = copy_to_user(argp, &ident, sizeof(ident)); | |
673 | return i ? -EFAULT : 0; | |
674 | ||
675 | case WDIOC_SETTIMEOUT: | |
676 | i = copy_from_user(&val, argp, sizeof(int)); | |
677 | if (i) | |
678 | return -EFAULT; | |
679 | timeout = val; | |
d1b29b97 | 680 | return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
1da177e4 LT |
681 | |
682 | case WDIOC_GETTIMEOUT: | |
683 | i = copy_to_user(argp, &timeout, sizeof(timeout)); | |
684 | if (i) | |
685 | return -EFAULT; | |
686 | return 0; | |
687 | ||
783e6bcd | 688 | case WDIOC_SETPRETIMEOUT: |
1da177e4 LT |
689 | i = copy_from_user(&val, argp, sizeof(int)); |
690 | if (i) | |
691 | return -EFAULT; | |
692 | pretimeout = val; | |
d1b29b97 | 693 | return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
1da177e4 | 694 | |
783e6bcd | 695 | case WDIOC_GETPRETIMEOUT: |
1da177e4 LT |
696 | i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); |
697 | if (i) | |
698 | return -EFAULT; | |
699 | return 0; | |
700 | ||
701 | case WDIOC_KEEPALIVE: | |
d1b29b97 | 702 | return _ipmi_heartbeat(); |
1da177e4 LT |
703 | |
704 | case WDIOC_SETOPTIONS: | |
705 | i = copy_from_user(&val, argp, sizeof(int)); | |
706 | if (i) | |
707 | return -EFAULT; | |
36c7dc44 | 708 | if (val & WDIOS_DISABLECARD) { |
1da177e4 | 709 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
d1b29b97 | 710 | _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
1da177e4 LT |
711 | ipmi_start_timer_on_heartbeat = 0; |
712 | } | |
713 | ||
36c7dc44 | 714 | if (val & WDIOS_ENABLECARD) { |
1da177e4 | 715 | ipmi_watchdog_state = action_val; |
d1b29b97 | 716 | _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); |
1da177e4 LT |
717 | } |
718 | return 0; | |
719 | ||
720 | case WDIOC_GETSTATUS: | |
721 | val = 0; | |
722 | i = copy_to_user(argp, &val, sizeof(val)); | |
723 | if (i) | |
724 | return -EFAULT; | |
725 | return 0; | |
726 | ||
727 | default: | |
728 | return -ENOIOCTLCMD; | |
729 | } | |
730 | } | |
731 | ||
55929332 AB |
732 | static long ipmi_unlocked_ioctl(struct file *file, |
733 | unsigned int cmd, | |
734 | unsigned long arg) | |
735 | { | |
736 | int ret; | |
737 | ||
609146fd | 738 | mutex_lock(&ipmi_watchdog_mutex); |
55929332 | 739 | ret = ipmi_ioctl(file, cmd, arg); |
609146fd | 740 | mutex_unlock(&ipmi_watchdog_mutex); |
55929332 AB |
741 | |
742 | return ret; | |
743 | } | |
744 | ||
1da177e4 LT |
745 | static ssize_t ipmi_write(struct file *file, |
746 | const char __user *buf, | |
747 | size_t len, | |
748 | loff_t *ppos) | |
749 | { | |
750 | int rv; | |
751 | ||
752 | if (len) { | |
36c7dc44 CM |
753 | if (!nowayout) { |
754 | size_t i; | |
1da177e4 LT |
755 | |
756 | /* In case it was set long ago */ | |
757 | expect_close = 0; | |
758 | ||
36c7dc44 | 759 | for (i = 0; i != len; i++) { |
1da177e4 LT |
760 | char c; |
761 | ||
762 | if (get_user(c, buf + i)) | |
763 | return -EFAULT; | |
764 | if (c == 'V') | |
765 | expect_close = 42; | |
766 | } | |
767 | } | |
768 | rv = ipmi_heartbeat(); | |
769 | if (rv) | |
770 | return rv; | |
1da177e4 | 771 | } |
3976df9b | 772 | return len; |
1da177e4 LT |
773 | } |
774 | ||
775 | static ssize_t ipmi_read(struct file *file, | |
776 | char __user *buf, | |
777 | size_t count, | |
778 | loff_t *ppos) | |
779 | { | |
780 | int rv = 0; | |
ac6424b9 | 781 | wait_queue_entry_t wait; |
1da177e4 LT |
782 | |
783 | if (count <= 0) | |
784 | return 0; | |
785 | ||
36c7dc44 CM |
786 | /* |
787 | * Reading returns if the pretimeout has gone off, and it only does | |
788 | * it once per pretimeout. | |
789 | */ | |
d1b29b97 | 790 | spin_lock_irq(&ipmi_read_lock); |
1da177e4 LT |
791 | if (!data_to_read) { |
792 | if (file->f_flags & O_NONBLOCK) { | |
793 | rv = -EAGAIN; | |
794 | goto out; | |
795 | } | |
36c7dc44 | 796 | |
1da177e4 LT |
797 | init_waitqueue_entry(&wait, current); |
798 | add_wait_queue(&read_q, &wait); | |
799 | while (!data_to_read) { | |
800 | set_current_state(TASK_INTERRUPTIBLE); | |
d1b29b97 | 801 | spin_unlock_irq(&ipmi_read_lock); |
1da177e4 | 802 | schedule(); |
d1b29b97 | 803 | spin_lock_irq(&ipmi_read_lock); |
1da177e4 LT |
804 | } |
805 | remove_wait_queue(&read_q, &wait); | |
36c7dc44 | 806 | |
1da177e4 LT |
807 | if (signal_pending(current)) { |
808 | rv = -ERESTARTSYS; | |
809 | goto out; | |
810 | } | |
811 | } | |
812 | data_to_read = 0; | |
813 | ||
814 | out: | |
d1b29b97 | 815 | spin_unlock_irq(&ipmi_read_lock); |
1da177e4 LT |
816 | |
817 | if (rv == 0) { | |
818 | if (copy_to_user(buf, &data_to_read, 1)) | |
819 | rv = -EFAULT; | |
820 | else | |
821 | rv = 1; | |
822 | } | |
823 | ||
824 | return rv; | |
825 | } | |
826 | ||
827 | static int ipmi_open(struct inode *ino, struct file *filep) | |
828 | { | |
36c7dc44 CM |
829 | switch (iminor(ino)) { |
830 | case WATCHDOG_MINOR: | |
e8b33617 | 831 | if (test_and_set_bit(0, &ipmi_wdog_open)) |
36c7dc44 | 832 | return -EBUSY; |
1da177e4 | 833 | |
af96f010 | 834 | |
36c7dc44 CM |
835 | /* |
836 | * Don't start the timer now, let it start on the | |
837 | * first heartbeat. | |
838 | */ | |
e8b33617 CM |
839 | ipmi_start_timer_on_heartbeat = 1; |
840 | return nonseekable_open(ino, filep); | |
1da177e4 | 841 | |
e8b33617 CM |
842 | default: |
843 | return (-ENODEV); | |
36c7dc44 | 844 | } |
1da177e4 LT |
845 | } |
846 | ||
afc9a42b | 847 | static __poll_t ipmi_poll(struct file *file, poll_table *wait) |
1da177e4 | 848 | { |
afc9a42b | 849 | __poll_t mask = 0; |
36c7dc44 | 850 | |
1da177e4 LT |
851 | poll_wait(file, &read_q, wait); |
852 | ||
d1b29b97 | 853 | spin_lock_irq(&ipmi_read_lock); |
1da177e4 | 854 | if (data_to_read) |
a9a08845 | 855 | mask |= (EPOLLIN | EPOLLRDNORM); |
d1b29b97 | 856 | spin_unlock_irq(&ipmi_read_lock); |
1da177e4 LT |
857 | |
858 | return mask; | |
859 | } | |
860 | ||
861 | static int ipmi_fasync(int fd, struct file *file, int on) | |
862 | { | |
863 | int result; | |
864 | ||
865 | result = fasync_helper(fd, file, on, &fasync_q); | |
866 | ||
867 | return (result); | |
868 | } | |
869 | ||
870 | static int ipmi_close(struct inode *ino, struct file *filep) | |
871 | { | |
8a3628d5 | 872 | if (iminor(ino) == WATCHDOG_MINOR) { |
1da177e4 | 873 | if (expect_close == 42) { |
d1b29b97 | 874 | mutex_lock(&ipmi_watchdog_mutex); |
1da177e4 | 875 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
d1b29b97 CM |
876 | _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
877 | mutex_unlock(&ipmi_watchdog_mutex); | |
1da177e4 | 878 | } else { |
e2384917 CM |
879 | pr_crit(PFX |
880 | "Unexpected close, not stopping watchdog!\n"); | |
1da177e4 LT |
881 | ipmi_heartbeat(); |
882 | } | |
ec26d79f | 883 | clear_bit(0, &ipmi_wdog_open); |
1da177e4 LT |
884 | } |
885 | ||
1da177e4 LT |
886 | expect_close = 0; |
887 | ||
888 | return 0; | |
889 | } | |
890 | ||
62322d25 | 891 | static const struct file_operations ipmi_wdog_fops = { |
1da177e4 LT |
892 | .owner = THIS_MODULE, |
893 | .read = ipmi_read, | |
894 | .poll = ipmi_poll, | |
895 | .write = ipmi_write, | |
55929332 | 896 | .unlocked_ioctl = ipmi_unlocked_ioctl, |
1da177e4 LT |
897 | .open = ipmi_open, |
898 | .release = ipmi_close, | |
899 | .fasync = ipmi_fasync, | |
6038f373 | 900 | .llseek = no_llseek, |
1da177e4 LT |
901 | }; |
902 | ||
903 | static struct miscdevice ipmi_wdog_miscdev = { | |
904 | .minor = WATCHDOG_MINOR, | |
905 | .name = "watchdog", | |
906 | .fops = &ipmi_wdog_fops | |
907 | }; | |
908 | ||
1da177e4 LT |
909 | static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, |
910 | void *handler_data) | |
911 | { | |
b75d91f7 CM |
912 | if (msg->msg.cmd == IPMI_WDOG_RESET_TIMER && |
913 | msg->msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) | |
e2384917 | 914 | pr_info(PFX "response: The IPMI controller appears to have been reset, will attempt to reinitialize the watchdog timer\n"); |
b75d91f7 | 915 | else if (msg->msg.data[0] != 0) |
e2384917 | 916 | pr_err(PFX "response: Error %x on cmd %x\n", |
1da177e4 LT |
917 | msg->msg.data[0], |
918 | msg->msg.cmd); | |
36c7dc44 | 919 | |
1da177e4 LT |
920 | ipmi_free_recv_msg(msg); |
921 | } | |
922 | ||
923 | static void ipmi_wdog_pretimeout_handler(void *handler_data) | |
924 | { | |
925 | if (preaction_val != WDOG_PRETIMEOUT_NONE) { | |
b385676b CM |
926 | if (preop_val == WDOG_PREOP_PANIC) { |
927 | if (atomic_inc_and_test(&preop_panic_excl)) | |
928 | panic("Watchdog pre-timeout"); | |
929 | } else if (preop_val == WDOG_PREOP_GIVE_DATA) { | |
d1b29b97 CM |
930 | unsigned long flags; |
931 | ||
932 | spin_lock_irqsave(&ipmi_read_lock, flags); | |
1da177e4 LT |
933 | data_to_read = 1; |
934 | wake_up_interruptible(&read_q); | |
935 | kill_fasync(&fasync_q, SIGIO, POLL_IN); | |
d1b29b97 | 936 | spin_unlock_irqrestore(&ipmi_read_lock, flags); |
1da177e4 LT |
937 | } |
938 | } | |
939 | ||
36c7dc44 CM |
940 | /* |
941 | * On some machines, the heartbeat will give an error and not | |
942 | * work unless we re-enable the timer. So do so. | |
943 | */ | |
d1b29b97 | 944 | atomic_set(&pretimeout_since_last_heartbeat, 1); |
1da177e4 LT |
945 | } |
946 | ||
5194970c CM |
947 | static void ipmi_wdog_panic_handler(void *user_data) |
948 | { | |
949 | static int panic_event_handled; | |
950 | ||
951 | /* | |
952 | * On a panic, if we have a panic timeout, make sure to extend | |
953 | * the watchdog timer to a reasonable value to complete the | |
954 | * panic, if the watchdog timer is running. Plus the | |
955 | * pretimeout is meaningless at panic time. | |
956 | */ | |
957 | if (watchdog_user && !panic_event_handled && | |
958 | ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
959 | /* Make sure we do this only once. */ | |
960 | panic_event_handled = 1; | |
961 | ||
962 | timeout = panic_wdt_timeout; | |
963 | pretimeout = 0; | |
964 | panic_halt_ipmi_set_timeout(); | |
965 | } | |
966 | } | |
967 | ||
210af2a5 | 968 | static const struct ipmi_user_hndl ipmi_hndlrs = { |
1da177e4 | 969 | .ipmi_recv_hndl = ipmi_wdog_msg_handler, |
5194970c CM |
970 | .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler, |
971 | .ipmi_panic_handler = ipmi_wdog_panic_handler | |
1da177e4 LT |
972 | }; |
973 | ||
974 | static void ipmi_register_watchdog(int ipmi_intf) | |
975 | { | |
976 | int rv = -EBUSY; | |
977 | ||
1da177e4 LT |
978 | if (watchdog_user) |
979 | goto out; | |
980 | ||
b2c03941 CM |
981 | if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf)) |
982 | goto out; | |
983 | ||
984 | watchdog_ifnum = ipmi_intf; | |
985 | ||
1da177e4 LT |
986 | rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); |
987 | if (rv < 0) { | |
e2384917 | 988 | pr_crit(PFX "Unable to register with ipmi\n"); |
1da177e4 LT |
989 | goto out; |
990 | } | |
991 | ||
511d57dc CM |
992 | rv = ipmi_get_version(watchdog_user, |
993 | &ipmi_version_major, | |
994 | &ipmi_version_minor); | |
995 | if (rv) { | |
996 | pr_warn(PFX "Unable to get IPMI version, assuming 1.0\n"); | |
997 | ipmi_version_major = 1; | |
998 | ipmi_version_minor = 0; | |
999 | } | |
1da177e4 LT |
1000 | |
1001 | rv = misc_register(&ipmi_wdog_miscdev); | |
1002 | if (rv < 0) { | |
1003 | ipmi_destroy_user(watchdog_user); | |
1004 | watchdog_user = NULL; | |
e2384917 | 1005 | pr_crit(PFX "Unable to register misc device\n"); |
1da177e4 LT |
1006 | } |
1007 | ||
612b5a8d CM |
1008 | #ifdef HAVE_DIE_NMI |
1009 | if (nmi_handler_registered) { | |
1010 | int old_pretimeout = pretimeout; | |
1011 | int old_timeout = timeout; | |
1012 | int old_preop_val = preop_val; | |
1013 | ||
36c7dc44 CM |
1014 | /* |
1015 | * Set the pretimeout to go off in a second and give | |
1016 | * ourselves plenty of time to stop the timer. | |
1017 | */ | |
612b5a8d CM |
1018 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; |
1019 | preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */ | |
1020 | pretimeout = 99; | |
1021 | timeout = 100; | |
1022 | ||
1023 | testing_nmi = 1; | |
1024 | ||
1025 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
1026 | if (rv) { | |
e2384917 CM |
1027 | pr_warn(PFX "Error starting timer to test NMI: 0x%x. The NMI pretimeout will likely not work\n", |
1028 | rv); | |
612b5a8d CM |
1029 | rv = 0; |
1030 | goto out_restore; | |
1031 | } | |
1032 | ||
1033 | msleep(1500); | |
1034 | ||
1035 | if (testing_nmi != 2) { | |
e2384917 | 1036 | pr_warn(PFX "IPMI NMI didn't seem to occur. The NMI pretimeout will likely not work\n"); |
612b5a8d | 1037 | } |
36c7dc44 | 1038 | out_restore: |
612b5a8d CM |
1039 | testing_nmi = 0; |
1040 | preop_val = old_preop_val; | |
1041 | pretimeout = old_pretimeout; | |
1042 | timeout = old_timeout; | |
1043 | } | |
1044 | #endif | |
1045 | ||
1da177e4 | 1046 | out: |
1da177e4 LT |
1047 | if ((start_now) && (rv == 0)) { |
1048 | /* Run from startup, so start the timer now. */ | |
1049 | start_now = 0; /* Disable this function after first startup. */ | |
1050 | ipmi_watchdog_state = action_val; | |
1051 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
e2384917 | 1052 | pr_info(PFX "Starting now!\n"); |
612b5a8d CM |
1053 | } else { |
1054 | /* Stop the timer now. */ | |
1055 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
1056 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
1da177e4 LT |
1057 | } |
1058 | } | |
1059 | ||
b2c03941 CM |
1060 | static void ipmi_unregister_watchdog(int ipmi_intf) |
1061 | { | |
1062 | int rv; | |
2911c988 | 1063 | struct ipmi_user *loc_user = watchdog_user; |
b2c03941 | 1064 | |
d1b29b97 CM |
1065 | if (!loc_user) |
1066 | return; | |
b2c03941 CM |
1067 | |
1068 | if (watchdog_ifnum != ipmi_intf) | |
d1b29b97 | 1069 | return; |
b2c03941 CM |
1070 | |
1071 | /* Make sure no one can call us any more. */ | |
1072 | misc_deregister(&ipmi_wdog_miscdev); | |
1073 | ||
d1b29b97 CM |
1074 | watchdog_user = NULL; |
1075 | ||
36c7dc44 CM |
1076 | /* |
1077 | * Wait to make sure the message makes it out. The lower layer has | |
1078 | * pointers to our buffers, we want to make sure they are done before | |
1079 | * we release our memory. | |
1080 | */ | |
d1b29b97 CM |
1081 | while (atomic_read(&msg_tofree)) |
1082 | msg_free_smi(NULL); | |
1083 | ||
1084 | mutex_lock(&ipmi_watchdog_mutex); | |
b2c03941 CM |
1085 | |
1086 | /* Disconnect from IPMI. */ | |
d1b29b97 CM |
1087 | rv = ipmi_destroy_user(loc_user); |
1088 | if (rv) | |
1089 | pr_warn(PFX "error unlinking from IPMI: %d\n", rv); | |
b2c03941 | 1090 | |
d1b29b97 CM |
1091 | /* If it comes back, restart it properly. */ |
1092 | ipmi_start_timer_on_heartbeat = 1; | |
1093 | ||
1094 | mutex_unlock(&ipmi_watchdog_mutex); | |
b2c03941 CM |
1095 | } |
1096 | ||
612b5a8d | 1097 | #ifdef HAVE_DIE_NMI |
1da177e4 | 1098 | static int |
9c48f1c6 | 1099 | ipmi_nmi(unsigned int val, struct pt_regs *regs) |
1da177e4 | 1100 | { |
612b5a8d CM |
1101 | /* |
1102 | * If we get here, it's an NMI that's not a memory or I/O | |
1103 | * error. We can't truly tell if it's from IPMI or not | |
1104 | * without sending a message, and sending a message is almost | |
1105 | * impossible because of locking. | |
1106 | */ | |
1107 | ||
1108 | if (testing_nmi) { | |
1109 | testing_nmi = 2; | |
9c48f1c6 | 1110 | return NMI_HANDLED; |
612b5a8d CM |
1111 | } |
1112 | ||
36c7dc44 | 1113 | /* If we are not expecting a timeout, ignore it. */ |
8f05ee9a | 1114 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
9c48f1c6 | 1115 | return NMI_DONE; |
612b5a8d CM |
1116 | |
1117 | if (preaction_val != WDOG_PRETIMEOUT_NMI) | |
9c48f1c6 | 1118 | return NMI_DONE; |
8f05ee9a | 1119 | |
36c7dc44 CM |
1120 | /* |
1121 | * If no one else handled the NMI, we assume it was the IPMI | |
1122 | * watchdog. | |
1123 | */ | |
612b5a8d | 1124 | if (preop_val == WDOG_PREOP_PANIC) { |
8f05ee9a CM |
1125 | /* On some machines, the heartbeat will give |
1126 | an error and not work unless we re-enable | |
1127 | the timer. So do so. */ | |
d1b29b97 | 1128 | atomic_set(&pretimeout_since_last_heartbeat, 1); |
b385676b | 1129 | if (atomic_inc_and_test(&preop_panic_excl)) |
73cbf4a1 | 1130 | nmi_panic(regs, PFX "pre-timeout"); |
8f05ee9a | 1131 | } |
1da177e4 | 1132 | |
9c48f1c6 | 1133 | return NMI_HANDLED; |
1da177e4 | 1134 | } |
1da177e4 LT |
1135 | #endif |
1136 | ||
1137 | static int wdog_reboot_handler(struct notifier_block *this, | |
1138 | unsigned long code, | |
1139 | void *unused) | |
1140 | { | |
36c7dc44 | 1141 | static int reboot_event_handled; |
1da177e4 LT |
1142 | |
1143 | if ((watchdog_user) && (!reboot_event_handled)) { | |
1144 | /* Make sure we only do this once. */ | |
1145 | reboot_event_handled = 1; | |
1146 | ||
fcfa4724 | 1147 | if (code == SYS_POWER_OFF || code == SYS_HALT) { |
1da177e4 LT |
1148 | /* Disable the WDT if we are shutting down. */ |
1149 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
423a5bb4 | 1150 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
96febe9f | 1151 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
860f01e9 VV |
1152 | /* Set a long timer to let the reboot happen or |
1153 | reset if it hangs, but only if the watchdog | |
96febe9f | 1154 | timer was already running. */ |
860f01e9 VV |
1155 | if (timeout < 120) |
1156 | timeout = 120; | |
1da177e4 LT |
1157 | pretimeout = 0; |
1158 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; | |
423a5bb4 | 1159 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
1da177e4 LT |
1160 | } |
1161 | } | |
1162 | return NOTIFY_OK; | |
1163 | } | |
1164 | ||
1165 | static struct notifier_block wdog_reboot_notifier = { | |
1166 | .notifier_call = wdog_reboot_handler, | |
1167 | .next = NULL, | |
1168 | .priority = 0 | |
1169 | }; | |
1170 | ||
50c812b2 | 1171 | static void ipmi_new_smi(int if_num, struct device *device) |
1da177e4 LT |
1172 | { |
1173 | ipmi_register_watchdog(if_num); | |
1174 | } | |
1175 | ||
1176 | static void ipmi_smi_gone(int if_num) | |
1177 | { | |
b2c03941 | 1178 | ipmi_unregister_watchdog(if_num); |
1da177e4 LT |
1179 | } |
1180 | ||
36c7dc44 | 1181 | static struct ipmi_smi_watcher smi_watcher = { |
1da177e4 LT |
1182 | .owner = THIS_MODULE, |
1183 | .new_smi = ipmi_new_smi, | |
1184 | .smi_gone = ipmi_smi_gone | |
1185 | }; | |
1186 | ||
cc4673ee | 1187 | static int action_op(const char *inval, char *outval) |
1da177e4 | 1188 | { |
cc4673ee CM |
1189 | if (outval) |
1190 | strcpy(outval, action); | |
1191 | ||
1192 | if (!inval) | |
1193 | return 0; | |
1da177e4 | 1194 | |
cc4673ee | 1195 | if (strcmp(inval, "reset") == 0) |
1da177e4 | 1196 | action_val = WDOG_TIMEOUT_RESET; |
cc4673ee | 1197 | else if (strcmp(inval, "none") == 0) |
1da177e4 | 1198 | action_val = WDOG_TIMEOUT_NONE; |
cc4673ee | 1199 | else if (strcmp(inval, "power_cycle") == 0) |
1da177e4 | 1200 | action_val = WDOG_TIMEOUT_POWER_CYCLE; |
cc4673ee | 1201 | else if (strcmp(inval, "power_off") == 0) |
1da177e4 | 1202 | action_val = WDOG_TIMEOUT_POWER_DOWN; |
cc4673ee CM |
1203 | else |
1204 | return -EINVAL; | |
1205 | strcpy(action, inval); | |
1206 | return 0; | |
1207 | } | |
1208 | ||
1209 | static int preaction_op(const char *inval, char *outval) | |
1210 | { | |
1211 | if (outval) | |
1212 | strcpy(outval, preaction); | |
1da177e4 | 1213 | |
cc4673ee CM |
1214 | if (!inval) |
1215 | return 0; | |
1216 | ||
1217 | if (strcmp(inval, "pre_none") == 0) | |
1da177e4 | 1218 | preaction_val = WDOG_PRETIMEOUT_NONE; |
cc4673ee | 1219 | else if (strcmp(inval, "pre_smi") == 0) |
1da177e4 | 1220 | preaction_val = WDOG_PRETIMEOUT_SMI; |
612b5a8d | 1221 | #ifdef HAVE_DIE_NMI |
cc4673ee | 1222 | else if (strcmp(inval, "pre_nmi") == 0) |
1da177e4 LT |
1223 | preaction_val = WDOG_PRETIMEOUT_NMI; |
1224 | #endif | |
cc4673ee | 1225 | else if (strcmp(inval, "pre_int") == 0) |
1da177e4 | 1226 | preaction_val = WDOG_PRETIMEOUT_MSG_INT; |
cc4673ee CM |
1227 | else |
1228 | return -EINVAL; | |
1229 | strcpy(preaction, inval); | |
1230 | return 0; | |
1231 | } | |
1232 | ||
1233 | static int preop_op(const char *inval, char *outval) | |
1234 | { | |
1235 | if (outval) | |
1236 | strcpy(outval, preop); | |
1da177e4 | 1237 | |
cc4673ee CM |
1238 | if (!inval) |
1239 | return 0; | |
1240 | ||
1241 | if (strcmp(inval, "preop_none") == 0) | |
1da177e4 | 1242 | preop_val = WDOG_PREOP_NONE; |
cc4673ee | 1243 | else if (strcmp(inval, "preop_panic") == 0) |
1da177e4 | 1244 | preop_val = WDOG_PREOP_PANIC; |
cc4673ee | 1245 | else if (strcmp(inval, "preop_give_data") == 0) |
1da177e4 | 1246 | preop_val = WDOG_PREOP_GIVE_DATA; |
cc4673ee CM |
1247 | else |
1248 | return -EINVAL; | |
1249 | strcpy(preop, inval); | |
1250 | return 0; | |
1251 | } | |
1da177e4 | 1252 | |
cc4673ee CM |
1253 | static void check_parms(void) |
1254 | { | |
612b5a8d | 1255 | #ifdef HAVE_DIE_NMI |
cc4673ee CM |
1256 | int do_nmi = 0; |
1257 | int rv; | |
1258 | ||
1da177e4 | 1259 | if (preaction_val == WDOG_PRETIMEOUT_NMI) { |
cc4673ee | 1260 | do_nmi = 1; |
1da177e4 | 1261 | if (preop_val == WDOG_PREOP_GIVE_DATA) { |
e2384917 | 1262 | pr_warn(PFX "Pretimeout op is to give data but NMI pretimeout is enabled, setting pretimeout op to none\n"); |
cc4673ee CM |
1263 | preop_op("preop_none", NULL); |
1264 | do_nmi = 0; | |
1da177e4 | 1265 | } |
cc4673ee CM |
1266 | } |
1267 | if (do_nmi && !nmi_handler_registered) { | |
9c48f1c6 DZ |
1268 | rv = register_nmi_handler(NMI_UNKNOWN, ipmi_nmi, 0, |
1269 | "ipmi"); | |
1da177e4 | 1270 | if (rv) { |
e2384917 | 1271 | pr_warn(PFX "Can't register nmi handler\n"); |
cc4673ee CM |
1272 | return; |
1273 | } else | |
1274 | nmi_handler_registered = 1; | |
1275 | } else if (!do_nmi && nmi_handler_registered) { | |
9c48f1c6 | 1276 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); |
cc4673ee | 1277 | nmi_handler_registered = 0; |
1da177e4 LT |
1278 | } |
1279 | #endif | |
cc4673ee CM |
1280 | } |
1281 | ||
1282 | static int __init ipmi_wdog_init(void) | |
1283 | { | |
1284 | int rv; | |
1285 | ||
1286 | if (action_op(action, NULL)) { | |
1287 | action_op("reset", NULL); | |
e2384917 CM |
1288 | pr_info(PFX "Unknown action '%s', defaulting to reset\n", |
1289 | action); | |
cc4673ee CM |
1290 | } |
1291 | ||
1292 | if (preaction_op(preaction, NULL)) { | |
1293 | preaction_op("pre_none", NULL); | |
e2384917 CM |
1294 | pr_info(PFX "Unknown preaction '%s', defaulting to none\n", |
1295 | preaction); | |
cc4673ee CM |
1296 | } |
1297 | ||
1298 | if (preop_op(preop, NULL)) { | |
1299 | preop_op("preop_none", NULL); | |
e2384917 | 1300 | pr_info(PFX "Unknown preop '%s', defaulting to none\n", preop); |
cc4673ee CM |
1301 | } |
1302 | ||
1303 | check_parms(); | |
1da177e4 | 1304 | |
b2c03941 | 1305 | register_reboot_notifier(&wdog_reboot_notifier); |
b2c03941 | 1306 | |
1da177e4 LT |
1307 | rv = ipmi_smi_watcher_register(&smi_watcher); |
1308 | if (rv) { | |
612b5a8d CM |
1309 | #ifdef HAVE_DIE_NMI |
1310 | if (nmi_handler_registered) | |
9c48f1c6 | 1311 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); |
1da177e4 | 1312 | #endif |
b2c03941 | 1313 | unregister_reboot_notifier(&wdog_reboot_notifier); |
e2384917 | 1314 | pr_warn(PFX "can't register smi watcher\n"); |
1da177e4 LT |
1315 | return rv; |
1316 | } | |
1317 | ||
e2384917 | 1318 | pr_info(PFX "driver initialized\n"); |
1fdd75bd | 1319 | |
1da177e4 LT |
1320 | return 0; |
1321 | } | |
1322 | ||
b2c03941 | 1323 | static void __exit ipmi_wdog_exit(void) |
1da177e4 | 1324 | { |
b2c03941 CM |
1325 | ipmi_smi_watcher_unregister(&smi_watcher); |
1326 | ipmi_unregister_watchdog(watchdog_ifnum); | |
1da177e4 | 1327 | |
612b5a8d | 1328 | #ifdef HAVE_DIE_NMI |
cc4673ee | 1329 | if (nmi_handler_registered) |
9c48f1c6 | 1330 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); |
1da177e4 LT |
1331 | #endif |
1332 | ||
1da177e4 | 1333 | unregister_reboot_notifier(&wdog_reboot_notifier); |
1da177e4 LT |
1334 | } |
1335 | module_exit(ipmi_wdog_exit); | |
1336 | module_init(ipmi_wdog_init); | |
1337 | MODULE_LICENSE("GPL"); | |
1fdd75bd CM |
1338 | MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); |
1339 | MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); |