Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 LT |
34 | #include <linux/module.h> |
35 | #include <linux/moduleparam.h> | |
36 | #include <linux/ipmi.h> | |
37 | #include <linux/ipmi_smi.h> | |
38 | #include <linux/watchdog.h> | |
39 | #include <linux/miscdevice.h> | |
40 | #include <linux/init.h> | |
d6dfd131 | 41 | #include <linux/completion.h> |
1eeb66a1 | 42 | #include <linux/kdebug.h> |
1da177e4 LT |
43 | #include <linux/rwsem.h> |
44 | #include <linux/errno.h> | |
45 | #include <asm/uaccess.h> | |
46 | #include <linux/notifier.h> | |
47 | #include <linux/nmi.h> | |
48 | #include <linux/reboot.h> | |
49 | #include <linux/wait.h> | |
50 | #include <linux/poll.h> | |
cc4673ee CM |
51 | #include <linux/string.h> |
52 | #include <linux/ctype.h> | |
f64da958 | 53 | #include <linux/delay.h> |
b385676b | 54 | #include <asm/atomic.h> |
f64da958 CM |
55 | |
56 | #ifdef CONFIG_X86 | |
57 | /* This is ugly, but I've determined that x86 is the only architecture | |
58 | that can reasonably support the IPMI NMI watchdog timeout at this | |
59 | time. If another architecture adds this capability somehow, it | |
60 | will have to be a somewhat different mechanism and I have no idea | |
61 | how it will work. So in the unlikely event that another | |
62 | architecture supports this, we can figure out a good generic | |
63 | mechanism for it at that time. */ | |
f64da958 | 64 | #define HAVE_DIE_NMI_POST |
1da177e4 LT |
65 | #endif |
66 | ||
67 | #define PFX "IPMI Watchdog: " | |
68 | ||
1da177e4 LT |
69 | /* |
70 | * The IPMI command/response information for the watchdog timer. | |
71 | */ | |
72 | ||
73 | /* values for byte 1 of the set command, byte 2 of the get response. */ | |
74 | #define WDOG_DONT_LOG (1 << 7) | |
75 | #define WDOG_DONT_STOP_ON_SET (1 << 6) | |
76 | #define WDOG_SET_TIMER_USE(byte, use) \ | |
77 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
78 | #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) | |
79 | #define WDOG_TIMER_USE_BIOS_FRB2 1 | |
80 | #define WDOG_TIMER_USE_BIOS_POST 2 | |
81 | #define WDOG_TIMER_USE_OS_LOAD 3 | |
82 | #define WDOG_TIMER_USE_SMS_OS 4 | |
83 | #define WDOG_TIMER_USE_OEM 5 | |
84 | ||
85 | /* values for byte 2 of the set command, byte 3 of the get response. */ | |
86 | #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ | |
87 | byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) | |
88 | #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) | |
89 | #define WDOG_PRETIMEOUT_NONE 0 | |
90 | #define WDOG_PRETIMEOUT_SMI 1 | |
91 | #define WDOG_PRETIMEOUT_NMI 2 | |
92 | #define WDOG_PRETIMEOUT_MSG_INT 3 | |
93 | ||
94 | /* Operations that can be performed on a pretimout. */ | |
95 | #define WDOG_PREOP_NONE 0 | |
96 | #define WDOG_PREOP_PANIC 1 | |
97 | #define WDOG_PREOP_GIVE_DATA 2 /* Cause data to be available to | |
98 | read. Doesn't work in NMI | |
99 | mode. */ | |
100 | ||
101 | /* Actions to perform on a full timeout. */ | |
102 | #define WDOG_SET_TIMEOUT_ACT(byte, use) \ | |
103 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
104 | #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) | |
105 | #define WDOG_TIMEOUT_NONE 0 | |
106 | #define WDOG_TIMEOUT_RESET 1 | |
107 | #define WDOG_TIMEOUT_POWER_DOWN 2 | |
108 | #define WDOG_TIMEOUT_POWER_CYCLE 3 | |
109 | ||
110 | /* Byte 3 of the get command, byte 4 of the get response is the | |
111 | pre-timeout in seconds. */ | |
112 | ||
113 | /* Bits for setting byte 4 of the set command, byte 5 of the get response. */ | |
114 | #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1) | |
115 | #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2) | |
116 | #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3) | |
117 | #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4) | |
118 | #define WDOG_EXPIRE_CLEAR_OEM (1 << 5) | |
119 | ||
120 | /* Setting/getting the watchdog timer value. This is for bytes 5 and | |
121 | 6 (the timeout time) of the set command, and bytes 6 and 7 (the | |
122 | timeout time) and 8 and 9 (the current countdown value) of the | |
123 | response. The timeout value is given in seconds (in the command it | |
124 | is 100ms intervals). */ | |
125 | #define WDOG_SET_TIMEOUT(byte1, byte2, val) \ | |
126 | (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8) | |
127 | #define WDOG_GET_TIMEOUT(byte1, byte2) \ | |
128 | (((byte1) | ((byte2) << 8)) / 10) | |
129 | ||
130 | #define IPMI_WDOG_RESET_TIMER 0x22 | |
131 | #define IPMI_WDOG_SET_TIMER 0x24 | |
132 | #define IPMI_WDOG_GET_TIMER 0x25 | |
133 | ||
134 | /* These are here until the real ones get into the watchdog.h interface. */ | |
135 | #ifndef WDIOC_GETTIMEOUT | |
136 | #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int) | |
137 | #endif | |
138 | #ifndef WDIOC_SET_PRETIMEOUT | |
139 | #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int) | |
140 | #endif | |
141 | #ifndef WDIOC_GET_PRETIMEOUT | |
142 | #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int) | |
143 | #endif | |
144 | ||
4bfdf378 | 145 | static int nowayout = WATCHDOG_NOWAYOUT; |
1da177e4 | 146 | |
0c8204b3 | 147 | static ipmi_user_t watchdog_user; |
b2c03941 | 148 | static int watchdog_ifnum; |
1da177e4 LT |
149 | |
150 | /* Default the timeout to 10 seconds. */ | |
151 | static int timeout = 10; | |
152 | ||
153 | /* The pre-timeout is disabled by default. */ | |
0c8204b3 | 154 | static int pretimeout; |
1da177e4 LT |
155 | |
156 | /* Default action is to reset the board on a timeout. */ | |
157 | static unsigned char action_val = WDOG_TIMEOUT_RESET; | |
158 | ||
159 | static char action[16] = "reset"; | |
160 | ||
161 | static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; | |
162 | ||
163 | static char preaction[16] = "pre_none"; | |
164 | ||
165 | static unsigned char preop_val = WDOG_PREOP_NONE; | |
166 | ||
167 | static char preop[16] = "preop_none"; | |
168 | static DEFINE_SPINLOCK(ipmi_read_lock); | |
0c8204b3 | 169 | static char data_to_read; |
1da177e4 | 170 | static DECLARE_WAIT_QUEUE_HEAD(read_q); |
0c8204b3 RD |
171 | static struct fasync_struct *fasync_q; |
172 | static char pretimeout_since_last_heartbeat; | |
1da177e4 LT |
173 | static char expect_close; |
174 | ||
b2c03941 CM |
175 | static int ifnum_to_use = -1; |
176 | ||
cc4673ee CM |
177 | static DECLARE_RWSEM(register_sem); |
178 | ||
179 | /* Parameters to ipmi_set_timeout */ | |
180 | #define IPMI_SET_TIMEOUT_NO_HB 0 | |
181 | #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 | |
182 | #define IPMI_SET_TIMEOUT_FORCE_HB 2 | |
183 | ||
184 | static int ipmi_set_timeout(int do_heartbeat); | |
b2c03941 CM |
185 | static void ipmi_register_watchdog(int ipmi_intf); |
186 | static void ipmi_unregister_watchdog(int ipmi_intf); | |
cc4673ee | 187 | |
1da177e4 LT |
188 | /* If true, the driver will start running as soon as it is configured |
189 | and ready. */ | |
0c8204b3 | 190 | static int start_now; |
1da177e4 | 191 | |
cc4673ee CM |
192 | static int set_param_int(const char *val, struct kernel_param *kp) |
193 | { | |
194 | char *endp; | |
195 | int l; | |
196 | int rv = 0; | |
197 | ||
198 | if (!val) | |
199 | return -EINVAL; | |
200 | l = simple_strtoul(val, &endp, 0); | |
201 | if (endp == val) | |
202 | return -EINVAL; | |
203 | ||
204 | down_read(®ister_sem); | |
205 | *((int *)kp->arg) = l; | |
206 | if (watchdog_user) | |
207 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
208 | up_read(®ister_sem); | |
209 | ||
210 | return rv; | |
211 | } | |
212 | ||
213 | static int get_param_int(char *buffer, struct kernel_param *kp) | |
214 | { | |
215 | return sprintf(buffer, "%i", *((int *)kp->arg)); | |
216 | } | |
217 | ||
218 | typedef int (*action_fn)(const char *intval, char *outval); | |
219 | ||
220 | static int action_op(const char *inval, char *outval); | |
221 | static int preaction_op(const char *inval, char *outval); | |
222 | static int preop_op(const char *inval, char *outval); | |
223 | static void check_parms(void); | |
224 | ||
225 | static int set_param_str(const char *val, struct kernel_param *kp) | |
226 | { | |
227 | action_fn fn = (action_fn) kp->arg; | |
228 | int rv = 0; | |
43cdff92 SD |
229 | char valcp[16]; |
230 | char *s; | |
66f969d0 | 231 | |
43cdff92 SD |
232 | strncpy(valcp, val, 16); |
233 | valcp[15] = '\0'; | |
66f969d0 | 234 | |
43cdff92 | 235 | s = strstrip(valcp); |
cc4673ee CM |
236 | |
237 | down_read(®ister_sem); | |
66f969d0 | 238 | rv = fn(s, NULL); |
cc4673ee CM |
239 | if (rv) |
240 | goto out_unlock; | |
241 | ||
242 | check_parms(); | |
243 | if (watchdog_user) | |
244 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
245 | ||
246 | out_unlock: | |
247 | up_read(®ister_sem); | |
248 | return rv; | |
249 | } | |
250 | ||
251 | static int get_param_str(char *buffer, struct kernel_param *kp) | |
252 | { | |
253 | action_fn fn = (action_fn) kp->arg; | |
254 | int rv; | |
255 | ||
256 | rv = fn(NULL, buffer); | |
257 | if (rv) | |
258 | return rv; | |
259 | return strlen(buffer); | |
260 | } | |
261 | ||
b2c03941 CM |
262 | |
263 | static int set_param_wdog_ifnum(const char *val, struct kernel_param *kp) | |
264 | { | |
265 | int rv = param_set_int(val, kp); | |
266 | if (rv) | |
267 | return rv; | |
268 | if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum)) | |
269 | return 0; | |
270 | ||
271 | ipmi_unregister_watchdog(watchdog_ifnum); | |
272 | ipmi_register_watchdog(ifnum_to_use); | |
273 | return 0; | |
274 | } | |
275 | ||
276 | module_param_call(ifnum_to_use, set_param_wdog_ifnum, get_param_int, | |
277 | &ifnum_to_use, 0644); | |
278 | MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog " | |
279 | "timer. Setting to -1 defaults to the first registered " | |
280 | "interface"); | |
281 | ||
cc4673ee | 282 | module_param_call(timeout, set_param_int, get_param_int, &timeout, 0644); |
1da177e4 | 283 | MODULE_PARM_DESC(timeout, "Timeout value in seconds."); |
cc4673ee CM |
284 | |
285 | module_param_call(pretimeout, set_param_int, get_param_int, &pretimeout, 0644); | |
1da177e4 | 286 | MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); |
cc4673ee CM |
287 | |
288 | module_param_call(action, set_param_str, get_param_str, action_op, 0644); | |
1da177e4 LT |
289 | MODULE_PARM_DESC(action, "Timeout action. One of: " |
290 | "reset, none, power_cycle, power_off."); | |
cc4673ee CM |
291 | |
292 | module_param_call(preaction, set_param_str, get_param_str, preaction_op, 0644); | |
1da177e4 LT |
293 | MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " |
294 | "pre_none, pre_smi, pre_nmi, pre_int."); | |
cc4673ee CM |
295 | |
296 | module_param_call(preop, set_param_str, get_param_str, preop_op, 0644); | |
1da177e4 LT |
297 | MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " |
298 | "preop_none, preop_panic, preop_give_data."); | |
cc4673ee | 299 | |
b2c03941 | 300 | module_param(start_now, int, 0444); |
1da177e4 LT |
301 | MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" |
302 | "soon as the driver is loaded."); | |
cc4673ee CM |
303 | |
304 | module_param(nowayout, int, 0644); | |
b2c03941 CM |
305 | MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started " |
306 | "(default=CONFIG_WATCHDOG_NOWAYOUT)"); | |
1da177e4 LT |
307 | |
308 | /* Default state of the timer. */ | |
309 | static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
310 | ||
311 | /* If shutting down via IPMI, we ignore the heartbeat. */ | |
0c8204b3 | 312 | static int ipmi_ignore_heartbeat; |
1da177e4 LT |
313 | |
314 | /* Is someone using the watchdog? Only one user is allowed. */ | |
0c8204b3 | 315 | static unsigned long ipmi_wdog_open; |
1da177e4 LT |
316 | |
317 | /* If set to 1, the heartbeat command will set the state to reset and | |
318 | start the timer. The timer doesn't normally run when the driver is | |
319 | first opened until the heartbeat is set the first time, this | |
320 | variable is used to accomplish this. */ | |
0c8204b3 | 321 | static int ipmi_start_timer_on_heartbeat; |
1da177e4 LT |
322 | |
323 | /* IPMI version of the BMC. */ | |
324 | static unsigned char ipmi_version_major; | |
325 | static unsigned char ipmi_version_minor; | |
326 | ||
b385676b CM |
327 | /* If a pretimeout occurs, this is used to allow only one panic to happen. */ |
328 | static atomic_t preop_panic_excl = ATOMIC_INIT(-1); | |
1da177e4 | 329 | |
f64da958 CM |
330 | #ifdef HAVE_DIE_NMI_POST |
331 | static int testing_nmi; | |
332 | static int nmi_handler_registered; | |
333 | #endif | |
334 | ||
1da177e4 LT |
335 | static int ipmi_heartbeat(void); |
336 | static void panic_halt_ipmi_heartbeat(void); | |
337 | ||
338 | ||
d6dfd131 | 339 | /* We use a mutex to make sure that only one thing can send a set |
1da177e4 | 340 | timeout at one time, because we only have one copy of the data. |
d6dfd131 | 341 | The mutex is claimed when the set_timeout is sent and freed |
1da177e4 LT |
342 | when both messages are free. */ |
343 | static atomic_t set_timeout_tofree = ATOMIC_INIT(0); | |
d6dfd131 CM |
344 | static DEFINE_MUTEX(set_timeout_lock); |
345 | static DECLARE_COMPLETION(set_timeout_wait); | |
1da177e4 LT |
346 | static void set_timeout_free_smi(struct ipmi_smi_msg *msg) |
347 | { | |
348 | if (atomic_dec_and_test(&set_timeout_tofree)) | |
d6dfd131 | 349 | complete(&set_timeout_wait); |
1da177e4 LT |
350 | } |
351 | static void set_timeout_free_recv(struct ipmi_recv_msg *msg) | |
352 | { | |
353 | if (atomic_dec_and_test(&set_timeout_tofree)) | |
d6dfd131 | 354 | complete(&set_timeout_wait); |
1da177e4 LT |
355 | } |
356 | static struct ipmi_smi_msg set_timeout_smi_msg = | |
357 | { | |
358 | .done = set_timeout_free_smi | |
359 | }; | |
360 | static struct ipmi_recv_msg set_timeout_recv_msg = | |
361 | { | |
362 | .done = set_timeout_free_recv | |
363 | }; | |
364 | ||
365 | static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, | |
366 | struct ipmi_recv_msg *recv_msg, | |
367 | int *send_heartbeat_now) | |
368 | { | |
369 | struct kernel_ipmi_msg msg; | |
370 | unsigned char data[6]; | |
371 | int rv; | |
372 | struct ipmi_system_interface_addr addr; | |
373 | int hbnow = 0; | |
374 | ||
375 | ||
f64da958 CM |
376 | /* These can be cleared as we are setting the timeout. */ |
377 | ipmi_start_timer_on_heartbeat = 0; | |
378 | pretimeout_since_last_heartbeat = 0; | |
379 | ||
1da177e4 LT |
380 | data[0] = 0; |
381 | WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); | |
382 | ||
383 | if ((ipmi_version_major > 1) | |
384 | || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) | |
385 | { | |
386 | /* This is an IPMI 1.5-only feature. */ | |
387 | data[0] |= WDOG_DONT_STOP_ON_SET; | |
388 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
389 | /* In ipmi 1.0, setting the timer stops the watchdog, we | |
390 | need to start it back up again. */ | |
391 | hbnow = 1; | |
392 | } | |
393 | ||
394 | data[1] = 0; | |
395 | WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); | |
8f05ee9a | 396 | if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { |
1da177e4 LT |
397 | WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); |
398 | data[2] = pretimeout; | |
399 | } else { | |
400 | WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); | |
401 | data[2] = 0; /* No pretimeout. */ | |
402 | } | |
403 | data[3] = 0; | |
404 | WDOG_SET_TIMEOUT(data[4], data[5], timeout); | |
405 | ||
406 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
407 | addr.channel = IPMI_BMC_CHANNEL; | |
408 | addr.lun = 0; | |
409 | ||
410 | msg.netfn = 0x06; | |
411 | msg.cmd = IPMI_WDOG_SET_TIMER; | |
412 | msg.data = data; | |
413 | msg.data_len = sizeof(data); | |
414 | rv = ipmi_request_supply_msgs(watchdog_user, | |
415 | (struct ipmi_addr *) &addr, | |
416 | 0, | |
417 | &msg, | |
418 | NULL, | |
419 | smi_msg, | |
420 | recv_msg, | |
421 | 1); | |
422 | if (rv) { | |
423 | printk(KERN_WARNING PFX "set timeout error: %d\n", | |
424 | rv); | |
425 | } | |
426 | ||
427 | if (send_heartbeat_now) | |
428 | *send_heartbeat_now = hbnow; | |
429 | ||
430 | return rv; | |
431 | } | |
432 | ||
1da177e4 LT |
433 | static int ipmi_set_timeout(int do_heartbeat) |
434 | { | |
435 | int send_heartbeat_now; | |
436 | int rv; | |
437 | ||
438 | ||
439 | /* We can only send one of these at a time. */ | |
d6dfd131 | 440 | mutex_lock(&set_timeout_lock); |
1da177e4 LT |
441 | |
442 | atomic_set(&set_timeout_tofree, 2); | |
443 | ||
444 | rv = i_ipmi_set_timeout(&set_timeout_smi_msg, | |
445 | &set_timeout_recv_msg, | |
446 | &send_heartbeat_now); | |
447 | if (rv) { | |
d6dfd131 CM |
448 | mutex_unlock(&set_timeout_lock); |
449 | goto out; | |
450 | } | |
451 | ||
452 | wait_for_completion(&set_timeout_wait); | |
453 | ||
f64da958 CM |
454 | mutex_unlock(&set_timeout_lock); |
455 | ||
d6dfd131 CM |
456 | if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) |
457 | || ((send_heartbeat_now) | |
458 | && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) | |
d6dfd131 | 459 | rv = ipmi_heartbeat(); |
1da177e4 | 460 | |
d6dfd131 | 461 | out: |
1da177e4 LT |
462 | return rv; |
463 | } | |
464 | ||
465 | static void dummy_smi_free(struct ipmi_smi_msg *msg) | |
466 | { | |
467 | } | |
468 | static void dummy_recv_free(struct ipmi_recv_msg *msg) | |
469 | { | |
470 | } | |
471 | static struct ipmi_smi_msg panic_halt_smi_msg = | |
472 | { | |
473 | .done = dummy_smi_free | |
474 | }; | |
475 | static struct ipmi_recv_msg panic_halt_recv_msg = | |
476 | { | |
477 | .done = dummy_recv_free | |
478 | }; | |
479 | ||
480 | /* Special call, doesn't claim any locks. This is only to be called | |
481 | at panic or halt time, in run-to-completion mode, when the caller | |
482 | is the only CPU and the only thing that will be going is these IPMI | |
483 | calls. */ | |
484 | static void panic_halt_ipmi_set_timeout(void) | |
485 | { | |
486 | int send_heartbeat_now; | |
487 | int rv; | |
488 | ||
489 | rv = i_ipmi_set_timeout(&panic_halt_smi_msg, | |
490 | &panic_halt_recv_msg, | |
491 | &send_heartbeat_now); | |
492 | if (!rv) { | |
493 | if (send_heartbeat_now) | |
494 | panic_halt_ipmi_heartbeat(); | |
495 | } | |
496 | } | |
497 | ||
498 | /* We use a semaphore to make sure that only one thing can send a | |
499 | heartbeat at one time, because we only have one copy of the data. | |
500 | The semaphore is claimed when the set_timeout is sent and freed | |
501 | when both messages are free. */ | |
502 | static atomic_t heartbeat_tofree = ATOMIC_INIT(0); | |
d6dfd131 CM |
503 | static DEFINE_MUTEX(heartbeat_lock); |
504 | static DECLARE_COMPLETION(heartbeat_wait); | |
1da177e4 LT |
505 | static void heartbeat_free_smi(struct ipmi_smi_msg *msg) |
506 | { | |
507 | if (atomic_dec_and_test(&heartbeat_tofree)) | |
d6dfd131 | 508 | complete(&heartbeat_wait); |
1da177e4 LT |
509 | } |
510 | static void heartbeat_free_recv(struct ipmi_recv_msg *msg) | |
511 | { | |
512 | if (atomic_dec_and_test(&heartbeat_tofree)) | |
d6dfd131 | 513 | complete(&heartbeat_wait); |
1da177e4 LT |
514 | } |
515 | static struct ipmi_smi_msg heartbeat_smi_msg = | |
516 | { | |
517 | .done = heartbeat_free_smi | |
518 | }; | |
519 | static struct ipmi_recv_msg heartbeat_recv_msg = | |
520 | { | |
521 | .done = heartbeat_free_recv | |
522 | }; | |
523 | ||
524 | static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = | |
525 | { | |
526 | .done = dummy_smi_free | |
527 | }; | |
528 | static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = | |
529 | { | |
530 | .done = dummy_recv_free | |
531 | }; | |
532 | ||
533 | static int ipmi_heartbeat(void) | |
534 | { | |
535 | struct kernel_ipmi_msg msg; | |
536 | int rv; | |
537 | struct ipmi_system_interface_addr addr; | |
538 | ||
f64da958 | 539 | if (ipmi_ignore_heartbeat) |
1da177e4 | 540 | return 0; |
1da177e4 LT |
541 | |
542 | if (ipmi_start_timer_on_heartbeat) { | |
1da177e4 LT |
543 | ipmi_watchdog_state = action_val; |
544 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
545 | } else if (pretimeout_since_last_heartbeat) { | |
546 | /* A pretimeout occurred, make sure we set the timeout. | |
547 | We don't want to set the action, though, we want to | |
548 | leave that alone (thus it can't be combined with the | |
549 | above operation. */ | |
1da177e4 LT |
550 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
551 | } | |
552 | ||
d6dfd131 | 553 | mutex_lock(&heartbeat_lock); |
1da177e4 LT |
554 | |
555 | atomic_set(&heartbeat_tofree, 2); | |
556 | ||
557 | /* Don't reset the timer if we have the timer turned off, that | |
558 | re-enables the watchdog. */ | |
559 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) { | |
d6dfd131 | 560 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
561 | return 0; |
562 | } | |
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(watchdog_user, | |
573 | (struct ipmi_addr *) &addr, | |
574 | 0, | |
575 | &msg, | |
576 | NULL, | |
577 | &heartbeat_smi_msg, | |
578 | &heartbeat_recv_msg, | |
579 | 1); | |
580 | if (rv) { | |
d6dfd131 | 581 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
582 | printk(KERN_WARNING PFX "heartbeat failure: %d\n", |
583 | rv); | |
584 | return rv; | |
585 | } | |
586 | ||
587 | /* Wait for the heartbeat to be sent. */ | |
d6dfd131 | 588 | wait_for_completion(&heartbeat_wait); |
1da177e4 LT |
589 | |
590 | if (heartbeat_recv_msg.msg.data[0] != 0) { | |
591 | /* Got an error in the heartbeat response. It was already | |
592 | reported in ipmi_wdog_msg_handler, but we should return | |
593 | an error here. */ | |
594 | rv = -EINVAL; | |
595 | } | |
596 | ||
d6dfd131 | 597 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
598 | |
599 | return rv; | |
600 | } | |
601 | ||
602 | static void panic_halt_ipmi_heartbeat(void) | |
603 | { | |
604 | struct kernel_ipmi_msg msg; | |
605 | struct ipmi_system_interface_addr addr; | |
606 | ||
607 | ||
608 | /* Don't reset the timer if we have the timer turned off, that | |
609 | re-enables the watchdog. */ | |
610 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) | |
611 | return; | |
612 | ||
613 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
614 | addr.channel = IPMI_BMC_CHANNEL; | |
615 | addr.lun = 0; | |
616 | ||
617 | msg.netfn = 0x06; | |
618 | msg.cmd = IPMI_WDOG_RESET_TIMER; | |
619 | msg.data = NULL; | |
620 | msg.data_len = 0; | |
621 | ipmi_request_supply_msgs(watchdog_user, | |
622 | (struct ipmi_addr *) &addr, | |
623 | 0, | |
624 | &msg, | |
625 | NULL, | |
626 | &panic_halt_heartbeat_smi_msg, | |
627 | &panic_halt_heartbeat_recv_msg, | |
628 | 1); | |
629 | } | |
630 | ||
8a3628d5 | 631 | static struct watchdog_info ident = |
1da177e4 LT |
632 | { |
633 | .options = 0, /* WDIOF_SETTIMEOUT, */ | |
634 | .firmware_version = 1, | |
635 | .identity = "IPMI" | |
636 | }; | |
637 | ||
638 | static int ipmi_ioctl(struct inode *inode, struct file *file, | |
639 | unsigned int cmd, unsigned long arg) | |
640 | { | |
641 | void __user *argp = (void __user *)arg; | |
642 | int i; | |
643 | int val; | |
644 | ||
645 | switch(cmd) { | |
646 | case WDIOC_GETSUPPORT: | |
647 | i = copy_to_user(argp, &ident, sizeof(ident)); | |
648 | return i ? -EFAULT : 0; | |
649 | ||
650 | case WDIOC_SETTIMEOUT: | |
651 | i = copy_from_user(&val, argp, sizeof(int)); | |
652 | if (i) | |
653 | return -EFAULT; | |
654 | timeout = val; | |
655 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
656 | ||
657 | case WDIOC_GETTIMEOUT: | |
658 | i = copy_to_user(argp, &timeout, sizeof(timeout)); | |
659 | if (i) | |
660 | return -EFAULT; | |
661 | return 0; | |
662 | ||
663 | case WDIOC_SET_PRETIMEOUT: | |
664 | i = copy_from_user(&val, argp, sizeof(int)); | |
665 | if (i) | |
666 | return -EFAULT; | |
667 | pretimeout = val; | |
668 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
669 | ||
670 | case WDIOC_GET_PRETIMEOUT: | |
671 | i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); | |
672 | if (i) | |
673 | return -EFAULT; | |
674 | return 0; | |
675 | ||
676 | case WDIOC_KEEPALIVE: | |
677 | return ipmi_heartbeat(); | |
678 | ||
679 | case WDIOC_SETOPTIONS: | |
680 | i = copy_from_user(&val, argp, sizeof(int)); | |
681 | if (i) | |
682 | return -EFAULT; | |
683 | if (val & WDIOS_DISABLECARD) | |
684 | { | |
685 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
686 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
687 | ipmi_start_timer_on_heartbeat = 0; | |
688 | } | |
689 | ||
690 | if (val & WDIOS_ENABLECARD) | |
691 | { | |
692 | ipmi_watchdog_state = action_val; | |
693 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
694 | } | |
695 | return 0; | |
696 | ||
697 | case WDIOC_GETSTATUS: | |
698 | val = 0; | |
699 | i = copy_to_user(argp, &val, sizeof(val)); | |
700 | if (i) | |
701 | return -EFAULT; | |
702 | return 0; | |
703 | ||
704 | default: | |
705 | return -ENOIOCTLCMD; | |
706 | } | |
707 | } | |
708 | ||
709 | static ssize_t ipmi_write(struct file *file, | |
710 | const char __user *buf, | |
711 | size_t len, | |
712 | loff_t *ppos) | |
713 | { | |
714 | int rv; | |
715 | ||
716 | if (len) { | |
717 | if (!nowayout) { | |
718 | size_t i; | |
719 | ||
720 | /* In case it was set long ago */ | |
721 | expect_close = 0; | |
722 | ||
723 | for (i = 0; i != len; i++) { | |
724 | char c; | |
725 | ||
726 | if (get_user(c, buf + i)) | |
727 | return -EFAULT; | |
728 | if (c == 'V') | |
729 | expect_close = 42; | |
730 | } | |
731 | } | |
732 | rv = ipmi_heartbeat(); | |
733 | if (rv) | |
734 | return rv; | |
735 | return 1; | |
736 | } | |
737 | return 0; | |
738 | } | |
739 | ||
740 | static ssize_t ipmi_read(struct file *file, | |
741 | char __user *buf, | |
742 | size_t count, | |
743 | loff_t *ppos) | |
744 | { | |
745 | int rv = 0; | |
746 | wait_queue_t wait; | |
747 | ||
748 | if (count <= 0) | |
749 | return 0; | |
750 | ||
751 | /* Reading returns if the pretimeout has gone off, and it only does | |
752 | it once per pretimeout. */ | |
753 | spin_lock(&ipmi_read_lock); | |
754 | if (!data_to_read) { | |
755 | if (file->f_flags & O_NONBLOCK) { | |
756 | rv = -EAGAIN; | |
757 | goto out; | |
758 | } | |
759 | ||
760 | init_waitqueue_entry(&wait, current); | |
761 | add_wait_queue(&read_q, &wait); | |
762 | while (!data_to_read) { | |
763 | set_current_state(TASK_INTERRUPTIBLE); | |
764 | spin_unlock(&ipmi_read_lock); | |
765 | schedule(); | |
766 | spin_lock(&ipmi_read_lock); | |
767 | } | |
768 | remove_wait_queue(&read_q, &wait); | |
769 | ||
770 | if (signal_pending(current)) { | |
771 | rv = -ERESTARTSYS; | |
772 | goto out; | |
773 | } | |
774 | } | |
775 | data_to_read = 0; | |
776 | ||
777 | out: | |
778 | spin_unlock(&ipmi_read_lock); | |
779 | ||
780 | if (rv == 0) { | |
781 | if (copy_to_user(buf, &data_to_read, 1)) | |
782 | rv = -EFAULT; | |
783 | else | |
784 | rv = 1; | |
785 | } | |
786 | ||
787 | return rv; | |
788 | } | |
789 | ||
790 | static int ipmi_open(struct inode *ino, struct file *filep) | |
791 | { | |
e8b33617 CM |
792 | switch (iminor(ino)) { |
793 | case WATCHDOG_MINOR: | |
794 | if (test_and_set_bit(0, &ipmi_wdog_open)) | |
1da177e4 LT |
795 | return -EBUSY; |
796 | ||
e8b33617 CM |
797 | /* Don't start the timer now, let it start on the |
798 | first heartbeat. */ | |
799 | ipmi_start_timer_on_heartbeat = 1; | |
800 | return nonseekable_open(ino, filep); | |
1da177e4 | 801 | |
e8b33617 CM |
802 | default: |
803 | return (-ENODEV); | |
1da177e4 LT |
804 | } |
805 | } | |
806 | ||
807 | static unsigned int ipmi_poll(struct file *file, poll_table *wait) | |
808 | { | |
809 | unsigned int mask = 0; | |
810 | ||
811 | poll_wait(file, &read_q, wait); | |
812 | ||
813 | spin_lock(&ipmi_read_lock); | |
814 | if (data_to_read) | |
815 | mask |= (POLLIN | POLLRDNORM); | |
816 | spin_unlock(&ipmi_read_lock); | |
817 | ||
818 | return mask; | |
819 | } | |
820 | ||
821 | static int ipmi_fasync(int fd, struct file *file, int on) | |
822 | { | |
823 | int result; | |
824 | ||
825 | result = fasync_helper(fd, file, on, &fasync_q); | |
826 | ||
827 | return (result); | |
828 | } | |
829 | ||
830 | static int ipmi_close(struct inode *ino, struct file *filep) | |
831 | { | |
8a3628d5 | 832 | if (iminor(ino) == WATCHDOG_MINOR) { |
1da177e4 LT |
833 | if (expect_close == 42) { |
834 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
835 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
1da177e4 | 836 | } else { |
8a3628d5 CM |
837 | printk(KERN_CRIT PFX |
838 | "Unexpected close, not stopping watchdog!\n"); | |
1da177e4 LT |
839 | ipmi_heartbeat(); |
840 | } | |
ec26d79f | 841 | clear_bit(0, &ipmi_wdog_open); |
1da177e4 LT |
842 | } |
843 | ||
844 | ipmi_fasync (-1, filep, 0); | |
845 | expect_close = 0; | |
846 | ||
847 | return 0; | |
848 | } | |
849 | ||
62322d25 | 850 | static const struct file_operations ipmi_wdog_fops = { |
1da177e4 LT |
851 | .owner = THIS_MODULE, |
852 | .read = ipmi_read, | |
853 | .poll = ipmi_poll, | |
854 | .write = ipmi_write, | |
855 | .ioctl = ipmi_ioctl, | |
856 | .open = ipmi_open, | |
857 | .release = ipmi_close, | |
858 | .fasync = ipmi_fasync, | |
859 | }; | |
860 | ||
861 | static struct miscdevice ipmi_wdog_miscdev = { | |
862 | .minor = WATCHDOG_MINOR, | |
863 | .name = "watchdog", | |
864 | .fops = &ipmi_wdog_fops | |
865 | }; | |
866 | ||
1da177e4 LT |
867 | static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, |
868 | void *handler_data) | |
869 | { | |
870 | if (msg->msg.data[0] != 0) { | |
871 | printk(KERN_ERR PFX "response: Error %x on cmd %x\n", | |
872 | msg->msg.data[0], | |
873 | msg->msg.cmd); | |
874 | } | |
875 | ||
876 | ipmi_free_recv_msg(msg); | |
877 | } | |
878 | ||
879 | static void ipmi_wdog_pretimeout_handler(void *handler_data) | |
880 | { | |
881 | if (preaction_val != WDOG_PRETIMEOUT_NONE) { | |
b385676b CM |
882 | if (preop_val == WDOG_PREOP_PANIC) { |
883 | if (atomic_inc_and_test(&preop_panic_excl)) | |
884 | panic("Watchdog pre-timeout"); | |
885 | } else if (preop_val == WDOG_PREOP_GIVE_DATA) { | |
1da177e4 LT |
886 | spin_lock(&ipmi_read_lock); |
887 | data_to_read = 1; | |
888 | wake_up_interruptible(&read_q); | |
889 | kill_fasync(&fasync_q, SIGIO, POLL_IN); | |
890 | ||
891 | spin_unlock(&ipmi_read_lock); | |
892 | } | |
893 | } | |
894 | ||
895 | /* On some machines, the heartbeat will give | |
896 | an error and not work unless we re-enable | |
897 | the timer. So do so. */ | |
898 | pretimeout_since_last_heartbeat = 1; | |
899 | } | |
900 | ||
901 | static struct ipmi_user_hndl ipmi_hndlrs = | |
902 | { | |
903 | .ipmi_recv_hndl = ipmi_wdog_msg_handler, | |
904 | .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler | |
905 | }; | |
906 | ||
907 | static void ipmi_register_watchdog(int ipmi_intf) | |
908 | { | |
909 | int rv = -EBUSY; | |
910 | ||
911 | down_write(®ister_sem); | |
912 | if (watchdog_user) | |
913 | goto out; | |
914 | ||
b2c03941 CM |
915 | if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf)) |
916 | goto out; | |
917 | ||
918 | watchdog_ifnum = ipmi_intf; | |
919 | ||
1da177e4 LT |
920 | rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); |
921 | if (rv < 0) { | |
922 | printk(KERN_CRIT PFX "Unable to register with ipmi\n"); | |
923 | goto out; | |
924 | } | |
925 | ||
926 | ipmi_get_version(watchdog_user, | |
927 | &ipmi_version_major, | |
928 | &ipmi_version_minor); | |
929 | ||
930 | rv = misc_register(&ipmi_wdog_miscdev); | |
931 | if (rv < 0) { | |
932 | ipmi_destroy_user(watchdog_user); | |
933 | watchdog_user = NULL; | |
934 | printk(KERN_CRIT PFX "Unable to register misc device\n"); | |
935 | } | |
936 | ||
f64da958 CM |
937 | #ifdef HAVE_DIE_NMI_POST |
938 | if (nmi_handler_registered) { | |
939 | int old_pretimeout = pretimeout; | |
940 | int old_timeout = timeout; | |
941 | int old_preop_val = preop_val; | |
942 | ||
943 | /* Set the pretimeout to go off in a second and give | |
944 | ourselves plenty of time to stop the timer. */ | |
945 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; | |
946 | preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */ | |
947 | pretimeout = 99; | |
948 | timeout = 100; | |
949 | ||
950 | testing_nmi = 1; | |
951 | ||
952 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
953 | if (rv) { | |
954 | printk(KERN_WARNING PFX "Error starting timer to" | |
955 | " test NMI: 0x%x. The NMI pretimeout will" | |
956 | " likely not work\n", rv); | |
957 | rv = 0; | |
958 | goto out_restore; | |
959 | } | |
960 | ||
961 | msleep(1500); | |
962 | ||
963 | if (testing_nmi != 2) { | |
964 | printk(KERN_WARNING PFX "IPMI NMI didn't seem to" | |
965 | " occur. The NMI pretimeout will" | |
966 | " likely not work\n"); | |
967 | } | |
968 | out_restore: | |
969 | testing_nmi = 0; | |
970 | preop_val = old_preop_val; | |
971 | pretimeout = old_pretimeout; | |
972 | timeout = old_timeout; | |
973 | } | |
974 | #endif | |
975 | ||
1da177e4 LT |
976 | out: |
977 | up_write(®ister_sem); | |
978 | ||
979 | if ((start_now) && (rv == 0)) { | |
980 | /* Run from startup, so start the timer now. */ | |
981 | start_now = 0; /* Disable this function after first startup. */ | |
982 | ipmi_watchdog_state = action_val; | |
983 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
984 | printk(KERN_INFO PFX "Starting now!\n"); | |
f64da958 CM |
985 | } else { |
986 | /* Stop the timer now. */ | |
987 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
988 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
1da177e4 LT |
989 | } |
990 | } | |
991 | ||
b2c03941 CM |
992 | static void ipmi_unregister_watchdog(int ipmi_intf) |
993 | { | |
994 | int rv; | |
995 | ||
996 | down_write(®ister_sem); | |
997 | ||
998 | if (!watchdog_user) | |
999 | goto out; | |
1000 | ||
1001 | if (watchdog_ifnum != ipmi_intf) | |
1002 | goto out; | |
1003 | ||
1004 | /* Make sure no one can call us any more. */ | |
1005 | misc_deregister(&ipmi_wdog_miscdev); | |
1006 | ||
1007 | /* Wait to make sure the message makes it out. The lower layer has | |
1008 | pointers to our buffers, we want to make sure they are done before | |
1009 | we release our memory. */ | |
1010 | while (atomic_read(&set_timeout_tofree)) | |
1011 | schedule_timeout_uninterruptible(1); | |
1012 | ||
1013 | /* Disconnect from IPMI. */ | |
1014 | rv = ipmi_destroy_user(watchdog_user); | |
1015 | if (rv) { | |
1016 | printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n", | |
1017 | rv); | |
1018 | } | |
1019 | watchdog_user = NULL; | |
1020 | ||
1021 | out: | |
1022 | up_write(®ister_sem); | |
1023 | } | |
1024 | ||
f64da958 | 1025 | #ifdef HAVE_DIE_NMI_POST |
1da177e4 | 1026 | static int |
f64da958 | 1027 | ipmi_nmi(struct notifier_block *self, unsigned long val, void *data) |
1da177e4 | 1028 | { |
f64da958 CM |
1029 | if (val != DIE_NMI_POST) |
1030 | return NOTIFY_OK; | |
1031 | ||
1032 | if (testing_nmi) { | |
1033 | testing_nmi = 2; | |
1034 | return NOTIFY_STOP; | |
1035 | } | |
1036 | ||
8f05ee9a CM |
1037 | /* If we are not expecting a timeout, ignore it. */ |
1038 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) | |
f64da958 CM |
1039 | return NOTIFY_OK; |
1040 | ||
1041 | if (preaction_val != WDOG_PRETIMEOUT_NMI) | |
1042 | return NOTIFY_OK; | |
8f05ee9a | 1043 | |
1da177e4 LT |
1044 | /* If no one else handled the NMI, we assume it was the IPMI |
1045 | watchdog. */ | |
f64da958 | 1046 | if (preop_val == WDOG_PREOP_PANIC) { |
8f05ee9a CM |
1047 | /* On some machines, the heartbeat will give |
1048 | an error and not work unless we re-enable | |
1049 | the timer. So do so. */ | |
1050 | pretimeout_since_last_heartbeat = 1; | |
b385676b CM |
1051 | if (atomic_inc_and_test(&preop_panic_excl)) |
1052 | panic(PFX "pre-timeout"); | |
8f05ee9a | 1053 | } |
1da177e4 | 1054 | |
f64da958 | 1055 | return NOTIFY_STOP; |
1da177e4 LT |
1056 | } |
1057 | ||
f64da958 CM |
1058 | static struct notifier_block ipmi_nmi_handler = { |
1059 | .notifier_call = ipmi_nmi | |
1da177e4 LT |
1060 | }; |
1061 | #endif | |
1062 | ||
1063 | static int wdog_reboot_handler(struct notifier_block *this, | |
1064 | unsigned long code, | |
1065 | void *unused) | |
1066 | { | |
1067 | static int reboot_event_handled = 0; | |
1068 | ||
1069 | if ((watchdog_user) && (!reboot_event_handled)) { | |
1070 | /* Make sure we only do this once. */ | |
1071 | reboot_event_handled = 1; | |
1072 | ||
1073 | if (code == SYS_DOWN || code == SYS_HALT) { | |
1074 | /* Disable the WDT if we are shutting down. */ | |
1075 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
1076 | panic_halt_ipmi_set_timeout(); | |
96febe9f | 1077 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
1da177e4 | 1078 | /* Set a long timer to let the reboot happens, but |
96febe9f CM |
1079 | reboot if it hangs, but only if the watchdog |
1080 | timer was already running. */ | |
1da177e4 LT |
1081 | timeout = 120; |
1082 | pretimeout = 0; | |
1083 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; | |
1084 | panic_halt_ipmi_set_timeout(); | |
1085 | } | |
1086 | } | |
1087 | return NOTIFY_OK; | |
1088 | } | |
1089 | ||
1090 | static struct notifier_block wdog_reboot_notifier = { | |
1091 | .notifier_call = wdog_reboot_handler, | |
1092 | .next = NULL, | |
1093 | .priority = 0 | |
1094 | }; | |
1095 | ||
1096 | static int wdog_panic_handler(struct notifier_block *this, | |
1097 | unsigned long event, | |
1098 | void *unused) | |
1099 | { | |
1100 | static int panic_event_handled = 0; | |
1101 | ||
96febe9f CM |
1102 | /* On a panic, if we have a panic timeout, make sure to extend |
1103 | the watchdog timer to a reasonable value to complete the | |
1104 | panic, if the watchdog timer is running. Plus the | |
1105 | pretimeout is meaningless at panic time. */ | |
1106 | if (watchdog_user && !panic_event_handled && | |
1107 | ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
1108 | /* Make sure we do this only once. */ | |
1da177e4 LT |
1109 | panic_event_handled = 1; |
1110 | ||
1111 | timeout = 255; | |
1112 | pretimeout = 0; | |
1da177e4 LT |
1113 | panic_halt_ipmi_set_timeout(); |
1114 | } | |
1115 | ||
1116 | return NOTIFY_OK; | |
1117 | } | |
1118 | ||
1119 | static struct notifier_block wdog_panic_notifier = { | |
1120 | .notifier_call = wdog_panic_handler, | |
1121 | .next = NULL, | |
1122 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ | |
1123 | }; | |
1124 | ||
1125 | ||
50c812b2 | 1126 | static void ipmi_new_smi(int if_num, struct device *device) |
1da177e4 LT |
1127 | { |
1128 | ipmi_register_watchdog(if_num); | |
1129 | } | |
1130 | ||
1131 | static void ipmi_smi_gone(int if_num) | |
1132 | { | |
b2c03941 | 1133 | ipmi_unregister_watchdog(if_num); |
1da177e4 LT |
1134 | } |
1135 | ||
1136 | static struct ipmi_smi_watcher smi_watcher = | |
1137 | { | |
1138 | .owner = THIS_MODULE, | |
1139 | .new_smi = ipmi_new_smi, | |
1140 | .smi_gone = ipmi_smi_gone | |
1141 | }; | |
1142 | ||
cc4673ee | 1143 | static int action_op(const char *inval, char *outval) |
1da177e4 | 1144 | { |
cc4673ee CM |
1145 | if (outval) |
1146 | strcpy(outval, action); | |
1147 | ||
1148 | if (!inval) | |
1149 | return 0; | |
1da177e4 | 1150 | |
cc4673ee | 1151 | if (strcmp(inval, "reset") == 0) |
1da177e4 | 1152 | action_val = WDOG_TIMEOUT_RESET; |
cc4673ee | 1153 | else if (strcmp(inval, "none") == 0) |
1da177e4 | 1154 | action_val = WDOG_TIMEOUT_NONE; |
cc4673ee | 1155 | else if (strcmp(inval, "power_cycle") == 0) |
1da177e4 | 1156 | action_val = WDOG_TIMEOUT_POWER_CYCLE; |
cc4673ee | 1157 | else if (strcmp(inval, "power_off") == 0) |
1da177e4 | 1158 | action_val = WDOG_TIMEOUT_POWER_DOWN; |
cc4673ee CM |
1159 | else |
1160 | return -EINVAL; | |
1161 | strcpy(action, inval); | |
1162 | return 0; | |
1163 | } | |
1164 | ||
1165 | static int preaction_op(const char *inval, char *outval) | |
1166 | { | |
1167 | if (outval) | |
1168 | strcpy(outval, preaction); | |
1da177e4 | 1169 | |
cc4673ee CM |
1170 | if (!inval) |
1171 | return 0; | |
1172 | ||
1173 | if (strcmp(inval, "pre_none") == 0) | |
1da177e4 | 1174 | preaction_val = WDOG_PRETIMEOUT_NONE; |
cc4673ee | 1175 | else if (strcmp(inval, "pre_smi") == 0) |
1da177e4 | 1176 | preaction_val = WDOG_PRETIMEOUT_SMI; |
f64da958 | 1177 | #ifdef HAVE_DIE_NMI_POST |
cc4673ee | 1178 | else if (strcmp(inval, "pre_nmi") == 0) |
1da177e4 LT |
1179 | preaction_val = WDOG_PRETIMEOUT_NMI; |
1180 | #endif | |
cc4673ee | 1181 | else if (strcmp(inval, "pre_int") == 0) |
1da177e4 | 1182 | preaction_val = WDOG_PRETIMEOUT_MSG_INT; |
cc4673ee CM |
1183 | else |
1184 | return -EINVAL; | |
1185 | strcpy(preaction, inval); | |
1186 | return 0; | |
1187 | } | |
1188 | ||
1189 | static int preop_op(const char *inval, char *outval) | |
1190 | { | |
1191 | if (outval) | |
1192 | strcpy(outval, preop); | |
1da177e4 | 1193 | |
cc4673ee CM |
1194 | if (!inval) |
1195 | return 0; | |
1196 | ||
1197 | if (strcmp(inval, "preop_none") == 0) | |
1da177e4 | 1198 | preop_val = WDOG_PREOP_NONE; |
cc4673ee | 1199 | else if (strcmp(inval, "preop_panic") == 0) |
1da177e4 | 1200 | preop_val = WDOG_PREOP_PANIC; |
cc4673ee | 1201 | else if (strcmp(inval, "preop_give_data") == 0) |
1da177e4 | 1202 | preop_val = WDOG_PREOP_GIVE_DATA; |
cc4673ee CM |
1203 | else |
1204 | return -EINVAL; | |
1205 | strcpy(preop, inval); | |
1206 | return 0; | |
1207 | } | |
1da177e4 | 1208 | |
cc4673ee CM |
1209 | static void check_parms(void) |
1210 | { | |
f64da958 | 1211 | #ifdef HAVE_DIE_NMI_POST |
cc4673ee CM |
1212 | int do_nmi = 0; |
1213 | int rv; | |
1214 | ||
1da177e4 | 1215 | if (preaction_val == WDOG_PRETIMEOUT_NMI) { |
cc4673ee | 1216 | do_nmi = 1; |
1da177e4 LT |
1217 | if (preop_val == WDOG_PREOP_GIVE_DATA) { |
1218 | printk(KERN_WARNING PFX "Pretimeout op is to give data" | |
1219 | " but NMI pretimeout is enabled, setting" | |
1220 | " pretimeout op to none\n"); | |
cc4673ee CM |
1221 | preop_op("preop_none", NULL); |
1222 | do_nmi = 0; | |
1da177e4 | 1223 | } |
cc4673ee CM |
1224 | } |
1225 | if (do_nmi && !nmi_handler_registered) { | |
f64da958 | 1226 | rv = register_die_notifier(&ipmi_nmi_handler); |
1da177e4 | 1227 | if (rv) { |
cc4673ee CM |
1228 | printk(KERN_WARNING PFX |
1229 | "Can't register nmi handler\n"); | |
1230 | return; | |
1231 | } else | |
1232 | nmi_handler_registered = 1; | |
1233 | } else if (!do_nmi && nmi_handler_registered) { | |
f64da958 | 1234 | unregister_die_notifier(&ipmi_nmi_handler); |
cc4673ee | 1235 | nmi_handler_registered = 0; |
1da177e4 LT |
1236 | } |
1237 | #endif | |
cc4673ee CM |
1238 | } |
1239 | ||
1240 | static int __init ipmi_wdog_init(void) | |
1241 | { | |
1242 | int rv; | |
1243 | ||
1244 | if (action_op(action, NULL)) { | |
1245 | action_op("reset", NULL); | |
1246 | printk(KERN_INFO PFX "Unknown action '%s', defaulting to" | |
1247 | " reset\n", action); | |
1248 | } | |
1249 | ||
1250 | if (preaction_op(preaction, NULL)) { | |
1251 | preaction_op("pre_none", NULL); | |
1252 | printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to" | |
1253 | " none\n", preaction); | |
1254 | } | |
1255 | ||
1256 | if (preop_op(preop, NULL)) { | |
1257 | preop_op("preop_none", NULL); | |
1258 | printk(KERN_INFO PFX "Unknown preop '%s', defaulting to" | |
1259 | " none\n", preop); | |
1260 | } | |
1261 | ||
1262 | check_parms(); | |
1da177e4 | 1263 | |
b2c03941 CM |
1264 | register_reboot_notifier(&wdog_reboot_notifier); |
1265 | atomic_notifier_chain_register(&panic_notifier_list, | |
1266 | &wdog_panic_notifier); | |
1267 | ||
1da177e4 LT |
1268 | rv = ipmi_smi_watcher_register(&smi_watcher); |
1269 | if (rv) { | |
f64da958 CM |
1270 | #ifdef HAVE_DIE_NMI_POST |
1271 | if (nmi_handler_registered) | |
1272 | unregister_die_notifier(&ipmi_nmi_handler); | |
1da177e4 | 1273 | #endif |
b2c03941 CM |
1274 | atomic_notifier_chain_unregister(&panic_notifier_list, |
1275 | &wdog_panic_notifier); | |
1276 | unregister_reboot_notifier(&wdog_reboot_notifier); | |
1da177e4 LT |
1277 | printk(KERN_WARNING PFX "can't register smi watcher\n"); |
1278 | return rv; | |
1279 | } | |
1280 | ||
1fdd75bd CM |
1281 | printk(KERN_INFO PFX "driver initialized\n"); |
1282 | ||
1da177e4 LT |
1283 | return 0; |
1284 | } | |
1285 | ||
b2c03941 | 1286 | static void __exit ipmi_wdog_exit(void) |
1da177e4 | 1287 | { |
b2c03941 CM |
1288 | ipmi_smi_watcher_unregister(&smi_watcher); |
1289 | ipmi_unregister_watchdog(watchdog_ifnum); | |
1da177e4 | 1290 | |
f64da958 | 1291 | #ifdef HAVE_DIE_NMI_POST |
cc4673ee | 1292 | if (nmi_handler_registered) |
f64da958 | 1293 | unregister_die_notifier(&ipmi_nmi_handler); |
1da177e4 LT |
1294 | #endif |
1295 | ||
e041c683 | 1296 | atomic_notifier_chain_unregister(&panic_notifier_list, |
b2c03941 | 1297 | &wdog_panic_notifier); |
1da177e4 | 1298 | unregister_reboot_notifier(&wdog_reboot_notifier); |
1da177e4 LT |
1299 | } |
1300 | module_exit(ipmi_wdog_exit); | |
1301 | module_init(ipmi_wdog_init); | |
1302 | MODULE_LICENSE("GPL"); | |
1fdd75bd CM |
1303 | MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); |
1304 | MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); |