]>
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> | |
609146fd | 38 | #include <linux/mutex.h> |
1da177e4 LT |
39 | #include <linux/watchdog.h> |
40 | #include <linux/miscdevice.h> | |
41 | #include <linux/init.h> | |
d6dfd131 | 42 | #include <linux/completion.h> |
1eeb66a1 | 43 | #include <linux/kdebug.h> |
1da177e4 LT |
44 | #include <linux/rwsem.h> |
45 | #include <linux/errno.h> | |
46 | #include <asm/uaccess.h> | |
47 | #include <linux/notifier.h> | |
48 | #include <linux/nmi.h> | |
49 | #include <linux/reboot.h> | |
50 | #include <linux/wait.h> | |
51 | #include <linux/poll.h> | |
cc4673ee CM |
52 | #include <linux/string.h> |
53 | #include <linux/ctype.h> | |
612b5a8d | 54 | #include <linux/delay.h> |
60063497 | 55 | #include <linux/atomic.h> |
f64da958 | 56 | |
612b5a8d | 57 | #ifdef CONFIG_X86 |
36c7dc44 CM |
58 | /* |
59 | * This is ugly, but I've determined that x86 is the only architecture | |
60 | * that can reasonably support the IPMI NMI watchdog timeout at this | |
61 | * time. If another architecture adds this capability somehow, it | |
62 | * will have to be a somewhat different mechanism and I have no idea | |
63 | * how it will work. So in the unlikely event that another | |
64 | * architecture supports this, we can figure out a good generic | |
65 | * mechanism for it at that time. | |
66 | */ | |
612b5a8d CM |
67 | #include <asm/kdebug.h> |
68 | #define HAVE_DIE_NMI | |
1da177e4 LT |
69 | #endif |
70 | ||
71 | #define PFX "IPMI Watchdog: " | |
72 | ||
1da177e4 LT |
73 | /* |
74 | * The IPMI command/response information for the watchdog timer. | |
75 | */ | |
76 | ||
77 | /* values for byte 1 of the set command, byte 2 of the get response. */ | |
78 | #define WDOG_DONT_LOG (1 << 7) | |
79 | #define WDOG_DONT_STOP_ON_SET (1 << 6) | |
80 | #define WDOG_SET_TIMER_USE(byte, use) \ | |
81 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
82 | #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) | |
83 | #define WDOG_TIMER_USE_BIOS_FRB2 1 | |
84 | #define WDOG_TIMER_USE_BIOS_POST 2 | |
85 | #define WDOG_TIMER_USE_OS_LOAD 3 | |
86 | #define WDOG_TIMER_USE_SMS_OS 4 | |
87 | #define WDOG_TIMER_USE_OEM 5 | |
88 | ||
89 | /* values for byte 2 of the set command, byte 3 of the get response. */ | |
90 | #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ | |
91 | byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) | |
92 | #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) | |
93 | #define WDOG_PRETIMEOUT_NONE 0 | |
94 | #define WDOG_PRETIMEOUT_SMI 1 | |
95 | #define WDOG_PRETIMEOUT_NMI 2 | |
96 | #define WDOG_PRETIMEOUT_MSG_INT 3 | |
97 | ||
98 | /* Operations that can be performed on a pretimout. */ | |
99 | #define WDOG_PREOP_NONE 0 | |
100 | #define WDOG_PREOP_PANIC 1 | |
36c7dc44 CM |
101 | /* Cause data to be available to read. Doesn't work in NMI mode. */ |
102 | #define WDOG_PREOP_GIVE_DATA 2 | |
1da177e4 LT |
103 | |
104 | /* Actions to perform on a full timeout. */ | |
105 | #define WDOG_SET_TIMEOUT_ACT(byte, use) \ | |
106 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
107 | #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) | |
108 | #define WDOG_TIMEOUT_NONE 0 | |
109 | #define WDOG_TIMEOUT_RESET 1 | |
110 | #define WDOG_TIMEOUT_POWER_DOWN 2 | |
111 | #define WDOG_TIMEOUT_POWER_CYCLE 3 | |
112 | ||
36c7dc44 CM |
113 | /* |
114 | * Byte 3 of the get command, byte 4 of the get response is the | |
115 | * pre-timeout in seconds. | |
116 | */ | |
1da177e4 LT |
117 | |
118 | /* Bits for setting byte 4 of the set command, byte 5 of the get response. */ | |
119 | #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1) | |
120 | #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2) | |
121 | #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3) | |
122 | #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4) | |
123 | #define WDOG_EXPIRE_CLEAR_OEM (1 << 5) | |
124 | ||
36c7dc44 CM |
125 | /* |
126 | * Setting/getting the watchdog timer value. This is for bytes 5 and | |
127 | * 6 (the timeout time) of the set command, and bytes 6 and 7 (the | |
128 | * timeout time) and 8 and 9 (the current countdown value) of the | |
129 | * response. The timeout value is given in seconds (in the command it | |
130 | * is 100ms intervals). | |
131 | */ | |
1da177e4 LT |
132 | #define WDOG_SET_TIMEOUT(byte1, byte2, val) \ |
133 | (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8) | |
134 | #define WDOG_GET_TIMEOUT(byte1, byte2) \ | |
135 | (((byte1) | ((byte2) << 8)) / 10) | |
136 | ||
137 | #define IPMI_WDOG_RESET_TIMER 0x22 | |
138 | #define IPMI_WDOG_SET_TIMER 0x24 | |
139 | #define IPMI_WDOG_GET_TIMER 0x25 | |
140 | ||
141 | /* These are here until the real ones get into the watchdog.h interface. */ | |
142 | #ifndef WDIOC_GETTIMEOUT | |
143 | #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int) | |
144 | #endif | |
145 | #ifndef WDIOC_SET_PRETIMEOUT | |
146 | #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int) | |
147 | #endif | |
148 | #ifndef WDIOC_GET_PRETIMEOUT | |
149 | #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int) | |
150 | #endif | |
151 | ||
609146fd | 152 | static DEFINE_MUTEX(ipmi_watchdog_mutex); |
4bfdf378 | 153 | static int nowayout = WATCHDOG_NOWAYOUT; |
1da177e4 | 154 | |
0c8204b3 | 155 | static ipmi_user_t watchdog_user; |
b2c03941 | 156 | static int watchdog_ifnum; |
1da177e4 LT |
157 | |
158 | /* Default the timeout to 10 seconds. */ | |
159 | static int timeout = 10; | |
160 | ||
161 | /* The pre-timeout is disabled by default. */ | |
0c8204b3 | 162 | static int pretimeout; |
1da177e4 LT |
163 | |
164 | /* Default action is to reset the board on a timeout. */ | |
165 | static unsigned char action_val = WDOG_TIMEOUT_RESET; | |
166 | ||
167 | static char action[16] = "reset"; | |
168 | ||
169 | static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; | |
170 | ||
171 | static char preaction[16] = "pre_none"; | |
172 | ||
173 | static unsigned char preop_val = WDOG_PREOP_NONE; | |
174 | ||
175 | static char preop[16] = "preop_none"; | |
176 | static DEFINE_SPINLOCK(ipmi_read_lock); | |
0c8204b3 | 177 | static char data_to_read; |
1da177e4 | 178 | static DECLARE_WAIT_QUEUE_HEAD(read_q); |
0c8204b3 RD |
179 | static struct fasync_struct *fasync_q; |
180 | static char pretimeout_since_last_heartbeat; | |
1da177e4 LT |
181 | static char expect_close; |
182 | ||
b2c03941 CM |
183 | static int ifnum_to_use = -1; |
184 | ||
cc4673ee CM |
185 | /* Parameters to ipmi_set_timeout */ |
186 | #define IPMI_SET_TIMEOUT_NO_HB 0 | |
187 | #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 | |
188 | #define IPMI_SET_TIMEOUT_FORCE_HB 2 | |
189 | ||
190 | static int ipmi_set_timeout(int do_heartbeat); | |
b2c03941 CM |
191 | static void ipmi_register_watchdog(int ipmi_intf); |
192 | static void ipmi_unregister_watchdog(int ipmi_intf); | |
cc4673ee | 193 | |
36c7dc44 CM |
194 | /* |
195 | * If true, the driver will start running as soon as it is configured | |
196 | * and ready. | |
197 | */ | |
0c8204b3 | 198 | static int start_now; |
1da177e4 | 199 | |
c8ba6c52 | 200 | static int set_param_timeout(const char *val, const struct kernel_param *kp) |
cc4673ee CM |
201 | { |
202 | char *endp; | |
203 | int l; | |
204 | int rv = 0; | |
205 | ||
206 | if (!val) | |
207 | return -EINVAL; | |
208 | l = simple_strtoul(val, &endp, 0); | |
209 | if (endp == val) | |
210 | return -EINVAL; | |
211 | ||
cc4673ee CM |
212 | *((int *)kp->arg) = l; |
213 | if (watchdog_user) | |
214 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
cc4673ee CM |
215 | |
216 | return rv; | |
217 | } | |
218 | ||
c8ba6c52 RR |
219 | static struct kernel_param_ops param_ops_timeout = { |
220 | .set = set_param_timeout, | |
221 | .get = param_get_int, | |
222 | }; | |
223 | #define param_check_timeout param_check_int | |
cc4673ee CM |
224 | |
225 | typedef int (*action_fn)(const char *intval, char *outval); | |
226 | ||
227 | static int action_op(const char *inval, char *outval); | |
228 | static int preaction_op(const char *inval, char *outval); | |
229 | static int preop_op(const char *inval, char *outval); | |
230 | static void check_parms(void); | |
231 | ||
c8ba6c52 | 232 | static int set_param_str(const char *val, const struct kernel_param *kp) |
cc4673ee CM |
233 | { |
234 | action_fn fn = (action_fn) kp->arg; | |
235 | int rv = 0; | |
43cdff92 SD |
236 | char valcp[16]; |
237 | char *s; | |
66f969d0 | 238 | |
43cdff92 SD |
239 | strncpy(valcp, val, 16); |
240 | valcp[15] = '\0'; | |
66f969d0 | 241 | |
43cdff92 | 242 | s = strstrip(valcp); |
cc4673ee | 243 | |
66f969d0 | 244 | rv = fn(s, NULL); |
cc4673ee | 245 | if (rv) |
f8fbcd3b | 246 | goto out; |
cc4673ee CM |
247 | |
248 | check_parms(); | |
249 | if (watchdog_user) | |
250 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
251 | ||
f8fbcd3b | 252 | out: |
cc4673ee CM |
253 | return rv; |
254 | } | |
255 | ||
c8ba6c52 | 256 | static int get_param_str(char *buffer, const struct kernel_param *kp) |
cc4673ee CM |
257 | { |
258 | action_fn fn = (action_fn) kp->arg; | |
259 | int rv; | |
260 | ||
261 | rv = fn(NULL, buffer); | |
262 | if (rv) | |
263 | return rv; | |
264 | return strlen(buffer); | |
265 | } | |
266 | ||
b2c03941 | 267 | |
c8ba6c52 | 268 | static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp) |
b2c03941 CM |
269 | { |
270 | int rv = param_set_int(val, kp); | |
271 | if (rv) | |
272 | return rv; | |
273 | if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum)) | |
274 | return 0; | |
275 | ||
276 | ipmi_unregister_watchdog(watchdog_ifnum); | |
277 | ipmi_register_watchdog(ifnum_to_use); | |
278 | return 0; | |
279 | } | |
280 | ||
c8ba6c52 RR |
281 | static struct kernel_param_ops param_ops_wdog_ifnum = { |
282 | .set = set_param_wdog_ifnum, | |
283 | .get = param_get_int, | |
284 | }; | |
285 | ||
286 | #define param_check_wdog_ifnum param_check_int | |
287 | ||
288 | static struct kernel_param_ops param_ops_str = { | |
289 | .set = set_param_str, | |
290 | .get = get_param_str, | |
291 | }; | |
292 | ||
293 | module_param(ifnum_to_use, wdog_ifnum, 0644); | |
b2c03941 CM |
294 | MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog " |
295 | "timer. Setting to -1 defaults to the first registered " | |
296 | "interface"); | |
297 | ||
c8ba6c52 | 298 | module_param(timeout, timeout, 0644); |
1da177e4 | 299 | MODULE_PARM_DESC(timeout, "Timeout value in seconds."); |
cc4673ee | 300 | |
c8ba6c52 | 301 | module_param(pretimeout, timeout, 0644); |
1da177e4 | 302 | MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); |
cc4673ee | 303 | |
c8ba6c52 | 304 | module_param_cb(action, ¶m_ops_str, action_op, 0644); |
1da177e4 LT |
305 | MODULE_PARM_DESC(action, "Timeout action. One of: " |
306 | "reset, none, power_cycle, power_off."); | |
cc4673ee | 307 | |
c8ba6c52 | 308 | module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644); |
1da177e4 LT |
309 | MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " |
310 | "pre_none, pre_smi, pre_nmi, pre_int."); | |
cc4673ee | 311 | |
c8ba6c52 | 312 | module_param_cb(preop, ¶m_ops_str, preop_op, 0644); |
1da177e4 LT |
313 | MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " |
314 | "preop_none, preop_panic, preop_give_data."); | |
cc4673ee | 315 | |
b2c03941 | 316 | module_param(start_now, int, 0444); |
1da177e4 LT |
317 | MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" |
318 | "soon as the driver is loaded."); | |
cc4673ee CM |
319 | |
320 | module_param(nowayout, int, 0644); | |
b2c03941 CM |
321 | MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started " |
322 | "(default=CONFIG_WATCHDOG_NOWAYOUT)"); | |
1da177e4 LT |
323 | |
324 | /* Default state of the timer. */ | |
325 | static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
326 | ||
327 | /* If shutting down via IPMI, we ignore the heartbeat. */ | |
0c8204b3 | 328 | static int ipmi_ignore_heartbeat; |
1da177e4 LT |
329 | |
330 | /* Is someone using the watchdog? Only one user is allowed. */ | |
0c8204b3 | 331 | static unsigned long ipmi_wdog_open; |
1da177e4 | 332 | |
36c7dc44 CM |
333 | /* |
334 | * If set to 1, the heartbeat command will set the state to reset and | |
335 | * start the timer. The timer doesn't normally run when the driver is | |
336 | * first opened until the heartbeat is set the first time, this | |
337 | * variable is used to accomplish this. | |
338 | */ | |
0c8204b3 | 339 | static int ipmi_start_timer_on_heartbeat; |
1da177e4 LT |
340 | |
341 | /* IPMI version of the BMC. */ | |
342 | static unsigned char ipmi_version_major; | |
343 | static unsigned char ipmi_version_minor; | |
344 | ||
b385676b CM |
345 | /* If a pretimeout occurs, this is used to allow only one panic to happen. */ |
346 | static atomic_t preop_panic_excl = ATOMIC_INIT(-1); | |
1da177e4 | 347 | |
612b5a8d CM |
348 | #ifdef HAVE_DIE_NMI |
349 | static int testing_nmi; | |
350 | static int nmi_handler_registered; | |
351 | #endif | |
352 | ||
1da177e4 | 353 | static int ipmi_heartbeat(void); |
1da177e4 | 354 | |
36c7dc44 CM |
355 | /* |
356 | * We use a mutex to make sure that only one thing can send a set | |
357 | * timeout at one time, because we only have one copy of the data. | |
358 | * The mutex is claimed when the set_timeout is sent and freed | |
359 | * when both messages are free. | |
360 | */ | |
1da177e4 | 361 | static atomic_t set_timeout_tofree = ATOMIC_INIT(0); |
d6dfd131 CM |
362 | static DEFINE_MUTEX(set_timeout_lock); |
363 | static DECLARE_COMPLETION(set_timeout_wait); | |
1da177e4 LT |
364 | static void set_timeout_free_smi(struct ipmi_smi_msg *msg) |
365 | { | |
366 | if (atomic_dec_and_test(&set_timeout_tofree)) | |
d6dfd131 | 367 | complete(&set_timeout_wait); |
1da177e4 LT |
368 | } |
369 | static void set_timeout_free_recv(struct ipmi_recv_msg *msg) | |
370 | { | |
371 | if (atomic_dec_and_test(&set_timeout_tofree)) | |
d6dfd131 | 372 | complete(&set_timeout_wait); |
1da177e4 | 373 | } |
36c7dc44 | 374 | static struct ipmi_smi_msg set_timeout_smi_msg = { |
1da177e4 LT |
375 | .done = set_timeout_free_smi |
376 | }; | |
36c7dc44 | 377 | static struct ipmi_recv_msg set_timeout_recv_msg = { |
1da177e4 LT |
378 | .done = set_timeout_free_recv |
379 | }; | |
36c7dc44 | 380 | |
1da177e4 LT |
381 | static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, |
382 | struct ipmi_recv_msg *recv_msg, | |
383 | int *send_heartbeat_now) | |
384 | { | |
385 | struct kernel_ipmi_msg msg; | |
386 | unsigned char data[6]; | |
387 | int rv; | |
388 | struct ipmi_system_interface_addr addr; | |
389 | int hbnow = 0; | |
390 | ||
391 | ||
612b5a8d CM |
392 | /* These can be cleared as we are setting the timeout. */ |
393 | pretimeout_since_last_heartbeat = 0; | |
394 | ||
1da177e4 LT |
395 | data[0] = 0; |
396 | WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); | |
397 | ||
398 | if ((ipmi_version_major > 1) | |
36c7dc44 | 399 | || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) { |
1da177e4 LT |
400 | /* This is an IPMI 1.5-only feature. */ |
401 | data[0] |= WDOG_DONT_STOP_ON_SET; | |
402 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
36c7dc44 CM |
403 | /* |
404 | * In ipmi 1.0, setting the timer stops the watchdog, we | |
405 | * need to start it back up again. | |
406 | */ | |
1da177e4 LT |
407 | hbnow = 1; |
408 | } | |
409 | ||
410 | data[1] = 0; | |
411 | WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); | |
8f05ee9a | 412 | if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { |
1da177e4 LT |
413 | WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); |
414 | data[2] = pretimeout; | |
415 | } else { | |
416 | WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); | |
417 | data[2] = 0; /* No pretimeout. */ | |
418 | } | |
419 | data[3] = 0; | |
420 | WDOG_SET_TIMEOUT(data[4], data[5], timeout); | |
421 | ||
422 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
423 | addr.channel = IPMI_BMC_CHANNEL; | |
424 | addr.lun = 0; | |
425 | ||
426 | msg.netfn = 0x06; | |
427 | msg.cmd = IPMI_WDOG_SET_TIMER; | |
428 | msg.data = data; | |
429 | msg.data_len = sizeof(data); | |
430 | rv = ipmi_request_supply_msgs(watchdog_user, | |
431 | (struct ipmi_addr *) &addr, | |
432 | 0, | |
433 | &msg, | |
434 | NULL, | |
435 | smi_msg, | |
436 | recv_msg, | |
437 | 1); | |
438 | if (rv) { | |
439 | printk(KERN_WARNING PFX "set timeout error: %d\n", | |
440 | rv); | |
441 | } | |
442 | ||
443 | if (send_heartbeat_now) | |
444 | *send_heartbeat_now = hbnow; | |
445 | ||
446 | return rv; | |
447 | } | |
448 | ||
1da177e4 LT |
449 | static int ipmi_set_timeout(int do_heartbeat) |
450 | { | |
451 | int send_heartbeat_now; | |
452 | int rv; | |
453 | ||
454 | ||
455 | /* We can only send one of these at a time. */ | |
d6dfd131 | 456 | mutex_lock(&set_timeout_lock); |
1da177e4 LT |
457 | |
458 | atomic_set(&set_timeout_tofree, 2); | |
459 | ||
460 | rv = i_ipmi_set_timeout(&set_timeout_smi_msg, | |
461 | &set_timeout_recv_msg, | |
462 | &send_heartbeat_now); | |
463 | if (rv) { | |
d6dfd131 CM |
464 | mutex_unlock(&set_timeout_lock); |
465 | goto out; | |
466 | } | |
467 | ||
468 | wait_for_completion(&set_timeout_wait); | |
469 | ||
612b5a8d CM |
470 | mutex_unlock(&set_timeout_lock); |
471 | ||
d6dfd131 CM |
472 | if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) |
473 | || ((send_heartbeat_now) | |
474 | && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) | |
d6dfd131 | 475 | rv = ipmi_heartbeat(); |
1da177e4 | 476 | |
d6dfd131 | 477 | out: |
1da177e4 LT |
478 | return rv; |
479 | } | |
480 | ||
fcfa4724 CM |
481 | static atomic_t panic_done_count = ATOMIC_INIT(0); |
482 | ||
483 | static void panic_smi_free(struct ipmi_smi_msg *msg) | |
1da177e4 | 484 | { |
fcfa4724 | 485 | atomic_dec(&panic_done_count); |
1da177e4 | 486 | } |
fcfa4724 CM |
487 | static void panic_recv_free(struct ipmi_recv_msg *msg) |
488 | { | |
489 | atomic_dec(&panic_done_count); | |
490 | } | |
491 | ||
36c7dc44 | 492 | static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = { |
fcfa4724 CM |
493 | .done = panic_smi_free |
494 | }; | |
36c7dc44 | 495 | static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = { |
fcfa4724 CM |
496 | .done = panic_recv_free |
497 | }; | |
498 | ||
499 | static void panic_halt_ipmi_heartbeat(void) | |
500 | { | |
501 | struct kernel_ipmi_msg msg; | |
502 | struct ipmi_system_interface_addr addr; | |
503 | int rv; | |
504 | ||
36c7dc44 CM |
505 | /* |
506 | * Don't reset the timer if we have the timer turned off, that | |
507 | * re-enables the watchdog. | |
508 | */ | |
fcfa4724 CM |
509 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
510 | return; | |
511 | ||
512 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
513 | addr.channel = IPMI_BMC_CHANNEL; | |
514 | addr.lun = 0; | |
515 | ||
516 | msg.netfn = 0x06; | |
517 | msg.cmd = IPMI_WDOG_RESET_TIMER; | |
518 | msg.data = NULL; | |
519 | msg.data_len = 0; | |
520 | rv = ipmi_request_supply_msgs(watchdog_user, | |
521 | (struct ipmi_addr *) &addr, | |
522 | 0, | |
523 | &msg, | |
524 | NULL, | |
525 | &panic_halt_heartbeat_smi_msg, | |
526 | &panic_halt_heartbeat_recv_msg, | |
527 | 1); | |
528 | if (!rv) | |
529 | atomic_add(2, &panic_done_count); | |
1da177e4 | 530 | } |
fcfa4724 | 531 | |
36c7dc44 | 532 | static struct ipmi_smi_msg panic_halt_smi_msg = { |
fcfa4724 | 533 | .done = panic_smi_free |
1da177e4 | 534 | }; |
36c7dc44 | 535 | static struct ipmi_recv_msg panic_halt_recv_msg = { |
fcfa4724 | 536 | .done = panic_recv_free |
1da177e4 LT |
537 | }; |
538 | ||
36c7dc44 CM |
539 | /* |
540 | * Special call, doesn't claim any locks. This is only to be called | |
541 | * at panic or halt time, in run-to-completion mode, when the caller | |
542 | * is the only CPU and the only thing that will be going is these IPMI | |
543 | * calls. | |
544 | */ | |
1da177e4 LT |
545 | static void panic_halt_ipmi_set_timeout(void) |
546 | { | |
547 | int send_heartbeat_now; | |
548 | int rv; | |
549 | ||
fcfa4724 CM |
550 | /* Wait for the messages to be free. */ |
551 | while (atomic_read(&panic_done_count) != 0) | |
552 | ipmi_poll_interface(watchdog_user); | |
1da177e4 LT |
553 | rv = i_ipmi_set_timeout(&panic_halt_smi_msg, |
554 | &panic_halt_recv_msg, | |
555 | &send_heartbeat_now); | |
556 | if (!rv) { | |
fcfa4724 | 557 | atomic_add(2, &panic_done_count); |
1da177e4 LT |
558 | if (send_heartbeat_now) |
559 | panic_halt_ipmi_heartbeat(); | |
fcfa4724 CM |
560 | } else |
561 | printk(KERN_WARNING PFX | |
562 | "Unable to extend the watchdog timeout."); | |
563 | while (atomic_read(&panic_done_count) != 0) | |
564 | ipmi_poll_interface(watchdog_user); | |
1da177e4 LT |
565 | } |
566 | ||
36c7dc44 CM |
567 | /* |
568 | * We use a mutex to make sure that only one thing can send a | |
569 | * heartbeat at one time, because we only have one copy of the data. | |
570 | * The semaphore is claimed when the set_timeout is sent and freed | |
571 | * when both messages are free. | |
572 | */ | |
1da177e4 | 573 | static atomic_t heartbeat_tofree = ATOMIC_INIT(0); |
d6dfd131 CM |
574 | static DEFINE_MUTEX(heartbeat_lock); |
575 | static DECLARE_COMPLETION(heartbeat_wait); | |
1da177e4 LT |
576 | static void heartbeat_free_smi(struct ipmi_smi_msg *msg) |
577 | { | |
578 | if (atomic_dec_and_test(&heartbeat_tofree)) | |
d6dfd131 | 579 | complete(&heartbeat_wait); |
1da177e4 LT |
580 | } |
581 | static void heartbeat_free_recv(struct ipmi_recv_msg *msg) | |
582 | { | |
583 | if (atomic_dec_and_test(&heartbeat_tofree)) | |
d6dfd131 | 584 | complete(&heartbeat_wait); |
1da177e4 | 585 | } |
36c7dc44 | 586 | static struct ipmi_smi_msg heartbeat_smi_msg = { |
1da177e4 LT |
587 | .done = heartbeat_free_smi |
588 | }; | |
36c7dc44 | 589 | static struct ipmi_recv_msg heartbeat_recv_msg = { |
1da177e4 LT |
590 | .done = heartbeat_free_recv |
591 | }; | |
36c7dc44 | 592 | |
1da177e4 LT |
593 | static int ipmi_heartbeat(void) |
594 | { | |
595 | struct kernel_ipmi_msg msg; | |
596 | int rv; | |
597 | struct ipmi_system_interface_addr addr; | |
598 | ||
612b5a8d | 599 | if (ipmi_ignore_heartbeat) |
1da177e4 | 600 | return 0; |
1da177e4 LT |
601 | |
602 | if (ipmi_start_timer_on_heartbeat) { | |
faa8b6c3 | 603 | ipmi_start_timer_on_heartbeat = 0; |
1da177e4 LT |
604 | ipmi_watchdog_state = action_val; |
605 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
606 | } else if (pretimeout_since_last_heartbeat) { | |
36c7dc44 CM |
607 | /* |
608 | * A pretimeout occurred, make sure we set the timeout. | |
609 | * We don't want to set the action, though, we want to | |
610 | * leave that alone (thus it can't be combined with the | |
611 | * above operation. | |
612 | */ | |
1da177e4 LT |
613 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
614 | } | |
615 | ||
d6dfd131 | 616 | mutex_lock(&heartbeat_lock); |
1da177e4 LT |
617 | |
618 | atomic_set(&heartbeat_tofree, 2); | |
619 | ||
36c7dc44 CM |
620 | /* |
621 | * Don't reset the timer if we have the timer turned off, that | |
622 | * re-enables the watchdog. | |
623 | */ | |
1da177e4 | 624 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) { |
d6dfd131 | 625 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
626 | return 0; |
627 | } | |
628 | ||
629 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
630 | addr.channel = IPMI_BMC_CHANNEL; | |
631 | addr.lun = 0; | |
632 | ||
633 | msg.netfn = 0x06; | |
634 | msg.cmd = IPMI_WDOG_RESET_TIMER; | |
635 | msg.data = NULL; | |
636 | msg.data_len = 0; | |
637 | rv = ipmi_request_supply_msgs(watchdog_user, | |
638 | (struct ipmi_addr *) &addr, | |
639 | 0, | |
640 | &msg, | |
641 | NULL, | |
642 | &heartbeat_smi_msg, | |
643 | &heartbeat_recv_msg, | |
644 | 1); | |
645 | if (rv) { | |
d6dfd131 | 646 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
647 | printk(KERN_WARNING PFX "heartbeat failure: %d\n", |
648 | rv); | |
649 | return rv; | |
650 | } | |
651 | ||
652 | /* Wait for the heartbeat to be sent. */ | |
d6dfd131 | 653 | wait_for_completion(&heartbeat_wait); |
1da177e4 LT |
654 | |
655 | if (heartbeat_recv_msg.msg.data[0] != 0) { | |
36c7dc44 CM |
656 | /* |
657 | * Got an error in the heartbeat response. It was already | |
658 | * reported in ipmi_wdog_msg_handler, but we should return | |
659 | * an error here. | |
660 | */ | |
661 | rv = -EINVAL; | |
1da177e4 LT |
662 | } |
663 | ||
d6dfd131 | 664 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
665 | |
666 | return rv; | |
667 | } | |
668 | ||
36c7dc44 | 669 | static struct watchdog_info ident = { |
1da177e4 LT |
670 | .options = 0, /* WDIOF_SETTIMEOUT, */ |
671 | .firmware_version = 1, | |
672 | .identity = "IPMI" | |
673 | }; | |
674 | ||
55929332 | 675 | static int ipmi_ioctl(struct file *file, |
1da177e4 LT |
676 | unsigned int cmd, unsigned long arg) |
677 | { | |
678 | void __user *argp = (void __user *)arg; | |
679 | int i; | |
680 | int val; | |
681 | ||
36c7dc44 | 682 | switch (cmd) { |
1da177e4 LT |
683 | case WDIOC_GETSUPPORT: |
684 | i = copy_to_user(argp, &ident, sizeof(ident)); | |
685 | return i ? -EFAULT : 0; | |
686 | ||
687 | case WDIOC_SETTIMEOUT: | |
688 | i = copy_from_user(&val, argp, sizeof(int)); | |
689 | if (i) | |
690 | return -EFAULT; | |
691 | timeout = val; | |
692 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
693 | ||
694 | case WDIOC_GETTIMEOUT: | |
695 | i = copy_to_user(argp, &timeout, sizeof(timeout)); | |
696 | if (i) | |
697 | return -EFAULT; | |
698 | return 0; | |
699 | ||
700 | case WDIOC_SET_PRETIMEOUT: | |
783e6bcd | 701 | case WDIOC_SETPRETIMEOUT: |
1da177e4 LT |
702 | i = copy_from_user(&val, argp, sizeof(int)); |
703 | if (i) | |
704 | return -EFAULT; | |
705 | pretimeout = val; | |
706 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
707 | ||
708 | case WDIOC_GET_PRETIMEOUT: | |
783e6bcd | 709 | case WDIOC_GETPRETIMEOUT: |
1da177e4 LT |
710 | i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); |
711 | if (i) | |
712 | return -EFAULT; | |
713 | return 0; | |
714 | ||
715 | case WDIOC_KEEPALIVE: | |
716 | return ipmi_heartbeat(); | |
717 | ||
718 | case WDIOC_SETOPTIONS: | |
719 | i = copy_from_user(&val, argp, sizeof(int)); | |
720 | if (i) | |
721 | return -EFAULT; | |
36c7dc44 | 722 | if (val & WDIOS_DISABLECARD) { |
1da177e4 LT |
723 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
724 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
725 | ipmi_start_timer_on_heartbeat = 0; | |
726 | } | |
727 | ||
36c7dc44 | 728 | if (val & WDIOS_ENABLECARD) { |
1da177e4 LT |
729 | ipmi_watchdog_state = action_val; |
730 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
731 | } | |
732 | return 0; | |
733 | ||
734 | case WDIOC_GETSTATUS: | |
735 | val = 0; | |
736 | i = copy_to_user(argp, &val, sizeof(val)); | |
737 | if (i) | |
738 | return -EFAULT; | |
739 | return 0; | |
740 | ||
741 | default: | |
742 | return -ENOIOCTLCMD; | |
743 | } | |
744 | } | |
745 | ||
55929332 AB |
746 | static long ipmi_unlocked_ioctl(struct file *file, |
747 | unsigned int cmd, | |
748 | unsigned long arg) | |
749 | { | |
750 | int ret; | |
751 | ||
609146fd | 752 | mutex_lock(&ipmi_watchdog_mutex); |
55929332 | 753 | ret = ipmi_ioctl(file, cmd, arg); |
609146fd | 754 | mutex_unlock(&ipmi_watchdog_mutex); |
55929332 AB |
755 | |
756 | return ret; | |
757 | } | |
758 | ||
1da177e4 LT |
759 | static ssize_t ipmi_write(struct file *file, |
760 | const char __user *buf, | |
761 | size_t len, | |
762 | loff_t *ppos) | |
763 | { | |
764 | int rv; | |
765 | ||
766 | if (len) { | |
36c7dc44 CM |
767 | if (!nowayout) { |
768 | size_t i; | |
1da177e4 LT |
769 | |
770 | /* In case it was set long ago */ | |
771 | expect_close = 0; | |
772 | ||
36c7dc44 | 773 | for (i = 0; i != len; i++) { |
1da177e4 LT |
774 | char c; |
775 | ||
776 | if (get_user(c, buf + i)) | |
777 | return -EFAULT; | |
778 | if (c == 'V') | |
779 | expect_close = 42; | |
780 | } | |
781 | } | |
782 | rv = ipmi_heartbeat(); | |
783 | if (rv) | |
784 | return rv; | |
1da177e4 | 785 | } |
3976df9b | 786 | return len; |
1da177e4 LT |
787 | } |
788 | ||
789 | static ssize_t ipmi_read(struct file *file, | |
790 | char __user *buf, | |
791 | size_t count, | |
792 | loff_t *ppos) | |
793 | { | |
794 | int rv = 0; | |
795 | wait_queue_t wait; | |
796 | ||
797 | if (count <= 0) | |
798 | return 0; | |
799 | ||
36c7dc44 CM |
800 | /* |
801 | * Reading returns if the pretimeout has gone off, and it only does | |
802 | * it once per pretimeout. | |
803 | */ | |
1da177e4 LT |
804 | spin_lock(&ipmi_read_lock); |
805 | if (!data_to_read) { | |
806 | if (file->f_flags & O_NONBLOCK) { | |
807 | rv = -EAGAIN; | |
808 | goto out; | |
809 | } | |
36c7dc44 | 810 | |
1da177e4 LT |
811 | init_waitqueue_entry(&wait, current); |
812 | add_wait_queue(&read_q, &wait); | |
813 | while (!data_to_read) { | |
814 | set_current_state(TASK_INTERRUPTIBLE); | |
815 | spin_unlock(&ipmi_read_lock); | |
816 | schedule(); | |
817 | spin_lock(&ipmi_read_lock); | |
818 | } | |
819 | remove_wait_queue(&read_q, &wait); | |
36c7dc44 | 820 | |
1da177e4 LT |
821 | if (signal_pending(current)) { |
822 | rv = -ERESTARTSYS; | |
823 | goto out; | |
824 | } | |
825 | } | |
826 | data_to_read = 0; | |
827 | ||
828 | out: | |
829 | spin_unlock(&ipmi_read_lock); | |
830 | ||
831 | if (rv == 0) { | |
832 | if (copy_to_user(buf, &data_to_read, 1)) | |
833 | rv = -EFAULT; | |
834 | else | |
835 | rv = 1; | |
836 | } | |
837 | ||
838 | return rv; | |
839 | } | |
840 | ||
841 | static int ipmi_open(struct inode *ino, struct file *filep) | |
842 | { | |
36c7dc44 CM |
843 | switch (iminor(ino)) { |
844 | case WATCHDOG_MINOR: | |
e8b33617 | 845 | if (test_and_set_bit(0, &ipmi_wdog_open)) |
36c7dc44 | 846 | return -EBUSY; |
1da177e4 | 847 | |
af96f010 | 848 | |
36c7dc44 CM |
849 | /* |
850 | * Don't start the timer now, let it start on the | |
851 | * first heartbeat. | |
852 | */ | |
e8b33617 CM |
853 | ipmi_start_timer_on_heartbeat = 1; |
854 | return nonseekable_open(ino, filep); | |
1da177e4 | 855 | |
e8b33617 CM |
856 | default: |
857 | return (-ENODEV); | |
36c7dc44 | 858 | } |
1da177e4 LT |
859 | } |
860 | ||
861 | static unsigned int ipmi_poll(struct file *file, poll_table *wait) | |
862 | { | |
863 | unsigned int mask = 0; | |
36c7dc44 | 864 | |
1da177e4 LT |
865 | poll_wait(file, &read_q, wait); |
866 | ||
867 | spin_lock(&ipmi_read_lock); | |
868 | if (data_to_read) | |
869 | mask |= (POLLIN | POLLRDNORM); | |
870 | spin_unlock(&ipmi_read_lock); | |
871 | ||
872 | return mask; | |
873 | } | |
874 | ||
875 | static int ipmi_fasync(int fd, struct file *file, int on) | |
876 | { | |
877 | int result; | |
878 | ||
879 | result = fasync_helper(fd, file, on, &fasync_q); | |
880 | ||
881 | return (result); | |
882 | } | |
883 | ||
884 | static int ipmi_close(struct inode *ino, struct file *filep) | |
885 | { | |
8a3628d5 | 886 | if (iminor(ino) == WATCHDOG_MINOR) { |
1da177e4 LT |
887 | if (expect_close == 42) { |
888 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
889 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
1da177e4 | 890 | } else { |
8a3628d5 CM |
891 | printk(KERN_CRIT PFX |
892 | "Unexpected close, not stopping watchdog!\n"); | |
1da177e4 LT |
893 | ipmi_heartbeat(); |
894 | } | |
ec26d79f | 895 | clear_bit(0, &ipmi_wdog_open); |
1da177e4 LT |
896 | } |
897 | ||
1da177e4 LT |
898 | expect_close = 0; |
899 | ||
900 | return 0; | |
901 | } | |
902 | ||
62322d25 | 903 | static const struct file_operations ipmi_wdog_fops = { |
1da177e4 LT |
904 | .owner = THIS_MODULE, |
905 | .read = ipmi_read, | |
906 | .poll = ipmi_poll, | |
907 | .write = ipmi_write, | |
55929332 | 908 | .unlocked_ioctl = ipmi_unlocked_ioctl, |
1da177e4 LT |
909 | .open = ipmi_open, |
910 | .release = ipmi_close, | |
911 | .fasync = ipmi_fasync, | |
6038f373 | 912 | .llseek = no_llseek, |
1da177e4 LT |
913 | }; |
914 | ||
915 | static struct miscdevice ipmi_wdog_miscdev = { | |
916 | .minor = WATCHDOG_MINOR, | |
917 | .name = "watchdog", | |
918 | .fops = &ipmi_wdog_fops | |
919 | }; | |
920 | ||
1da177e4 LT |
921 | static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, |
922 | void *handler_data) | |
923 | { | |
924 | if (msg->msg.data[0] != 0) { | |
925 | printk(KERN_ERR PFX "response: Error %x on cmd %x\n", | |
926 | msg->msg.data[0], | |
927 | msg->msg.cmd); | |
928 | } | |
36c7dc44 | 929 | |
1da177e4 LT |
930 | ipmi_free_recv_msg(msg); |
931 | } | |
932 | ||
933 | static void ipmi_wdog_pretimeout_handler(void *handler_data) | |
934 | { | |
935 | if (preaction_val != WDOG_PRETIMEOUT_NONE) { | |
b385676b CM |
936 | if (preop_val == WDOG_PREOP_PANIC) { |
937 | if (atomic_inc_and_test(&preop_panic_excl)) | |
938 | panic("Watchdog pre-timeout"); | |
939 | } else if (preop_val == WDOG_PREOP_GIVE_DATA) { | |
1da177e4 LT |
940 | spin_lock(&ipmi_read_lock); |
941 | data_to_read = 1; | |
942 | wake_up_interruptible(&read_q); | |
943 | kill_fasync(&fasync_q, SIGIO, POLL_IN); | |
944 | ||
945 | spin_unlock(&ipmi_read_lock); | |
946 | } | |
947 | } | |
948 | ||
36c7dc44 CM |
949 | /* |
950 | * On some machines, the heartbeat will give an error and not | |
951 | * work unless we re-enable the timer. So do so. | |
952 | */ | |
1da177e4 LT |
953 | pretimeout_since_last_heartbeat = 1; |
954 | } | |
955 | ||
36c7dc44 | 956 | static struct ipmi_user_hndl ipmi_hndlrs = { |
1da177e4 LT |
957 | .ipmi_recv_hndl = ipmi_wdog_msg_handler, |
958 | .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler | |
959 | }; | |
960 | ||
961 | static void ipmi_register_watchdog(int ipmi_intf) | |
962 | { | |
963 | int rv = -EBUSY; | |
964 | ||
1da177e4 LT |
965 | if (watchdog_user) |
966 | goto out; | |
967 | ||
b2c03941 CM |
968 | if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf)) |
969 | goto out; | |
970 | ||
971 | watchdog_ifnum = ipmi_intf; | |
972 | ||
1da177e4 LT |
973 | rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); |
974 | if (rv < 0) { | |
975 | printk(KERN_CRIT PFX "Unable to register with ipmi\n"); | |
976 | goto out; | |
977 | } | |
978 | ||
979 | ipmi_get_version(watchdog_user, | |
980 | &ipmi_version_major, | |
981 | &ipmi_version_minor); | |
982 | ||
983 | rv = misc_register(&ipmi_wdog_miscdev); | |
984 | if (rv < 0) { | |
985 | ipmi_destroy_user(watchdog_user); | |
986 | watchdog_user = NULL; | |
987 | printk(KERN_CRIT PFX "Unable to register misc device\n"); | |
988 | } | |
989 | ||
612b5a8d CM |
990 | #ifdef HAVE_DIE_NMI |
991 | if (nmi_handler_registered) { | |
992 | int old_pretimeout = pretimeout; | |
993 | int old_timeout = timeout; | |
994 | int old_preop_val = preop_val; | |
995 | ||
36c7dc44 CM |
996 | /* |
997 | * Set the pretimeout to go off in a second and give | |
998 | * ourselves plenty of time to stop the timer. | |
999 | */ | |
612b5a8d CM |
1000 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; |
1001 | preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */ | |
1002 | pretimeout = 99; | |
1003 | timeout = 100; | |
1004 | ||
1005 | testing_nmi = 1; | |
1006 | ||
1007 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
1008 | if (rv) { | |
1009 | printk(KERN_WARNING PFX "Error starting timer to" | |
1010 | " test NMI: 0x%x. The NMI pretimeout will" | |
1011 | " likely not work\n", rv); | |
1012 | rv = 0; | |
1013 | goto out_restore; | |
1014 | } | |
1015 | ||
1016 | msleep(1500); | |
1017 | ||
1018 | if (testing_nmi != 2) { | |
1019 | printk(KERN_WARNING PFX "IPMI NMI didn't seem to" | |
1020 | " occur. The NMI pretimeout will" | |
1021 | " likely not work\n"); | |
1022 | } | |
36c7dc44 | 1023 | out_restore: |
612b5a8d CM |
1024 | testing_nmi = 0; |
1025 | preop_val = old_preop_val; | |
1026 | pretimeout = old_pretimeout; | |
1027 | timeout = old_timeout; | |
1028 | } | |
1029 | #endif | |
1030 | ||
1da177e4 | 1031 | out: |
1da177e4 LT |
1032 | if ((start_now) && (rv == 0)) { |
1033 | /* Run from startup, so start the timer now. */ | |
1034 | start_now = 0; /* Disable this function after first startup. */ | |
1035 | ipmi_watchdog_state = action_val; | |
1036 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
1037 | printk(KERN_INFO PFX "Starting now!\n"); | |
612b5a8d CM |
1038 | } else { |
1039 | /* Stop the timer now. */ | |
1040 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
1041 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
1da177e4 LT |
1042 | } |
1043 | } | |
1044 | ||
b2c03941 CM |
1045 | static void ipmi_unregister_watchdog(int ipmi_intf) |
1046 | { | |
1047 | int rv; | |
1048 | ||
b2c03941 CM |
1049 | if (!watchdog_user) |
1050 | goto out; | |
1051 | ||
1052 | if (watchdog_ifnum != ipmi_intf) | |
1053 | goto out; | |
1054 | ||
1055 | /* Make sure no one can call us any more. */ | |
1056 | misc_deregister(&ipmi_wdog_miscdev); | |
1057 | ||
36c7dc44 CM |
1058 | /* |
1059 | * Wait to make sure the message makes it out. The lower layer has | |
1060 | * pointers to our buffers, we want to make sure they are done before | |
1061 | * we release our memory. | |
1062 | */ | |
b2c03941 CM |
1063 | while (atomic_read(&set_timeout_tofree)) |
1064 | schedule_timeout_uninterruptible(1); | |
1065 | ||
1066 | /* Disconnect from IPMI. */ | |
1067 | rv = ipmi_destroy_user(watchdog_user); | |
1068 | if (rv) { | |
1069 | printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n", | |
1070 | rv); | |
1071 | } | |
1072 | watchdog_user = NULL; | |
1073 | ||
1074 | out: | |
f8fbcd3b | 1075 | return; |
b2c03941 CM |
1076 | } |
1077 | ||
612b5a8d | 1078 | #ifdef HAVE_DIE_NMI |
1da177e4 | 1079 | static int |
612b5a8d | 1080 | ipmi_nmi(struct notifier_block *self, unsigned long val, void *data) |
1da177e4 | 1081 | { |
612b5a8d CM |
1082 | struct die_args *args = data; |
1083 | ||
673a6092 | 1084 | if (val != DIE_NMIUNKNOWN) |
612b5a8d CM |
1085 | return NOTIFY_OK; |
1086 | ||
1087 | /* Hack, if it's a memory or I/O error, ignore it. */ | |
1088 | if (args->err & 0xc0) | |
1089 | return NOTIFY_OK; | |
1090 | ||
1091 | /* | |
1092 | * If we get here, it's an NMI that's not a memory or I/O | |
1093 | * error. We can't truly tell if it's from IPMI or not | |
1094 | * without sending a message, and sending a message is almost | |
1095 | * impossible because of locking. | |
1096 | */ | |
1097 | ||
1098 | if (testing_nmi) { | |
1099 | testing_nmi = 2; | |
1100 | return NOTIFY_STOP; | |
1101 | } | |
1102 | ||
36c7dc44 | 1103 | /* If we are not expecting a timeout, ignore it. */ |
8f05ee9a | 1104 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
612b5a8d CM |
1105 | return NOTIFY_OK; |
1106 | ||
1107 | if (preaction_val != WDOG_PRETIMEOUT_NMI) | |
1108 | return NOTIFY_OK; | |
8f05ee9a | 1109 | |
36c7dc44 CM |
1110 | /* |
1111 | * If no one else handled the NMI, we assume it was the IPMI | |
1112 | * watchdog. | |
1113 | */ | |
612b5a8d | 1114 | if (preop_val == WDOG_PREOP_PANIC) { |
8f05ee9a CM |
1115 | /* On some machines, the heartbeat will give |
1116 | an error and not work unless we re-enable | |
1117 | the timer. So do so. */ | |
1118 | pretimeout_since_last_heartbeat = 1; | |
b385676b CM |
1119 | if (atomic_inc_and_test(&preop_panic_excl)) |
1120 | panic(PFX "pre-timeout"); | |
8f05ee9a | 1121 | } |
1da177e4 | 1122 | |
612b5a8d | 1123 | return NOTIFY_STOP; |
1da177e4 LT |
1124 | } |
1125 | ||
612b5a8d CM |
1126 | static struct notifier_block ipmi_nmi_handler = { |
1127 | .notifier_call = ipmi_nmi | |
1da177e4 LT |
1128 | }; |
1129 | #endif | |
1130 | ||
1131 | static int wdog_reboot_handler(struct notifier_block *this, | |
1132 | unsigned long code, | |
1133 | void *unused) | |
1134 | { | |
36c7dc44 | 1135 | static int reboot_event_handled; |
1da177e4 LT |
1136 | |
1137 | if ((watchdog_user) && (!reboot_event_handled)) { | |
1138 | /* Make sure we only do this once. */ | |
1139 | reboot_event_handled = 1; | |
1140 | ||
fcfa4724 | 1141 | if (code == SYS_POWER_OFF || code == SYS_HALT) { |
1da177e4 LT |
1142 | /* Disable the WDT if we are shutting down. */ |
1143 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
1144 | panic_halt_ipmi_set_timeout(); | |
96febe9f | 1145 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
1da177e4 | 1146 | /* Set a long timer to let the reboot happens, but |
96febe9f CM |
1147 | reboot if it hangs, but only if the watchdog |
1148 | timer was already running. */ | |
1da177e4 LT |
1149 | timeout = 120; |
1150 | pretimeout = 0; | |
1151 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; | |
1152 | panic_halt_ipmi_set_timeout(); | |
1153 | } | |
1154 | } | |
1155 | return NOTIFY_OK; | |
1156 | } | |
1157 | ||
1158 | static struct notifier_block wdog_reboot_notifier = { | |
1159 | .notifier_call = wdog_reboot_handler, | |
1160 | .next = NULL, | |
1161 | .priority = 0 | |
1162 | }; | |
1163 | ||
1164 | static int wdog_panic_handler(struct notifier_block *this, | |
1165 | unsigned long event, | |
1166 | void *unused) | |
1167 | { | |
36c7dc44 | 1168 | static int panic_event_handled; |
1da177e4 | 1169 | |
96febe9f CM |
1170 | /* On a panic, if we have a panic timeout, make sure to extend |
1171 | the watchdog timer to a reasonable value to complete the | |
1172 | panic, if the watchdog timer is running. Plus the | |
1173 | pretimeout is meaningless at panic time. */ | |
1174 | if (watchdog_user && !panic_event_handled && | |
1175 | ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
1176 | /* Make sure we do this only once. */ | |
1da177e4 | 1177 | panic_event_handled = 1; |
36c7dc44 | 1178 | |
1da177e4 LT |
1179 | timeout = 255; |
1180 | pretimeout = 0; | |
1da177e4 LT |
1181 | panic_halt_ipmi_set_timeout(); |
1182 | } | |
1183 | ||
1184 | return NOTIFY_OK; | |
1185 | } | |
1186 | ||
1187 | static struct notifier_block wdog_panic_notifier = { | |
1188 | .notifier_call = wdog_panic_handler, | |
1189 | .next = NULL, | |
1190 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ | |
1191 | }; | |
1192 | ||
1193 | ||
50c812b2 | 1194 | static void ipmi_new_smi(int if_num, struct device *device) |
1da177e4 LT |
1195 | { |
1196 | ipmi_register_watchdog(if_num); | |
1197 | } | |
1198 | ||
1199 | static void ipmi_smi_gone(int if_num) | |
1200 | { | |
b2c03941 | 1201 | ipmi_unregister_watchdog(if_num); |
1da177e4 LT |
1202 | } |
1203 | ||
36c7dc44 | 1204 | static struct ipmi_smi_watcher smi_watcher = { |
1da177e4 LT |
1205 | .owner = THIS_MODULE, |
1206 | .new_smi = ipmi_new_smi, | |
1207 | .smi_gone = ipmi_smi_gone | |
1208 | }; | |
1209 | ||
cc4673ee | 1210 | static int action_op(const char *inval, char *outval) |
1da177e4 | 1211 | { |
cc4673ee CM |
1212 | if (outval) |
1213 | strcpy(outval, action); | |
1214 | ||
1215 | if (!inval) | |
1216 | return 0; | |
1da177e4 | 1217 | |
cc4673ee | 1218 | if (strcmp(inval, "reset") == 0) |
1da177e4 | 1219 | action_val = WDOG_TIMEOUT_RESET; |
cc4673ee | 1220 | else if (strcmp(inval, "none") == 0) |
1da177e4 | 1221 | action_val = WDOG_TIMEOUT_NONE; |
cc4673ee | 1222 | else if (strcmp(inval, "power_cycle") == 0) |
1da177e4 | 1223 | action_val = WDOG_TIMEOUT_POWER_CYCLE; |
cc4673ee | 1224 | else if (strcmp(inval, "power_off") == 0) |
1da177e4 | 1225 | action_val = WDOG_TIMEOUT_POWER_DOWN; |
cc4673ee CM |
1226 | else |
1227 | return -EINVAL; | |
1228 | strcpy(action, inval); | |
1229 | return 0; | |
1230 | } | |
1231 | ||
1232 | static int preaction_op(const char *inval, char *outval) | |
1233 | { | |
1234 | if (outval) | |
1235 | strcpy(outval, preaction); | |
1da177e4 | 1236 | |
cc4673ee CM |
1237 | if (!inval) |
1238 | return 0; | |
1239 | ||
1240 | if (strcmp(inval, "pre_none") == 0) | |
1da177e4 | 1241 | preaction_val = WDOG_PRETIMEOUT_NONE; |
cc4673ee | 1242 | else if (strcmp(inval, "pre_smi") == 0) |
1da177e4 | 1243 | preaction_val = WDOG_PRETIMEOUT_SMI; |
612b5a8d | 1244 | #ifdef HAVE_DIE_NMI |
cc4673ee | 1245 | else if (strcmp(inval, "pre_nmi") == 0) |
1da177e4 LT |
1246 | preaction_val = WDOG_PRETIMEOUT_NMI; |
1247 | #endif | |
cc4673ee | 1248 | else if (strcmp(inval, "pre_int") == 0) |
1da177e4 | 1249 | preaction_val = WDOG_PRETIMEOUT_MSG_INT; |
cc4673ee CM |
1250 | else |
1251 | return -EINVAL; | |
1252 | strcpy(preaction, inval); | |
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | static int preop_op(const char *inval, char *outval) | |
1257 | { | |
1258 | if (outval) | |
1259 | strcpy(outval, preop); | |
1da177e4 | 1260 | |
cc4673ee CM |
1261 | if (!inval) |
1262 | return 0; | |
1263 | ||
1264 | if (strcmp(inval, "preop_none") == 0) | |
1da177e4 | 1265 | preop_val = WDOG_PREOP_NONE; |
cc4673ee | 1266 | else if (strcmp(inval, "preop_panic") == 0) |
1da177e4 | 1267 | preop_val = WDOG_PREOP_PANIC; |
cc4673ee | 1268 | else if (strcmp(inval, "preop_give_data") == 0) |
1da177e4 | 1269 | preop_val = WDOG_PREOP_GIVE_DATA; |
cc4673ee CM |
1270 | else |
1271 | return -EINVAL; | |
1272 | strcpy(preop, inval); | |
1273 | return 0; | |
1274 | } | |
1da177e4 | 1275 | |
cc4673ee CM |
1276 | static void check_parms(void) |
1277 | { | |
612b5a8d | 1278 | #ifdef HAVE_DIE_NMI |
cc4673ee CM |
1279 | int do_nmi = 0; |
1280 | int rv; | |
1281 | ||
1da177e4 | 1282 | if (preaction_val == WDOG_PRETIMEOUT_NMI) { |
cc4673ee | 1283 | do_nmi = 1; |
1da177e4 LT |
1284 | if (preop_val == WDOG_PREOP_GIVE_DATA) { |
1285 | printk(KERN_WARNING PFX "Pretimeout op is to give data" | |
1286 | " but NMI pretimeout is enabled, setting" | |
1287 | " pretimeout op to none\n"); | |
cc4673ee CM |
1288 | preop_op("preop_none", NULL); |
1289 | do_nmi = 0; | |
1da177e4 | 1290 | } |
cc4673ee CM |
1291 | } |
1292 | if (do_nmi && !nmi_handler_registered) { | |
612b5a8d | 1293 | rv = register_die_notifier(&ipmi_nmi_handler); |
1da177e4 | 1294 | if (rv) { |
cc4673ee CM |
1295 | printk(KERN_WARNING PFX |
1296 | "Can't register nmi handler\n"); | |
1297 | return; | |
1298 | } else | |
1299 | nmi_handler_registered = 1; | |
1300 | } else if (!do_nmi && nmi_handler_registered) { | |
612b5a8d | 1301 | unregister_die_notifier(&ipmi_nmi_handler); |
cc4673ee | 1302 | nmi_handler_registered = 0; |
1da177e4 LT |
1303 | } |
1304 | #endif | |
cc4673ee CM |
1305 | } |
1306 | ||
1307 | static int __init ipmi_wdog_init(void) | |
1308 | { | |
1309 | int rv; | |
1310 | ||
1311 | if (action_op(action, NULL)) { | |
1312 | action_op("reset", NULL); | |
1313 | printk(KERN_INFO PFX "Unknown action '%s', defaulting to" | |
1314 | " reset\n", action); | |
1315 | } | |
1316 | ||
1317 | if (preaction_op(preaction, NULL)) { | |
1318 | preaction_op("pre_none", NULL); | |
1319 | printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to" | |
1320 | " none\n", preaction); | |
1321 | } | |
1322 | ||
1323 | if (preop_op(preop, NULL)) { | |
1324 | preop_op("preop_none", NULL); | |
1325 | printk(KERN_INFO PFX "Unknown preop '%s', defaulting to" | |
1326 | " none\n", preop); | |
1327 | } | |
1328 | ||
1329 | check_parms(); | |
1da177e4 | 1330 | |
b2c03941 CM |
1331 | register_reboot_notifier(&wdog_reboot_notifier); |
1332 | atomic_notifier_chain_register(&panic_notifier_list, | |
1333 | &wdog_panic_notifier); | |
1334 | ||
1da177e4 LT |
1335 | rv = ipmi_smi_watcher_register(&smi_watcher); |
1336 | if (rv) { | |
612b5a8d CM |
1337 | #ifdef HAVE_DIE_NMI |
1338 | if (nmi_handler_registered) | |
1339 | unregister_die_notifier(&ipmi_nmi_handler); | |
1da177e4 | 1340 | #endif |
b2c03941 CM |
1341 | atomic_notifier_chain_unregister(&panic_notifier_list, |
1342 | &wdog_panic_notifier); | |
1343 | unregister_reboot_notifier(&wdog_reboot_notifier); | |
1da177e4 LT |
1344 | printk(KERN_WARNING PFX "can't register smi watcher\n"); |
1345 | return rv; | |
1346 | } | |
1347 | ||
1fdd75bd CM |
1348 | printk(KERN_INFO PFX "driver initialized\n"); |
1349 | ||
1da177e4 LT |
1350 | return 0; |
1351 | } | |
1352 | ||
b2c03941 | 1353 | static void __exit ipmi_wdog_exit(void) |
1da177e4 | 1354 | { |
b2c03941 CM |
1355 | ipmi_smi_watcher_unregister(&smi_watcher); |
1356 | ipmi_unregister_watchdog(watchdog_ifnum); | |
1da177e4 | 1357 | |
612b5a8d | 1358 | #ifdef HAVE_DIE_NMI |
cc4673ee | 1359 | if (nmi_handler_registered) |
612b5a8d | 1360 | unregister_die_notifier(&ipmi_nmi_handler); |
1da177e4 LT |
1361 | #endif |
1362 | ||
e041c683 | 1363 | atomic_notifier_chain_unregister(&panic_notifier_list, |
b2c03941 | 1364 | &wdog_panic_notifier); |
1da177e4 | 1365 | unregister_reboot_notifier(&wdog_reboot_notifier); |
1da177e4 LT |
1366 | } |
1367 | module_exit(ipmi_wdog_exit); | |
1368 | module_init(ipmi_wdog_init); | |
1369 | MODULE_LICENSE("GPL"); | |
1fdd75bd CM |
1370 | MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); |
1371 | MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); |