]>
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 | 67 | #include <asm/kdebug.h> |
9c48f1c6 | 68 | #include <asm/nmi.h> |
612b5a8d | 69 | #define HAVE_DIE_NMI |
1da177e4 LT |
70 | #endif |
71 | ||
72 | #define PFX "IPMI Watchdog: " | |
73 | ||
1da177e4 LT |
74 | /* |
75 | * The IPMI command/response information for the watchdog timer. | |
76 | */ | |
77 | ||
78 | /* values for byte 1 of the set command, byte 2 of the get response. */ | |
79 | #define WDOG_DONT_LOG (1 << 7) | |
80 | #define WDOG_DONT_STOP_ON_SET (1 << 6) | |
81 | #define WDOG_SET_TIMER_USE(byte, use) \ | |
82 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
83 | #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7) | |
84 | #define WDOG_TIMER_USE_BIOS_FRB2 1 | |
85 | #define WDOG_TIMER_USE_BIOS_POST 2 | |
86 | #define WDOG_TIMER_USE_OS_LOAD 3 | |
87 | #define WDOG_TIMER_USE_SMS_OS 4 | |
88 | #define WDOG_TIMER_USE_OEM 5 | |
89 | ||
90 | /* values for byte 2 of the set command, byte 3 of the get response. */ | |
91 | #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \ | |
92 | byte = ((byte) & 0x8f) | (((use) & 0x7) << 4) | |
93 | #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7) | |
94 | #define WDOG_PRETIMEOUT_NONE 0 | |
95 | #define WDOG_PRETIMEOUT_SMI 1 | |
96 | #define WDOG_PRETIMEOUT_NMI 2 | |
97 | #define WDOG_PRETIMEOUT_MSG_INT 3 | |
98 | ||
99 | /* Operations that can be performed on a pretimout. */ | |
100 | #define WDOG_PREOP_NONE 0 | |
101 | #define WDOG_PREOP_PANIC 1 | |
36c7dc44 CM |
102 | /* Cause data to be available to read. Doesn't work in NMI mode. */ |
103 | #define WDOG_PREOP_GIVE_DATA 2 | |
1da177e4 LT |
104 | |
105 | /* Actions to perform on a full timeout. */ | |
106 | #define WDOG_SET_TIMEOUT_ACT(byte, use) \ | |
107 | byte = ((byte) & 0xf8) | ((use) & 0x7) | |
108 | #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7) | |
109 | #define WDOG_TIMEOUT_NONE 0 | |
110 | #define WDOG_TIMEOUT_RESET 1 | |
111 | #define WDOG_TIMEOUT_POWER_DOWN 2 | |
112 | #define WDOG_TIMEOUT_POWER_CYCLE 3 | |
113 | ||
36c7dc44 CM |
114 | /* |
115 | * Byte 3 of the get command, byte 4 of the get response is the | |
116 | * pre-timeout in seconds. | |
117 | */ | |
1da177e4 LT |
118 | |
119 | /* Bits for setting byte 4 of the set command, byte 5 of the get response. */ | |
120 | #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1) | |
121 | #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2) | |
122 | #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3) | |
123 | #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4) | |
124 | #define WDOG_EXPIRE_CLEAR_OEM (1 << 5) | |
125 | ||
36c7dc44 CM |
126 | /* |
127 | * Setting/getting the watchdog timer value. This is for bytes 5 and | |
128 | * 6 (the timeout time) of the set command, and bytes 6 and 7 (the | |
129 | * timeout time) and 8 and 9 (the current countdown value) of the | |
130 | * response. The timeout value is given in seconds (in the command it | |
131 | * is 100ms intervals). | |
132 | */ | |
1da177e4 LT |
133 | #define WDOG_SET_TIMEOUT(byte1, byte2, val) \ |
134 | (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8) | |
135 | #define WDOG_GET_TIMEOUT(byte1, byte2) \ | |
136 | (((byte1) | ((byte2) << 8)) / 10) | |
137 | ||
138 | #define IPMI_WDOG_RESET_TIMER 0x22 | |
139 | #define IPMI_WDOG_SET_TIMER 0x24 | |
140 | #define IPMI_WDOG_GET_TIMER 0x25 | |
141 | ||
b75d91f7 CM |
142 | #define IPMI_WDOG_TIMER_NOT_INIT_RESP 0x80 |
143 | ||
609146fd | 144 | static DEFINE_MUTEX(ipmi_watchdog_mutex); |
86a1e189 | 145 | static bool 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 | 155 | |
c7f42c63 JYF |
156 | /* Default timeout to set on panic */ |
157 | static int panic_wdt_timeout = 255; | |
158 | ||
1da177e4 LT |
159 | /* Default action is to reset the board on a timeout. */ |
160 | static unsigned char action_val = WDOG_TIMEOUT_RESET; | |
161 | ||
162 | static char action[16] = "reset"; | |
163 | ||
164 | static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE; | |
165 | ||
166 | static char preaction[16] = "pre_none"; | |
167 | ||
168 | static unsigned char preop_val = WDOG_PREOP_NONE; | |
169 | ||
170 | static char preop[16] = "preop_none"; | |
171 | static DEFINE_SPINLOCK(ipmi_read_lock); | |
0c8204b3 | 172 | static char data_to_read; |
1da177e4 | 173 | static DECLARE_WAIT_QUEUE_HEAD(read_q); |
0c8204b3 RD |
174 | static struct fasync_struct *fasync_q; |
175 | static char pretimeout_since_last_heartbeat; | |
1da177e4 LT |
176 | static char expect_close; |
177 | ||
b2c03941 CM |
178 | static int ifnum_to_use = -1; |
179 | ||
cc4673ee CM |
180 | /* Parameters to ipmi_set_timeout */ |
181 | #define IPMI_SET_TIMEOUT_NO_HB 0 | |
182 | #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 | |
183 | #define IPMI_SET_TIMEOUT_FORCE_HB 2 | |
184 | ||
185 | static int ipmi_set_timeout(int do_heartbeat); | |
b2c03941 CM |
186 | static void ipmi_register_watchdog(int ipmi_intf); |
187 | static void ipmi_unregister_watchdog(int ipmi_intf); | |
cc4673ee | 188 | |
36c7dc44 CM |
189 | /* |
190 | * If true, the driver will start running as soon as it is configured | |
191 | * and ready. | |
192 | */ | |
0c8204b3 | 193 | static int start_now; |
1da177e4 | 194 | |
c8ba6c52 | 195 | static int set_param_timeout(const char *val, const struct kernel_param *kp) |
cc4673ee CM |
196 | { |
197 | char *endp; | |
198 | int l; | |
199 | int rv = 0; | |
200 | ||
201 | if (!val) | |
202 | return -EINVAL; | |
203 | l = simple_strtoul(val, &endp, 0); | |
204 | if (endp == val) | |
205 | return -EINVAL; | |
206 | ||
cc4673ee CM |
207 | *((int *)kp->arg) = l; |
208 | if (watchdog_user) | |
209 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
cc4673ee CM |
210 | |
211 | return rv; | |
212 | } | |
213 | ||
9c27847d | 214 | static const struct kernel_param_ops param_ops_timeout = { |
c8ba6c52 RR |
215 | .set = set_param_timeout, |
216 | .get = param_get_int, | |
217 | }; | |
218 | #define param_check_timeout param_check_int | |
cc4673ee CM |
219 | |
220 | typedef int (*action_fn)(const char *intval, char *outval); | |
221 | ||
222 | static int action_op(const char *inval, char *outval); | |
223 | static int preaction_op(const char *inval, char *outval); | |
224 | static int preop_op(const char *inval, char *outval); | |
225 | static void check_parms(void); | |
226 | ||
c8ba6c52 | 227 | static int set_param_str(const char *val, const struct kernel_param *kp) |
cc4673ee CM |
228 | { |
229 | action_fn fn = (action_fn) kp->arg; | |
230 | int rv = 0; | |
43cdff92 SD |
231 | char valcp[16]; |
232 | char *s; | |
66f969d0 | 233 | |
43cdff92 SD |
234 | strncpy(valcp, val, 16); |
235 | valcp[15] = '\0'; | |
66f969d0 | 236 | |
43cdff92 | 237 | s = strstrip(valcp); |
cc4673ee | 238 | |
66f969d0 | 239 | rv = fn(s, NULL); |
cc4673ee | 240 | if (rv) |
f8fbcd3b | 241 | goto out; |
cc4673ee CM |
242 | |
243 | check_parms(); | |
244 | if (watchdog_user) | |
245 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
246 | ||
f8fbcd3b | 247 | out: |
cc4673ee CM |
248 | return rv; |
249 | } | |
250 | ||
c8ba6c52 | 251 | static int get_param_str(char *buffer, const struct kernel_param *kp) |
cc4673ee CM |
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 | 262 | |
c8ba6c52 | 263 | static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp) |
b2c03941 CM |
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 | ||
9c27847d | 276 | static const struct kernel_param_ops param_ops_wdog_ifnum = { |
c8ba6c52 RR |
277 | .set = set_param_wdog_ifnum, |
278 | .get = param_get_int, | |
279 | }; | |
280 | ||
281 | #define param_check_wdog_ifnum param_check_int | |
282 | ||
9c27847d | 283 | static const struct kernel_param_ops param_ops_str = { |
c8ba6c52 RR |
284 | .set = set_param_str, |
285 | .get = get_param_str, | |
286 | }; | |
287 | ||
288 | module_param(ifnum_to_use, wdog_ifnum, 0644); | |
b2c03941 CM |
289 | MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog " |
290 | "timer. Setting to -1 defaults to the first registered " | |
291 | "interface"); | |
292 | ||
c8ba6c52 | 293 | module_param(timeout, timeout, 0644); |
1da177e4 | 294 | MODULE_PARM_DESC(timeout, "Timeout value in seconds."); |
cc4673ee | 295 | |
c8ba6c52 | 296 | module_param(pretimeout, timeout, 0644); |
1da177e4 | 297 | MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds."); |
cc4673ee | 298 | |
c7f42c63 JYF |
299 | module_param(panic_wdt_timeout, timeout, 0644); |
300 | MODULE_PARM_DESC(timeout, "Timeout value on kernel panic in seconds."); | |
301 | ||
c8ba6c52 | 302 | module_param_cb(action, ¶m_ops_str, action_op, 0644); |
1da177e4 LT |
303 | MODULE_PARM_DESC(action, "Timeout action. One of: " |
304 | "reset, none, power_cycle, power_off."); | |
cc4673ee | 305 | |
c8ba6c52 | 306 | module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644); |
1da177e4 LT |
307 | MODULE_PARM_DESC(preaction, "Pretimeout action. One of: " |
308 | "pre_none, pre_smi, pre_nmi, pre_int."); | |
cc4673ee | 309 | |
c8ba6c52 | 310 | module_param_cb(preop, ¶m_ops_str, preop_op, 0644); |
1da177e4 LT |
311 | MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: " |
312 | "preop_none, preop_panic, preop_give_data."); | |
cc4673ee | 313 | |
b2c03941 | 314 | module_param(start_now, int, 0444); |
1da177e4 LT |
315 | MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as" |
316 | "soon as the driver is loaded."); | |
cc4673ee | 317 | |
86a1e189 | 318 | module_param(nowayout, bool, 0644); |
b2c03941 CM |
319 | MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started " |
320 | "(default=CONFIG_WATCHDOG_NOWAYOUT)"); | |
1da177e4 LT |
321 | |
322 | /* Default state of the timer. */ | |
323 | static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
324 | ||
325 | /* If shutting down via IPMI, we ignore the heartbeat. */ | |
0c8204b3 | 326 | static int ipmi_ignore_heartbeat; |
1da177e4 LT |
327 | |
328 | /* Is someone using the watchdog? Only one user is allowed. */ | |
0c8204b3 | 329 | static unsigned long ipmi_wdog_open; |
1da177e4 | 330 | |
36c7dc44 CM |
331 | /* |
332 | * If set to 1, the heartbeat command will set the state to reset and | |
333 | * start the timer. The timer doesn't normally run when the driver is | |
334 | * first opened until the heartbeat is set the first time, this | |
335 | * variable is used to accomplish this. | |
336 | */ | |
0c8204b3 | 337 | static int ipmi_start_timer_on_heartbeat; |
1da177e4 LT |
338 | |
339 | /* IPMI version of the BMC. */ | |
340 | static unsigned char ipmi_version_major; | |
341 | static unsigned char ipmi_version_minor; | |
342 | ||
b385676b CM |
343 | /* If a pretimeout occurs, this is used to allow only one panic to happen. */ |
344 | static atomic_t preop_panic_excl = ATOMIC_INIT(-1); | |
1da177e4 | 345 | |
612b5a8d CM |
346 | #ifdef HAVE_DIE_NMI |
347 | static int testing_nmi; | |
348 | static int nmi_handler_registered; | |
349 | #endif | |
350 | ||
1da177e4 | 351 | static int ipmi_heartbeat(void); |
1da177e4 | 352 | |
36c7dc44 CM |
353 | /* |
354 | * We use a mutex to make sure that only one thing can send a set | |
355 | * timeout at one time, because we only have one copy of the data. | |
356 | * The mutex is claimed when the set_timeout is sent and freed | |
357 | * when both messages are free. | |
358 | */ | |
1da177e4 | 359 | static atomic_t set_timeout_tofree = ATOMIC_INIT(0); |
d6dfd131 CM |
360 | static DEFINE_MUTEX(set_timeout_lock); |
361 | static DECLARE_COMPLETION(set_timeout_wait); | |
1da177e4 LT |
362 | static void set_timeout_free_smi(struct ipmi_smi_msg *msg) |
363 | { | |
364 | if (atomic_dec_and_test(&set_timeout_tofree)) | |
d6dfd131 | 365 | complete(&set_timeout_wait); |
1da177e4 LT |
366 | } |
367 | static void set_timeout_free_recv(struct ipmi_recv_msg *msg) | |
368 | { | |
369 | if (atomic_dec_and_test(&set_timeout_tofree)) | |
d6dfd131 | 370 | complete(&set_timeout_wait); |
1da177e4 | 371 | } |
36c7dc44 | 372 | static struct ipmi_smi_msg set_timeout_smi_msg = { |
1da177e4 LT |
373 | .done = set_timeout_free_smi |
374 | }; | |
36c7dc44 | 375 | static struct ipmi_recv_msg set_timeout_recv_msg = { |
1da177e4 LT |
376 | .done = set_timeout_free_recv |
377 | }; | |
36c7dc44 | 378 | |
1da177e4 LT |
379 | static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, |
380 | struct ipmi_recv_msg *recv_msg, | |
381 | int *send_heartbeat_now) | |
382 | { | |
383 | struct kernel_ipmi_msg msg; | |
384 | unsigned char data[6]; | |
385 | int rv; | |
386 | struct ipmi_system_interface_addr addr; | |
387 | int hbnow = 0; | |
388 | ||
389 | ||
612b5a8d CM |
390 | /* These can be cleared as we are setting the timeout. */ |
391 | pretimeout_since_last_heartbeat = 0; | |
392 | ||
1da177e4 LT |
393 | data[0] = 0; |
394 | WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS); | |
395 | ||
396 | if ((ipmi_version_major > 1) | |
36c7dc44 | 397 | || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) { |
1da177e4 LT |
398 | /* This is an IPMI 1.5-only feature. */ |
399 | data[0] |= WDOG_DONT_STOP_ON_SET; | |
400 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
36c7dc44 CM |
401 | /* |
402 | * In ipmi 1.0, setting the timer stops the watchdog, we | |
403 | * need to start it back up again. | |
404 | */ | |
1da177e4 LT |
405 | hbnow = 1; |
406 | } | |
407 | ||
408 | data[1] = 0; | |
409 | WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state); | |
8f05ee9a | 410 | if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) { |
1da177e4 LT |
411 | WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val); |
412 | data[2] = pretimeout; | |
413 | } else { | |
414 | WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE); | |
415 | data[2] = 0; /* No pretimeout. */ | |
416 | } | |
417 | data[3] = 0; | |
418 | WDOG_SET_TIMEOUT(data[4], data[5], timeout); | |
419 | ||
420 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
421 | addr.channel = IPMI_BMC_CHANNEL; | |
422 | addr.lun = 0; | |
423 | ||
424 | msg.netfn = 0x06; | |
425 | msg.cmd = IPMI_WDOG_SET_TIMER; | |
426 | msg.data = data; | |
427 | msg.data_len = sizeof(data); | |
428 | rv = ipmi_request_supply_msgs(watchdog_user, | |
429 | (struct ipmi_addr *) &addr, | |
430 | 0, | |
431 | &msg, | |
432 | NULL, | |
433 | smi_msg, | |
434 | recv_msg, | |
435 | 1); | |
436 | if (rv) { | |
437 | printk(KERN_WARNING PFX "set timeout error: %d\n", | |
438 | rv); | |
439 | } | |
440 | ||
441 | if (send_heartbeat_now) | |
442 | *send_heartbeat_now = hbnow; | |
443 | ||
444 | return rv; | |
445 | } | |
446 | ||
1da177e4 LT |
447 | static int ipmi_set_timeout(int do_heartbeat) |
448 | { | |
449 | int send_heartbeat_now; | |
450 | int rv; | |
451 | ||
452 | ||
453 | /* We can only send one of these at a time. */ | |
d6dfd131 | 454 | mutex_lock(&set_timeout_lock); |
1da177e4 LT |
455 | |
456 | atomic_set(&set_timeout_tofree, 2); | |
457 | ||
458 | rv = i_ipmi_set_timeout(&set_timeout_smi_msg, | |
459 | &set_timeout_recv_msg, | |
460 | &send_heartbeat_now); | |
461 | if (rv) { | |
d6dfd131 CM |
462 | mutex_unlock(&set_timeout_lock); |
463 | goto out; | |
464 | } | |
465 | ||
466 | wait_for_completion(&set_timeout_wait); | |
467 | ||
612b5a8d CM |
468 | mutex_unlock(&set_timeout_lock); |
469 | ||
d6dfd131 CM |
470 | if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB) |
471 | || ((send_heartbeat_now) | |
472 | && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY))) | |
d6dfd131 | 473 | rv = ipmi_heartbeat(); |
1da177e4 | 474 | |
d6dfd131 | 475 | out: |
1da177e4 LT |
476 | return rv; |
477 | } | |
478 | ||
fcfa4724 CM |
479 | static atomic_t panic_done_count = ATOMIC_INIT(0); |
480 | ||
481 | static void panic_smi_free(struct ipmi_smi_msg *msg) | |
1da177e4 | 482 | { |
fcfa4724 | 483 | atomic_dec(&panic_done_count); |
1da177e4 | 484 | } |
fcfa4724 CM |
485 | static void panic_recv_free(struct ipmi_recv_msg *msg) |
486 | { | |
487 | atomic_dec(&panic_done_count); | |
488 | } | |
489 | ||
36c7dc44 | 490 | static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = { |
fcfa4724 CM |
491 | .done = panic_smi_free |
492 | }; | |
36c7dc44 | 493 | static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = { |
fcfa4724 CM |
494 | .done = panic_recv_free |
495 | }; | |
496 | ||
497 | static void panic_halt_ipmi_heartbeat(void) | |
498 | { | |
499 | struct kernel_ipmi_msg msg; | |
500 | struct ipmi_system_interface_addr addr; | |
501 | int rv; | |
502 | ||
36c7dc44 CM |
503 | /* |
504 | * Don't reset the timer if we have the timer turned off, that | |
505 | * re-enables the watchdog. | |
506 | */ | |
fcfa4724 CM |
507 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
508 | return; | |
509 | ||
510 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
511 | addr.channel = IPMI_BMC_CHANNEL; | |
512 | addr.lun = 0; | |
513 | ||
514 | msg.netfn = 0x06; | |
515 | msg.cmd = IPMI_WDOG_RESET_TIMER; | |
516 | msg.data = NULL; | |
517 | msg.data_len = 0; | |
895dcfd1 | 518 | atomic_add(2, &panic_done_count); |
fcfa4724 CM |
519 | rv = ipmi_request_supply_msgs(watchdog_user, |
520 | (struct ipmi_addr *) &addr, | |
521 | 0, | |
522 | &msg, | |
523 | NULL, | |
524 | &panic_halt_heartbeat_smi_msg, | |
525 | &panic_halt_heartbeat_recv_msg, | |
526 | 1); | |
895dcfd1 CM |
527 | if (rv) |
528 | atomic_sub(2, &panic_done_count); | |
1da177e4 | 529 | } |
fcfa4724 | 530 | |
36c7dc44 | 531 | static struct ipmi_smi_msg panic_halt_smi_msg = { |
fcfa4724 | 532 | .done = panic_smi_free |
1da177e4 | 533 | }; |
36c7dc44 | 534 | static struct ipmi_recv_msg panic_halt_recv_msg = { |
fcfa4724 | 535 | .done = panic_recv_free |
1da177e4 LT |
536 | }; |
537 | ||
36c7dc44 CM |
538 | /* |
539 | * Special call, doesn't claim any locks. This is only to be called | |
540 | * at panic or halt time, in run-to-completion mode, when the caller | |
541 | * is the only CPU and the only thing that will be going is these IPMI | |
542 | * calls. | |
543 | */ | |
1da177e4 LT |
544 | static void panic_halt_ipmi_set_timeout(void) |
545 | { | |
546 | int send_heartbeat_now; | |
547 | int rv; | |
548 | ||
fcfa4724 CM |
549 | /* Wait for the messages to be free. */ |
550 | while (atomic_read(&panic_done_count) != 0) | |
551 | ipmi_poll_interface(watchdog_user); | |
895dcfd1 | 552 | atomic_add(2, &panic_done_count); |
1da177e4 LT |
553 | rv = i_ipmi_set_timeout(&panic_halt_smi_msg, |
554 | &panic_halt_recv_msg, | |
555 | &send_heartbeat_now); | |
895dcfd1 CM |
556 | if (rv) { |
557 | atomic_sub(2, &panic_done_count); | |
fcfa4724 CM |
558 | printk(KERN_WARNING PFX |
559 | "Unable to extend the watchdog timeout."); | |
895dcfd1 CM |
560 | } else { |
561 | if (send_heartbeat_now) | |
562 | panic_halt_ipmi_heartbeat(); | |
563 | } | |
fcfa4724 CM |
564 | while (atomic_read(&panic_done_count) != 0) |
565 | ipmi_poll_interface(watchdog_user); | |
1da177e4 LT |
566 | } |
567 | ||
36c7dc44 CM |
568 | /* |
569 | * We use a mutex to make sure that only one thing can send a | |
570 | * heartbeat at one time, because we only have one copy of the data. | |
571 | * The semaphore is claimed when the set_timeout is sent and freed | |
572 | * when both messages are free. | |
573 | */ | |
1da177e4 | 574 | static atomic_t heartbeat_tofree = ATOMIC_INIT(0); |
d6dfd131 CM |
575 | static DEFINE_MUTEX(heartbeat_lock); |
576 | static DECLARE_COMPLETION(heartbeat_wait); | |
1da177e4 LT |
577 | static void heartbeat_free_smi(struct ipmi_smi_msg *msg) |
578 | { | |
579 | if (atomic_dec_and_test(&heartbeat_tofree)) | |
d6dfd131 | 580 | complete(&heartbeat_wait); |
1da177e4 LT |
581 | } |
582 | static void heartbeat_free_recv(struct ipmi_recv_msg *msg) | |
583 | { | |
584 | if (atomic_dec_and_test(&heartbeat_tofree)) | |
d6dfd131 | 585 | complete(&heartbeat_wait); |
1da177e4 | 586 | } |
36c7dc44 | 587 | static struct ipmi_smi_msg heartbeat_smi_msg = { |
1da177e4 LT |
588 | .done = heartbeat_free_smi |
589 | }; | |
36c7dc44 | 590 | static struct ipmi_recv_msg heartbeat_recv_msg = { |
1da177e4 LT |
591 | .done = heartbeat_free_recv |
592 | }; | |
36c7dc44 | 593 | |
1da177e4 LT |
594 | static int ipmi_heartbeat(void) |
595 | { | |
596 | struct kernel_ipmi_msg msg; | |
597 | int rv; | |
598 | struct ipmi_system_interface_addr addr; | |
b75d91f7 | 599 | int timeout_retries = 0; |
1da177e4 | 600 | |
612b5a8d | 601 | if (ipmi_ignore_heartbeat) |
1da177e4 | 602 | return 0; |
1da177e4 LT |
603 | |
604 | if (ipmi_start_timer_on_heartbeat) { | |
faa8b6c3 | 605 | ipmi_start_timer_on_heartbeat = 0; |
1da177e4 LT |
606 | ipmi_watchdog_state = action_val; |
607 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
608 | } else if (pretimeout_since_last_heartbeat) { | |
36c7dc44 CM |
609 | /* |
610 | * A pretimeout occurred, make sure we set the timeout. | |
611 | * We don't want to set the action, though, we want to | |
612 | * leave that alone (thus it can't be combined with the | |
613 | * above operation. | |
614 | */ | |
1da177e4 LT |
615 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); |
616 | } | |
617 | ||
d6dfd131 | 618 | mutex_lock(&heartbeat_lock); |
1da177e4 | 619 | |
b75d91f7 | 620 | restart: |
1da177e4 LT |
621 | atomic_set(&heartbeat_tofree, 2); |
622 | ||
36c7dc44 CM |
623 | /* |
624 | * Don't reset the timer if we have the timer turned off, that | |
625 | * re-enables the watchdog. | |
626 | */ | |
1da177e4 | 627 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) { |
d6dfd131 | 628 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
629 | return 0; |
630 | } | |
631 | ||
632 | addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; | |
633 | addr.channel = IPMI_BMC_CHANNEL; | |
634 | addr.lun = 0; | |
635 | ||
636 | msg.netfn = 0x06; | |
637 | msg.cmd = IPMI_WDOG_RESET_TIMER; | |
638 | msg.data = NULL; | |
639 | msg.data_len = 0; | |
640 | rv = ipmi_request_supply_msgs(watchdog_user, | |
641 | (struct ipmi_addr *) &addr, | |
642 | 0, | |
643 | &msg, | |
644 | NULL, | |
645 | &heartbeat_smi_msg, | |
646 | &heartbeat_recv_msg, | |
647 | 1); | |
648 | if (rv) { | |
d6dfd131 | 649 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
650 | printk(KERN_WARNING PFX "heartbeat failure: %d\n", |
651 | rv); | |
652 | return rv; | |
653 | } | |
654 | ||
655 | /* Wait for the heartbeat to be sent. */ | |
d6dfd131 | 656 | wait_for_completion(&heartbeat_wait); |
1da177e4 | 657 | |
b75d91f7 CM |
658 | if (heartbeat_recv_msg.msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) { |
659 | timeout_retries++; | |
660 | if (timeout_retries > 3) { | |
661 | printk(KERN_ERR PFX ": Unable to restore the IPMI" | |
662 | " watchdog's settings, giving up.\n"); | |
663 | rv = -EIO; | |
664 | goto out_unlock; | |
665 | } | |
666 | ||
667 | /* | |
668 | * The timer was not initialized, that means the BMC was | |
669 | * probably reset and lost the watchdog information. Attempt | |
670 | * to restore the timer's info. Note that we still hold | |
671 | * the heartbeat lock, to keep a heartbeat from happening | |
672 | * in this process, so must say no heartbeat to avoid a | |
673 | * deadlock on this mutex. | |
674 | */ | |
675 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
676 | if (rv) { | |
677 | printk(KERN_ERR PFX ": Unable to send the command to" | |
678 | " set the watchdog's settings, giving up.\n"); | |
679 | goto out_unlock; | |
680 | } | |
681 | ||
682 | /* We might need a new heartbeat, so do it now */ | |
683 | goto restart; | |
684 | } else if (heartbeat_recv_msg.msg.data[0] != 0) { | |
36c7dc44 CM |
685 | /* |
686 | * Got an error in the heartbeat response. It was already | |
687 | * reported in ipmi_wdog_msg_handler, but we should return | |
688 | * an error here. | |
689 | */ | |
690 | rv = -EINVAL; | |
1da177e4 LT |
691 | } |
692 | ||
b75d91f7 | 693 | out_unlock: |
d6dfd131 | 694 | mutex_unlock(&heartbeat_lock); |
1da177e4 LT |
695 | |
696 | return rv; | |
697 | } | |
698 | ||
36c7dc44 | 699 | static struct watchdog_info ident = { |
1da177e4 LT |
700 | .options = 0, /* WDIOF_SETTIMEOUT, */ |
701 | .firmware_version = 1, | |
702 | .identity = "IPMI" | |
703 | }; | |
704 | ||
55929332 | 705 | static int ipmi_ioctl(struct file *file, |
1da177e4 LT |
706 | unsigned int cmd, unsigned long arg) |
707 | { | |
708 | void __user *argp = (void __user *)arg; | |
709 | int i; | |
710 | int val; | |
711 | ||
36c7dc44 | 712 | switch (cmd) { |
1da177e4 LT |
713 | case WDIOC_GETSUPPORT: |
714 | i = copy_to_user(argp, &ident, sizeof(ident)); | |
715 | return i ? -EFAULT : 0; | |
716 | ||
717 | case WDIOC_SETTIMEOUT: | |
718 | i = copy_from_user(&val, argp, sizeof(int)); | |
719 | if (i) | |
720 | return -EFAULT; | |
721 | timeout = val; | |
722 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
723 | ||
724 | case WDIOC_GETTIMEOUT: | |
725 | i = copy_to_user(argp, &timeout, sizeof(timeout)); | |
726 | if (i) | |
727 | return -EFAULT; | |
728 | return 0; | |
729 | ||
783e6bcd | 730 | case WDIOC_SETPRETIMEOUT: |
1da177e4 LT |
731 | i = copy_from_user(&val, argp, sizeof(int)); |
732 | if (i) | |
733 | return -EFAULT; | |
734 | pretimeout = val; | |
735 | return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); | |
736 | ||
783e6bcd | 737 | case WDIOC_GETPRETIMEOUT: |
1da177e4 LT |
738 | i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); |
739 | if (i) | |
740 | return -EFAULT; | |
741 | return 0; | |
742 | ||
743 | case WDIOC_KEEPALIVE: | |
744 | return ipmi_heartbeat(); | |
745 | ||
746 | case WDIOC_SETOPTIONS: | |
747 | i = copy_from_user(&val, argp, sizeof(int)); | |
748 | if (i) | |
749 | return -EFAULT; | |
36c7dc44 | 750 | if (val & WDIOS_DISABLECARD) { |
1da177e4 LT |
751 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; |
752 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
753 | ipmi_start_timer_on_heartbeat = 0; | |
754 | } | |
755 | ||
36c7dc44 | 756 | if (val & WDIOS_ENABLECARD) { |
1da177e4 LT |
757 | ipmi_watchdog_state = action_val; |
758 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
759 | } | |
760 | return 0; | |
761 | ||
762 | case WDIOC_GETSTATUS: | |
763 | val = 0; | |
764 | i = copy_to_user(argp, &val, sizeof(val)); | |
765 | if (i) | |
766 | return -EFAULT; | |
767 | return 0; | |
768 | ||
769 | default: | |
770 | return -ENOIOCTLCMD; | |
771 | } | |
772 | } | |
773 | ||
55929332 AB |
774 | static long ipmi_unlocked_ioctl(struct file *file, |
775 | unsigned int cmd, | |
776 | unsigned long arg) | |
777 | { | |
778 | int ret; | |
779 | ||
609146fd | 780 | mutex_lock(&ipmi_watchdog_mutex); |
55929332 | 781 | ret = ipmi_ioctl(file, cmd, arg); |
609146fd | 782 | mutex_unlock(&ipmi_watchdog_mutex); |
55929332 AB |
783 | |
784 | return ret; | |
785 | } | |
786 | ||
1da177e4 LT |
787 | static ssize_t ipmi_write(struct file *file, |
788 | const char __user *buf, | |
789 | size_t len, | |
790 | loff_t *ppos) | |
791 | { | |
792 | int rv; | |
793 | ||
794 | if (len) { | |
36c7dc44 CM |
795 | if (!nowayout) { |
796 | size_t i; | |
1da177e4 LT |
797 | |
798 | /* In case it was set long ago */ | |
799 | expect_close = 0; | |
800 | ||
36c7dc44 | 801 | for (i = 0; i != len; i++) { |
1da177e4 LT |
802 | char c; |
803 | ||
804 | if (get_user(c, buf + i)) | |
805 | return -EFAULT; | |
806 | if (c == 'V') | |
807 | expect_close = 42; | |
808 | } | |
809 | } | |
810 | rv = ipmi_heartbeat(); | |
811 | if (rv) | |
812 | return rv; | |
1da177e4 | 813 | } |
3976df9b | 814 | return len; |
1da177e4 LT |
815 | } |
816 | ||
817 | static ssize_t ipmi_read(struct file *file, | |
818 | char __user *buf, | |
819 | size_t count, | |
820 | loff_t *ppos) | |
821 | { | |
822 | int rv = 0; | |
823 | wait_queue_t wait; | |
824 | ||
825 | if (count <= 0) | |
826 | return 0; | |
827 | ||
36c7dc44 CM |
828 | /* |
829 | * Reading returns if the pretimeout has gone off, and it only does | |
830 | * it once per pretimeout. | |
831 | */ | |
1da177e4 LT |
832 | spin_lock(&ipmi_read_lock); |
833 | if (!data_to_read) { | |
834 | if (file->f_flags & O_NONBLOCK) { | |
835 | rv = -EAGAIN; | |
836 | goto out; | |
837 | } | |
36c7dc44 | 838 | |
1da177e4 LT |
839 | init_waitqueue_entry(&wait, current); |
840 | add_wait_queue(&read_q, &wait); | |
841 | while (!data_to_read) { | |
842 | set_current_state(TASK_INTERRUPTIBLE); | |
843 | spin_unlock(&ipmi_read_lock); | |
844 | schedule(); | |
845 | spin_lock(&ipmi_read_lock); | |
846 | } | |
847 | remove_wait_queue(&read_q, &wait); | |
36c7dc44 | 848 | |
1da177e4 LT |
849 | if (signal_pending(current)) { |
850 | rv = -ERESTARTSYS; | |
851 | goto out; | |
852 | } | |
853 | } | |
854 | data_to_read = 0; | |
855 | ||
856 | out: | |
857 | spin_unlock(&ipmi_read_lock); | |
858 | ||
859 | if (rv == 0) { | |
860 | if (copy_to_user(buf, &data_to_read, 1)) | |
861 | rv = -EFAULT; | |
862 | else | |
863 | rv = 1; | |
864 | } | |
865 | ||
866 | return rv; | |
867 | } | |
868 | ||
869 | static int ipmi_open(struct inode *ino, struct file *filep) | |
870 | { | |
36c7dc44 CM |
871 | switch (iminor(ino)) { |
872 | case WATCHDOG_MINOR: | |
e8b33617 | 873 | if (test_and_set_bit(0, &ipmi_wdog_open)) |
36c7dc44 | 874 | return -EBUSY; |
1da177e4 | 875 | |
af96f010 | 876 | |
36c7dc44 CM |
877 | /* |
878 | * Don't start the timer now, let it start on the | |
879 | * first heartbeat. | |
880 | */ | |
e8b33617 CM |
881 | ipmi_start_timer_on_heartbeat = 1; |
882 | return nonseekable_open(ino, filep); | |
1da177e4 | 883 | |
e8b33617 CM |
884 | default: |
885 | return (-ENODEV); | |
36c7dc44 | 886 | } |
1da177e4 LT |
887 | } |
888 | ||
889 | static unsigned int ipmi_poll(struct file *file, poll_table *wait) | |
890 | { | |
891 | unsigned int mask = 0; | |
36c7dc44 | 892 | |
1da177e4 LT |
893 | poll_wait(file, &read_q, wait); |
894 | ||
895 | spin_lock(&ipmi_read_lock); | |
896 | if (data_to_read) | |
897 | mask |= (POLLIN | POLLRDNORM); | |
898 | spin_unlock(&ipmi_read_lock); | |
899 | ||
900 | return mask; | |
901 | } | |
902 | ||
903 | static int ipmi_fasync(int fd, struct file *file, int on) | |
904 | { | |
905 | int result; | |
906 | ||
907 | result = fasync_helper(fd, file, on, &fasync_q); | |
908 | ||
909 | return (result); | |
910 | } | |
911 | ||
912 | static int ipmi_close(struct inode *ino, struct file *filep) | |
913 | { | |
8a3628d5 | 914 | if (iminor(ino) == WATCHDOG_MINOR) { |
1da177e4 LT |
915 | if (expect_close == 42) { |
916 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
917 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
1da177e4 | 918 | } else { |
8a3628d5 CM |
919 | printk(KERN_CRIT PFX |
920 | "Unexpected close, not stopping watchdog!\n"); | |
1da177e4 LT |
921 | ipmi_heartbeat(); |
922 | } | |
ec26d79f | 923 | clear_bit(0, &ipmi_wdog_open); |
1da177e4 LT |
924 | } |
925 | ||
1da177e4 LT |
926 | expect_close = 0; |
927 | ||
928 | return 0; | |
929 | } | |
930 | ||
62322d25 | 931 | static const struct file_operations ipmi_wdog_fops = { |
1da177e4 LT |
932 | .owner = THIS_MODULE, |
933 | .read = ipmi_read, | |
934 | .poll = ipmi_poll, | |
935 | .write = ipmi_write, | |
55929332 | 936 | .unlocked_ioctl = ipmi_unlocked_ioctl, |
1da177e4 LT |
937 | .open = ipmi_open, |
938 | .release = ipmi_close, | |
939 | .fasync = ipmi_fasync, | |
6038f373 | 940 | .llseek = no_llseek, |
1da177e4 LT |
941 | }; |
942 | ||
943 | static struct miscdevice ipmi_wdog_miscdev = { | |
944 | .minor = WATCHDOG_MINOR, | |
945 | .name = "watchdog", | |
946 | .fops = &ipmi_wdog_fops | |
947 | }; | |
948 | ||
1da177e4 LT |
949 | static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg, |
950 | void *handler_data) | |
951 | { | |
b75d91f7 CM |
952 | if (msg->msg.cmd == IPMI_WDOG_RESET_TIMER && |
953 | msg->msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) | |
954 | printk(KERN_INFO PFX "response: The IPMI controller appears" | |
955 | " to have been reset, will attempt to reinitialize" | |
956 | " the watchdog timer\n"); | |
957 | else if (msg->msg.data[0] != 0) | |
1da177e4 LT |
958 | printk(KERN_ERR PFX "response: Error %x on cmd %x\n", |
959 | msg->msg.data[0], | |
960 | msg->msg.cmd); | |
36c7dc44 | 961 | |
1da177e4 LT |
962 | ipmi_free_recv_msg(msg); |
963 | } | |
964 | ||
965 | static void ipmi_wdog_pretimeout_handler(void *handler_data) | |
966 | { | |
967 | if (preaction_val != WDOG_PRETIMEOUT_NONE) { | |
b385676b CM |
968 | if (preop_val == WDOG_PREOP_PANIC) { |
969 | if (atomic_inc_and_test(&preop_panic_excl)) | |
970 | panic("Watchdog pre-timeout"); | |
971 | } else if (preop_val == WDOG_PREOP_GIVE_DATA) { | |
1da177e4 LT |
972 | spin_lock(&ipmi_read_lock); |
973 | data_to_read = 1; | |
974 | wake_up_interruptible(&read_q); | |
975 | kill_fasync(&fasync_q, SIGIO, POLL_IN); | |
976 | ||
977 | spin_unlock(&ipmi_read_lock); | |
978 | } | |
979 | } | |
980 | ||
36c7dc44 CM |
981 | /* |
982 | * On some machines, the heartbeat will give an error and not | |
983 | * work unless we re-enable the timer. So do so. | |
984 | */ | |
1da177e4 LT |
985 | pretimeout_since_last_heartbeat = 1; |
986 | } | |
987 | ||
36c7dc44 | 988 | static struct ipmi_user_hndl ipmi_hndlrs = { |
1da177e4 LT |
989 | .ipmi_recv_hndl = ipmi_wdog_msg_handler, |
990 | .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler | |
991 | }; | |
992 | ||
993 | static void ipmi_register_watchdog(int ipmi_intf) | |
994 | { | |
995 | int rv = -EBUSY; | |
996 | ||
1da177e4 LT |
997 | if (watchdog_user) |
998 | goto out; | |
999 | ||
b2c03941 CM |
1000 | if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf)) |
1001 | goto out; | |
1002 | ||
1003 | watchdog_ifnum = ipmi_intf; | |
1004 | ||
1da177e4 LT |
1005 | rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user); |
1006 | if (rv < 0) { | |
1007 | printk(KERN_CRIT PFX "Unable to register with ipmi\n"); | |
1008 | goto out; | |
1009 | } | |
1010 | ||
1011 | ipmi_get_version(watchdog_user, | |
1012 | &ipmi_version_major, | |
1013 | &ipmi_version_minor); | |
1014 | ||
1015 | rv = misc_register(&ipmi_wdog_miscdev); | |
1016 | if (rv < 0) { | |
1017 | ipmi_destroy_user(watchdog_user); | |
1018 | watchdog_user = NULL; | |
1019 | printk(KERN_CRIT PFX "Unable to register misc device\n"); | |
1020 | } | |
1021 | ||
612b5a8d CM |
1022 | #ifdef HAVE_DIE_NMI |
1023 | if (nmi_handler_registered) { | |
1024 | int old_pretimeout = pretimeout; | |
1025 | int old_timeout = timeout; | |
1026 | int old_preop_val = preop_val; | |
1027 | ||
36c7dc44 CM |
1028 | /* |
1029 | * Set the pretimeout to go off in a second and give | |
1030 | * ourselves plenty of time to stop the timer. | |
1031 | */ | |
612b5a8d CM |
1032 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; |
1033 | preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */ | |
1034 | pretimeout = 99; | |
1035 | timeout = 100; | |
1036 | ||
1037 | testing_nmi = 1; | |
1038 | ||
1039 | rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
1040 | if (rv) { | |
1041 | printk(KERN_WARNING PFX "Error starting timer to" | |
1042 | " test NMI: 0x%x. The NMI pretimeout will" | |
1043 | " likely not work\n", rv); | |
1044 | rv = 0; | |
1045 | goto out_restore; | |
1046 | } | |
1047 | ||
1048 | msleep(1500); | |
1049 | ||
1050 | if (testing_nmi != 2) { | |
1051 | printk(KERN_WARNING PFX "IPMI NMI didn't seem to" | |
1052 | " occur. The NMI pretimeout will" | |
1053 | " likely not work\n"); | |
1054 | } | |
36c7dc44 | 1055 | out_restore: |
612b5a8d CM |
1056 | testing_nmi = 0; |
1057 | preop_val = old_preop_val; | |
1058 | pretimeout = old_pretimeout; | |
1059 | timeout = old_timeout; | |
1060 | } | |
1061 | #endif | |
1062 | ||
1da177e4 | 1063 | out: |
1da177e4 LT |
1064 | if ((start_now) && (rv == 0)) { |
1065 | /* Run from startup, so start the timer now. */ | |
1066 | start_now = 0; /* Disable this function after first startup. */ | |
1067 | ipmi_watchdog_state = action_val; | |
1068 | ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); | |
1069 | printk(KERN_INFO PFX "Starting now!\n"); | |
612b5a8d CM |
1070 | } else { |
1071 | /* Stop the timer now. */ | |
1072 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
1073 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); | |
1da177e4 LT |
1074 | } |
1075 | } | |
1076 | ||
b2c03941 CM |
1077 | static void ipmi_unregister_watchdog(int ipmi_intf) |
1078 | { | |
1079 | int rv; | |
1080 | ||
b2c03941 CM |
1081 | if (!watchdog_user) |
1082 | goto out; | |
1083 | ||
1084 | if (watchdog_ifnum != ipmi_intf) | |
1085 | goto out; | |
1086 | ||
1087 | /* Make sure no one can call us any more. */ | |
1088 | misc_deregister(&ipmi_wdog_miscdev); | |
1089 | ||
36c7dc44 CM |
1090 | /* |
1091 | * Wait to make sure the message makes it out. The lower layer has | |
1092 | * pointers to our buffers, we want to make sure they are done before | |
1093 | * we release our memory. | |
1094 | */ | |
b2c03941 CM |
1095 | while (atomic_read(&set_timeout_tofree)) |
1096 | schedule_timeout_uninterruptible(1); | |
1097 | ||
1098 | /* Disconnect from IPMI. */ | |
1099 | rv = ipmi_destroy_user(watchdog_user); | |
1100 | if (rv) { | |
1101 | printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n", | |
1102 | rv); | |
1103 | } | |
1104 | watchdog_user = NULL; | |
1105 | ||
1106 | out: | |
f8fbcd3b | 1107 | return; |
b2c03941 CM |
1108 | } |
1109 | ||
612b5a8d | 1110 | #ifdef HAVE_DIE_NMI |
1da177e4 | 1111 | static int |
9c48f1c6 | 1112 | ipmi_nmi(unsigned int val, struct pt_regs *regs) |
1da177e4 | 1113 | { |
612b5a8d CM |
1114 | /* |
1115 | * If we get here, it's an NMI that's not a memory or I/O | |
1116 | * error. We can't truly tell if it's from IPMI or not | |
1117 | * without sending a message, and sending a message is almost | |
1118 | * impossible because of locking. | |
1119 | */ | |
1120 | ||
1121 | if (testing_nmi) { | |
1122 | testing_nmi = 2; | |
9c48f1c6 | 1123 | return NMI_HANDLED; |
612b5a8d CM |
1124 | } |
1125 | ||
36c7dc44 | 1126 | /* If we are not expecting a timeout, ignore it. */ |
8f05ee9a | 1127 | if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) |
9c48f1c6 | 1128 | return NMI_DONE; |
612b5a8d CM |
1129 | |
1130 | if (preaction_val != WDOG_PRETIMEOUT_NMI) | |
9c48f1c6 | 1131 | return NMI_DONE; |
8f05ee9a | 1132 | |
36c7dc44 CM |
1133 | /* |
1134 | * If no one else handled the NMI, we assume it was the IPMI | |
1135 | * watchdog. | |
1136 | */ | |
612b5a8d | 1137 | if (preop_val == WDOG_PREOP_PANIC) { |
8f05ee9a CM |
1138 | /* On some machines, the heartbeat will give |
1139 | an error and not work unless we re-enable | |
1140 | the timer. So do so. */ | |
1141 | pretimeout_since_last_heartbeat = 1; | |
b385676b CM |
1142 | if (atomic_inc_and_test(&preop_panic_excl)) |
1143 | panic(PFX "pre-timeout"); | |
8f05ee9a | 1144 | } |
1da177e4 | 1145 | |
9c48f1c6 | 1146 | return NMI_HANDLED; |
1da177e4 | 1147 | } |
1da177e4 LT |
1148 | #endif |
1149 | ||
1150 | static int wdog_reboot_handler(struct notifier_block *this, | |
1151 | unsigned long code, | |
1152 | void *unused) | |
1153 | { | |
36c7dc44 | 1154 | static int reboot_event_handled; |
1da177e4 LT |
1155 | |
1156 | if ((watchdog_user) && (!reboot_event_handled)) { | |
1157 | /* Make sure we only do this once. */ | |
1158 | reboot_event_handled = 1; | |
1159 | ||
fcfa4724 | 1160 | if (code == SYS_POWER_OFF || code == SYS_HALT) { |
1da177e4 LT |
1161 | /* Disable the WDT if we are shutting down. */ |
1162 | ipmi_watchdog_state = WDOG_TIMEOUT_NONE; | |
423a5bb4 | 1163 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
96febe9f | 1164 | } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { |
1da177e4 | 1165 | /* Set a long timer to let the reboot happens, but |
96febe9f CM |
1166 | reboot if it hangs, but only if the watchdog |
1167 | timer was already running. */ | |
1da177e4 LT |
1168 | timeout = 120; |
1169 | pretimeout = 0; | |
1170 | ipmi_watchdog_state = WDOG_TIMEOUT_RESET; | |
423a5bb4 | 1171 | ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB); |
1da177e4 LT |
1172 | } |
1173 | } | |
1174 | return NOTIFY_OK; | |
1175 | } | |
1176 | ||
1177 | static struct notifier_block wdog_reboot_notifier = { | |
1178 | .notifier_call = wdog_reboot_handler, | |
1179 | .next = NULL, | |
1180 | .priority = 0 | |
1181 | }; | |
1182 | ||
1183 | static int wdog_panic_handler(struct notifier_block *this, | |
1184 | unsigned long event, | |
1185 | void *unused) | |
1186 | { | |
36c7dc44 | 1187 | static int panic_event_handled; |
1da177e4 | 1188 | |
96febe9f CM |
1189 | /* On a panic, if we have a panic timeout, make sure to extend |
1190 | the watchdog timer to a reasonable value to complete the | |
1191 | panic, if the watchdog timer is running. Plus the | |
1192 | pretimeout is meaningless at panic time. */ | |
1193 | if (watchdog_user && !panic_event_handled && | |
1194 | ipmi_watchdog_state != WDOG_TIMEOUT_NONE) { | |
1195 | /* Make sure we do this only once. */ | |
1da177e4 | 1196 | panic_event_handled = 1; |
36c7dc44 | 1197 | |
c7f42c63 | 1198 | timeout = panic_wdt_timeout; |
1da177e4 | 1199 | pretimeout = 0; |
1da177e4 LT |
1200 | panic_halt_ipmi_set_timeout(); |
1201 | } | |
1202 | ||
1203 | return NOTIFY_OK; | |
1204 | } | |
1205 | ||
1206 | static struct notifier_block wdog_panic_notifier = { | |
1207 | .notifier_call = wdog_panic_handler, | |
1208 | .next = NULL, | |
1209 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ | |
1210 | }; | |
1211 | ||
1212 | ||
50c812b2 | 1213 | static void ipmi_new_smi(int if_num, struct device *device) |
1da177e4 LT |
1214 | { |
1215 | ipmi_register_watchdog(if_num); | |
1216 | } | |
1217 | ||
1218 | static void ipmi_smi_gone(int if_num) | |
1219 | { | |
b2c03941 | 1220 | ipmi_unregister_watchdog(if_num); |
1da177e4 LT |
1221 | } |
1222 | ||
36c7dc44 | 1223 | static struct ipmi_smi_watcher smi_watcher = { |
1da177e4 LT |
1224 | .owner = THIS_MODULE, |
1225 | .new_smi = ipmi_new_smi, | |
1226 | .smi_gone = ipmi_smi_gone | |
1227 | }; | |
1228 | ||
cc4673ee | 1229 | static int action_op(const char *inval, char *outval) |
1da177e4 | 1230 | { |
cc4673ee CM |
1231 | if (outval) |
1232 | strcpy(outval, action); | |
1233 | ||
1234 | if (!inval) | |
1235 | return 0; | |
1da177e4 | 1236 | |
cc4673ee | 1237 | if (strcmp(inval, "reset") == 0) |
1da177e4 | 1238 | action_val = WDOG_TIMEOUT_RESET; |
cc4673ee | 1239 | else if (strcmp(inval, "none") == 0) |
1da177e4 | 1240 | action_val = WDOG_TIMEOUT_NONE; |
cc4673ee | 1241 | else if (strcmp(inval, "power_cycle") == 0) |
1da177e4 | 1242 | action_val = WDOG_TIMEOUT_POWER_CYCLE; |
cc4673ee | 1243 | else if (strcmp(inval, "power_off") == 0) |
1da177e4 | 1244 | action_val = WDOG_TIMEOUT_POWER_DOWN; |
cc4673ee CM |
1245 | else |
1246 | return -EINVAL; | |
1247 | strcpy(action, inval); | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | static int preaction_op(const char *inval, char *outval) | |
1252 | { | |
1253 | if (outval) | |
1254 | strcpy(outval, preaction); | |
1da177e4 | 1255 | |
cc4673ee CM |
1256 | if (!inval) |
1257 | return 0; | |
1258 | ||
1259 | if (strcmp(inval, "pre_none") == 0) | |
1da177e4 | 1260 | preaction_val = WDOG_PRETIMEOUT_NONE; |
cc4673ee | 1261 | else if (strcmp(inval, "pre_smi") == 0) |
1da177e4 | 1262 | preaction_val = WDOG_PRETIMEOUT_SMI; |
612b5a8d | 1263 | #ifdef HAVE_DIE_NMI |
cc4673ee | 1264 | else if (strcmp(inval, "pre_nmi") == 0) |
1da177e4 LT |
1265 | preaction_val = WDOG_PRETIMEOUT_NMI; |
1266 | #endif | |
cc4673ee | 1267 | else if (strcmp(inval, "pre_int") == 0) |
1da177e4 | 1268 | preaction_val = WDOG_PRETIMEOUT_MSG_INT; |
cc4673ee CM |
1269 | else |
1270 | return -EINVAL; | |
1271 | strcpy(preaction, inval); | |
1272 | return 0; | |
1273 | } | |
1274 | ||
1275 | static int preop_op(const char *inval, char *outval) | |
1276 | { | |
1277 | if (outval) | |
1278 | strcpy(outval, preop); | |
1da177e4 | 1279 | |
cc4673ee CM |
1280 | if (!inval) |
1281 | return 0; | |
1282 | ||
1283 | if (strcmp(inval, "preop_none") == 0) | |
1da177e4 | 1284 | preop_val = WDOG_PREOP_NONE; |
cc4673ee | 1285 | else if (strcmp(inval, "preop_panic") == 0) |
1da177e4 | 1286 | preop_val = WDOG_PREOP_PANIC; |
cc4673ee | 1287 | else if (strcmp(inval, "preop_give_data") == 0) |
1da177e4 | 1288 | preop_val = WDOG_PREOP_GIVE_DATA; |
cc4673ee CM |
1289 | else |
1290 | return -EINVAL; | |
1291 | strcpy(preop, inval); | |
1292 | return 0; | |
1293 | } | |
1da177e4 | 1294 | |
cc4673ee CM |
1295 | static void check_parms(void) |
1296 | { | |
612b5a8d | 1297 | #ifdef HAVE_DIE_NMI |
cc4673ee CM |
1298 | int do_nmi = 0; |
1299 | int rv; | |
1300 | ||
1da177e4 | 1301 | if (preaction_val == WDOG_PRETIMEOUT_NMI) { |
cc4673ee | 1302 | do_nmi = 1; |
1da177e4 LT |
1303 | if (preop_val == WDOG_PREOP_GIVE_DATA) { |
1304 | printk(KERN_WARNING PFX "Pretimeout op is to give data" | |
1305 | " but NMI pretimeout is enabled, setting" | |
1306 | " pretimeout op to none\n"); | |
cc4673ee CM |
1307 | preop_op("preop_none", NULL); |
1308 | do_nmi = 0; | |
1da177e4 | 1309 | } |
cc4673ee CM |
1310 | } |
1311 | if (do_nmi && !nmi_handler_registered) { | |
9c48f1c6 DZ |
1312 | rv = register_nmi_handler(NMI_UNKNOWN, ipmi_nmi, 0, |
1313 | "ipmi"); | |
1da177e4 | 1314 | if (rv) { |
cc4673ee CM |
1315 | printk(KERN_WARNING PFX |
1316 | "Can't register nmi handler\n"); | |
1317 | return; | |
1318 | } else | |
1319 | nmi_handler_registered = 1; | |
1320 | } else if (!do_nmi && nmi_handler_registered) { | |
9c48f1c6 | 1321 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); |
cc4673ee | 1322 | nmi_handler_registered = 0; |
1da177e4 LT |
1323 | } |
1324 | #endif | |
cc4673ee CM |
1325 | } |
1326 | ||
1327 | static int __init ipmi_wdog_init(void) | |
1328 | { | |
1329 | int rv; | |
1330 | ||
1331 | if (action_op(action, NULL)) { | |
1332 | action_op("reset", NULL); | |
1333 | printk(KERN_INFO PFX "Unknown action '%s', defaulting to" | |
1334 | " reset\n", action); | |
1335 | } | |
1336 | ||
1337 | if (preaction_op(preaction, NULL)) { | |
1338 | preaction_op("pre_none", NULL); | |
1339 | printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to" | |
1340 | " none\n", preaction); | |
1341 | } | |
1342 | ||
1343 | if (preop_op(preop, NULL)) { | |
1344 | preop_op("preop_none", NULL); | |
1345 | printk(KERN_INFO PFX "Unknown preop '%s', defaulting to" | |
1346 | " none\n", preop); | |
1347 | } | |
1348 | ||
1349 | check_parms(); | |
1da177e4 | 1350 | |
b2c03941 CM |
1351 | register_reboot_notifier(&wdog_reboot_notifier); |
1352 | atomic_notifier_chain_register(&panic_notifier_list, | |
1353 | &wdog_panic_notifier); | |
1354 | ||
1da177e4 LT |
1355 | rv = ipmi_smi_watcher_register(&smi_watcher); |
1356 | if (rv) { | |
612b5a8d CM |
1357 | #ifdef HAVE_DIE_NMI |
1358 | if (nmi_handler_registered) | |
9c48f1c6 | 1359 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); |
1da177e4 | 1360 | #endif |
b2c03941 CM |
1361 | atomic_notifier_chain_unregister(&panic_notifier_list, |
1362 | &wdog_panic_notifier); | |
1363 | unregister_reboot_notifier(&wdog_reboot_notifier); | |
1da177e4 LT |
1364 | printk(KERN_WARNING PFX "can't register smi watcher\n"); |
1365 | return rv; | |
1366 | } | |
1367 | ||
1fdd75bd CM |
1368 | printk(KERN_INFO PFX "driver initialized\n"); |
1369 | ||
1da177e4 LT |
1370 | return 0; |
1371 | } | |
1372 | ||
b2c03941 | 1373 | static void __exit ipmi_wdog_exit(void) |
1da177e4 | 1374 | { |
b2c03941 CM |
1375 | ipmi_smi_watcher_unregister(&smi_watcher); |
1376 | ipmi_unregister_watchdog(watchdog_ifnum); | |
1da177e4 | 1377 | |
612b5a8d | 1378 | #ifdef HAVE_DIE_NMI |
cc4673ee | 1379 | if (nmi_handler_registered) |
9c48f1c6 | 1380 | unregister_nmi_handler(NMI_UNKNOWN, "ipmi"); |
1da177e4 LT |
1381 | #endif |
1382 | ||
e041c683 | 1383 | atomic_notifier_chain_unregister(&panic_notifier_list, |
b2c03941 | 1384 | &wdog_panic_notifier); |
1da177e4 | 1385 | unregister_reboot_notifier(&wdog_reboot_notifier); |
1da177e4 LT |
1386 | } |
1387 | module_exit(ipmi_wdog_exit); | |
1388 | module_init(ipmi_wdog_init); | |
1389 | MODULE_LICENSE("GPL"); | |
1fdd75bd CM |
1390 | MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>"); |
1391 | MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface."); |