blob: 8d16f6f3c4eccbbba5e28f1e7f60db24ad942fa7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001The Linux Watchdog driver API.
2
3Copyright 2002 Christer Weingel <wingel@nano-system.com>
4
5Some parts of this document are copied verbatim from the sbc60xxwdt
6driver which is (c) Copyright 2000 Jakob Oestergaard <jakob@ostenfeld.dk>
7
8This document describes the state of the Linux 2.4.18 kernel.
9
10Introduction:
11
12A Watchdog Timer (WDT) is a hardware circuit that can reset the
13computer system in case of a software fault. You probably knew that
14already.
15
16Usually a userspace daemon will notify the kernel watchdog driver via the
17/dev/watchdog special device file that userspace is still alive, at
18regular intervals. When such a notification occurs, the driver will
19usually tell the hardware watchdog that everything is in order, and
20that the watchdog should wait for yet another little while to reset
21the system. If userspace fails (RAM error, kernel bug, whatever), the
22notifications cease to occur, and the hardware watchdog will reset the
23system (causing a reboot) after the timeout occurs.
24
25The Linux watchdog API is a rather AD hoc construction and different
26drivers implement different, and sometimes incompatible, parts of it.
27This file is an attempt to document the existing usage and allow
28future driver writers to use it as a reference.
29
30The simplest API:
31
32All drivers support the basic mode of operation, where the watchdog
33activates as soon as /dev/watchdog is opened and will reboot unless
34the watchdog is pinged within a certain time, this time is called the
35timeout or margin. The simplest way to ping the watchdog is to write
36some data to the device. So a very simple watchdog daemon would look
Randy Dunlap56fb9e52006-05-21 20:58:10 -070037like this source file: see Documentation/watchdog/src/watchdog-simple.c
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39A more advanced driver could for example check that a HTTP server is
40still responding before doing the write call to ping the watchdog.
41
42When the device is closed, the watchdog is disabled. This is not
43always such a good idea, since if there is a bug in the watchdog
44daemon and it crashes the system will not reboot. Because of this,
45some of the drivers support the configuration option "Disable watchdog
46shutdown on close", CONFIG_WATCHDOG_NOWAYOUT. If it is set to Y when
47compiling the kernel, there is no way of disabling the watchdog once
Matt LaPlante2fe0ae72006-10-03 22:50:39 +020048it has been started. So, if the watchdog daemon crashes, the system
Linus Torvalds1da177e2005-04-16 15:20:36 -070049will reboot after the timeout has passed.
50
51Some other drivers will not disable the watchdog, unless a specific
52magic character 'V' has been sent /dev/watchdog just before closing
53the file. If the userspace daemon closes the file without sending
54this special character, the driver will assume that the daemon (and
55userspace in general) died, and will stop pinging the watchdog without
56disabling it first. This will then cause a reboot.
57
58The ioctl API:
59
60All conforming drivers also support an ioctl API.
61
62Pinging the watchdog using an ioctl:
63
64All drivers that have an ioctl interface support at least one ioctl,
65KEEPALIVE. This ioctl does exactly the same thing as a write to the
66watchdog device, so the main loop in the above program could be
67replaced with:
68
69 while (1) {
70 ioctl(fd, WDIOC_KEEPALIVE, 0);
71 sleep(10);
72 }
73
74the argument to the ioctl is ignored.
75
76Setting and getting the timeout:
77
78For some drivers it is possible to modify the watchdog timeout on the
79fly with the SETTIMEOUT ioctl, those drivers have the WDIOF_SETTIMEOUT
80flag set in their option field. The argument is an integer
81representing the timeout in seconds. The driver returns the real
82timeout used in the same variable, and this timeout might differ from
83the requested one due to limitation of the hardware.
84
85 int timeout = 45;
86 ioctl(fd, WDIOC_SETTIMEOUT, &timeout);
87 printf("The timeout was set to %d seconds\n", timeout);
88
89This example might actually print "The timeout was set to 60 seconds"
90if the device has a granularity of minutes for its timeout.
91
92Starting with the Linux 2.4.18 kernel, it is possible to query the
93current timeout using the GETTIMEOUT ioctl.
94
95 ioctl(fd, WDIOC_GETTIMEOUT, &timeout);
96 printf("The timeout was is %d seconds\n", timeout);
97
Corey Minyarde05b59f2006-04-19 22:40:53 +020098Pretimeouts:
99
100Some watchdog timers can be set to have a trigger go off before the
101actual time they will reset the system. This can be done with an NMI,
102interrupt, or other mechanism. This allows Linux to record useful
103information (like panic information and kernel coredumps) before it
104resets.
105
106 pretimeout = 10;
107 ioctl(fd, WDIOC_SETPRETIMEOUT, &pretimeout);
108
109Note that the pretimeout is the number of seconds before the time
110when the timeout will go off. It is not the number of seconds until
111the pretimeout. So, for instance, if you set the timeout to 60 seconds
112and the pretimeout to 10 seconds, the pretimout will go of in 50
113seconds. Setting a pretimeout to zero disables it.
114
115There is also a get function for getting the pretimeout:
116
117 ioctl(fd, WDIOC_GETPRETIMEOUT, &timeout);
118 printf("The pretimeout was is %d seconds\n", timeout);
119
120Not all watchdog drivers will support a pretimeout.
121
Wim Van Sebroeck58b519f2006-05-21 12:48:44 +0200122Get the number of seconds before reboot:
123
124Some watchdog drivers have the ability to report the remaining time
125before the system will reboot. The WDIOC_GETTIMELEFT is the ioctl
126that returns the number of seconds before reboot.
127
128 ioctl(fd, WDIOC_GETTIMELEFT, &timeleft);
129 printf("The timeout was is %d seconds\n", timeleft);
130
Corey Minyarde05b59f2006-04-19 22:40:53 +0200131Environmental monitoring:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133All watchdog drivers are required return more information about the system,
134some do temperature, fan and power level monitoring, some can tell you
135the reason for the last reboot of the system. The GETSUPPORT ioctl is
136available to ask what the device can do:
137
138 struct watchdog_info ident;
139 ioctl(fd, WDIOC_GETSUPPORT, &ident);
140
141the fields returned in the ident struct are:
142
143 identity a string identifying the watchdog driver
144 firmware_version the firmware version of the card if available
145 options a flags describing what the device supports
146
147the options field can have the following bits set, and describes what
148kind of information that the GET_STATUS and GET_BOOT_STATUS ioctls can
149return. [FIXME -- Is this correct?]
150
151 WDIOF_OVERHEAT Reset due to CPU overheat
152
153The machine was last rebooted by the watchdog because the thermal limit was
154exceeded
155
156 WDIOF_FANFAULT Fan failed
157
158A system fan monitored by the watchdog card has failed
159
160 WDIOF_EXTERN1 External relay 1
161
162External monitoring relay/source 1 was triggered. Controllers intended for
163real world applications include external monitoring pins that will trigger
164a reset.
165
166 WDIOF_EXTERN2 External relay 2
167
168External monitoring relay/source 2 was triggered
169
170 WDIOF_POWERUNDER Power bad/power fault
171
172The machine is showing an undervoltage status
173
174 WDIOF_CARDRESET Card previously reset the CPU
175
176The last reboot was caused by the watchdog card
177
178 WDIOF_POWEROVER Power over voltage
179
180The machine is showing an overvoltage status. Note that if one level is
181under and one over both bits will be set - this may seem odd but makes
182sense.
183
184 WDIOF_KEEPALIVEPING Keep alive ping reply
185
186The watchdog saw a keepalive ping since it was last queried.
187
188 WDIOF_SETTIMEOUT Can set/get the timeout
189
Corey Minyarde05b59f2006-04-19 22:40:53 +0200190The watchdog can do pretimeouts.
191
192 WDIOF_PRETIMEOUT Pretimeout (in seconds), get/set
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195For those drivers that return any bits set in the option field, the
196GETSTATUS and GETBOOTSTATUS ioctls can be used to ask for the current
197status, and the status at the last reboot, respectively.
198
199 int flags;
200 ioctl(fd, WDIOC_GETSTATUS, &flags);
201
202 or
203
204 ioctl(fd, WDIOC_GETBOOTSTATUS, &flags);
205
206Note that not all devices support these two calls, and some only
207support the GETBOOTSTATUS call.
208
209Some drivers can measure the temperature using the GETTEMP ioctl. The
Matt LaPlantea2ffd272006-10-03 22:49:15 +0200210returned value is the temperature in degrees fahrenheit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212 int temperature;
213 ioctl(fd, WDIOC_GETTEMP, &temperature);
214
215Finally the SETOPTIONS ioctl can be used to control some aspects of
216the cards operation; right now the pcwd driver is the only one
Matt LaPlantefa00e7e2006-11-30 04:55:36 +0100217supporting this ioctl.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 int options = 0;
220 ioctl(fd, WDIOC_SETOPTIONS, options);
221
222The following options are available:
223
224 WDIOS_DISABLECARD Turn off the watchdog timer
225 WDIOS_ENABLECARD Turn on the watchdog timer
226 WDIOS_TEMPPANIC Kernel panic on temperature trip
227
228[FIXME -- better explanations]
229
230Implementations in the current drivers in the kernel tree:
231
232Here I have tried to summarize what the different drivers support and
233where they do strange things compared to the other drivers.
234
235acquirewdt.c -- Acquire Single Board Computer
236
237 This driver has a hardcoded timeout of 1 minute
238
239 Supports CONFIG_WATCHDOG_NOWAYOUT
240
241 GETSUPPORT returns KEEPALIVEPING. GETSTATUS will return 1 if
242 the device is open, 0 if not. [FIXME -- isn't this rather
243 silly? To be able to use the ioctl, the device must be open
244 and so GETSTATUS will always return 1].
245
246advantechwdt.c -- Advantech Single Board Computer
247
248 Timeout that defaults to 60 seconds, supports SETTIMEOUT.
249
250 Supports CONFIG_WATCHDOG_NOWAYOUT
251
252 GETSUPPORT returns WDIOF_KEEPALIVEPING and WDIOF_SETTIMEOUT.
253 The GETSTATUS call returns if the device is open or not.
254 [FIXME -- silliness again?]
255
Kumar Galaa2f40cc2005-09-03 15:55:33 -0700256booke_wdt.c -- PowerPC BookE Watchdog Timer
257
258 Timeout default varies according to frequency, supports
259 SETTIMEOUT
260
Matt LaPlante84eb8d02006-10-03 22:53:09 +0200261 Watchdog cannot be turned off, CONFIG_WATCHDOG_NOWAYOUT
Kumar Galaa2f40cc2005-09-03 15:55:33 -0700262 does not make sense
263
264 GETSUPPORT returns the watchdog_info struct, and
265 GETSTATUS returns the supported options. GETBOOTSTATUS
266 returns a 1 if the last reset was caused by the
Matt LaPlante84eb8d02006-10-03 22:53:09 +0200267 watchdog and a 0 otherwise. This watchdog cannot be
Kumar Galaa2f40cc2005-09-03 15:55:33 -0700268 disabled once it has been started. The wdt_period kernel
269 parameter selects which bit of the time base changing
270 from 0->1 will trigger the watchdog exception. Changing
271 the timeout from the ioctl calls will change the
272 wdt_period as defined above. Finally if you would like to
273 replace the default Watchdog Handler you can implement the
274 WatchdogHandler() function in your own code.
275
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276eurotechwdt.c -- Eurotech CPU-1220/1410
277
278 The timeout can be set using the SETTIMEOUT ioctl and defaults
279 to 60 seconds.
280
281 Also has a module parameter "ev", event type which controls
282 what should happen on a timeout, the string "int" or anything
283 else that causes a reboot. [FIXME -- better description]
284
285 Supports CONFIG_WATCHDOG_NOWAYOUT
286
287 GETSUPPORT returns CARDRESET and WDIOF_SETTIMEOUT but
288 GETSTATUS is not supported and GETBOOTSTATUS just returns 0.
289
290i810-tco.c -- Intel 810 chipset
291
292 Also has support for a lot of other i8x0 stuff, but the
293 watchdog is one of the things.
294
295 The timeout is set using the module parameter "i810_margin",
296 which is in steps of 0.6 seconds where 2<i810_margin<64. The
297 driver supports the SETTIMEOUT ioctl.
298
299 Supports CONFIG_WATCHDOG_NOWAYOUT.
300
301 GETSUPPORT returns WDIOF_SETTIMEOUT. The GETSTATUS call
302 returns some kind of timer value which ist not compatible with
303 the other drivers. GETBOOT status returns some kind of
304 hardware specific boot status. [FIXME -- describe this]
305
306ib700wdt.c -- IB700 Single Board Computer
307
308 Default timeout of 30 seconds and the timeout is settable
309 using the SETTIMEOUT ioctl. Note that only a few timeout
310 values are supported.
311
312 Supports CONFIG_WATCHDOG_NOWAYOUT
313
314 GETSUPPORT returns WDIOF_KEEPALIVEPING and WDIOF_SETTIMEOUT.
315 The GETSTATUS call returns if the device is open or not.
316 [FIXME -- silliness again?]
317
318machzwd.c -- MachZ ZF-Logic
319
320 Hardcoded timeout of 10 seconds
321
322 Has a module parameter "action" that controls what happens
323 when the timeout runs out which can be 0 = RESET (default),
324 1 = SMI, 2 = NMI, 3 = SCI.
325
326 Supports CONFIG_WATCHDOG_NOWAYOUT and the magic character
327 'V' close handling.
328
329 GETSUPPORT returns WDIOF_KEEPALIVEPING, and the GETSTATUS call
330 returns if the device is open or not. [FIXME -- silliness
331 again?]
332
333mixcomwd.c -- MixCom Watchdog
334
335 [FIXME -- I'm unable to tell what the timeout is]
336
337 Supports CONFIG_WATCHDOG_NOWAYOUT
338
339 GETSUPPORT returns WDIOF_KEEPALIVEPING, GETSTATUS returns if
340 the device is opened or not [FIXME -- I'm not really sure how
341 this works, there seems to be some magic connected to
342 CONFIG_WATCHDOG_NOWAYOUT]
343
344pcwd.c -- Berkshire PC Watchdog
345
346 Hardcoded timeout of 1.5 seconds
347
348 Supports CONFIG_WATCHDOG_NOWAYOUT
349
350 GETSUPPORT returns WDIOF_OVERHEAT|WDIOF_CARDRESET and both
351 GETSTATUS and GETBOOTSTATUS return something useful.
352
353 The SETOPTIONS call can be used to enable and disable the card
354 and to ask the driver to call panic if the system overheats.
355
356sbc60xxwdt.c -- 60xx Single Board Computer
357
358 Hardcoded timeout of 10 seconds
359
360 Does not support CONFIG_WATCHDOG_NOWAYOUT, but has the magic
361 character 'V' close handling.
362
363 No bits set in GETSUPPORT
364
365scx200.c -- National SCx200 CPUs
366
367 Not in the kernel yet.
368
369 The timeout is set using a module parameter "margin" which
370 defaults to 60 seconds. The timeout can also be set using
371 SETTIMEOUT and read using GETTIMEOUT.
372
373 Supports a module parameter "nowayout" that is initialized
374 with the value of CONFIG_WATCHDOG_NOWAYOUT. Also supports the
375 magic character 'V' handling.
376
377shwdt.c -- SuperH 3/4 processors
378
379 [FIXME -- I'm unable to tell what the timeout is]
380
381 Supports CONFIG_WATCHDOG_NOWAYOUT
382
383 GETSUPPORT returns WDIOF_KEEPALIVEPING, and the GETSTATUS call
384 returns if the device is open or not. [FIXME -- silliness
385 again?]
386
387softdog.c -- Software watchdog
388
389 The timeout is set with the module parameter "soft_margin"
390 which defaults to 60 seconds, the timeout is also settable
391 using the SETTIMEOUT ioctl.
392
393 Supports CONFIG_WATCHDOG_NOWAYOUT
394
395 WDIOF_SETTIMEOUT bit set in GETSUPPORT
396
397w83877f_wdt.c -- W83877F Computer
398
399 Hardcoded timeout of 30 seconds
400
401 Does not support CONFIG_WATCHDOG_NOWAYOUT, but has the magic
402 character 'V' close handling.
403
404 No bits set in GETSUPPORT
405
406w83627hf_wdt.c -- w83627hf watchdog
407
408 Timeout that defaults to 60 seconds, supports SETTIMEOUT.
409
410 Supports CONFIG_WATCHDOG_NOWAYOUT
411
412 GETSUPPORT returns WDIOF_KEEPALIVEPING and WDIOF_SETTIMEOUT.
413 The GETSTATUS call returns if the device is open or not.
414
415wdt.c -- ICS WDT500/501 ISA and
416wdt_pci.c -- ICS WDT500/501 PCI
417
418 Default timeout of 60 seconds. The timeout is also settable
419 using the SETTIMEOUT ioctl.
420
421 Supports CONFIG_WATCHDOG_NOWAYOUT
422
423 GETSUPPORT returns with bits set depending on the actual
424 card. The WDT501 supports a lot of external monitoring, the
425 WDT500 much less.
426
427wdt285.c -- Footbridge watchdog
428
429 The timeout is set with the module parameter "soft_margin"
430 which defaults to 60 seconds. The timeout is also settable
431 using the SETTIMEOUT ioctl.
432
433 Does not support CONFIG_WATCHDOG_NOWAYOUT
434
435 WDIOF_SETTIMEOUT bit set in GETSUPPORT
436
437wdt977.c -- Netwinder W83977AF chip
438
439 Hardcoded timeout of 3 minutes
440
441 Supports CONFIG_WATCHDOG_NOWAYOUT
442
443 Does not support any ioctls at all.
444