Documentation/watchdog/watchdog-kernel-api.txt

   1 The Linux WatchDog Timer Driver Core kernel API.
   2 ===============================================
   3 Last reviewed: 12-Feb-2013
   4
   5 Wim Van Sebroeck <wim@iguana.be>
   6
   7 Introduction
   8 ------------
   9 This document does not describe what a WatchDog Timer (WDT) Driver or Device is.
  10 It also does not describe the API which can be used by user space to communicate
  11 with a WatchDog Timer. If you want to know this then please read the following
  12 file: Documentation/watchdog/watchdog-api.txt .
  13
  14 So what does this document describe? It describes the API that can be used by
  15 WatchDog Timer Drivers that want to use the WatchDog Timer Driver Core
  16 Framework. This framework provides all interfacing towards user space so that
  17 the same code does not have to be reproduced each time. This also means that
  18 a watchdog timer driver then only needs to provide the different routines
  19 (operations) that control the watchdog timer (WDT).
  20
  21 The API
  22 -------
  23 Each watchdog timer driver that wants to use the WatchDog Timer Driver Core
  24 must #include <linux/watchdog.h> (you would have to do this anyway when
  25 writing a watchdog device driver). This include file contains following
  26 register/unregister routines:
  27
  28 extern int watchdog_register_device(struct watchdog_device *);
  29 extern void watchdog_unregister_device(struct watchdog_device *);
  30
  31 The watchdog_register_device routine registers a watchdog timer device.
  32 The parameter of this routine is a pointer to a watchdog_device structure.
  33 This routine returns zero on success and a negative errno code for failure.
  34
  35 The watchdog_unregister_device routine deregisters a registered watchdog timer
  36 device. The parameter of this routine is the pointer to the registered
  37 watchdog_device structure.
  38
  39 The watchdog subsystem includes an registration deferral mechanism,
  40 which allows you to register an watchdog as early as you wish during
  41 the boot process.
  42
  43 The watchdog device structure looks like this:
  44
  45 struct watchdog_device {
  46         int id;
  47         struct device *parent;
  48         const struct attribute_group **groups;
  49         const struct watchdog_info *info;
  50         const struct watchdog_ops *ops;
  51         const struct watchdog_governor *gov;
  52         unsigned int bootstatus;
  53         unsigned int timeout;
  54         unsigned int pretimeout;
  55         unsigned int min_timeout;
  56         unsigned int max_timeout;
  57         unsigned int min_hw_heartbeat_ms;
  58         unsigned int max_hw_heartbeat_ms;
  59         struct notifier_block reboot_nb;
  60         struct notifier_block restart_nb;
  61         void *driver_data;
  62         struct watchdog_core_data *wd_data;
  63         unsigned long status;
  64         struct list_head deferred;
  65 };
  66
  67 It contains following fields:
  68 * id: set by watchdog_register_device, id 0 is special. It has both a
  69   /dev/watchdog0 cdev (dynamic major, minor 0) as well as the old
  70   /dev/watchdog miscdev. The id is set automatically when calling
  71   watchdog_register_device.
  72 * parent: set this to the parent device (or NULL) before calling
  73   watchdog_register_device.
  74 * groups: List of sysfs attribute groups to create when creating the watchdog
  75   device.
  76 * info: a pointer to a watchdog_info structure. This structure gives some
  77   additional information about the watchdog timer itself. (Like it's unique name)
  78 * ops: a pointer to the list of watchdog operations that the watchdog supports.
  79 * gov: a pointer to the assigned watchdog device pretimeout governor or NULL.
  80 * timeout: the watchdog timer's timeout value (in seconds).
  81   This is the time after which the system will reboot if user space does
  82   not send a heartbeat request if WDOG_ACTIVE is set.
  83 * pretimeout: the watchdog timer's pretimeout value (in seconds).
  84 * min_timeout: the watchdog timer's minimum timeout value (in seconds).
  85   If set, the minimum configurable value for 'timeout'.
  86 * max_timeout: the watchdog timer's maximum timeout value (in seconds),
  87   as seen from userspace. If set, the maximum configurable value for
  88   'timeout'. Not used if max_hw_heartbeat_ms is non-zero.
  89 * min_hw_heartbeat_ms: Hardware limit for minimum time between heartbeats,
  90   in milli-seconds. This value is normally 0; it should only be provided
  91   if the hardware can not tolerate lower intervals between heartbeats.
  92 * max_hw_heartbeat_ms: Maximum hardware heartbeat, in milli-seconds.
  93   If set, the infrastructure will send heartbeats to the watchdog driver
  94   if 'timeout' is larger than max_hw_heartbeat_ms, unless WDOG_ACTIVE
  95   is set and userspace failed to send a heartbeat for at least 'timeout'
  96   seconds. max_hw_heartbeat_ms must be set if a driver does not implement
  97   the stop function.
  98 * reboot_nb: notifier block that is registered for reboot notifications, for
  99   internal use only. If the driver calls watchdog_stop_on_reboot, watchdog core
 100   will stop the watchdog on such notifications.
 101 * restart_nb: notifier block that is registered for machine restart, for
 102   internal use only. If a watchdog is capable of restarting the machine, it
 103   should define ops->restart. Priority can be changed through
 104   watchdog_set_restart_priority.
 105 * bootstatus: status of the device after booting (reported with watchdog
 106   WDIOF_* status bits).
 107 * driver_data: a pointer to the drivers private data of a watchdog device.
 108   This data should only be accessed via the watchdog_set_drvdata and
 109   watchdog_get_drvdata routines.
 110 * wd_data: a pointer to watchdog core internal data.
 111 * status: this field contains a number of status bits that give extra
 112   information about the status of the device (Like: is the watchdog timer
 113   running/active, or is the nowayout bit set).
 114 * deferred: entry in wtd_deferred_reg_list which is used to
 115   register early initialized watchdogs.
 116
 117 The list of watchdog operations is defined as:
 118
 119 struct watchdog_ops {
 120         struct module *owner;
 121         /* mandatory operations */
 122         int (*start)(struct watchdog_device *);
 123         int (*stop)(struct watchdog_device *);
 124         /* optional operations */
 125         int (*ping)(struct watchdog_device *);
 126         unsigned int (*status)(struct watchdog_device *);
 127         int (*set_timeout)(struct watchdog_device *, unsigned int);
 128         int (*set_pretimeout)(struct watchdog_device *, unsigned int);
 129         unsigned int (*get_timeleft)(struct watchdog_device *);
 130         int (*restart)(struct watchdog_device *);
 131         void (*ref)(struct watchdog_device *) __deprecated;
 132         void (*unref)(struct watchdog_device *) __deprecated;
 133         long (*ioctl)(struct watchdog_device *, unsigned int, unsigned long);
 134 };
 135
 136 It is important that you first define the module owner of the watchdog timer
 137 driver's operations. This module owner will be used to lock the module when
 138 the watchdog is active. (This to avoid a system crash when you unload the
 139 module and /dev/watchdog is still open).
 140
 141 Some operations are mandatory and some are optional. The mandatory operations
 142 are:
 143 * start: this is a pointer to the routine that starts the watchdog timer
 144   device.
 145   The routine needs a pointer to the watchdog timer device structure as a
 146   parameter. It returns zero on success or a negative errno code for failure.
 147
 148 Not all watchdog timer hardware supports the same functionality. That's why
 149 all other routines/operations are optional. They only need to be provided if
 150 they are supported. These optional routines/operations are:
 151 * stop: with this routine the watchdog timer device is being stopped.
 152   The routine needs a pointer to the watchdog timer device structure as a
 153   parameter. It returns zero on success or a negative errno code for failure.
 154   Some watchdog timer hardware can only be started and not be stopped. A
 155   driver supporting such hardware does not have to implement the stop routine.
 156   If a driver has no stop function, the watchdog core will set WDOG_HW_RUNNING
 157   and start calling the driver's keepalive pings function after the watchdog
 158   device is closed.
 159   If a watchdog driver does not implement the stop function, it must set
 160   max_hw_heartbeat_ms.
 161 * ping: this is the routine that sends a keepalive ping to the watchdog timer
 162   hardware.
 163   The routine needs a pointer to the watchdog timer device structure as a
 164   parameter. It returns zero on success or a negative errno code for failure.
 165   Most hardware that does not support this as a separate function uses the
 166   start function to restart the watchdog timer hardware. And that's also what
 167   the watchdog timer driver core does: to send a keepalive ping to the watchdog
 168   timer hardware it will either use the ping operation (when available) or the
 169   start operation (when the ping operation is not available).
 170   (Note: the WDIOC_KEEPALIVE ioctl call will only be active when the
 171   WDIOF_KEEPALIVEPING bit has been set in the option field on the watchdog's
 172   info structure).
 173 * status: this routine checks the status of the watchdog timer device. The
 174   status of the device is reported with watchdog WDIOF_* status flags/bits.
 175   WDIOF_MAGICCLOSE and WDIOF_KEEPALIVEPING are reported by the watchdog core;
 176   it is not necessary to report those bits from the driver. Also, if no status
 177   function is provided by the driver, the watchdog core reports the status bits
 178   provided in the bootstatus variable of struct watchdog_device.
 179 * set_timeout: this routine checks and changes the timeout of the watchdog
 180   timer device. It returns 0 on success, -EINVAL for "parameter out of range"
 181   and -EIO for "could not write value to the watchdog". On success this
 182   routine should set the timeout value of the watchdog_device to the
 183   achieved timeout value (which may be different from the requested one
 184   because the watchdog does not necessarily have a 1 second resolution).
 185   Drivers implementing max_hw_heartbeat_ms set the hardware watchdog heartbeat
 186   to the minimum of timeout and max_hw_heartbeat_ms. Those drivers set the
 187   timeout value of the watchdog_device either to the requested timeout value
 188   (if it is larger than max_hw_heartbeat_ms), or to the achieved timeout value.
 189   (Note: the WDIOF_SETTIMEOUT needs to be set in the options field of the
 190   watchdog's info structure).
 191   If the watchdog driver does not have to perform any action but setting the
 192   watchdog_device.timeout, this callback can be omitted.
 193   If set_timeout is not provided but, WDIOF_SETTIMEOUT is set, the watchdog
 194   infrastructure updates the timeout value of the watchdog_device internally
 195   to the requested value.
 196   If the pretimeout feature is used (WDIOF_PRETIMEOUT), then set_timeout must
 197   also take care of checking if pretimeout is still valid and set up the timer
 198   accordingly. This can't be done in the core without races, so it is the
 199   duty of the driver.
 200 * set_pretimeout: this routine checks and changes the pretimeout value of
 201   the watchdog. It is optional because not all watchdogs support pretimeout
 202   notification. The timeout value is not an absolute time, but the number of
 203   seconds before the actual timeout would happen. It returns 0 on success,
 204   -EINVAL for "parameter out of range" and -EIO for "could not write value to
 205   the watchdog". A value of 0 disables pretimeout notification.
 206   (Note: the WDIOF_PRETIMEOUT needs to be set in the options field of the
 207   watchdog's info structure).
 208   If the watchdog driver does not have to perform any action but setting the
 209   watchdog_device.pretimeout, this callback can be omitted. That means if
 210   set_pretimeout is not provided but WDIOF_PRETIMEOUT is set, the watchdog
 211   infrastructure updates the pretimeout value of the watchdog_device internally
 212   to the requested value.
 213 * get_timeleft: this routines returns the time that's left before a reset.
 214 * restart: this routine restarts the machine. It returns 0 on success or a
 215   negative errno code for failure.
 216 * ioctl: if this routine is present then it will be called first before we do
 217   our own internal ioctl call handling. This routine should return -ENOIOCTLCMD
 218   if a command is not supported. The parameters that are passed to the ioctl
 219   call are: watchdog_device, cmd and arg.
 220
 221 The 'ref' and 'unref' operations are no longer used and deprecated.
 222
 223 The status bits should (preferably) be set with the set_bit and clear_bit alike
 224 bit-operations. The status bits that are defined are:
 225 * WDOG_ACTIVE: this status bit indicates whether or not a watchdog timer device
 226   is active or not from user perspective. User space is expected to send
 227   heartbeat requests to the driver while this flag is set.
 228 * WDOG_NO_WAY_OUT: this bit stores the nowayout setting for the watchdog.
 229   If this bit is set then the watchdog timer will not be able to stop.
 230 * WDOG_HW_RUNNING: Set by the watchdog driver if the hardware watchdog is
 231   running. The bit must be set if the watchdog timer hardware can not be
 232   stopped. The bit may also be set if the watchdog timer is running after
 233   booting, before the watchdog device is opened. If set, the watchdog
 234   infrastructure will send keepalives to the watchdog hardware while
 235   WDOG_ACTIVE is not set.
 236   Note: when you register the watchdog timer device with this bit set,
 237   then opening /dev/watchdog will skip the start operation but send a keepalive
 238   request instead.
 239
 240   To set the WDOG_NO_WAY_OUT status bit (before registering your watchdog
 241   timer device) you can either:
 242   * set it statically in your watchdog_device struct with
 243         .status = WATCHDOG_NOWAYOUT_INIT_STATUS,
 244     (this will set the value the same as CONFIG_WATCHDOG_NOWAYOUT) or
 245   * use the following helper function:
 246   static inline void watchdog_set_nowayout(struct watchdog_device *wdd, int nowayout)
 247
 248 Note: The WatchDog Timer Driver Core supports the magic close feature and
 249 the nowayout feature. To use the magic close feature you must set the
 250 WDIOF_MAGICCLOSE bit in the options field of the watchdog's info structure.
 251 The nowayout feature will overrule the magic close feature.
 252
 253 To get or set driver specific data the following two helper functions should be
 254 used:
 255
 256 static inline void watchdog_set_drvdata(struct watchdog_device *wdd, void *data)
 257 static inline void *watchdog_get_drvdata(struct watchdog_device *wdd)
 258
 259 The watchdog_set_drvdata function allows you to add driver specific data. The
 260 arguments of this function are the watchdog device where you want to add the
 261 driver specific data to and a pointer to the data itself.
 262
 263 The watchdog_get_drvdata function allows you to retrieve driver specific data.
 264 The argument of this function is the watchdog device where you want to retrieve
 265 data from. The function returns the pointer to the driver specific data.
 266
 267 To initialize the timeout field, the following function can be used:
 268
 269 extern int watchdog_init_timeout(struct watchdog_device *wdd,
 270                                   unsigned int timeout_parm, struct device *dev);
 271
 272 The watchdog_init_timeout function allows you to initialize the timeout field
 273 using the module timeout parameter or by retrieving the timeout-sec property from
 274 the device tree (if the module timeout parameter is invalid). Best practice is
 275 to set the default timeout value as timeout value in the watchdog_device and
 276 then use this function to set the user "preferred" timeout value.
 277 This routine returns zero on success and a negative errno code for failure.
 278
 279 To disable the watchdog on reboot, the user must call the following helper:
 280
 281 static inline void watchdog_stop_on_reboot(struct watchdog_device *wdd);
 282
 283 To disable the watchdog when unregistering the watchdog, the user must call
 284 the following helper. Note that this will only stop the watchdog if the
 285 nowayout flag is not set.
 286
 287 static inline void watchdog_stop_on_unregister(struct watchdog_device *wdd);
 288
 289 To change the priority of the restart handler the following helper should be
 290 used:
 291
 292 void watchdog_set_restart_priority(struct watchdog_device *wdd, int priority);
 293
 294 User should follow the following guidelines for setting the priority:
 295 * 0: should be called in last resort, has limited restart capabilities
 296 * 128: default restart handler, use if no other handler is expected to be
 297   available, and/or if restart is sufficient to restart the entire system
 298 * 255: highest priority, will preempt all other restart handlers
 299
 300 To raise a pretimeout notification, the following function should be used:
 301
 302 void watchdog_notify_pretimeout(struct watchdog_device *wdd)
 303
 304 The function can be called in the interrupt context. If watchdog pretimeout
 305 governor framework (kbuild CONFIG_WATCHDOG_PRETIMEOUT_GOV symbol) is enabled,
 306 an action is taken by a preconfigured pretimeout governor preassigned to
 307 the watchdog device. If watchdog pretimeout governor framework is not
 308 enabled, watchdog_notify_pretimeout() prints a notification message to
 309 the kernel log buffer.