4 * A watchdog timer based upon the IPMI interface.
6 * Author: MontaVista Software, Inc.
7 * Corey Minyard <minyard@mvista.com>
10 * Copyright 2002 MontaVista Software Inc.
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.
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.
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.
34 #include <linux/module.h>
35 #include <linux/moduleparam.h>
36 #include <linux/ipmi.h>
37 #include <linux/ipmi_smi.h>
38 #include <linux/mutex.h>
39 #include <linux/watchdog.h>
40 #include <linux/miscdevice.h>
41 #include <linux/init.h>
42 #include <linux/completion.h>
43 #include <linux/kdebug.h>
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>
52 #include <linux/string.h>
53 #include <linux/ctype.h>
54 #include <linux/delay.h>
55 #include <linux/atomic.h>
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.
67 #include <asm/kdebug.h>
72 #define PFX "IPMI Watchdog: "
75 * The IPMI command/response information for the watchdog timer.
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
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
99 /* Operations that can be performed on a pretimout. */
100 #define WDOG_PREOP_NONE 0
101 #define WDOG_PREOP_PANIC 1
102 /* Cause data to be available to read. Doesn't work in NMI mode. */
103 #define WDOG_PREOP_GIVE_DATA 2
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
115 * Byte 3 of the get command, byte 4 of the get response is the
116 * pre-timeout in seconds.
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)
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).
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)
138 #define IPMI_WDOG_RESET_TIMER 0x22
139 #define IPMI_WDOG_SET_TIMER 0x24
140 #define IPMI_WDOG_GET_TIMER 0x25
142 /* These are here until the real ones get into the watchdog.h interface. */
143 #ifndef WDIOC_GETTIMEOUT
144 #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int)
146 #ifndef WDIOC_SET_PRETIMEOUT
147 #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int)
149 #ifndef WDIOC_GET_PRETIMEOUT
150 #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int)
153 static DEFINE_MUTEX(ipmi_watchdog_mutex
);
154 static int nowayout
= WATCHDOG_NOWAYOUT
;
156 static ipmi_user_t watchdog_user
;
157 static int watchdog_ifnum
;
159 /* Default the timeout to 10 seconds. */
160 static int timeout
= 10;
162 /* The pre-timeout is disabled by default. */
163 static int pretimeout
;
165 /* Default action is to reset the board on a timeout. */
166 static unsigned char action_val
= WDOG_TIMEOUT_RESET
;
168 static char action
[16] = "reset";
170 static unsigned char preaction_val
= WDOG_PRETIMEOUT_NONE
;
172 static char preaction
[16] = "pre_none";
174 static unsigned char preop_val
= WDOG_PREOP_NONE
;
176 static char preop
[16] = "preop_none";
177 static DEFINE_SPINLOCK(ipmi_read_lock
);
178 static char data_to_read
;
179 static DECLARE_WAIT_QUEUE_HEAD(read_q
);
180 static struct fasync_struct
*fasync_q
;
181 static char pretimeout_since_last_heartbeat
;
182 static char expect_close
;
184 static int ifnum_to_use
= -1;
186 /* Parameters to ipmi_set_timeout */
187 #define IPMI_SET_TIMEOUT_NO_HB 0
188 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
189 #define IPMI_SET_TIMEOUT_FORCE_HB 2
191 static int ipmi_set_timeout(int do_heartbeat
);
192 static void ipmi_register_watchdog(int ipmi_intf
);
193 static void ipmi_unregister_watchdog(int ipmi_intf
);
196 * If true, the driver will start running as soon as it is configured
199 static int start_now
;
201 static int set_param_timeout(const char *val
, const struct kernel_param
*kp
)
209 l
= simple_strtoul(val
, &endp
, 0);
213 *((int *)kp
->arg
) = l
;
215 rv
= ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
220 static struct kernel_param_ops param_ops_timeout
= {
221 .set
= set_param_timeout
,
222 .get
= param_get_int
,
224 #define param_check_timeout param_check_int
226 typedef int (*action_fn
)(const char *intval
, char *outval
);
228 static int action_op(const char *inval
, char *outval
);
229 static int preaction_op(const char *inval
, char *outval
);
230 static int preop_op(const char *inval
, char *outval
);
231 static void check_parms(void);
233 static int set_param_str(const char *val
, const struct kernel_param
*kp
)
235 action_fn fn
= (action_fn
) kp
->arg
;
240 strncpy(valcp
, val
, 16);
251 rv
= ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
257 static int get_param_str(char *buffer
, const struct kernel_param
*kp
)
259 action_fn fn
= (action_fn
) kp
->arg
;
262 rv
= fn(NULL
, buffer
);
265 return strlen(buffer
);
269 static int set_param_wdog_ifnum(const char *val
, const struct kernel_param
*kp
)
271 int rv
= param_set_int(val
, kp
);
274 if ((ifnum_to_use
< 0) || (ifnum_to_use
== watchdog_ifnum
))
277 ipmi_unregister_watchdog(watchdog_ifnum
);
278 ipmi_register_watchdog(ifnum_to_use
);
282 static struct kernel_param_ops param_ops_wdog_ifnum
= {
283 .set
= set_param_wdog_ifnum
,
284 .get
= param_get_int
,
287 #define param_check_wdog_ifnum param_check_int
289 static struct kernel_param_ops param_ops_str
= {
290 .set
= set_param_str
,
291 .get
= get_param_str
,
294 module_param(ifnum_to_use
, wdog_ifnum
, 0644);
295 MODULE_PARM_DESC(ifnum_to_use
, "The interface number to use for the watchdog "
296 "timer. Setting to -1 defaults to the first registered "
299 module_param(timeout
, timeout
, 0644);
300 MODULE_PARM_DESC(timeout
, "Timeout value in seconds.");
302 module_param(pretimeout
, timeout
, 0644);
303 MODULE_PARM_DESC(pretimeout
, "Pretimeout value in seconds.");
305 module_param_cb(action
, ¶m_ops_str
, action_op
, 0644);
306 MODULE_PARM_DESC(action
, "Timeout action. One of: "
307 "reset, none, power_cycle, power_off.");
309 module_param_cb(preaction
, ¶m_ops_str
, preaction_op
, 0644);
310 MODULE_PARM_DESC(preaction
, "Pretimeout action. One of: "
311 "pre_none, pre_smi, pre_nmi, pre_int.");
313 module_param_cb(preop
, ¶m_ops_str
, preop_op
, 0644);
314 MODULE_PARM_DESC(preop
, "Pretimeout driver operation. One of: "
315 "preop_none, preop_panic, preop_give_data.");
317 module_param(start_now
, int, 0444);
318 MODULE_PARM_DESC(start_now
, "Set to 1 to start the watchdog as"
319 "soon as the driver is loaded.");
321 module_param(nowayout
, int, 0644);
322 MODULE_PARM_DESC(nowayout
, "Watchdog cannot be stopped once started "
323 "(default=CONFIG_WATCHDOG_NOWAYOUT)");
325 /* Default state of the timer. */
326 static unsigned char ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
328 /* If shutting down via IPMI, we ignore the heartbeat. */
329 static int ipmi_ignore_heartbeat
;
331 /* Is someone using the watchdog? Only one user is allowed. */
332 static unsigned long ipmi_wdog_open
;
335 * If set to 1, the heartbeat command will set the state to reset and
336 * start the timer. The timer doesn't normally run when the driver is
337 * first opened until the heartbeat is set the first time, this
338 * variable is used to accomplish this.
340 static int ipmi_start_timer_on_heartbeat
;
342 /* IPMI version of the BMC. */
343 static unsigned char ipmi_version_major
;
344 static unsigned char ipmi_version_minor
;
346 /* If a pretimeout occurs, this is used to allow only one panic to happen. */
347 static atomic_t preop_panic_excl
= ATOMIC_INIT(-1);
350 static int testing_nmi
;
351 static int nmi_handler_registered
;
354 static int ipmi_heartbeat(void);
357 * We use a mutex to make sure that only one thing can send a set
358 * timeout at one time, because we only have one copy of the data.
359 * The mutex is claimed when the set_timeout is sent and freed
360 * when both messages are free.
362 static atomic_t set_timeout_tofree
= ATOMIC_INIT(0);
363 static DEFINE_MUTEX(set_timeout_lock
);
364 static DECLARE_COMPLETION(set_timeout_wait
);
365 static void set_timeout_free_smi(struct ipmi_smi_msg
*msg
)
367 if (atomic_dec_and_test(&set_timeout_tofree
))
368 complete(&set_timeout_wait
);
370 static void set_timeout_free_recv(struct ipmi_recv_msg
*msg
)
372 if (atomic_dec_and_test(&set_timeout_tofree
))
373 complete(&set_timeout_wait
);
375 static struct ipmi_smi_msg set_timeout_smi_msg
= {
376 .done
= set_timeout_free_smi
378 static struct ipmi_recv_msg set_timeout_recv_msg
= {
379 .done
= set_timeout_free_recv
382 static int i_ipmi_set_timeout(struct ipmi_smi_msg
*smi_msg
,
383 struct ipmi_recv_msg
*recv_msg
,
384 int *send_heartbeat_now
)
386 struct kernel_ipmi_msg msg
;
387 unsigned char data
[6];
389 struct ipmi_system_interface_addr addr
;
393 /* These can be cleared as we are setting the timeout. */
394 pretimeout_since_last_heartbeat
= 0;
397 WDOG_SET_TIMER_USE(data
[0], WDOG_TIMER_USE_SMS_OS
);
399 if ((ipmi_version_major
> 1)
400 || ((ipmi_version_major
== 1) && (ipmi_version_minor
>= 5))) {
401 /* This is an IPMI 1.5-only feature. */
402 data
[0] |= WDOG_DONT_STOP_ON_SET
;
403 } else if (ipmi_watchdog_state
!= WDOG_TIMEOUT_NONE
) {
405 * In ipmi 1.0, setting the timer stops the watchdog, we
406 * need to start it back up again.
412 WDOG_SET_TIMEOUT_ACT(data
[1], ipmi_watchdog_state
);
413 if ((pretimeout
> 0) && (ipmi_watchdog_state
!= WDOG_TIMEOUT_NONE
)) {
414 WDOG_SET_PRETIMEOUT_ACT(data
[1], preaction_val
);
415 data
[2] = pretimeout
;
417 WDOG_SET_PRETIMEOUT_ACT(data
[1], WDOG_PRETIMEOUT_NONE
);
418 data
[2] = 0; /* No pretimeout. */
421 WDOG_SET_TIMEOUT(data
[4], data
[5], timeout
);
423 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
424 addr
.channel
= IPMI_BMC_CHANNEL
;
428 msg
.cmd
= IPMI_WDOG_SET_TIMER
;
430 msg
.data_len
= sizeof(data
);
431 rv
= ipmi_request_supply_msgs(watchdog_user
,
432 (struct ipmi_addr
*) &addr
,
440 printk(KERN_WARNING PFX
"set timeout error: %d\n",
444 if (send_heartbeat_now
)
445 *send_heartbeat_now
= hbnow
;
450 static int ipmi_set_timeout(int do_heartbeat
)
452 int send_heartbeat_now
;
456 /* We can only send one of these at a time. */
457 mutex_lock(&set_timeout_lock
);
459 atomic_set(&set_timeout_tofree
, 2);
461 rv
= i_ipmi_set_timeout(&set_timeout_smi_msg
,
462 &set_timeout_recv_msg
,
463 &send_heartbeat_now
);
465 mutex_unlock(&set_timeout_lock
);
469 wait_for_completion(&set_timeout_wait
);
471 mutex_unlock(&set_timeout_lock
);
473 if ((do_heartbeat
== IPMI_SET_TIMEOUT_FORCE_HB
)
474 || ((send_heartbeat_now
)
475 && (do_heartbeat
== IPMI_SET_TIMEOUT_HB_IF_NECESSARY
)))
476 rv
= ipmi_heartbeat();
482 static atomic_t panic_done_count
= ATOMIC_INIT(0);
484 static void panic_smi_free(struct ipmi_smi_msg
*msg
)
486 atomic_dec(&panic_done_count
);
488 static void panic_recv_free(struct ipmi_recv_msg
*msg
)
490 atomic_dec(&panic_done_count
);
493 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg
= {
494 .done
= panic_smi_free
496 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg
= {
497 .done
= panic_recv_free
500 static void panic_halt_ipmi_heartbeat(void)
502 struct kernel_ipmi_msg msg
;
503 struct ipmi_system_interface_addr addr
;
507 * Don't reset the timer if we have the timer turned off, that
508 * re-enables the watchdog.
510 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
)
513 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
514 addr
.channel
= IPMI_BMC_CHANNEL
;
518 msg
.cmd
= IPMI_WDOG_RESET_TIMER
;
521 rv
= ipmi_request_supply_msgs(watchdog_user
,
522 (struct ipmi_addr
*) &addr
,
526 &panic_halt_heartbeat_smi_msg
,
527 &panic_halt_heartbeat_recv_msg
,
530 atomic_add(2, &panic_done_count
);
533 static struct ipmi_smi_msg panic_halt_smi_msg
= {
534 .done
= panic_smi_free
536 static struct ipmi_recv_msg panic_halt_recv_msg
= {
537 .done
= panic_recv_free
541 * Special call, doesn't claim any locks. This is only to be called
542 * at panic or halt time, in run-to-completion mode, when the caller
543 * is the only CPU and the only thing that will be going is these IPMI
546 static void panic_halt_ipmi_set_timeout(void)
548 int send_heartbeat_now
;
551 /* Wait for the messages to be free. */
552 while (atomic_read(&panic_done_count
) != 0)
553 ipmi_poll_interface(watchdog_user
);
554 rv
= i_ipmi_set_timeout(&panic_halt_smi_msg
,
555 &panic_halt_recv_msg
,
556 &send_heartbeat_now
);
558 atomic_add(2, &panic_done_count
);
559 if (send_heartbeat_now
)
560 panic_halt_ipmi_heartbeat();
562 printk(KERN_WARNING PFX
563 "Unable to extend the watchdog timeout.");
564 while (atomic_read(&panic_done_count
) != 0)
565 ipmi_poll_interface(watchdog_user
);
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.
574 static atomic_t heartbeat_tofree
= ATOMIC_INIT(0);
575 static DEFINE_MUTEX(heartbeat_lock
);
576 static DECLARE_COMPLETION(heartbeat_wait
);
577 static void heartbeat_free_smi(struct ipmi_smi_msg
*msg
)
579 if (atomic_dec_and_test(&heartbeat_tofree
))
580 complete(&heartbeat_wait
);
582 static void heartbeat_free_recv(struct ipmi_recv_msg
*msg
)
584 if (atomic_dec_and_test(&heartbeat_tofree
))
585 complete(&heartbeat_wait
);
587 static struct ipmi_smi_msg heartbeat_smi_msg
= {
588 .done
= heartbeat_free_smi
590 static struct ipmi_recv_msg heartbeat_recv_msg
= {
591 .done
= heartbeat_free_recv
594 static int ipmi_heartbeat(void)
596 struct kernel_ipmi_msg msg
;
598 struct ipmi_system_interface_addr addr
;
600 if (ipmi_ignore_heartbeat
)
603 if (ipmi_start_timer_on_heartbeat
) {
604 ipmi_start_timer_on_heartbeat
= 0;
605 ipmi_watchdog_state
= action_val
;
606 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
607 } else if (pretimeout_since_last_heartbeat
) {
609 * A pretimeout occurred, make sure we set the timeout.
610 * We don't want to set the action, though, we want to
611 * leave that alone (thus it can't be combined with the
614 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
617 mutex_lock(&heartbeat_lock
);
619 atomic_set(&heartbeat_tofree
, 2);
622 * Don't reset the timer if we have the timer turned off, that
623 * re-enables the watchdog.
625 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
) {
626 mutex_unlock(&heartbeat_lock
);
630 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
631 addr
.channel
= IPMI_BMC_CHANNEL
;
635 msg
.cmd
= IPMI_WDOG_RESET_TIMER
;
638 rv
= ipmi_request_supply_msgs(watchdog_user
,
639 (struct ipmi_addr
*) &addr
,
647 mutex_unlock(&heartbeat_lock
);
648 printk(KERN_WARNING PFX
"heartbeat failure: %d\n",
653 /* Wait for the heartbeat to be sent. */
654 wait_for_completion(&heartbeat_wait
);
656 if (heartbeat_recv_msg
.msg
.data
[0] != 0) {
658 * Got an error in the heartbeat response. It was already
659 * reported in ipmi_wdog_msg_handler, but we should return
665 mutex_unlock(&heartbeat_lock
);
670 static struct watchdog_info ident
= {
671 .options
= 0, /* WDIOF_SETTIMEOUT, */
672 .firmware_version
= 1,
676 static int ipmi_ioctl(struct file
*file
,
677 unsigned int cmd
, unsigned long arg
)
679 void __user
*argp
= (void __user
*)arg
;
684 case WDIOC_GETSUPPORT
:
685 i
= copy_to_user(argp
, &ident
, sizeof(ident
));
686 return i
? -EFAULT
: 0;
688 case WDIOC_SETTIMEOUT
:
689 i
= copy_from_user(&val
, argp
, sizeof(int));
693 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
695 case WDIOC_GETTIMEOUT
:
696 i
= copy_to_user(argp
, &timeout
, sizeof(timeout
));
701 case WDIOC_SET_PRETIMEOUT
:
702 case WDIOC_SETPRETIMEOUT
:
703 i
= copy_from_user(&val
, argp
, sizeof(int));
707 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
709 case WDIOC_GET_PRETIMEOUT
:
710 case WDIOC_GETPRETIMEOUT
:
711 i
= copy_to_user(argp
, &pretimeout
, sizeof(pretimeout
));
716 case WDIOC_KEEPALIVE
:
717 return ipmi_heartbeat();
719 case WDIOC_SETOPTIONS
:
720 i
= copy_from_user(&val
, argp
, sizeof(int));
723 if (val
& WDIOS_DISABLECARD
) {
724 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
725 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB
);
726 ipmi_start_timer_on_heartbeat
= 0;
729 if (val
& WDIOS_ENABLECARD
) {
730 ipmi_watchdog_state
= action_val
;
731 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
735 case WDIOC_GETSTATUS
:
737 i
= copy_to_user(argp
, &val
, sizeof(val
));
747 static long ipmi_unlocked_ioctl(struct file
*file
,
753 mutex_lock(&ipmi_watchdog_mutex
);
754 ret
= ipmi_ioctl(file
, cmd
, arg
);
755 mutex_unlock(&ipmi_watchdog_mutex
);
760 static ssize_t
ipmi_write(struct file
*file
,
761 const char __user
*buf
,
771 /* In case it was set long ago */
774 for (i
= 0; i
!= len
; i
++) {
777 if (get_user(c
, buf
+ i
))
783 rv
= ipmi_heartbeat();
790 static ssize_t
ipmi_read(struct file
*file
,
802 * Reading returns if the pretimeout has gone off, and it only does
803 * it once per pretimeout.
805 spin_lock(&ipmi_read_lock
);
807 if (file
->f_flags
& O_NONBLOCK
) {
812 init_waitqueue_entry(&wait
, current
);
813 add_wait_queue(&read_q
, &wait
);
814 while (!data_to_read
) {
815 set_current_state(TASK_INTERRUPTIBLE
);
816 spin_unlock(&ipmi_read_lock
);
818 spin_lock(&ipmi_read_lock
);
820 remove_wait_queue(&read_q
, &wait
);
822 if (signal_pending(current
)) {
830 spin_unlock(&ipmi_read_lock
);
833 if (copy_to_user(buf
, &data_to_read
, 1))
842 static int ipmi_open(struct inode
*ino
, struct file
*filep
)
844 switch (iminor(ino
)) {
846 if (test_and_set_bit(0, &ipmi_wdog_open
))
851 * Don't start the timer now, let it start on the
854 ipmi_start_timer_on_heartbeat
= 1;
855 return nonseekable_open(ino
, filep
);
862 static unsigned int ipmi_poll(struct file
*file
, poll_table
*wait
)
864 unsigned int mask
= 0;
866 poll_wait(file
, &read_q
, wait
);
868 spin_lock(&ipmi_read_lock
);
870 mask
|= (POLLIN
| POLLRDNORM
);
871 spin_unlock(&ipmi_read_lock
);
876 static int ipmi_fasync(int fd
, struct file
*file
, int on
)
880 result
= fasync_helper(fd
, file
, on
, &fasync_q
);
885 static int ipmi_close(struct inode
*ino
, struct file
*filep
)
887 if (iminor(ino
) == WATCHDOG_MINOR
) {
888 if (expect_close
== 42) {
889 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
890 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB
);
893 "Unexpected close, not stopping watchdog!\n");
896 clear_bit(0, &ipmi_wdog_open
);
904 static const struct file_operations ipmi_wdog_fops
= {
905 .owner
= THIS_MODULE
,
909 .unlocked_ioctl
= ipmi_unlocked_ioctl
,
911 .release
= ipmi_close
,
912 .fasync
= ipmi_fasync
,
916 static struct miscdevice ipmi_wdog_miscdev
= {
917 .minor
= WATCHDOG_MINOR
,
919 .fops
= &ipmi_wdog_fops
922 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg
*msg
,
925 if (msg
->msg
.data
[0] != 0) {
926 printk(KERN_ERR PFX
"response: Error %x on cmd %x\n",
931 ipmi_free_recv_msg(msg
);
934 static void ipmi_wdog_pretimeout_handler(void *handler_data
)
936 if (preaction_val
!= WDOG_PRETIMEOUT_NONE
) {
937 if (preop_val
== WDOG_PREOP_PANIC
) {
938 if (atomic_inc_and_test(&preop_panic_excl
))
939 panic("Watchdog pre-timeout");
940 } else if (preop_val
== WDOG_PREOP_GIVE_DATA
) {
941 spin_lock(&ipmi_read_lock
);
943 wake_up_interruptible(&read_q
);
944 kill_fasync(&fasync_q
, SIGIO
, POLL_IN
);
946 spin_unlock(&ipmi_read_lock
);
951 * On some machines, the heartbeat will give an error and not
952 * work unless we re-enable the timer. So do so.
954 pretimeout_since_last_heartbeat
= 1;
957 static struct ipmi_user_hndl ipmi_hndlrs
= {
958 .ipmi_recv_hndl
= ipmi_wdog_msg_handler
,
959 .ipmi_watchdog_pretimeout
= ipmi_wdog_pretimeout_handler
962 static void ipmi_register_watchdog(int ipmi_intf
)
969 if ((ifnum_to_use
>= 0) && (ifnum_to_use
!= ipmi_intf
))
972 watchdog_ifnum
= ipmi_intf
;
974 rv
= ipmi_create_user(ipmi_intf
, &ipmi_hndlrs
, NULL
, &watchdog_user
);
976 printk(KERN_CRIT PFX
"Unable to register with ipmi\n");
980 ipmi_get_version(watchdog_user
,
982 &ipmi_version_minor
);
984 rv
= misc_register(&ipmi_wdog_miscdev
);
986 ipmi_destroy_user(watchdog_user
);
987 watchdog_user
= NULL
;
988 printk(KERN_CRIT PFX
"Unable to register misc device\n");
992 if (nmi_handler_registered
) {
993 int old_pretimeout
= pretimeout
;
994 int old_timeout
= timeout
;
995 int old_preop_val
= preop_val
;
998 * Set the pretimeout to go off in a second and give
999 * ourselves plenty of time to stop the timer.
1001 ipmi_watchdog_state
= WDOG_TIMEOUT_RESET
;
1002 preop_val
= WDOG_PREOP_NONE
; /* Make sure nothing happens */
1008 rv
= ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
1010 printk(KERN_WARNING PFX
"Error starting timer to"
1011 " test NMI: 0x%x. The NMI pretimeout will"
1012 " likely not work\n", rv
);
1019 if (testing_nmi
!= 2) {
1020 printk(KERN_WARNING PFX
"IPMI NMI didn't seem to"
1021 " occur. The NMI pretimeout will"
1022 " likely not work\n");
1026 preop_val
= old_preop_val
;
1027 pretimeout
= old_pretimeout
;
1028 timeout
= old_timeout
;
1033 if ((start_now
) && (rv
== 0)) {
1034 /* Run from startup, so start the timer now. */
1035 start_now
= 0; /* Disable this function after first startup. */
1036 ipmi_watchdog_state
= action_val
;
1037 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
1038 printk(KERN_INFO PFX
"Starting now!\n");
1040 /* Stop the timer now. */
1041 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
1042 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB
);
1046 static void ipmi_unregister_watchdog(int ipmi_intf
)
1053 if (watchdog_ifnum
!= ipmi_intf
)
1056 /* Make sure no one can call us any more. */
1057 misc_deregister(&ipmi_wdog_miscdev
);
1060 * Wait to make sure the message makes it out. The lower layer has
1061 * pointers to our buffers, we want to make sure they are done before
1062 * we release our memory.
1064 while (atomic_read(&set_timeout_tofree
))
1065 schedule_timeout_uninterruptible(1);
1067 /* Disconnect from IPMI. */
1068 rv
= ipmi_destroy_user(watchdog_user
);
1070 printk(KERN_WARNING PFX
"error unlinking from IPMI: %d\n",
1073 watchdog_user
= NULL
;
1081 ipmi_nmi(unsigned int val
, struct pt_regs
*regs
)
1084 * If we get here, it's an NMI that's not a memory or I/O
1085 * error. We can't truly tell if it's from IPMI or not
1086 * without sending a message, and sending a message is almost
1087 * impossible because of locking.
1095 /* If we are not expecting a timeout, ignore it. */
1096 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
)
1099 if (preaction_val
!= WDOG_PRETIMEOUT_NMI
)
1103 * If no one else handled the NMI, we assume it was the IPMI
1106 if (preop_val
== WDOG_PREOP_PANIC
) {
1107 /* On some machines, the heartbeat will give
1108 an error and not work unless we re-enable
1109 the timer. So do so. */
1110 pretimeout_since_last_heartbeat
= 1;
1111 if (atomic_inc_and_test(&preop_panic_excl
))
1112 panic(PFX
"pre-timeout");
1119 static int wdog_reboot_handler(struct notifier_block
*this,
1123 static int reboot_event_handled
;
1125 if ((watchdog_user
) && (!reboot_event_handled
)) {
1126 /* Make sure we only do this once. */
1127 reboot_event_handled
= 1;
1129 if (code
== SYS_POWER_OFF
|| code
== SYS_HALT
) {
1130 /* Disable the WDT if we are shutting down. */
1131 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
1132 panic_halt_ipmi_set_timeout();
1133 } else if (ipmi_watchdog_state
!= WDOG_TIMEOUT_NONE
) {
1134 /* Set a long timer to let the reboot happens, but
1135 reboot if it hangs, but only if the watchdog
1136 timer was already running. */
1139 ipmi_watchdog_state
= WDOG_TIMEOUT_RESET
;
1140 panic_halt_ipmi_set_timeout();
1146 static struct notifier_block wdog_reboot_notifier
= {
1147 .notifier_call
= wdog_reboot_handler
,
1152 static int wdog_panic_handler(struct notifier_block
*this,
1153 unsigned long event
,
1156 static int panic_event_handled
;
1158 /* On a panic, if we have a panic timeout, make sure to extend
1159 the watchdog timer to a reasonable value to complete the
1160 panic, if the watchdog timer is running. Plus the
1161 pretimeout is meaningless at panic time. */
1162 if (watchdog_user
&& !panic_event_handled
&&
1163 ipmi_watchdog_state
!= WDOG_TIMEOUT_NONE
) {
1164 /* Make sure we do this only once. */
1165 panic_event_handled
= 1;
1169 panic_halt_ipmi_set_timeout();
1175 static struct notifier_block wdog_panic_notifier
= {
1176 .notifier_call
= wdog_panic_handler
,
1178 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
1182 static void ipmi_new_smi(int if_num
, struct device
*device
)
1184 ipmi_register_watchdog(if_num
);
1187 static void ipmi_smi_gone(int if_num
)
1189 ipmi_unregister_watchdog(if_num
);
1192 static struct ipmi_smi_watcher smi_watcher
= {
1193 .owner
= THIS_MODULE
,
1194 .new_smi
= ipmi_new_smi
,
1195 .smi_gone
= ipmi_smi_gone
1198 static int action_op(const char *inval
, char *outval
)
1201 strcpy(outval
, action
);
1206 if (strcmp(inval
, "reset") == 0)
1207 action_val
= WDOG_TIMEOUT_RESET
;
1208 else if (strcmp(inval
, "none") == 0)
1209 action_val
= WDOG_TIMEOUT_NONE
;
1210 else if (strcmp(inval
, "power_cycle") == 0)
1211 action_val
= WDOG_TIMEOUT_POWER_CYCLE
;
1212 else if (strcmp(inval
, "power_off") == 0)
1213 action_val
= WDOG_TIMEOUT_POWER_DOWN
;
1216 strcpy(action
, inval
);
1220 static int preaction_op(const char *inval
, char *outval
)
1223 strcpy(outval
, preaction
);
1228 if (strcmp(inval
, "pre_none") == 0)
1229 preaction_val
= WDOG_PRETIMEOUT_NONE
;
1230 else if (strcmp(inval
, "pre_smi") == 0)
1231 preaction_val
= WDOG_PRETIMEOUT_SMI
;
1233 else if (strcmp(inval
, "pre_nmi") == 0)
1234 preaction_val
= WDOG_PRETIMEOUT_NMI
;
1236 else if (strcmp(inval
, "pre_int") == 0)
1237 preaction_val
= WDOG_PRETIMEOUT_MSG_INT
;
1240 strcpy(preaction
, inval
);
1244 static int preop_op(const char *inval
, char *outval
)
1247 strcpy(outval
, preop
);
1252 if (strcmp(inval
, "preop_none") == 0)
1253 preop_val
= WDOG_PREOP_NONE
;
1254 else if (strcmp(inval
, "preop_panic") == 0)
1255 preop_val
= WDOG_PREOP_PANIC
;
1256 else if (strcmp(inval
, "preop_give_data") == 0)
1257 preop_val
= WDOG_PREOP_GIVE_DATA
;
1260 strcpy(preop
, inval
);
1264 static void check_parms(void)
1270 if (preaction_val
== WDOG_PRETIMEOUT_NMI
) {
1272 if (preop_val
== WDOG_PREOP_GIVE_DATA
) {
1273 printk(KERN_WARNING PFX
"Pretimeout op is to give data"
1274 " but NMI pretimeout is enabled, setting"
1275 " pretimeout op to none\n");
1276 preop_op("preop_none", NULL
);
1280 if (do_nmi
&& !nmi_handler_registered
) {
1281 rv
= register_nmi_handler(NMI_UNKNOWN
, ipmi_nmi
, 0,
1284 printk(KERN_WARNING PFX
1285 "Can't register nmi handler\n");
1288 nmi_handler_registered
= 1;
1289 } else if (!do_nmi
&& nmi_handler_registered
) {
1290 unregister_nmi_handler(NMI_UNKNOWN
, "ipmi");
1291 nmi_handler_registered
= 0;
1296 static int __init
ipmi_wdog_init(void)
1300 if (action_op(action
, NULL
)) {
1301 action_op("reset", NULL
);
1302 printk(KERN_INFO PFX
"Unknown action '%s', defaulting to"
1303 " reset\n", action
);
1306 if (preaction_op(preaction
, NULL
)) {
1307 preaction_op("pre_none", NULL
);
1308 printk(KERN_INFO PFX
"Unknown preaction '%s', defaulting to"
1309 " none\n", preaction
);
1312 if (preop_op(preop
, NULL
)) {
1313 preop_op("preop_none", NULL
);
1314 printk(KERN_INFO PFX
"Unknown preop '%s', defaulting to"
1320 register_reboot_notifier(&wdog_reboot_notifier
);
1321 atomic_notifier_chain_register(&panic_notifier_list
,
1322 &wdog_panic_notifier
);
1324 rv
= ipmi_smi_watcher_register(&smi_watcher
);
1327 if (nmi_handler_registered
)
1328 unregister_nmi_handler(NMI_UNKNOWN
, "ipmi");
1330 atomic_notifier_chain_unregister(&panic_notifier_list
,
1331 &wdog_panic_notifier
);
1332 unregister_reboot_notifier(&wdog_reboot_notifier
);
1333 printk(KERN_WARNING PFX
"can't register smi watcher\n");
1337 printk(KERN_INFO PFX
"driver initialized\n");
1342 static void __exit
ipmi_wdog_exit(void)
1344 ipmi_smi_watcher_unregister(&smi_watcher
);
1345 ipmi_unregister_watchdog(watchdog_ifnum
);
1348 if (nmi_handler_registered
)
1349 unregister_nmi_handler(NMI_UNKNOWN
, "ipmi");
1352 atomic_notifier_chain_unregister(&panic_notifier_list
,
1353 &wdog_panic_notifier
);
1354 unregister_reboot_notifier(&wdog_reboot_notifier
);
1356 module_exit(ipmi_wdog_exit
);
1357 module_init(ipmi_wdog_init
);
1358 MODULE_LICENSE("GPL");
1359 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
1360 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface.");