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/config.h>
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/ipmi.h>
38 #include <linux/ipmi_smi.h>
39 #include <linux/watchdog.h>
40 #include <linux/miscdevice.h>
41 #include <linux/init.h>
42 #include <linux/rwsem.h>
43 #include <linux/errno.h>
44 #include <asm/uaccess.h>
45 #include <linux/notifier.h>
46 #include <linux/nmi.h>
47 #include <linux/reboot.h>
48 #include <linux/wait.h>
49 #include <linux/poll.h>
50 #ifdef CONFIG_X86_LOCAL_APIC
54 #define PFX "IPMI Watchdog: "
56 #define IPMI_WATCHDOG_VERSION "v33"
59 * The IPMI command/response information for the watchdog timer.
62 /* values for byte 1 of the set command, byte 2 of the get response. */
63 #define WDOG_DONT_LOG (1 << 7)
64 #define WDOG_DONT_STOP_ON_SET (1 << 6)
65 #define WDOG_SET_TIMER_USE(byte, use) \
66 byte = ((byte) & 0xf8) | ((use) & 0x7)
67 #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7)
68 #define WDOG_TIMER_USE_BIOS_FRB2 1
69 #define WDOG_TIMER_USE_BIOS_POST 2
70 #define WDOG_TIMER_USE_OS_LOAD 3
71 #define WDOG_TIMER_USE_SMS_OS 4
72 #define WDOG_TIMER_USE_OEM 5
74 /* values for byte 2 of the set command, byte 3 of the get response. */
75 #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \
76 byte = ((byte) & 0x8f) | (((use) & 0x7) << 4)
77 #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7)
78 #define WDOG_PRETIMEOUT_NONE 0
79 #define WDOG_PRETIMEOUT_SMI 1
80 #define WDOG_PRETIMEOUT_NMI 2
81 #define WDOG_PRETIMEOUT_MSG_INT 3
83 /* Operations that can be performed on a pretimout. */
84 #define WDOG_PREOP_NONE 0
85 #define WDOG_PREOP_PANIC 1
86 #define WDOG_PREOP_GIVE_DATA 2 /* Cause data to be available to
87 read. Doesn't work in NMI
90 /* Actions to perform on a full timeout. */
91 #define WDOG_SET_TIMEOUT_ACT(byte, use) \
92 byte = ((byte) & 0xf8) | ((use) & 0x7)
93 #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7)
94 #define WDOG_TIMEOUT_NONE 0
95 #define WDOG_TIMEOUT_RESET 1
96 #define WDOG_TIMEOUT_POWER_DOWN 2
97 #define WDOG_TIMEOUT_POWER_CYCLE 3
99 /* Byte 3 of the get command, byte 4 of the get response is the
100 pre-timeout in seconds. */
102 /* Bits for setting byte 4 of the set command, byte 5 of the get response. */
103 #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1)
104 #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2)
105 #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3)
106 #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4)
107 #define WDOG_EXPIRE_CLEAR_OEM (1 << 5)
109 /* Setting/getting the watchdog timer value. This is for bytes 5 and
110 6 (the timeout time) of the set command, and bytes 6 and 7 (the
111 timeout time) and 8 and 9 (the current countdown value) of the
112 response. The timeout value is given in seconds (in the command it
113 is 100ms intervals). */
114 #define WDOG_SET_TIMEOUT(byte1, byte2, val) \
115 (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8)
116 #define WDOG_GET_TIMEOUT(byte1, byte2) \
117 (((byte1) | ((byte2) << 8)) / 10)
119 #define IPMI_WDOG_RESET_TIMER 0x22
120 #define IPMI_WDOG_SET_TIMER 0x24
121 #define IPMI_WDOG_GET_TIMER 0x25
123 /* These are here until the real ones get into the watchdog.h interface. */
124 #ifndef WDIOC_GETTIMEOUT
125 #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int)
127 #ifndef WDIOC_SET_PRETIMEOUT
128 #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int)
130 #ifndef WDIOC_GET_PRETIMEOUT
131 #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int)
134 #ifdef CONFIG_WATCHDOG_NOWAYOUT
135 static int nowayout
= 1;
140 static ipmi_user_t watchdog_user
= NULL
;
142 /* Default the timeout to 10 seconds. */
143 static int timeout
= 10;
145 /* The pre-timeout is disabled by default. */
146 static int pretimeout
= 0;
148 /* Default action is to reset the board on a timeout. */
149 static unsigned char action_val
= WDOG_TIMEOUT_RESET
;
151 static char action
[16] = "reset";
153 static unsigned char preaction_val
= WDOG_PRETIMEOUT_NONE
;
155 static char preaction
[16] = "pre_none";
157 static unsigned char preop_val
= WDOG_PREOP_NONE
;
159 static char preop
[16] = "preop_none";
160 static DEFINE_SPINLOCK(ipmi_read_lock
);
161 static char data_to_read
= 0;
162 static DECLARE_WAIT_QUEUE_HEAD(read_q
);
163 static struct fasync_struct
*fasync_q
= NULL
;
164 static char pretimeout_since_last_heartbeat
= 0;
165 static char expect_close
;
167 /* If true, the driver will start running as soon as it is configured
169 static int start_now
= 0;
171 module_param(timeout
, int, 0);
172 MODULE_PARM_DESC(timeout
, "Timeout value in seconds.");
173 module_param(pretimeout
, int, 0);
174 MODULE_PARM_DESC(pretimeout
, "Pretimeout value in seconds.");
175 module_param_string(action
, action
, sizeof(action
), 0);
176 MODULE_PARM_DESC(action
, "Timeout action. One of: "
177 "reset, none, power_cycle, power_off.");
178 module_param_string(preaction
, preaction
, sizeof(preaction
), 0);
179 MODULE_PARM_DESC(preaction
, "Pretimeout action. One of: "
180 "pre_none, pre_smi, pre_nmi, pre_int.");
181 module_param_string(preop
, preop
, sizeof(preop
), 0);
182 MODULE_PARM_DESC(preop
, "Pretimeout driver operation. One of: "
183 "preop_none, preop_panic, preop_give_data.");
184 module_param(start_now
, int, 0);
185 MODULE_PARM_DESC(start_now
, "Set to 1 to start the watchdog as"
186 "soon as the driver is loaded.");
187 module_param(nowayout
, int, 0);
188 MODULE_PARM_DESC(nowayout
, "Watchdog cannot be stopped once started (default=CONFIG_WATCHDOG_NOWAYOUT)");
190 /* Default state of the timer. */
191 static unsigned char ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
193 /* If shutting down via IPMI, we ignore the heartbeat. */
194 static int ipmi_ignore_heartbeat
= 0;
196 /* Is someone using the watchdog? Only one user is allowed. */
197 static unsigned long ipmi_wdog_open
= 0;
199 /* If set to 1, the heartbeat command will set the state to reset and
200 start the timer. The timer doesn't normally run when the driver is
201 first opened until the heartbeat is set the first time, this
202 variable is used to accomplish this. */
203 static int ipmi_start_timer_on_heartbeat
= 0;
205 /* IPMI version of the BMC. */
206 static unsigned char ipmi_version_major
;
207 static unsigned char ipmi_version_minor
;
210 static int ipmi_heartbeat(void);
211 static void panic_halt_ipmi_heartbeat(void);
214 /* We use a semaphore to make sure that only one thing can send a set
215 timeout at one time, because we only have one copy of the data.
216 The semaphore is claimed when the set_timeout is sent and freed
217 when both messages are free. */
218 static atomic_t set_timeout_tofree
= ATOMIC_INIT(0);
219 static DECLARE_MUTEX(set_timeout_lock
);
220 static void set_timeout_free_smi(struct ipmi_smi_msg
*msg
)
222 if (atomic_dec_and_test(&set_timeout_tofree
))
223 up(&set_timeout_lock
);
225 static void set_timeout_free_recv(struct ipmi_recv_msg
*msg
)
227 if (atomic_dec_and_test(&set_timeout_tofree
))
228 up(&set_timeout_lock
);
230 static struct ipmi_smi_msg set_timeout_smi_msg
=
232 .done
= set_timeout_free_smi
234 static struct ipmi_recv_msg set_timeout_recv_msg
=
236 .done
= set_timeout_free_recv
239 static int i_ipmi_set_timeout(struct ipmi_smi_msg
*smi_msg
,
240 struct ipmi_recv_msg
*recv_msg
,
241 int *send_heartbeat_now
)
243 struct kernel_ipmi_msg msg
;
244 unsigned char data
[6];
246 struct ipmi_system_interface_addr addr
;
251 WDOG_SET_TIMER_USE(data
[0], WDOG_TIMER_USE_SMS_OS
);
253 if ((ipmi_version_major
> 1)
254 || ((ipmi_version_major
== 1) && (ipmi_version_minor
>= 5)))
256 /* This is an IPMI 1.5-only feature. */
257 data
[0] |= WDOG_DONT_STOP_ON_SET
;
258 } else if (ipmi_watchdog_state
!= WDOG_TIMEOUT_NONE
) {
259 /* In ipmi 1.0, setting the timer stops the watchdog, we
260 need to start it back up again. */
265 WDOG_SET_TIMEOUT_ACT(data
[1], ipmi_watchdog_state
);
266 if (pretimeout
> 0) {
267 WDOG_SET_PRETIMEOUT_ACT(data
[1], preaction_val
);
268 data
[2] = pretimeout
;
270 WDOG_SET_PRETIMEOUT_ACT(data
[1], WDOG_PRETIMEOUT_NONE
);
271 data
[2] = 0; /* No pretimeout. */
274 WDOG_SET_TIMEOUT(data
[4], data
[5], timeout
);
276 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
277 addr
.channel
= IPMI_BMC_CHANNEL
;
281 msg
.cmd
= IPMI_WDOG_SET_TIMER
;
283 msg
.data_len
= sizeof(data
);
284 rv
= ipmi_request_supply_msgs(watchdog_user
,
285 (struct ipmi_addr
*) &addr
,
293 printk(KERN_WARNING PFX
"set timeout error: %d\n",
297 if (send_heartbeat_now
)
298 *send_heartbeat_now
= hbnow
;
303 /* Parameters to ipmi_set_timeout */
304 #define IPMI_SET_TIMEOUT_NO_HB 0
305 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
306 #define IPMI_SET_TIMEOUT_FORCE_HB 2
308 static int ipmi_set_timeout(int do_heartbeat
)
310 int send_heartbeat_now
;
314 /* We can only send one of these at a time. */
315 down(&set_timeout_lock
);
317 atomic_set(&set_timeout_tofree
, 2);
319 rv
= i_ipmi_set_timeout(&set_timeout_smi_msg
,
320 &set_timeout_recv_msg
,
321 &send_heartbeat_now
);
323 up(&set_timeout_lock
);
325 if ((do_heartbeat
== IPMI_SET_TIMEOUT_FORCE_HB
)
326 || ((send_heartbeat_now
)
327 && (do_heartbeat
== IPMI_SET_TIMEOUT_HB_IF_NECESSARY
)))
329 rv
= ipmi_heartbeat();
336 static void dummy_smi_free(struct ipmi_smi_msg
*msg
)
339 static void dummy_recv_free(struct ipmi_recv_msg
*msg
)
342 static struct ipmi_smi_msg panic_halt_smi_msg
=
344 .done
= dummy_smi_free
346 static struct ipmi_recv_msg panic_halt_recv_msg
=
348 .done
= dummy_recv_free
351 /* Special call, doesn't claim any locks. This is only to be called
352 at panic or halt time, in run-to-completion mode, when the caller
353 is the only CPU and the only thing that will be going is these IPMI
355 static void panic_halt_ipmi_set_timeout(void)
357 int send_heartbeat_now
;
360 rv
= i_ipmi_set_timeout(&panic_halt_smi_msg
,
361 &panic_halt_recv_msg
,
362 &send_heartbeat_now
);
364 if (send_heartbeat_now
)
365 panic_halt_ipmi_heartbeat();
369 /* We use a semaphore to make sure that only one thing can send a
370 heartbeat at one time, because we only have one copy of the data.
371 The semaphore is claimed when the set_timeout is sent and freed
372 when both messages are free. */
373 static atomic_t heartbeat_tofree
= ATOMIC_INIT(0);
374 static DECLARE_MUTEX(heartbeat_lock
);
375 static DECLARE_MUTEX_LOCKED(heartbeat_wait_lock
);
376 static void heartbeat_free_smi(struct ipmi_smi_msg
*msg
)
378 if (atomic_dec_and_test(&heartbeat_tofree
))
379 up(&heartbeat_wait_lock
);
381 static void heartbeat_free_recv(struct ipmi_recv_msg
*msg
)
383 if (atomic_dec_and_test(&heartbeat_tofree
))
384 up(&heartbeat_wait_lock
);
386 static struct ipmi_smi_msg heartbeat_smi_msg
=
388 .done
= heartbeat_free_smi
390 static struct ipmi_recv_msg heartbeat_recv_msg
=
392 .done
= heartbeat_free_recv
395 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg
=
397 .done
= dummy_smi_free
399 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg
=
401 .done
= dummy_recv_free
404 static int ipmi_heartbeat(void)
406 struct kernel_ipmi_msg msg
;
408 struct ipmi_system_interface_addr addr
;
410 if (ipmi_ignore_heartbeat
) {
414 if (ipmi_start_timer_on_heartbeat
) {
415 ipmi_start_timer_on_heartbeat
= 0;
416 ipmi_watchdog_state
= action_val
;
417 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
418 } else if (pretimeout_since_last_heartbeat
) {
419 /* A pretimeout occurred, make sure we set the timeout.
420 We don't want to set the action, though, we want to
421 leave that alone (thus it can't be combined with the
423 pretimeout_since_last_heartbeat
= 0;
424 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
427 down(&heartbeat_lock
);
429 atomic_set(&heartbeat_tofree
, 2);
431 /* Don't reset the timer if we have the timer turned off, that
432 re-enables the watchdog. */
433 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
) {
438 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
439 addr
.channel
= IPMI_BMC_CHANNEL
;
443 msg
.cmd
= IPMI_WDOG_RESET_TIMER
;
446 rv
= ipmi_request_supply_msgs(watchdog_user
,
447 (struct ipmi_addr
*) &addr
,
456 printk(KERN_WARNING PFX
"heartbeat failure: %d\n",
461 /* Wait for the heartbeat to be sent. */
462 down(&heartbeat_wait_lock
);
464 if (heartbeat_recv_msg
.msg
.data
[0] != 0) {
465 /* Got an error in the heartbeat response. It was already
466 reported in ipmi_wdog_msg_handler, but we should return
476 static void panic_halt_ipmi_heartbeat(void)
478 struct kernel_ipmi_msg msg
;
479 struct ipmi_system_interface_addr addr
;
482 /* Don't reset the timer if we have the timer turned off, that
483 re-enables the watchdog. */
484 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
)
487 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
488 addr
.channel
= IPMI_BMC_CHANNEL
;
492 msg
.cmd
= IPMI_WDOG_RESET_TIMER
;
495 ipmi_request_supply_msgs(watchdog_user
,
496 (struct ipmi_addr
*) &addr
,
500 &panic_halt_heartbeat_smi_msg
,
501 &panic_halt_heartbeat_recv_msg
,
505 static struct watchdog_info ident
=
507 .options
= 0, /* WDIOF_SETTIMEOUT, */
508 .firmware_version
= 1,
512 static int ipmi_ioctl(struct inode
*inode
, struct file
*file
,
513 unsigned int cmd
, unsigned long arg
)
515 void __user
*argp
= (void __user
*)arg
;
520 case WDIOC_GETSUPPORT
:
521 i
= copy_to_user(argp
, &ident
, sizeof(ident
));
522 return i
? -EFAULT
: 0;
524 case WDIOC_SETTIMEOUT
:
525 i
= copy_from_user(&val
, argp
, sizeof(int));
529 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
531 case WDIOC_GETTIMEOUT
:
532 i
= copy_to_user(argp
, &timeout
, sizeof(timeout
));
537 case WDIOC_SET_PRETIMEOUT
:
538 i
= copy_from_user(&val
, argp
, sizeof(int));
542 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
544 case WDIOC_GET_PRETIMEOUT
:
545 i
= copy_to_user(argp
, &pretimeout
, sizeof(pretimeout
));
550 case WDIOC_KEEPALIVE
:
551 return ipmi_heartbeat();
553 case WDIOC_SETOPTIONS
:
554 i
= copy_from_user(&val
, argp
, sizeof(int));
557 if (val
& WDIOS_DISABLECARD
)
559 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
560 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB
);
561 ipmi_start_timer_on_heartbeat
= 0;
564 if (val
& WDIOS_ENABLECARD
)
566 ipmi_watchdog_state
= action_val
;
567 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
571 case WDIOC_GETSTATUS
:
573 i
= copy_to_user(argp
, &val
, sizeof(val
));
583 static ssize_t
ipmi_write(struct file
*file
,
584 const char __user
*buf
,
594 /* In case it was set long ago */
597 for (i
= 0; i
!= len
; i
++) {
600 if (get_user(c
, buf
+ i
))
606 rv
= ipmi_heartbeat();
614 static ssize_t
ipmi_read(struct file
*file
,
625 /* Reading returns if the pretimeout has gone off, and it only does
626 it once per pretimeout. */
627 spin_lock(&ipmi_read_lock
);
629 if (file
->f_flags
& O_NONBLOCK
) {
634 init_waitqueue_entry(&wait
, current
);
635 add_wait_queue(&read_q
, &wait
);
636 while (!data_to_read
) {
637 set_current_state(TASK_INTERRUPTIBLE
);
638 spin_unlock(&ipmi_read_lock
);
640 spin_lock(&ipmi_read_lock
);
642 remove_wait_queue(&read_q
, &wait
);
644 if (signal_pending(current
)) {
652 spin_unlock(&ipmi_read_lock
);
655 if (copy_to_user(buf
, &data_to_read
, 1))
664 static int ipmi_open(struct inode
*ino
, struct file
*filep
)
669 if(test_and_set_bit(0, &ipmi_wdog_open
))
672 /* Don't start the timer now, let it start on the
674 ipmi_start_timer_on_heartbeat
= 1;
675 return nonseekable_open(ino
, filep
);
682 static unsigned int ipmi_poll(struct file
*file
, poll_table
*wait
)
684 unsigned int mask
= 0;
686 poll_wait(file
, &read_q
, wait
);
688 spin_lock(&ipmi_read_lock
);
690 mask
|= (POLLIN
| POLLRDNORM
);
691 spin_unlock(&ipmi_read_lock
);
696 static int ipmi_fasync(int fd
, struct file
*file
, int on
)
700 result
= fasync_helper(fd
, file
, on
, &fasync_q
);
705 static int ipmi_close(struct inode
*ino
, struct file
*filep
)
707 if (iminor(ino
)==WATCHDOG_MINOR
)
709 if (expect_close
== 42) {
710 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
711 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB
);
713 printk(KERN_CRIT PFX
"Unexpected close, not stopping watchdog!\n");
716 clear_bit(0, &ipmi_wdog_open
);
719 ipmi_fasync (-1, filep
, 0);
725 static struct file_operations ipmi_wdog_fops
= {
726 .owner
= THIS_MODULE
,
732 .release
= ipmi_close
,
733 .fasync
= ipmi_fasync
,
736 static struct miscdevice ipmi_wdog_miscdev
= {
737 .minor
= WATCHDOG_MINOR
,
739 .fops
= &ipmi_wdog_fops
742 static DECLARE_RWSEM(register_sem
);
744 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg
*msg
,
747 if (msg
->msg
.data
[0] != 0) {
748 printk(KERN_ERR PFX
"response: Error %x on cmd %x\n",
753 ipmi_free_recv_msg(msg
);
756 static void ipmi_wdog_pretimeout_handler(void *handler_data
)
758 if (preaction_val
!= WDOG_PRETIMEOUT_NONE
) {
759 if (preop_val
== WDOG_PREOP_PANIC
)
760 panic("Watchdog pre-timeout");
761 else if (preop_val
== WDOG_PREOP_GIVE_DATA
) {
762 spin_lock(&ipmi_read_lock
);
764 wake_up_interruptible(&read_q
);
765 kill_fasync(&fasync_q
, SIGIO
, POLL_IN
);
767 spin_unlock(&ipmi_read_lock
);
771 /* On some machines, the heartbeat will give
772 an error and not work unless we re-enable
773 the timer. So do so. */
774 pretimeout_since_last_heartbeat
= 1;
777 static struct ipmi_user_hndl ipmi_hndlrs
=
779 .ipmi_recv_hndl
= ipmi_wdog_msg_handler
,
780 .ipmi_watchdog_pretimeout
= ipmi_wdog_pretimeout_handler
783 static void ipmi_register_watchdog(int ipmi_intf
)
787 down_write(®ister_sem
);
791 rv
= ipmi_create_user(ipmi_intf
, &ipmi_hndlrs
, NULL
, &watchdog_user
);
793 printk(KERN_CRIT PFX
"Unable to register with ipmi\n");
797 ipmi_get_version(watchdog_user
,
799 &ipmi_version_minor
);
801 rv
= misc_register(&ipmi_wdog_miscdev
);
803 ipmi_destroy_user(watchdog_user
);
804 watchdog_user
= NULL
;
805 printk(KERN_CRIT PFX
"Unable to register misc device\n");
809 up_write(®ister_sem
);
811 if ((start_now
) && (rv
== 0)) {
812 /* Run from startup, so start the timer now. */
813 start_now
= 0; /* Disable this function after first startup. */
814 ipmi_watchdog_state
= action_val
;
815 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
816 printk(KERN_INFO PFX
"Starting now!\n");
820 #ifdef HAVE_NMI_HANDLER
822 ipmi_nmi(void *dev_id
, struct pt_regs
*regs
, int cpu
, int handled
)
824 /* If no one else handled the NMI, we assume it was the IPMI
826 if ((!handled
) && (preop_val
== WDOG_PREOP_PANIC
))
827 panic(PFX
"pre-timeout");
829 /* On some machines, the heartbeat will give
830 an error and not work unless we re-enable
831 the timer. So do so. */
832 pretimeout_since_last_heartbeat
= 1;
837 static struct nmi_handler ipmi_nmi_handler
=
839 .link
= LIST_HEAD_INIT(ipmi_nmi_handler
.link
),
840 .dev_name
= "ipmi_watchdog",
843 .priority
= 0, /* Call us last. */
847 static int wdog_reboot_handler(struct notifier_block
*this,
851 static int reboot_event_handled
= 0;
853 if ((watchdog_user
) && (!reboot_event_handled
)) {
854 /* Make sure we only do this once. */
855 reboot_event_handled
= 1;
857 if (code
== SYS_DOWN
|| code
== SYS_HALT
) {
858 /* Disable the WDT if we are shutting down. */
859 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
860 panic_halt_ipmi_set_timeout();
862 /* Set a long timer to let the reboot happens, but
863 reboot if it hangs. */
866 ipmi_watchdog_state
= WDOG_TIMEOUT_RESET
;
867 panic_halt_ipmi_set_timeout();
873 static struct notifier_block wdog_reboot_notifier
= {
874 .notifier_call
= wdog_reboot_handler
,
879 static int wdog_panic_handler(struct notifier_block
*this,
883 static int panic_event_handled
= 0;
885 /* On a panic, if we have a panic timeout, make sure that the thing
886 reboots, even if it hangs during that panic. */
887 if (watchdog_user
&& !panic_event_handled
) {
888 /* Make sure the panic doesn't hang, and make sure we
889 do this only once. */
890 panic_event_handled
= 1;
894 ipmi_watchdog_state
= WDOG_TIMEOUT_RESET
;
895 panic_halt_ipmi_set_timeout();
901 static struct notifier_block wdog_panic_notifier
= {
902 .notifier_call
= wdog_panic_handler
,
904 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
908 static void ipmi_new_smi(int if_num
)
910 ipmi_register_watchdog(if_num
);
913 static void ipmi_smi_gone(int if_num
)
915 /* This can never be called, because once the watchdog is
916 registered, the interface can't go away until the watchdog
920 static struct ipmi_smi_watcher smi_watcher
=
922 .owner
= THIS_MODULE
,
923 .new_smi
= ipmi_new_smi
,
924 .smi_gone
= ipmi_smi_gone
927 static int __init
ipmi_wdog_init(void)
931 printk(KERN_INFO PFX
"driver version "
932 IPMI_WATCHDOG_VERSION
"\n");
934 if (strcmp(action
, "reset") == 0) {
935 action_val
= WDOG_TIMEOUT_RESET
;
936 } else if (strcmp(action
, "none") == 0) {
937 action_val
= WDOG_TIMEOUT_NONE
;
938 } else if (strcmp(action
, "power_cycle") == 0) {
939 action_val
= WDOG_TIMEOUT_POWER_CYCLE
;
940 } else if (strcmp(action
, "power_off") == 0) {
941 action_val
= WDOG_TIMEOUT_POWER_DOWN
;
943 action_val
= WDOG_TIMEOUT_RESET
;
944 printk(KERN_INFO PFX
"Unknown action '%s', defaulting to"
948 if (strcmp(preaction
, "pre_none") == 0) {
949 preaction_val
= WDOG_PRETIMEOUT_NONE
;
950 } else if (strcmp(preaction
, "pre_smi") == 0) {
951 preaction_val
= WDOG_PRETIMEOUT_SMI
;
952 #ifdef HAVE_NMI_HANDLER
953 } else if (strcmp(preaction
, "pre_nmi") == 0) {
954 preaction_val
= WDOG_PRETIMEOUT_NMI
;
956 } else if (strcmp(preaction
, "pre_int") == 0) {
957 preaction_val
= WDOG_PRETIMEOUT_MSG_INT
;
959 preaction_val
= WDOG_PRETIMEOUT_NONE
;
960 printk(KERN_INFO PFX
"Unknown preaction '%s', defaulting to"
961 " none\n", preaction
);
964 if (strcmp(preop
, "preop_none") == 0) {
965 preop_val
= WDOG_PREOP_NONE
;
966 } else if (strcmp(preop
, "preop_panic") == 0) {
967 preop_val
= WDOG_PREOP_PANIC
;
968 } else if (strcmp(preop
, "preop_give_data") == 0) {
969 preop_val
= WDOG_PREOP_GIVE_DATA
;
971 preop_val
= WDOG_PREOP_NONE
;
972 printk(KERN_INFO PFX
"Unknown preop '%s', defaulting to"
976 #ifdef HAVE_NMI_HANDLER
977 if (preaction_val
== WDOG_PRETIMEOUT_NMI
) {
978 if (preop_val
== WDOG_PREOP_GIVE_DATA
) {
979 printk(KERN_WARNING PFX
"Pretimeout op is to give data"
980 " but NMI pretimeout is enabled, setting"
981 " pretimeout op to none\n");
982 preop_val
= WDOG_PREOP_NONE
;
984 #ifdef CONFIG_X86_LOCAL_APIC
985 if (nmi_watchdog
== NMI_IO_APIC
) {
986 printk(KERN_WARNING PFX
"nmi_watchdog is set to IO APIC"
987 " mode (value is %d), that is incompatible"
988 " with using NMI in the IPMI watchdog."
989 " Disabling IPMI nmi pretimeout.\n",
991 preaction_val
= WDOG_PRETIMEOUT_NONE
;
994 rv
= request_nmi(&ipmi_nmi_handler
);
996 printk(KERN_WARNING PFX
"Can't register nmi handler\n");
999 #ifdef CONFIG_X86_LOCAL_APIC
1005 rv
= ipmi_smi_watcher_register(&smi_watcher
);
1007 #ifdef HAVE_NMI_HANDLER
1008 if (preaction_val
== WDOG_PRETIMEOUT_NMI
)
1009 release_nmi(&ipmi_nmi_handler
);
1011 printk(KERN_WARNING PFX
"can't register smi watcher\n");
1015 register_reboot_notifier(&wdog_reboot_notifier
);
1016 notifier_chain_register(&panic_notifier_list
, &wdog_panic_notifier
);
1021 static __exit
void ipmi_unregister_watchdog(void)
1025 down_write(®ister_sem
);
1027 #ifdef HAVE_NMI_HANDLER
1028 if (preaction_val
== WDOG_PRETIMEOUT_NMI
)
1029 release_nmi(&ipmi_nmi_handler
);
1032 notifier_chain_unregister(&panic_notifier_list
, &wdog_panic_notifier
);
1033 unregister_reboot_notifier(&wdog_reboot_notifier
);
1035 if (! watchdog_user
)
1038 /* Make sure no one can call us any more. */
1039 misc_deregister(&ipmi_wdog_miscdev
);
1041 /* Wait to make sure the message makes it out. The lower layer has
1042 pointers to our buffers, we want to make sure they are done before
1043 we release our memory. */
1044 while (atomic_read(&set_timeout_tofree
)) {
1045 set_current_state(TASK_UNINTERRUPTIBLE
);
1046 schedule_timeout(1);
1049 /* Disconnect from IPMI. */
1050 rv
= ipmi_destroy_user(watchdog_user
);
1052 printk(KERN_WARNING PFX
"error unlinking from IPMI: %d\n",
1055 watchdog_user
= NULL
;
1058 up_write(®ister_sem
);
1061 static void __exit
ipmi_wdog_exit(void)
1063 ipmi_smi_watcher_unregister(&smi_watcher
);
1064 ipmi_unregister_watchdog();
1066 module_exit(ipmi_wdog_exit
);
1067 module_init(ipmi_wdog_init
);
1068 MODULE_LICENSE("GPL");