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 static int nowayout
= WATCHDOG_NOWAYOUT
;
136 static ipmi_user_t watchdog_user
= NULL
;
138 /* Default the timeout to 10 seconds. */
139 static int timeout
= 10;
141 /* The pre-timeout is disabled by default. */
142 static int pretimeout
= 0;
144 /* Default action is to reset the board on a timeout. */
145 static unsigned char action_val
= WDOG_TIMEOUT_RESET
;
147 static char action
[16] = "reset";
149 static unsigned char preaction_val
= WDOG_PRETIMEOUT_NONE
;
151 static char preaction
[16] = "pre_none";
153 static unsigned char preop_val
= WDOG_PREOP_NONE
;
155 static char preop
[16] = "preop_none";
156 static DEFINE_SPINLOCK(ipmi_read_lock
);
157 static char data_to_read
= 0;
158 static DECLARE_WAIT_QUEUE_HEAD(read_q
);
159 static struct fasync_struct
*fasync_q
= NULL
;
160 static char pretimeout_since_last_heartbeat
= 0;
161 static char expect_close
;
163 /* If true, the driver will start running as soon as it is configured
165 static int start_now
= 0;
167 module_param(timeout
, int, 0);
168 MODULE_PARM_DESC(timeout
, "Timeout value in seconds.");
169 module_param(pretimeout
, int, 0);
170 MODULE_PARM_DESC(pretimeout
, "Pretimeout value in seconds.");
171 module_param_string(action
, action
, sizeof(action
), 0);
172 MODULE_PARM_DESC(action
, "Timeout action. One of: "
173 "reset, none, power_cycle, power_off.");
174 module_param_string(preaction
, preaction
, sizeof(preaction
), 0);
175 MODULE_PARM_DESC(preaction
, "Pretimeout action. One of: "
176 "pre_none, pre_smi, pre_nmi, pre_int.");
177 module_param_string(preop
, preop
, sizeof(preop
), 0);
178 MODULE_PARM_DESC(preop
, "Pretimeout driver operation. One of: "
179 "preop_none, preop_panic, preop_give_data.");
180 module_param(start_now
, int, 0);
181 MODULE_PARM_DESC(start_now
, "Set to 1 to start the watchdog as"
182 "soon as the driver is loaded.");
183 module_param(nowayout
, int, 0);
184 MODULE_PARM_DESC(nowayout
, "Watchdog cannot be stopped once started (default=CONFIG_WATCHDOG_NOWAYOUT)");
186 /* Default state of the timer. */
187 static unsigned char ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
189 /* If shutting down via IPMI, we ignore the heartbeat. */
190 static int ipmi_ignore_heartbeat
= 0;
192 /* Is someone using the watchdog? Only one user is allowed. */
193 static unsigned long ipmi_wdog_open
= 0;
195 /* If set to 1, the heartbeat command will set the state to reset and
196 start the timer. The timer doesn't normally run when the driver is
197 first opened until the heartbeat is set the first time, this
198 variable is used to accomplish this. */
199 static int ipmi_start_timer_on_heartbeat
= 0;
201 /* IPMI version of the BMC. */
202 static unsigned char ipmi_version_major
;
203 static unsigned char ipmi_version_minor
;
206 static int ipmi_heartbeat(void);
207 static void panic_halt_ipmi_heartbeat(void);
210 /* We use a semaphore to make sure that only one thing can send a set
211 timeout at one time, because we only have one copy of the data.
212 The semaphore is claimed when the set_timeout is sent and freed
213 when both messages are free. */
214 static atomic_t set_timeout_tofree
= ATOMIC_INIT(0);
215 static DECLARE_MUTEX(set_timeout_lock
);
216 static void set_timeout_free_smi(struct ipmi_smi_msg
*msg
)
218 if (atomic_dec_and_test(&set_timeout_tofree
))
219 up(&set_timeout_lock
);
221 static void set_timeout_free_recv(struct ipmi_recv_msg
*msg
)
223 if (atomic_dec_and_test(&set_timeout_tofree
))
224 up(&set_timeout_lock
);
226 static struct ipmi_smi_msg set_timeout_smi_msg
=
228 .done
= set_timeout_free_smi
230 static struct ipmi_recv_msg set_timeout_recv_msg
=
232 .done
= set_timeout_free_recv
235 static int i_ipmi_set_timeout(struct ipmi_smi_msg
*smi_msg
,
236 struct ipmi_recv_msg
*recv_msg
,
237 int *send_heartbeat_now
)
239 struct kernel_ipmi_msg msg
;
240 unsigned char data
[6];
242 struct ipmi_system_interface_addr addr
;
247 WDOG_SET_TIMER_USE(data
[0], WDOG_TIMER_USE_SMS_OS
);
249 if ((ipmi_version_major
> 1)
250 || ((ipmi_version_major
== 1) && (ipmi_version_minor
>= 5)))
252 /* This is an IPMI 1.5-only feature. */
253 data
[0] |= WDOG_DONT_STOP_ON_SET
;
254 } else if (ipmi_watchdog_state
!= WDOG_TIMEOUT_NONE
) {
255 /* In ipmi 1.0, setting the timer stops the watchdog, we
256 need to start it back up again. */
261 WDOG_SET_TIMEOUT_ACT(data
[1], ipmi_watchdog_state
);
262 if ((pretimeout
> 0) && (ipmi_watchdog_state
!= WDOG_TIMEOUT_NONE
)) {
263 WDOG_SET_PRETIMEOUT_ACT(data
[1], preaction_val
);
264 data
[2] = pretimeout
;
266 WDOG_SET_PRETIMEOUT_ACT(data
[1], WDOG_PRETIMEOUT_NONE
);
267 data
[2] = 0; /* No pretimeout. */
270 WDOG_SET_TIMEOUT(data
[4], data
[5], timeout
);
272 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
273 addr
.channel
= IPMI_BMC_CHANNEL
;
277 msg
.cmd
= IPMI_WDOG_SET_TIMER
;
279 msg
.data_len
= sizeof(data
);
280 rv
= ipmi_request_supply_msgs(watchdog_user
,
281 (struct ipmi_addr
*) &addr
,
289 printk(KERN_WARNING PFX
"set timeout error: %d\n",
293 if (send_heartbeat_now
)
294 *send_heartbeat_now
= hbnow
;
299 /* Parameters to ipmi_set_timeout */
300 #define IPMI_SET_TIMEOUT_NO_HB 0
301 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
302 #define IPMI_SET_TIMEOUT_FORCE_HB 2
304 static int ipmi_set_timeout(int do_heartbeat
)
306 int send_heartbeat_now
;
310 /* We can only send one of these at a time. */
311 down(&set_timeout_lock
);
313 atomic_set(&set_timeout_tofree
, 2);
315 rv
= i_ipmi_set_timeout(&set_timeout_smi_msg
,
316 &set_timeout_recv_msg
,
317 &send_heartbeat_now
);
319 up(&set_timeout_lock
);
321 if ((do_heartbeat
== IPMI_SET_TIMEOUT_FORCE_HB
)
322 || ((send_heartbeat_now
)
323 && (do_heartbeat
== IPMI_SET_TIMEOUT_HB_IF_NECESSARY
)))
325 rv
= ipmi_heartbeat();
332 static void dummy_smi_free(struct ipmi_smi_msg
*msg
)
335 static void dummy_recv_free(struct ipmi_recv_msg
*msg
)
338 static struct ipmi_smi_msg panic_halt_smi_msg
=
340 .done
= dummy_smi_free
342 static struct ipmi_recv_msg panic_halt_recv_msg
=
344 .done
= dummy_recv_free
347 /* Special call, doesn't claim any locks. This is only to be called
348 at panic or halt time, in run-to-completion mode, when the caller
349 is the only CPU and the only thing that will be going is these IPMI
351 static void panic_halt_ipmi_set_timeout(void)
353 int send_heartbeat_now
;
356 rv
= i_ipmi_set_timeout(&panic_halt_smi_msg
,
357 &panic_halt_recv_msg
,
358 &send_heartbeat_now
);
360 if (send_heartbeat_now
)
361 panic_halt_ipmi_heartbeat();
365 /* We use a semaphore to make sure that only one thing can send a
366 heartbeat at one time, because we only have one copy of the data.
367 The semaphore is claimed when the set_timeout is sent and freed
368 when both messages are free. */
369 static atomic_t heartbeat_tofree
= ATOMIC_INIT(0);
370 static DECLARE_MUTEX(heartbeat_lock
);
371 static DECLARE_MUTEX_LOCKED(heartbeat_wait_lock
);
372 static void heartbeat_free_smi(struct ipmi_smi_msg
*msg
)
374 if (atomic_dec_and_test(&heartbeat_tofree
))
375 up(&heartbeat_wait_lock
);
377 static void heartbeat_free_recv(struct ipmi_recv_msg
*msg
)
379 if (atomic_dec_and_test(&heartbeat_tofree
))
380 up(&heartbeat_wait_lock
);
382 static struct ipmi_smi_msg heartbeat_smi_msg
=
384 .done
= heartbeat_free_smi
386 static struct ipmi_recv_msg heartbeat_recv_msg
=
388 .done
= heartbeat_free_recv
391 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg
=
393 .done
= dummy_smi_free
395 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg
=
397 .done
= dummy_recv_free
400 static int ipmi_heartbeat(void)
402 struct kernel_ipmi_msg msg
;
404 struct ipmi_system_interface_addr addr
;
406 if (ipmi_ignore_heartbeat
) {
410 if (ipmi_start_timer_on_heartbeat
) {
411 ipmi_start_timer_on_heartbeat
= 0;
412 ipmi_watchdog_state
= action_val
;
413 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
414 } else if (pretimeout_since_last_heartbeat
) {
415 /* A pretimeout occurred, make sure we set the timeout.
416 We don't want to set the action, though, we want to
417 leave that alone (thus it can't be combined with the
419 pretimeout_since_last_heartbeat
= 0;
420 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
423 down(&heartbeat_lock
);
425 atomic_set(&heartbeat_tofree
, 2);
427 /* Don't reset the timer if we have the timer turned off, that
428 re-enables the watchdog. */
429 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
) {
434 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
435 addr
.channel
= IPMI_BMC_CHANNEL
;
439 msg
.cmd
= IPMI_WDOG_RESET_TIMER
;
442 rv
= ipmi_request_supply_msgs(watchdog_user
,
443 (struct ipmi_addr
*) &addr
,
452 printk(KERN_WARNING PFX
"heartbeat failure: %d\n",
457 /* Wait for the heartbeat to be sent. */
458 down(&heartbeat_wait_lock
);
460 if (heartbeat_recv_msg
.msg
.data
[0] != 0) {
461 /* Got an error in the heartbeat response. It was already
462 reported in ipmi_wdog_msg_handler, but we should return
472 static void panic_halt_ipmi_heartbeat(void)
474 struct kernel_ipmi_msg msg
;
475 struct ipmi_system_interface_addr addr
;
478 /* Don't reset the timer if we have the timer turned off, that
479 re-enables the watchdog. */
480 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
)
483 addr
.addr_type
= IPMI_SYSTEM_INTERFACE_ADDR_TYPE
;
484 addr
.channel
= IPMI_BMC_CHANNEL
;
488 msg
.cmd
= IPMI_WDOG_RESET_TIMER
;
491 ipmi_request_supply_msgs(watchdog_user
,
492 (struct ipmi_addr
*) &addr
,
496 &panic_halt_heartbeat_smi_msg
,
497 &panic_halt_heartbeat_recv_msg
,
501 static struct watchdog_info ident
=
503 .options
= 0, /* WDIOF_SETTIMEOUT, */
504 .firmware_version
= 1,
508 static int ipmi_ioctl(struct inode
*inode
, struct file
*file
,
509 unsigned int cmd
, unsigned long arg
)
511 void __user
*argp
= (void __user
*)arg
;
516 case WDIOC_GETSUPPORT
:
517 i
= copy_to_user(argp
, &ident
, sizeof(ident
));
518 return i
? -EFAULT
: 0;
520 case WDIOC_SETTIMEOUT
:
521 i
= copy_from_user(&val
, argp
, sizeof(int));
525 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
527 case WDIOC_GETTIMEOUT
:
528 i
= copy_to_user(argp
, &timeout
, sizeof(timeout
));
533 case WDIOC_SET_PRETIMEOUT
:
534 i
= copy_from_user(&val
, argp
, sizeof(int));
538 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY
);
540 case WDIOC_GET_PRETIMEOUT
:
541 i
= copy_to_user(argp
, &pretimeout
, sizeof(pretimeout
));
546 case WDIOC_KEEPALIVE
:
547 return ipmi_heartbeat();
549 case WDIOC_SETOPTIONS
:
550 i
= copy_from_user(&val
, argp
, sizeof(int));
553 if (val
& WDIOS_DISABLECARD
)
555 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
556 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB
);
557 ipmi_start_timer_on_heartbeat
= 0;
560 if (val
& WDIOS_ENABLECARD
)
562 ipmi_watchdog_state
= action_val
;
563 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
567 case WDIOC_GETSTATUS
:
569 i
= copy_to_user(argp
, &val
, sizeof(val
));
579 static ssize_t
ipmi_write(struct file
*file
,
580 const char __user
*buf
,
590 /* In case it was set long ago */
593 for (i
= 0; i
!= len
; i
++) {
596 if (get_user(c
, buf
+ i
))
602 rv
= ipmi_heartbeat();
610 static ssize_t
ipmi_read(struct file
*file
,
621 /* Reading returns if the pretimeout has gone off, and it only does
622 it once per pretimeout. */
623 spin_lock(&ipmi_read_lock
);
625 if (file
->f_flags
& O_NONBLOCK
) {
630 init_waitqueue_entry(&wait
, current
);
631 add_wait_queue(&read_q
, &wait
);
632 while (!data_to_read
) {
633 set_current_state(TASK_INTERRUPTIBLE
);
634 spin_unlock(&ipmi_read_lock
);
636 spin_lock(&ipmi_read_lock
);
638 remove_wait_queue(&read_q
, &wait
);
640 if (signal_pending(current
)) {
648 spin_unlock(&ipmi_read_lock
);
651 if (copy_to_user(buf
, &data_to_read
, 1))
660 static int ipmi_open(struct inode
*ino
, struct file
*filep
)
665 if(test_and_set_bit(0, &ipmi_wdog_open
))
668 /* Don't start the timer now, let it start on the
670 ipmi_start_timer_on_heartbeat
= 1;
671 return nonseekable_open(ino
, filep
);
678 static unsigned int ipmi_poll(struct file
*file
, poll_table
*wait
)
680 unsigned int mask
= 0;
682 poll_wait(file
, &read_q
, wait
);
684 spin_lock(&ipmi_read_lock
);
686 mask
|= (POLLIN
| POLLRDNORM
);
687 spin_unlock(&ipmi_read_lock
);
692 static int ipmi_fasync(int fd
, struct file
*file
, int on
)
696 result
= fasync_helper(fd
, file
, on
, &fasync_q
);
701 static int ipmi_close(struct inode
*ino
, struct file
*filep
)
703 if (iminor(ino
)==WATCHDOG_MINOR
)
705 if (expect_close
== 42) {
706 ipmi_watchdog_state
= WDOG_TIMEOUT_NONE
;
707 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB
);
709 printk(KERN_CRIT PFX
"Unexpected close, not stopping watchdog!\n");
712 clear_bit(0, &ipmi_wdog_open
);
715 ipmi_fasync (-1, filep
, 0);
721 static struct file_operations ipmi_wdog_fops
= {
722 .owner
= THIS_MODULE
,
728 .release
= ipmi_close
,
729 .fasync
= ipmi_fasync
,
732 static struct miscdevice ipmi_wdog_miscdev
= {
733 .minor
= WATCHDOG_MINOR
,
735 .fops
= &ipmi_wdog_fops
738 static DECLARE_RWSEM(register_sem
);
740 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg
*msg
,
743 if (msg
->msg
.data
[0] != 0) {
744 printk(KERN_ERR PFX
"response: Error %x on cmd %x\n",
749 ipmi_free_recv_msg(msg
);
752 static void ipmi_wdog_pretimeout_handler(void *handler_data
)
754 if (preaction_val
!= WDOG_PRETIMEOUT_NONE
) {
755 if (preop_val
== WDOG_PREOP_PANIC
)
756 panic("Watchdog pre-timeout");
757 else if (preop_val
== WDOG_PREOP_GIVE_DATA
) {
758 spin_lock(&ipmi_read_lock
);
760 wake_up_interruptible(&read_q
);
761 kill_fasync(&fasync_q
, SIGIO
, POLL_IN
);
763 spin_unlock(&ipmi_read_lock
);
767 /* On some machines, the heartbeat will give
768 an error and not work unless we re-enable
769 the timer. So do so. */
770 pretimeout_since_last_heartbeat
= 1;
773 static struct ipmi_user_hndl ipmi_hndlrs
=
775 .ipmi_recv_hndl
= ipmi_wdog_msg_handler
,
776 .ipmi_watchdog_pretimeout
= ipmi_wdog_pretimeout_handler
779 static void ipmi_register_watchdog(int ipmi_intf
)
783 down_write(®ister_sem
);
787 rv
= ipmi_create_user(ipmi_intf
, &ipmi_hndlrs
, NULL
, &watchdog_user
);
789 printk(KERN_CRIT PFX
"Unable to register with ipmi\n");
793 ipmi_get_version(watchdog_user
,
795 &ipmi_version_minor
);
797 rv
= misc_register(&ipmi_wdog_miscdev
);
799 ipmi_destroy_user(watchdog_user
);
800 watchdog_user
= NULL
;
801 printk(KERN_CRIT PFX
"Unable to register misc device\n");
805 up_write(®ister_sem
);
807 if ((start_now
) && (rv
== 0)) {
808 /* Run from startup, so start the timer now. */
809 start_now
= 0; /* Disable this function after first startup. */
810 ipmi_watchdog_state
= action_val
;
811 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB
);
812 printk(KERN_INFO PFX
"Starting now!\n");
816 #ifdef HAVE_NMI_HANDLER
818 ipmi_nmi(void *dev_id
, struct pt_regs
*regs
, int cpu
, int handled
)
820 /* If we are not expecting a timeout, ignore it. */
821 if (ipmi_watchdog_state
== WDOG_TIMEOUT_NONE
)
824 /* If no one else handled the NMI, we assume it was the IPMI
826 if ((!handled
) && (preop_val
== WDOG_PREOP_PANIC
)) {
827 /* On some machines, the heartbeat will give
828 an error and not work unless we re-enable
829 the timer. So do so. */
830 pretimeout_since_last_heartbeat
= 1;
831 panic(PFX
"pre-timeout");
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");