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 #define IPMI_WDOG_TIMER_NOT_INIT_RESP 0x80
144 static DEFINE_MUTEX(ipmi_watchdog_mutex);
145 static bool nowayout = WATCHDOG_NOWAYOUT;
147 static ipmi_user_t watchdog_user;
148 static int watchdog_ifnum;
150 /* Default the timeout to 10 seconds. */
151 static int timeout = 10;
153 /* The pre-timeout is disabled by default. */
154 static int pretimeout;
156 /* Default timeout to set on panic */
157 static int panic_wdt_timeout = 255;
159 /* Default action is to reset the board on a timeout. */
160 static unsigned char action_val = WDOG_TIMEOUT_RESET;
162 static char action[16] = "reset";
164 static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE;
166 static char preaction[16] = "pre_none";
168 static unsigned char preop_val = WDOG_PREOP_NONE;
170 static char preop[16] = "preop_none";
171 static DEFINE_SPINLOCK(ipmi_read_lock);
172 static char data_to_read;
173 static DECLARE_WAIT_QUEUE_HEAD(read_q);
174 static struct fasync_struct *fasync_q;
175 static char pretimeout_since_last_heartbeat;
176 static char expect_close;
178 static int ifnum_to_use = -1;
180 /* Parameters to ipmi_set_timeout */
181 #define IPMI_SET_TIMEOUT_NO_HB 0
182 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
183 #define IPMI_SET_TIMEOUT_FORCE_HB 2
185 static int ipmi_set_timeout(int do_heartbeat);
186 static void ipmi_register_watchdog(int ipmi_intf);
187 static void ipmi_unregister_watchdog(int ipmi_intf);
190 * If true, the driver will start running as soon as it is configured
193 static int start_now;
195 static int set_param_timeout(const char *val, const struct kernel_param *kp)
203 l = simple_strtoul(val, &endp, 0);
207 *((int *)kp->arg) = l;
209 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
214 static const struct kernel_param_ops param_ops_timeout = {
215 .set = set_param_timeout,
216 .get = param_get_int,
218 #define param_check_timeout param_check_int
220 typedef int (*action_fn)(const char *intval, char *outval);
222 static int action_op(const char *inval, char *outval);
223 static int preaction_op(const char *inval, char *outval);
224 static int preop_op(const char *inval, char *outval);
225 static void check_parms(void);
227 static int set_param_str(const char *val, const struct kernel_param *kp)
229 action_fn fn = (action_fn) kp->arg;
234 strncpy(valcp, val, 16);
245 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
251 static int get_param_str(char *buffer, const struct kernel_param *kp)
253 action_fn fn = (action_fn) kp->arg;
256 rv = fn(NULL, buffer);
259 return strlen(buffer);
263 static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp)
265 int rv = param_set_int(val, kp);
268 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum))
271 ipmi_unregister_watchdog(watchdog_ifnum);
272 ipmi_register_watchdog(ifnum_to_use);
276 static const struct kernel_param_ops param_ops_wdog_ifnum = {
277 .set = set_param_wdog_ifnum,
278 .get = param_get_int,
281 #define param_check_wdog_ifnum param_check_int
283 static const struct kernel_param_ops param_ops_str = {
284 .set = set_param_str,
285 .get = get_param_str,
288 module_param(ifnum_to_use, wdog_ifnum, 0644);
289 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog "
290 "timer. Setting to -1 defaults to the first registered "
293 module_param(timeout, timeout, 0644);
294 MODULE_PARM_DESC(timeout, "Timeout value in seconds.");
296 module_param(pretimeout, timeout, 0644);
297 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds.");
299 module_param(panic_wdt_timeout, timeout, 0644);
300 MODULE_PARM_DESC(timeout, "Timeout value on kernel panic in seconds.");
302 module_param_cb(action, ¶m_ops_str, action_op, 0644);
303 MODULE_PARM_DESC(action, "Timeout action. One of: "
304 "reset, none, power_cycle, power_off.");
306 module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644);
307 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: "
308 "pre_none, pre_smi, pre_nmi, pre_int.");
310 module_param_cb(preop, ¶m_ops_str, preop_op, 0644);
311 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: "
312 "preop_none, preop_panic, preop_give_data.");
314 module_param(start_now, int, 0444);
315 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as"
316 "soon as the driver is loaded.");
318 module_param(nowayout, bool, 0644);
319 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started "
320 "(default=CONFIG_WATCHDOG_NOWAYOUT)");
322 /* Default state of the timer. */
323 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
325 /* If shutting down via IPMI, we ignore the heartbeat. */
326 static int ipmi_ignore_heartbeat;
328 /* Is someone using the watchdog? Only one user is allowed. */
329 static unsigned long ipmi_wdog_open;
332 * If set to 1, the heartbeat command will set the state to reset and
333 * start the timer. The timer doesn't normally run when the driver is
334 * first opened until the heartbeat is set the first time, this
335 * variable is used to accomplish this.
337 static int ipmi_start_timer_on_heartbeat;
339 /* IPMI version of the BMC. */
340 static unsigned char ipmi_version_major;
341 static unsigned char ipmi_version_minor;
343 /* If a pretimeout occurs, this is used to allow only one panic to happen. */
344 static atomic_t preop_panic_excl = ATOMIC_INIT(-1);
347 static int testing_nmi;
348 static int nmi_handler_registered;
351 static int ipmi_heartbeat(void);
354 * We use a mutex to make sure that only one thing can send a set
355 * timeout at one time, because we only have one copy of the data.
356 * The mutex is claimed when the set_timeout is sent and freed
357 * when both messages are free.
359 static atomic_t set_timeout_tofree = ATOMIC_INIT(0);
360 static DEFINE_MUTEX(set_timeout_lock);
361 static DECLARE_COMPLETION(set_timeout_wait);
362 static void set_timeout_free_smi(struct ipmi_smi_msg *msg)
364 if (atomic_dec_and_test(&set_timeout_tofree))
365 complete(&set_timeout_wait);
367 static void set_timeout_free_recv(struct ipmi_recv_msg *msg)
369 if (atomic_dec_and_test(&set_timeout_tofree))
370 complete(&set_timeout_wait);
372 static struct ipmi_smi_msg set_timeout_smi_msg = {
373 .done = set_timeout_free_smi
375 static struct ipmi_recv_msg set_timeout_recv_msg = {
376 .done = set_timeout_free_recv
379 static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg,
380 struct ipmi_recv_msg *recv_msg,
381 int *send_heartbeat_now)
383 struct kernel_ipmi_msg msg;
384 unsigned char data[6];
386 struct ipmi_system_interface_addr addr;
390 /* These can be cleared as we are setting the timeout. */
391 pretimeout_since_last_heartbeat = 0;
394 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
396 if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
397 if ((ipmi_version_major > 1) ||
398 ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) {
399 /* This is an IPMI 1.5-only feature. */
400 data[0] |= WDOG_DONT_STOP_ON_SET;
403 * In ipmi 1.0, setting the timer stops the watchdog, we
404 * need to start it back up again.
411 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state);
412 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) {
413 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val);
414 data[2] = pretimeout;
416 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE);
417 data[2] = 0; /* No pretimeout. */
420 WDOG_SET_TIMEOUT(data[4], data[5], timeout);
422 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
423 addr.channel = IPMI_BMC_CHANNEL;
427 msg.cmd = IPMI_WDOG_SET_TIMER;
429 msg.data_len = sizeof(data);
430 rv = ipmi_request_supply_msgs(watchdog_user,
431 (struct ipmi_addr *) &addr,
439 printk(KERN_WARNING PFX "set timeout error: %d\n",
443 if (send_heartbeat_now)
444 *send_heartbeat_now = hbnow;
449 static int ipmi_set_timeout(int do_heartbeat)
451 int send_heartbeat_now;
455 /* We can only send one of these at a time. */
456 mutex_lock(&set_timeout_lock);
458 atomic_set(&set_timeout_tofree, 2);
460 rv = i_ipmi_set_timeout(&set_timeout_smi_msg,
461 &set_timeout_recv_msg,
462 &send_heartbeat_now);
464 mutex_unlock(&set_timeout_lock);
468 wait_for_completion(&set_timeout_wait);
470 mutex_unlock(&set_timeout_lock);
472 if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB)
473 || ((send_heartbeat_now)
474 && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY)))
475 rv = ipmi_heartbeat();
481 static atomic_t panic_done_count = ATOMIC_INIT(0);
483 static void panic_smi_free(struct ipmi_smi_msg *msg)
485 atomic_dec(&panic_done_count);
487 static void panic_recv_free(struct ipmi_recv_msg *msg)
489 atomic_dec(&panic_done_count);
492 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = {
493 .done = panic_smi_free
495 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = {
496 .done = panic_recv_free
499 static void panic_halt_ipmi_heartbeat(void)
501 struct kernel_ipmi_msg msg;
502 struct ipmi_system_interface_addr addr;
506 * Don't reset the timer if we have the timer turned off, that
507 * re-enables the watchdog.
509 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
512 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
513 addr.channel = IPMI_BMC_CHANNEL;
517 msg.cmd = IPMI_WDOG_RESET_TIMER;
520 atomic_add(1, &panic_done_count);
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_sub(1, &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 atomic_add(1, &panic_done_count);
555 rv = i_ipmi_set_timeout(&panic_halt_smi_msg,
556 &panic_halt_recv_msg,
557 &send_heartbeat_now);
559 atomic_sub(1, &panic_done_count);
560 printk(KERN_WARNING PFX
561 "Unable to extend the watchdog timeout.");
563 if (send_heartbeat_now)
564 panic_halt_ipmi_heartbeat();
566 while (atomic_read(&panic_done_count) != 0)
567 ipmi_poll_interface(watchdog_user);
571 * We use a mutex to make sure that only one thing can send a
572 * heartbeat at one time, because we only have one copy of the data.
573 * The semaphore is claimed when the set_timeout is sent and freed
574 * when both messages are free.
576 static atomic_t heartbeat_tofree = ATOMIC_INIT(0);
577 static DEFINE_MUTEX(heartbeat_lock);
578 static DECLARE_COMPLETION(heartbeat_wait);
579 static void heartbeat_free_smi(struct ipmi_smi_msg *msg)
581 if (atomic_dec_and_test(&heartbeat_tofree))
582 complete(&heartbeat_wait);
584 static void heartbeat_free_recv(struct ipmi_recv_msg *msg)
586 if (atomic_dec_and_test(&heartbeat_tofree))
587 complete(&heartbeat_wait);
589 static struct ipmi_smi_msg heartbeat_smi_msg = {
590 .done = heartbeat_free_smi
592 static struct ipmi_recv_msg heartbeat_recv_msg = {
593 .done = heartbeat_free_recv
596 static int ipmi_heartbeat(void)
598 struct kernel_ipmi_msg msg;
600 struct ipmi_system_interface_addr addr;
601 int timeout_retries = 0;
603 if (ipmi_ignore_heartbeat)
606 if (ipmi_start_timer_on_heartbeat) {
607 ipmi_start_timer_on_heartbeat = 0;
608 ipmi_watchdog_state = action_val;
609 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
610 } else if (pretimeout_since_last_heartbeat) {
612 * A pretimeout occurred, make sure we set the timeout.
613 * We don't want to set the action, though, we want to
614 * leave that alone (thus it can't be combined with the
617 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
620 mutex_lock(&heartbeat_lock);
623 atomic_set(&heartbeat_tofree, 2);
626 * Don't reset the timer if we have the timer turned off, that
627 * re-enables the watchdog.
629 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) {
630 mutex_unlock(&heartbeat_lock);
634 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
635 addr.channel = IPMI_BMC_CHANNEL;
639 msg.cmd = IPMI_WDOG_RESET_TIMER;
642 rv = ipmi_request_supply_msgs(watchdog_user,
643 (struct ipmi_addr *) &addr,
651 mutex_unlock(&heartbeat_lock);
652 printk(KERN_WARNING PFX "heartbeat failure: %d\n",
657 /* Wait for the heartbeat to be sent. */
658 wait_for_completion(&heartbeat_wait);
660 if (heartbeat_recv_msg.msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) {
662 if (timeout_retries > 3) {
663 printk(KERN_ERR PFX ": Unable to restore the IPMI"
664 " watchdog's settings, giving up.\n");
670 * The timer was not initialized, that means the BMC was
671 * probably reset and lost the watchdog information. Attempt
672 * to restore the timer's info. Note that we still hold
673 * the heartbeat lock, to keep a heartbeat from happening
674 * in this process, so must say no heartbeat to avoid a
675 * deadlock on this mutex.
677 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
679 printk(KERN_ERR PFX ": Unable to send the command to"
680 " set the watchdog's settings, giving up.\n");
684 /* We might need a new heartbeat, so do it now */
686 } else if (heartbeat_recv_msg.msg.data[0] != 0) {
688 * Got an error in the heartbeat response. It was already
689 * reported in ipmi_wdog_msg_handler, but we should return
696 mutex_unlock(&heartbeat_lock);
701 static struct watchdog_info ident = {
702 .options = 0, /* WDIOF_SETTIMEOUT, */
703 .firmware_version = 1,
707 static int ipmi_ioctl(struct file *file,
708 unsigned int cmd, unsigned long arg)
710 void __user *argp = (void __user *)arg;
715 case WDIOC_GETSUPPORT:
716 i = copy_to_user(argp, &ident, sizeof(ident));
717 return i ? -EFAULT : 0;
719 case WDIOC_SETTIMEOUT:
720 i = copy_from_user(&val, argp, sizeof(int));
724 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
726 case WDIOC_GETTIMEOUT:
727 i = copy_to_user(argp, &timeout, sizeof(timeout));
732 case WDIOC_SETPRETIMEOUT:
733 i = copy_from_user(&val, argp, sizeof(int));
737 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
739 case WDIOC_GETPRETIMEOUT:
740 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout));
745 case WDIOC_KEEPALIVE:
746 return ipmi_heartbeat();
748 case WDIOC_SETOPTIONS:
749 i = copy_from_user(&val, argp, sizeof(int));
752 if (val & WDIOS_DISABLECARD) {
753 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
754 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
755 ipmi_start_timer_on_heartbeat = 0;
758 if (val & WDIOS_ENABLECARD) {
759 ipmi_watchdog_state = action_val;
760 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
764 case WDIOC_GETSTATUS:
766 i = copy_to_user(argp, &val, sizeof(val));
776 static long ipmi_unlocked_ioctl(struct file *file,
782 mutex_lock(&ipmi_watchdog_mutex);
783 ret = ipmi_ioctl(file, cmd, arg);
784 mutex_unlock(&ipmi_watchdog_mutex);
789 static ssize_t ipmi_write(struct file *file,
790 const char __user *buf,
800 /* In case it was set long ago */
803 for (i = 0; i != len; i++) {
806 if (get_user(c, buf + i))
812 rv = ipmi_heartbeat();
819 static ssize_t ipmi_read(struct file *file,
831 * Reading returns if the pretimeout has gone off, and it only does
832 * it once per pretimeout.
834 spin_lock(&ipmi_read_lock);
836 if (file->f_flags & O_NONBLOCK) {
841 init_waitqueue_entry(&wait, current);
842 add_wait_queue(&read_q, &wait);
843 while (!data_to_read) {
844 set_current_state(TASK_INTERRUPTIBLE);
845 spin_unlock(&ipmi_read_lock);
847 spin_lock(&ipmi_read_lock);
849 remove_wait_queue(&read_q, &wait);
851 if (signal_pending(current)) {
859 spin_unlock(&ipmi_read_lock);
862 if (copy_to_user(buf, &data_to_read, 1))
871 static int ipmi_open(struct inode *ino, struct file *filep)
873 switch (iminor(ino)) {
875 if (test_and_set_bit(0, &ipmi_wdog_open))
880 * Don't start the timer now, let it start on the
883 ipmi_start_timer_on_heartbeat = 1;
884 return nonseekable_open(ino, filep);
891 static unsigned int ipmi_poll(struct file *file, poll_table *wait)
893 unsigned int mask = 0;
895 poll_wait(file, &read_q, wait);
897 spin_lock(&ipmi_read_lock);
899 mask |= (POLLIN | POLLRDNORM);
900 spin_unlock(&ipmi_read_lock);
905 static int ipmi_fasync(int fd, struct file *file, int on)
909 result = fasync_helper(fd, file, on, &fasync_q);
914 static int ipmi_close(struct inode *ino, struct file *filep)
916 if (iminor(ino) == WATCHDOG_MINOR) {
917 if (expect_close == 42) {
918 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
919 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
922 "Unexpected close, not stopping watchdog!\n");
925 clear_bit(0, &ipmi_wdog_open);
933 static const struct file_operations ipmi_wdog_fops = {
934 .owner = THIS_MODULE,
938 .unlocked_ioctl = ipmi_unlocked_ioctl,
940 .release = ipmi_close,
941 .fasync = ipmi_fasync,
945 static struct miscdevice ipmi_wdog_miscdev = {
946 .minor = WATCHDOG_MINOR,
948 .fops = &ipmi_wdog_fops
951 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg,
954 if (msg->msg.cmd == IPMI_WDOG_RESET_TIMER &&
955 msg->msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP)
956 printk(KERN_INFO PFX "response: The IPMI controller appears"
957 " to have been reset, will attempt to reinitialize"
958 " the watchdog timer\n");
959 else if (msg->msg.data[0] != 0)
960 printk(KERN_ERR PFX "response: Error %x on cmd %x\n",
964 ipmi_free_recv_msg(msg);
967 static void ipmi_wdog_pretimeout_handler(void *handler_data)
969 if (preaction_val != WDOG_PRETIMEOUT_NONE) {
970 if (preop_val == WDOG_PREOP_PANIC) {
971 if (atomic_inc_and_test(&preop_panic_excl))
972 panic("Watchdog pre-timeout");
973 } else if (preop_val == WDOG_PREOP_GIVE_DATA) {
974 spin_lock(&ipmi_read_lock);
976 wake_up_interruptible(&read_q);
977 kill_fasync(&fasync_q, SIGIO, POLL_IN);
979 spin_unlock(&ipmi_read_lock);
984 * On some machines, the heartbeat will give an error and not
985 * work unless we re-enable the timer. So do so.
987 pretimeout_since_last_heartbeat = 1;
990 static struct ipmi_user_hndl ipmi_hndlrs = {
991 .ipmi_recv_hndl = ipmi_wdog_msg_handler,
992 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler
995 static void ipmi_register_watchdog(int ipmi_intf)
1002 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf))
1005 watchdog_ifnum = ipmi_intf;
1007 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user);
1009 printk(KERN_CRIT PFX "Unable to register with ipmi\n");
1013 ipmi_get_version(watchdog_user,
1014 &ipmi_version_major,
1015 &ipmi_version_minor);
1017 rv = misc_register(&ipmi_wdog_miscdev);
1019 ipmi_destroy_user(watchdog_user);
1020 watchdog_user = NULL;
1021 printk(KERN_CRIT PFX "Unable to register misc device\n");
1025 if (nmi_handler_registered) {
1026 int old_pretimeout = pretimeout;
1027 int old_timeout = timeout;
1028 int old_preop_val = preop_val;
1031 * Set the pretimeout to go off in a second and give
1032 * ourselves plenty of time to stop the timer.
1034 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1035 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */
1041 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1043 printk(KERN_WARNING PFX "Error starting timer to"
1044 " test NMI: 0x%x. The NMI pretimeout will"
1045 " likely not work\n", rv);
1052 if (testing_nmi != 2) {
1053 printk(KERN_WARNING PFX "IPMI NMI didn't seem to"
1054 " occur. The NMI pretimeout will"
1055 " likely not work\n");
1059 preop_val = old_preop_val;
1060 pretimeout = old_pretimeout;
1061 timeout = old_timeout;
1066 if ((start_now) && (rv == 0)) {
1067 /* Run from startup, so start the timer now. */
1068 start_now = 0; /* Disable this function after first startup. */
1069 ipmi_watchdog_state = action_val;
1070 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1071 printk(KERN_INFO PFX "Starting now!\n");
1073 /* Stop the timer now. */
1074 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1075 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1079 static void ipmi_unregister_watchdog(int ipmi_intf)
1086 if (watchdog_ifnum != ipmi_intf)
1089 /* Make sure no one can call us any more. */
1090 misc_deregister(&ipmi_wdog_miscdev);
1093 * Wait to make sure the message makes it out. The lower layer has
1094 * pointers to our buffers, we want to make sure they are done before
1095 * we release our memory.
1097 while (atomic_read(&set_timeout_tofree))
1098 schedule_timeout_uninterruptible(1);
1100 /* Disconnect from IPMI. */
1101 rv = ipmi_destroy_user(watchdog_user);
1103 printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n",
1106 watchdog_user = NULL;
1114 ipmi_nmi(unsigned int val, struct pt_regs *regs)
1117 * If we get here, it's an NMI that's not a memory or I/O
1118 * error. We can't truly tell if it's from IPMI or not
1119 * without sending a message, and sending a message is almost
1120 * impossible because of locking.
1128 /* If we are not expecting a timeout, ignore it. */
1129 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
1132 if (preaction_val != WDOG_PRETIMEOUT_NMI)
1136 * If no one else handled the NMI, we assume it was the IPMI
1139 if (preop_val == WDOG_PREOP_PANIC) {
1140 /* On some machines, the heartbeat will give
1141 an error and not work unless we re-enable
1142 the timer. So do so. */
1143 pretimeout_since_last_heartbeat = 1;
1144 if (atomic_inc_and_test(&preop_panic_excl))
1145 nmi_panic(regs, PFX "pre-timeout");
1152 static int wdog_reboot_handler(struct notifier_block *this,
1156 static int reboot_event_handled;
1158 if ((watchdog_user) && (!reboot_event_handled)) {
1159 /* Make sure we only do this once. */
1160 reboot_event_handled = 1;
1162 if (code == SYS_POWER_OFF || code == SYS_HALT) {
1163 /* Disable the WDT if we are shutting down. */
1164 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1165 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1166 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1167 /* Set a long timer to let the reboot happen or
1168 reset if it hangs, but only if the watchdog
1169 timer was already running. */
1173 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1174 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1180 static struct notifier_block wdog_reboot_notifier = {
1181 .notifier_call = wdog_reboot_handler,
1186 static int wdog_panic_handler(struct notifier_block *this,
1187 unsigned long event,
1190 static int panic_event_handled;
1192 /* On a panic, if we have a panic timeout, make sure to extend
1193 the watchdog timer to a reasonable value to complete the
1194 panic, if the watchdog timer is running. Plus the
1195 pretimeout is meaningless at panic time. */
1196 if (watchdog_user && !panic_event_handled &&
1197 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1198 /* Make sure we do this only once. */
1199 panic_event_handled = 1;
1201 timeout = panic_wdt_timeout;
1203 panic_halt_ipmi_set_timeout();
1209 static struct notifier_block wdog_panic_notifier = {
1210 .notifier_call = wdog_panic_handler,
1212 .priority = 150 /* priority: INT_MAX >= x >= 0 */
1216 static void ipmi_new_smi(int if_num, struct device *device)
1218 ipmi_register_watchdog(if_num);
1221 static void ipmi_smi_gone(int if_num)
1223 ipmi_unregister_watchdog(if_num);
1226 static struct ipmi_smi_watcher smi_watcher = {
1227 .owner = THIS_MODULE,
1228 .new_smi = ipmi_new_smi,
1229 .smi_gone = ipmi_smi_gone
1232 static int action_op(const char *inval, char *outval)
1235 strcpy(outval, action);
1240 if (strcmp(inval, "reset") == 0)
1241 action_val = WDOG_TIMEOUT_RESET;
1242 else if (strcmp(inval, "none") == 0)
1243 action_val = WDOG_TIMEOUT_NONE;
1244 else if (strcmp(inval, "power_cycle") == 0)
1245 action_val = WDOG_TIMEOUT_POWER_CYCLE;
1246 else if (strcmp(inval, "power_off") == 0)
1247 action_val = WDOG_TIMEOUT_POWER_DOWN;
1250 strcpy(action, inval);
1254 static int preaction_op(const char *inval, char *outval)
1257 strcpy(outval, preaction);
1262 if (strcmp(inval, "pre_none") == 0)
1263 preaction_val = WDOG_PRETIMEOUT_NONE;
1264 else if (strcmp(inval, "pre_smi") == 0)
1265 preaction_val = WDOG_PRETIMEOUT_SMI;
1267 else if (strcmp(inval, "pre_nmi") == 0)
1268 preaction_val = WDOG_PRETIMEOUT_NMI;
1270 else if (strcmp(inval, "pre_int") == 0)
1271 preaction_val = WDOG_PRETIMEOUT_MSG_INT;
1274 strcpy(preaction, inval);
1278 static int preop_op(const char *inval, char *outval)
1281 strcpy(outval, preop);
1286 if (strcmp(inval, "preop_none") == 0)
1287 preop_val = WDOG_PREOP_NONE;
1288 else if (strcmp(inval, "preop_panic") == 0)
1289 preop_val = WDOG_PREOP_PANIC;
1290 else if (strcmp(inval, "preop_give_data") == 0)
1291 preop_val = WDOG_PREOP_GIVE_DATA;
1294 strcpy(preop, inval);
1298 static void check_parms(void)
1304 if (preaction_val == WDOG_PRETIMEOUT_NMI) {
1306 if (preop_val == WDOG_PREOP_GIVE_DATA) {
1307 printk(KERN_WARNING PFX "Pretimeout op is to give data"
1308 " but NMI pretimeout is enabled, setting"
1309 " pretimeout op to none\n");
1310 preop_op("preop_none", NULL);
1314 if (do_nmi && !nmi_handler_registered) {
1315 rv = register_nmi_handler(NMI_UNKNOWN, ipmi_nmi, 0,
1318 printk(KERN_WARNING PFX
1319 "Can't register nmi handler\n");
1322 nmi_handler_registered = 1;
1323 } else if (!do_nmi && nmi_handler_registered) {
1324 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1325 nmi_handler_registered = 0;
1330 static int __init ipmi_wdog_init(void)
1334 if (action_op(action, NULL)) {
1335 action_op("reset", NULL);
1336 printk(KERN_INFO PFX "Unknown action '%s', defaulting to"
1337 " reset\n", action);
1340 if (preaction_op(preaction, NULL)) {
1341 preaction_op("pre_none", NULL);
1342 printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to"
1343 " none\n", preaction);
1346 if (preop_op(preop, NULL)) {
1347 preop_op("preop_none", NULL);
1348 printk(KERN_INFO PFX "Unknown preop '%s', defaulting to"
1354 register_reboot_notifier(&wdog_reboot_notifier);
1355 atomic_notifier_chain_register(&panic_notifier_list,
1356 &wdog_panic_notifier);
1358 rv = ipmi_smi_watcher_register(&smi_watcher);
1361 if (nmi_handler_registered)
1362 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1364 atomic_notifier_chain_unregister(&panic_notifier_list,
1365 &wdog_panic_notifier);
1366 unregister_reboot_notifier(&wdog_reboot_notifier);
1367 printk(KERN_WARNING PFX "can't register smi watcher\n");
1371 printk(KERN_INFO PFX "driver initialized\n");
1376 static void __exit ipmi_wdog_exit(void)
1378 ipmi_smi_watcher_unregister(&smi_watcher);
1379 ipmi_unregister_watchdog(watchdog_ifnum);
1382 if (nmi_handler_registered)
1383 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1386 atomic_notifier_chain_unregister(&panic_notifier_list,
1387 &wdog_panic_notifier);
1388 unregister_reboot_notifier(&wdog_reboot_notifier);
1390 module_exit(ipmi_wdog_exit);
1391 module_init(ipmi_wdog_init);
1392 MODULE_LICENSE("GPL");
1393 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
1394 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface.");