1 // SPDX-License-Identifier: GPL-2.0+
5 * A watchdog timer based upon the IPMI interface.
7 * Author: MontaVista Software, Inc.
8 * Corey Minyard <minyard@mvista.com>
11 * Copyright 2002 MontaVista Software Inc.
14 #include <linux/module.h>
15 #include <linux/moduleparam.h>
16 #include <linux/ipmi.h>
17 #include <linux/ipmi_smi.h>
18 #include <linux/mutex.h>
19 #include <linux/watchdog.h>
20 #include <linux/miscdevice.h>
21 #include <linux/init.h>
22 #include <linux/completion.h>
23 #include <linux/kdebug.h>
24 #include <linux/rwsem.h>
25 #include <linux/errno.h>
26 #include <linux/uaccess.h>
27 #include <linux/notifier.h>
28 #include <linux/nmi.h>
29 #include <linux/reboot.h>
30 #include <linux/wait.h>
31 #include <linux/poll.h>
32 #include <linux/string.h>
33 #include <linux/ctype.h>
34 #include <linux/delay.h>
35 #include <linux/atomic.h>
36 #include <linux/sched/signal.h>
40 * This is ugly, but I've determined that x86 is the only architecture
41 * that can reasonably support the IPMI NMI watchdog timeout at this
42 * time. If another architecture adds this capability somehow, it
43 * will have to be a somewhat different mechanism and I have no idea
44 * how it will work. So in the unlikely event that another
45 * architecture supports this, we can figure out a good generic
46 * mechanism for it at that time.
48 #include <asm/kdebug.h>
53 #define PFX "IPMI Watchdog: "
56 * The IPMI command/response information for the watchdog timer.
59 /* values for byte 1 of the set command, byte 2 of the get response. */
60 #define WDOG_DONT_LOG (1 << 7)
61 #define WDOG_DONT_STOP_ON_SET (1 << 6)
62 #define WDOG_SET_TIMER_USE(byte, use) \
63 byte = ((byte) & 0xf8) | ((use) & 0x7)
64 #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7)
65 #define WDOG_TIMER_USE_BIOS_FRB2 1
66 #define WDOG_TIMER_USE_BIOS_POST 2
67 #define WDOG_TIMER_USE_OS_LOAD 3
68 #define WDOG_TIMER_USE_SMS_OS 4
69 #define WDOG_TIMER_USE_OEM 5
71 /* values for byte 2 of the set command, byte 3 of the get response. */
72 #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \
73 byte = ((byte) & 0x8f) | (((use) & 0x7) << 4)
74 #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7)
75 #define WDOG_PRETIMEOUT_NONE 0
76 #define WDOG_PRETIMEOUT_SMI 1
77 #define WDOG_PRETIMEOUT_NMI 2
78 #define WDOG_PRETIMEOUT_MSG_INT 3
80 /* Operations that can be performed on a pretimout. */
81 #define WDOG_PREOP_NONE 0
82 #define WDOG_PREOP_PANIC 1
83 /* Cause data to be available to read. Doesn't work in NMI mode. */
84 #define WDOG_PREOP_GIVE_DATA 2
86 /* Actions to perform on a full timeout. */
87 #define WDOG_SET_TIMEOUT_ACT(byte, use) \
88 byte = ((byte) & 0xf8) | ((use) & 0x7)
89 #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7)
90 #define WDOG_TIMEOUT_NONE 0
91 #define WDOG_TIMEOUT_RESET 1
92 #define WDOG_TIMEOUT_POWER_DOWN 2
93 #define WDOG_TIMEOUT_POWER_CYCLE 3
96 * Byte 3 of the get command, byte 4 of the get response is the
97 * pre-timeout in seconds.
100 /* Bits for setting byte 4 of the set command, byte 5 of the get response. */
101 #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1)
102 #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2)
103 #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3)
104 #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4)
105 #define WDOG_EXPIRE_CLEAR_OEM (1 << 5)
108 * Setting/getting the watchdog timer value. This is for bytes 5 and
109 * 6 (the timeout time) of the set command, and bytes 6 and 7 (the
110 * timeout time) and 8 and 9 (the current countdown value) of the
111 * response. The timeout value is given in seconds (in the command it
112 * 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 #define IPMI_WDOG_TIMER_NOT_INIT_RESP 0x80
125 static DEFINE_MUTEX(ipmi_watchdog_mutex);
126 static bool nowayout = WATCHDOG_NOWAYOUT;
128 static struct ipmi_user *watchdog_user;
129 static int watchdog_ifnum;
131 /* Default the timeout to 10 seconds. */
132 static int timeout = 10;
134 /* The pre-timeout is disabled by default. */
135 static int pretimeout;
137 /* Default timeout to set on panic */
138 static int panic_wdt_timeout = 255;
140 /* Default action is to reset the board on a timeout. */
141 static unsigned char action_val = WDOG_TIMEOUT_RESET;
143 static char action[16] = "reset";
145 static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE;
147 static char preaction[16] = "pre_none";
149 static unsigned char preop_val = WDOG_PREOP_NONE;
151 static char preop[16] = "preop_none";
152 static DEFINE_SPINLOCK(ipmi_read_lock);
153 static char data_to_read;
154 static DECLARE_WAIT_QUEUE_HEAD(read_q);
155 static struct fasync_struct *fasync_q;
156 static atomic_t pretimeout_since_last_heartbeat;
157 static char expect_close;
159 static int ifnum_to_use = -1;
161 /* Parameters to ipmi_set_timeout */
162 #define IPMI_SET_TIMEOUT_NO_HB 0
163 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
164 #define IPMI_SET_TIMEOUT_FORCE_HB 2
166 static int ipmi_set_timeout(int do_heartbeat);
167 static void ipmi_register_watchdog(int ipmi_intf);
168 static void ipmi_unregister_watchdog(int ipmi_intf);
171 * If true, the driver will start running as soon as it is configured
174 static int start_now;
176 static int set_param_timeout(const char *val, const struct kernel_param *kp)
184 l = simple_strtoul(val, &endp, 0);
188 *((int *)kp->arg) = l;
190 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
195 static const struct kernel_param_ops param_ops_timeout = {
196 .set = set_param_timeout,
197 .get = param_get_int,
199 #define param_check_timeout param_check_int
201 typedef int (*action_fn)(const char *intval, char *outval);
203 static int action_op(const char *inval, char *outval);
204 static int preaction_op(const char *inval, char *outval);
205 static int preop_op(const char *inval, char *outval);
206 static void check_parms(void);
208 static int set_param_str(const char *val, const struct kernel_param *kp)
210 action_fn fn = (action_fn) kp->arg;
215 strncpy(valcp, val, 15);
226 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
232 static int get_param_str(char *buffer, const struct kernel_param *kp)
234 action_fn fn = (action_fn) kp->arg;
237 rv = fn(NULL, buffer);
240 return strlen(buffer);
244 static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp)
246 int rv = param_set_int(val, kp);
249 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum))
252 ipmi_unregister_watchdog(watchdog_ifnum);
253 ipmi_register_watchdog(ifnum_to_use);
257 static const struct kernel_param_ops param_ops_wdog_ifnum = {
258 .set = set_param_wdog_ifnum,
259 .get = param_get_int,
262 #define param_check_wdog_ifnum param_check_int
264 static const struct kernel_param_ops param_ops_str = {
265 .set = set_param_str,
266 .get = get_param_str,
269 module_param(ifnum_to_use, wdog_ifnum, 0644);
270 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog "
271 "timer. Setting to -1 defaults to the first registered "
274 module_param(timeout, timeout, 0644);
275 MODULE_PARM_DESC(timeout, "Timeout value in seconds.");
277 module_param(pretimeout, timeout, 0644);
278 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds.");
280 module_param(panic_wdt_timeout, timeout, 0644);
281 MODULE_PARM_DESC(panic_wdt_timeout, "Timeout value on kernel panic in seconds.");
283 module_param_cb(action, ¶m_ops_str, action_op, 0644);
284 MODULE_PARM_DESC(action, "Timeout action. One of: "
285 "reset, none, power_cycle, power_off.");
287 module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644);
288 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: "
289 "pre_none, pre_smi, pre_nmi, pre_int.");
291 module_param_cb(preop, ¶m_ops_str, preop_op, 0644);
292 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: "
293 "preop_none, preop_panic, preop_give_data.");
295 module_param(start_now, int, 0444);
296 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as"
297 "soon as the driver is loaded.");
299 module_param(nowayout, bool, 0644);
300 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started "
301 "(default=CONFIG_WATCHDOG_NOWAYOUT)");
303 /* Default state of the timer. */
304 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
306 /* Is someone using the watchdog? Only one user is allowed. */
307 static unsigned long ipmi_wdog_open;
310 * If set to 1, the heartbeat command will set the state to reset and
311 * start the timer. The timer doesn't normally run when the driver is
312 * first opened until the heartbeat is set the first time, this
313 * variable is used to accomplish this.
315 static int ipmi_start_timer_on_heartbeat;
317 /* IPMI version of the BMC. */
318 static unsigned char ipmi_version_major;
319 static unsigned char ipmi_version_minor;
321 /* If a pretimeout occurs, this is used to allow only one panic to happen. */
322 static atomic_t preop_panic_excl = ATOMIC_INIT(-1);
325 static int testing_nmi;
326 static int nmi_handler_registered;
329 static int __ipmi_heartbeat(void);
332 * We use a mutex to make sure that only one thing can send a set a
333 * message at one time. The mutex is claimed when a message is sent
334 * and freed when both the send and receive messages are free.
336 static atomic_t msg_tofree = ATOMIC_INIT(0);
337 static DECLARE_COMPLETION(msg_wait);
338 static void msg_free_smi(struct ipmi_smi_msg *msg)
340 if (atomic_dec_and_test(&msg_tofree))
343 static void msg_free_recv(struct ipmi_recv_msg *msg)
345 if (atomic_dec_and_test(&msg_tofree))
348 static struct ipmi_smi_msg smi_msg = {
351 static struct ipmi_recv_msg recv_msg = {
352 .done = msg_free_recv
355 static int __ipmi_set_timeout(struct ipmi_smi_msg *smi_msg,
356 struct ipmi_recv_msg *recv_msg,
357 int *send_heartbeat_now)
359 struct kernel_ipmi_msg msg;
360 unsigned char data[6];
362 struct ipmi_system_interface_addr addr;
367 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
369 if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
370 if ((ipmi_version_major > 1) ||
371 ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) {
372 /* This is an IPMI 1.5-only feature. */
373 data[0] |= WDOG_DONT_STOP_ON_SET;
376 * In ipmi 1.0, setting the timer stops the watchdog, we
377 * need to start it back up again.
384 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state);
385 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) {
386 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val);
387 data[2] = pretimeout;
389 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE);
390 data[2] = 0; /* No pretimeout. */
393 WDOG_SET_TIMEOUT(data[4], data[5], timeout);
395 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
396 addr.channel = IPMI_BMC_CHANNEL;
400 msg.cmd = IPMI_WDOG_SET_TIMER;
402 msg.data_len = sizeof(data);
403 rv = ipmi_request_supply_msgs(watchdog_user,
404 (struct ipmi_addr *) &addr,
412 pr_warn(PFX "set timeout error: %d\n", rv);
413 else if (send_heartbeat_now)
414 *send_heartbeat_now = hbnow;
419 static int _ipmi_set_timeout(int do_heartbeat)
421 int send_heartbeat_now;
427 atomic_set(&msg_tofree, 2);
429 rv = __ipmi_set_timeout(&smi_msg,
431 &send_heartbeat_now);
435 wait_for_completion(&msg_wait);
437 if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB)
438 || ((send_heartbeat_now)
439 && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY)))
440 rv = __ipmi_heartbeat();
445 static int ipmi_set_timeout(int do_heartbeat)
449 mutex_lock(&ipmi_watchdog_mutex);
450 rv = _ipmi_set_timeout(do_heartbeat);
451 mutex_unlock(&ipmi_watchdog_mutex);
456 static atomic_t panic_done_count = ATOMIC_INIT(0);
458 static void panic_smi_free(struct ipmi_smi_msg *msg)
460 atomic_dec(&panic_done_count);
462 static void panic_recv_free(struct ipmi_recv_msg *msg)
464 atomic_dec(&panic_done_count);
467 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = {
468 .done = panic_smi_free
470 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = {
471 .done = panic_recv_free
474 static void panic_halt_ipmi_heartbeat(void)
476 struct kernel_ipmi_msg msg;
477 struct ipmi_system_interface_addr addr;
481 * Don't reset the timer if we have the timer turned off, that
482 * 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 atomic_add(1, &panic_done_count);
496 rv = ipmi_request_supply_msgs(watchdog_user,
497 (struct ipmi_addr *) &addr,
501 &panic_halt_heartbeat_smi_msg,
502 &panic_halt_heartbeat_recv_msg,
505 atomic_sub(1, &panic_done_count);
508 static struct ipmi_smi_msg panic_halt_smi_msg = {
509 .done = panic_smi_free
511 static struct ipmi_recv_msg panic_halt_recv_msg = {
512 .done = panic_recv_free
516 * Special call, doesn't claim any locks. This is only to be called
517 * at panic or halt time, in run-to-completion mode, when the caller
518 * is the only CPU and the only thing that will be going is these IPMI
521 static void panic_halt_ipmi_set_timeout(void)
523 int send_heartbeat_now;
526 /* Wait for the messages to be free. */
527 while (atomic_read(&panic_done_count) != 0)
528 ipmi_poll_interface(watchdog_user);
529 atomic_add(1, &panic_done_count);
530 rv = __ipmi_set_timeout(&panic_halt_smi_msg,
531 &panic_halt_recv_msg,
532 &send_heartbeat_now);
534 atomic_sub(1, &panic_done_count);
535 pr_warn(PFX "Unable to extend the watchdog timeout.");
537 if (send_heartbeat_now)
538 panic_halt_ipmi_heartbeat();
540 while (atomic_read(&panic_done_count) != 0)
541 ipmi_poll_interface(watchdog_user);
544 static int __ipmi_heartbeat(void)
546 struct kernel_ipmi_msg msg;
548 struct ipmi_system_interface_addr addr;
549 int timeout_retries = 0;
553 * Don't reset the timer if we have the timer turned off, that
554 * re-enables the watchdog.
556 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
559 atomic_set(&msg_tofree, 2);
561 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
562 addr.channel = IPMI_BMC_CHANNEL;
566 msg.cmd = IPMI_WDOG_RESET_TIMER;
569 rv = ipmi_request_supply_msgs(watchdog_user,
570 (struct ipmi_addr *) &addr,
578 pr_warn(PFX "heartbeat send failure: %d\n", rv);
582 /* Wait for the heartbeat to be sent. */
583 wait_for_completion(&msg_wait);
585 if (recv_msg.msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) {
587 if (timeout_retries > 3) {
588 pr_err(PFX ": Unable to restore the IPMI watchdog's settings, giving up.\n");
594 * The timer was not initialized, that means the BMC was
595 * probably reset and lost the watchdog information. Attempt
596 * to restore the timer's info. Note that we still hold
597 * the heartbeat lock, to keep a heartbeat from happening
598 * in this process, so must say no heartbeat to avoid a
599 * deadlock on this mutex
601 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
603 pr_err(PFX ": Unable to send the command to set the watchdog's settings, giving up.\n");
607 /* Might need a heartbeat send, go ahead and do it. */
609 } else if (recv_msg.msg.data[0] != 0) {
611 * Got an error in the heartbeat response. It was already
612 * reported in ipmi_wdog_msg_handler, but we should return
622 static int _ipmi_heartbeat(void)
629 if (ipmi_start_timer_on_heartbeat) {
630 ipmi_start_timer_on_heartbeat = 0;
631 ipmi_watchdog_state = action_val;
632 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
633 } else if (atomic_cmpxchg(&pretimeout_since_last_heartbeat, 1, 0)) {
635 * A pretimeout occurred, make sure we set the timeout.
636 * We don't want to set the action, though, we want to
637 * leave that alone (thus it can't be combined with the
640 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
642 rv = __ipmi_heartbeat();
648 static int ipmi_heartbeat(void)
652 mutex_lock(&ipmi_watchdog_mutex);
653 rv = _ipmi_heartbeat();
654 mutex_unlock(&ipmi_watchdog_mutex);
659 static struct watchdog_info ident = {
660 .options = 0, /* WDIOF_SETTIMEOUT, */
661 .firmware_version = 1,
665 static int ipmi_ioctl(struct file *file,
666 unsigned int cmd, unsigned long arg)
668 void __user *argp = (void __user *)arg;
673 case WDIOC_GETSUPPORT:
674 i = copy_to_user(argp, &ident, sizeof(ident));
675 return i ? -EFAULT : 0;
677 case WDIOC_SETTIMEOUT:
678 i = copy_from_user(&val, argp, sizeof(int));
682 return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
684 case WDIOC_GETTIMEOUT:
685 i = copy_to_user(argp, &timeout, sizeof(timeout));
690 case WDIOC_SETPRETIMEOUT:
691 i = copy_from_user(&val, argp, sizeof(int));
695 return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
697 case WDIOC_GETPRETIMEOUT:
698 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout));
703 case WDIOC_KEEPALIVE:
704 return _ipmi_heartbeat();
706 case WDIOC_SETOPTIONS:
707 i = copy_from_user(&val, argp, sizeof(int));
710 if (val & WDIOS_DISABLECARD) {
711 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
712 _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
713 ipmi_start_timer_on_heartbeat = 0;
716 if (val & WDIOS_ENABLECARD) {
717 ipmi_watchdog_state = action_val;
718 _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
722 case WDIOC_GETSTATUS:
724 i = copy_to_user(argp, &val, sizeof(val));
734 static long ipmi_unlocked_ioctl(struct file *file,
740 mutex_lock(&ipmi_watchdog_mutex);
741 ret = ipmi_ioctl(file, cmd, arg);
742 mutex_unlock(&ipmi_watchdog_mutex);
747 static ssize_t ipmi_write(struct file *file,
748 const char __user *buf,
758 /* In case it was set long ago */
761 for (i = 0; i != len; i++) {
764 if (get_user(c, buf + i))
770 rv = ipmi_heartbeat();
777 static ssize_t ipmi_read(struct file *file,
783 wait_queue_entry_t wait;
789 * Reading returns if the pretimeout has gone off, and it only does
790 * it once per pretimeout.
792 spin_lock_irq(&ipmi_read_lock);
794 if (file->f_flags & O_NONBLOCK) {
799 init_waitqueue_entry(&wait, current);
800 add_wait_queue(&read_q, &wait);
801 while (!data_to_read) {
802 set_current_state(TASK_INTERRUPTIBLE);
803 spin_unlock_irq(&ipmi_read_lock);
805 spin_lock_irq(&ipmi_read_lock);
807 remove_wait_queue(&read_q, &wait);
809 if (signal_pending(current)) {
817 spin_unlock_irq(&ipmi_read_lock);
820 if (copy_to_user(buf, &data_to_read, 1))
829 static int ipmi_open(struct inode *ino, struct file *filep)
831 switch (iminor(ino)) {
833 if (test_and_set_bit(0, &ipmi_wdog_open))
838 * Don't start the timer now, let it start on the
841 ipmi_start_timer_on_heartbeat = 1;
842 return nonseekable_open(ino, filep);
849 static __poll_t ipmi_poll(struct file *file, poll_table *wait)
853 poll_wait(file, &read_q, wait);
855 spin_lock_irq(&ipmi_read_lock);
857 mask |= (EPOLLIN | EPOLLRDNORM);
858 spin_unlock_irq(&ipmi_read_lock);
863 static int ipmi_fasync(int fd, struct file *file, int on)
867 result = fasync_helper(fd, file, on, &fasync_q);
872 static int ipmi_close(struct inode *ino, struct file *filep)
874 if (iminor(ino) == WATCHDOG_MINOR) {
875 if (expect_close == 42) {
876 mutex_lock(&ipmi_watchdog_mutex);
877 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
878 _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
879 mutex_unlock(&ipmi_watchdog_mutex);
882 "Unexpected close, not stopping watchdog!\n");
885 clear_bit(0, &ipmi_wdog_open);
893 static const struct file_operations ipmi_wdog_fops = {
894 .owner = THIS_MODULE,
898 .unlocked_ioctl = ipmi_unlocked_ioctl,
900 .release = ipmi_close,
901 .fasync = ipmi_fasync,
905 static struct miscdevice ipmi_wdog_miscdev = {
906 .minor = WATCHDOG_MINOR,
908 .fops = &ipmi_wdog_fops
911 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg,
914 if (msg->msg.cmd == IPMI_WDOG_RESET_TIMER &&
915 msg->msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP)
916 pr_info(PFX "response: The IPMI controller appears to have been reset, will attempt to reinitialize the watchdog timer\n");
917 else if (msg->msg.data[0] != 0)
918 pr_err(PFX "response: Error %x on cmd %x\n",
922 ipmi_free_recv_msg(msg);
925 static void ipmi_wdog_pretimeout_handler(void *handler_data)
927 if (preaction_val != WDOG_PRETIMEOUT_NONE) {
928 if (preop_val == WDOG_PREOP_PANIC) {
929 if (atomic_inc_and_test(&preop_panic_excl))
930 panic("Watchdog pre-timeout");
931 } else if (preop_val == WDOG_PREOP_GIVE_DATA) {
934 spin_lock_irqsave(&ipmi_read_lock, flags);
936 wake_up_interruptible(&read_q);
937 kill_fasync(&fasync_q, SIGIO, POLL_IN);
938 spin_unlock_irqrestore(&ipmi_read_lock, flags);
943 * On some machines, the heartbeat will give an error and not
944 * work unless we re-enable the timer. So do so.
946 atomic_set(&pretimeout_since_last_heartbeat, 1);
949 static void ipmi_wdog_panic_handler(void *user_data)
951 static int panic_event_handled;
954 * On a panic, if we have a panic timeout, make sure to extend
955 * the watchdog timer to a reasonable value to complete the
956 * panic, if the watchdog timer is running. Plus the
957 * pretimeout is meaningless at panic time.
959 if (watchdog_user && !panic_event_handled &&
960 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
961 /* Make sure we do this only once. */
962 panic_event_handled = 1;
964 timeout = panic_wdt_timeout;
966 panic_halt_ipmi_set_timeout();
970 static const struct ipmi_user_hndl ipmi_hndlrs = {
971 .ipmi_recv_hndl = ipmi_wdog_msg_handler,
972 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler,
973 .ipmi_panic_handler = ipmi_wdog_panic_handler
976 static void ipmi_register_watchdog(int ipmi_intf)
983 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf))
986 watchdog_ifnum = ipmi_intf;
988 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user);
990 pr_crit(PFX "Unable to register with ipmi\n");
994 rv = ipmi_get_version(watchdog_user,
996 &ipmi_version_minor);
998 pr_warn(PFX "Unable to get IPMI version, assuming 1.0\n");
999 ipmi_version_major = 1;
1000 ipmi_version_minor = 0;
1003 rv = misc_register(&ipmi_wdog_miscdev);
1005 ipmi_destroy_user(watchdog_user);
1006 watchdog_user = NULL;
1007 pr_crit(PFX "Unable to register misc device\n");
1011 if (nmi_handler_registered) {
1012 int old_pretimeout = pretimeout;
1013 int old_timeout = timeout;
1014 int old_preop_val = preop_val;
1017 * Set the pretimeout to go off in a second and give
1018 * ourselves plenty of time to stop the timer.
1020 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1021 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */
1027 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1029 pr_warn(PFX "Error starting timer to test NMI: 0x%x. The NMI pretimeout will likely not work\n",
1037 if (testing_nmi != 2) {
1038 pr_warn(PFX "IPMI NMI didn't seem to occur. The NMI pretimeout will likely not work\n");
1042 preop_val = old_preop_val;
1043 pretimeout = old_pretimeout;
1044 timeout = old_timeout;
1049 if ((start_now) && (rv == 0)) {
1050 /* Run from startup, so start the timer now. */
1051 start_now = 0; /* Disable this function after first startup. */
1052 ipmi_watchdog_state = action_val;
1053 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1054 pr_info(PFX "Starting now!\n");
1056 /* Stop the timer now. */
1057 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1058 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1062 static void ipmi_unregister_watchdog(int ipmi_intf)
1065 struct ipmi_user *loc_user = watchdog_user;
1070 if (watchdog_ifnum != ipmi_intf)
1073 /* Make sure no one can call us any more. */
1074 misc_deregister(&ipmi_wdog_miscdev);
1076 watchdog_user = NULL;
1079 * Wait to make sure the message makes it out. The lower layer has
1080 * pointers to our buffers, we want to make sure they are done before
1081 * we release our memory.
1083 while (atomic_read(&msg_tofree))
1086 mutex_lock(&ipmi_watchdog_mutex);
1088 /* Disconnect from IPMI. */
1089 rv = ipmi_destroy_user(loc_user);
1091 pr_warn(PFX "error unlinking from IPMI: %d\n", rv);
1093 /* If it comes back, restart it properly. */
1094 ipmi_start_timer_on_heartbeat = 1;
1096 mutex_unlock(&ipmi_watchdog_mutex);
1101 ipmi_nmi(unsigned int val, struct pt_regs *regs)
1104 * If we get here, it's an NMI that's not a memory or I/O
1105 * error. We can't truly tell if it's from IPMI or not
1106 * without sending a message, and sending a message is almost
1107 * impossible because of locking.
1115 /* If we are not expecting a timeout, ignore it. */
1116 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
1119 if (preaction_val != WDOG_PRETIMEOUT_NMI)
1123 * If no one else handled the NMI, we assume it was the IPMI
1126 if (preop_val == WDOG_PREOP_PANIC) {
1127 /* On some machines, the heartbeat will give
1128 an error and not work unless we re-enable
1129 the timer. So do so. */
1130 atomic_set(&pretimeout_since_last_heartbeat, 1);
1131 if (atomic_inc_and_test(&preop_panic_excl))
1132 nmi_panic(regs, PFX "pre-timeout");
1139 static int wdog_reboot_handler(struct notifier_block *this,
1143 static int reboot_event_handled;
1145 if ((watchdog_user) && (!reboot_event_handled)) {
1146 /* Make sure we only do this once. */
1147 reboot_event_handled = 1;
1149 if (code == SYS_POWER_OFF || code == SYS_HALT) {
1150 /* Disable the WDT if we are shutting down. */
1151 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1152 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1153 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1154 /* Set a long timer to let the reboot happen or
1155 reset if it hangs, but only if the watchdog
1156 timer was already running. */
1160 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1161 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1167 static struct notifier_block wdog_reboot_notifier = {
1168 .notifier_call = wdog_reboot_handler,
1173 static void ipmi_new_smi(int if_num, struct device *device)
1175 ipmi_register_watchdog(if_num);
1178 static void ipmi_smi_gone(int if_num)
1180 ipmi_unregister_watchdog(if_num);
1183 static struct ipmi_smi_watcher smi_watcher = {
1184 .owner = THIS_MODULE,
1185 .new_smi = ipmi_new_smi,
1186 .smi_gone = ipmi_smi_gone
1189 static int action_op(const char *inval, char *outval)
1192 strcpy(outval, action);
1197 if (strcmp(inval, "reset") == 0)
1198 action_val = WDOG_TIMEOUT_RESET;
1199 else if (strcmp(inval, "none") == 0)
1200 action_val = WDOG_TIMEOUT_NONE;
1201 else if (strcmp(inval, "power_cycle") == 0)
1202 action_val = WDOG_TIMEOUT_POWER_CYCLE;
1203 else if (strcmp(inval, "power_off") == 0)
1204 action_val = WDOG_TIMEOUT_POWER_DOWN;
1207 strcpy(action, inval);
1211 static int preaction_op(const char *inval, char *outval)
1214 strcpy(outval, preaction);
1219 if (strcmp(inval, "pre_none") == 0)
1220 preaction_val = WDOG_PRETIMEOUT_NONE;
1221 else if (strcmp(inval, "pre_smi") == 0)
1222 preaction_val = WDOG_PRETIMEOUT_SMI;
1224 else if (strcmp(inval, "pre_nmi") == 0)
1225 preaction_val = WDOG_PRETIMEOUT_NMI;
1227 else if (strcmp(inval, "pre_int") == 0)
1228 preaction_val = WDOG_PRETIMEOUT_MSG_INT;
1231 strcpy(preaction, inval);
1235 static int preop_op(const char *inval, char *outval)
1238 strcpy(outval, preop);
1243 if (strcmp(inval, "preop_none") == 0)
1244 preop_val = WDOG_PREOP_NONE;
1245 else if (strcmp(inval, "preop_panic") == 0)
1246 preop_val = WDOG_PREOP_PANIC;
1247 else if (strcmp(inval, "preop_give_data") == 0)
1248 preop_val = WDOG_PREOP_GIVE_DATA;
1251 strcpy(preop, inval);
1255 static void check_parms(void)
1261 if (preaction_val == WDOG_PRETIMEOUT_NMI) {
1263 if (preop_val == WDOG_PREOP_GIVE_DATA) {
1264 pr_warn(PFX "Pretimeout op is to give data but NMI pretimeout is enabled, setting pretimeout op to none\n");
1265 preop_op("preop_none", NULL);
1269 if (do_nmi && !nmi_handler_registered) {
1270 rv = register_nmi_handler(NMI_UNKNOWN, ipmi_nmi, 0,
1273 pr_warn(PFX "Can't register nmi handler\n");
1276 nmi_handler_registered = 1;
1277 } else if (!do_nmi && nmi_handler_registered) {
1278 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1279 nmi_handler_registered = 0;
1284 static int __init ipmi_wdog_init(void)
1288 if (action_op(action, NULL)) {
1289 action_op("reset", NULL);
1290 pr_info(PFX "Unknown action '%s', defaulting to reset\n",
1294 if (preaction_op(preaction, NULL)) {
1295 preaction_op("pre_none", NULL);
1296 pr_info(PFX "Unknown preaction '%s', defaulting to none\n",
1300 if (preop_op(preop, NULL)) {
1301 preop_op("preop_none", NULL);
1302 pr_info(PFX "Unknown preop '%s', defaulting to none\n", preop);
1307 register_reboot_notifier(&wdog_reboot_notifier);
1309 rv = ipmi_smi_watcher_register(&smi_watcher);
1312 if (nmi_handler_registered)
1313 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1315 unregister_reboot_notifier(&wdog_reboot_notifier);
1316 pr_warn(PFX "can't register smi watcher\n");
1320 pr_info(PFX "driver initialized\n");
1325 static void __exit ipmi_wdog_exit(void)
1327 ipmi_smi_watcher_unregister(&smi_watcher);
1328 ipmi_unregister_watchdog(watchdog_ifnum);
1331 if (nmi_handler_registered)
1332 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1335 unregister_reboot_notifier(&wdog_reboot_notifier);
1337 module_exit(ipmi_wdog_exit);
1338 module_init(ipmi_wdog_init);
1339 MODULE_LICENSE("GPL");
1340 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
1341 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface.");