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 #define pr_fmt(fmt) "IPMI Watchdog: " fmt
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/ipmi.h>
19 #include <linux/ipmi_smi.h>
20 #include <linux/mutex.h>
21 #include <linux/watchdog.h>
22 #include <linux/miscdevice.h>
23 #include <linux/init.h>
24 #include <linux/completion.h>
25 #include <linux/kdebug.h>
26 #include <linux/rwsem.h>
27 #include <linux/errno.h>
28 #include <linux/uaccess.h>
29 #include <linux/notifier.h>
30 #include <linux/nmi.h>
31 #include <linux/reboot.h>
32 #include <linux/wait.h>
33 #include <linux/poll.h>
34 #include <linux/string.h>
35 #include <linux/ctype.h>
36 #include <linux/delay.h>
37 #include <linux/atomic.h>
38 #include <linux/sched/signal.h>
42 * This is ugly, but I've determined that x86 is the only architecture
43 * that can reasonably support the IPMI NMI watchdog timeout at this
44 * time. If another architecture adds this capability somehow, it
45 * will have to be a somewhat different mechanism and I have no idea
46 * how it will work. So in the unlikely event that another
47 * architecture supports this, we can figure out a good generic
48 * mechanism for it at that time.
50 #include <asm/kdebug.h>
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);
241 len = strlen(buffer);
242 buffer[len++] = '\n';
249 static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp)
251 int rv = param_set_int(val, kp);
254 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum))
257 ipmi_unregister_watchdog(watchdog_ifnum);
258 ipmi_register_watchdog(ifnum_to_use);
262 static const struct kernel_param_ops param_ops_wdog_ifnum = {
263 .set = set_param_wdog_ifnum,
264 .get = param_get_int,
267 #define param_check_wdog_ifnum param_check_int
269 static const struct kernel_param_ops param_ops_str = {
270 .set = set_param_str,
271 .get = get_param_str,
274 module_param(ifnum_to_use, wdog_ifnum, 0644);
275 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog "
276 "timer. Setting to -1 defaults to the first registered "
279 module_param(timeout, timeout, 0644);
280 MODULE_PARM_DESC(timeout, "Timeout value in seconds.");
282 module_param(pretimeout, timeout, 0644);
283 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds.");
285 module_param(panic_wdt_timeout, timeout, 0644);
286 MODULE_PARM_DESC(panic_wdt_timeout, "Timeout value on kernel panic in seconds.");
288 module_param_cb(action, ¶m_ops_str, action_op, 0644);
289 MODULE_PARM_DESC(action, "Timeout action. One of: "
290 "reset, none, power_cycle, power_off.");
292 module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644);
293 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: "
294 "pre_none, pre_smi, pre_nmi, pre_int.");
296 module_param_cb(preop, ¶m_ops_str, preop_op, 0644);
297 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: "
298 "preop_none, preop_panic, preop_give_data.");
300 module_param(start_now, int, 0444);
301 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as"
302 "soon as the driver is loaded.");
304 module_param(nowayout, bool, 0644);
305 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started "
306 "(default=CONFIG_WATCHDOG_NOWAYOUT)");
308 /* Default state of the timer. */
309 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
311 /* Is someone using the watchdog? Only one user is allowed. */
312 static unsigned long ipmi_wdog_open;
315 * If set to 1, the heartbeat command will set the state to reset and
316 * start the timer. The timer doesn't normally run when the driver is
317 * first opened until the heartbeat is set the first time, this
318 * variable is used to accomplish this.
320 static int ipmi_start_timer_on_heartbeat;
322 /* IPMI version of the BMC. */
323 static unsigned char ipmi_version_major;
324 static unsigned char ipmi_version_minor;
326 /* If a pretimeout occurs, this is used to allow only one panic to happen. */
327 static atomic_t preop_panic_excl = ATOMIC_INIT(-1);
330 static int testing_nmi;
331 static int nmi_handler_registered;
334 static int __ipmi_heartbeat(void);
337 * We use a mutex to make sure that only one thing can send a set a
338 * message at one time. The mutex is claimed when a message is sent
339 * and freed when both the send and receive messages are free.
341 static atomic_t msg_tofree = ATOMIC_INIT(0);
342 static DECLARE_COMPLETION(msg_wait);
343 static void msg_free_smi(struct ipmi_smi_msg *msg)
345 if (atomic_dec_and_test(&msg_tofree)) {
346 if (!oops_in_progress)
350 static void msg_free_recv(struct ipmi_recv_msg *msg)
352 if (atomic_dec_and_test(&msg_tofree)) {
353 if (!oops_in_progress)
357 static struct ipmi_smi_msg smi_msg = {
360 static struct ipmi_recv_msg recv_msg = {
361 .done = msg_free_recv
364 static int __ipmi_set_timeout(struct ipmi_smi_msg *smi_msg,
365 struct ipmi_recv_msg *recv_msg,
366 int *send_heartbeat_now)
368 struct kernel_ipmi_msg msg;
369 unsigned char data[6];
371 struct ipmi_system_interface_addr addr;
376 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
378 if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
379 if ((ipmi_version_major > 1) ||
380 ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) {
381 /* This is an IPMI 1.5-only feature. */
382 data[0] |= WDOG_DONT_STOP_ON_SET;
385 * In ipmi 1.0, setting the timer stops the watchdog, we
386 * need to start it back up again.
393 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state);
394 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) {
395 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val);
396 data[2] = pretimeout;
398 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE);
399 data[2] = 0; /* No pretimeout. */
402 WDOG_SET_TIMEOUT(data[4], data[5], timeout);
404 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
405 addr.channel = IPMI_BMC_CHANNEL;
409 msg.cmd = IPMI_WDOG_SET_TIMER;
411 msg.data_len = sizeof(data);
412 rv = ipmi_request_supply_msgs(watchdog_user,
413 (struct ipmi_addr *) &addr,
421 pr_warn("set timeout error: %d\n", rv);
422 else if (send_heartbeat_now)
423 *send_heartbeat_now = hbnow;
428 static int _ipmi_set_timeout(int do_heartbeat)
430 int send_heartbeat_now;
436 atomic_set(&msg_tofree, 2);
438 rv = __ipmi_set_timeout(&smi_msg,
440 &send_heartbeat_now);
442 atomic_set(&msg_tofree, 0);
446 wait_for_completion(&msg_wait);
448 if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB)
449 || ((send_heartbeat_now)
450 && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY)))
451 rv = __ipmi_heartbeat();
456 static int ipmi_set_timeout(int do_heartbeat)
460 mutex_lock(&ipmi_watchdog_mutex);
461 rv = _ipmi_set_timeout(do_heartbeat);
462 mutex_unlock(&ipmi_watchdog_mutex);
467 static atomic_t panic_done_count = ATOMIC_INIT(0);
469 static void panic_smi_free(struct ipmi_smi_msg *msg)
471 atomic_dec(&panic_done_count);
473 static void panic_recv_free(struct ipmi_recv_msg *msg)
475 atomic_dec(&panic_done_count);
478 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = {
479 .done = panic_smi_free
481 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = {
482 .done = panic_recv_free
485 static void panic_halt_ipmi_heartbeat(void)
487 struct kernel_ipmi_msg msg;
488 struct ipmi_system_interface_addr addr;
492 * Don't reset the timer if we have the timer turned off, that
493 * re-enables the watchdog.
495 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
498 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
499 addr.channel = IPMI_BMC_CHANNEL;
503 msg.cmd = IPMI_WDOG_RESET_TIMER;
506 atomic_add(1, &panic_done_count);
507 rv = ipmi_request_supply_msgs(watchdog_user,
508 (struct ipmi_addr *) &addr,
512 &panic_halt_heartbeat_smi_msg,
513 &panic_halt_heartbeat_recv_msg,
516 atomic_sub(1, &panic_done_count);
519 static struct ipmi_smi_msg panic_halt_smi_msg = {
520 .done = panic_smi_free
522 static struct ipmi_recv_msg panic_halt_recv_msg = {
523 .done = panic_recv_free
527 * Special call, doesn't claim any locks. This is only to be called
528 * at panic or halt time, in run-to-completion mode, when the caller
529 * is the only CPU and the only thing that will be going is these IPMI
532 static void panic_halt_ipmi_set_timeout(void)
534 int send_heartbeat_now;
537 /* Wait for the messages to be free. */
538 while (atomic_read(&panic_done_count) != 0)
539 ipmi_poll_interface(watchdog_user);
540 atomic_add(1, &panic_done_count);
541 rv = __ipmi_set_timeout(&panic_halt_smi_msg,
542 &panic_halt_recv_msg,
543 &send_heartbeat_now);
545 atomic_sub(1, &panic_done_count);
546 pr_warn("Unable to extend the watchdog timeout\n");
548 if (send_heartbeat_now)
549 panic_halt_ipmi_heartbeat();
551 while (atomic_read(&panic_done_count) != 0)
552 ipmi_poll_interface(watchdog_user);
555 static int __ipmi_heartbeat(void)
557 struct kernel_ipmi_msg msg;
559 struct ipmi_system_interface_addr addr;
560 int timeout_retries = 0;
564 * Don't reset the timer if we have the timer turned off, that
565 * re-enables the watchdog.
567 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
570 atomic_set(&msg_tofree, 2);
572 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
573 addr.channel = IPMI_BMC_CHANNEL;
577 msg.cmd = IPMI_WDOG_RESET_TIMER;
580 rv = ipmi_request_supply_msgs(watchdog_user,
581 (struct ipmi_addr *) &addr,
589 atomic_set(&msg_tofree, 0);
590 pr_warn("heartbeat send failure: %d\n", rv);
594 /* Wait for the heartbeat to be sent. */
595 wait_for_completion(&msg_wait);
597 if (recv_msg.msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP) {
599 if (timeout_retries > 3) {
600 pr_err("Unable to restore the IPMI watchdog's settings, giving up\n");
606 * The timer was not initialized, that means the BMC was
607 * probably reset and lost the watchdog information. Attempt
608 * to restore the timer's info. Note that we still hold
609 * the heartbeat lock, to keep a heartbeat from happening
610 * in this process, so must say no heartbeat to avoid a
611 * deadlock on this mutex
613 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
615 pr_err("Unable to send the command to set the watchdog's settings, giving up\n");
619 /* Might need a heartbeat send, go ahead and do it. */
621 } else if (recv_msg.msg.data[0] != 0) {
623 * Got an error in the heartbeat response. It was already
624 * reported in ipmi_wdog_msg_handler, but we should return
634 static int _ipmi_heartbeat(void)
641 if (ipmi_start_timer_on_heartbeat) {
642 ipmi_start_timer_on_heartbeat = 0;
643 ipmi_watchdog_state = action_val;
644 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
645 } else if (atomic_cmpxchg(&pretimeout_since_last_heartbeat, 1, 0)) {
647 * A pretimeout occurred, make sure we set the timeout.
648 * We don't want to set the action, though, we want to
649 * leave that alone (thus it can't be combined with the
652 rv = _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
654 rv = __ipmi_heartbeat();
660 static int ipmi_heartbeat(void)
664 mutex_lock(&ipmi_watchdog_mutex);
665 rv = _ipmi_heartbeat();
666 mutex_unlock(&ipmi_watchdog_mutex);
671 static struct watchdog_info ident = {
672 .options = 0, /* WDIOF_SETTIMEOUT, */
673 .firmware_version = 1,
677 static int ipmi_ioctl(struct file *file,
678 unsigned int cmd, unsigned long arg)
680 void __user *argp = (void __user *)arg;
685 case WDIOC_GETSUPPORT:
686 i = copy_to_user(argp, &ident, sizeof(ident));
687 return i ? -EFAULT : 0;
689 case WDIOC_SETTIMEOUT:
690 i = copy_from_user(&val, argp, sizeof(int));
694 return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
696 case WDIOC_GETTIMEOUT:
697 i = copy_to_user(argp, &timeout, sizeof(timeout));
702 case WDIOC_SETPRETIMEOUT:
703 i = copy_from_user(&val, argp, sizeof(int));
707 return _ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
709 case WDIOC_GETPRETIMEOUT:
710 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout));
715 case WDIOC_KEEPALIVE:
716 return _ipmi_heartbeat();
718 case WDIOC_SETOPTIONS:
719 i = copy_from_user(&val, argp, sizeof(int));
722 if (val & WDIOS_DISABLECARD) {
723 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
724 _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
725 ipmi_start_timer_on_heartbeat = 0;
728 if (val & WDIOS_ENABLECARD) {
729 ipmi_watchdog_state = action_val;
730 _ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
734 case WDIOC_GETSTATUS:
736 i = copy_to_user(argp, &val, sizeof(val));
746 static long ipmi_unlocked_ioctl(struct file *file,
752 mutex_lock(&ipmi_watchdog_mutex);
753 ret = ipmi_ioctl(file, cmd, arg);
754 mutex_unlock(&ipmi_watchdog_mutex);
759 static ssize_t ipmi_write(struct file *file,
760 const char __user *buf,
770 /* In case it was set long ago */
773 for (i = 0; i != len; i++) {
776 if (get_user(c, buf + i))
782 rv = ipmi_heartbeat();
789 static ssize_t ipmi_read(struct file *file,
795 wait_queue_entry_t wait;
801 * Reading returns if the pretimeout has gone off, and it only does
802 * it once per pretimeout.
804 spin_lock_irq(&ipmi_read_lock);
806 if (file->f_flags & O_NONBLOCK) {
811 init_waitqueue_entry(&wait, current);
812 add_wait_queue(&read_q, &wait);
813 while (!data_to_read) {
814 set_current_state(TASK_INTERRUPTIBLE);
815 spin_unlock_irq(&ipmi_read_lock);
817 spin_lock_irq(&ipmi_read_lock);
819 remove_wait_queue(&read_q, &wait);
821 if (signal_pending(current)) {
829 spin_unlock_irq(&ipmi_read_lock);
832 if (copy_to_user(buf, &data_to_read, 1))
841 static int ipmi_open(struct inode *ino, struct file *filep)
843 switch (iminor(ino)) {
845 if (test_and_set_bit(0, &ipmi_wdog_open))
850 * Don't start the timer now, let it start on the
853 ipmi_start_timer_on_heartbeat = 1;
854 return stream_open(ino, filep);
861 static __poll_t ipmi_poll(struct file *file, poll_table *wait)
865 poll_wait(file, &read_q, wait);
867 spin_lock_irq(&ipmi_read_lock);
869 mask |= (EPOLLIN | EPOLLRDNORM);
870 spin_unlock_irq(&ipmi_read_lock);
875 static int ipmi_fasync(int fd, struct file *file, int on)
879 result = fasync_helper(fd, file, on, &fasync_q);
884 static int ipmi_close(struct inode *ino, struct file *filep)
886 if (iminor(ino) == WATCHDOG_MINOR) {
887 if (expect_close == 42) {
888 mutex_lock(&ipmi_watchdog_mutex);
889 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
890 _ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
891 mutex_unlock(&ipmi_watchdog_mutex);
893 pr_crit("Unexpected close, not stopping watchdog!\n");
896 clear_bit(0, &ipmi_wdog_open);
904 static const struct file_operations ipmi_wdog_fops = {
905 .owner = THIS_MODULE,
909 .unlocked_ioctl = ipmi_unlocked_ioctl,
910 .compat_ioctl = compat_ptr_ioctl,
912 .release = ipmi_close,
913 .fasync = ipmi_fasync,
917 static struct miscdevice ipmi_wdog_miscdev = {
918 .minor = WATCHDOG_MINOR,
920 .fops = &ipmi_wdog_fops
923 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg,
926 if (msg->msg.cmd == IPMI_WDOG_RESET_TIMER &&
927 msg->msg.data[0] == IPMI_WDOG_TIMER_NOT_INIT_RESP)
928 pr_info("response: The IPMI controller appears to have been reset, will attempt to reinitialize the watchdog timer\n");
929 else if (msg->msg.data[0] != 0)
930 pr_err("response: Error %x on cmd %x\n",
934 ipmi_free_recv_msg(msg);
937 static void ipmi_wdog_pretimeout_handler(void *handler_data)
939 if (preaction_val != WDOG_PRETIMEOUT_NONE) {
940 if (preop_val == WDOG_PREOP_PANIC) {
941 if (atomic_inc_and_test(&preop_panic_excl))
942 panic("Watchdog pre-timeout");
943 } else if (preop_val == WDOG_PREOP_GIVE_DATA) {
946 spin_lock_irqsave(&ipmi_read_lock, flags);
948 wake_up_interruptible(&read_q);
949 kill_fasync(&fasync_q, SIGIO, POLL_IN);
950 spin_unlock_irqrestore(&ipmi_read_lock, flags);
955 * On some machines, the heartbeat will give an error and not
956 * work unless we re-enable the timer. So do so.
958 atomic_set(&pretimeout_since_last_heartbeat, 1);
961 static void ipmi_wdog_panic_handler(void *user_data)
963 static int panic_event_handled;
966 * On a panic, if we have a panic timeout, make sure to extend
967 * the watchdog timer to a reasonable value to complete the
968 * panic, if the watchdog timer is running. Plus the
969 * pretimeout is meaningless at panic time.
971 if (watchdog_user && !panic_event_handled &&
972 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
973 /* Make sure we do this only once. */
974 panic_event_handled = 1;
976 timeout = panic_wdt_timeout;
978 panic_halt_ipmi_set_timeout();
982 static const struct ipmi_user_hndl ipmi_hndlrs = {
983 .ipmi_recv_hndl = ipmi_wdog_msg_handler,
984 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler,
985 .ipmi_panic_handler = ipmi_wdog_panic_handler
988 static void ipmi_register_watchdog(int ipmi_intf)
995 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf))
998 watchdog_ifnum = ipmi_intf;
1000 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user);
1002 pr_crit("Unable to register with ipmi\n");
1006 rv = ipmi_get_version(watchdog_user,
1007 &ipmi_version_major,
1008 &ipmi_version_minor);
1010 pr_warn("Unable to get IPMI version, assuming 1.0\n");
1011 ipmi_version_major = 1;
1012 ipmi_version_minor = 0;
1015 rv = misc_register(&ipmi_wdog_miscdev);
1017 ipmi_destroy_user(watchdog_user);
1018 watchdog_user = NULL;
1019 pr_crit("Unable to register misc device\n");
1023 if (nmi_handler_registered) {
1024 int old_pretimeout = pretimeout;
1025 int old_timeout = timeout;
1026 int old_preop_val = preop_val;
1029 * Set the pretimeout to go off in a second and give
1030 * ourselves plenty of time to stop the timer.
1032 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1033 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */
1039 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1041 pr_warn("Error starting timer to test NMI: 0x%x. The NMI pretimeout will likely not work\n",
1049 if (testing_nmi != 2) {
1050 pr_warn("IPMI NMI didn't seem to occur. The NMI pretimeout will likely not work\n");
1054 preop_val = old_preop_val;
1055 pretimeout = old_pretimeout;
1056 timeout = old_timeout;
1061 if ((start_now) && (rv == 0)) {
1062 /* Run from startup, so start the timer now. */
1063 start_now = 0; /* Disable this function after first startup. */
1064 ipmi_watchdog_state = action_val;
1065 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1066 pr_info("Starting now!\n");
1068 /* Stop the timer now. */
1069 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1070 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1074 static void ipmi_unregister_watchdog(int ipmi_intf)
1077 struct ipmi_user *loc_user = watchdog_user;
1082 if (watchdog_ifnum != ipmi_intf)
1085 /* Make sure no one can call us any more. */
1086 misc_deregister(&ipmi_wdog_miscdev);
1088 watchdog_user = NULL;
1091 * Wait to make sure the message makes it out. The lower layer has
1092 * pointers to our buffers, we want to make sure they are done before
1093 * we release our memory.
1095 while (atomic_read(&msg_tofree))
1098 mutex_lock(&ipmi_watchdog_mutex);
1100 /* Disconnect from IPMI. */
1101 rv = ipmi_destroy_user(loc_user);
1103 pr_warn("error unlinking from IPMI: %d\n", rv);
1105 /* If it comes back, restart it properly. */
1106 ipmi_start_timer_on_heartbeat = 1;
1108 mutex_unlock(&ipmi_watchdog_mutex);
1113 ipmi_nmi(unsigned int val, struct pt_regs *regs)
1116 * If we get here, it's an NMI that's not a memory or I/O
1117 * error. We can't truly tell if it's from IPMI or not
1118 * without sending a message, and sending a message is almost
1119 * impossible because of locking.
1127 /* If we are not expecting a timeout, ignore it. */
1128 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
1131 if (preaction_val != WDOG_PRETIMEOUT_NMI)
1135 * If no one else handled the NMI, we assume it was the IPMI
1138 if (preop_val == WDOG_PREOP_PANIC) {
1139 /* On some machines, the heartbeat will give
1140 an error and not work unless we re-enable
1141 the timer. So do so. */
1142 atomic_set(&pretimeout_since_last_heartbeat, 1);
1143 if (atomic_inc_and_test(&preop_panic_excl))
1144 nmi_panic(regs, "pre-timeout");
1151 static int wdog_reboot_handler(struct notifier_block *this,
1155 static int reboot_event_handled;
1157 if ((watchdog_user) && (!reboot_event_handled)) {
1158 /* Make sure we only do this once. */
1159 reboot_event_handled = 1;
1161 if (code == SYS_POWER_OFF || code == SYS_HALT) {
1162 /* Disable the WDT if we are shutting down. */
1163 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1164 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1165 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1166 /* Set a long timer to let the reboot happen or
1167 reset if it hangs, but only if the watchdog
1168 timer was already running. */
1172 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1173 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1179 static struct notifier_block wdog_reboot_notifier = {
1180 .notifier_call = wdog_reboot_handler,
1185 static void ipmi_new_smi(int if_num, struct device *device)
1187 ipmi_register_watchdog(if_num);
1190 static void ipmi_smi_gone(int if_num)
1192 ipmi_unregister_watchdog(if_num);
1195 static struct ipmi_smi_watcher smi_watcher = {
1196 .owner = THIS_MODULE,
1197 .new_smi = ipmi_new_smi,
1198 .smi_gone = ipmi_smi_gone
1201 static int action_op(const char *inval, char *outval)
1204 strcpy(outval, action);
1209 if (strcmp(inval, "reset") == 0)
1210 action_val = WDOG_TIMEOUT_RESET;
1211 else if (strcmp(inval, "none") == 0)
1212 action_val = WDOG_TIMEOUT_NONE;
1213 else if (strcmp(inval, "power_cycle") == 0)
1214 action_val = WDOG_TIMEOUT_POWER_CYCLE;
1215 else if (strcmp(inval, "power_off") == 0)
1216 action_val = WDOG_TIMEOUT_POWER_DOWN;
1219 strcpy(action, inval);
1223 static int preaction_op(const char *inval, char *outval)
1226 strcpy(outval, preaction);
1231 if (strcmp(inval, "pre_none") == 0)
1232 preaction_val = WDOG_PRETIMEOUT_NONE;
1233 else if (strcmp(inval, "pre_smi") == 0)
1234 preaction_val = WDOG_PRETIMEOUT_SMI;
1236 else if (strcmp(inval, "pre_nmi") == 0)
1237 preaction_val = WDOG_PRETIMEOUT_NMI;
1239 else if (strcmp(inval, "pre_int") == 0)
1240 preaction_val = WDOG_PRETIMEOUT_MSG_INT;
1243 strcpy(preaction, inval);
1247 static int preop_op(const char *inval, char *outval)
1250 strcpy(outval, preop);
1255 if (strcmp(inval, "preop_none") == 0)
1256 preop_val = WDOG_PREOP_NONE;
1257 else if (strcmp(inval, "preop_panic") == 0)
1258 preop_val = WDOG_PREOP_PANIC;
1259 else if (strcmp(inval, "preop_give_data") == 0)
1260 preop_val = WDOG_PREOP_GIVE_DATA;
1263 strcpy(preop, inval);
1267 static void check_parms(void)
1273 if (preaction_val == WDOG_PRETIMEOUT_NMI) {
1275 if (preop_val == WDOG_PREOP_GIVE_DATA) {
1276 pr_warn("Pretimeout op is to give data but NMI pretimeout is enabled, setting pretimeout op to none\n");
1277 preop_op("preop_none", NULL);
1281 if (do_nmi && !nmi_handler_registered) {
1282 rv = register_nmi_handler(NMI_UNKNOWN, ipmi_nmi, 0,
1285 pr_warn("Can't register nmi handler\n");
1288 nmi_handler_registered = 1;
1289 } else if (!do_nmi && nmi_handler_registered) {
1290 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1291 nmi_handler_registered = 0;
1296 static int __init ipmi_wdog_init(void)
1300 if (action_op(action, NULL)) {
1301 action_op("reset", NULL);
1302 pr_info("Unknown action '%s', defaulting to reset\n", action);
1305 if (preaction_op(preaction, NULL)) {
1306 preaction_op("pre_none", NULL);
1307 pr_info("Unknown preaction '%s', defaulting to none\n",
1311 if (preop_op(preop, NULL)) {
1312 preop_op("preop_none", NULL);
1313 pr_info("Unknown preop '%s', defaulting to none\n", preop);
1318 register_reboot_notifier(&wdog_reboot_notifier);
1320 rv = ipmi_smi_watcher_register(&smi_watcher);
1323 if (nmi_handler_registered)
1324 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1326 unregister_reboot_notifier(&wdog_reboot_notifier);
1327 pr_warn("can't register smi watcher\n");
1331 pr_info("driver initialized\n");
1336 static void __exit ipmi_wdog_exit(void)
1338 ipmi_smi_watcher_unregister(&smi_watcher);
1339 ipmi_unregister_watchdog(watchdog_ifnum);
1342 if (nmi_handler_registered)
1343 unregister_nmi_handler(NMI_UNKNOWN, "ipmi");
1346 unregister_reboot_notifier(&wdog_reboot_notifier);
1348 module_exit(ipmi_wdog_exit);
1349 module_init(ipmi_wdog_init);
1350 MODULE_LICENSE("GPL");
1351 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
1352 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface.");