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 <asm/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>
71 #define PFX "IPMI Watchdog: "
74 * The IPMI command/response information for the watchdog timer.
77 /* values for byte 1 of the set command, byte 2 of the get response. */
78 #define WDOG_DONT_LOG (1 << 7)
79 #define WDOG_DONT_STOP_ON_SET (1 << 6)
80 #define WDOG_SET_TIMER_USE(byte, use) \
81 byte = ((byte) & 0xf8) | ((use) & 0x7)
82 #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7)
83 #define WDOG_TIMER_USE_BIOS_FRB2 1
84 #define WDOG_TIMER_USE_BIOS_POST 2
85 #define WDOG_TIMER_USE_OS_LOAD 3
86 #define WDOG_TIMER_USE_SMS_OS 4
87 #define WDOG_TIMER_USE_OEM 5
89 /* values for byte 2 of the set command, byte 3 of the get response. */
90 #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \
91 byte = ((byte) & 0x8f) | (((use) & 0x7) << 4)
92 #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7)
93 #define WDOG_PRETIMEOUT_NONE 0
94 #define WDOG_PRETIMEOUT_SMI 1
95 #define WDOG_PRETIMEOUT_NMI 2
96 #define WDOG_PRETIMEOUT_MSG_INT 3
98 /* Operations that can be performed on a pretimout. */
99 #define WDOG_PREOP_NONE 0
100 #define WDOG_PREOP_PANIC 1
101 /* Cause data to be available to read. Doesn't work in NMI mode. */
102 #define WDOG_PREOP_GIVE_DATA 2
104 /* Actions to perform on a full timeout. */
105 #define WDOG_SET_TIMEOUT_ACT(byte, use) \
106 byte = ((byte) & 0xf8) | ((use) & 0x7)
107 #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7)
108 #define WDOG_TIMEOUT_NONE 0
109 #define WDOG_TIMEOUT_RESET 1
110 #define WDOG_TIMEOUT_POWER_DOWN 2
111 #define WDOG_TIMEOUT_POWER_CYCLE 3
114 * Byte 3 of the get command, byte 4 of the get response is the
115 * pre-timeout in seconds.
118 /* Bits for setting byte 4 of the set command, byte 5 of the get response. */
119 #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1)
120 #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2)
121 #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3)
122 #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4)
123 #define WDOG_EXPIRE_CLEAR_OEM (1 << 5)
126 * Setting/getting the watchdog timer value. This is for bytes 5 and
127 * 6 (the timeout time) of the set command, and bytes 6 and 7 (the
128 * timeout time) and 8 and 9 (the current countdown value) of the
129 * response. The timeout value is given in seconds (in the command it
130 * is 100ms intervals).
132 #define WDOG_SET_TIMEOUT(byte1, byte2, val) \
133 (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8)
134 #define WDOG_GET_TIMEOUT(byte1, byte2) \
135 (((byte1) | ((byte2) << 8)) / 10)
137 #define IPMI_WDOG_RESET_TIMER 0x22
138 #define IPMI_WDOG_SET_TIMER 0x24
139 #define IPMI_WDOG_GET_TIMER 0x25
141 /* These are here until the real ones get into the watchdog.h interface. */
142 #ifndef WDIOC_GETTIMEOUT
143 #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int)
145 #ifndef WDIOC_SET_PRETIMEOUT
146 #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int)
148 #ifndef WDIOC_GET_PRETIMEOUT
149 #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int)
152 static DEFINE_MUTEX(ipmi_watchdog_mutex);
153 static int nowayout = WATCHDOG_NOWAYOUT;
155 static ipmi_user_t watchdog_user;
156 static int watchdog_ifnum;
158 /* Default the timeout to 10 seconds. */
159 static int timeout = 10;
161 /* The pre-timeout is disabled by default. */
162 static int pretimeout;
164 /* Default action is to reset the board on a timeout. */
165 static unsigned char action_val = WDOG_TIMEOUT_RESET;
167 static char action[16] = "reset";
169 static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE;
171 static char preaction[16] = "pre_none";
173 static unsigned char preop_val = WDOG_PREOP_NONE;
175 static char preop[16] = "preop_none";
176 static DEFINE_SPINLOCK(ipmi_read_lock);
177 static char data_to_read;
178 static DECLARE_WAIT_QUEUE_HEAD(read_q);
179 static struct fasync_struct *fasync_q;
180 static char pretimeout_since_last_heartbeat;
181 static char expect_close;
183 static int ifnum_to_use = -1;
185 /* Parameters to ipmi_set_timeout */
186 #define IPMI_SET_TIMEOUT_NO_HB 0
187 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
188 #define IPMI_SET_TIMEOUT_FORCE_HB 2
190 static int ipmi_set_timeout(int do_heartbeat);
191 static void ipmi_register_watchdog(int ipmi_intf);
192 static void ipmi_unregister_watchdog(int ipmi_intf);
195 * If true, the driver will start running as soon as it is configured
198 static int start_now;
200 static int set_param_timeout(const char *val, const struct kernel_param *kp)
208 l = simple_strtoul(val, &endp, 0);
212 *((int *)kp->arg) = l;
214 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
219 static struct kernel_param_ops param_ops_timeout = {
220 .set = set_param_timeout,
221 .get = param_get_int,
223 #define param_check_timeout param_check_int
225 typedef int (*action_fn)(const char *intval, char *outval);
227 static int action_op(const char *inval, char *outval);
228 static int preaction_op(const char *inval, char *outval);
229 static int preop_op(const char *inval, char *outval);
230 static void check_parms(void);
232 static int set_param_str(const char *val, const struct kernel_param *kp)
234 action_fn fn = (action_fn) kp->arg;
239 strncpy(valcp, val, 16);
250 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
256 static int get_param_str(char *buffer, const struct kernel_param *kp)
258 action_fn fn = (action_fn) kp->arg;
261 rv = fn(NULL, buffer);
264 return strlen(buffer);
268 static int set_param_wdog_ifnum(const char *val, const struct kernel_param *kp)
270 int rv = param_set_int(val, kp);
273 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum))
276 ipmi_unregister_watchdog(watchdog_ifnum);
277 ipmi_register_watchdog(ifnum_to_use);
281 static struct kernel_param_ops param_ops_wdog_ifnum = {
282 .set = set_param_wdog_ifnum,
283 .get = param_get_int,
286 #define param_check_wdog_ifnum param_check_int
288 static struct kernel_param_ops param_ops_str = {
289 .set = set_param_str,
290 .get = get_param_str,
293 module_param(ifnum_to_use, wdog_ifnum, 0644);
294 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog "
295 "timer. Setting to -1 defaults to the first registered "
298 module_param(timeout, timeout, 0644);
299 MODULE_PARM_DESC(timeout, "Timeout value in seconds.");
301 module_param(pretimeout, timeout, 0644);
302 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds.");
304 module_param_cb(action, ¶m_ops_str, action_op, 0644);
305 MODULE_PARM_DESC(action, "Timeout action. One of: "
306 "reset, none, power_cycle, power_off.");
308 module_param_cb(preaction, ¶m_ops_str, preaction_op, 0644);
309 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: "
310 "pre_none, pre_smi, pre_nmi, pre_int.");
312 module_param_cb(preop, ¶m_ops_str, preop_op, 0644);
313 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: "
314 "preop_none, preop_panic, preop_give_data.");
316 module_param(start_now, int, 0444);
317 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as"
318 "soon as the driver is loaded.");
320 module_param(nowayout, int, 0644);
321 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started "
322 "(default=CONFIG_WATCHDOG_NOWAYOUT)");
324 /* Default state of the timer. */
325 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
327 /* If shutting down via IPMI, we ignore the heartbeat. */
328 static int ipmi_ignore_heartbeat;
330 /* Is someone using the watchdog? Only one user is allowed. */
331 static unsigned long ipmi_wdog_open;
334 * If set to 1, the heartbeat command will set the state to reset and
335 * start the timer. The timer doesn't normally run when the driver is
336 * first opened until the heartbeat is set the first time, this
337 * variable is used to accomplish this.
339 static int ipmi_start_timer_on_heartbeat;
341 /* IPMI version of the BMC. */
342 static unsigned char ipmi_version_major;
343 static unsigned char ipmi_version_minor;
345 /* If a pretimeout occurs, this is used to allow only one panic to happen. */
346 static atomic_t preop_panic_excl = ATOMIC_INIT(-1);
349 static int testing_nmi;
350 static int nmi_handler_registered;
353 static int ipmi_heartbeat(void);
356 * We use a mutex to make sure that only one thing can send a set
357 * timeout at one time, because we only have one copy of the data.
358 * The mutex is claimed when the set_timeout is sent and freed
359 * when both messages are free.
361 static atomic_t set_timeout_tofree = ATOMIC_INIT(0);
362 static DEFINE_MUTEX(set_timeout_lock);
363 static DECLARE_COMPLETION(set_timeout_wait);
364 static void set_timeout_free_smi(struct ipmi_smi_msg *msg)
366 if (atomic_dec_and_test(&set_timeout_tofree))
367 complete(&set_timeout_wait);
369 static void set_timeout_free_recv(struct ipmi_recv_msg *msg)
371 if (atomic_dec_and_test(&set_timeout_tofree))
372 complete(&set_timeout_wait);
374 static struct ipmi_smi_msg set_timeout_smi_msg = {
375 .done = set_timeout_free_smi
377 static struct ipmi_recv_msg set_timeout_recv_msg = {
378 .done = set_timeout_free_recv
381 static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg,
382 struct ipmi_recv_msg *recv_msg,
383 int *send_heartbeat_now)
385 struct kernel_ipmi_msg msg;
386 unsigned char data[6];
388 struct ipmi_system_interface_addr addr;
392 /* These can be cleared as we are setting the timeout. */
393 pretimeout_since_last_heartbeat = 0;
396 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
398 if ((ipmi_version_major > 1)
399 || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5))) {
400 /* This is an IPMI 1.5-only feature. */
401 data[0] |= WDOG_DONT_STOP_ON_SET;
402 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
404 * In ipmi 1.0, setting the timer stops the watchdog, we
405 * 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 rv = ipmi_request_supply_msgs(watchdog_user,
521 (struct ipmi_addr *) &addr,
525 &panic_halt_heartbeat_smi_msg,
526 &panic_halt_heartbeat_recv_msg,
529 atomic_add(2, &panic_done_count);
532 static struct ipmi_smi_msg panic_halt_smi_msg = {
533 .done = panic_smi_free
535 static struct ipmi_recv_msg panic_halt_recv_msg = {
536 .done = panic_recv_free
540 * Special call, doesn't claim any locks. This is only to be called
541 * at panic or halt time, in run-to-completion mode, when the caller
542 * is the only CPU and the only thing that will be going is these IPMI
545 static void panic_halt_ipmi_set_timeout(void)
547 int send_heartbeat_now;
550 /* Wait for the messages to be free. */
551 while (atomic_read(&panic_done_count) != 0)
552 ipmi_poll_interface(watchdog_user);
553 rv = i_ipmi_set_timeout(&panic_halt_smi_msg,
554 &panic_halt_recv_msg,
555 &send_heartbeat_now);
557 atomic_add(2, &panic_done_count);
558 if (send_heartbeat_now)
559 panic_halt_ipmi_heartbeat();
561 printk(KERN_WARNING PFX
562 "Unable to extend the watchdog timeout.");
563 while (atomic_read(&panic_done_count) != 0)
564 ipmi_poll_interface(watchdog_user);
568 * We use a mutex to make sure that only one thing can send a
569 * heartbeat at one time, because we only have one copy of the data.
570 * The semaphore is claimed when the set_timeout is sent and freed
571 * when both messages are free.
573 static atomic_t heartbeat_tofree = ATOMIC_INIT(0);
574 static DEFINE_MUTEX(heartbeat_lock);
575 static DECLARE_COMPLETION(heartbeat_wait);
576 static void heartbeat_free_smi(struct ipmi_smi_msg *msg)
578 if (atomic_dec_and_test(&heartbeat_tofree))
579 complete(&heartbeat_wait);
581 static void heartbeat_free_recv(struct ipmi_recv_msg *msg)
583 if (atomic_dec_and_test(&heartbeat_tofree))
584 complete(&heartbeat_wait);
586 static struct ipmi_smi_msg heartbeat_smi_msg = {
587 .done = heartbeat_free_smi
589 static struct ipmi_recv_msg heartbeat_recv_msg = {
590 .done = heartbeat_free_recv
593 static int ipmi_heartbeat(void)
595 struct kernel_ipmi_msg msg;
597 struct ipmi_system_interface_addr addr;
599 if (ipmi_ignore_heartbeat)
602 if (ipmi_start_timer_on_heartbeat) {
603 ipmi_start_timer_on_heartbeat = 0;
604 ipmi_watchdog_state = action_val;
605 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
606 } else if (pretimeout_since_last_heartbeat) {
608 * A pretimeout occurred, make sure we set the timeout.
609 * We don't want to set the action, though, we want to
610 * leave that alone (thus it can't be combined with the
613 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
616 mutex_lock(&heartbeat_lock);
618 atomic_set(&heartbeat_tofree, 2);
621 * Don't reset the timer if we have the timer turned off, that
622 * re-enables the watchdog.
624 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) {
625 mutex_unlock(&heartbeat_lock);
629 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
630 addr.channel = IPMI_BMC_CHANNEL;
634 msg.cmd = IPMI_WDOG_RESET_TIMER;
637 rv = ipmi_request_supply_msgs(watchdog_user,
638 (struct ipmi_addr *) &addr,
646 mutex_unlock(&heartbeat_lock);
647 printk(KERN_WARNING PFX "heartbeat failure: %d\n",
652 /* Wait for the heartbeat to be sent. */
653 wait_for_completion(&heartbeat_wait);
655 if (heartbeat_recv_msg.msg.data[0] != 0) {
657 * Got an error in the heartbeat response. It was already
658 * reported in ipmi_wdog_msg_handler, but we should return
664 mutex_unlock(&heartbeat_lock);
669 static struct watchdog_info ident = {
670 .options = 0, /* WDIOF_SETTIMEOUT, */
671 .firmware_version = 1,
675 static int ipmi_ioctl(struct file *file,
676 unsigned int cmd, unsigned long arg)
678 void __user *argp = (void __user *)arg;
683 case WDIOC_GETSUPPORT:
684 i = copy_to_user(argp, &ident, sizeof(ident));
685 return i ? -EFAULT : 0;
687 case WDIOC_SETTIMEOUT:
688 i = copy_from_user(&val, argp, sizeof(int));
692 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
694 case WDIOC_GETTIMEOUT:
695 i = copy_to_user(argp, &timeout, sizeof(timeout));
700 case WDIOC_SET_PRETIMEOUT:
701 case WDIOC_SETPRETIMEOUT:
702 i = copy_from_user(&val, argp, sizeof(int));
706 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
708 case WDIOC_GET_PRETIMEOUT:
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,
801 * Reading returns if the pretimeout has gone off, and it only does
802 * it once per pretimeout.
804 spin_lock(&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(&ipmi_read_lock);
817 spin_lock(&ipmi_read_lock);
819 remove_wait_queue(&read_q, &wait);
821 if (signal_pending(current)) {
829 spin_unlock(&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 nonseekable_open(ino, filep);
861 static unsigned int ipmi_poll(struct file *file, poll_table *wait)
863 unsigned int mask = 0;
865 poll_wait(file, &read_q, wait);
867 spin_lock(&ipmi_read_lock);
869 mask |= (POLLIN | POLLRDNORM);
870 spin_unlock(&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 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
889 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
892 "Unexpected close, not stopping watchdog!\n");
895 clear_bit(0, &ipmi_wdog_open);
903 static const struct file_operations ipmi_wdog_fops = {
904 .owner = THIS_MODULE,
908 .unlocked_ioctl = ipmi_unlocked_ioctl,
910 .release = ipmi_close,
911 .fasync = ipmi_fasync,
915 static struct miscdevice ipmi_wdog_miscdev = {
916 .minor = WATCHDOG_MINOR,
918 .fops = &ipmi_wdog_fops
921 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg,
924 if (msg->msg.data[0] != 0) {
925 printk(KERN_ERR PFX "response: Error %x on cmd %x\n",
930 ipmi_free_recv_msg(msg);
933 static void ipmi_wdog_pretimeout_handler(void *handler_data)
935 if (preaction_val != WDOG_PRETIMEOUT_NONE) {
936 if (preop_val == WDOG_PREOP_PANIC) {
937 if (atomic_inc_and_test(&preop_panic_excl))
938 panic("Watchdog pre-timeout");
939 } else if (preop_val == WDOG_PREOP_GIVE_DATA) {
940 spin_lock(&ipmi_read_lock);
942 wake_up_interruptible(&read_q);
943 kill_fasync(&fasync_q, SIGIO, POLL_IN);
945 spin_unlock(&ipmi_read_lock);
950 * On some machines, the heartbeat will give an error and not
951 * work unless we re-enable the timer. So do so.
953 pretimeout_since_last_heartbeat = 1;
956 static struct ipmi_user_hndl ipmi_hndlrs = {
957 .ipmi_recv_hndl = ipmi_wdog_msg_handler,
958 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler
961 static void ipmi_register_watchdog(int ipmi_intf)
968 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf))
971 watchdog_ifnum = ipmi_intf;
973 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user);
975 printk(KERN_CRIT PFX "Unable to register with ipmi\n");
979 ipmi_get_version(watchdog_user,
981 &ipmi_version_minor);
983 rv = misc_register(&ipmi_wdog_miscdev);
985 ipmi_destroy_user(watchdog_user);
986 watchdog_user = NULL;
987 printk(KERN_CRIT PFX "Unable to register misc device\n");
991 if (nmi_handler_registered) {
992 int old_pretimeout = pretimeout;
993 int old_timeout = timeout;
994 int old_preop_val = preop_val;
997 * Set the pretimeout to go off in a second and give
998 * ourselves plenty of time to stop the timer.
1000 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1001 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */
1007 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1009 printk(KERN_WARNING PFX "Error starting timer to"
1010 " test NMI: 0x%x. The NMI pretimeout will"
1011 " likely not work\n", rv);
1018 if (testing_nmi != 2) {
1019 printk(KERN_WARNING PFX "IPMI NMI didn't seem to"
1020 " occur. The NMI pretimeout will"
1021 " likely not work\n");
1025 preop_val = old_preop_val;
1026 pretimeout = old_pretimeout;
1027 timeout = old_timeout;
1032 if ((start_now) && (rv == 0)) {
1033 /* Run from startup, so start the timer now. */
1034 start_now = 0; /* Disable this function after first startup. */
1035 ipmi_watchdog_state = action_val;
1036 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
1037 printk(KERN_INFO PFX "Starting now!\n");
1039 /* Stop the timer now. */
1040 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1041 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
1045 static void ipmi_unregister_watchdog(int ipmi_intf)
1052 if (watchdog_ifnum != ipmi_intf)
1055 /* Make sure no one can call us any more. */
1056 misc_deregister(&ipmi_wdog_miscdev);
1059 * Wait to make sure the message makes it out. The lower layer has
1060 * pointers to our buffers, we want to make sure they are done before
1061 * we release our memory.
1063 while (atomic_read(&set_timeout_tofree))
1064 schedule_timeout_uninterruptible(1);
1066 /* Disconnect from IPMI. */
1067 rv = ipmi_destroy_user(watchdog_user);
1069 printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n",
1072 watchdog_user = NULL;
1080 ipmi_nmi(struct notifier_block *self, unsigned long val, void *data)
1082 struct die_args *args = data;
1084 if (val != DIE_NMIUNKNOWN)
1087 /* Hack, if it's a memory or I/O error, ignore it. */
1088 if (args->err & 0xc0)
1092 * If we get here, it's an NMI that's not a memory or I/O
1093 * error. We can't truly tell if it's from IPMI or not
1094 * without sending a message, and sending a message is almost
1095 * impossible because of locking.
1103 /* If we are not expecting a timeout, ignore it. */
1104 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
1107 if (preaction_val != WDOG_PRETIMEOUT_NMI)
1111 * If no one else handled the NMI, we assume it was the IPMI
1114 if (preop_val == WDOG_PREOP_PANIC) {
1115 /* On some machines, the heartbeat will give
1116 an error and not work unless we re-enable
1117 the timer. So do so. */
1118 pretimeout_since_last_heartbeat = 1;
1119 if (atomic_inc_and_test(&preop_panic_excl))
1120 panic(PFX "pre-timeout");
1126 static struct notifier_block ipmi_nmi_handler = {
1127 .notifier_call = ipmi_nmi
1131 static int wdog_reboot_handler(struct notifier_block *this,
1135 static int reboot_event_handled;
1137 if ((watchdog_user) && (!reboot_event_handled)) {
1138 /* Make sure we only do this once. */
1139 reboot_event_handled = 1;
1141 if (code == SYS_POWER_OFF || code == SYS_HALT) {
1142 /* Disable the WDT if we are shutting down. */
1143 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1144 panic_halt_ipmi_set_timeout();
1145 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1146 /* Set a long timer to let the reboot happens, but
1147 reboot if it hangs, but only if the watchdog
1148 timer was already running. */
1151 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1152 panic_halt_ipmi_set_timeout();
1158 static struct notifier_block wdog_reboot_notifier = {
1159 .notifier_call = wdog_reboot_handler,
1164 static int wdog_panic_handler(struct notifier_block *this,
1165 unsigned long event,
1168 static int panic_event_handled;
1170 /* On a panic, if we have a panic timeout, make sure to extend
1171 the watchdog timer to a reasonable value to complete the
1172 panic, if the watchdog timer is running. Plus the
1173 pretimeout is meaningless at panic time. */
1174 if (watchdog_user && !panic_event_handled &&
1175 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1176 /* Make sure we do this only once. */
1177 panic_event_handled = 1;
1181 panic_halt_ipmi_set_timeout();
1187 static struct notifier_block wdog_panic_notifier = {
1188 .notifier_call = wdog_panic_handler,
1190 .priority = 150 /* priority: INT_MAX >= x >= 0 */
1194 static void ipmi_new_smi(int if_num, struct device *device)
1196 ipmi_register_watchdog(if_num);
1199 static void ipmi_smi_gone(int if_num)
1201 ipmi_unregister_watchdog(if_num);
1204 static struct ipmi_smi_watcher smi_watcher = {
1205 .owner = THIS_MODULE,
1206 .new_smi = ipmi_new_smi,
1207 .smi_gone = ipmi_smi_gone
1210 static int action_op(const char *inval, char *outval)
1213 strcpy(outval, action);
1218 if (strcmp(inval, "reset") == 0)
1219 action_val = WDOG_TIMEOUT_RESET;
1220 else if (strcmp(inval, "none") == 0)
1221 action_val = WDOG_TIMEOUT_NONE;
1222 else if (strcmp(inval, "power_cycle") == 0)
1223 action_val = WDOG_TIMEOUT_POWER_CYCLE;
1224 else if (strcmp(inval, "power_off") == 0)
1225 action_val = WDOG_TIMEOUT_POWER_DOWN;
1228 strcpy(action, inval);
1232 static int preaction_op(const char *inval, char *outval)
1235 strcpy(outval, preaction);
1240 if (strcmp(inval, "pre_none") == 0)
1241 preaction_val = WDOG_PRETIMEOUT_NONE;
1242 else if (strcmp(inval, "pre_smi") == 0)
1243 preaction_val = WDOG_PRETIMEOUT_SMI;
1245 else if (strcmp(inval, "pre_nmi") == 0)
1246 preaction_val = WDOG_PRETIMEOUT_NMI;
1248 else if (strcmp(inval, "pre_int") == 0)
1249 preaction_val = WDOG_PRETIMEOUT_MSG_INT;
1252 strcpy(preaction, inval);
1256 static int preop_op(const char *inval, char *outval)
1259 strcpy(outval, preop);
1264 if (strcmp(inval, "preop_none") == 0)
1265 preop_val = WDOG_PREOP_NONE;
1266 else if (strcmp(inval, "preop_panic") == 0)
1267 preop_val = WDOG_PREOP_PANIC;
1268 else if (strcmp(inval, "preop_give_data") == 0)
1269 preop_val = WDOG_PREOP_GIVE_DATA;
1272 strcpy(preop, inval);
1276 static void check_parms(void)
1282 if (preaction_val == WDOG_PRETIMEOUT_NMI) {
1284 if (preop_val == WDOG_PREOP_GIVE_DATA) {
1285 printk(KERN_WARNING PFX "Pretimeout op is to give data"
1286 " but NMI pretimeout is enabled, setting"
1287 " pretimeout op to none\n");
1288 preop_op("preop_none", NULL);
1292 if (do_nmi && !nmi_handler_registered) {
1293 rv = register_die_notifier(&ipmi_nmi_handler);
1295 printk(KERN_WARNING PFX
1296 "Can't register nmi handler\n");
1299 nmi_handler_registered = 1;
1300 } else if (!do_nmi && nmi_handler_registered) {
1301 unregister_die_notifier(&ipmi_nmi_handler);
1302 nmi_handler_registered = 0;
1307 static int __init ipmi_wdog_init(void)
1311 if (action_op(action, NULL)) {
1312 action_op("reset", NULL);
1313 printk(KERN_INFO PFX "Unknown action '%s', defaulting to"
1314 " reset\n", action);
1317 if (preaction_op(preaction, NULL)) {
1318 preaction_op("pre_none", NULL);
1319 printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to"
1320 " none\n", preaction);
1323 if (preop_op(preop, NULL)) {
1324 preop_op("preop_none", NULL);
1325 printk(KERN_INFO PFX "Unknown preop '%s', defaulting to"
1331 register_reboot_notifier(&wdog_reboot_notifier);
1332 atomic_notifier_chain_register(&panic_notifier_list,
1333 &wdog_panic_notifier);
1335 rv = ipmi_smi_watcher_register(&smi_watcher);
1338 if (nmi_handler_registered)
1339 unregister_die_notifier(&ipmi_nmi_handler);
1341 atomic_notifier_chain_unregister(&panic_notifier_list,
1342 &wdog_panic_notifier);
1343 unregister_reboot_notifier(&wdog_reboot_notifier);
1344 printk(KERN_WARNING PFX "can't register smi watcher\n");
1348 printk(KERN_INFO PFX "driver initialized\n");
1353 static void __exit ipmi_wdog_exit(void)
1355 ipmi_smi_watcher_unregister(&smi_watcher);
1356 ipmi_unregister_watchdog(watchdog_ifnum);
1359 if (nmi_handler_registered)
1360 unregister_die_notifier(&ipmi_nmi_handler);
1363 atomic_notifier_chain_unregister(&panic_notifier_list,
1364 &wdog_panic_notifier);
1365 unregister_reboot_notifier(&wdog_reboot_notifier);
1367 module_exit(ipmi_wdog_exit);
1368 module_init(ipmi_wdog_init);
1369 MODULE_LICENSE("GPL");
1370 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
1371 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface.");