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/watchdog.h>
39 #include <linux/miscdevice.h>
40 #include <linux/init.h>
41 #include <linux/completion.h>
42 #include <linux/kdebug.h>
43 #include <linux/rwsem.h>
44 #include <linux/errno.h>
45 #include <asm/uaccess.h>
46 #include <linux/notifier.h>
47 #include <linux/nmi.h>
48 #include <linux/reboot.h>
49 #include <linux/wait.h>
50 #include <linux/poll.h>
51 #include <linux/string.h>
52 #include <linux/ctype.h>
53 #include <linux/delay.h>
54 #include <asm/atomic.h>
57 /* This is ugly, but I've determined that x86 is the only architecture
58 that can reasonably support the IPMI NMI watchdog timeout at this
59 time. If another architecture adds this capability somehow, it
60 will have to be a somewhat different mechanism and I have no idea
61 how it will work. So in the unlikely event that another
62 architecture supports this, we can figure out a good generic
63 mechanism for it at that time. */
64 #include <asm/kdebug.h>
68 #define PFX "IPMI Watchdog: "
71 * The IPMI command/response information for the watchdog timer.
74 /* values for byte 1 of the set command, byte 2 of the get response. */
75 #define WDOG_DONT_LOG (1 << 7)
76 #define WDOG_DONT_STOP_ON_SET (1 << 6)
77 #define WDOG_SET_TIMER_USE(byte, use) \
78 byte = ((byte) & 0xf8) | ((use) & 0x7)
79 #define WDOG_GET_TIMER_USE(byte) ((byte) & 0x7)
80 #define WDOG_TIMER_USE_BIOS_FRB2 1
81 #define WDOG_TIMER_USE_BIOS_POST 2
82 #define WDOG_TIMER_USE_OS_LOAD 3
83 #define WDOG_TIMER_USE_SMS_OS 4
84 #define WDOG_TIMER_USE_OEM 5
86 /* values for byte 2 of the set command, byte 3 of the get response. */
87 #define WDOG_SET_PRETIMEOUT_ACT(byte, use) \
88 byte = ((byte) & 0x8f) | (((use) & 0x7) << 4)
89 #define WDOG_GET_PRETIMEOUT_ACT(byte) (((byte) >> 4) & 0x7)
90 #define WDOG_PRETIMEOUT_NONE 0
91 #define WDOG_PRETIMEOUT_SMI 1
92 #define WDOG_PRETIMEOUT_NMI 2
93 #define WDOG_PRETIMEOUT_MSG_INT 3
95 /* Operations that can be performed on a pretimout. */
96 #define WDOG_PREOP_NONE 0
97 #define WDOG_PREOP_PANIC 1
98 #define WDOG_PREOP_GIVE_DATA 2 /* Cause data to be available to
99 read. Doesn't work in NMI
102 /* Actions to perform on a full timeout. */
103 #define WDOG_SET_TIMEOUT_ACT(byte, use) \
104 byte = ((byte) & 0xf8) | ((use) & 0x7)
105 #define WDOG_GET_TIMEOUT_ACT(byte) ((byte) & 0x7)
106 #define WDOG_TIMEOUT_NONE 0
107 #define WDOG_TIMEOUT_RESET 1
108 #define WDOG_TIMEOUT_POWER_DOWN 2
109 #define WDOG_TIMEOUT_POWER_CYCLE 3
111 /* Byte 3 of the get command, byte 4 of the get response is the
112 pre-timeout in seconds. */
114 /* Bits for setting byte 4 of the set command, byte 5 of the get response. */
115 #define WDOG_EXPIRE_CLEAR_BIOS_FRB2 (1 << 1)
116 #define WDOG_EXPIRE_CLEAR_BIOS_POST (1 << 2)
117 #define WDOG_EXPIRE_CLEAR_OS_LOAD (1 << 3)
118 #define WDOG_EXPIRE_CLEAR_SMS_OS (1 << 4)
119 #define WDOG_EXPIRE_CLEAR_OEM (1 << 5)
121 /* Setting/getting the watchdog timer value. This is for bytes 5 and
122 6 (the timeout time) of the set command, and bytes 6 and 7 (the
123 timeout time) and 8 and 9 (the current countdown value) of the
124 response. The timeout value is given in seconds (in the command it
125 is 100ms intervals). */
126 #define WDOG_SET_TIMEOUT(byte1, byte2, val) \
127 (byte1) = (((val) * 10) & 0xff), (byte2) = (((val) * 10) >> 8)
128 #define WDOG_GET_TIMEOUT(byte1, byte2) \
129 (((byte1) | ((byte2) << 8)) / 10)
131 #define IPMI_WDOG_RESET_TIMER 0x22
132 #define IPMI_WDOG_SET_TIMER 0x24
133 #define IPMI_WDOG_GET_TIMER 0x25
135 /* These are here until the real ones get into the watchdog.h interface. */
136 #ifndef WDIOC_GETTIMEOUT
137 #define WDIOC_GETTIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 20, int)
139 #ifndef WDIOC_SET_PRETIMEOUT
140 #define WDIOC_SET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 21, int)
142 #ifndef WDIOC_GET_PRETIMEOUT
143 #define WDIOC_GET_PRETIMEOUT _IOW(WATCHDOG_IOCTL_BASE, 22, int)
146 static int nowayout = WATCHDOG_NOWAYOUT;
148 static ipmi_user_t watchdog_user;
149 static int watchdog_ifnum;
151 /* Default the timeout to 10 seconds. */
152 static int timeout = 10;
154 /* The pre-timeout is disabled by default. */
155 static int pretimeout;
157 /* Default action is to reset the board on a timeout. */
158 static unsigned char action_val = WDOG_TIMEOUT_RESET;
160 static char action[16] = "reset";
162 static unsigned char preaction_val = WDOG_PRETIMEOUT_NONE;
164 static char preaction[16] = "pre_none";
166 static unsigned char preop_val = WDOG_PREOP_NONE;
168 static char preop[16] = "preop_none";
169 static DEFINE_SPINLOCK(ipmi_read_lock);
170 static char data_to_read;
171 static DECLARE_WAIT_QUEUE_HEAD(read_q);
172 static struct fasync_struct *fasync_q;
173 static char pretimeout_since_last_heartbeat;
174 static char expect_close;
176 static int ifnum_to_use = -1;
178 /* Parameters to ipmi_set_timeout */
179 #define IPMI_SET_TIMEOUT_NO_HB 0
180 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
181 #define IPMI_SET_TIMEOUT_FORCE_HB 2
183 static int ipmi_set_timeout(int do_heartbeat);
184 static void ipmi_register_watchdog(int ipmi_intf);
185 static void ipmi_unregister_watchdog(int ipmi_intf);
187 /* If true, the driver will start running as soon as it is configured
189 static int start_now;
191 static int set_param_int(const char *val, struct kernel_param *kp)
199 l = simple_strtoul(val, &endp, 0);
203 *((int *)kp->arg) = l;
205 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
210 static int get_param_int(char *buffer, struct kernel_param *kp)
212 return sprintf(buffer, "%i", *((int *)kp->arg));
215 typedef int (*action_fn)(const char *intval, char *outval);
217 static int action_op(const char *inval, char *outval);
218 static int preaction_op(const char *inval, char *outval);
219 static int preop_op(const char *inval, char *outval);
220 static void check_parms(void);
222 static int set_param_str(const char *val, struct kernel_param *kp)
224 action_fn fn = (action_fn) kp->arg;
229 strncpy(valcp, val, 16);
240 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
246 static int get_param_str(char *buffer, struct kernel_param *kp)
248 action_fn fn = (action_fn) kp->arg;
251 rv = fn(NULL, buffer);
254 return strlen(buffer);
258 static int set_param_wdog_ifnum(const char *val, struct kernel_param *kp)
260 int rv = param_set_int(val, kp);
263 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum))
266 ipmi_unregister_watchdog(watchdog_ifnum);
267 ipmi_register_watchdog(ifnum_to_use);
271 module_param_call(ifnum_to_use, set_param_wdog_ifnum, get_param_int,
272 &ifnum_to_use, 0644);
273 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog "
274 "timer. Setting to -1 defaults to the first registered "
277 module_param_call(timeout, set_param_int, get_param_int, &timeout, 0644);
278 MODULE_PARM_DESC(timeout, "Timeout value in seconds.");
280 module_param_call(pretimeout, set_param_int, get_param_int, &pretimeout, 0644);
281 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds.");
283 module_param_call(action, set_param_str, get_param_str, action_op, 0644);
284 MODULE_PARM_DESC(action, "Timeout action. One of: "
285 "reset, none, power_cycle, power_off.");
287 module_param_call(preaction, set_param_str, get_param_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_call(preop, set_param_str, get_param_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, int, 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 /* If shutting down via IPMI, we ignore the heartbeat. */
307 static int ipmi_ignore_heartbeat;
309 /* Is someone using the watchdog? Only one user is allowed. */
310 static unsigned long ipmi_wdog_open;
312 /* If set to 1, the heartbeat command will set the state to reset and
313 start the timer. The timer doesn't normally run when the driver is
314 first opened until the heartbeat is set the first time, this
315 variable is used to accomplish this. */
316 static int ipmi_start_timer_on_heartbeat;
318 /* IPMI version of the BMC. */
319 static unsigned char ipmi_version_major;
320 static unsigned char ipmi_version_minor;
322 /* If a pretimeout occurs, this is used to allow only one panic to happen. */
323 static atomic_t preop_panic_excl = ATOMIC_INIT(-1);
326 static int testing_nmi;
327 static int nmi_handler_registered;
330 static int ipmi_heartbeat(void);
332 /* We use a mutex to make sure that only one thing can send a set
333 timeout at one time, because we only have one copy of the data.
334 The mutex is claimed when the set_timeout is sent and freed
335 when both messages are free. */
336 static atomic_t set_timeout_tofree = ATOMIC_INIT(0);
337 static DEFINE_MUTEX(set_timeout_lock);
338 static DECLARE_COMPLETION(set_timeout_wait);
339 static void set_timeout_free_smi(struct ipmi_smi_msg *msg)
341 if (atomic_dec_and_test(&set_timeout_tofree))
342 complete(&set_timeout_wait);
344 static void set_timeout_free_recv(struct ipmi_recv_msg *msg)
346 if (atomic_dec_and_test(&set_timeout_tofree))
347 complete(&set_timeout_wait);
349 static struct ipmi_smi_msg set_timeout_smi_msg =
351 .done = set_timeout_free_smi
353 static struct ipmi_recv_msg set_timeout_recv_msg =
355 .done = set_timeout_free_recv
358 static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg,
359 struct ipmi_recv_msg *recv_msg,
360 int *send_heartbeat_now)
362 struct kernel_ipmi_msg msg;
363 unsigned char data[6];
365 struct ipmi_system_interface_addr addr;
369 /* These can be cleared as we are setting the timeout. */
370 pretimeout_since_last_heartbeat = 0;
373 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
375 if ((ipmi_version_major > 1)
376 || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5)))
378 /* This is an IPMI 1.5-only feature. */
379 data[0] |= WDOG_DONT_STOP_ON_SET;
380 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
381 /* In ipmi 1.0, setting the timer stops the watchdog, we
382 need to start it back up again. */
387 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state);
388 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) {
389 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val);
390 data[2] = pretimeout;
392 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE);
393 data[2] = 0; /* No pretimeout. */
396 WDOG_SET_TIMEOUT(data[4], data[5], timeout);
398 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
399 addr.channel = IPMI_BMC_CHANNEL;
403 msg.cmd = IPMI_WDOG_SET_TIMER;
405 msg.data_len = sizeof(data);
406 rv = ipmi_request_supply_msgs(watchdog_user,
407 (struct ipmi_addr *) &addr,
415 printk(KERN_WARNING PFX "set timeout error: %d\n",
419 if (send_heartbeat_now)
420 *send_heartbeat_now = hbnow;
425 static int ipmi_set_timeout(int do_heartbeat)
427 int send_heartbeat_now;
431 /* We can only send one of these at a time. */
432 mutex_lock(&set_timeout_lock);
434 atomic_set(&set_timeout_tofree, 2);
436 rv = i_ipmi_set_timeout(&set_timeout_smi_msg,
437 &set_timeout_recv_msg,
438 &send_heartbeat_now);
440 mutex_unlock(&set_timeout_lock);
444 wait_for_completion(&set_timeout_wait);
446 mutex_unlock(&set_timeout_lock);
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();
457 static atomic_t panic_done_count = ATOMIC_INIT(0);
459 static void panic_smi_free(struct ipmi_smi_msg *msg)
461 atomic_dec(&panic_done_count);
463 static void panic_recv_free(struct ipmi_recv_msg *msg)
465 atomic_dec(&panic_done_count);
468 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg =
470 .done = panic_smi_free
472 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg =
474 .done = panic_recv_free
477 static void panic_halt_ipmi_heartbeat(void)
479 struct kernel_ipmi_msg msg;
480 struct ipmi_system_interface_addr addr;
483 /* Don't reset the timer if we have the timer turned off, that
484 re-enables the watchdog. */
485 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
488 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
489 addr.channel = IPMI_BMC_CHANNEL;
493 msg.cmd = IPMI_WDOG_RESET_TIMER;
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_add(2, &panic_done_count);
508 static struct ipmi_smi_msg panic_halt_smi_msg =
510 .done = panic_smi_free
512 static struct ipmi_recv_msg panic_halt_recv_msg =
514 .done = panic_recv_free
517 /* Special call, doesn't claim any locks. This is only to be called
518 at panic or halt time, in run-to-completion mode, when the caller
519 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 rv = i_ipmi_set_timeout(&panic_halt_smi_msg,
530 &panic_halt_recv_msg,
531 &send_heartbeat_now);
533 atomic_add(2, &panic_done_count);
534 if (send_heartbeat_now)
535 panic_halt_ipmi_heartbeat();
537 printk(KERN_WARNING PFX
538 "Unable to extend the watchdog timeout.");
539 while (atomic_read(&panic_done_count) != 0)
540 ipmi_poll_interface(watchdog_user);
543 /* We use a semaphore to make sure that only one thing can send a
544 heartbeat at one time, because we only have one copy of the data.
545 The semaphore is claimed when the set_timeout is sent and freed
546 when both messages are free. */
547 static atomic_t heartbeat_tofree = ATOMIC_INIT(0);
548 static DEFINE_MUTEX(heartbeat_lock);
549 static DECLARE_COMPLETION(heartbeat_wait);
550 static void heartbeat_free_smi(struct ipmi_smi_msg *msg)
552 if (atomic_dec_and_test(&heartbeat_tofree))
553 complete(&heartbeat_wait);
555 static void heartbeat_free_recv(struct ipmi_recv_msg *msg)
557 if (atomic_dec_and_test(&heartbeat_tofree))
558 complete(&heartbeat_wait);
560 static struct ipmi_smi_msg heartbeat_smi_msg =
562 .done = heartbeat_free_smi
564 static struct ipmi_recv_msg heartbeat_recv_msg =
566 .done = heartbeat_free_recv
569 static int ipmi_heartbeat(void)
571 struct kernel_ipmi_msg msg;
573 struct ipmi_system_interface_addr addr;
575 if (ipmi_ignore_heartbeat)
578 if (ipmi_start_timer_on_heartbeat) {
579 ipmi_start_timer_on_heartbeat = 0;
580 ipmi_watchdog_state = action_val;
581 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
582 } else if (pretimeout_since_last_heartbeat) {
583 /* A pretimeout occurred, make sure we set the timeout.
584 We don't want to set the action, though, we want to
585 leave that alone (thus it can't be combined with the
587 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
590 mutex_lock(&heartbeat_lock);
592 atomic_set(&heartbeat_tofree, 2);
594 /* Don't reset the timer if we have the timer turned off, that
595 re-enables the watchdog. */
596 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) {
597 mutex_unlock(&heartbeat_lock);
601 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
602 addr.channel = IPMI_BMC_CHANNEL;
606 msg.cmd = IPMI_WDOG_RESET_TIMER;
609 rv = ipmi_request_supply_msgs(watchdog_user,
610 (struct ipmi_addr *) &addr,
618 mutex_unlock(&heartbeat_lock);
619 printk(KERN_WARNING PFX "heartbeat failure: %d\n",
624 /* Wait for the heartbeat to be sent. */
625 wait_for_completion(&heartbeat_wait);
627 if (heartbeat_recv_msg.msg.data[0] != 0) {
628 /* Got an error in the heartbeat response. It was already
629 reported in ipmi_wdog_msg_handler, but we should return
634 mutex_unlock(&heartbeat_lock);
639 static struct watchdog_info ident =
641 .options = 0, /* WDIOF_SETTIMEOUT, */
642 .firmware_version = 1,
646 static int ipmi_ioctl(struct inode *inode, struct file *file,
647 unsigned int cmd, unsigned long arg)
649 void __user *argp = (void __user *)arg;
654 case WDIOC_GETSUPPORT:
655 i = copy_to_user(argp, &ident, sizeof(ident));
656 return i ? -EFAULT : 0;
658 case WDIOC_SETTIMEOUT:
659 i = copy_from_user(&val, argp, sizeof(int));
663 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
665 case WDIOC_GETTIMEOUT:
666 i = copy_to_user(argp, &timeout, sizeof(timeout));
671 case WDIOC_SET_PRETIMEOUT:
672 i = copy_from_user(&val, argp, sizeof(int));
676 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
678 case WDIOC_GET_PRETIMEOUT:
679 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout));
684 case WDIOC_KEEPALIVE:
685 return ipmi_heartbeat();
687 case WDIOC_SETOPTIONS:
688 i = copy_from_user(&val, argp, sizeof(int));
691 if (val & WDIOS_DISABLECARD)
693 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
694 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
695 ipmi_start_timer_on_heartbeat = 0;
698 if (val & WDIOS_ENABLECARD)
700 ipmi_watchdog_state = action_val;
701 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
705 case WDIOC_GETSTATUS:
707 i = copy_to_user(argp, &val, sizeof(val));
717 static ssize_t ipmi_write(struct file *file,
718 const char __user *buf,
728 /* In case it was set long ago */
731 for (i = 0; i != len; i++) {
734 if (get_user(c, buf + i))
740 rv = ipmi_heartbeat();
748 static ssize_t ipmi_read(struct file *file,
759 /* Reading returns if the pretimeout has gone off, and it only does
760 it once per pretimeout. */
761 spin_lock(&ipmi_read_lock);
763 if (file->f_flags & O_NONBLOCK) {
768 init_waitqueue_entry(&wait, current);
769 add_wait_queue(&read_q, &wait);
770 while (!data_to_read) {
771 set_current_state(TASK_INTERRUPTIBLE);
772 spin_unlock(&ipmi_read_lock);
774 spin_lock(&ipmi_read_lock);
776 remove_wait_queue(&read_q, &wait);
778 if (signal_pending(current)) {
786 spin_unlock(&ipmi_read_lock);
789 if (copy_to_user(buf, &data_to_read, 1))
798 static int ipmi_open(struct inode *ino, struct file *filep)
800 switch (iminor(ino)) {
802 if (test_and_set_bit(0, &ipmi_wdog_open))
805 /* Don't start the timer now, let it start on the
807 ipmi_start_timer_on_heartbeat = 1;
808 return nonseekable_open(ino, filep);
815 static unsigned int ipmi_poll(struct file *file, poll_table *wait)
817 unsigned int mask = 0;
819 poll_wait(file, &read_q, wait);
821 spin_lock(&ipmi_read_lock);
823 mask |= (POLLIN | POLLRDNORM);
824 spin_unlock(&ipmi_read_lock);
829 static int ipmi_fasync(int fd, struct file *file, int on)
833 result = fasync_helper(fd, file, on, &fasync_q);
838 static int ipmi_close(struct inode *ino, struct file *filep)
840 if (iminor(ino) == WATCHDOG_MINOR) {
841 if (expect_close == 42) {
842 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
843 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
846 "Unexpected close, not stopping watchdog!\n");
849 clear_bit(0, &ipmi_wdog_open);
852 ipmi_fasync (-1, filep, 0);
858 static const struct file_operations ipmi_wdog_fops = {
859 .owner = THIS_MODULE,
865 .release = ipmi_close,
866 .fasync = ipmi_fasync,
869 static struct miscdevice ipmi_wdog_miscdev = {
870 .minor = WATCHDOG_MINOR,
872 .fops = &ipmi_wdog_fops
875 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg,
878 if (msg->msg.data[0] != 0) {
879 printk(KERN_ERR PFX "response: Error %x on cmd %x\n",
884 ipmi_free_recv_msg(msg);
887 static void ipmi_wdog_pretimeout_handler(void *handler_data)
889 if (preaction_val != WDOG_PRETIMEOUT_NONE) {
890 if (preop_val == WDOG_PREOP_PANIC) {
891 if (atomic_inc_and_test(&preop_panic_excl))
892 panic("Watchdog pre-timeout");
893 } else if (preop_val == WDOG_PREOP_GIVE_DATA) {
894 spin_lock(&ipmi_read_lock);
896 wake_up_interruptible(&read_q);
897 kill_fasync(&fasync_q, SIGIO, POLL_IN);
899 spin_unlock(&ipmi_read_lock);
903 /* On some machines, the heartbeat will give
904 an error and not work unless we re-enable
905 the timer. So do so. */
906 pretimeout_since_last_heartbeat = 1;
909 static struct ipmi_user_hndl ipmi_hndlrs =
911 .ipmi_recv_hndl = ipmi_wdog_msg_handler,
912 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler
915 static void ipmi_register_watchdog(int ipmi_intf)
922 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf))
925 watchdog_ifnum = ipmi_intf;
927 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user);
929 printk(KERN_CRIT PFX "Unable to register with ipmi\n");
933 ipmi_get_version(watchdog_user,
935 &ipmi_version_minor);
937 rv = misc_register(&ipmi_wdog_miscdev);
939 ipmi_destroy_user(watchdog_user);
940 watchdog_user = NULL;
941 printk(KERN_CRIT PFX "Unable to register misc device\n");
945 if (nmi_handler_registered) {
946 int old_pretimeout = pretimeout;
947 int old_timeout = timeout;
948 int old_preop_val = preop_val;
950 /* Set the pretimeout to go off in a second and give
951 ourselves plenty of time to stop the timer. */
952 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
953 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */
959 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
961 printk(KERN_WARNING PFX "Error starting timer to"
962 " test NMI: 0x%x. The NMI pretimeout will"
963 " likely not work\n", rv);
970 if (testing_nmi != 2) {
971 printk(KERN_WARNING PFX "IPMI NMI didn't seem to"
972 " occur. The NMI pretimeout will"
973 " likely not work\n");
977 preop_val = old_preop_val;
978 pretimeout = old_pretimeout;
979 timeout = old_timeout;
984 if ((start_now) && (rv == 0)) {
985 /* Run from startup, so start the timer now. */
986 start_now = 0; /* Disable this function after first startup. */
987 ipmi_watchdog_state = action_val;
988 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
989 printk(KERN_INFO PFX "Starting now!\n");
991 /* Stop the timer now. */
992 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
993 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
997 static void ipmi_unregister_watchdog(int ipmi_intf)
1004 if (watchdog_ifnum != ipmi_intf)
1007 /* Make sure no one can call us any more. */
1008 misc_deregister(&ipmi_wdog_miscdev);
1010 /* Wait to make sure the message makes it out. The lower layer has
1011 pointers to our buffers, we want to make sure they are done before
1012 we release our memory. */
1013 while (atomic_read(&set_timeout_tofree))
1014 schedule_timeout_uninterruptible(1);
1016 /* Disconnect from IPMI. */
1017 rv = ipmi_destroy_user(watchdog_user);
1019 printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n",
1022 watchdog_user = NULL;
1030 ipmi_nmi(struct notifier_block *self, unsigned long val, void *data)
1032 struct die_args *args = data;
1037 /* Hack, if it's a memory or I/O error, ignore it. */
1038 if (args->err & 0xc0)
1042 * If we get here, it's an NMI that's not a memory or I/O
1043 * error. We can't truly tell if it's from IPMI or not
1044 * without sending a message, and sending a message is almost
1045 * impossible because of locking.
1053 /* If we are not expecting a timeout, ignore it. */
1054 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
1057 if (preaction_val != WDOG_PRETIMEOUT_NMI)
1060 /* If no one else handled the NMI, we assume it was the IPMI
1062 if (preop_val == WDOG_PREOP_PANIC) {
1063 /* On some machines, the heartbeat will give
1064 an error and not work unless we re-enable
1065 the timer. So do so. */
1066 pretimeout_since_last_heartbeat = 1;
1067 if (atomic_inc_and_test(&preop_panic_excl))
1068 panic(PFX "pre-timeout");
1074 static struct notifier_block ipmi_nmi_handler = {
1075 .notifier_call = ipmi_nmi
1079 static int wdog_reboot_handler(struct notifier_block *this,
1083 static int reboot_event_handled = 0;
1085 if ((watchdog_user) && (!reboot_event_handled)) {
1086 /* Make sure we only do this once. */
1087 reboot_event_handled = 1;
1089 if (code == SYS_POWER_OFF || code == SYS_HALT) {
1090 /* Disable the WDT if we are shutting down. */
1091 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1092 panic_halt_ipmi_set_timeout();
1093 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1094 /* Set a long timer to let the reboot happens, but
1095 reboot if it hangs, but only if the watchdog
1096 timer was already running. */
1099 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1100 panic_halt_ipmi_set_timeout();
1106 static struct notifier_block wdog_reboot_notifier = {
1107 .notifier_call = wdog_reboot_handler,
1112 static int wdog_panic_handler(struct notifier_block *this,
1113 unsigned long event,
1116 static int panic_event_handled = 0;
1118 /* On a panic, if we have a panic timeout, make sure to extend
1119 the watchdog timer to a reasonable value to complete the
1120 panic, if the watchdog timer is running. Plus the
1121 pretimeout is meaningless at panic time. */
1122 if (watchdog_user && !panic_event_handled &&
1123 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1124 /* Make sure we do this only once. */
1125 panic_event_handled = 1;
1129 panic_halt_ipmi_set_timeout();
1135 static struct notifier_block wdog_panic_notifier = {
1136 .notifier_call = wdog_panic_handler,
1138 .priority = 150 /* priority: INT_MAX >= x >= 0 */
1142 static void ipmi_new_smi(int if_num, struct device *device)
1144 ipmi_register_watchdog(if_num);
1147 static void ipmi_smi_gone(int if_num)
1149 ipmi_unregister_watchdog(if_num);
1152 static struct ipmi_smi_watcher smi_watcher =
1154 .owner = THIS_MODULE,
1155 .new_smi = ipmi_new_smi,
1156 .smi_gone = ipmi_smi_gone
1159 static int action_op(const char *inval, char *outval)
1162 strcpy(outval, action);
1167 if (strcmp(inval, "reset") == 0)
1168 action_val = WDOG_TIMEOUT_RESET;
1169 else if (strcmp(inval, "none") == 0)
1170 action_val = WDOG_TIMEOUT_NONE;
1171 else if (strcmp(inval, "power_cycle") == 0)
1172 action_val = WDOG_TIMEOUT_POWER_CYCLE;
1173 else if (strcmp(inval, "power_off") == 0)
1174 action_val = WDOG_TIMEOUT_POWER_DOWN;
1177 strcpy(action, inval);
1181 static int preaction_op(const char *inval, char *outval)
1184 strcpy(outval, preaction);
1189 if (strcmp(inval, "pre_none") == 0)
1190 preaction_val = WDOG_PRETIMEOUT_NONE;
1191 else if (strcmp(inval, "pre_smi") == 0)
1192 preaction_val = WDOG_PRETIMEOUT_SMI;
1194 else if (strcmp(inval, "pre_nmi") == 0)
1195 preaction_val = WDOG_PRETIMEOUT_NMI;
1197 else if (strcmp(inval, "pre_int") == 0)
1198 preaction_val = WDOG_PRETIMEOUT_MSG_INT;
1201 strcpy(preaction, inval);
1205 static int preop_op(const char *inval, char *outval)
1208 strcpy(outval, preop);
1213 if (strcmp(inval, "preop_none") == 0)
1214 preop_val = WDOG_PREOP_NONE;
1215 else if (strcmp(inval, "preop_panic") == 0)
1216 preop_val = WDOG_PREOP_PANIC;
1217 else if (strcmp(inval, "preop_give_data") == 0)
1218 preop_val = WDOG_PREOP_GIVE_DATA;
1221 strcpy(preop, inval);
1225 static void check_parms(void)
1231 if (preaction_val == WDOG_PRETIMEOUT_NMI) {
1233 if (preop_val == WDOG_PREOP_GIVE_DATA) {
1234 printk(KERN_WARNING PFX "Pretimeout op is to give data"
1235 " but NMI pretimeout is enabled, setting"
1236 " pretimeout op to none\n");
1237 preop_op("preop_none", NULL);
1241 if (do_nmi && !nmi_handler_registered) {
1242 rv = register_die_notifier(&ipmi_nmi_handler);
1244 printk(KERN_WARNING PFX
1245 "Can't register nmi handler\n");
1248 nmi_handler_registered = 1;
1249 } else if (!do_nmi && nmi_handler_registered) {
1250 unregister_die_notifier(&ipmi_nmi_handler);
1251 nmi_handler_registered = 0;
1256 static int __init ipmi_wdog_init(void)
1260 if (action_op(action, NULL)) {
1261 action_op("reset", NULL);
1262 printk(KERN_INFO PFX "Unknown action '%s', defaulting to"
1263 " reset\n", action);
1266 if (preaction_op(preaction, NULL)) {
1267 preaction_op("pre_none", NULL);
1268 printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to"
1269 " none\n", preaction);
1272 if (preop_op(preop, NULL)) {
1273 preop_op("preop_none", NULL);
1274 printk(KERN_INFO PFX "Unknown preop '%s', defaulting to"
1280 register_reboot_notifier(&wdog_reboot_notifier);
1281 atomic_notifier_chain_register(&panic_notifier_list,
1282 &wdog_panic_notifier);
1284 rv = ipmi_smi_watcher_register(&smi_watcher);
1287 if (nmi_handler_registered)
1288 unregister_die_notifier(&ipmi_nmi_handler);
1290 atomic_notifier_chain_unregister(&panic_notifier_list,
1291 &wdog_panic_notifier);
1292 unregister_reboot_notifier(&wdog_reboot_notifier);
1293 printk(KERN_WARNING PFX "can't register smi watcher\n");
1297 printk(KERN_INFO PFX "driver initialized\n");
1302 static void __exit ipmi_wdog_exit(void)
1304 ipmi_smi_watcher_unregister(&smi_watcher);
1305 ipmi_unregister_watchdog(watchdog_ifnum);
1308 if (nmi_handler_registered)
1309 unregister_die_notifier(&ipmi_nmi_handler);
1312 atomic_notifier_chain_unregister(&panic_notifier_list,
1313 &wdog_panic_notifier);
1314 unregister_reboot_notifier(&wdog_reboot_notifier);
1316 module_exit(ipmi_wdog_exit);
1317 module_init(ipmi_wdog_init);
1318 MODULE_LICENSE("GPL");
1319 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
1320 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface.");