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>
65 #define HAVE_DIE_NMI_POST
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 static DECLARE_RWSEM(register_sem);
180 /* Parameters to ipmi_set_timeout */
181 #define IPMI_SET_TIMEOUT_NO_HB 0
182 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1
183 #define IPMI_SET_TIMEOUT_FORCE_HB 2
185 static int ipmi_set_timeout(int do_heartbeat);
186 static void ipmi_register_watchdog(int ipmi_intf);
187 static void ipmi_unregister_watchdog(int ipmi_intf);
189 /* If true, the driver will start running as soon as it is configured
191 static int start_now;
193 static int set_param_int(const char *val, struct kernel_param *kp)
201 l = simple_strtoul(val, &endp, 0);
205 down_read(®ister_sem);
206 *((int *)kp->arg) = l;
208 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
209 up_read(®ister_sem);
214 static int get_param_int(char *buffer, struct kernel_param *kp)
216 return sprintf(buffer, "%i", *((int *)kp->arg));
219 typedef int (*action_fn)(const char *intval, char *outval);
221 static int action_op(const char *inval, char *outval);
222 static int preaction_op(const char *inval, char *outval);
223 static int preop_op(const char *inval, char *outval);
224 static void check_parms(void);
226 static int set_param_str(const char *val, struct kernel_param *kp)
228 action_fn fn = (action_fn) kp->arg;
233 strncpy(valcp, val, 16);
238 down_read(®ister_sem);
245 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
248 up_read(®ister_sem);
252 static int get_param_str(char *buffer, struct kernel_param *kp)
254 action_fn fn = (action_fn) kp->arg;
257 rv = fn(NULL, buffer);
260 return strlen(buffer);
264 static int set_param_wdog_ifnum(const char *val, struct kernel_param *kp)
266 int rv = param_set_int(val, kp);
269 if ((ifnum_to_use < 0) || (ifnum_to_use == watchdog_ifnum))
272 ipmi_unregister_watchdog(watchdog_ifnum);
273 ipmi_register_watchdog(ifnum_to_use);
277 module_param_call(ifnum_to_use, set_param_wdog_ifnum, get_param_int,
278 &ifnum_to_use, 0644);
279 MODULE_PARM_DESC(ifnum_to_use, "The interface number to use for the watchdog "
280 "timer. Setting to -1 defaults to the first registered "
283 module_param_call(timeout, set_param_int, get_param_int, &timeout, 0644);
284 MODULE_PARM_DESC(timeout, "Timeout value in seconds.");
286 module_param_call(pretimeout, set_param_int, get_param_int, &pretimeout, 0644);
287 MODULE_PARM_DESC(pretimeout, "Pretimeout value in seconds.");
289 module_param_call(action, set_param_str, get_param_str, action_op, 0644);
290 MODULE_PARM_DESC(action, "Timeout action. One of: "
291 "reset, none, power_cycle, power_off.");
293 module_param_call(preaction, set_param_str, get_param_str, preaction_op, 0644);
294 MODULE_PARM_DESC(preaction, "Pretimeout action. One of: "
295 "pre_none, pre_smi, pre_nmi, pre_int.");
297 module_param_call(preop, set_param_str, get_param_str, preop_op, 0644);
298 MODULE_PARM_DESC(preop, "Pretimeout driver operation. One of: "
299 "preop_none, preop_panic, preop_give_data.");
301 module_param(start_now, int, 0444);
302 MODULE_PARM_DESC(start_now, "Set to 1 to start the watchdog as"
303 "soon as the driver is loaded.");
305 module_param(nowayout, int, 0644);
306 MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started "
307 "(default=CONFIG_WATCHDOG_NOWAYOUT)");
309 /* Default state of the timer. */
310 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
312 /* If shutting down via IPMI, we ignore the heartbeat. */
313 static int ipmi_ignore_heartbeat;
315 /* Is someone using the watchdog? Only one user is allowed. */
316 static unsigned long ipmi_wdog_open;
318 /* If set to 1, the heartbeat command will set the state to reset and
319 start the timer. The timer doesn't normally run when the driver is
320 first opened until the heartbeat is set the first time, this
321 variable is used to accomplish this. */
322 static int ipmi_start_timer_on_heartbeat;
324 /* IPMI version of the BMC. */
325 static unsigned char ipmi_version_major;
326 static unsigned char ipmi_version_minor;
328 /* If a pretimeout occurs, this is used to allow only one panic to happen. */
329 static atomic_t preop_panic_excl = ATOMIC_INIT(-1);
331 #ifdef HAVE_DIE_NMI_POST
332 static int testing_nmi;
333 static int nmi_handler_registered;
336 static int ipmi_heartbeat(void);
337 static void panic_halt_ipmi_heartbeat(void);
340 /* We use a mutex to make sure that only one thing can send a set
341 timeout at one time, because we only have one copy of the data.
342 The mutex is claimed when the set_timeout is sent and freed
343 when both messages are free. */
344 static atomic_t set_timeout_tofree = ATOMIC_INIT(0);
345 static DEFINE_MUTEX(set_timeout_lock);
346 static DECLARE_COMPLETION(set_timeout_wait);
347 static void set_timeout_free_smi(struct ipmi_smi_msg *msg)
349 if (atomic_dec_and_test(&set_timeout_tofree))
350 complete(&set_timeout_wait);
352 static void set_timeout_free_recv(struct ipmi_recv_msg *msg)
354 if (atomic_dec_and_test(&set_timeout_tofree))
355 complete(&set_timeout_wait);
357 static struct ipmi_smi_msg set_timeout_smi_msg =
359 .done = set_timeout_free_smi
361 static struct ipmi_recv_msg set_timeout_recv_msg =
363 .done = set_timeout_free_recv
366 static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg,
367 struct ipmi_recv_msg *recv_msg,
368 int *send_heartbeat_now)
370 struct kernel_ipmi_msg msg;
371 unsigned char data[6];
373 struct ipmi_system_interface_addr addr;
377 /* These can be cleared as we are setting the timeout. */
378 ipmi_start_timer_on_heartbeat = 0;
379 pretimeout_since_last_heartbeat = 0;
382 WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
384 if ((ipmi_version_major > 1)
385 || ((ipmi_version_major == 1) && (ipmi_version_minor >= 5)))
387 /* This is an IPMI 1.5-only feature. */
388 data[0] |= WDOG_DONT_STOP_ON_SET;
389 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
390 /* In ipmi 1.0, setting the timer stops the watchdog, we
391 need to start it back up again. */
396 WDOG_SET_TIMEOUT_ACT(data[1], ipmi_watchdog_state);
397 if ((pretimeout > 0) && (ipmi_watchdog_state != WDOG_TIMEOUT_NONE)) {
398 WDOG_SET_PRETIMEOUT_ACT(data[1], preaction_val);
399 data[2] = pretimeout;
401 WDOG_SET_PRETIMEOUT_ACT(data[1], WDOG_PRETIMEOUT_NONE);
402 data[2] = 0; /* No pretimeout. */
405 WDOG_SET_TIMEOUT(data[4], data[5], timeout);
407 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
408 addr.channel = IPMI_BMC_CHANNEL;
412 msg.cmd = IPMI_WDOG_SET_TIMER;
414 msg.data_len = sizeof(data);
415 rv = ipmi_request_supply_msgs(watchdog_user,
416 (struct ipmi_addr *) &addr,
424 printk(KERN_WARNING PFX "set timeout error: %d\n",
428 if (send_heartbeat_now)
429 *send_heartbeat_now = hbnow;
434 static int ipmi_set_timeout(int do_heartbeat)
436 int send_heartbeat_now;
440 /* We can only send one of these at a time. */
441 mutex_lock(&set_timeout_lock);
443 atomic_set(&set_timeout_tofree, 2);
445 rv = i_ipmi_set_timeout(&set_timeout_smi_msg,
446 &set_timeout_recv_msg,
447 &send_heartbeat_now);
449 mutex_unlock(&set_timeout_lock);
453 wait_for_completion(&set_timeout_wait);
455 mutex_unlock(&set_timeout_lock);
457 if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB)
458 || ((send_heartbeat_now)
459 && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY)))
460 rv = ipmi_heartbeat();
466 static void dummy_smi_free(struct ipmi_smi_msg *msg)
469 static void dummy_recv_free(struct ipmi_recv_msg *msg)
472 static struct ipmi_smi_msg panic_halt_smi_msg =
474 .done = dummy_smi_free
476 static struct ipmi_recv_msg panic_halt_recv_msg =
478 .done = dummy_recv_free
481 /* Special call, doesn't claim any locks. This is only to be called
482 at panic or halt time, in run-to-completion mode, when the caller
483 is the only CPU and the only thing that will be going is these IPMI
485 static void panic_halt_ipmi_set_timeout(void)
487 int send_heartbeat_now;
490 rv = i_ipmi_set_timeout(&panic_halt_smi_msg,
491 &panic_halt_recv_msg,
492 &send_heartbeat_now);
494 if (send_heartbeat_now)
495 panic_halt_ipmi_heartbeat();
499 /* We use a semaphore to make sure that only one thing can send a
500 heartbeat at one time, because we only have one copy of the data.
501 The semaphore is claimed when the set_timeout is sent and freed
502 when both messages are free. */
503 static atomic_t heartbeat_tofree = ATOMIC_INIT(0);
504 static DEFINE_MUTEX(heartbeat_lock);
505 static DECLARE_COMPLETION(heartbeat_wait);
506 static void heartbeat_free_smi(struct ipmi_smi_msg *msg)
508 if (atomic_dec_and_test(&heartbeat_tofree))
509 complete(&heartbeat_wait);
511 static void heartbeat_free_recv(struct ipmi_recv_msg *msg)
513 if (atomic_dec_and_test(&heartbeat_tofree))
514 complete(&heartbeat_wait);
516 static struct ipmi_smi_msg heartbeat_smi_msg =
518 .done = heartbeat_free_smi
520 static struct ipmi_recv_msg heartbeat_recv_msg =
522 .done = heartbeat_free_recv
525 static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg =
527 .done = dummy_smi_free
529 static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg =
531 .done = dummy_recv_free
534 static int ipmi_heartbeat(void)
536 struct kernel_ipmi_msg msg;
538 struct ipmi_system_interface_addr addr;
540 if (ipmi_ignore_heartbeat)
543 if (ipmi_start_timer_on_heartbeat) {
544 ipmi_watchdog_state = action_val;
545 return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
546 } else if (pretimeout_since_last_heartbeat) {
547 /* A pretimeout occurred, make sure we set the timeout.
548 We don't want to set the action, though, we want to
549 leave that alone (thus it can't be combined with the
551 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
554 mutex_lock(&heartbeat_lock);
556 atomic_set(&heartbeat_tofree, 2);
558 /* Don't reset the timer if we have the timer turned off, that
559 re-enables the watchdog. */
560 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) {
561 mutex_unlock(&heartbeat_lock);
565 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
566 addr.channel = IPMI_BMC_CHANNEL;
570 msg.cmd = IPMI_WDOG_RESET_TIMER;
573 rv = ipmi_request_supply_msgs(watchdog_user,
574 (struct ipmi_addr *) &addr,
582 mutex_unlock(&heartbeat_lock);
583 printk(KERN_WARNING PFX "heartbeat failure: %d\n",
588 /* Wait for the heartbeat to be sent. */
589 wait_for_completion(&heartbeat_wait);
591 if (heartbeat_recv_msg.msg.data[0] != 0) {
592 /* Got an error in the heartbeat response. It was already
593 reported in ipmi_wdog_msg_handler, but we should return
598 mutex_unlock(&heartbeat_lock);
603 static void panic_halt_ipmi_heartbeat(void)
605 struct kernel_ipmi_msg msg;
606 struct ipmi_system_interface_addr addr;
609 /* Don't reset the timer if we have the timer turned off, that
610 re-enables the watchdog. */
611 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
614 addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE;
615 addr.channel = IPMI_BMC_CHANNEL;
619 msg.cmd = IPMI_WDOG_RESET_TIMER;
622 ipmi_request_supply_msgs(watchdog_user,
623 (struct ipmi_addr *) &addr,
627 &panic_halt_heartbeat_smi_msg,
628 &panic_halt_heartbeat_recv_msg,
632 static struct watchdog_info ident =
634 .options = 0, /* WDIOF_SETTIMEOUT, */
635 .firmware_version = 1,
639 static int ipmi_ioctl(struct inode *inode, struct file *file,
640 unsigned int cmd, unsigned long arg)
642 void __user *argp = (void __user *)arg;
647 case WDIOC_GETSUPPORT:
648 i = copy_to_user(argp, &ident, sizeof(ident));
649 return i ? -EFAULT : 0;
651 case WDIOC_SETTIMEOUT:
652 i = copy_from_user(&val, argp, sizeof(int));
656 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
658 case WDIOC_GETTIMEOUT:
659 i = copy_to_user(argp, &timeout, sizeof(timeout));
664 case WDIOC_SET_PRETIMEOUT:
665 i = copy_from_user(&val, argp, sizeof(int));
669 return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
671 case WDIOC_GET_PRETIMEOUT:
672 i = copy_to_user(argp, &pretimeout, sizeof(pretimeout));
677 case WDIOC_KEEPALIVE:
678 return ipmi_heartbeat();
680 case WDIOC_SETOPTIONS:
681 i = copy_from_user(&val, argp, sizeof(int));
684 if (val & WDIOS_DISABLECARD)
686 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
687 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
688 ipmi_start_timer_on_heartbeat = 0;
691 if (val & WDIOS_ENABLECARD)
693 ipmi_watchdog_state = action_val;
694 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
698 case WDIOC_GETSTATUS:
700 i = copy_to_user(argp, &val, sizeof(val));
710 static ssize_t ipmi_write(struct file *file,
711 const char __user *buf,
721 /* In case it was set long ago */
724 for (i = 0; i != len; i++) {
727 if (get_user(c, buf + i))
733 rv = ipmi_heartbeat();
741 static ssize_t ipmi_read(struct file *file,
752 /* Reading returns if the pretimeout has gone off, and it only does
753 it once per pretimeout. */
754 spin_lock(&ipmi_read_lock);
756 if (file->f_flags & O_NONBLOCK) {
761 init_waitqueue_entry(&wait, current);
762 add_wait_queue(&read_q, &wait);
763 while (!data_to_read) {
764 set_current_state(TASK_INTERRUPTIBLE);
765 spin_unlock(&ipmi_read_lock);
767 spin_lock(&ipmi_read_lock);
769 remove_wait_queue(&read_q, &wait);
771 if (signal_pending(current)) {
779 spin_unlock(&ipmi_read_lock);
782 if (copy_to_user(buf, &data_to_read, 1))
791 static int ipmi_open(struct inode *ino, struct file *filep)
793 switch (iminor(ino)) {
795 if (test_and_set_bit(0, &ipmi_wdog_open))
798 /* Don't start the timer now, let it start on the
800 ipmi_start_timer_on_heartbeat = 1;
801 return nonseekable_open(ino, filep);
808 static unsigned int ipmi_poll(struct file *file, poll_table *wait)
810 unsigned int mask = 0;
812 poll_wait(file, &read_q, wait);
814 spin_lock(&ipmi_read_lock);
816 mask |= (POLLIN | POLLRDNORM);
817 spin_unlock(&ipmi_read_lock);
822 static int ipmi_fasync(int fd, struct file *file, int on)
826 result = fasync_helper(fd, file, on, &fasync_q);
831 static int ipmi_close(struct inode *ino, struct file *filep)
833 if (iminor(ino) == WATCHDOG_MINOR) {
834 if (expect_close == 42) {
835 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
836 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
839 "Unexpected close, not stopping watchdog!\n");
842 clear_bit(0, &ipmi_wdog_open);
845 ipmi_fasync (-1, filep, 0);
851 static const struct file_operations ipmi_wdog_fops = {
852 .owner = THIS_MODULE,
858 .release = ipmi_close,
859 .fasync = ipmi_fasync,
862 static struct miscdevice ipmi_wdog_miscdev = {
863 .minor = WATCHDOG_MINOR,
865 .fops = &ipmi_wdog_fops
868 static void ipmi_wdog_msg_handler(struct ipmi_recv_msg *msg,
871 if (msg->msg.data[0] != 0) {
872 printk(KERN_ERR PFX "response: Error %x on cmd %x\n",
877 ipmi_free_recv_msg(msg);
880 static void ipmi_wdog_pretimeout_handler(void *handler_data)
882 if (preaction_val != WDOG_PRETIMEOUT_NONE) {
883 if (preop_val == WDOG_PREOP_PANIC) {
884 if (atomic_inc_and_test(&preop_panic_excl))
885 panic("Watchdog pre-timeout");
886 } else if (preop_val == WDOG_PREOP_GIVE_DATA) {
887 spin_lock(&ipmi_read_lock);
889 wake_up_interruptible(&read_q);
890 kill_fasync(&fasync_q, SIGIO, POLL_IN);
892 spin_unlock(&ipmi_read_lock);
896 /* On some machines, the heartbeat will give
897 an error and not work unless we re-enable
898 the timer. So do so. */
899 pretimeout_since_last_heartbeat = 1;
902 static struct ipmi_user_hndl ipmi_hndlrs =
904 .ipmi_recv_hndl = ipmi_wdog_msg_handler,
905 .ipmi_watchdog_pretimeout = ipmi_wdog_pretimeout_handler
908 static void ipmi_register_watchdog(int ipmi_intf)
912 down_write(®ister_sem);
916 if ((ifnum_to_use >= 0) && (ifnum_to_use != ipmi_intf))
919 watchdog_ifnum = ipmi_intf;
921 rv = ipmi_create_user(ipmi_intf, &ipmi_hndlrs, NULL, &watchdog_user);
923 printk(KERN_CRIT PFX "Unable to register with ipmi\n");
927 ipmi_get_version(watchdog_user,
929 &ipmi_version_minor);
931 rv = misc_register(&ipmi_wdog_miscdev);
933 ipmi_destroy_user(watchdog_user);
934 watchdog_user = NULL;
935 printk(KERN_CRIT PFX "Unable to register misc device\n");
938 #ifdef HAVE_DIE_NMI_POST
939 if (nmi_handler_registered) {
940 int old_pretimeout = pretimeout;
941 int old_timeout = timeout;
942 int old_preop_val = preop_val;
944 /* Set the pretimeout to go off in a second and give
945 ourselves plenty of time to stop the timer. */
946 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
947 preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */
953 rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
955 printk(KERN_WARNING PFX "Error starting timer to"
956 " test NMI: 0x%x. The NMI pretimeout will"
957 " likely not work\n", rv);
964 if (testing_nmi != 2) {
965 printk(KERN_WARNING PFX "IPMI NMI didn't seem to"
966 " occur. The NMI pretimeout will"
967 " likely not work\n");
971 preop_val = old_preop_val;
972 pretimeout = old_pretimeout;
973 timeout = old_timeout;
978 up_write(®ister_sem);
980 if ((start_now) && (rv == 0)) {
981 /* Run from startup, so start the timer now. */
982 start_now = 0; /* Disable this function after first startup. */
983 ipmi_watchdog_state = action_val;
984 ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
985 printk(KERN_INFO PFX "Starting now!\n");
987 /* Stop the timer now. */
988 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
989 ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
993 static void ipmi_unregister_watchdog(int ipmi_intf)
997 down_write(®ister_sem);
1002 if (watchdog_ifnum != ipmi_intf)
1005 /* Make sure no one can call us any more. */
1006 misc_deregister(&ipmi_wdog_miscdev);
1008 /* Wait to make sure the message makes it out. The lower layer has
1009 pointers to our buffers, we want to make sure they are done before
1010 we release our memory. */
1011 while (atomic_read(&set_timeout_tofree))
1012 schedule_timeout_uninterruptible(1);
1014 /* Disconnect from IPMI. */
1015 rv = ipmi_destroy_user(watchdog_user);
1017 printk(KERN_WARNING PFX "error unlinking from IPMI: %d\n",
1020 watchdog_user = NULL;
1023 up_write(®ister_sem);
1026 #ifdef HAVE_DIE_NMI_POST
1028 ipmi_nmi(struct notifier_block *self, unsigned long val, void *data)
1030 if (val != DIE_NMI_POST)
1038 /* If we are not expecting a timeout, ignore it. */
1039 if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
1042 if (preaction_val != WDOG_PRETIMEOUT_NMI)
1045 /* If no one else handled the NMI, we assume it was the IPMI
1047 if (preop_val == WDOG_PREOP_PANIC) {
1048 /* On some machines, the heartbeat will give
1049 an error and not work unless we re-enable
1050 the timer. So do so. */
1051 pretimeout_since_last_heartbeat = 1;
1052 if (atomic_inc_and_test(&preop_panic_excl))
1053 panic(PFX "pre-timeout");
1059 static struct notifier_block ipmi_nmi_handler = {
1060 .notifier_call = ipmi_nmi
1064 static int wdog_reboot_handler(struct notifier_block *this,
1068 static int reboot_event_handled = 0;
1070 if ((watchdog_user) && (!reboot_event_handled)) {
1071 /* Make sure we only do this once. */
1072 reboot_event_handled = 1;
1074 if (code == SYS_DOWN || code == SYS_HALT) {
1075 /* Disable the WDT if we are shutting down. */
1076 ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
1077 panic_halt_ipmi_set_timeout();
1078 } else if (ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1079 /* Set a long timer to let the reboot happens, but
1080 reboot if it hangs, but only if the watchdog
1081 timer was already running. */
1084 ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
1085 panic_halt_ipmi_set_timeout();
1091 static struct notifier_block wdog_reboot_notifier = {
1092 .notifier_call = wdog_reboot_handler,
1097 static int wdog_panic_handler(struct notifier_block *this,
1098 unsigned long event,
1101 static int panic_event_handled = 0;
1103 /* On a panic, if we have a panic timeout, make sure to extend
1104 the watchdog timer to a reasonable value to complete the
1105 panic, if the watchdog timer is running. Plus the
1106 pretimeout is meaningless at panic time. */
1107 if (watchdog_user && !panic_event_handled &&
1108 ipmi_watchdog_state != WDOG_TIMEOUT_NONE) {
1109 /* Make sure we do this only once. */
1110 panic_event_handled = 1;
1114 panic_halt_ipmi_set_timeout();
1120 static struct notifier_block wdog_panic_notifier = {
1121 .notifier_call = wdog_panic_handler,
1123 .priority = 150 /* priority: INT_MAX >= x >= 0 */
1127 static void ipmi_new_smi(int if_num, struct device *device)
1129 ipmi_register_watchdog(if_num);
1132 static void ipmi_smi_gone(int if_num)
1134 ipmi_unregister_watchdog(if_num);
1137 static struct ipmi_smi_watcher smi_watcher =
1139 .owner = THIS_MODULE,
1140 .new_smi = ipmi_new_smi,
1141 .smi_gone = ipmi_smi_gone
1144 static int action_op(const char *inval, char *outval)
1147 strcpy(outval, action);
1152 if (strcmp(inval, "reset") == 0)
1153 action_val = WDOG_TIMEOUT_RESET;
1154 else if (strcmp(inval, "none") == 0)
1155 action_val = WDOG_TIMEOUT_NONE;
1156 else if (strcmp(inval, "power_cycle") == 0)
1157 action_val = WDOG_TIMEOUT_POWER_CYCLE;
1158 else if (strcmp(inval, "power_off") == 0)
1159 action_val = WDOG_TIMEOUT_POWER_DOWN;
1162 strcpy(action, inval);
1166 static int preaction_op(const char *inval, char *outval)
1169 strcpy(outval, preaction);
1174 if (strcmp(inval, "pre_none") == 0)
1175 preaction_val = WDOG_PRETIMEOUT_NONE;
1176 else if (strcmp(inval, "pre_smi") == 0)
1177 preaction_val = WDOG_PRETIMEOUT_SMI;
1178 #ifdef HAVE_DIE_NMI_POST
1179 else if (strcmp(inval, "pre_nmi") == 0)
1180 preaction_val = WDOG_PRETIMEOUT_NMI;
1182 else if (strcmp(inval, "pre_int") == 0)
1183 preaction_val = WDOG_PRETIMEOUT_MSG_INT;
1186 strcpy(preaction, inval);
1190 static int preop_op(const char *inval, char *outval)
1193 strcpy(outval, preop);
1198 if (strcmp(inval, "preop_none") == 0)
1199 preop_val = WDOG_PREOP_NONE;
1200 else if (strcmp(inval, "preop_panic") == 0)
1201 preop_val = WDOG_PREOP_PANIC;
1202 else if (strcmp(inval, "preop_give_data") == 0)
1203 preop_val = WDOG_PREOP_GIVE_DATA;
1206 strcpy(preop, inval);
1210 static void check_parms(void)
1212 #ifdef HAVE_DIE_NMI_POST
1216 if (preaction_val == WDOG_PRETIMEOUT_NMI) {
1218 if (preop_val == WDOG_PREOP_GIVE_DATA) {
1219 printk(KERN_WARNING PFX "Pretimeout op is to give data"
1220 " but NMI pretimeout is enabled, setting"
1221 " pretimeout op to none\n");
1222 preop_op("preop_none", NULL);
1226 if (do_nmi && !nmi_handler_registered) {
1227 rv = register_die_notifier(&ipmi_nmi_handler);
1229 printk(KERN_WARNING PFX
1230 "Can't register nmi handler\n");
1233 nmi_handler_registered = 1;
1234 } else if (!do_nmi && nmi_handler_registered) {
1235 unregister_die_notifier(&ipmi_nmi_handler);
1236 nmi_handler_registered = 0;
1241 static int __init ipmi_wdog_init(void)
1245 if (action_op(action, NULL)) {
1246 action_op("reset", NULL);
1247 printk(KERN_INFO PFX "Unknown action '%s', defaulting to"
1248 " reset\n", action);
1251 if (preaction_op(preaction, NULL)) {
1252 preaction_op("pre_none", NULL);
1253 printk(KERN_INFO PFX "Unknown preaction '%s', defaulting to"
1254 " none\n", preaction);
1257 if (preop_op(preop, NULL)) {
1258 preop_op("preop_none", NULL);
1259 printk(KERN_INFO PFX "Unknown preop '%s', defaulting to"
1265 register_reboot_notifier(&wdog_reboot_notifier);
1266 atomic_notifier_chain_register(&panic_notifier_list,
1267 &wdog_panic_notifier);
1269 rv = ipmi_smi_watcher_register(&smi_watcher);
1271 #ifdef HAVE_DIE_NMI_POST
1272 if (nmi_handler_registered)
1273 unregister_die_notifier(&ipmi_nmi_handler);
1275 atomic_notifier_chain_unregister(&panic_notifier_list,
1276 &wdog_panic_notifier);
1277 unregister_reboot_notifier(&wdog_reboot_notifier);
1278 printk(KERN_WARNING PFX "can't register smi watcher\n");
1282 printk(KERN_INFO PFX "driver initialized\n");
1287 static void __exit ipmi_wdog_exit(void)
1289 ipmi_smi_watcher_unregister(&smi_watcher);
1290 ipmi_unregister_watchdog(watchdog_ifnum);
1292 #ifdef HAVE_DIE_NMI_POST
1293 if (nmi_handler_registered)
1294 unregister_die_notifier(&ipmi_nmi_handler);
1297 atomic_notifier_chain_unregister(&panic_notifier_list,
1298 &wdog_panic_notifier);
1299 unregister_reboot_notifier(&wdog_reboot_notifier);
1301 module_exit(ipmi_wdog_exit);
1302 module_init(ipmi_wdog_init);
1303 MODULE_LICENSE("GPL");
1304 MODULE_AUTHOR("Corey Minyard <minyard@mvista.com>");
1305 MODULE_DESCRIPTION("watchdog timer based upon the IPMI interface.");