X-Git-Url: https://err.no/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=drivers%2Fchar%2Fipmi%2Fipmi_watchdog.c;h=8f45ca9235ad10fb41ece3d803d776843bb3ea53;hb=75659ca0c10992dcb39258518368a0f6f56e935d;hp=147c12047cf3d83b6b345c4b8348f43e4ffc6a5d;hpb=f697b677620d04d8c77841745727de85f7e948b1;p=linux-2.6 diff --git a/drivers/char/ipmi/ipmi_watchdog.c b/drivers/char/ipmi/ipmi_watchdog.c index 147c12047c..8f45ca9235 100644 --- a/drivers/char/ipmi/ipmi_watchdog.c +++ b/drivers/char/ipmi/ipmi_watchdog.c @@ -61,7 +61,8 @@ how it will work. So in the unlikely event that another architecture supports this, we can figure out a good generic mechanism for it at that time. */ -#define HAVE_DIE_NMI_POST +#include +#define HAVE_DIE_NMI #endif #define PFX "IPMI Watchdog: " @@ -174,8 +175,6 @@ static char expect_close; static int ifnum_to_use = -1; -static DECLARE_RWSEM(register_sem); - /* Parameters to ipmi_set_timeout */ #define IPMI_SET_TIMEOUT_NO_HB 0 #define IPMI_SET_TIMEOUT_HB_IF_NECESSARY 1 @@ -201,11 +200,9 @@ static int set_param_int(const char *val, struct kernel_param *kp) if (endp == val) return -EINVAL; - down_read(®ister_sem); *((int *)kp->arg) = l; if (watchdog_user) rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); - up_read(®ister_sem); return rv; } @@ -234,17 +231,15 @@ static int set_param_str(const char *val, struct kernel_param *kp) s = strstrip(valcp); - down_read(®ister_sem); rv = fn(s, NULL); if (rv) - goto out_unlock; + goto out; check_parms(); if (watchdog_user) rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); - out_unlock: - up_read(®ister_sem); + out: return rv; } @@ -327,14 +322,12 @@ static unsigned char ipmi_version_minor; /* If a pretimeout occurs, this is used to allow only one panic to happen. */ static atomic_t preop_panic_excl = ATOMIC_INIT(-1); -#ifdef HAVE_DIE_NMI_POST +#ifdef HAVE_DIE_NMI static int testing_nmi; static int nmi_handler_registered; #endif static int ipmi_heartbeat(void); -static void panic_halt_ipmi_heartbeat(void); - /* We use a mutex to make sure that only one thing can send a set timeout at one time, because we only have one copy of the data. @@ -374,7 +367,6 @@ static int i_ipmi_set_timeout(struct ipmi_smi_msg *smi_msg, /* These can be cleared as we are setting the timeout. */ - ipmi_start_timer_on_heartbeat = 0; pretimeout_since_last_heartbeat = 0; data[0] = 0; @@ -462,19 +454,64 @@ out: return rv; } -static void dummy_smi_free(struct ipmi_smi_msg *msg) +static atomic_t panic_done_count = ATOMIC_INIT(0); + +static void panic_smi_free(struct ipmi_smi_msg *msg) { + atomic_dec(&panic_done_count); } -static void dummy_recv_free(struct ipmi_recv_msg *msg) +static void panic_recv_free(struct ipmi_recv_msg *msg) { + atomic_dec(&panic_done_count); } + +static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = +{ + .done = panic_smi_free +}; +static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = +{ + .done = panic_recv_free +}; + +static void panic_halt_ipmi_heartbeat(void) +{ + struct kernel_ipmi_msg msg; + struct ipmi_system_interface_addr addr; + int rv; + + /* Don't reset the timer if we have the timer turned off, that + re-enables the watchdog. */ + if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) + return; + + addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; + addr.channel = IPMI_BMC_CHANNEL; + addr.lun = 0; + + msg.netfn = 0x06; + msg.cmd = IPMI_WDOG_RESET_TIMER; + msg.data = NULL; + msg.data_len = 0; + rv = ipmi_request_supply_msgs(watchdog_user, + (struct ipmi_addr *) &addr, + 0, + &msg, + NULL, + &panic_halt_heartbeat_smi_msg, + &panic_halt_heartbeat_recv_msg, + 1); + if (!rv) + atomic_add(2, &panic_done_count); +} + static struct ipmi_smi_msg panic_halt_smi_msg = { - .done = dummy_smi_free + .done = panic_smi_free }; static struct ipmi_recv_msg panic_halt_recv_msg = { - .done = dummy_recv_free + .done = panic_recv_free }; /* Special call, doesn't claim any locks. This is only to be called @@ -486,13 +523,21 @@ static void panic_halt_ipmi_set_timeout(void) int send_heartbeat_now; int rv; + /* Wait for the messages to be free. */ + while (atomic_read(&panic_done_count) != 0) + ipmi_poll_interface(watchdog_user); rv = i_ipmi_set_timeout(&panic_halt_smi_msg, &panic_halt_recv_msg, &send_heartbeat_now); if (!rv) { + atomic_add(2, &panic_done_count); if (send_heartbeat_now) panic_halt_ipmi_heartbeat(); - } + } else + printk(KERN_WARNING PFX + "Unable to extend the watchdog timeout."); + while (atomic_read(&panic_done_count) != 0) + ipmi_poll_interface(watchdog_user); } /* We use a semaphore to make sure that only one thing can send a @@ -521,15 +566,6 @@ static struct ipmi_recv_msg heartbeat_recv_msg = .done = heartbeat_free_recv }; -static struct ipmi_smi_msg panic_halt_heartbeat_smi_msg = -{ - .done = dummy_smi_free -}; -static struct ipmi_recv_msg panic_halt_heartbeat_recv_msg = -{ - .done = dummy_recv_free -}; - static int ipmi_heartbeat(void) { struct kernel_ipmi_msg msg; @@ -540,6 +576,7 @@ static int ipmi_heartbeat(void) return 0; if (ipmi_start_timer_on_heartbeat) { + ipmi_start_timer_on_heartbeat = 0; ipmi_watchdog_state = action_val; return ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB); } else if (pretimeout_since_last_heartbeat) { @@ -599,35 +636,6 @@ static int ipmi_heartbeat(void) return rv; } -static void panic_halt_ipmi_heartbeat(void) -{ - struct kernel_ipmi_msg msg; - struct ipmi_system_interface_addr addr; - - - /* Don't reset the timer if we have the timer turned off, that - re-enables the watchdog. */ - if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE) - return; - - addr.addr_type = IPMI_SYSTEM_INTERFACE_ADDR_TYPE; - addr.channel = IPMI_BMC_CHANNEL; - addr.lun = 0; - - msg.netfn = 0x06; - msg.cmd = IPMI_WDOG_RESET_TIMER; - msg.data = NULL; - msg.data_len = 0; - ipmi_request_supply_msgs(watchdog_user, - (struct ipmi_addr *) &addr, - 0, - &msg, - NULL, - &panic_halt_heartbeat_smi_msg, - &panic_halt_heartbeat_recv_msg, - 1); -} - static struct watchdog_info ident = { .options = 0, /* WDIOF_SETTIMEOUT, */ @@ -661,6 +669,7 @@ static int ipmi_ioctl(struct inode *inode, struct file *file, return 0; case WDIOC_SET_PRETIMEOUT: + case WDIOC_SETPRETIMEOUT: i = copy_from_user(&val, argp, sizeof(int)); if (i) return -EFAULT; @@ -668,6 +677,7 @@ static int ipmi_ioctl(struct inode *inode, struct file *file, return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY); case WDIOC_GET_PRETIMEOUT: + case WDIOC_GETPRETIMEOUT: i = copy_to_user(argp, &pretimeout, sizeof(pretimeout)); if (i) return -EFAULT; @@ -908,7 +918,6 @@ static void ipmi_register_watchdog(int ipmi_intf) { int rv = -EBUSY; - down_write(®ister_sem); if (watchdog_user) goto out; @@ -934,7 +943,7 @@ static void ipmi_register_watchdog(int ipmi_intf) printk(KERN_CRIT PFX "Unable to register misc device\n"); } -#ifdef HAVE_DIE_NMI_POST +#ifdef HAVE_DIE_NMI if (nmi_handler_registered) { int old_pretimeout = pretimeout; int old_timeout = timeout; @@ -974,8 +983,6 @@ static void ipmi_register_watchdog(int ipmi_intf) #endif out: - up_write(®ister_sem); - if ((start_now) && (rv == 0)) { /* Run from startup, so start the timer now. */ start_now = 0; /* Disable this function after first startup. */ @@ -993,8 +1000,6 @@ static void ipmi_unregister_watchdog(int ipmi_intf) { int rv; - down_write(®ister_sem); - if (!watchdog_user) goto out; @@ -1019,16 +1024,29 @@ static void ipmi_unregister_watchdog(int ipmi_intf) watchdog_user = NULL; out: - up_write(®ister_sem); + return; } -#ifdef HAVE_DIE_NMI_POST +#ifdef HAVE_DIE_NMI static int ipmi_nmi(struct notifier_block *self, unsigned long val, void *data) { - if (val != DIE_NMI_POST) + struct die_args *args = data; + + if (val != DIE_NMI) + return NOTIFY_OK; + + /* Hack, if it's a memory or I/O error, ignore it. */ + if (args->err & 0xc0) return NOTIFY_OK; + /* + * If we get here, it's an NMI that's not a memory or I/O + * error. We can't truly tell if it's from IPMI or not + * without sending a message, and sending a message is almost + * impossible because of locking. + */ + if (testing_nmi) { testing_nmi = 2; return NOTIFY_STOP; @@ -1070,7 +1088,7 @@ static int wdog_reboot_handler(struct notifier_block *this, /* Make sure we only do this once. */ reboot_event_handled = 1; - if (code == SYS_DOWN || code == SYS_HALT) { + if (code == SYS_POWER_OFF || code == SYS_HALT) { /* Disable the WDT if we are shutting down. */ ipmi_watchdog_state = WDOG_TIMEOUT_NONE; panic_halt_ipmi_set_timeout(); @@ -1174,7 +1192,7 @@ static int preaction_op(const char *inval, char *outval) preaction_val = WDOG_PRETIMEOUT_NONE; else if (strcmp(inval, "pre_smi") == 0) preaction_val = WDOG_PRETIMEOUT_SMI; -#ifdef HAVE_DIE_NMI_POST +#ifdef HAVE_DIE_NMI else if (strcmp(inval, "pre_nmi") == 0) preaction_val = WDOG_PRETIMEOUT_NMI; #endif @@ -1208,7 +1226,7 @@ static int preop_op(const char *inval, char *outval) static void check_parms(void) { -#ifdef HAVE_DIE_NMI_POST +#ifdef HAVE_DIE_NMI int do_nmi = 0; int rv; @@ -1267,7 +1285,7 @@ static int __init ipmi_wdog_init(void) rv = ipmi_smi_watcher_register(&smi_watcher); if (rv) { -#ifdef HAVE_DIE_NMI_POST +#ifdef HAVE_DIE_NMI if (nmi_handler_registered) unregister_die_notifier(&ipmi_nmi_handler); #endif @@ -1288,7 +1306,7 @@ static void __exit ipmi_wdog_exit(void) ipmi_smi_watcher_unregister(&smi_watcher); ipmi_unregister_watchdog(watchdog_ifnum); -#ifdef HAVE_DIE_NMI_POST +#ifdef HAVE_DIE_NMI if (nmi_handler_registered) unregister_die_notifier(&ipmi_nmi_handler); #endif