move die notifier handling to common code
[linux-2.6-block.git] / drivers / char / ipmi / ipmi_watchdog.c
index 90fb2a541916e2efba863662d4e3cffa35698af4..b6a972ed5bb32590aaf5bef2dc91b14dc159ab05 100644 (file)
@@ -39,6 +39,7 @@
 #include <linux/miscdevice.h>
 #include <linux/init.h>
 #include <linux/completion.h>
+#include <linux/kdebug.h>
 #include <linux/rwsem.h>
 #include <linux/errno.h>
 #include <asm/uaccess.h>
 #include <linux/poll.h>
 #include <linux/string.h>
 #include <linux/ctype.h>
+#include <linux/delay.h>
 #include <asm/atomic.h>
-#ifdef CONFIG_X86_LOCAL_APIC
-#include <asm/apic.h>
+
+#ifdef CONFIG_X86
+/* This is ugly, but I've determined that x86 is the only architecture
+   that can reasonably support the IPMI NMI watchdog timeout at this
+   time.  If another architecture adds this capability somehow, it
+   will have to be a somewhat different mechanism and I have no idea
+   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. */
+#include <asm/kdebug.h>
+#define HAVE_DIE_NMI_POST
 #endif
 
 #define        PFX "IPMI Watchdog: "
 
 static int nowayout = WATCHDOG_NOWAYOUT;
 
-static ipmi_user_t watchdog_user = NULL;
+static ipmi_user_t watchdog_user;
 static int watchdog_ifnum;
 
 /* Default the timeout to 10 seconds. */
 static int timeout = 10;
 
 /* The pre-timeout is disabled by default. */
-static int pretimeout = 0;
+static int pretimeout;
 
 /* Default action is to reset the board on a timeout. */
 static unsigned char action_val = WDOG_TIMEOUT_RESET;
@@ -156,10 +167,10 @@ static unsigned char preop_val = WDOG_PREOP_NONE;
 
 static char preop[16] = "preop_none";
 static DEFINE_SPINLOCK(ipmi_read_lock);
-static char data_to_read = 0;
+static char data_to_read;
 static DECLARE_WAIT_QUEUE_HEAD(read_q);
-static struct fasync_struct *fasync_q = NULL;
-static char pretimeout_since_last_heartbeat = 0;
+static struct fasync_struct *fasync_q;
+static char pretimeout_since_last_heartbeat;
 static char expect_close;
 
 static int ifnum_to_use = -1;
@@ -177,7 +188,7 @@ static void ipmi_unregister_watchdog(int ipmi_intf);
 
 /* If true, the driver will start running as soon as it is configured
    and ready. */
-static int start_now = 0;
+static int start_now;
 
 static int set_param_int(const char *val, struct kernel_param *kp)
 {
@@ -216,13 +227,13 @@ static int set_param_str(const char *val, struct kernel_param *kp)
 {
        action_fn  fn = (action_fn) kp->arg;
        int        rv = 0;
-       char       *dup, *s;
+       char       valcp[16];
+       char       *s;
 
-       dup = kstrdup(val, GFP_KERNEL);
-       if (!dup)
-               return -ENOMEM;
+       strncpy(valcp, val, 16);
+       valcp[15] = '\0';
 
-       s = strstrip(dup);
+       s = strstrip(valcp);
 
        down_read(&register_sem);
        rv = fn(s, NULL);
@@ -235,7 +246,6 @@ static int set_param_str(const char *val, struct kernel_param *kp)
 
  out_unlock:
        up_read(&register_sem);
-       kfree(dup);
        return rv;
 }
 
@@ -300,16 +310,16 @@ MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started "
 static unsigned char ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
 
 /* If shutting down via IPMI, we ignore the heartbeat. */
-static int ipmi_ignore_heartbeat = 0;
+static int ipmi_ignore_heartbeat;
 
 /* Is someone using the watchdog?  Only one user is allowed. */
-static unsigned long ipmi_wdog_open = 0;
+static unsigned long ipmi_wdog_open;
 
 /* If set to 1, the heartbeat command will set the state to reset and
    start the timer.  The timer doesn't normally run when the driver is
    first opened until the heartbeat is set the first time, this
    variable is used to accomplish this. */
-static int ipmi_start_timer_on_heartbeat = 0;
+static int ipmi_start_timer_on_heartbeat;
 
 /* IPMI version of the BMC. */
 static unsigned char ipmi_version_major;
@@ -318,6 +328,11 @@ 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
+static int testing_nmi;
+static int nmi_handler_registered;
+#endif
+
 static int ipmi_heartbeat(void);
 static void panic_halt_ipmi_heartbeat(void);
 
@@ -359,6 +374,10 @@ static int i_ipmi_set_timeout(struct ipmi_smi_msg  *smi_msg,
        int                               hbnow = 0;
 
 
+       /* These can be cleared as we are setting the timeout. */
+       ipmi_start_timer_on_heartbeat = 0;
+       pretimeout_since_last_heartbeat = 0;
+
        data[0] = 0;
        WDOG_SET_TIMER_USE(data[0], WDOG_TIMER_USE_SMS_OS);
 
@@ -433,13 +452,12 @@ static int ipmi_set_timeout(int do_heartbeat)
 
        wait_for_completion(&set_timeout_wait);
 
+       mutex_unlock(&set_timeout_lock);
+
        if ((do_heartbeat == IPMI_SET_TIMEOUT_FORCE_HB)
            || ((send_heartbeat_now)
                && (do_heartbeat == IPMI_SET_TIMEOUT_HB_IF_NECESSARY)))
-       {
                rv = ipmi_heartbeat();
-       }
-       mutex_unlock(&set_timeout_lock);
 
 out:
        return rv;
@@ -519,12 +537,10 @@ static int ipmi_heartbeat(void)
        int                               rv;
        struct ipmi_system_interface_addr addr;
 
-       if (ipmi_ignore_heartbeat) {
+       if (ipmi_ignore_heartbeat)
                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) {
@@ -532,7 +548,6 @@ static int ipmi_heartbeat(void)
                   We don't want to set the action, though, we want to
                   leave that alone (thus it can't be combined with the
                   above operation. */
-               pretimeout_since_last_heartbeat = 0;
                return ipmi_set_timeout(IPMI_SET_TIMEOUT_HB_IF_NECESSARY);
        }
 
@@ -920,6 +935,45 @@ static void ipmi_register_watchdog(int ipmi_intf)
                printk(KERN_CRIT PFX "Unable to register misc device\n");
        }
 
+#ifdef HAVE_DIE_NMI_POST
+       if (nmi_handler_registered) {
+               int old_pretimeout = pretimeout;
+               int old_timeout = timeout;
+               int old_preop_val = preop_val;
+
+               /* Set the pretimeout to go off in a second and give
+                  ourselves plenty of time to stop the timer. */
+               ipmi_watchdog_state = WDOG_TIMEOUT_RESET;
+               preop_val = WDOG_PREOP_NONE; /* Make sure nothing happens */
+               pretimeout = 99;
+               timeout = 100;
+
+               testing_nmi = 1;
+
+               rv = ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
+               if (rv) {
+                       printk(KERN_WARNING PFX "Error starting timer to"
+                              " test NMI: 0x%x.  The NMI pretimeout will"
+                              " likely not work\n", rv);
+                       rv = 0;
+                       goto out_restore;
+               }
+
+               msleep(1500);
+
+               if (testing_nmi != 2) {
+                       printk(KERN_WARNING PFX "IPMI NMI didn't seem to"
+                              " occur.  The NMI pretimeout will"
+                              " likely not work\n");
+               }
+       out_restore:
+               testing_nmi = 0;
+               preop_val = old_preop_val;
+               pretimeout = old_pretimeout;
+               timeout = old_timeout;
+       }
+#endif
+
  out:
        up_write(&register_sem);
 
@@ -929,6 +983,10 @@ static void ipmi_register_watchdog(int ipmi_intf)
                ipmi_watchdog_state = action_val;
                ipmi_set_timeout(IPMI_SET_TIMEOUT_FORCE_HB);
                printk(KERN_INFO PFX "Starting now!\n");
+       } else {
+               /* Stop the timer now. */
+               ipmi_watchdog_state = WDOG_TIMEOUT_NONE;
+               ipmi_set_timeout(IPMI_SET_TIMEOUT_NO_HB);
        }
 }
 
@@ -965,17 +1023,28 @@ static void ipmi_unregister_watchdog(int ipmi_intf)
        up_write(&register_sem);
 }
 
-#ifdef HAVE_NMI_HANDLER
+#ifdef HAVE_DIE_NMI_POST
 static int
-ipmi_nmi(void *dev_id, int cpu, int handled)
+ipmi_nmi(struct notifier_block *self, unsigned long val, void *data)
 {
+       if (val != DIE_NMI_POST)
+               return NOTIFY_OK;
+
+       if (testing_nmi) {
+               testing_nmi = 2;
+               return NOTIFY_STOP;
+       }
+
         /* If we are not expecting a timeout, ignore it. */
        if (ipmi_watchdog_state == WDOG_TIMEOUT_NONE)
-               return NOTIFY_DONE;
+               return NOTIFY_OK;
+
+       if (preaction_val != WDOG_PRETIMEOUT_NMI)
+               return NOTIFY_OK;
 
        /* If no one else handled the NMI, we assume it was the IPMI
            watchdog. */
-       if ((!handled) && (preop_val == WDOG_PREOP_PANIC)) {
+       if (preop_val == WDOG_PREOP_PANIC) {
                /* On some machines, the heartbeat will give
                   an error and not work unless we re-enable
                   the timer.   So do so. */
@@ -984,18 +1053,12 @@ ipmi_nmi(void *dev_id, int cpu, int handled)
                        panic(PFX "pre-timeout");
        }
 
-       return NOTIFY_DONE;
+       return NOTIFY_STOP;
 }
 
-static struct nmi_handler ipmi_nmi_handler =
-{
-       .link     = LIST_HEAD_INIT(ipmi_nmi_handler.link),
-       .dev_name = "ipmi_watchdog",
-       .dev_id   = NULL,
-       .handler  = ipmi_nmi,
-       .priority = 0, /* Call us last. */
+static struct notifier_block ipmi_nmi_handler = {
+       .notifier_call = ipmi_nmi
 };
-int nmi_handler_registered;
 #endif
 
 static int wdog_reboot_handler(struct notifier_block *this,
@@ -1112,7 +1175,7 @@ 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_NMI_HANDLER
+#ifdef HAVE_DIE_NMI_POST
        else if (strcmp(inval, "pre_nmi") == 0)
                preaction_val = WDOG_PRETIMEOUT_NMI;
 #endif
@@ -1146,7 +1209,7 @@ static int preop_op(const char *inval, char *outval)
 
 static void check_parms(void)
 {
-#ifdef HAVE_NMI_HANDLER
+#ifdef HAVE_DIE_NMI_POST
        int do_nmi = 0;
        int rv;
 
@@ -1159,20 +1222,9 @@ static void check_parms(void)
                        preop_op("preop_none", NULL);
                        do_nmi = 0;
                }
-#ifdef CONFIG_X86_LOCAL_APIC
-               if (nmi_watchdog == NMI_IO_APIC) {
-                       printk(KERN_WARNING PFX "nmi_watchdog is set to IO APIC"
-                              " mode (value is %d), that is incompatible"
-                              " with using NMI in the IPMI watchdog."
-                              " Disabling IPMI nmi pretimeout.\n",
-                              nmi_watchdog);
-                       preaction_val = WDOG_PRETIMEOUT_NONE;
-                       do_nmi = 0;
-               }
-#endif
        }
        if (do_nmi && !nmi_handler_registered) {
-               rv = request_nmi(&ipmi_nmi_handler);
+               rv = register_die_notifier(&ipmi_nmi_handler);
                if (rv) {
                        printk(KERN_WARNING PFX
                               "Can't register nmi handler\n");
@@ -1180,7 +1232,7 @@ static void check_parms(void)
                } else
                        nmi_handler_registered = 1;
        } else if (!do_nmi && nmi_handler_registered) {
-               release_nmi(&ipmi_nmi_handler);
+               unregister_die_notifier(&ipmi_nmi_handler);
                nmi_handler_registered = 0;
        }
 #endif
@@ -1216,9 +1268,9 @@ static int __init ipmi_wdog_init(void)
 
        rv = ipmi_smi_watcher_register(&smi_watcher);
        if (rv) {
-#ifdef HAVE_NMI_HANDLER
-               if (preaction_val == WDOG_PRETIMEOUT_NMI)
-                       release_nmi(&ipmi_nmi_handler);
+#ifdef HAVE_DIE_NMI_POST
+               if (nmi_handler_registered)
+                       unregister_die_notifier(&ipmi_nmi_handler);
 #endif
                atomic_notifier_chain_unregister(&panic_notifier_list,
                                                 &wdog_panic_notifier);
@@ -1237,9 +1289,9 @@ static void __exit ipmi_wdog_exit(void)
        ipmi_smi_watcher_unregister(&smi_watcher);
        ipmi_unregister_watchdog(watchdog_ifnum);
 
-#ifdef HAVE_NMI_HANDLER
+#ifdef HAVE_DIE_NMI_POST
        if (nmi_handler_registered)
-               release_nmi(&ipmi_nmi_handler);
+               unregister_die_notifier(&ipmi_nmi_handler);
 #endif
 
        atomic_notifier_chain_unregister(&panic_notifier_list,