#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>
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
+#define HAVE_DIE_NMI
#endif
#define PFX "IPMI Watchdog: "
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
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;
}
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;
}
/* 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.
/* These can be cleared as we are setting the timeout. */
- ipmi_start_timer_on_heartbeat = 0;
pretimeout_since_last_heartbeat = 0;
data[0] = 0;
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
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
.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;
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) {
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, */
{
int rv = -EBUSY;
- down_write(®ister_sem);
if (watchdog_user)
goto out;
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;
#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. */
{
int rv;
- down_write(®ister_sem);
-
if (!watchdog_user)
goto out;
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;
/* 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();
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
static void check_parms(void)
{
-#ifdef HAVE_DIE_NMI_POST
+#ifdef HAVE_DIE_NMI
int do_nmi = 0;
int rv;
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
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