struct lc_watchdog {
cfs_timer_t lcw_timer; /* kernel timer */
struct list_head lcw_list;
- struct timeval lcw_last_touched;
+ cfs_time_t lcw_last_touched;
cfs_task_t *lcw_task;
void (*lcw_callback)(pid_t, void *);
void *lcw_data;
pid_t lcw_pid;
- cfs_duration_t lcw_time; /* time until watchdog fires, jiffies */
enum {
LC_WATCHDOG_DISABLED,
static void
lcw_dump(struct lc_watchdog *lcw)
{
- cfs_task_t *tsk;
+ ENTRY;
#if defined(HAVE_TASKLIST_LOCK)
read_lock(&tasklist_lock);
#elif defined(HAVE_TASK_RCU)
rcu_read_lock();
#else
CERROR("unable to dump stack because of missing export\n");
- return;
+ RETURN_EXIT;
#endif
- ENTRY;
-
- tsk = find_task_by_pid(lcw->lcw_pid);
-
- if (tsk == NULL) {
+ if (lcw->lcw_task == NULL) {
CWARN("Process " LPPID " was not found in the task list; "
- "watchdog callback may be incomplete\n", lcw->lcw_pid);
- } else if (tsk != lcw->lcw_task) {
- CWARN("The current process " LPPID " did not set the watchdog; "
- "watchdog callback may be incomplete\n", lcw->lcw_pid);
+ "watchdog callback may be incomplete\n",
+ (int)lcw->lcw_pid);
} else {
- libcfs_debug_dumpstack(tsk);
+ libcfs_debug_dumpstack(lcw->lcw_task);
}
#if defined(HAVE_TASKLIST_LOCK)
struct lc_watchdog *lcw = (struct lc_watchdog *)data;
cfs_time_t current_time;
cfs_duration_t delta_time;
+ struct timeval timediff;
ENTRY;
lcw->lcw_state = LC_WATCHDOG_EXPIRED;
current_time = cfs_time_current();
+ delta_time = cfs_time_sub(current_time, lcw->lcw_last_touched);
+ cfs_duration_usec(delta_time, &timediff);
+
/* Check to see if we should throttle the watchdog timer to avoid
* too many dumps going to the console thus triggering an NMI.
* Normally we would not hold the spin lock over the CWARN but in
* this case we hold it to ensure non ratelimited lcw_dumps are not
* interleaved on the console making them hard to read. */
spin_lock_bh(&lcw_last_watchdog_lock);
- delta_time = cfs_duration_sec(current_time - lcw_last_watchdog_time);
+ delta_time = cfs_duration_sec(cfs_time_sub(current_time,
+ lcw_last_watchdog_time));
- if (delta_time < libcfs_watchdog_ratelimit && lcw_recent_watchdog_count > 3) {
+ if (delta_time < libcfs_watchdog_ratelimit &&
+ lcw_recent_watchdog_count > 3) {
CWARN("Refusing to fire watchdog for pid %d: it was inactive "
- "for %ldms. Rate limiting 1 per %d seconds.\n",
- (int)lcw->lcw_pid,cfs_duration_sec(lcw->lcw_time) * 1000,
- libcfs_watchdog_ratelimit);
+ "for %lu.%.02lus. Rate limiting 3 per %d seconds.\n",
+ (int)lcw->lcw_pid, timediff.tv_sec,
+ timediff.tv_usec / 10000, libcfs_watchdog_ratelimit);
} else {
if (delta_time < libcfs_watchdog_ratelimit) {
lcw_recent_watchdog_count++;
/* This warning should appear on the console, but may not get
* into the logs since we're running in a softirq handler */
- CWARN("Watchdog triggered for pid %d: it was inactive for %lds\n",
- (int)lcw->lcw_pid, cfs_duration_sec(lcw->lcw_time));
+ CWARN("Watchdog triggered for pid %d: it was inactive for "
+ "%lu.%.02lus\n", (int)lcw->lcw_pid, timediff.tv_sec,
+ timediff.tv_usec / 10000);
lcw_dump(lcw);
}
list_del_init(&lcw->lcw_list);
spin_unlock_bh(&lcw_pending_timers_lock);
- CDEBUG(D_INFO, "found lcw for pid " LPPID ": inactive for "
- "%lds\n", lcw->lcw_pid, cfs_duration_sec(lcw->lcw_time));
+ CDEBUG(D_INFO, "found lcw for pid " LPPID "\n", lcw->lcw_pid);
if (lcw->lcw_state != LC_WATCHDOG_DISABLED)
lcw->lcw_callback(lcw->lcw_pid, lcw->lcw_data);
EXIT;
}
-struct lc_watchdog *lc_watchdog_add(int timeout_ms,
+struct lc_watchdog *lc_watchdog_add(int timeout,
void (*callback)(pid_t, void *),
void *data)
{
lcw->lcw_task = cfs_current();
lcw->lcw_pid = cfs_curproc_pid();
- lcw->lcw_time = cfs_time_seconds(timeout_ms) / 1000;
lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
lcw->lcw_data = data;
lcw->lcw_state = LC_WATCHDOG_DISABLED;
/* Keep this working in case we enable them by default */
if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
- do_gettimeofday(&lcw->lcw_last_touched);
- cfs_timer_arm(&lcw->lcw_timer, lcw->lcw_time +
+ lcw->lcw_last_touched = cfs_time_current();
+ cfs_timer_arm(&lcw->lcw_timer, cfs_time_seconds(timeout) +
cfs_time_current());
}
static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
{
- struct timeval newtime;
- struct timeval timediff;
+ cfs_time_t newtime = cfs_time_current();;
- do_gettimeofday(&newtime);
if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
- cfs_timeval_sub(&newtime, &lcw->lcw_last_touched, &timediff);
- CWARN("Expired watchdog for pid " LPPID " %s after %lu.%.4lus\n",
- lcw->lcw_pid,
- message,
- timediff.tv_sec,
- timediff.tv_usec / 100);
+ struct timeval timediff;
+ cfs_time_t delta_time = cfs_time_sub(newtime,
+ lcw->lcw_last_touched);
+ cfs_duration_usec(delta_time, &timediff);
+
+ CWARN("Expired watchdog for pid " LPPID " %s after %lu.%.02lus\n",
+ lcw->lcw_pid, message, timediff.tv_sec,
+ timediff.tv_usec / 10000);
}
lcw->lcw_last_touched = newtime;
}
-void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
+void lc_watchdog_touch(struct lc_watchdog *lcw, int timeout)
{
ENTRY;
LASSERT(lcw != NULL);
lcw->lcw_state = LC_WATCHDOG_ENABLED;
cfs_timer_arm(&lcw->lcw_timer, cfs_time_current() +
- cfs_time_seconds(timeout_ms) / 1000);
+ cfs_time_seconds(timeout));
EXIT;
}
-EXPORT_SYMBOL(lc_watchdog_touch_ms);
-
-/* deprecated - use above instead */
-void lc_watchdog_touch(struct lc_watchdog *lcw)
-{
- lc_watchdog_touch_ms(lcw, (int)cfs_duration_sec(lcw->lcw_time) * 1000);
-}
EXPORT_SYMBOL(lc_watchdog_touch);
void lc_watchdog_disable(struct lc_watchdog *lcw)
#else /* !defined(WITH_WATCHDOG) */
-struct lc_watchdog *lc_watchdog_add(int timeout_ms,
+struct lc_watchdog *lc_watchdog_add(int timeout,
void (*callback)(pid_t pid, void *),
void *data)
{
}
EXPORT_SYMBOL(lc_watchdog_add);
-void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
-{
-}
-EXPORT_SYMBOL(lc_watchdog_touch_ms);
-
-void lc_watchdog_touch(struct lc_watchdog *lcw)
+void lc_watchdog_touch(struct lc_watchdog *lcw, int timeout)
{
}
EXPORT_SYMBOL(lc_watchdog_touch);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff