[PATCH] improve timekeeping resume robustness
authorjohn stultz <johnstul@us.ibm.com>
Fri, 14 Jul 2006 07:24:17 +0000 (00:24 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Sat, 15 Jul 2006 04:53:54 +0000 (21:53 -0700)
Resolve problems seen w/ APM suspend.

Due to resume initialization ordering, its possible we could get a timer
interrupt before the timekeeping resume() function is called.  This patch
ensures we don't do any timekeeping accounting before we're fully resumed.

(akpm: fixes the machine-freezes-on-APM-resume bug)

Signed-off-by: John Stultz <johnstul@us.ibm.com>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
kernel/timer.c

index acfa557..05809c2 100644 (file)
@@ -969,6 +969,7 @@ void __init timekeeping_init(void)
 }
 
 
+static int timekeeping_suspended;
 /*
  * timekeeping_resume - Resumes the generic timekeeping subsystem.
  * @dev:       unused
@@ -984,6 +985,18 @@ static int timekeeping_resume(struct sys_device *dev)
        write_seqlock_irqsave(&xtime_lock, flags);
        /* restart the last cycle value */
        clock->cycle_last = clocksource_read(clock);
+       clock->error = 0;
+       timekeeping_suspended = 0;
+       write_sequnlock_irqrestore(&xtime_lock, flags);
+       return 0;
+}
+
+static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
+{
+       unsigned long flags;
+
+       write_seqlock_irqsave(&xtime_lock, flags);
+       timekeeping_suspended = 1;
        write_sequnlock_irqrestore(&xtime_lock, flags);
        return 0;
 }
@@ -991,6 +1004,7 @@ static int timekeeping_resume(struct sys_device *dev)
 /* sysfs resume/suspend bits for timekeeping */
 static struct sysdev_class timekeeping_sysclass = {
        .resume         = timekeeping_resume,
+       .suspend        = timekeeping_suspend,
        set_kset_name("timekeeping"),
 };
 
@@ -1101,13 +1115,16 @@ static void update_wall_time(void)
 {
        cycle_t offset;
 
-       clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
+       /* Make sure we're fully resumed: */
+       if (unlikely(timekeeping_suspended))
+               return;
 
 #ifdef CONFIG_GENERIC_TIME
        offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
 #else
        offset = clock->cycle_interval;
 #endif
+       clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
 
        /* normally this loop will run just once, however in the
         * case of lost or late ticks, it will accumulate correctly.