From: Oleg Nesterov <oleg@tv-sign.ru>
I think it is supposed that timers are serialized wrt to itself,
but I can't find any documentation about it.
If CPU_0 does mod_timer(jiffies+1) while the timer is currently
running on CPU 1, it is quite possible that local interrupt on
CPU_0 will start that timer before it finished on CPU_1.
With this patch __mod_timer() locks old_base even if timer is not
pending, like del_timer_sync() does. If the timer is still running,
and old_base != new_base, __mod_timer() retries.
Now we have:
1. the timer's handler can't run on different cpus at
the same time.
2. the timer's base can't be changed during execution
of the timer's handler.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
---
25-akpm/kernel/timer.c | 17 ++++++++++-------
1 files changed, 10 insertions(+), 7 deletions(-)
diff -puN kernel/timer.c~timers-serialize-timers kernel/timer.c
--- 25/kernel/timer.c~timers-serialize-timers 2005-03-19 14:19:54.000000000 -0800
+++ 25-akpm/kernel/timer.c 2005-03-19 14:19:54.000000000 -0800
@@ -193,7 +193,7 @@ int __mod_timer(struct timer_list *timer
spin_lock_irqsave(&timer->lock, flags);
new_base = &__get_cpu_var(tvec_bases);
repeat:
- old_base = __get_base(timer);
+ old_base = timer_base(timer);
/*
* Prevent deadlocks via ordering by old_base < new_base.
@@ -208,29 +208,32 @@ repeat:
}
/*
* The timer base might have been cancelled while we were
- * trying to take the lock(s):
+ * trying to take the lock(s), or can still be running on
+ * old_base's CPU.
*/
- if (__get_base(timer) != old_base) {
+ if (timer_base(timer) != old_base
+ || old_base->running_timer == timer) {
spin_unlock(&new_base->lock);
spin_unlock(&old_base->lock);
goto repeat;
}
} else {
spin_lock(&new_base->lock);
- if (__get_base(timer) != old_base) {
+ if (timer_base(timer) != old_base) {
spin_unlock(&new_base->lock);
goto repeat;
}
}
/*
- * Delete the previous timeout (if there was any), and install
- * the new one:
+ * Delete the previous timeout (if there was any).
+ * We hold timer->lock, no need to check old_base != 0.
*/
- if (old_base) {
+ if (timer_pending(timer)) {
list_del(&timer->entry);
ret = 1;
}
+
timer->expires = expires;
internal_add_timer(new_base, timer);
__set_base(timer, new_base, 1);
_