clockevent: Don't remove broadcast device when cpu is dead
authorXiaotian Feng <dfeng@redhat.com>
Thu, 7 Jan 2010 03:22:44 +0000 (11:22 +0800)
committerThomas Gleixner <tglx@linutronix.de>
Mon, 18 Jan 2010 13:44:50 +0000 (14:44 +0100)
Marc reported that the BUG_ON in clockevents_notify() triggers on his
system. This happens because the kernel tries to remove an active
clock event device (used for broadcasting) from the device list.

The handling of devices which can be used as per cpu device and as a
global broadcast device is suboptimal.

The simplest solution for now (and for stable) is to check whether the
device is used as global broadcast device, but this needs to be
revisited.

[ tglx: restored the cpuweight check and massaged the changelog ]

Reported-by: Marc Dionne <marc.c.dionne@gmail.com>
Tested-by: Marc Dionne <marc.c.dionne@gmail.com>
Signed-off-by: Xiaotian Feng <dfeng@redhat.com>
LKML-Reference: <1262834564-13033-1-git-send-email-dfeng@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@kernel.org
kernel/time/clockevents.c

index 6f740d9f09485a3843086d20472c169b86c232aa..d7395fdfb9f38f931761e29d14534143dbe3799c 100644 (file)
@@ -259,7 +259,8 @@ void clockevents_notify(unsigned long reason, void *arg)
                cpu = *((int *)arg);
                list_for_each_entry_safe(dev, tmp, &clockevent_devices, list) {
                        if (cpumask_test_cpu(cpu, dev->cpumask) &&
-                           cpumask_weight(dev->cpumask) == 1) {
+                           cpumask_weight(dev->cpumask) == 1 &&
+                           !tick_is_broadcast_device(dev)) {
                                BUG_ON(dev->mode != CLOCK_EVT_MODE_UNUSED);
                                list_del(&dev->list);
                        }