From: Steven Rostedt <rostedt@xxxxxxxxxxx>
The raw_spin_unlock() is not a full memory barrier. It only keeps
things from leaking past it, but does not prevent leaks from entering
the critical section. That is:
p = 1;
raw_spin_lock();
[...]
raw_spin_unlock();
y = x
Can turn into:
p = 1;
raw_spin_lock();
load x
store p = 1
raw_spin_unlock();
y = x
This means that the condition check in __swait_event() (and friends)
can be seen before the h->list is set.
raw_spin_lock();
load condition;
store h->list;
raw_spin_unlock();
And the other CPU can see h->list as empty, and this CPU see condition
as not set, and possibly miss the wake up.
To prevent this from happening, add an mb() after setting the h->list.
Reviewed-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Steven Rostedt <rostedt@xxxxxxxxxxx>
---
kernel/wait-simple.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/kernel/wait-simple.c b/kernel/wait-simple.c
index 9725a11..2c85626 100644
--- a/kernel/wait-simple.c
+++ b/kernel/wait-simple.c
@@ -16,6 +16,8 @@
static inline void __swait_enqueue(struct swait_head *head, struct swaiter *w)
{
list_add(&w->node, &head->list);
+ /* We can't let the condition leak before the setting of head */
+ smp_mb();
}
/* Removes w from head->list. Must be called with head->lock locked. */
--
1.7.10.4
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
