call_function_many: fix list delete vs add race
authorMilton Miller <miltonm@bga.com>
Tue, 15 Mar 2011 19:27:16 +0000 (13:27 -0600)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 17 Mar 2011 23:58:10 +0000 (16:58 -0700)
commite6cd1e07a185d5f9b0aa75e020df02d3c1c44940
tree74f58672a37de933f8fa3456c0ec19a73b483b65
parentef2b4b95a63a1d23958dcb99eb2c6898eddc87d0
call_function_many: fix list delete vs add race

Peter pointed out there was nothing preventing the list_del_rcu in
smp_call_function_interrupt from running before the list_add_rcu in
smp_call_function_many.

Fix this by not setting refs until we have gotten the lock for the list.
Take advantage of the wmb in list_add_rcu to save an explicit additional
one.

I tried to force this race with a udelay before the lock & list_add and
by mixing all 64 online cpus with just 3 random cpus in the mask, but
was unsuccessful.  Still, inspection shows a valid race, and the fix is
a extension of the existing protection window in the current code.

Cc: stable@kernel.org (v2.6.32 and later)
Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/smp.c