On Fri, 24 Apr 2009, Stephen Rothwell wrote:

On Fri, 24 Apr 2009 12:25:41 +0530 Sachin Sant [off-list ref] wrote:

quoted

While booting today's next tree on a powerpc box [ power 6 blade]
observed the following :

khelper used greatest stack depth: 10176 bytes left

=====================================
[ BUG: lock held at task exit time! ]
-------------------------------------
khelper/21 is exiting with locks still held!
2 locks held by khelper/21:
 #0:  (rcu_read_lock){.+.+.+}, at: [<c0000000001382fc>] 
.check_unsafe_exec+0x44/0x148
 #1:  (rcu_read_lock){.+.+.+}, at: [<c000000000138368>] 
.check_unsafe_exec+0xb0/0x148

stack backtrace:
Call Trace:
[c000000044483cf0] [c000000000011a54] .show_stack+0x6c/0x16c (unreliable)
[c000000044483da0] [c00000000009ae14] .debug_check_no_locks_held+0x98/0xb4
[c000000044483e20] [c000000000073b1c] .do_exit+0x758/0x7b0
[c000000044483f00] [c0000000000853d8] .____call_usermodehelper+0x170/0x174
[c000000044483f90] [c00000000002bd8c] .kernel_thread+0x54/0x70
net_namespace: 2000 bytes

Complete dmesg attached. Let me know if you need any other info. I will 
try yesterday's next
tree to check if this problem can be recreated.

Almost certainly commit 874a9e18f25c86dbc199ad32ddd9ca44d25290e8
("check_unsafe_exec: s/lock_task_sighand/rcu_read_lock/") which has a
typo (two locks instead of lock/unlock) as pointed out by Hugh Dickins
([off-list ref] on LKML).

Indeed, thanks for the headsup Stephen.  My own config gives, not
Sachin's message (or not still visibly on screen anyway), but an
outright panic.  Shame that leaked out into the big world, we'd
all have preferred a quiet fixup!  Here's a patch, which I'll
also send as reply to the relevant thread.

[PATCH] check_unsafe_exec: rcu_read_unlock

Fix typo in previous commit: second rcu_read_lock should be rcu_read_unlock.

Signed-off-by: Hugh Dickins <redacted>
---

 fs/exec.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- 2.6.30-rc3-next-20090424/fs/exec.c	2009-04-24 12:23:43.000000000 +0100
+++ linux/fs/exec.c	2009-04-24 12:26:10.000000000 +0100

@@ -1043,7 +1043,7 @@ int check_unsafe_exec(struct linux_binpr
 		if (t->fs == p->fs)
 			n_fs++;
 	}
-	rcu_read_lock();
+	rcu_read_unlock();
 
 	if (p->fs->users > n_fs) {
 		bprm->unsafe |= LSM_UNSAFE_SHARE;