Prev: [PATCH 03/10] Use percpu stats
Next: [PATCH 05/10] Reduce per table entry overhead by 4 bytes
From: Paul E. McKenney on 9 Aug 2010 13:30
On Tue, Aug 03, 2010 at 11:20:58PM -0400, Miles Lane wrote:
> [ INFO: suspicious rcu_dereference_check() usage. ]
> ---------------------------------------------------
> kernel/exit.c:1387 invoked rcu_dereference_check() without protection!
>
> other info that might help us debug this:
>
> rcu_scheduler_active = 1, debug_locks = 1
> 2 locks held by init/1:
> #0: (tasklist_lock){.+.+..}, at: [<ffffffff81045ca8>] do_wait+0xa9/0x1fa
> #1: (&(&sighand->siglock)->rlock){......}, at: [<ffffffff810457e8>]
> wait_consider_task+0x5e1/0x9f8
>
> stack backtrace:
> Pid: 1, comm: init Not tainted 2.6.35 #15
> Call Trace:
> [<ffffffff8106759c>] lockdep_rcu_dereference+0x9d/0xa6
> [<ffffffff81045877>] wait_consider_task+0x670/0x9f8
> [<ffffffff81045d14>] do_wait+0x115/0x1fa
> [<ffffffff81045f41>] sys_waitid+0x7f/0x178
> [<ffffffff81009cba>] ? sysret_check+0x2e/0x69
> [<ffffffff8104454e>] ? child_wait_callback+0x0/0x53
> [<ffffffff81009c82>] system_call_fastpath+0x16/0x1b

This one is interesting. The ->sighand->siglock is held, but the
rcu_dereference_check() check condition requires that either the
task is dead or that we are in an RCU read-side critical section.
The comment preceding the call to __task_cred() claims that we
"don't need the RCU readlock here as we're holding a spinlock."
This comment dates back to 2008, so might be obsolete.

David, should we enclose the __task_cred() in wait_task_stopped()
with rcu_read_lock()? Or would it be better to add a check to
__task_cred() checking for ->sighand->siglock? Or do we need to
do something else entirely? ;-)

Thanx, Paul
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: [PATCH 03/10] Use percpu stats
Next: [PATCH 05/10] Reduce per table entry overhead by 4 bytes