[PATCH] do_sched_setscheduler: don't take tasklist_lock

[PATCH] do_sched_setscheduler: don't take tasklist_lock

[Oleg Nesterov]

We don't need to take tasklist_lock or disable irqs for
find_task_by_pid() + get_task_struct(). Use RCU locks
instead.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>

--- 2.6.18-rc3/kernel/sched.c~1_dss	2006-07-16 01:53:08.000000000 +0400
+++ 2.6.18-rc3/kernel/sched.c	2006-08-13 20:19:02.000000000 +0400
@@ -4156,14 +4156,15 @@ do_sched_setscheduler(pid_t pid, int pol
 		return -EINVAL;
 	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
 		return -EFAULT;
-	read_lock_irq(&tasklist_lock);
+
+	rcu_read_lock();
 	p = find_process_by_pid(pid);
-	if (!p) {
-		read_unlock_irq(&tasklist_lock);
+	if (p)
+		get_task_struct(p);
+	rcu_read_unlock();
+	if (!p)
 		return -ESRCH;
-	}
-	get_task_struct(p);
-	read_unlock_irq(&tasklist_lock);
+
 	retval = sched_setscheduler(p, policy, &lparam);
 	put_task_struct(p);
 

Re: [PATCH] do_sched_setscheduler: don't take tasklist_lock

[Oleg Nesterov]

On 08/13, Oleg Nesterov wrote:
>
> We don't need to take tasklist_lock or disable irqs for
> find_task_by_pid() + get_task_struct(). Use RCU locks
> instead.

On the other hand, I think sched_setscheduler() does need tasklist_lock!

It is unsafe do dereference ->signal unless tasklist_lock or ->siglock
is held (or p == current). Yes, we pin the task structure, but this can't
prevent from release_task()->__exit_signal() which sets ->signal = NULL.

So, I think this patch

	[PATCH] Drop tasklist lock in do_sched_setscheduler
	commit e74c69f46d93d29eea0ad8647863d1c6488f0f55

is not correct.

Am I missed something?

Oleg.