[PATCH] for account_group_exec_runtime(), make sure ->signal can't be freed under rq->lock

[Oleg Nesterov <oleg@redhat.com>]

The patch is ugly, but I don't see the better fix for now. Needs
the review from Peter/Ingo.

Unlike other similar routines, account_group_exec_runtime() could be
called "implicitly" from within scheduler after exit_notify(). This
means we can race with the parent doing release_task(), we can't just
check ->signal != NULL.

Change __exit_signal() to do spin_unlock_wait(&task_rq(tsk)->lock)
before __cleanup_signal() to make sure ->signal can't be freed under
task_rq(tsk)->lock. Note that task_rq_unlock_wait() doesn't care
about the case when tsk changes cpu/rq under us, this should be OK.

Thanks to Ingo who nacked my previous buggy patch.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reported-by: Doug Chapman <doug.chapman@hp.com>

--- K-28/include/linux/sched.h~SIG_RQ_LOCK	2008-11-06 19:12:44.000000000 +0100
+++ K-28/include/linux/sched.h	2008-11-10 13:13:07.000000000 +0100
@@ -247,6 +247,7 @@ extern void init_idle(struct task_struct
 extern void init_idle_bootup_task(struct task_struct *idle);
 
 extern int runqueue_is_locked(void);
+extern void task_rq_unlock_wait(struct task_struct *p);
 
 extern cpumask_t nohz_cpu_mask;
 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ)
--- K-28/kernel/sched.c~SIG_RQ_LOCK	2008-11-06 19:12:44.000000000 +0100
+++ K-28/kernel/sched.c	2008-11-10 13:05:09.000000000 +0100
@@ -969,6 +969,14 @@ static struct rq *task_rq_lock(struct ta
 	}
 }
 
+void task_rq_unlock_wait(struct task_struct *p)
+{
+	struct rq *rq = task_rq(p);
+
+	smp_mb();
+	spin_unlock_wait(&rq->lock);
+}
+
 static void __task_rq_unlock(struct rq *rq)
 	__releases(rq->lock)
 {
--- K-28/kernel/exit.c~SIG_RQ_LOCK	2008-11-06 19:11:02.000000000 +0100
+++ K-28/kernel/exit.c	2008-11-10 15:07:22.000000000 +0100
@@ -141,6 +141,11 @@ static void __exit_signal(struct task_st
 	if (sig) {
 		flush_sigqueue(&sig->shared_pending);
 		taskstats_tgid_free(sig);
+		/*
+		 * Make sure ->signal can't go away under rq->lock,
+		 * see account_group_exec_runtime().
+		 */
+		task_rq_unlock_wait(tsk);
 		__cleanup_signal(sig);
 	}
 }