[PATCH] Hotplug CPU vs TASK_ZOMBIEs: The Sequel to Hotplug CPU vs TASK_DEAD

release_task can sleep.  Sleeping allows a CPU to go down underneath you.
release_task removes you from the tasklist, so you don't get migrated off the
CPU: BUG() in sched.c.

In last week's episode, our dashing hero (Ingo Molnar) solved this for
self-reaping tasks by grabbing the hotplug cpu lock to prevent this. 
However, in an unexpected twist, the problem remains for tasks whose
parents call release_task on them: the zombies are off the task list, and
lurk on the dead CPU.

Fortunately, the comedic sidekick (Rusty Russell) has an answer: let's make
the hotplug callback walk the runqueue of the dead CPU as well, taking care
of the zombies.

1) Restore exit.c to its former form.  The comment is incorrect: sched.c
   checks PF_DEAD, not the state, to decide to do the final
   put_task_struct(), and it does it for all tasks, self-reaping or no.

2) Implement migrate_dead_tasks() in the sched.c hotplug CPU callback.

3) Rename migrate_all_tasks() to migrate_live_tasks().

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

2 files changed, 50 insertions, 27 deletions

diff --git a/kernel/exit.c b/kernel/exit.c
index b21436b147f1..f0d1ff3c30f2 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -754,8 +754,8 @@ static void exit_notify(struct task_struct *tsk)
 	state = TASK_ZOMBIE;
 	if (tsk->exit_signal == -1 && tsk->ptrace == 0)
 		state = TASK_DEAD;
-	else
-		tsk->state = state;
+	tsk->state = state;
+
 	/*
 	 * Clear these here so that update_process_times() won't try to deliver
 	 * itimer, profile or rlimit signals to this task while it is in late exit.
@@ -764,14 +764,6 @@ static void exit_notify(struct task_struct *tsk)
 	tsk->it_prof_value = 0;
 	tsk->rlim[RLIMIT_CPU].rlim_cur = RLIM_INFINITY;
 
-	/*
-	 * Get a reference to it so that we can set the state
-	 * as the last step. The state-setting only matters if the
-	 * current task is releasing itself, to trigger the final
-	 * put_task_struct() in finish_task_switch(). (thread self-reap)
-	 */
-	get_task_struct(tsk);
-
 	write_unlock_irq(&tasklist_lock);
 
 	list_for_each_safe(_p, _n, &ptrace_dead) {
@@ -781,23 +773,12 @@ static void exit_notify(struct task_struct *tsk)
 	}
 
 	/* If the process is dead, release it - nobody will wait for it */
-	if (state == TASK_DEAD) {
-		lock_cpu_hotplug();
+	if (state == TASK_DEAD)
 		release_task(tsk);
-		write_lock_irq(&tasklist_lock);
-		/*
-		 * No preemption may happen from this point on,
-		 * or CPU hotplug (and task exit) breaks:
-		 */
-		unlock_cpu_hotplug();
-		tsk->state = state;
-		_raw_write_unlock(&tasklist_lock);
-		local_irq_enable();
-	} else
-		preempt_disable();
 
+	/* PF_DEAD causes final put_task_struct after we schedule. */
+	preempt_disable();
 	tsk->flags |= PF_DEAD;
-	put_task_struct(tsk);
 }
 
 asmlinkage NORET_TYPE void do_exit(long code)
diff --git a/kernel/sched.c b/kernel/sched.c
index 732344a7be1b..4b44d5219c36 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -3990,8 +3990,8 @@ static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *tsk)
 	__migrate_task(tsk, dead_cpu, dest_cpu);
 }
 
-/* migrate_all_tasks - function to migrate all tasks from the dead cpu. */
-static void migrate_all_tasks(int src_cpu)
+/* Run through task list and migrate tasks from the dead cpu. */
+static void migrate_live_tasks(int src_cpu)
 {
 	struct task_struct *tsk, *t;
 
@@ -4033,6 +4033,47 @@ void sched_idle_next(void)
 
 	spin_unlock_irqrestore(&rq->lock, flags);
 }
+
+static void migrate_dead(unsigned int dead_cpu, task_t *tsk)
+{
+	struct runqueue *rq = cpu_rq(dead_cpu);
+
+	/* Must be exiting, otherwise would be on tasklist. */
+	BUG_ON(tsk->state != TASK_ZOMBIE && tsk->state != TASK_DEAD);
+
+	/* Cannot have done final schedule yet: would have vanished. */
+	BUG_ON(tsk->flags & PF_DEAD);
+
+	get_task_struct(tsk);
+
+	/*
+	 * Drop lock around migration; if someone else moves it,
+	 * that's OK.  No task can be added to this CPU, so iteration is
+	 * fine.
+	 */
+	spin_unlock_irq(&rq->lock);
+	move_task_off_dead_cpu(dead_cpu, tsk);
+	spin_lock_irq(&rq->lock);
+
+	put_task_struct(tsk);
+}
+
+/* release_task() removes task from tasklist, so we won't find dead tasks. */
+static void migrate_dead_tasks(unsigned int dead_cpu)
+{
+	unsigned arr, i;
+	struct runqueue *rq = cpu_rq(dead_cpu);
+
+	for (arr = 0; arr < 2; arr++) {
+		for (i = 0; i < MAX_PRIO; i++) {
+			struct list_head *list = &rq->arrays[arr].queue[i];
+			while (!list_empty(list))
+				migrate_dead(dead_cpu,
+					     list_entry(list->next, task_t,
+							run_list));
+		}
+	}
+}
 #endif /* CONFIG_HOTPLUG_CPU */
 
 /*
@@ -4072,7 +4113,7 @@ static int migration_call(struct notifier_block *nfb, unsigned long action,
 		cpu_rq(cpu)->migration_thread = NULL;
 		break;
 	case CPU_DEAD:
-		migrate_all_tasks(cpu);
+		migrate_live_tasks(cpu);
 		rq = cpu_rq(cpu);
 		kthread_stop(rq->migration_thread);
 		rq->migration_thread = NULL;
@@ -4081,6 +4122,7 @@ static int migration_call(struct notifier_block *nfb, unsigned long action,
 		deactivate_task(rq->idle, rq);
 		rq->idle->static_prio = MAX_PRIO;
 		__setscheduler(rq->idle, SCHED_NORMAL, 0);
+		migrate_dead_tasks(cpu);
 		task_rq_unlock(rq, &flags);
 		BUG_ON(rq->nr_running != 0);