move_task_off_dead_cpu()->select_fallback_rq() reads/updates ->cpus_allowed
lockless. We can race with set_cpus_allowed() running in parallel.

Change it to take rq->lock around select_fallback_rq(). Note that it is not
trivial to move this spin_lock() into select_fallback_rq(), we must recheck
the task was not migrated after we take the lock and other callers do not
need this lock.

To avoid the races with other callers of select_fallback_rq() which rely on
TASK_WAKING, we also check p->state != TASK_WAKING and do nothing otherwise.
The owner of TASK_WAKING must update ->cpus_allowed and choose the correct
CPU anyway, and the subsequent __migrate_task() is just meaningless because
p->se.on_rq must be false.

Alternatively, we could change select_task_rq() to take rq->lock right
after it calls sched_class->select_task_rq(), but this looks a bit ugly.

Also, change it to not assume irqs are disabled and absorb __migrate_task_irq().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---

 kernel/sched.c |   30 +++++++++++++++---------------
 1 file changed, 15 insertions(+), 15 deletions(-)

--- 34-rc1/kernel/sched.c~2_MTODC_TAKE_RQ_LOCK	2010-03-15 09:40:16.000000000 +0100
+++ 34-rc1/kernel/sched.c	2010-03-15 09:40:31.000000000 +0100
@@ -5509,29 +5509,29 @@ static int migration_thread(void *data)
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
-
-static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu)
-{
-	int ret;
-
-	local_irq_disable();
-	ret = __migrate_task(p, src_cpu, dest_cpu);
-	local_irq_enable();
-	return ret;
-}
-
 /*
  * Figure out where task on dead CPU should go, use force if necessary.
  */
 static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p)
 {
-	int dest_cpu;
-
+	struct rq *rq = cpu_rq(dead_cpu);
+	int needs_cpu, dest_cpu;
+	unsigned long flags;
 again:
-	dest_cpu = select_fallback_rq(dead_cpu, p);
+	local_irq_save(flags);
+
+	raw_spin_lock(&rq->lock);
+	needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != ...
[ message continues ]

^ 
--

[ message continues ]

Hmm. looks like my gcc is more friendly...

OK. certainly I'll send the updated patch, if this series passes
your review otherwise.

Oleg.

--

[ message continues ]

Hrm, that and I'm apparently unable to read, it said dest_cpu, not

Yeah, you made a few good points in 0/6, am now staring at the code on
how to close those holes, hope to post something sensible soon.
--

[ message continues ]

Yes, great.

Speaking of 0/6, I forgot to ask a couple more question...

try_to_wake_up() does task_rq_lock() which checks TASK_WAKING. Perhaps
it shouldn't ? I mean, perhaps try_to_wake_up() can take rq->lock without
checking task->state. It can never race with the owner of TASK_WAKING,
before anything else we check "p->state & state".

And. Without the change above, any owner of TASK_WAKING must disable
preemption and clear irqs.

What do you think?


And a stupid question. While doing these changes I was really, really
puzzled by task_rq_lock() which does

	local_irq_save(*flags);
	rq = task_rq(p);
	raw_spin_lock(&rq->lock);

to the point, I even tried to read the comment which says:

	Note the ordering: we can safely lookup the task_rq without
	explicitly disabling preemption.

Could you please explain what does this mean? IOW, why can't we do

	rq = task_rq(p);
	raw_spin_lock_irqsave(&rq->lock, flags);

instead?

Of course, this doesn't really matter, but I'd like to understand
what I have missed here.

Thanks,

Oleg.

--

[ message continues ]

You're right, but creating a special task_rq_lock() for ttwu() went a
little far, but now that we can remove all that again, this too should

I'm not sure why that is the case, v2.6.14:kernel/sched.c already has
that. Ingo can you remember any reason for this or should we change the
code like Oleg suggests?

--

[ message continues ]

Commit-ID:  1445c08d06c5594895b4fae952ef8a457e89c390
Gitweb:     http://git.kernel.org/tip/1445c08d06c5594895b4fae952ef8a457e89c390
Author:     Oleg Nesterov <oleg@redhat.com>
AuthorDate: Mon, 15 Mar 2010 10:10:10 +0100
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Fri, 2 Apr 2010 20:12:01 +0200

sched: move_task_off_dead_cpu(): Take rq->lock around select_fallback_rq()

move_task_off_dead_cpu()->select_fallback_rq() reads/updates ->cpus_allowed
lockless. We can race with set_cpus_allowed() running in parallel.

Change it to take rq->lock around select_fallback_rq(). Note that it is not
trivial to move this spin_lock() into select_fallback_rq(), we must recheck
the task was not migrated after we take the lock and other callers do not
need this lock.

To avoid the races with other callers of select_fallback_rq() which rely on
TASK_WAKING, we also check p->state != TASK_WAKING and do nothing otherwise.
The owner of TASK_WAKING must update ->cpus_allowed and choose the correct
CPU anyway, and the subsequent __migrate_task() is just meaningless because
p->se.on_rq must be false.

Alternatively, we could change select_task_rq() to take rq->lock right
after it calls sched_class->select_task_rq(), but this looks a bit ugly.

Also, change it to not assume irqs are disabled and absorb __migrate_task_irq().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100315091010.GA9131@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 kernel/sched.c |   30 +++++++++++++++---------------
 1 files changed, 15 insertions(+), 15 deletions(-)

diff --git a/kernel/sched.c b/kernel/sched.c
index c0b3ebc..27774b5 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -5448,29 +5448,29 @@ static int migration_thread(void *data)
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
-
-static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu)
-{
-	int ret;
-
-	local_irq_disable();
-	ret = __migrate_task(p, ...
[ message continues ]