Re: INFO: possible circular locking dependency at cleanup_workqueue_thread

[Oleg Nesterov <oleg@redhat.com>]

On 05/17, Johannes Berg wrote:
>
> I'm not entirely sure yet, but I would think the problem might be a
> false positive in the workqueue code -- remember this report only
> triggers because cleanup_workqueue_thread() acquires the fake lock for
> the workqueue.

I spent a lot of time, but I can't explain this report too :( Even
if it is false positive, I don't understand why lockdep complains.

> Maybe it shouldn't do that from the CPU_POST_DEAD
> notifier?

Well, in any case we should understand why we have the problem, before
changing the code. And CPU_POST_DEAD is not special, why should we treat
it specially and skip lock_map_acquire(wq->lockdep_map) ?


But, I am starting to suspect we have some problems with lockdep too.
OK, I can't explain what I mean... But consider this code:

	DEFINE_SPINLOCK(Z);
	DEFINE_SPINLOCK(L1);
	DEFINE_SPINLOCK(L2);

	#define L(l)	spin_lock(&l)
	#define U(l)	spin_unlock(&l)

	void t1(void)
	{
		L(L1);
		L(L2);

		U(L2);
		U(L1);
	}

	void t2(void)
	{
		L(L2);
			L(Z);
		L(L1);

		U(L1);
			U(Z);
		U(L2);
	}

	void tst(void)
	{
		t1();
		t2();
	}

We have the trivial AB-BA deadlock with L1 and L2, but lockdep says:

	[ INFO: possible circular locking dependency detected ]
	2.6.30-rc6-00043-g22ef37e-dirty #3
	-------------------------------------------------------
	perl/676 is trying to acquire lock:
	 (L1){+.+...}, at: [<ffffffff802522b8>] t2+0x28/0x50

	but task is already holding lock:
	 (Z){+.+...}, at: [<ffffffff802522ac>] t2+0x1c/0x50

	which lock already depends on the new lock.


	the existing dependency chain (in reverse order) is:

	-> #2 (Z){+.+...}:

	-> #1 (L2){+.+...}:

	-> #0 (L1){+.+...}:

	other info that might help us debug this:

	2 locks held by perl/676:
	 #0:  (L2){+.+...}, at: [<ffffffff802522a0>] t2+0x10/0x50
	 #1:  (Z){+.+...}, at: [<ffffffff802522ac>] t2+0x1c/0x50

This output looks obviously wrong, Z does not depend on L1 or any
other lock.

Oleg.