Re: [PATCH 7/8] Clean up workqueue.c with respect to the freezer based cpu-hotplug

[Oleg Nesterov <oleg@tv-sign.ru>]

On 04/12, Srivatsa Vaddagiri wrote:
>
> On Tue, Apr 03, 2007 at 10:48:20PM +0530, Srivatsa Vaddagiri wrote:
> > > Actually, we should do this before destroy_workqueue() calls flush_workqueue().
> > > Otherwise flush_cpu_workqueue() can hang forever in a similar manner.
> > 
> > Yep. I guess these are a class of freezer deadlocks very similar to vfork
> > parent waiting on child case. I get a feeling these should become common
> > outside of kthread too (A waits on B for something, B gets frozen, which
> > means A won't freeze causing freezer to fail). Can freezer detect this
> > dependency somehow and thaw B automatically? Probably not that easy ..
> 
> I wonder if there is some value in "enforcing" an order in which
> processes get frozen i.e freeze A first before B. That may solve the
> deadlocks we have been discussing wrt kthread_stop and flush_workqueue
> as well.

Perhaps we can add "atomic_t xxx" to task_struct.

	int freezing(struct task_struct *p)
	{
		return test_tsk_thread_flag(p, TIF_FREEZE)
			&& atomic_read(&p->xxx) == 0;
	}

	void xxx_start(struct task_struct *p)
	{
		atomic_inc(p->xxx);
		thaw_process(p);
	}

	xxx_end(struct task_struct *p)
	{
		atomic_dec(p->xxx);
	}

Now,

	xxx_start(p);
	... wait for something which depends on p...
	xxx_end(p);

Of course we need other changes, freeze_process() should check ->xxx, etc.
I am not sure this makes sense.

Oleg.