Re: [PATCH 3/3] kref: Remove the memory barriers

[Oliver Neukum <oneukum@suse.de>]

Am Montag, 12. Dezember 2011, 11:32:58 schrieb Ming Lei:
> On Mon, Dec 12, 2011 at 6:12 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> 
> > I don't know the driver model, and I don't plan to start learning it
> > now. But if what you said is possible its broken and no memory barriers
> > will fix it.
> 
> IMO, you don't need to learn it, and my example is very simple and common
> kref usage in device drivers, :-)
> 
> Could we only focus on it and see what is problem? and why won't memory
> barrier fix it?

You don't have a CPU ordering problem. If CPU A can do a kfree() you need to
make sure CPU B doesn't get a pointer to that object. Basically your race is:

CPU A							CPU B

p = a;
								p = a;
p->counter--;
if (!p->counter) kfree(p);
a = NULL;
								p->counter++;

This is not an ordering problem. You have a real critical section here.
It doesn't matter when CPU B sees the decrement.
You must make sure there are no pointers to objects you might free
if you intend to use the pointers without locks.

	Regards
		Oliver