Re: [PATCH RFC] drivers/core: Replace lockdep_set_novalidate_class() with unique class keys

[Boqun Feng]

On Mon, Feb 13, 2023 at 09:03:14PM -0500, Alan Stern wrote:
> On Mon, Feb 13, 2023 at 05:51:11PM -0800, Boqun Feng wrote:
> > Basically if you have two lock instances A and B with the same class,
> > and you know that locking ordering is always A -> B, then you can do
> > 
> > 	mutex_lock(A);
> > 	mutex_lock_nest_lock(B, A); // lock B.
> > 
> > to tell the lockdep this is not deadlock, plus lockdep will treat the
> > acquisition of A and the precondition of acquisition B, so the following
> > is not a deadlock as well:
> > 
> > T1:
> > 	mutex_lock(A);
> > 	mutex_lock(C);
> > 	mutex_lock_nest_lock(B, A);
> > 
> > T2:
> > 	mutex_lock(A);
> > 	mutex_lock_nest_lock(B, A);
> > 	mutex_lock(C);
> 
> Why isn't this treated as a deadlock?  It sure looks like a deadlock to 
> me.  Is this an example where lockdep just doesn't get the right answer?
> 

Because A serializes B and C, so that particular piece of code doesn't
cause deadlock. Note that you can still use you normal mutex_lock() for
B, so if there is more code:

T3:
	mutex_lock(C);
	mutex_lock(B);

lockdep will report deadlock.

Regards,
Boqun

> Alan Stern