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

[Kent Overstreet]

On Tue, Feb 14, 2023 at 12:05:27PM +0100, Peter Zijlstra wrote:
> This is lock order per decree, if you get the order function wrong
> lockdep will not see the inversion but you *will* deadlock.

Yeah, that's what I mean. Given that a subclass isn't a fixed thing you
assign to a lock, just something you magic up as needed - I just don't
see what this gets us?

Why not just tell lockdep what the order function is directly?

(I know, I've been saying I'd write a patch for this, I'll get around to
it, I swear :)