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

[Kent Overstreet <kent.overstreet@linux.dev>]

On Sat, Feb 11, 2023 at 09:40:58PM -0500, Alan Stern wrote:
> On Sat, Feb 11, 2023 at 06:24:42PM -0500, Kent Overstreet wrote:
> > After scanning the rest of the thread: I don't think you want to create
> > separate lockdep classes for each bus and device type, that's defeating
> > how lockdep works.
> 
> Not at all.  In fact, exactly the opposite: lockdep works by creating a 
> class for each lock-inside-a-data-structure-type combination.  A struct 
> device-bus_type/device_type combination is pretty much the same kind of 
> thing.
> 
> >  Maybe if it was only a small, _static_ number of new
> > classes,
> 
> The collection of bus_types and device_types _is_ static, in the sense 
> that each one is a structure defined in a driver source file.  Whether 
> the number is "small" depends on your tolerance for large numbers; the 
> kernel has a lot of source files.  :-)
> 
> Mind you, I'm not saying that having lockdep classes for each bus_type 
> or device_type is always the right thing to do.  There definitely are 
> cases where it wouldn't do what we want.  But perhaps in some cases it 
> would work.
> 
> > but the basic premesis of lockdep is that there are static
> > human understandable lock ordering rules, so lockdep figures out what
> > they are and checks them: if you create a bunch of dynamic classes, the
> > classes are going to be different for everyone in practice and won't
> > have any real bearing on the structure of the code
> 
> As a rule, bus_type's and device_type's aren't dynamic.  Maybe Greg KH 
> once published an example of such a thing; IIRC it was more like a 
> proof-of-principle rather than a serious recommendation on how to write 
> drivers.  (Or else I'm misremembering and it was actually an example of 
> creating dynamic sysfs attributes.)
> 
> Or maybe you're referring to what this patch does?  It does indeed 
> create a bunch of dynamic classes -- one for each struct device.  The 
> ordering rules derived by lockdep will be somewhat arbitrary, as you 
> say.  But some of them certainly will be related to the structure of the 
> source code.

I could be :) I haven't been able to find the patch in question - have a
link?

If you're talking about making lock_class_key dynamic, I think I stand
by what I said though - OTOH, if all you're doing is lifting that to the
caller of the device object init function, so it'll still be a static
object in the driver, that would be totally fine.

I probably should've found the patch before commenting :)