[PATCH 3/5] driver core: make struct device_type.uevent() take a const *

[Greg Kroah-Hartman]

On Wed, Nov 23, 2022 at 02:52:59PM +0000, Matthew Wilcox wrote:
> On Wed, Nov 23, 2022 at 02:59:00PM +0100, Maximilian Luz wrote:
> > On 11/23/22 14:34, Andy Shevchenko wrote:
> > > On Wed, Nov 23, 2022 at 02:14:31PM +0100, Maximilian Luz wrote:
> > > > On 11/23/22 13:25, Greg Kroah-Hartman wrote:
> > > > > The uevent() callback in struct device_type should not be modifying the
> > > > > device that is passed into it, so mark it as a const * and propagate the
> > > > > function signature changes out into all relevant subsystems that use
> > > > > this callback.
> > > 
> > > [...]
> > > 
> > > > > -static inline struct ssam_device *to_ssam_device(struct device *d)
> > > > > +static inline struct ssam_device *to_ssam_device(const struct device *d)
> > > > >    {
> > > > >    	return container_of(d, struct ssam_device, dev);
> > > > >    }
> > > > 
> > > > I am slightly conflicted about this change as that now more or less
> > > > implicitly drops the const. So I'm wondering if it wouldn't be better to
> > > > either create a function specifically for const pointers or to just
> > > > open-code it in the instance above.
> > > > 
> > > > I guess we could also convert this to a macro. Then at least there
> > > > wouldn't be an explicit and potentially misleading const-conversion
> > > > indicated in the function signature.
> > > 
> > > This is an intermediate step as far as I know since moving container_of to
> > > recognize const is a bit noisy right now. I guess you can find a discussion
> > > on the topic between Greg and Sakari.
> > 
> > Thanks! I assume you are referring to the following?
> > 
> > 	https://lore.kernel.org/lkml/4218173bd72b4f1899d4c41a8e251f0d@AcuMS.aculab.com/T/
> > 
> > As far as I can tell this is only a warning in documentation, not
> > compile time (which would probably be impossible?).
> > 
> > As I've said I'd be fine with converting the function to a macro (and
> > preferably adding a similar warning like the one proposed in that
> > thread). The point that irks me up is just that, as proposed, the
> > function signature would now advertise a conversion that should never be
> > happening.
> > 
> > Having two separate functions would create a compile-time guarantee, so
> > I'd prefer that, but I can understand if that might be considered too
> > noisy in code. Or if there is a push to make container_of() emit a
> > compile-time warning I'd also be perfectly happy with converting it to a
> > macro now as that'd alleviate the need for functions in the future.
> 
> Can't we do:
> 
> static inline const struct ssam_device *to_ssam_device(const struct device *d)
> {
> 	return container_of(d, const struct ssam_device, dev);
> }
> 

You could, if you can always handle a const pointer coming out of this
function, but I don't think you can.

What you might want to do instead, and I'll be glad to do it for all of
the functions like this I change, is to do what we have for struct
device now:

static inline struct device *__kobj_to_dev(struct kobject *kobj)
{
        return container_of(kobj, struct device, kobj);
}

static inline const struct device *__kobj_to_dev_const(const struct kobject *kobj)
{
        return container_of(kobj, const struct device, kobj);
}

/*
 * container_of() will happily take a const * and spit back a non-const * as it
 * is just doing pointer math.  But we want to be a bit more careful in the
 * driver code, so manually force any const * of a kobject to also be a const *
 * to a device.
 */
#define kobj_to_dev(kobj)                                       \
        _Generic((kobj),                                        \
                 const struct kobject *: __kobj_to_dev_const,   \
                 struct kobject *: __kobj_to_dev)(kobj)


Want me to do the same thing here as well?

thanks,

greg k-h