Re: [PATCH 1/3] driver core: add a faux bus for use when a simple device/bus is needed

[Andy Shevchenko]

On Mon, Feb 03, 2025 at 04:35:45PM +0100, Greg Kroah-Hartman wrote:
> On Mon, Feb 03, 2025 at 05:11:03PM +0200, Andy Shevchenko wrote:
> > On Mon, Feb 03, 2025 at 03:25:17PM +0100, Greg Kroah-Hartman wrote:

...

> > > +#include <linux/device/faux.h>
> > 
> > I would rather think that this goes after generic inclusions...
> > 
> > > +#include <linux/err.h>
> > > +#include <linux/init.h>
> > > +#include <linux/slab.h>
> > > +#include <linux/string.h>
> > 
> > ...somewhere here.
> > 
> > But looking into organisation of device.h and device/*.h,
> > I would rather think of the linux/faux_device.h.
> 
> It can go anywhere, there is no need to sort things :)

It's not about sorting, it's about grouping from more generic to less generic.

> > > +#include "base.h"
> > 
> > I don't remember by heart what it does include, I would go with IWYU principle
> > and list above all what we use.
> > 
> > container_of.h
> > device.h
> > export.h
> > printk.h
> > types.h
> 
> That's not what the driver core ever did, so no need to worry about it,
> thanks.

But it doesn't mean that driver code does its best. No big worries, of course.

...

> > > +	return (strcmp(faux_obj->name, drv->name) == 0);
> > 
> > Outer parentheses are not needed.
> 
> Makes me feel good as it is an assignment test, and that's what
> platform.c has for the past few decades.

Sure, it also can be written as

	return !strcmp(faux_obj->name, drv->name);

that makes the same without explicit comparing to 0. But it's matter of taste.

...

> > > +/**
> > > + * __faux_device_create - create and register a faux device and driver
> > > + * @name: name of the device and driver we are adding
> > > + * @faux_ops: struct faux_driver_ops that the new device will call back into, can be NULL
> > > + * @owner: module owner of the device/driver
> > > + *
> > > + * Create a new faux device and driver, both with the same name, and register
> > > + * them in the driver core properly.  The probe() callback of @faux_ops will be
> > > + * called with the new device that is created for the caller to do something
> > > + * with.
> > 
> > The kernel-doc will complain on missing Return: section.
> 
> Is that new?  Does that mean platform.c has lots of complaints in it as
> well?  What does platform_find_device_by_driver() give you for a
> documentation issue?
> 
> And as I didn't hook this up to the kernel documentation build yet, it
> shouldn't produce any warnings anywhere :)

I would rather say it's old.

Run

	kernel-doc -Wall -none -v ...your file...

and find the warning. During the kernel builds this is moved to extra warnings.

> > > + */

...

> > > +	faux_obj = kzalloc(sizeof(*faux_obj) + strlen(name) + 1, GFP_KERNEL);
> > 
> > Potential overflow. To avoid one may use struct_size() from overflow.h.
> 
> Users should not be providing the string here.  Again, this comes from
> platform.c.

I'm not sure I follow. The name parameter is not limited anyhow, so one may
provide non-terminated string and strlen() will return an arbitrary number.
Potentially this can lead to big numbers and become an overflow when gets
to a parameter for kmalloc(). This most likely never happen in real life,
but still the overflow is possible.

-- 
With Best Regards,
Andy Shevchenko