Hi Dmitry,

> > Isn't it actually the other way around? platform_get_drvdata() is a
> > convenience function to access driver_data which is embedded in struct
> > device?
> 
> I guess it depends on how you read it. I always considered it separate
> because none (?) of the bus implementation assert this in comments to
> XXX_get_drvdata().

Well, even in the case somebody will implement a custom driver_data for
platform_devices, this person will need to convert all current users to
'dev_get_drvdata(&pdev->dev);' first in order to avoid regressions, I'd
think. This is what my patch does right now (but merely for overhead
reasons). Or?

> > > in the future, so I'd prefer keep using the proper accessors for the
> > > objects we are dealing with.
> > 
> > Exactly. I'd just argue, the object we are dealing with, declared in the
> > PM functions, is a struct device.
> 
> No, the driver does not create a generic device, it actually creates a
> platform device, or i2c client, or spi, or something else. The fact that

True.

> suspend and resume routines have generic device as their argument has
> more to do with the language limitation rather than reflection of true
> type of the objects we are dealing with.

Ok, can be argued. I'd personally still go for the gain, but I won't
push harder than this mail.

Regards,

   Wolfram