Re: [PATCH] MFD: ab8500: New ab8500_gpadc APIs and reentrance

[Samuel Ortiz <sameo@linux.intel.com>]

Hi Linus,

On Wed, Mar 02, 2011 at 03:12:31PM +0100, Linus Walleij wrote:
> 2011/3/2 Samuel Ortiz <sameo@linux.intel.com>:
> 
> >> +/**
> >> + * ab8500_gpadc_get() - returns a reference to the primary AB8500 GPADC
> >> + * (i.e. the first GPADC in the instance list)
> >> + */
> >> +struct ab8500_gpadc *ab8500_gpadc_get(void)
> >> +{
> >> +     struct ab8500_gpadc *gpadc;
> >> +     gpadc = list_first_entry(&ab8500_gpadc_list, struct ab8500_gpadc, node);
> >> +
> >> +     return gpadc;
> >> +}
> 
> > This seems really arbitrary. We argued with Mattias about it, and giving
> > drivers access to your ADCs means they should somehow have a pointer back to
> > the right ADC. That's not the case here, and while it will just work fine
> > whenever you have one ADC on your board, you'll probably be relying on some
> > sort of device probe order otherwise.
> 
> I guess the solution is to rewrite that function to take a parameter
> then, such as:
> 
> struct ab8500_gpadc *ab8500_gpadc_get(char *name)
> 
> If name is then just a strcmpm(dev_name(gpadc->dev), name)
> the client use will be something like:
> 
> struct ab8500_gpadc *gpadc = ab8500_gpadc_get("ab8500-gpadc.0");
> 
> For the first GPDC in the system (unless .init_name is specified
> by the ab8500-core when creating the device).
> 
> Fair enough?
That's an original solution. It really is hacking around the device model, but
at least it's nicer than the original proposal. So, fair enough, yes.

Cheers,
Samuel.

-- 
Intel Open Source Technology Centre
http://oss.intel.com/