Re: [RFC v5 net-next 01/13] mfd: ocelot: add support for external mfd control over SPI for the VSC7512

[Lee Jones <lee.jones@linaro.org>]

On Tue, 11 Jan 2022, Colin Foster wrote:

> Hi Mark and Lee,
> 
> On Tue, Jan 11, 2022 at 01:44:28PM +0000, Mark Brown wrote:
> > On Tue, Jan 11, 2022 at 10:13:43AM +0000, Lee Jones wrote:
> > 
> > > Unless something has changed or my understanding is not correct,
> > > regmap does not support over-lapping register ranges.
> > 
> > If there's no caches and we're always going direct to hardware it will
> > work a lot of the time since the buses generally have concurrency
> > protection at the lowest level, though if the drivers ever do any
> > read/modify/write operations the underlying hardware bus isn't going to
> > know about it so you could get data corruption if two drivers decide to
> > try to operate on the same register.  If there's caches things will
> > probably go badly since the cache will tend to amplify the
> > read/modify/write issues.
> 
> Good point about caches. No, nothing in these drivers utilize caches
> currently, but that doesn't mean it shouldn't... or won't. Any
> concurrency in this specific case would be around the SPI bus.
> 
> I think the "overlapping regmaps" already exist in the current drivers... 
> but actually I'm not so sure anymore.
> 
> Either way, this is helping nudge me in the right direction.
> 
> > 
> > > However, even if that is required, I still think we can come up with
> > > something cleaner than creating a whole API based around creating
> > > and fetching different regmap configurations depending on how the
> > > system was initialised.
> > 
> > Yeah, I'd expect the usual pattern is to have wrapper drivers that
> > instantiate a regmap then have the bulk of the driver be a library that
> > they call into should work.
> 
> Understood. And I think this can make sense and clean things up. The
> "ocelot_core" mfd will register every regmap range, regardless of
> whether any child actually uses them. Every child can then get regmaps
> by name, via dev_get_regmap. That'll get rid of the back-and-forth
> regmap hooks.

I was under the impression that MFD would not always be used?

Didn't you have a use-case where the child devices could be used
independently of anything else?

If not, why don't you just register a single Regmap covering the whole
range?  Then let the Regmap API deal with the concurrency handling.

-- 
Lee Jones [李琼斯]
Principal Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog