[RFC PATCH v3 0/2] drivers: mfd: Versatile Express SPC support

Pawel Moll pawel.moll at arm.com
Thu Jun 13 19:45:57 EST 2013


On Thu, 2013-06-13 at 01:13 +0100, Samuel Ortiz wrote:
> Now, about the driver itself, besides the really odd code design, the
> static variables all over the place, the nasty init hacks and the
> unneeded long function names, someone should refresh my memory and explain
> to me why is this guy under mfd. I can see it somehow supports IP blocks
> providing different functions, but the design is not sharing anything with
> most of the rest of the mfd drivers.

I belive the vexpress-sysreg.c is a Multi Function Device by all means.
It does so many things that only a water fountain is missing ;-)

If you feel strongly about it, I'm ready to split it into mfd_cells and
move the gpio and leds code into separate drivers, however I'm not
convinced that it's worth the effort.

Now, as to the vexpress-config.c... The first time I've posted the
series, all parts lived in "driver/misc(/vexpress)", but (if I remember
correctly) Arnd had some feelings about "misc" existence at all... I was
thinking about a separate directory for random "system/platform/machine
configuration" drivers, but the idea didn't get any traction.

> Not only that, but the whole vexpress-config code design is not the
> nicest piece of code I've ever seen. And I'm usually not picky. e.g. the
> whole vexpress-config ad-hoc API is awkward and I wonder if it could be
> implemented as a bus instead.

Funny you mention this :-) Again, the first version actually was a
vexpress-config bus. The feedback was - a whole bus_type is over the top
(I'm simplifying the letter slightly but this was the spirit).

> FWIW I take the blame here for not reviewing the initial driver
> submission that Arnd kindly sent to me...But now that I'm looking at it,
> I think it really is on the edge of being staging material. Any thought
> on that ?

I'm more than happy to improve it. The infrastructure (as in: the
hardware) itself is slightly strange and the code pretty much reflects
the situation. There is also a very good reason for some of the oddities
like static bridges array etc - the infrastructure must be functional
very early, long before slab is available (this also caused a lot of
issues with the bus-based implementation, as the device model does
kmalloc all over the place).

So to summarize - I'm open to any suggestions and ready to spend time on
this stuff.

Regards and thanks for your time!

Pawel




More information about the devicetree-discuss mailing list