[Ksummit-2013-discuss] [ATTEND] ACPI vs DT

[Guenter Roeck]

On Mon, Jul 22, 2013 at 03:42:34PM +0200, Linus Walleij wrote:
> On Mon, Jul 22, 2013 at 11:32 AM, Guenter Roeck <linux at roeck-us.net> wrote:
> 
> > Another interesting one: Intel CPUs in embedded systems. Should those use DT or
> > ACPI ? Any interest or plan to make DT fully available for Intel CPUs ?
> 
> Judging from the patches I've receieved for different x86 embedded GPIO
> controllers recently, there seems to be an ambition to use ACPI (for
> example BayTrail is doing an honest attempt).
> 
> For embedded systems not coming from Intel the ambition seems to
> be lower, and basically revert to do what the probe() function was named
> after: probe around in the x86 IOport range, brining us back to the
> ISA times.
> 
> Example:
> http://marc.info/?l=linux-kernel&m=137269614004968&w=2
> 
That is how Super-IO chips work, unfortunately. After all, they implement
functionality which is typically found on the ISA bus. Problem with the above
patches is that there should really be a mfd driver to act as parent for its
various functions. But that is not different to the implementation for other
Super-IO chips, who don't do it either. Only reason you have not seen this
much in the gpio subsystem is that most of the time there is no gpio driver
for the gpio function of those chips.

Using ACPI does not work well for Super-IO chips - the chips are way too complex
to handle with ACPI (or at least that is how it seems), and the available ACPI
bindings are too limited. For example, a Super-IO chip may have dozens of
registers to program sophistitaced fan control. If accessed through ACPI,
all one can typically do is to turn a fan on or off.

Anyway, I don't count that as embedded use. Embedded systems with Intel CPUs
rarely use Super-IO chips. You'll find one on (almost) every PC board, though.

Guenter