Re: Can't pinctrl-baytrail be a module?

[Mathias Nyman <mathias.nyman@linux.intel.com>]

On 11.02.2016 15:08, Takashi Iwai wrote:
> On Thu, 11 Feb 2016 13:56:45 +0100,
> Mika Westerberg wrote:
>>
>> +Jean and Mathias
>>
>> On Thu, Feb 11, 2016 at 01:45:16PM +0100, Takashi Iwai wrote:
>>> Hi,
>>>
>>> while discussing whether to enable pinctrl drivers on openSUSE distro
>>> kernels, I wonder why pinctrl-baytrail is built-in only.  Basically
>>> the forced built-in is the only reason against the enablement in
>>> distro kernels.
>>>
>>> The other Intel pinctrl drivers seem to be tristate.
>>> Any missing piece there?
>>
>> Jean sent a patch against this couple of hours ago.
>
> I noticed it right after I send my mail, too, sorry :)
>
>> I explained to him
>> that there is some ACPI GPIO magic happening on Baytrail-T based
>> machines such as Asus T100 where the GPIO driver is needed early at
>> boot, or something along those lines. I've CC'd Mathias Nyman the
>> original author who hopefully remembers this better.
>
> I know of pinctrl-cherryview is needed before others like shdci, but
> for Baytrail-T, it's more than that?
>

If I remember correctly the reason was that the gpio driver was supposed
to work with Hardware reduced ACPI events (in ACPI 5).
Basically ACPI telling operating system it wants an ACPI event handler run
when a certain gpio interrupt is triggered, and the HW reduced ACPI expects the OS gpio driver to do it.

I haven't followed up on this at all. Can't say if we have any HW reduced baytrail variants out there.

For Lynxpoint there was also some issue that pnpacpi reserved the IO resources before a modular gpio driver could get it,
I think it was fixed, might be that baytrail would face same issue.

-Mathias