Re: [PATCH v1 6/8] gpio: acpi: Explain how to get GPIO descriptors in ACPI case

[Dmitry Torokhov]

On Tue, Apr 04, 2017 at 08:59:11PM +0300, Andy Shevchenko wrote:
> On Tue, 2017-04-04 at 10:31 -0700, Dmitry Torokhov wrote:
> > On Tue, Apr 04, 2017 at 07:11:17PM +0300, Andy Shevchenko wrote:
> > > On Wed, 2017-03-29 at 18:04 +0300, Andy Shevchenko wrote:
> > > > On Wed, 2017-03-29 at 00:12 -0700, Dmitry Torokhov wrote:
> > > > > On Tue, Mar 28, 2017 at 07:39:23PM +0300, Andy Shevchenko wrote:
> > > > > > On Thu, 2017-03-23 at 13:28 -0700, Dmitry Torokhov wrote:
> > > > > > > On Thu, Mar 23, 2017 at 09:46:16PM +0200, Andy Shevchenko
> > > > > > > wrote:
> 
> > > > Otherwise I'm reading something like this:
> > > > "If we have platform driverX.c which has DT/platform and ACPI
> > > > enumeration, we must split ACPI part out, duplicate a lot of code
> > > > and
> > > > use platform driver as a library."
> > 
> > No. You need to split the part that augments incomplete ACPI data, and
> > move it somewhere (drivers/platform/x86/<platform>-crap.c; the driver
> > stays the same: a driver that is useful across multiple platforms.
> 
> > > > Is that what you mean?
> 
> So, it means to spread IDs in two places.

For completely different purposes, yes. One takes DMI data to identify
platform, and ACPI _instance_ ID to identify the particular ACPI device;
there theoretically could be several of them.  If you have a better
option to identify the instance, we can switch to them.

The driver uses HID or CID to bind to one or more devices.

> Looking into silead_dmi.c I
> can say it looks as a hack, we aren't supposed to use "ACPIXXXX:YY" in
> the drivers AFAIK. Besides the fact of notifier and arch_initcall().
> 
> It indeed feels like a crap and looks like a crap.

It is supposed to be crap. We have a vendor that neglected to describe
the device in firmware properly and instead expects the driver to be
recompiled for each device shipped. Bad, bad vendor. So we have crap in
platform/x86/... At least we do not smear this shit all over generic
driver.

> 
> Rafael, Mika, what are your opinions about proposed approach?
> 
> > > > P.S. This all _CRS fallback shouldn't be allowed in the first
> > > > place.
> > 
> > It does work in many cases. By disallowing it completely you force
> > much
> > more platform stuff knowledge in the kernel, whereas before you needed
> > to deal with exceptions.
> 
> It works due to luck, not otherwise.

As far as I see it works often enough.

Thanks.

-- 
Dmitry