[PATCH 1/2] drivers: PL061: add support for platform driver probing

[Graeme Gregory]

On Thu, 13 Aug 2015, at 01:14 PM, Linus Walleij wrote:
> On Mon, Aug 3, 2015 at 5:13 PM, Graeme Gregory <gg at slimlogic.co.uk>
> wrote:
> > On Mon, 3 Aug 2015, at 01:40 PM, Linus Walleij wrote:
> >> On Mon, Aug 3, 2015 at 8:59 AM, Shannon Zhao <zhaoshenglong at huawei.com>
> >> wrote:
> >>
> >> > From: Shannon Zhao <shannon.zhao at linaro.org>
> >> >
> >> > Since PL061 currently only supports AMBA driver, to support using GPIO
> >> > PL061 by DT or ACPI, it needs to add support for platform driver.
> >> > A DT binding is provided with this patch, ACPI support is added in a
> >> > separate one.
> >> >
> >> > Signed-off-by: Shannon Zhao <zhaoshenglong at huawei.com>
> >> > Signed-off-by: Shannon Zhao <shannon.zhao at linaro.org>
> >>
> >> I already stated in 0/2 what the problem is with this, and it's obviously
> >> adding
> >> a huge codechunk, very similar to the AMBA probe path.
> >>
> >> We need to investigate if ACPI can just spawn AMBA/PrimeCell type devices
> >> instead.
> >
> > AMBA DT support is a hideous hack, we had kind of hoped to avoid doing
> > that in ACPI.
> 
> Define what you mean with "hideous hack".
> 
drivers/of/platform.c search for #ifdef CONFIG_ARM_AMBA

> I think it is elegant. It combines the DT aspect of telling where the
> device is with the Plug-and-play aspect of reading the 0xB105F00D
> magic in the PrimeCells very nicely. With an optional override
> mechanism. It's nice.
> 
> The runtime PM stuff and pclk handling that Russell and also Ulf
> has worked on is very helpful and takes out a lot of codelines
> in a lot of drivers. Also very nice.
> 
I have no objection to AMBA devices that are children of an AMBA bus
object. Its the hack above that is nasty!  It is also un-scalable as it
sets the precedent that the platform.c in both DT and ACPI should learn
about all the different types of device probing.

> > If you notice ARM already updated pl011 driver to support platform
> > device type probing this seems to be the way to go.
> 
> Just because there is a precedent does not mean it is good.
> 
> Continuing with this adds 100+ lines of hairy initialization to
> every AMBA driver.
> 
> What needs to happen is stop this and make ACPI spawn
> AMBA devices for PrimeCells, and move the serial port over
> to that.
> 
> A compromise may be to have something that probes a list
> of these platform devices in drivers/acpi/ and spawn a
> corresponding AMBA device for it. It will still save codelines in
> all other subsystems and lower maintenance for the
> subsystem maintainers.
> 
I am currently working on exactly what you describe above.

Graeme