Re: [PATCH v2 4/9] arm64: Add platform selection for BCM2835.

From: Eric Anholt
Date: Wed Jun 15 2016 - 02:49:08 EST


Catalin Marinas <catalin.marinas@xxxxxxx> writes:

> On Thu, Jun 09, 2016 at 05:21:35PM -0700, Eric Anholt wrote:
>> Catalin Marinas <catalin.marinas@xxxxxxx> writes:
>> > On Sat, Jun 04, 2016 at 12:55:15PM -0700, Eric Anholt wrote:
>> >> Catalin Marinas <catalin.marinas@xxxxxxx> writes:
>> >> > On Fri, Jun 03, 2016 at 08:18:23AM +0200, Gerd Hoffmann wrote:
>> >> >> + This SoC is used in the Raspberry Pi 3 device.
>> >> >
>> >> > I thought we would just use ARCH_BCM, or is it too generic?
>> >>
>> >> Consensus last time around seemed to be to drop adding ARCH_BCM, in
>> >> favor of patch 1 of the series.
>> >
>> > I may have missed that discussion. My point was about consistency with
>> > existing ARCH_* definitions in the arm64 Kconfig.platforms. I can see
>> > why it's easier for you since some drivers are built based on
>> > ARCH_BCM2835. Looking at drivers/clk/bcm/Makefile, there is an
>> > inconsistent mix of CLK_BCM_* and ARCH_BCM_*. I would rather have a new
>> > CLK_BCM2835 that's selected/enabled accordingly (maybe simply depending
>> > on ARCH_BCM).
>>
>> So I introduce a new ARCH_BCM here, that selects the just the 283x
>> family's core drivers? That seems strange, but I'm willing if that's
>> what you want.
>
> I'll leave this decision to the arm-soc guys. What I want to avoid is
> another ARCH_BCM283[89] when some clock or other device changes in a
> future revision of this board (RPi4?). I also don't want fine-grained
> SoC configuration *within* the arch/arm64 Kconfigs but rather just a
> family ARCH_* entry with selectable individual drivers based on the SoC
> revision you target (in case you want to avoid single Image).
>
> We should in general try to give drivers their own Kconfig entries
> separate from ARCH_* ones (with a "depend on ARCH_*" and default y if
> you want it enabled).

OK, we haven't added separate ARCH_BCM283* for the 3 chip revs so far,
so I think what you want is actually the status quo, and we're in
serious agreement. The name for the family just happens to be
ARCH_BCM2835.

Any chance we could get an ack on this?

Attachment: signature.asc
Description: PGP signature