[U-Boot] [PATCH 2/3] vexpress64: remove #error

[Ryan Harkin]

On 27 October 2015 at 11:36, Linus Walleij <linus.walleij at linaro.org> wrote:
> On Wed, Oct 21, 2015 at 11:54 AM, Ryan Harkin <ryan.harkin at linaro.org> wrote:
>
>> This patch allows vexpress64 targets to be compiled when
>> CONFIG_SYS_FLASH_CFI is enabled.
>
> I can't see what this has to do with enabling CFI?

The errors happen when I enable CFI on Juno or FVP.  Enabling CFI
support includes envcrc.c into the build, which then includes
include/config.h, which in turn includes
include/configs/vexpress_aemv8a.h, but without a board defined,
despite building a Juno or FVP configuration.  I don't really know why
the board isn't defined at that point.

Looking deeper into why envcrc is included into the build, it was
another #define that included it.

I can see that tools/Makefile adds envcrc if CONFIG_BUILD_ENVCRC is
defined, which in turn is enabled if ENVCRC-y is defined, which seems
to happen if CONFIG_ENV_IS_IN_FLASH is defined.

So my comment is wrong.

>
>> I considered using #warning instead of #error, but this just clutters up
>> the build output and hides real warnings.
>>
>> Without this patch, you see errors during compilation like this:
>>
>> include/configs/vexpress_aemv8a.h:42:2: error: #error "Unknown board
>> variant"
>>  #error "Unknown board variant"
>
> The #error is there because noone could answer the question I
> posed: what AEMv8 boards are there that are neither FVP nor
> Juno?

AEMv8 may be the wrong term anyway.  I *think* AEMv8 really only
refers to one flavour of the FVP:  AEMv8 is an representation of the
complete ARMv8 architecture, not of a specific CPU variant such as
Cortex A57, which omits some of the arch.

Everything we're including in this vexpress_aemv8a file is really more
like just vexpress64, with variants inside it for Juno and FVP.

And with a bit of code to detect and handle the platform differences,
we could probably create a single binary that works on everything
covered by this config.  Although I think the #define for the CFI base
address could be a sticking point there.

Currently, ARM have:

Models:
- FVP Foundation models
- FVP AEMv8 models
- FVP Cortex A57/A53 models

^ the same binary runs on all three, although I never test on Cortex
models.  There is a semihosting variant and I've created a DRAM
variant.

Real boards:
- Juno R0
- Juno R1

^ the same binary runs on both

There are no other public platforms, to my knowledge.  ARM have some
internal models that have different peripheral sets, CPU
configurations, etc... but these aren't covered by the public code at
all.

>
> So what board variant are you compiling for here? I suspect a
> non-upstream thingofabob? Maybe there is a better way to cater
> for that than this magic catch-all?

I was building for Juno.


>
> Yours,
> Linus Walleij