[coreboot] Why is `MMCONF_SUPPORT_DEFAULT` not enabled for more devices?

[Kyösti Mälkki]

On Sat, 2013-06-29 at 13:37 +0200, Paul Menzel wrote:
> Dear coreboot folks,
> 

<clip>

> Searching the code for that Kconfig variable, I get the following
> results
> 
>         $ git grep MMCONF_SUPPORT_DEFAULT
>         src/Kconfig:config MMCONF_SUPPORT_DEFAULT
>         src/arch/x86/include/arch/io.h:#if CONFIG_MMCONF_SUPPORT_DEFAULT
>         src/arch/x86/include/arch/io.h:#if CONFIG_MMCONF_SUPPORT_DEFAULT

<clip>

> and I wonder why this option is not selected for more platforms or even
> enabled by default (for systems having `MMCONF_SUPPORT`)? Reading the
> AMD whitepaper it claims that the Linux kernel can determine if MMIO is
> there during runtime [3]. Though I have no idea if that is only possible
> because of the firmware setting things up correctly beforehand.
> 
> Additionally for Intel systems `MMCONF_SUPPORT` (without `_DEFAULT`)
> seems to be selected in each mainboard where for AMD systems it is done
> under `src/cpu`.
> 

To use MMCONF_SUPPORT_DEFAULT you need to enable MMCONF as one of the
very first things you do.

For Intel see haswell/bootblock.c or gm45/early_init.c. Both use
explicit PCI IO style config access to do the job. This explicit early
write is missing on some other chipsets, thus you cannot blindly enable
MMCONF_SUPPORT_DEFAULT. There is no hardware restriction AFAIK, patches
welcome.

For AMD the same enable function is in cpu/amd/car/cache_as_ram.inc.
One really should not be using PCI IO configs on AMD multiprocessor
platforms, please submit patch to select MMCONF_SUPPORT_DEFAULT for any
AMD board with MMCONF_SUPPORT.

IMHO, MMCONF_SUPPORT and MMCONF_SUPPORT_DEFAULT belong under
northbridge/ while MMCONF_BASE_ADDRESS belongs under mainboard/.

Kyösti