[coreboot] When should we retire newconfig?
patrick at georgi-clan.de
Tue Jan 12 13:18:59 CET 2010
Am 11.01.2010 23:41, schrieb Peter Stuge:
> Patrick Georgi wrote:
>> 4. Implement it like in southbridge/amd/amd8111:
>> 4a. Add a new file (I recommend
>> "southbridge/$vendor/$device/bootblock.c") with a static void
>> bootblock_southbridge_init(void) function that enables rom mapping
>> 4b. Set BOOTBLOCK_SOUTHBRIDGE_INIT to point to the file
> Please can some of these be renamed?
> map_rom.c (or map_full_rom.c or possibly enable_rom_decode.c)
> southbridge_map_rom() (or similar per filename above)
We already have $chipsetid_enable_rom.c in some directories, which can't
easily be reused for tinybootblock (romcc vs. gcc differences).
I used such generic names because we might decide that we also want to
have serial enable in the bootblock.
> My reasoning is that the build system is a great place to describe
> _when_ files are used, but code and filenames need only say _what_
> they do - if there is one single convention.
> I think 4b. should be made general. I don't know if I prefer always
> including the file and having an empty function, or allowing the file
> to be included only for southbridges where it actually exists. Empty
> functions really suck; source code noise. Automagic build rules do
> too; rename a file and get a runtime error instead of a buildtime
> error. I have to go with the lesser evil; the empty function.
If the config variable and the filename disagree, the compiler fails to
find the file. That's a buildtime error, as far as I can see.
Your proposal would mean that there's a default value for
BOOTBLOCK_SOUTHBRIDGE_INIT, probably in src/arch/i386/Kconfig, which
points to the dummy. You forget to change the value and it uses the
empty function, instead of what it actually should use - that's a silent
(but instant) runtime error.
More information about the coreboot