[coreboot] GCC update broke AMD Fam10h boot

Julius Werner jwerner at chromium.org
Thu Mar 19 23:54:54 CET 2015

> You are right that it would work, but back solving for which alignment
> the compiler decided is hard. It's definitely whack-a-mole in that
> case. It could have very well decided 16 too. Without any insight as
> to why that breaks down. Or you go the route of putting alignments on
> structures just for that behavior which I consider not the best
> approach.

Well, we only need to pick an alignment *larger* than what the
compiler decides, right? I'd hazard a guess that 8 would be enough for
a long time (GCC probably just got confused between 32 and 64 bit
modes or something... there's no other reason I could think of to
explain this).

>> From that I'd conclude that if you did:
>> extern struct boot_state_init_entry _bs_init_begin[];
>> extern struct boot_state_init_entry _bs_init_end[];
>> for (cur = _bs_init_begin; cur < _bs_init_end; cur++) ...
>> the compiler shouldn't be able to guarantee that the first array
>> wasn't empty and therefore pointing to "one after the end" which is
>> the start of the next array in the first iteration.
> I think you are right on that point -- using '<' instead of equality.
> But those aren't pointer types. Not sure if the spec things arrays
> like that are poitners or a separate type. Either way, hopefully there
> isn't some language in the spec that suggests comparisons like that
> can be optimized away.

The array degenerates to a pointer at that point AFAIK. You could also
write &_bs_init_begin... for an array type the two should be pretty
much equivalent.

More information about the coreboot mailing list