Re: [coreboot] [Coreboot] Misalignment with GCC 4.9.1

[Aaron Durbin]

On Mon, Oct 13, 2014 at 5:59 PM, Lin, Ryan  wrote:
> Hi Peter,
>
> It should be unpatched GCC 4.9, and I found out that .bs_init is force
> 32-byte alignment with GCC 4.9.
>
> Dump from ramstage.o
>
> 12 .bs_init      000000c8  00000000  00000000  00036160  2**5
>               CONTENTS, ALLOC, LOAD, RELOC, DATA

Do you know where the default alignment is coming from? What
ramstage.o did you look at? What did the symbols say for these
objects? Are the sizes for the OBJECT types set to 20 for sure (I'd
expect it to be)?

The following could be related:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61296

We might be playing whack-a-mole and need to align all of our special
sections up to a typical cacheline (64) in the linker scripts. Or we
switch all of our objects in special sections to be pointers to
objects so that we can pointer align them.
> As the result, padding datd is added to mrc_cache_update, pch_log and
> cbmem_bscb.
>
> 0001d1d0 T _bs_init_begin
>
> 0001d1e0 t mrc_cache_update <- 32 bytes
>
> 0001d200 t pch_log <- 32 bytes
>
> 0001d220 t finalize
>
> 0001d248 t spi_init_bscb
>
> 0001d25c t cbmem_bscb <- 36 bytes
>
> 0001d280 t disable_rom_cache_bscb
>
> 0001d2a8 T _bs_init_end
>
> If we add aligned(4) to the define as the following :
>
> #define BOOT_STATE_INIT_ATTR  __attribute__ ((used, aligned(4), section
> (“.bs_init)))
>
> We can get 4-byte alignment for .bs_init.
>
> What is your thoughts on this?
>
> Regards,
>
> Ryan