shouldn't asm-generic/barebox.lds.h be arch-independent?

[Antony Pavlov]

On 18 November 2012 21:52, Robert P. J. Day <rpjday at crashcourse.ca> wrote:
>
>   perusing the code to see how the final barebox executable is created
> and ran across this opening to include/asm-generic/barebox.lds.h:
>
> #if defined CONFIG_ARCH_IMX25 || \
>         defined CONFIG_ARCH_IMX35 || \
>         defined CONFIG_ARCH_IMX51 || \
>         defined CONFIG_ARCH_IMX53 || \
>         defined CONFIG_ARCH_IMX6 || \
>         defined CONFIG_X86 || \
>         defined CONFIG_ARCH_EP93XX
> #include <mach/barebox.lds.h>
> #endif
>
> #ifndef PRE_IMAGE
> #define PRE_IMAGE
> #endif
>
>   i find that more than a little confusing and inappropriate -- it
> strikes me that the allegedly "generic" lds.h file shouldn't be
> testing for an arbitrary set of inexplicable machine/arch
> combinations, especially when there is absolutely no explanation about
> what makes that particular set of symbols magic with respect to this
> file.
>
>   a minute or two of perusing shows that what it's for is inserting
> some special "pre-image" content into the executable, but that means
> that for each new selection that needs that, this file will have to be
> changed and that's just messy.
>
>   the linux kernel has a simpler strategy for
> asm-generic/vmlinux.lds.h, as you see in this patch snippet that added
> bss first section functionality to that file earlier this year:
>
> +/*
> + * Allow archectures to redefine BSS_FIRST_SECTIONS to add extra
> + * sections to the front of bss.
> + */
> +#ifndef BSS_FIRST_SECTIONS
> +#define BSS_FIRST_SECTIONS
> +#endif
> +
>  #define BSS(bss_align)                                                 \
>         . = ALIGN(bss_align);                                           \
>         .bss : AT(ADDR(.bss) - LOAD_OFFSET) {                           \
> +               BSS_FIRST_SECTIONS                                      \
>                 *(.bss..page_aligned)                                   \
>                 *(.dynbss)                                              \
>                 *(.bss)                                                 \
>
>   that would seem to be a much nicer solution as it pushes the
> selection of pre-image content back to the architectures where it
> belongs.
>
>   the fact that the current approach is prone to "error" can be seen
> in the file arch/mips/lib/barebox.lds.S:
>
> #include <asm-generic/barebox.lds.h>
>
> OUTPUT_ARCH(mips)
> ENTRY(_start)
> SECTIONS
> {
>         . = TEXT_BASE;
>
>         . = ALIGN(4);
>         .text      :
>         {
>                 _start = .;
>                 *(.text_entry*)
>                 _stext = .;
>                 _text = .;
>                 __bare_init_start = .;
>                 *(.text_bare_init*)
>                 __bare_init_end = .;
>                 *(.text*)
>         }
>         BAREBOX_BARE_INIT_SIZE
>
>         PRE_IMAGE  <--- ?????
>
> apparently, the MIPS architecture is including the "PRE_IMAGE"
> content, despite the fact that it can't possibly be defined.  it's not
> wrong, it's just pointless and potentially confusing.

Unfortunately nowadays MIPS lacks lowlevel initialization support.
Yesterday I mailed RFC about pbl (prebootloader) support. It needs
much work, but IMHO after incorporating MIPS pbl support the status of
"PRE_IMAGE" will be more clear.

-- 
Best regards,
  Antony Pavlov