gcc 4.8.1, powerpc, .data.rel.ro.local entry with no corresponding .fixup entry

gcc 4.8.1, powerpc, .data.rel.ro.local entry with no corresponding .fixup entry

[Joerg Albert]

Hi,

we've recently compiled "Das U-Boot" with gcc 4.8.1 for powerpc target. This
bootloader relocates itself into RAM, using the content of the .fixup section.

A codeline similar to "printf("%p\n", "qwerty");" inside a static function
creates an address entry (for "qwerty") in .data.rel.ro.local, but I cannot
find a corresponding entry in the .fixup section of this object file. This
let the bootloader reference "qwerty" in the flash after it relocated to RAM.

Could someone please explain how the .fixup section is supposed to work?
Shall every .rela.data.* entry in the elf file have a corresponding entry in
the .fixup section?
Are there any command line options which may suppress the generation of the
data.rel.ro.* sections (as U-Boot does not write-protect parts of the RAM)?

The code works fine with gcc 4.7.x, but I haven't looked into the details of
the generated assembly code and section layout.


Thanks,
Joerg

details:
- gcc version: powerpc-softfloat-linux-gnuspe-gcc (Gentoo 4.8.1 p1.0,
pie-0.5.6) 4.8.1
- gcc command line (some -I removed):
  -g -gdwarf-2  -Os   -fpic -mrelocatable \
  -meabi \
  -D__KERNEL__ -DCONFIG_SYS_TEXT_BASE=0xef770000 \
  -fno-builtin  -ffreestanding \
  -isystem /usr/lib/gcc/powerpc-softfloat-linux-gnuspe/4.8.1/include \
  -nostdinc -pipe  -DCONFIG_PPC -D__powerpc__ -ffixed-r2 -Wa,-me500 \
  -msoft-float -mno-string -mspe=yes -mno-spe -Wall -Wstrict-prototypes \
  -fno-stack-protector -Wno-format-nonliteral -Wno-format-security \
  -fstack-usage"

Re: gcc 4.8.1, powerpc, .data.rel.ro.local entry with no corresponding .fixup entry

[Joerg Albert]

Hi,

switched to gcc 4.8.2, but the problem remains.

It seems like .data.rel.ro.local contains pointers only
and its entries are not covered by any entries in .fixup.
All other .data.rel.ro.* sections contain arbitrary data -
if there are any pointers, there is always a corresponding
entry in .fixup.

Therefore I put .data.rel.ro.local between .got2 and .got in the
linker script file:

    KEEP(*(.got2))
    KEEP(*(.data.rel.ro.local))
    KEEP(*(.got))

which seems to solve the problem.

Is the above assumption valid? Or may it break with some other code
or with the next compiler version? It's quite hard to find documentation
about the usage of .data.rel.ro.local.

Thanks,
Joerg