PR13616, linker should pad executable sections with nops, not zeros
authorAlan Modra <amodra@gmail.com>
Wed, 16 Oct 2019 10:53:29 +0000 (21:23 +1030)
committerAlan Modra <amodra@gmail.com>
Wed, 16 Oct 2019 12:37:27 +0000 (23:07 +1030)
commit22216541c1796e9e1331d6f4e16b03a6f02e7381
tree4591c034f7c483dc18ff09eb257d19d99e916f30
parent12234dfd5f70086fde87e1f6d56cf2aa2ce5ec4c
PR13616, linker should pad executable sections with nops, not zeros

This implements padding of orphan executable sections for PowerPC.
Of course, the simple implementation of bfd_arch_ppc_nop_fill and
removing the NOP definition didn't work, with powerpc64 hitting a
testsuite failure linking to S-records.  That's because the srec
target is BFD_ENDIAN_UNKNOWN so the test of bfd_big_endian (abfd) in
default_data_link_order therefore returned false, resulting in a
little-endian nop pattern.  The rest of the patch fixes that problem
by adding a new field to bfd_link_info that can be used to determine
actual endianness on targets like srec.

PR 13616
include/
* bfdlink.h (struct bfd_link_info <big_endian>): New field.
bfd/
* cpu-powerpc.c (bfd_arch_ppc_nop_fill): New function, use it
for all ppc arch info.
* linker.c (default_data_link_order): Pass info->big_endian to
arch_info->fill function.
ld/
* emulparams/elf64lppc.sh (NOP): Don't define.
* emulparams/elf64ppc.sh (NOP): Don't define.
* ldwrite.c (build_link_order): Use link_info.big_endian.  Move
code determining endian to use for data_statement to..
* ldemul.c (after_open_default): ..here.  Set link_info.big_endian.
bfd/ChangeLog
bfd/cpu-powerpc.c
bfd/linker.c
include/ChangeLog
include/bfdlink.h
ld/ChangeLog
ld/emulparams/elf64lppc.sh
ld/emulparams/elf64ppc.sh
ld/ldemul.c
ld/ldwrite.c