binutils-gdb/mips-abiflags-0r.d at b3adc24a0713411ab38a21dc894dd40dbc5c8f4f - binutils-gdb - ClassFun

mirrors/binutils-gdb

mirror of https://sourceware.org/git/binutils-gdb.git synced 2024-12-11 19:33:33 +08:00

Maciej W. Rozycki 6798f8bfa6 MIPS/BFD: Fix the size of .reginfo' and .MIPS.abiflags' sections

Use the SEC_FIXED_SIZE flag to actually fix the size of `.reginfo' and
`.MIPS.abiflags' sections in `_bfd_mips_elf_always_size_sections', as
originally intended, removing link failures such as:

ld: final link failed: Section has no contents

or:

ld: final link failed: Bad value

or:

ld: foo: .reginfo section size should be 24 bytes, actual size is 32

and assertion failures like:

ld: BFD (GNU Binutils) 2.30.51.20180131 assertion fail .../bfd/elfxx-mips.c:14322

in link scenarios involving a linker script that either creates an
output `.reginfo' or `.MIPS.abiflags' section from scratch or produces
either section from different sections.  If such an output section's
size turns out to be incorrect according to the psABI, then the section
is either truncated or padded out to the correct size, as relevant.

This allows people to handle these sections in a link in an unusual way,
while still addressing the issue covered by commit 58807c48a5
("_bfd_mips_elf_final_link: Notify user about wrong .reginfo size").

The original arrangement, coming from an unindentified change made to
what was called `mips_elf_always_size_sections' back then, between
commit 02650bd0a9 ("This adds ABI flags to MIPS/ELF object files.")
and commit 252b5132c7 ("19990502 sourceware import"), also missing
from BFD ChangeLog files, assumed that the output section size is not
going to change after return from `bfd_elf_size_dynamic_sections', the
caller of that function, called in turn from `ldemul_before_allocation'
via `gld${EMULATION_NAME}_before_allocation' in ld/emultempl/elf32.em,
and ultimately from `lang_process'.  This is because later on in
`lang_process' processing `lang_size_sections' is called , happily
recalculating the section size, and it has actually already been the
case at the time of commit 252b5132c7 ("19990502 sourceware import"),
so the assumption was clearly incorrect right from the beginning.

	bfd/
	* elfxx-mips.c (_bfd_mips_elf_always_size_sections): Set
	SEC_FIXED_SIZE and SEC_HAS_CONTENTS flags for `.reginfo' and
	`.MIPS.abiflags' sections.
	(_bfd_mips_elf_final_link): Avoid reading beyond `.reginfo'
	section's end.

	ld/
	* testsuite/ld-mips-elf/reginfo-0.d: New test.
	* testsuite/ld-mips-elf/reginfo-0r.d: New test.
	* testsuite/ld-mips-elf/reginfo-1.d: New test.
	* testsuite/ld-mips-elf/reginfo-1r.d: New test.
	* testsuite/ld-mips-elf/reginfo-2.d: New test.
	* testsuite/ld-mips-elf/reginfo-2r.d: New test.
	* testsuite/ld-mips-elf/mips-abiflags-0.d: New test.
	* testsuite/ld-mips-elf/mips-abiflags-0r.d: New test.
	* testsuite/ld-mips-elf/mips-abiflags-1.d: New test.
	* testsuite/ld-mips-elf/mips-abiflags-1r.d: New test.
	* testsuite/ld-mips-elf/mips-abiflags-2.d: New test.
	* testsuite/ld-mips-elf/mips-abiflags-2r.d: New test.
	* testsuite/ld-mips-elf/reginfo-0.ld: New test linker script.
	* testsuite/ld-mips-elf/reginfo-1.ld: New test linker script.
	* testsuite/ld-mips-elf/mips-abiflags-0.ld: New test linker
	script.
	* testsuite/ld-mips-elf/mips-abiflags-1.ld: New test linker
	script.
	* testsuite/ld-mips-elf/reginfo-1.s: New test source.
	* testsuite/ld-mips-elf/reginfo-2.s: New test source.
	* testsuite/ld-mips-elf/mips-elf.exp: Run the new tests.

2018-02-19 18:38:41 +00:00

7 lines

204 B

Makefile

Raw Blame History

 #readelf: -x .MIPS.abiflags
 #name: MIPS .MIPS.abiflags section size 0 (relocatable)
 #source: empty.s RUN_OBJCOPY
 #objcopy_objects: -R .MIPS.abiflags
 #ld: -r -T mips-abiflags-0.ld
 #dump: mips-abiflags-0.d