mirror of
https://sourceware.org/git/binutils-gdb.git
synced 2024-12-21 04:42:53 +08:00
506b86a4f5
The strip-13 test runs into difficulty using dc.a on some targets. dc.a writes a power of 2 number of bytes large enough to contain a target address. On some targets, eg. avr-elf, this can be 2 bytes but the ELF format used require words of 4 bytes to make up a relocation entry. There was a hack in the test, duplicating the reloc type into what is normally the addend to make the test work when little-endian, but that hack fails for similar big-endian targets. This patch fixes that problem by arranging to emit 32-bit and 64-bit ELF relocs using .4byte and .8byte directives, chosen as appropriate for the ELF size. I've also bumped the reloc number tested to a higher unused value, and made the number more easily parameterised by target should that be necessary in the future. Whether REL or RELA relocs are used is now chosen by a new is_rela readelf test of an object file, rather than by lists of targets. * testsuite/lib/binutils-common.exp (is_elf64): Use directory of input file for readelf.out. (is_rela): New proc. * testsuite/binutils-all/objcopy.exp (elf64): Set new variable from first result of is_elf64 test. (reloc_format): Set using is_rela. (strip-13): Pass RELOC and ELF64 to assembler. (strip-14, strip-15): Use elf64 rather than calling is_elf64 again. * testsuite/binutils-all/strip-13.d: Run for more targets. * testsuite/binutils-all/strip-13mips64.s: Use RELOC and set addend to zero. * testsuite/binutils-all/strip-13rel.s: Use RELOC and ELF64. Don't add _NONE reloc. * testsuite/binutils-all/strip-13rela.s: Likewise.
13 lines
131 B
ArmAsm
13 lines
131 B
ArmAsm
.text
|
|
foo:
|
|
.dc.l 0x12345678
|
|
|
|
.section .rel.text
|
|
.ifdef ELF64
|
|
.8byte 0
|
|
.8byte RELOC
|
|
.else
|
|
.4byte 0
|
|
.4byte RELOC
|
|
.endif
|