I'm working on Solaris 10 on an opteron machine, with 64-bit support enabled. Using binutils-2.16.1, objcopy changes the alignment fields in the elf program header of a shared library to inappropriate values for the platform.
The shared library used to test was generated with gcc-4.0.1, configured to use the gnu assembler and the solaris linker (as suggested in the config notes at: http://gcc.gnu.org/install/specific.html#ix86-x-solaris210). objcopy libfizzle32.so libfizzle32.so.copied objdump -x libfizzle32.so > libfizzle32.so.dump objdump -x libfizzle32.so.copied > libfizzle32.so.copied.dump diff libfizzle32.so.dump libfizzle32.so.copied.dump 2,3c2,3 < libfizzle32.so: file format elf32-i386 < libfizzle32.so --- > libfizzle32.so.copied: file format elf32-i386 > libfizzle32.so.copied 9c9 < LOAD off 0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**16 --- > LOAD off 0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**12 11c11 < LOAD off 0x00000564 vaddr 0x00010564 paddr 0x00000000 align 2**16 --- > LOAD off 0x00000564 vaddr 0x00010564 paddr 0x00010564 align 2**12 13,14c13,14 < DYNAMIC off 0x00000588 vaddr 0x00010588 paddr 0x00000000 align 2**0 < filesz 0x000000b0 memsz 0x00000000 flags rwx --- > DYNAMIC off 0x00000588 vaddr 0x00010588 paddr 0x00010588 align 2**2 > filesz 0x000000b0 memsz 0x000000b0 flags rwx The same happens for a 64 bit target: diff libfizzle64.so.dump libfizzle64.so.copied.dump 2,3c2,3 < libfizzle64.so: file format elf64-x86-64 < libfizzle64.so --- > libfizzle64.so.copied: file format elf64-x86-64 > libfizzle64.so.copied 9c9 < LOAD off 0x0000000000000000 vaddr 0x0000000000000000 paddr 0x0000000000000000 align 2**16 --- > LOAD off 0x0000000000000000 vaddr 0x0000000000000000 paddr > 0x0000000000000000 align 2**4 11c11 < LOAD off 0x00000000000007f8 vaddr 0x00000000000107f8 paddr 0x0000000000000000 align 2**16 --- > LOAD off 0x00000000000007f8 vaddr 0x00000000000107f8 paddr > 0x00000000000107f8 align 2**5 13,15c13,15 < DYNAMIC off 0x0000000000000840 vaddr 0x0000000000010840 paddr 0x0000000000000000 align 2**0 < filesz 0x0000000000000160 memsz 0x0000000000000000 flags rw- < 0x6464e550 off 0x0000000000000120 vaddr 0x0000000000000120 paddr 0x0000000000000000 align 2**3 --- > DYNAMIC off 0x0000000000000840 vaddr 0x0000000000010840 paddr > 0x0000000000010840 align 2**3 > filesz 0x0000000000000160 memsz 0x0000000000000160 flags rw- > 0x6464e550 off 0x0000000000000120 vaddr 0x0000000000000120 paddr > 0x0000000000000120 align 2**3 According to the solaris linker and library guide http://docs.sun.com/app/docs/doc/817-1984/6mhm7pl2a?a=view#chapter6-34713 p_align should be 2**16 for 32 bit segments, and 2**20 for 64 bit segments. There seems to be a little inconsistency here, because on x86-64, the solaris linker generates p_align as 2**16, but the solaris documentation suggests it should be 2**20. Either way, the p_align values after objcopy do not meet the requirements. I also tried this on Solaris on SPARC, but objcopy correctly handled the alignment for both 32 bit targets and 64 bit targets, so this seems to be x86 specific. I have also verified that this probem exists in the 07/22 binutils snapshot. Thanks, Andrew _______________________________________________ bug-binutils mailing list bug-binutils@gnu.org http://lists.gnu.org/mailman/listinfo/bug-binutils
