https://sourceware.org/bugzilla/show_bug.cgi?id=32624
--- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> --- (In reply to Jan Beulich from comment #10) > I'm going solely from the output shown in patch 5. The first of the two > VPERMB clearly looks wrong in the linked executable. I may be missing > something, as I'm certainly aware of the conditional around the invocation > of elf_i386_convert_load_reloc(). > > With the first LEA in the example also using R_386_GOT32 (it's only the > first ADD which gets to use R_386_GOT32X as per the provided output), how > come it is being transformed? Afaict it's not elf_i386_convert_load_reloc() > that's responsible; this happens right in elf_i386_relocate_section() > afaict. Search for occurrences of 0x8d there. R_386_GOT32 is resolved, not transformed, which is independent of instructions. If you believe R_386_GOT32 is mishandled for some instructions, please file a bug against ld, not gas, with a proper testcase. -- You are receiving this mail because: You are on the CC list for the bug.
