Skip to content

Fix MIPS n64 ELF relocation parsing - #1367

Open
plusls wants to merge 1 commit into
lief-project:mainfrom
plusls:fix/mips-n64-reloc
Open

Fix MIPS n64 ELF relocation parsing#1367
plusls wants to merge 1 commit into
lief-project:mainfrom
plusls:fix/mips-n64-reloc

Conversation

@plusls

@plusls plusls commented Jul 28, 2026

Copy link
Copy Markdown

MIPS n64 (ELFCLASS64 + EM_MIPS) uses a non-standard on-disk r_info layout: r_info = r_sym[4] | r_ssym[1] | r_type3[1] | r_type2[1] | r_type[1] instead of the generic ELF64 (sym << 32) | type. LIEF decoded all ELF64 relocations with the generic mask, which is wrong for MIPS n64:

  • Little-endian: r_sym is in the low 32 bits, so r_info >> 32 produces garbage symbol indices. This caused the AUTO dynsym count to explode (196 instead of 10) and relocation types to decode as UNKNOWN.

  • Big-endian: r_info & 0xffffffff drags r_ssym/r_type3/r_type2 into the type field. Compound relocations (e.g. R_MIPS_GPREL16 with non-zero Type2/Type3) decode as garbage; simple ones match by accident.

The MIPS n64 encode/decode is now isolated in an internal header src/ELF/reloc_utils.hpp, keeping the public Relocation class format-agnostic (same pattern as the existing Android SLEB128 handling at the Parser/Builder boundary).

The judgment key arch == MIPS && class == ELF64 matches binutils ABI_64_P macro. n32/o32 are ELFCLASS32 and handled by the Elf32 code path at compile time, so they cannot trigger the MIPS n64 branch.

Non-MIPS architectures: zero change in decode/encode paths.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant