| Summary: | [binutils] x86_64-objdump -dr fails too disassmble correctly | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Danny van Dyk (RETIRED) <kugelfang> |
| Component: | New packages | Assignee: | Gentoo Toolchain Maintainers <toolchain> |
| Status: | RESOLVED INVALID | ||
| Severity: | normal | CC: | amd64 |
| Priority: | High | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
INVALID. Gcc emmits this and gas groks this sample. Seems like some undocumented/poorly documented "feature". My bad, the relocation is of type 'R_X86_64_PC32', emphasis on PC32 as in Program Counter (%rip). Didn't see that before. Sorry for all the fuss. |
Consider the following listing (taken from a proprietary piece of objectcode, thus the masquerading of symbols and offsets): XXc0: 66 3b 05 00 00 00 00 cmp 0(%rip),%ax # XXc7 <SOMESYMBOL+0x18> XXc3: R_X86_64_PC32 .rodata+0xSOMECONST Let's dissec the first instruction: 66 <-- Operand size override (to 16bit) 3b <-- CMP for one register, stored in reg part of the ModR/M byte 05 <-- ModR/M byte: 00 000 101 Mod 00 <-- Director Addressing using 32bit addresses directly following the ModR/M byte OR indirect addressing using the register specified by R/M reg 000 <-- Select the accumulator as register. For 16 bit, this is %ax. R/M 101 <-- Treat the the following byte as _direct_ address. 00 00 00 00 <-- The direct address (to be substituted by the following relocation) The binary should thus be read as: cmp ($0x00000000),%ax I consider this a bug. Alas, I don't know if this kind of notation is used purposefully, as using %rip there makes absolutely no sense. IIRC you can't use %rip as source of a comparison. You'd need to first mov its contents to a GPR.