ddb disassembler: handle multi-byte nops

[markj]

Summary

x86_64 has supported multibyte nops for a while; they have an opcode of 0f 1f, sometimes prefixed with 0x66 (operand size override) and 0x2e (cs segment override). The ddb disassembler doesn't support them, but they seem to appear fairly frequently (see cam_strvis() for several examples).

This change just adds an instruction table for 0x0f1, with a single entry for 1f. I use "nopl" by default, and "nopw" if the 0x66 prefix is used, for consistency with binutils.

Test Plan

ddb is able to disassemble all the instructions in cam_strvis() and sysctl_sysctl_next_ls() (the latter contains a 15-byte nop on my laptop running HEAD).

[kib]

I never read the ddb disassembler code, but I have no objections against the change.

Does i386 need the same patch ?

[markj]

• Detect multibyte nops on i386.

[markj]

In D1830#3, @kostikbel wrote:

I never read the ddb disassembler code, but I have no objections against the change.

Does i386 need the same patch ?

Yes, according to the ia32 manual. For some reason clang doesn't appear to emit them though (looking at ref11-i386's kernel). I uploaded the corresponding change anyway, it shouldn't hurt.

[markj]

Closed by commit rS278655 (authored by @markj).