dis

The x86-64 disassembler in the kernel, kernel/dis.c, behind xray, jitdis, dis and disr.

Using it

addr n dis          n instructions starting at addr
addr len disr       every instruction in len bytes from addr
'fib jitdis         compile fib and disassemble all of it
'fib xray           fib's source, then its disassembly

dis and disr take any address at all, not just compiled code. 0x100000 20 dis disassembles the start of the kernel itself. Point it at data and you get nonsense, since it can't tell code from data.

What it knows

It decodes the general-purpose integer instruction set: moves, arithmetic, logic, shifts and rotates, lea, imul, idiv, conditional jumps, setcc and cmovcc, call, ret, push and pop, the string instructions, and the system ones the kernel itself uses, like lgdt, lidt, invlpg, rdmsr, wrmsr, cpuid, rdtsc, cli, sti, hlt and iretq. That covers everything the JIT generates and everything gcc emits for the kernel, which is compiled with no SSE at all.

It doesn't know SSE, AVX, x87 or anything else exotic. Bytes it can't decode come out as something obviously wrong, rather than something subtly wrong.

Output

One instruction per line: the address, the bytes, and the instruction in Intel syntax, destination first. Relative jumps and calls are resolved to their absolute targets, so you can follow them by hand with another dis.

How it's checked

make host builds the interpreter as a Linux program, and its @dis name directive writes a compiled word's raw bytes to build/dis.bin and the disassembler's own listing to build/dis.txt. Compare the .txt with objdump -D -b binary -m i386:x86-64 -M intel on the .bin to check it. See testing for the other directives.