Debugging Assembly with GDB

GDB's default behavior assumes source-level debugging: it steps by source line, shows local variables by name, and hides the underlying instructions. When you're debugging hand-written assembly, or chasing a bug that only manifests in a compiler's generated machine code, you need to switch GDB's mental model down a level — stepping by individual instructions rather than source lines, and reading raw register and memory state rather than named variables. The key command for this is layout asm (or layout regs for a combined register-and-code view), which switches GDB's text-mode UI into a disassembly view showing the actual instruction stream with the current instruction pointer highlighted.

At the instruction level, stepi (si) and nexti (ni) replace step and next — si executes exactly one machine instruction, stepping into calls, while ni executes one instruction but steps over calls, running the entire called function before stopping again. info registers dumps the full general-purpose register set (RAX, RBX, RDI, RIP, RFLAGS, and so on) in one shot, while print $rax or p/x $rax prints a single register, and x/4xg $rsp examines memory at RSP as four hexadecimal giant (64-bit) words — the x command's format letters (x for hex, d for decimal, i for disassembled instruction) combined with size letters (b byte, h halfword, w word, g giant) give precise control over how raw memory is displayed.

break *0x401234 sets a breakpoint at a raw address rather than a source line, which is essential in stripped or hand-written binaries with no symbol table; watch *(int*)0x404050 sets a hardware watchpoint that halts execution the instant that memory location's value changes, invaluable for catching exactly which instruction corrupts a variable when the culprit isn't obvious from reading the code. Because modern Linux binaries are typically compiled as position-independent executables and loaded at a randomized base address (ASLR), raw addresses from a previous run, a disassembly listing, or objdump output will not match the addresses GDB sees at runtime unless you either disable ASLR for the debug session (set disable-randomization on, GDB's default) or compute offsets relative to the load base using info proc mappings.