Google is committed to advancing racial equity for Black communities. See how.

Assembly language in zxdb

Disassembly

The disassemble command (di for short) disassembles from the current location. If available, the instructions and call destinations are annotated with source line information:

[zxdb] di
miscsvc.cc:118
 ▶ 0x20bc1c7aa60a  mov     dword ptr [rbx + 0x10c], eax
miscsvc.cc:122
   0x20bc1c7aa610  movabs  rax, -0x5555555555555556
   0x20bc1c7aa61a  mov     qword ptr [rbx + 0xe8], rax
   0x20bc1c7aa621  mov     qword ptr [rbx + 0xe8], 0x0
   0x20bc1c7aa62c  mov     rdi, qword ptr [rbx + 0xb0]
   0x20bc1c7aa633  mov     rax, qword ptr [rbx + 0xe8]
   0x20bc1c7aa63a  mov     qword ptr [rbx + 0x20], rax
   0x20bc1c7aa63e  call    0x20d    ➔ std::__2::size<>()

The di command can also accept an address or symbol as a parameter. If given a function name, it disassembles the entire function:

[zxdb] di main
miscsvc.cc:88
   0x20bc1c7aa000  push    rbp
   0x20bc1c7aa001  mov     rbp, rsp
   0x20bc1c7aa004  push    rbx
   0x20bc1c7aa005  and     rsp, -0x20
   0x20bc1c7aa009  sub     rsp, 0x140
   0x20bc1c7aa010  mov     rbx, rsp
   0x20bc1c7aa013  mov     rax, qword ptr fs:[0x10]
   ...

Switches

The disassemble command accepts these switches:

  • --num=<lines> or -n <lines>: The number of lines or instructions to emit. Defaults to the instructions in the given function (if the location is a function name), or 16 otherwise.

  • --raw or -r: Output raw bytes in addition to the decoded instructions.

Stepping in machine instructions

Machine instructions can be stepped using the following Zxdb commands:

  • nexti / ni: Step to the next instruction, stepping over function calls.

  • stepi / si: Step the next instruction, following function calls.

For example:

[zxdb] ni
🛑 main(int, const char**) • main.cc:102
main.cc:99
 ▶ 0x23f711346233  mov   edx, 0x20
   0x23f711346238  call  0x35a3a3  ➔ __asan_memcpy
   0x23f71134623d  mov   rdi, qword ptr [rbx + 0x258]
   0x23f711346244  call  0x1677    ➔ $anon::DecodeCommandLine

[zxdb] ni
🛑 main(int, const char**) • main.cc:102
main.cc:99
 ▶ 0x23f711346238  call  0x35a3a3 ➔ __asan_memcpy
   0x23f71134623d  mov   rdi, qword ptr [rbx + 0x258]
   0x23f711346244  call  0x1677   ➔ $anon::DecodeCommandLine
   0x23f711346249  mov   rdi, qword ptr [rbx + 0x260]

Zxdb maintains information about whether the last command was an assembly command or a source-code and shows that information on stepping or breakpoint hits. To switch to assembly-language mode, type disassemble, and to switch back to source-code mode, type list.

Registers

The regs command shows the most common CPU registers.

[zxdb] regs
General Purpose Registers
      rax  0xfffffffffffffffa = -6
      rbx          0x50b7085b
      rcx                 0x0 = 0
      rdx      0x2023de8c87a0
      rsi  0x7fffffffffffffff
      rdi          0x50b7085b
      rbp      0x224bb1e0b950
      rsp      0x224bb1e0b928
      ...

There are other categories and options for CPU registers that can be shown by switches to the regs command:

  • --all or -a: Enable all register categories (does not imply -e).

  • --float or -f: Prints the dedicated floating-point registers. In most cases you should use --vector instead because all 64-bit ARM code and most x64 code uses vector registers for floating point.

  • --vector or -v: Prints the vector registers. See below for more details.

  • --debug or -d: Prints the debug registers.

  • --extended or -e: Enables more verbose flag decoding. This enables more information that is not normally useful for everyday debugging. This includes information such as the system level flags within the rflags register for x64.

Registers in expressions

Registers can be used in expressions like variables. The canonical name of a register is $reg(register name).

[zxdb] print $reg(x3)
79

In addition, the raw register name can be used if there is no variable with the same name:

[zxdb] print x3
79

Registers can be assigned using the normal expression evaluation syntax:

[zxdb] print x3 = 0
0

Vector registers

The regs --vector command displays vector registers in a table according to the current vector-format setting. Use get vector-format to see the current value and documentation, and set vector-format <new-value> to set a new vector format. Possible values are:

  • i8 (signed) or u8 (unsigned): Array of 8-bit integers.
  • i16 (signed) or u16 (unsigned): Array of 16-bit integers.
  • i32 (signed) or u32 (unsigned): Array of 32-bit integers.
  • i64 (signed) or u64 (unsigned): Array of 64-bit integers.
  • i128 (signed) or u128 (unsigned): Array of 128-bit integers.
  • float: Array of single-precision floating point.
  • double: Array of double-precision floating point. This is the default.
[zxdb] set vector-format double

[zxdb] regs -v
Vector Registers
  mxcsr 0x1fa0 = 8096

   Name [3] [2] [1]       [0]
   ymm0   0   0   0         0
   ymm1   0   0   0   3.14159
   ymm2   0   0   0         0
   ymm3   0   0   0         0
   ...

Vector registers can also be used like arrays in expressions. The vector-format setting controls how each register is converted into an array value. For example, to show the low 32 bits interpreted as a floating-point value of the x86 vector register ymm1:

[zxdb] set vector-format float

[zxdb] print ymm1[0]
3.14159

When converting to an array, the low bits are assigned to be index 0, increasing from there. Note that the vector register table in regs are displayed with the low values on the right side.