CSE3101 Assembly Language Introduction

Why use assembly language?

• Size optimization
• Priveliged code for operating systems
• Writing a compiler
• Reverse engineering
• Optimization for speed (but most C/C++ compilers can do better)

Intel architecture history

• 8088/8086 (1979)
  • Single tasking (real mode)
  • 1M (20 bit) address space (overlapping 64K segments)
  • 16 bit registers: AX, BX, CX, DX, SI, DI, SP, BP
  • 5 MHz (0.33 MIPS)
  • Used in original 1981 IBM-PC running BASIC
• 80386 (1985)
  • Adds protected mode to support multitasking
  • 4 GB (32 bit) flat address space
  • 64 TB (46 bit) virtual memory support
  • Adds 32 bit registers: EAX, EBX, ECX, EDX, ESI, EDI, ESP, EBP
  • 16 MHz (5-6 MIPS)
• 80486 (1989)
  • Adds 80-bit floating point registers on-chip: ST(0) - ST(7)
  • 25 MHz (20 MIPS)
• Pentium MMX (1997)
  • Adds 64-bit MMX (packed 8x8, 4x16, 2x32 bit integer operations): MM0 - MM7 (overlapping ST(0) - ST(7))
  • 166-233 MHz
• Pentium III (1999)
  • Adds 128-bit SSE (packed 4x32 float operations): XMM0 - XMM7
  • 450-800 MHz
• Pentium 4 (2000)
  • Adds SSE2 (packed 16x8 8x16, 4x32 bit integer and 2x64 bit double operations)
  • Up to 3.4 GHz, 11000 MIPS in 2004
• AMD Opteron, Athlon64 (2003)
  • Adds XMM8 - XMM15
  • Intel to add to 64-bit processors in 2004
• All versions are backwards compatible back to 8088

IA-32 Architecture

General purpose registers

  
  31         16 15  8 7   0
  +------------+-----+-----+
  |    EAX     | AH  | AL  |  Accumulator (EAX = 32 bits, AX = 16 bits, AH, AL = 8 bits)
  +------------+-----+-----+
  |    EBX     | BH  | BL  |  Table base
  +------------+-----+-----+
  |    ECX     | CH  | CL  |  Counter
  +------------+-----+-----+
  |    EDX     | DH  | DL  |  High bits of AX, EAX
  +------------+-----+-----+
  |    ESI     |    SI     |  Source index
  +------------+-----------+
  |    EDI     |    DI     |  Destination index
  +------------+-----------+
  |    EBP     |    BP     |  Base pointer
  +------------+-----------+
  |    ESP     |    SP     |  Stack pointer
  +------------+-----------+

• CF = carry or borrow (unsigned overflow)
• OF = overflow (signed)
• ZF = zero
• SF = sign (1 = negative)
• DI = direction (set to 1 = backwards) for array operations

• ST(0) - ST(7): 80 bit FPU floating point stack, ST(0) is top
• FPU status flags (rounding control, exceptions)
• MM0 - MM7: 64 bit packed integers, overlaps ST(0) - ST(7)
• XMM0 - XMM7: 128 bit packed float (SSE), and int and double (SSE2) (unrelated to MM0-7)
• CS, SS, DS, ES, FS, GS - 16 bit segment registers (priveliged)
• Test, debug, and other priveliged registers

Real mode

• 1 MB memory in 64K segments addressed by segment registers
• Pointers and int are 2 bytes (4 bytes for FAR pointers)
• Programs have full control over hardware
• For single tasking OS (MS-DOS)
• Initial mode after boot

• 4 GB flat memory model
• Pointers and int are 4 bytes
• For multitasking OS (Windows, Linux, UNIX)
• 4 privelige levels (OS = 0, user = 3)
• OS assigns read, write, execute permissions to memory segments
• Traps priveliged instructions in user mode (IN, OUT, CLI, MOV CS, etc.)
• Real model programs run in virtual 8086 mode

Program organization

Windows .EXE 32-bit (flat model)

• Protected mode (4 GB memory)
• Can link to C programs and libraries
• Can make Windows system calls, but not BIOS or MSDOS
• Code, data and stack segments

 0ffffffffh  +-------------+  <-- SS limit
             |             |  <-- ESP
             |   Stack     |
             |             |
             +-------------+  <-- SS base
             |             |
             |             |
             +-------------+  <-- DS, ES, FS, GS limits
             |  Dynamic    |
             |  data       |
  +-------+  +-------------+
  |       |  |  Static     |
  | .exe  |  |  data       |
  |       |  +-------------+  <-- DS, ES, FS, GS bases
  | File  |  |             |  <-- CS limit
  |       |  |  Code       |
  |       |  |             |  <-- EIP
  +-------+  +-------------+  <-- CS base
             |             |
         0h  +-------------+ 

MSDOS .EXE 16-bit (large model)

• Real mode (1 MB memory)
• Can link to 16-bit C or libraries (MARS, commercial Borland, Microsoft, Intel)
• Can call BIOS or MSDOS, but not Windows
• May have more than one code and data segment, one stack segment

    0fffffh  +-------------+  <-- SP
             |             |
             |   Stack     |
             |             |
             +-------------+  <-- SS
             |             |
             |             |
             +-------------+
             |  Dynamic    |
             |  data       |
  +-------+  +-------------+
  |       |  |  Static     |
  | .exe  |  |  data       |
  |       |  +-------------+  <-- DS, ES
  | File  |  |             |
  |       |  |  Code       |
  |       |  |             |  <-- IP
  +-------+  +-------------+  <-- CS
             |             |
         0h  +-------------+

• Real mode (64K memory)
• File contains pure code, no header
• Not linked to anything
• Can call BIOS and MSDOS, but not Windows
• Flat 64K memory model (all segments overlap)
• Execution starts at address 100h (first byte of file)

 0ffffh +------------+
        |            | <-- SP (stack grows downward)
        | free       |
        | memory     |
        |            |
        +------------+
        | copy of    |
        | prog.com   |
  100h  +------------+  <-- IP (instruction pointer)
   80h  |  PSP       |  <-- (command line arguments at 80h)
   0h   +------------+  <-- CS, SS, DS, ES

Software development

• MASM - Microsoft assembler (included with book)
  • Links to Visual C/C++, Intel C/C++, Borland C/C++, Digital MARS, MINGW, but not DJGPP
• TASM - Borland assembler (not included with free version of Borland C++)
  • Mostly MASM compatible except for some advanced features
• NASM (free) Documentation
  • For Windows or Linux
  • Largely incompatible with MASM, but uses familiar syntax
  • Links to Borland, MINGW, DJGPP C/C++ (but not MARS)
• AS (free, included with DJGPP and MINGW C/C++ compilers) Documentation
  • Uses AT&T syntax (completely different from MASM)
  • Compatible with Linux
  • Links to gcc only (DJGPP and MINGW)
• Inline assembler in C
  • Uses asm statement in C
  • Compiler dependent syntax (not portable, even on same hardware)
  • Borland requires TASM (not included with free version)
  • MARS requires its own assembler (not included with free version)
  • DJGPP and MINGW use AT&T syntax and AS
• C/C++ compilers: Borland, MARS, g++ are free at www.cplusplus.com/info/compilers/