1 //:: simulated x86 registers
  2 
  3 :(before "End Types")
  4 enum {
  5   EAX,
  6   ECX,
  7   EDX,
  8   EBX,
  9   ESP,
 10   EBP,
 11   ESI,
 12   EDI,
 13   NUM_INT_REGISTERS,
 14 };
 15 union reg {
 16   int32_t i;
 17   uint32_t u;
 18 };
 19 :(before "End Globals")
 20 reg Reg[NUM_INT_REGISTERS] = { {0} };
 21 uint32_t EIP = 0;
 22 :(before "End Reset")
 23 bzero(Reg, sizeof(Reg));
 24 EIP = 0;
 25 
 26 //:: simulated flag registers; just a subset that we care about
 27 
 28 :(before "End Globals")
 29 bool SF = false;  // sign flag
 30 bool ZF = false;  // zero flag
 31 bool OF = false;  // overflow flag
 32 :(before "End Reset")
 33 SF = ZF = OF = false;
 34 
 35 //: how the flag registers are updated after each instruction
 36 
 37 :(before "End Includes")
 38 // Combine 'arg1' and 'arg2' with arithmetic operation 'op' and store the
 39 // result in 'arg1', then update flags.
 40 // beware: no side-effects in args
 41 #define BINARY_ARITHMETIC_OP(op, arg1, arg2) { \
 42   /* arg1 and arg2 must be signed */ \
 43   int64_t tmp = arg1 op arg2; \
 44   arg1 = arg1 op arg2; \
 45   trace(2, "run") << "storing 0x" << HEXWORD << arg1 << end(); \
 46   SF = (arg1 < 0); \
 47   ZF = (arg1 == 0); \
 48   OF = (arg1 != tmp); \
 49 }
 50 
 51 // Combine 'arg1' and 'arg2' with bitwise operation 'op' and store the result
 52 // in 'arg1', then update flags.
 53 #define BINARY_BITWISE_OP(op, arg1, arg2) { \
 54   /* arg1 and arg2 must be unsigned */ \
 55   arg1 = arg1 op arg2; \
 56   trace(2, "run") << "storing 0x" << HEXWORD << arg1 << end(); \
 57   SF = (arg1 >> 31); \
 58   ZF = (arg1 == 0); \
 59   OF = false; \
 60 }
 61 
 62 //:: simulated RAM
 63 
 64 :(before "End Globals")
 65 vector<uint8_t> Mem;
 66 uint32_t End_of_program = 0;
 67 :(before "End Reset")
 68 Mem.clear();
 69 Mem.resize(1024);
 70 End_of_program = 0;
 71 :(before "End Includes")
 72 // depends on Mem being laid out contiguously (so you can't use a map, etc.)
 73 // and on the host also being little-endian
 74 #define SET_WORD_IN_MEM(addr, val)  *reinterpret_cast<int32_t*>(&Mem.at(addr)) = val;
 75 
 76 //:: core interpreter loop
 77 
 78 :(scenario add_imm32_to_eax)
 79 # In scenarios, programs are a series of hex bytes, each (variable-length)
 80 # instruction on one line.
 81 #
 82 # x86 instructions consist of the following parts (see cheatsheet.pdf):
 83 #   opcode        ModR/M                    SIB                   displacement    immediate
 84 #   instruction   mod, reg, Reg/Mem bits    scale, index, base
 85 #   1-3 bytes     0/1 byte                  0/1 byte              0/1/2/4 bytes   0/1/2/4 bytes
 86   ¦ 05                                                                            0a 0b 0c 0d  # add 0x0d0c0b0a to EAX
 87 # All hex bytes must be exactly 2 characters each. No '0x' prefixes.
 88 +load: 1 -> 05
 89 +load: 2 -> 0a
 90 +load: 3 -> 0b
 91 +load: 4 -> 0c
 92 +load: 5 -> 0d
 93 +run: add imm32 0x0d0c0b0a to reg EAX
 94 +run: storing 0x0d0c0b0a
 95 
 96 :(code)
 97 // helper for tests: load a program into memory from a textual representation
 98 // of its bytes, and run it
 99 void run(const string& text_bytes) {
100   load_program(text_bytes);
101   EIP = 1;  // preserve null pointer
102   while (EIP < End_of_program)
103   ¦ run_one_instruction();
104 }
105 
106 // skeleton of how x86 instructions are decoded
107 void run_one_instruction() {
108   uint8_t op=0, op2=0, op3=0;
109   trace(2, "run") << "inst: 0x" << HEXWORD << EIP << end();
110   switch (op = next()) {
111   case 0xf4:  // hlt
112   ¦ EIP = End_of_program;
113   ¦ break;
114   // our first opcode
115   case 0x05: {  // add imm32 to EAX
116   ¦ int32_t arg2 = imm32();
117   ¦ trace(2, "run") << "add imm32 0x" << HEXWORD << arg2 << " to reg EAX" << end();
118   ¦ BINARY_ARITHMETIC_OP(+, Reg[EAX].i, arg2);
119   ¦ break;
120   }
121   // End Single-Byte Opcodes
122   case 0x0f:
123   ¦ switch(op2 = next()) {
124   ¦ // End Two-Byte Opcodes Starting With 0f
125   ¦ default:
126   ¦ ¦ cerr << "unrecognized second opcode after 0f: " << HEXBYTE << NUM(op2) << '\n';
127   ¦ ¦ exit(1);
128   ¦ }
129   ¦ break;
130   case 0xf3:
131   ¦ switch(op2 = next()) {
132   ¦ // End Two-Byte Opcodes Starting With f3
133   ¦ case 0x0f:
134   ¦ ¦ switch(op3 = next()) {
135   ¦ ¦ // End Three-Byte Opcodes Starting With f3 0f
136   ¦ ¦ default:
137   ¦ ¦ ¦ cerr << "unrecognized third opcode after f3 0f: " << HEXBYTE << NUM(op3) << '\n';
138   ¦ ¦ ¦ exit(1);
139   ¦ ¦ }
140   ¦ ¦ break;
141   ¦ default:
142   ¦ ¦ cerr << "unrecognized second opcode after f3: " << HEXBYTE << NUM(op2) << '\n';
143   ¦ ¦ exit(1);
144   ¦ }
145   ¦ break;
146   default:
147   ¦ cerr << "unrecognized opcode: " << HEXBYTE << NUM(op) << '\n';
148   ¦ exit(1);
149   }
150 }
151 
152 void load_program(const string& text_bytes) {
153   uint32_t addr = 1;
154   istringstream in(text_bytes);
155   in >> std::noskipws;
156   while (has_data(in)) {
157   ¦ char c1 = next_hex_byte(in);
158   ¦ if (c1 == '\0') break;
159   ¦ if (!has_data(in)) {
160   ¦ ¦ raise << "input program truncated mid-byte\n" << end();
161   ¦ ¦ return;
162   ¦ }
163   ¦ char c2 = next_hex_byte(in);
164   ¦ if (c2 == '\0') {
165   ¦ ¦ raise << "input program truncated mid-byte\n" << end();
166   ¦ ¦ return;
167   ¦ }
168   ¦ Mem.at(addr) = to_byte(c1, c2);
169   ¦ trace(99, "load") << addr << " -> " << HEXBYTE << NUM(Mem.at(addr)) << end();
170   ¦ addr++;
171   }
172   End_of_program = addr;
173 }
174 
175 char next_hex_byte(istream& in) {
176   while (has_data(in)) {
177   ¦ char c = '\0';
178   ¦ in >> c;
179   ¦ if (c == ' ' || c == '\n') continue;
180   ¦ while (c == '#') {
181   ¦ ¦ while (has_data(in)) {
182   ¦ ¦ ¦ in >> c;
183   ¦ ¦ ¦ if (c == '\n') {
184   ¦ ¦ ¦ ¦ in >> c;
185   ¦ ¦ ¦ ¦ break;
186   ¦ ¦ ¦ }
187   ¦ ¦ }
188   ¦ }
189   ¦ if (c == '\0') return c;
190   ¦ if (c >= '0' && c <= '9') return c;
191   ¦ if (c >= 'a' && c <= 'f') return c;
192   ¦ if (c >= 'A' && c <= 'F') return tolower(c);
193   ¦ // disallow any non-hex characters, including a '0x' prefix
194   ¦ if (!isspace(c)) {
195   ¦ ¦ raise << "invalid non-hex character " << NUM(c) << "\n" << end();
196   ¦ ¦ break;
197   ¦ }
198   }
199   return '\0';
200 }
201 
202 uint8_t to_byte(char hex_byte1, char hex_byte2) {
203   return to_hex_num(hex_byte1)*16 + to_hex_num(hex_byte2);
204 }
205 uint8_t to_hex_num(char c) {
206   if (c >= '0' && c <= '9') return c - '0';
207   if (c >= 'a' && c <= 'f') return c - 'a' + 10;
208   assert(false);
209   return 0;
210 }
211 
212 inline uint8_t next() {
213   return Mem.at(EIP++);
214 }
215 
216 // read a 32-bit immediate in little-endian order from the instruction stream
217 int32_t imm32() {
218   int32_t result = next();
219   result |= (next()<<8);
220   result |= (next()<<16);
221   result |= (next()<<24);
222   return result;
223 }
224 
225 :(before "End Includes")
226 #include <iomanip>
227 #define HEXBYTE  std::hex << std::setw(2) << std::setfill('0')
228 #define HEXWORD  std::hex << std::setw(8) << std::setfill('0')
229 // ugly that iostream doesn't print uint8_t as an integer
230 #define NUM(X) static_cast<int>(X)
231 #include <stdint.h>