| ... | @@ -1,4 +1,8 @@ | ... | @@ -1,4 +1,8 @@ |
| 1 | const std = @import("std"); | 1 | const std = @import("std"); |
| | 2 | const testing = std.testing; |
| | 3 | const mem = std.mem; |
| | 4 | const assert = std.debug.assert; |
| | 5 | const ArrayList = std.ArrayList; |
| 2 | const Type = @import("../Type.zig"); | 6 | const Type = @import("../Type.zig"); |
| 3 | const DW = std.dwarf; | 7 | const DW = std.dwarf; |
| 4 | | 8 | |
| ... | @@ -68,6 +72,11 @@ pub const Register = enum(u8) { | ... | @@ -68,6 +72,11 @@ pub const Register = enum(u8) { |
| 68 | return @truncate(u4, @enumToInt(self)); | 72 | return @truncate(u4, @enumToInt(self)); |
| 69 | } | 73 | } |
| 70 | | 74 | |
| | 75 | /// Like id, but only returns the lower 3 bits. |
| | 76 | pub fn low_id(self: Register) u3 { |
| | 77 | return @truncate(u3, @enumToInt(self)); |
| | 78 | } |
| | 79 | |
| 71 | /// Returns the index into `callee_preserved_regs`. | 80 | /// Returns the index into `callee_preserved_regs`. |
| 72 | pub fn allocIndex(self: Register) ?u4 { | 81 | pub fn allocIndex(self: Register) ?u4 { |
| 73 | return switch (self) { | 82 | return switch (self) { |
| ... | @@ -136,6 +145,418 @@ pub const callee_preserved_regs = [_]Register{ .rax, .rcx, .rdx, .rsi, .rdi, .r8 | ... | @@ -136,6 +145,418 @@ pub const callee_preserved_regs = [_]Register{ .rax, .rcx, .rdx, .rsi, .rdi, .r8 |
| 136 | pub const c_abi_int_param_regs = [_]Register{ .rdi, .rsi, .rdx, .rcx, .r8, .r9 }; | 145 | pub const c_abi_int_param_regs = [_]Register{ .rdi, .rsi, .rdx, .rcx, .r8, .r9 }; |
| 137 | pub const c_abi_int_return_regs = [_]Register{ .rax, .rdx }; | 146 | pub const c_abi_int_return_regs = [_]Register{ .rax, .rdx }; |
| 138 | | 147 | |
| | 148 | /// Represents an unencoded x86 instruction. |
| | 149 | /// |
| | 150 | /// Roughly based on the table headings at http://ref.x86asm.net/coder64.html |
| | 151 | pub const Instruction = struct { |
| | 152 | /// Opcode prefix, needed for certain rare ops (e.g. MOVSS) |
| | 153 | opcode_prefix: ?u8 = null, |
| | 154 | |
| | 155 | /// One-byte primary opcode |
| | 156 | primary_opcode_1b: ?u8 = null, |
| | 157 | /// Two-byte primary opcode (always prefixed with 0f) |
| | 158 | primary_opcode_2b: ?u8 = null, |
| | 159 | // TODO: Support 3-byte opcodes |
| | 160 | |
| | 161 | /// Secondary opcode |
| | 162 | secondary_opcode: ?u8 = null, |
| | 163 | |
| | 164 | /// Opcode extension (to be placed in the ModR/M byte in place of reg) |
| | 165 | opcode_extension: ?u3 = null, |
| | 166 | |
| | 167 | /// Legacy prefixes to use with this instruction |
| | 168 | /// Most of the time, this field will be 0 and no prefixes are added. |
| | 169 | /// Otherwise, a prefix will be added for each field set. |
| | 170 | legacy_prefixes: LegacyPrefixes = .{}, |
| | 171 | |
| | 172 | /// 64-bit operand size |
| | 173 | operand_size_64: bool = false, |
| | 174 | |
| | 175 | /// The opcode-reg field, |
| | 176 | /// stored in the 3 least significant bits of the opcode |
| | 177 | /// on certain instructions + REX if extended |
| | 178 | opcode_reg: ?Register = null, |
| | 179 | |
| | 180 | /// The reg field |
| | 181 | reg: ?Register = null, |
| | 182 | /// The mod + r/m field |
| | 183 | modrm: ?ModrmEffectiveAddress = null, |
| | 184 | /// Location of the 3rd operand, if applicable |
| | 185 | sib: ?SibEffectiveAddress = null, |
| | 186 | |
| | 187 | /// Number of bytes of immediate |
| | 188 | immediate_bytes: u8 = 0, |
| | 189 | /// The value of the immediate |
| | 190 | immediate: u64 = 0, |
| | 191 | |
| | 192 | /// See legacy_prefixes |
| | 193 | pub const LegacyPrefixes = packed struct { |
| | 194 | /// LOCK |
| | 195 | prefix_f0: bool = false, |
| | 196 | /// REPNZ, REPNE, REP, Scalar Double-precision |
| | 197 | prefix_f2: bool = false, |
| | 198 | /// REPZ, REPE, REP, Scalar Single-precision |
| | 199 | prefix_f3: bool = false, |
| | 200 | |
| | 201 | /// CS segment override or Branch not taken |
| | 202 | prefix_2e: bool = false, |
| | 203 | /// DS segment override |
| | 204 | prefix_36: bool = false, |
| | 205 | /// ES segment override |
| | 206 | prefix_26: bool = false, |
| | 207 | /// FS segment override |
| | 208 | prefix_64: bool = false, |
| | 209 | /// GS segment override |
| | 210 | prefix_65: bool = false, |
| | 211 | |
| | 212 | /// Branch taken |
| | 213 | prefix_3e: bool = false, |
| | 214 | |
| | 215 | /// Operand size override |
| | 216 | prefix_66: bool = false, |
| | 217 | |
| | 218 | /// Address size override |
| | 219 | prefix_67: bool = false, |
| | 220 | |
| | 221 | padding: u5 = 0, |
| | 222 | }; |
| | 223 | |
| | 224 | /// Encodes an effective address for the Mod + R/M part of the ModR/M byte |
| | 225 | /// |
| | 226 | /// Note that depending on the instruction, not all effective addresses are allowed. |
| | 227 | /// |
| | 228 | /// Examples: |
| | 229 | /// eax: .reg = .eax |
| | 230 | /// [eax]: .mem = .eax |
| | 231 | /// [eax + 8]: .mem_disp = .{ .reg = .eax, .disp = 8 } |
| | 232 | /// [eax - 8]: .mem_disp = .{ .reg = .eax, .disp = -8 } |
| | 233 | /// [55]: .disp32 = 55 |
| | 234 | pub const ModrmEffectiveAddress = union(enum) { |
| | 235 | reg: Register, |
| | 236 | mem: Register, |
| | 237 | mem_disp: struct { |
| | 238 | reg: Register, |
| | 239 | disp: i32, |
| | 240 | }, |
| | 241 | disp32: u32, |
| | 242 | |
| | 243 | pub fn isExtended(self: @This()) bool { |
| | 244 | return switch (self) { |
| | 245 | .reg => |reg| reg.isExtended(), |
| | 246 | .mem => |memea| memea.isExtended(), |
| | 247 | .mem_disp => |mem_disp| mem_disp.reg.isExtended(), |
| | 248 | .disp32 => false, |
| | 249 | }; |
| | 250 | } |
| | 251 | }; |
| | 252 | |
| | 253 | /// Encodes an effective address for the SIB byte |
| | 254 | /// |
| | 255 | /// Note that depending on the instruction, not all effective addresses are allowed. |
| | 256 | /// |
| | 257 | /// Examples: |
| | 258 | /// [eax + ebx * 2]: .base_index = .{ .base = .eax, .index = .ebx, .scale = 2 } |
| | 259 | /// [eax]: .base_index = .{ .base = .eax, .index = null, .scale = 1 } |
| | 260 | /// [ebx * 2 + 256]: .index_disp = .{ .index = .ebx, .scale = 2, .disp = 256 } |
| | 261 | /// [[ebp] + ebx * 2 + 8]: .ebp_index_disp = .{ .index = .ebx, .scale = 2, .disp = 8 } |
| | 262 | pub const SibEffectiveAddress = union(enum) { |
| | 263 | base_index: struct { |
| | 264 | base: Register, |
| | 265 | index: ?Register, |
| | 266 | scale: u8, // 1, 2, 4, or 8 |
| | 267 | }, |
| | 268 | index_disp: struct { |
| | 269 | index: ?Register, |
| | 270 | scale: u8, // 1, 2, 4, or 8 |
| | 271 | disp: u32, |
| | 272 | }, |
| | 273 | ebp_index_disp: struct { |
| | 274 | index: ?Register, |
| | 275 | scale: u8, // 1, 2, 4, or 8 |
| | 276 | disp: u32, |
| | 277 | }, |
| | 278 | |
| | 279 | pub fn baseIsExtended(self: @This()) bool { |
| | 280 | return switch (self) { |
| | 281 | .base_index => |base_index| base_index.base.isExtended(), |
| | 282 | .index_disp, .ebp_index_disp => false, |
| | 283 | }; |
| | 284 | } |
| | 285 | |
| | 286 | pub fn indexIsExtended(self: @This()) bool { |
| | 287 | return switch (self) { |
| | 288 | .base_index => |base_index| if (base_index.index) |idx| idx.isExtended() else false, |
| | 289 | .index_disp => |index_disp| if (index_disp.index) |idx| idx.isExtended() else false, |
| | 290 | .ebp_index_disp => |ebp_index_disp| if (ebp_index_disp.index) |idx| idx.isExtended() else false, |
| | 291 | }; |
| | 292 | } |
| | 293 | }; |
| | 294 | |
| | 295 | /// Writes the encoded Instruction to the code ArrayList |
| | 296 | pub fn encodeInto(inst: Instruction, code: *ArrayList(u8)) !void { |
| | 297 | // We need to write the following, in that order: |
| | 298 | // - Legacy prefixes (0 to 13 bytes) |
| | 299 | // - REX prefix (0 to 1 byte) |
| | 300 | // - Opcode (1, 2, or 3 bytes) |
| | 301 | // - ModR/M (0 or 1 byte) |
| | 302 | // - SIB (0 or 1 byte) |
| | 303 | // - Displacement (0, 1, 2, or 4 bytes) |
| | 304 | // - Immediate (0, 1, 2, 4, or 8 bytes) |
| | 305 | |
| | 306 | // By this calculation, an instruction could be up to 31 bytes long (will probably not happen) |
| | 307 | try code.ensureCapacity(code.items.len + 31); |
| | 308 | |
| | 309 | // Legacy prefixes |
| | 310 | if (@bitCast(u16, inst.legacy_prefixes) != 0) { |
| | 311 | // Hopefully this path isn't taken very often, so we'll do it the slow way for now |
| | 312 | |
| | 313 | // LOCK |
| | 314 | if (inst.legacy_prefixes.prefix_f0) code.appendAssumeCapacity(0xf0); |
| | 315 | // REPNZ, REPNE, REP, Scalar Double-precision |
| | 316 | if (inst.legacy_prefixes.prefix_f2) code.appendAssumeCapacity(0xf2); |
| | 317 | // REPZ, REPE, REP, Scalar Single-precision |
| | 318 | if (inst.legacy_prefixes.prefix_f3) code.appendAssumeCapacity(0xf3); |
| | 319 | |
| | 320 | // CS segment override or Branch not taken |
| | 321 | if (inst.legacy_prefixes.prefix_2e) code.appendAssumeCapacity(0x2e); |
| | 322 | // DS segment override |
| | 323 | if (inst.legacy_prefixes.prefix_36) code.appendAssumeCapacity(0x36); |
| | 324 | // ES segment override |
| | 325 | if (inst.legacy_prefixes.prefix_26) code.appendAssumeCapacity(0x26); |
| | 326 | // FS segment override |
| | 327 | if (inst.legacy_prefixes.prefix_64) code.appendAssumeCapacity(0x64); |
| | 328 | // GS segment override |
| | 329 | if (inst.legacy_prefixes.prefix_65) code.appendAssumeCapacity(0x65); |
| | 330 | |
| | 331 | // Branch taken |
| | 332 | if (inst.legacy_prefixes.prefix_3e) code.appendAssumeCapacity(0x3e); |
| | 333 | |
| | 334 | // Operand size override |
| | 335 | if (inst.legacy_prefixes.prefix_66) code.appendAssumeCapacity(0x66); |
| | 336 | |
| | 337 | // Address size override |
| | 338 | if (inst.legacy_prefixes.prefix_67) code.appendAssumeCapacity(0x67); |
| | 339 | } |
| | 340 | |
| | 341 | // REX prefix |
| | 342 | // |
| | 343 | // A REX prefix has the following form: |
| | 344 | // 0b0100_WRXB |
| | 345 | // 0100: fixed bits |
| | 346 | // W: stands for "wide", indicates that the instruction uses 64-bit operands. |
| | 347 | // R, X, and B each contain the 4th bit of a register |
| | 348 | // these have to be set when using registers 8-15. |
| | 349 | // R: stands for "reg", extends the reg field in the ModR/M byte. |
| | 350 | // X: stands for "index", extends the index field in the SIB byte. |
| | 351 | // B: stands for "base", extends either the r/m field in the ModR/M byte, |
| | 352 | // the base field in the SIB byte, |
| | 353 | // or the opcode reg field in the Opcode byte. |
| | 354 | { |
| | 355 | var value: u8 = 0x40; |
| | 356 | if (inst.opcode_reg) |opcode_reg| { |
| | 357 | if (opcode_reg.isExtended()) { |
| | 358 | value |= 0x1; |
| | 359 | } |
| | 360 | } |
| | 361 | if (inst.modrm) |modrm| { |
| | 362 | if (modrm.isExtended()) { |
| | 363 | value |= 0x1; |
| | 364 | } |
| | 365 | } |
| | 366 | if (inst.sib) |sib| { |
| | 367 | if (sib.baseIsExtended()) { |
| | 368 | value |= 0x1; |
| | 369 | } |
| | 370 | if (sib.indexIsExtended()) { |
| | 371 | value |= 0x2; |
| | 372 | } |
| | 373 | } |
| | 374 | if (inst.reg) |reg| { |
| | 375 | if (reg.isExtended()) { |
| | 376 | value |= 0x4; |
| | 377 | } |
| | 378 | } |
| | 379 | if (inst.operand_size_64) { |
| | 380 | value |= 0x8; |
| | 381 | } |
| | 382 | if (value != 0x40) { |
| | 383 | code.appendAssumeCapacity(value); |
| | 384 | } |
| | 385 | } |
| | 386 | |
| | 387 | // Opcode |
| | 388 | if (inst.primary_opcode_1b) |opcode| { |
| | 389 | var value = opcode; |
| | 390 | if (inst.opcode_reg) |opcode_reg| { |
| | 391 | value |= opcode_reg.low_id(); |
| | 392 | } |
| | 393 | code.appendAssumeCapacity(value); |
| | 394 | } else if (inst.primary_opcode_2b) |opcode| { |
| | 395 | code.appendAssumeCapacity(0x0f); |
| | 396 | var value = opcode; |
| | 397 | if (inst.opcode_reg) |opcode_reg| { |
| | 398 | value |= opcode_reg.low_id(); |
| | 399 | } |
| | 400 | code.appendAssumeCapacity(value); |
| | 401 | } |
| | 402 | |
| | 403 | var disp8: ?u8 = null; |
| | 404 | var disp16: ?u16 = null; |
| | 405 | var disp32: ?u32 = null; |
| | 406 | |
| | 407 | // ModR/M |
| | 408 | // |
| | 409 | // Example ModR/M byte: |
| | 410 | // c7: ModR/M byte that contains: |
| | 411 | // 11 000 111: |
| | 412 | // ^ ^ ^ |
| | 413 | // mod | | |
| | 414 | // reg | |
| | 415 | // r/m |
| | 416 | // where mod = 11 indicates that both operands are registers, |
| | 417 | // reg = 000 indicates that the first operand is register EAX |
| | 418 | // r/m = 111 indicates that the second operand is register EDI (since mod = 11) |
| | 419 | if (inst.modrm != null or inst.reg != null or inst.opcode_extension != null) { |
| | 420 | var value: u8 = 0; |
| | 421 | |
| | 422 | // mod + rm |
| | 423 | if (inst.modrm) |modrm| { |
| | 424 | switch (modrm) { |
| | 425 | .reg => |reg| { |
| | 426 | value |= reg.low_id(); |
| | 427 | value |= 0b11_000_000; |
| | 428 | }, |
| | 429 | .mem => |memea| { |
| | 430 | assert(memea.low_id() != 4 and memea.low_id() != 5); |
| | 431 | value |= memea.low_id(); |
| | 432 | // value |= 0b00_000_000; |
| | 433 | }, |
| | 434 | .mem_disp => |mem_disp| { |
| | 435 | assert(mem_disp.reg.low_id() != 4); |
| | 436 | value |= mem_disp.reg.low_id(); |
| | 437 | if (mem_disp.disp < 128) { |
| | 438 | // Use 1 byte of displacement |
| | 439 | value |= 0b01_000_000; |
| | 440 | disp8 = @bitCast(u8, @intCast(i8, mem_disp.disp)); |
| | 441 | } else { |
| | 442 | // Use all 4 bytes of displacement |
| | 443 | value |= 0b10_000_000; |
| | 444 | disp32 = @bitCast(u32, mem_disp.disp); |
| | 445 | } |
| | 446 | }, |
| | 447 | .disp32 => |d| { |
| | 448 | value |= 0b00_000_101; |
| | 449 | disp32 = d; |
| | 450 | }, |
| | 451 | } |
| | 452 | } |
| | 453 | |
| | 454 | // reg |
| | 455 | if (inst.reg) |reg| { |
| | 456 | value |= @as(u8, reg.low_id()) << 3; |
| | 457 | } else if (inst.opcode_extension) |ext| { |
| | 458 | value |= @as(u8, ext) << 3; |
| | 459 | } |
| | 460 | |
| | 461 | code.appendAssumeCapacity(value); |
| | 462 | } |
| | 463 | |
| | 464 | // SIB |
| | 465 | { |
| | 466 | if (inst.sib) |sib| { |
| | 467 | return error.TODOSIBByteForX8664; |
| | 468 | } |
| | 469 | } |
| | 470 | |
| | 471 | // Displacement |
| | 472 | // |
| | 473 | // The size of the displacement depends on the instruction used and is very fragile. |
| | 474 | // The bytes are simply written in LE order. |
| | 475 | { |
| | 476 | |
| | 477 | // These writes won't fail because we ensured capacity earlier. |
| | 478 | if (disp8) |d| |
| | 479 | code.appendAssumeCapacity(d) |
| | 480 | else if (disp16) |d| |
| | 481 | mem.writeIntLittle(u16, code.addManyAsArrayAssumeCapacity(2), d) |
| | 482 | else if (disp32) |d| |
| | 483 | mem.writeIntLittle(u32, code.addManyAsArrayAssumeCapacity(4), d); |
| | 484 | } |
| | 485 | |
| | 486 | // Immediate |
| | 487 | // |
| | 488 | // The size of the immediate depends on the instruction used and is very fragile. |
| | 489 | // The bytes are simply written in LE order. |
| | 490 | { |
| | 491 | // These writes won't fail because we ensured capacity earlier. |
| | 492 | if (inst.immediate_bytes == 1) |
| | 493 | code.appendAssumeCapacity(@intCast(u8, inst.immediate)) |
| | 494 | else if (inst.immediate_bytes == 2) |
| | 495 | mem.writeIntLittle(u16, code.addManyAsArrayAssumeCapacity(2), @intCast(u16, inst.immediate)) |
| | 496 | else if (inst.immediate_bytes == 4) |
| | 497 | mem.writeIntLittle(u32, code.addManyAsArrayAssumeCapacity(4), @intCast(u32, inst.immediate)) |
| | 498 | else if (inst.immediate_bytes == 8) |
| | 499 | mem.writeIntLittle(u64, code.addManyAsArrayAssumeCapacity(8), inst.immediate); |
| | 500 | } |
| | 501 | } |
| | 502 | }; |
| | 503 | |
| | 504 | fn expectEncoded(inst: Instruction, expected: []const u8) !void { |
| | 505 | var code = ArrayList(u8).init(testing.allocator); |
| | 506 | defer code.deinit(); |
| | 507 | try inst.encodeInto(&code); |
| | 508 | testing.expectEqualSlices(u8, expected, code.items); |
| | 509 | } |
| | 510 | |
| | 511 | test "x86_64 Instruction.encodeInto" { |
| | 512 | // simple integer multiplication |
| | 513 | |
| | 514 | // imul eax,edi |
| | 515 | // 0faf c7 |
| | 516 | try expectEncoded(Instruction{ |
| | 517 | .primary_opcode_2b = 0xaf, // imul |
| | 518 | .reg = .eax, // destination |
| | 519 | .modrm = .{ .reg = .edi }, // source |
| | 520 | }, &[_]u8{ 0x0f, 0xaf, 0xc7 }); |
| | 521 | |
| | 522 | // simple mov |
| | 523 | |
| | 524 | // mov eax,edi |
| | 525 | // 89 f8 |
| | 526 | try expectEncoded(Instruction{ |
| | 527 | .primary_opcode_1b = 0x89, // mov (with rm as destination) |
| | 528 | .reg = .edi, // source |
| | 529 | .modrm = .{ .reg = .eax }, // destination |
| | 530 | }, &[_]u8{ 0x89, 0xf8 }); |
| | 531 | |
| | 532 | // signed integer addition of 32-bit sign extended immediate to 64 bit register |
| | 533 | |
| | 534 | // add rcx, 2147483647 |
| | 535 | // |
| | 536 | // Using the following opcode: REX.W + 81 /0 id, we expect the following encoding |
| | 537 | // |
| | 538 | // 48 : REX.W set for 64 bit operand (*r*cx) |
| | 539 | // 81 : opcode for "<arithmetic> with immediate" |
| | 540 | // c1 : id = rcx, |
| | 541 | // : c1 = 11 <-- mod = 11 indicates r/m is register (rcx) |
| | 542 | // : 000 <-- opcode_extension = 0 because opcode extension is /0. /0 specifies ADD |
| | 543 | // : 001 <-- 001 is rcx |
| | 544 | // ffffff7f : 2147483647 |
| | 545 | try expectEncoded(Instruction{ |
| | 546 | // REX.W + |
| | 547 | .operand_size_64 = true, |
| | 548 | // 81 |
| | 549 | .primary_opcode_1b = 0x81, |
| | 550 | // /0 |
| | 551 | .opcode_extension = 0, |
| | 552 | // rcx |
| | 553 | .modrm = .{ .reg = .rcx }, |
| | 554 | // immediate |
| | 555 | .immediate_bytes = 4, |
| | 556 | .immediate = 2147483647, |
| | 557 | }, &[_]u8{ 0x48, 0x81, 0xc1, 0xff, 0xff, 0xff, 0x7f }); |
| | 558 | } |
| | 559 | |
| 139 | // TODO add these registers to the enum and populate dwarfLocOp | 560 | // TODO add these registers to the enum and populate dwarfLocOp |
| 140 | // // Return Address register. This is stored in `0(%rsp, "")` and is not a physical register. | 561 | // // Return Address register. This is stored in `0(%rsp, "")` and is not a physical register. |
| 141 | // RA = (16, "RA"), | 562 | // RA = (16, "RA"), |
gracefu <81774659+gracefuu@users.noreply.github.com> 2021-04-08 17:05:48+08:00