| 1 | // Licensed to the .NET Foundation under one or more agreements. |
| 2 | // The .NET Foundation licenses this file to you under the MIT license. |
| 3 | // See the LICENSE file in the project root for more information. |
| 4 | /*********************************************************************** |
| 5 | * |
| 6 | * File: dis.cpp |
| 7 | * |
| 8 | |
| 9 | * |
| 10 | * File Comments: |
| 11 | * |
| 12 | * This file handles disassembly. It is adapted from the MS linker. |
| 13 | * |
| 14 | ***********************************************************************/ |
| 15 | |
| 16 | #include "jitpch.h" |
| 17 | #ifdef _MSC_VER |
| 18 | #pragma hdrstop |
| 19 | #endif |
| 20 | |
| 21 | /*****************************************************************************/ |
| 22 | #ifdef LATE_DISASM |
| 23 | /*****************************************************************************/ |
| 24 | |
| 25 | // Define DISASM_DEBUG to get verbose output of late disassembler inner workings. |
| 26 | //#define DISASM_DEBUG |
| 27 | #ifdef DISASM_DEBUG |
| 28 | #ifdef DEBUG |
| 29 | #define DISASM_DUMP(...) \ |
| 30 | if (VERBOSE) \ |
| 31 | printf(__VA_ARGS__) |
| 32 | #else // !DEBUG |
| 33 | #define DISASM_DUMP(...) printf(__VA_ARGS__) |
| 34 | #endif // !DEBUG |
| 35 | #else // !DISASM_DEBUG |
| 36 | #define DISASM_DUMP(...) |
| 37 | #endif // !DISASM_DEBUG |
| 38 | |
| 39 | /*****************************************************************************/ |
| 40 | |
| 41 | #define MAX_CLASSNAME_LENGTH 1024 |
| 42 | |
| 43 | #if defined(_AMD64_) |
| 44 | |
| 45 | #pragma comment(linker, \ |
| 46 | "/ALTERNATENAME:__imp_?CchFormatAddr@DIS@@QEBA_K_KPEAG0@Z=__imp_?CchFormatAddr@DIS@@QEBA_K_KPEA_W0@Z") |
| 47 | #pragma comment(linker, \ |
| 48 | "/ALTERNATENAME:__imp_?CchFormatInstr@DIS@@QEBA_KPEAG_K@Z=__imp_?CchFormatInstr@DIS@@QEBA_KPEA_W_K@Z") |
| 49 | #pragma comment( \ |
| 50 | linker, \ |
| 51 | "/ALTERNATENAME:__imp_?PfncchaddrSet@DIS@@QEAAP6A_KPEBV1@_KPEAG1PEA_K@ZP6A_K01213@Z@Z=__imp_?PfncchaddrSet@DIS@@QEAAP6A_KPEBV1@_KPEA_W1PEA_K@ZP6A_K01213@Z@Z") |
| 52 | #pragma comment( \ |
| 53 | linker, \ |
| 54 | "/ALTERNATENAME:__imp_?PfncchregSet@DIS@@QEAAP6A_KPEBV1@W4REGA@1@PEAG_K@ZP6A_K0123@Z@Z=__imp_?PfncchregSet@DIS@@QEAAP6A_KPEBV1@W4REGA@1@PEA_W_K@ZP6A_K0123@Z@Z") |
| 55 | #pragma comment( \ |
| 56 | linker, \ |
| 57 | "/ALTERNATENAME:__imp_?PfncchregrelSet@DIS@@QEAAP6A_KPEBV1@W4REGA@1@KPEAG_KPEAK@ZP6A_K01K234@Z@Z=__imp_?PfncchregrelSet@DIS@@QEAAP6A_KPEBV1@W4REGA@1@KPEA_W_KPEAK@ZP6A_K01K234@Z@Z") |
| 58 | #pragma comment( \ |
| 59 | linker, \ |
| 60 | "/ALTERNATENAME:__imp_?PfncchfixupSet@DIS@@QEAAP6A_KPEBV1@_K1PEAG1PEA_K@ZP6A_K011213@Z@Z=__imp_?PfncchfixupSet@DIS@@QEAAP6A_KPEBV1@_K1PEA_W1PEA_K@ZP6A_K011213@Z@Z") |
| 61 | |
| 62 | #elif defined(_X86_) |
| 63 | |
| 64 | #pragma comment(linker, "/ALTERNATENAME:__imp_?CchFormatAddr@DIS@@QBEI_KPAGI@Z=__imp_?CchFormatAddr@DIS@@QBEI_KPA_WI@Z") |
| 65 | #pragma comment(linker, "/ALTERNATENAME:__imp_?CchFormatInstr@DIS@@QBEIPAGI@Z=__imp_?CchFormatInstr@DIS@@QBEIPA_WI@Z") |
| 66 | #pragma comment( \ |
| 67 | linker, \ |
| 68 | "/ALTERNATENAME:__imp_?PfncchaddrSet@DIS@@QAEP6GIPBV1@_KPAGIPA_K@ZP6GI012I3@Z@Z=__imp_?PfncchaddrSet@DIS@@QAEP6GIPBV1@_KPA_WIPA_K@ZP6GI012I3@Z@Z") |
| 69 | #pragma comment( \ |
| 70 | linker, \ |
| 71 | "/ALTERNATENAME:__imp_?PfncchregSet@DIS@@QAEP6GIPBV1@W4REGA@1@PAGI@ZP6GI012I@Z@Z=__imp_?PfncchregSet@DIS@@QAEP6GIPBV1@W4REGA@1@PA_WI@ZP6GI012I@Z@Z") |
| 72 | #pragma comment( \ |
| 73 | linker, \ |
| 74 | "/ALTERNATENAME:__imp_?PfncchregrelSet@DIS@@QAEP6GIPBV1@W4REGA@1@KPAGIPAK@ZP6GI01K2I3@Z@Z=__imp_?PfncchregrelSet@DIS@@QAEP6GIPBV1@W4REGA@1@KPA_WIPAK@ZP6GI01K2I3@Z@Z") |
| 75 | #pragma comment( \ |
| 76 | linker, \ |
| 77 | "/ALTERNATENAME:__imp_?PfncchfixupSet@DIS@@QAEP6GIPBV1@_KIPAGIPA_K@ZP6GI01I2I3@Z@Z=__imp_?PfncchfixupSet@DIS@@QAEP6GIPBV1@_KIPA_WIPA_K@ZP6GI01I2I3@Z@Z") |
| 78 | |
| 79 | #endif |
| 80 | |
| 81 | /***************************************************************************** |
| 82 | * Given an absolute address from the beginning of the code |
| 83 | * find the corresponding emitter block and the relative offset |
| 84 | * of the current address in that block |
| 85 | * Was used to get to the fixup list of each block. The new emitter has |
| 86 | * no such fixups. Something needs to be added for this. |
| 87 | */ |
| 88 | |
| 89 | // These structs were defined in emit.h. Fake them here so DisAsm.cpp can compile |
| 90 | |
| 91 | typedef struct codeFix |
| 92 | { |
| 93 | codeFix* cfNext; |
| 94 | unsigned cfFixup; |
| 95 | } * codeFixPtr; |
| 96 | |
| 97 | typedef struct codeBlk |
| 98 | { |
| 99 | codeFix* cbFixupLst; |
| 100 | } * codeBlkPtr; |
| 101 | |
| 102 | /***************************************************************************** |
| 103 | * The following is the callback for jump label and direct function calls fixups. |
| 104 | * "addr" represents the address of jump that has to be |
| 105 | * replaced with a label or function name. |
| 106 | * |
| 107 | * Return 1 if a name was written representing the address, 0 otherwise. |
| 108 | */ |
| 109 | |
| 110 | /* static */ |
| 111 | size_t __stdcall DisAssembler::disCchAddr( |
| 112 | const DIS* pdis, DIS::ADDR addr, __in_ecount(cchMax) wchar_t* wz, size_t cchMax, DWORDLONG* pdwDisp) |
| 113 | { |
| 114 | DisAssembler* pDisAsm = (DisAssembler*)pdis->PvClient(); |
| 115 | assert(pDisAsm); |
| 116 | return pDisAsm->disCchAddrMember(pdis, addr, wz, cchMax, pdwDisp); |
| 117 | } |
| 118 | |
| 119 | size_t DisAssembler::disCchAddrMember( |
| 120 | const DIS* pdis, DIS::ADDR addr, __in_ecount(cchMax) wchar_t* wz, size_t cchMax, DWORDLONG* pdwDisp) |
| 121 | { |
| 122 | /* First check the termination type of the instruction |
| 123 | * because this might be a helper or static function call |
| 124 | * check to see if we have a fixup for the current address */ |
| 125 | |
| 126 | size_t retval = 0; // assume we don't know |
| 127 | |
| 128 | #if defined(_TARGET_XARCH_) |
| 129 | |
| 130 | DISX86::TRMTA terminationType = DISX86::TRMTA(pdis->Trmta()); |
| 131 | |
| 132 | DISASM_DUMP("AddrMember %p (%p), termType %u\n" , addr, disGetLinearAddr((size_t)addr), terminationType); |
| 133 | |
| 134 | switch (terminationType) |
| 135 | { |
| 136 | // int disCallSize; |
| 137 | |
| 138 | case DISX86::trmtaJmpShort: |
| 139 | case DISX86::trmtaJmpCcShort: |
| 140 | |
| 141 | /* We have a short jump in the current code block - generate the label to which we jump */ |
| 142 | |
| 143 | assert(0 <= disTarget && disTarget < disTotalCodeSize); |
| 144 | swprintf_s(wz, cchMax, W("short L_%02u" ), disLabels[disTarget]); |
| 145 | retval = 1; |
| 146 | break; |
| 147 | |
| 148 | case DISX86::trmtaJmpNear: |
| 149 | case DISX86::trmtaJmpCcNear: |
| 150 | |
| 151 | /* We have a near jump. Check if is in the current code block. |
| 152 | * Otherwise we have no target for it. */ |
| 153 | |
| 154 | if (0 <= disTarget && disTarget < disTotalCodeSize) |
| 155 | { |
| 156 | swprintf_s(wz, cchMax, W("L_%02u" ), disLabels[disTarget]); |
| 157 | retval = 1; |
| 158 | } |
| 159 | break; |
| 160 | |
| 161 | case DISX86::trmtaCallNear16: |
| 162 | case DISX86::trmtaCallNear32: |
| 163 | |
| 164 | /* check for local calls (i.e. CALL label) */ |
| 165 | |
| 166 | if (0 <= disTarget && disTarget < disTotalCodeSize) |
| 167 | { |
| 168 | /* not a "call ds:[0000]" - go ahead */ |
| 169 | /* disTarget within block boundary -> local call */ |
| 170 | |
| 171 | swprintf_s(wz, cchMax, W("short L_%02u" ), disLabels[disTarget]); |
| 172 | retval = 1; |
| 173 | break; |
| 174 | } |
| 175 | |
| 176 | /* this is a near call - in our case usually VM helper functions */ |
| 177 | |
| 178 | /* find the emitter block and the offset of the call fixup */ |
| 179 | /* for the fixup offset we have to add the opcode size for the call - in the case of a near call is 1 */ |
| 180 | |
| 181 | // disCallSize = 1; |
| 182 | |
| 183 | { |
| 184 | size_t absoluteTarget = (size_t)disGetLinearAddr(disTarget); |
| 185 | const char* name = disGetMethodFullName(absoluteTarget); |
| 186 | if (name != nullptr) |
| 187 | { |
| 188 | swprintf_s(wz, cchMax, W("%zx %S" ), dspAddr(absoluteTarget), name); |
| 189 | retval = 1; |
| 190 | break; |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | break; |
| 195 | |
| 196 | #ifdef _TARGET_AMD64_ |
| 197 | |
| 198 | case DISX86::trmtaFallThrough: |
| 199 | |
| 200 | /* memory indirect case. Could be for an LEA for the base address of a switch table, which is an arbitrary |
| 201 | * address, currently of the first block after the prolog. */ |
| 202 | |
| 203 | /* find the emitter block and the offset for the fixup |
| 204 | * "addr" is the address of the immediate */ |
| 205 | |
| 206 | break; |
| 207 | |
| 208 | #endif // _TARGET_AMD64_ |
| 209 | |
| 210 | default: |
| 211 | |
| 212 | printf("Termination type is %d\n" , (int)terminationType); |
| 213 | assert(!"treat this case\n" ); |
| 214 | break; |
| 215 | } |
| 216 | |
| 217 | #elif defined(_TARGET_ARM64_) |
| 218 | |
| 219 | DISARM64::TRMTA terminationType = DISARM64::TRMTA(pdis->Trmta()); |
| 220 | |
| 221 | DISASM_DUMP("AddrMember %p (%p), termType %u\n" , addr, disGetLinearAddr((size_t)addr), terminationType); |
| 222 | |
| 223 | switch (terminationType) |
| 224 | { |
| 225 | // int disCallSize; |
| 226 | |
| 227 | case DISARM64::TRMTA::trmtaBra: |
| 228 | case DISARM64::TRMTA::trmtaBraCase: |
| 229 | case DISARM64::TRMTA::trmtaBraCc: |
| 230 | case DISARM64::TRMTA::trmtaBraCcCase: |
| 231 | case DISARM64::TRMTA::trmtaBraCcInd: |
| 232 | case DISARM64::TRMTA::trmtaBraInd: |
| 233 | |
| 234 | /* We have a jump. Check if is in the current code block. |
| 235 | * Otherwise we have no target for it. */ |
| 236 | |
| 237 | if (0 <= disTarget && disTarget < disTotalCodeSize) |
| 238 | { |
| 239 | swprintf_s(wz, cchMax, W("L_%02u" ), disLabels[disTarget]); |
| 240 | retval = 1; |
| 241 | } |
| 242 | break; |
| 243 | |
| 244 | case DISARM64::trmtaCall: |
| 245 | case DISARM64::trmtaCallCc: |
| 246 | case DISARM64::trmtaCallCcInd: |
| 247 | case DISARM64::trmtaCallInd: |
| 248 | |
| 249 | /* check for local calls (i.e. CALL label) */ |
| 250 | |
| 251 | if (0 <= disTarget && disTarget < disTotalCodeSize) |
| 252 | { |
| 253 | /* not a "call [0000]" - go ahead */ |
| 254 | /* disTarget within block boundary -> local call */ |
| 255 | |
| 256 | swprintf_s(wz, cchMax, W("L_%02u" ), disLabels[disTarget]); |
| 257 | retval = 1; |
| 258 | break; |
| 259 | } |
| 260 | |
| 261 | /* this is a near call - in our case usually VM helper functions */ |
| 262 | |
| 263 | /* find the emitter block and the offset of the call fixup */ |
| 264 | /* for the fixup offset we have to add the opcode size for the call - in the case of a near call is 1 */ |
| 265 | |
| 266 | // disCallSize = 1; |
| 267 | |
| 268 | { |
| 269 | size_t absoluteTarget = (size_t)disGetLinearAddr(disTarget); |
| 270 | const char* name = disGetMethodFullName(absoluteTarget); |
| 271 | if (name != nullptr) |
| 272 | { |
| 273 | swprintf_s(wz, cchMax, W("%zx %S" ), dspAddr(absoluteTarget), name); |
| 274 | retval = 1; |
| 275 | break; |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | break; |
| 280 | |
| 281 | case DISARM64::trmtaFallThrough: |
| 282 | |
| 283 | /* memory indirect case. Could be for an LEA for the base address of a switch table, which is an arbitrary |
| 284 | * address, currently of the first block after the prolog. */ |
| 285 | |
| 286 | /* find the emitter block and the offset for the fixup |
| 287 | * "addr" is the address of the immediate */ |
| 288 | |
| 289 | { |
| 290 | DIS::INSTRUCTION instr; |
| 291 | DIS::OPERAND ops[DISARM64::coperandMax]; |
| 292 | bool ok = pdis->FDecode(&instr, ops, ArrLen(ops)); |
| 293 | if (ok) |
| 294 | { |
| 295 | bool isAddress = false; |
| 296 | switch ((DISARM64::OPA)instr.opa) |
| 297 | { |
| 298 | case DISARM64::opaAdr: |
| 299 | case DISARM64::opaAdrp: |
| 300 | isAddress = true; |
| 301 | break; |
| 302 | default: |
| 303 | break; |
| 304 | } |
| 305 | |
| 306 | if (isAddress && 0 <= addr && addr < disTotalCodeSize) |
| 307 | { |
| 308 | swprintf_s(wz, cchMax, W("L_%02u" ), disLabels[addr]); |
| 309 | retval = 1; |
| 310 | } |
| 311 | } |
| 312 | } |
| 313 | break; |
| 314 | |
| 315 | default: |
| 316 | |
| 317 | printf("Termination type is %d\n" , (int)terminationType); |
| 318 | assert(!"treat this case\n" ); |
| 319 | break; |
| 320 | } |
| 321 | |
| 322 | #else // _TARGET_* |
| 323 | #error Unsupported or unset target architecture |
| 324 | #endif // _TARGET_* |
| 325 | |
| 326 | if (retval == 0) |
| 327 | { |
| 328 | if (disDiffable) |
| 329 | { |
| 330 | swprintf_s(wz, cchMax, W("%p" ), dspAddr((void*)1)); |
| 331 | } |
| 332 | } |
| 333 | else |
| 334 | { |
| 335 | /* no displacement */ |
| 336 | |
| 337 | *pdwDisp = 0x0; |
| 338 | } |
| 339 | |
| 340 | return retval; |
| 341 | } |
| 342 | |
| 343 | /***************************************************************************** |
| 344 | * We annotate some instructions to get info needed to display the symbols |
| 345 | * for that instruction. |
| 346 | * |
| 347 | * Return 1 if a name was written representing the address, 0 otherwise. |
| 348 | */ |
| 349 | |
| 350 | /* static */ |
| 351 | size_t __stdcall DisAssembler::disCchFixup( |
| 352 | const DIS* pdis, DIS::ADDR addr, size_t size, __in_ecount(cchMax) wchar_t* wz, size_t cchMax, DWORDLONG* pdwDisp) |
| 353 | { |
| 354 | DisAssembler* pDisAsm = (DisAssembler*)pdis->PvClient(); |
| 355 | assert(pDisAsm); |
| 356 | |
| 357 | return pDisAsm->disCchFixupMember(pdis, addr, size, wz, cchMax, pdwDisp); |
| 358 | } |
| 359 | |
| 360 | size_t DisAssembler::disCchFixupMember( |
| 361 | const DIS* pdis, DIS::ADDR addr, size_t size, __in_ecount(cchMax) wchar_t* wz, size_t cchMax, DWORDLONG* pdwDisp) |
| 362 | { |
| 363 | #if defined(_TARGET_XARCH_) |
| 364 | |
| 365 | DISX86::TRMTA terminationType = DISX86::TRMTA(pdis->Trmta()); |
| 366 | // DIS::ADDR disIndAddr; |
| 367 | |
| 368 | DISASM_DUMP("FixupMember %016I64X (%08IX), size %d, termType %u\n" , addr, disGetLinearAddr((size_t)addr), size, |
| 369 | terminationType); |
| 370 | |
| 371 | // Is there a relocation registered for the address? |
| 372 | |
| 373 | size_t absoluteAddr = (size_t)disGetLinearAddr((size_t)addr); |
| 374 | size_t targetAddr; |
| 375 | bool anyReloc = GetRelocationMap()->Lookup(absoluteAddr, &targetAddr); |
| 376 | |
| 377 | switch (terminationType) |
| 378 | { |
| 379 | DIS::ADDR disCallSize; |
| 380 | |
| 381 | case DISX86::trmtaFallThrough: |
| 382 | |
| 383 | /* memory indirect case */ |
| 384 | |
| 385 | assert(addr > pdis->Addr()); |
| 386 | |
| 387 | /* find the emitter block and the offset for the fixup |
| 388 | * "addr" is the address of the immediate */ |
| 389 | |
| 390 | if (anyReloc) |
| 391 | { |
| 392 | // Make instructions like "mov rcx, 7FE8247A638h" diffable. |
| 393 | swprintf_s(wz, cchMax, W("%IXh" ), dspAddr(targetAddr)); |
| 394 | break; |
| 395 | } |
| 396 | |
| 397 | return 0; |
| 398 | |
| 399 | case DISX86::trmtaJmpInd: |
| 400 | |
| 401 | /* pretty rare case - something like "jmp [eax*4]" |
| 402 | * not a function call or anything worth annotating */ |
| 403 | |
| 404 | return 0; |
| 405 | |
| 406 | case DISX86::trmtaTrap: |
| 407 | case DISX86::trmtaTrapCc: |
| 408 | |
| 409 | /* some instructions like division have a TRAP termination type - ignore it */ |
| 410 | |
| 411 | return 0; |
| 412 | |
| 413 | case DISX86::trmtaJmpShort: |
| 414 | case DISX86::trmtaJmpCcShort: |
| 415 | |
| 416 | case DISX86::trmtaJmpNear: |
| 417 | case DISX86::trmtaJmpCcNear: |
| 418 | |
| 419 | /* these are treated by the CchAddr callback - skip them */ |
| 420 | |
| 421 | return 0; |
| 422 | |
| 423 | case DISX86::trmtaCallNear16: |
| 424 | case DISX86::trmtaCallNear32: |
| 425 | |
| 426 | if (anyReloc) |
| 427 | { |
| 428 | const char* name = disGetMethodFullName(targetAddr); |
| 429 | if (name != nullptr) |
| 430 | { |
| 431 | swprintf_s(wz, cchMax, W("%zx %S" ), dspAddr(targetAddr), name); |
| 432 | break; |
| 433 | } |
| 434 | } |
| 435 | |
| 436 | /* these are treated by the CchAddr callback - skip them */ |
| 437 | |
| 438 | return 0; |
| 439 | |
| 440 | case DISX86::trmtaCallInd: |
| 441 | |
| 442 | /* here we have an indirect call - find the indirect address */ |
| 443 | |
| 444 | // BYTE * code = disGetLinearAddr((size_t)addr); |
| 445 | // disIndAddr = (DIS::ADDR) (code+0); |
| 446 | |
| 447 | /* find the size of the call opcode - less the immediate */ |
| 448 | /* for the fixup offset we have to add the opcode size for the call */ |
| 449 | /* addr is the address of the immediate, pdis->Addr() returns the address of the disassembled instruction */ |
| 450 | |
| 451 | assert(addr > pdis->Addr()); |
| 452 | disCallSize = addr - pdis->Addr(); |
| 453 | |
| 454 | /* find the emitter block and the offset of the call fixup */ |
| 455 | |
| 456 | return 0; |
| 457 | |
| 458 | default: |
| 459 | |
| 460 | printf("Termination type is %d\n" , (int)terminationType); |
| 461 | assert(!"treat this case\n" ); |
| 462 | break; |
| 463 | } |
| 464 | |
| 465 | #elif defined(_TARGET_ARM64_) |
| 466 | |
| 467 | DISARM64::TRMTA terminationType = DISARM64::TRMTA(pdis->Trmta()); |
| 468 | // DIS::ADDR disIndAddr; |
| 469 | |
| 470 | DISASM_DUMP("FixupMember %016I64X (%08IX), size %d, termType %u\n" , addr, disGetLinearAddr((size_t)addr), size, |
| 471 | terminationType); |
| 472 | |
| 473 | // Is there a relocation registered for the address? |
| 474 | |
| 475 | size_t absoluteAddr = (size_t)disGetLinearAddr((size_t)addr); |
| 476 | size_t targetAddr; |
| 477 | bool anyReloc = GetRelocationMap()->Lookup(absoluteAddr, &targetAddr); |
| 478 | |
| 479 | switch (terminationType) |
| 480 | { |
| 481 | DIS::ADDR disCallSize; |
| 482 | |
| 483 | case DISARM64::TRMTA::trmtaUnknown: |
| 484 | return 0; |
| 485 | |
| 486 | case DISARM64::TRMTA::trmtaFallThrough: |
| 487 | |
| 488 | if (anyReloc) |
| 489 | { |
| 490 | /* memory indirect case */ |
| 491 | |
| 492 | assert(addr > pdis->Addr()); |
| 493 | |
| 494 | /* find the emitter block and the offset for the fixup |
| 495 | * "addr" is the address of the immediate */ |
| 496 | |
| 497 | // Make instructions like "mov rcx, 7FE8247A638h" diffable. |
| 498 | swprintf_s(wz, cchMax, W("%IXh" ), dspAddr(targetAddr)); |
| 499 | break; |
| 500 | } |
| 501 | |
| 502 | return 0; |
| 503 | |
| 504 | case DISARM64::TRMTA::trmtaBraInd: |
| 505 | case DISARM64::TRMTA::trmtaBraCcInd: |
| 506 | |
| 507 | /* pretty rare case - something like "jmp [eax*4]" |
| 508 | * not a function call or anything worth annotating */ |
| 509 | |
| 510 | return 0; |
| 511 | |
| 512 | case DISARM64::TRMTA::trmtaTrap: |
| 513 | case DISARM64::TRMTA::trmtaTrapCc: |
| 514 | |
| 515 | /* some instructions like division have a TRAP termination type - ignore it */ |
| 516 | |
| 517 | return 0; |
| 518 | |
| 519 | case DISARM64::TRMTA::trmtaBra: |
| 520 | case DISARM64::TRMTA::trmtaBraCase: |
| 521 | case DISARM64::TRMTA::trmtaBraCc: |
| 522 | case DISARM64::TRMTA::trmtaBraCcCase: |
| 523 | |
| 524 | /* these are treated by the CchAddr callback - skip them */ |
| 525 | |
| 526 | return 0; |
| 527 | |
| 528 | case DISARM64::TRMTA::trmtaCall: |
| 529 | case DISARM64::TRMTA::trmtaCallCc: |
| 530 | |
| 531 | if (anyReloc) |
| 532 | { |
| 533 | const char* name = disGetMethodFullName(targetAddr); |
| 534 | if (name != nullptr) |
| 535 | { |
| 536 | swprintf_s(wz, cchMax, W("%zx %S" ), dspAddr(targetAddr), name); |
| 537 | break; |
| 538 | } |
| 539 | } |
| 540 | |
| 541 | /* these are treated by the CchAddr callback - skip them */ |
| 542 | |
| 543 | return 0; |
| 544 | |
| 545 | case DISARM64::TRMTA::trmtaCallInd: |
| 546 | case DISARM64::TRMTA::trmtaCallCcInd: |
| 547 | |
| 548 | /* here we have an indirect call - find the indirect address */ |
| 549 | |
| 550 | // BYTE * code = disGetLinearAddr((size_t)addr); |
| 551 | // disIndAddr = (DIS::ADDR) (code+0); |
| 552 | |
| 553 | /* find the size of the call opcode - less the immediate */ |
| 554 | /* for the fixup offset we have to add the opcode size for the call */ |
| 555 | /* addr is the address of the immediate, pdis->Addr() returns the address of the disassembled instruction */ |
| 556 | |
| 557 | assert(addr > pdis->Addr()); |
| 558 | disCallSize = addr - pdis->Addr(); |
| 559 | |
| 560 | /* find the emitter block and the offset of the call fixup */ |
| 561 | |
| 562 | return 0; |
| 563 | |
| 564 | default: |
| 565 | |
| 566 | printf("Termination type is %d\n" , (int)terminationType); |
| 567 | assert(!"treat this case\n" ); |
| 568 | break; |
| 569 | } |
| 570 | |
| 571 | #else // _TARGET_* |
| 572 | #error Unsupported or unset target architecture |
| 573 | #endif // _TARGET_* |
| 574 | |
| 575 | /* no displacement */ |
| 576 | |
| 577 | *pdwDisp = 0x0; |
| 578 | |
| 579 | return 1; |
| 580 | } |
| 581 | |
| 582 | /***************************************************************************** |
| 583 | * This the callback for register-relative operands in an instruction. |
| 584 | * If the register is ESP or EBP, the operand may be a local variable |
| 585 | * or a parameter, else the operand may be an instance variable |
| 586 | * |
| 587 | * Return 1 if a name was written representing the register-relative operand, 0 otherwise. |
| 588 | */ |
| 589 | |
| 590 | /* static */ |
| 591 | size_t __stdcall DisAssembler::disCchRegRel( |
| 592 | const DIS* pdis, DIS::REGA reg, DWORD disp, __in_ecount(cchMax) wchar_t* wz, size_t cchMax, DWORD* pdwDisp) |
| 593 | { |
| 594 | DisAssembler* pDisAsm = (DisAssembler*)pdis->PvClient(); |
| 595 | assert(pDisAsm); |
| 596 | |
| 597 | return pDisAsm->disCchRegRelMember(pdis, reg, disp, wz, cchMax, pdwDisp); |
| 598 | } |
| 599 | |
| 600 | size_t DisAssembler::disCchRegRelMember( |
| 601 | const DIS* pdis, DIS::REGA reg, DWORD disp, __in_ecount(cchMax) wchar_t* wz, size_t cchMax, DWORD* pdwDisp) |
| 602 | { |
| 603 | #if defined(_TARGET_XARCH_) |
| 604 | |
| 605 | DISX86::TRMTA terminationType = DISX86::TRMTA(pdis->Trmta()); |
| 606 | // DIS::ADDR disIndAddr; |
| 607 | |
| 608 | DISASM_DUMP("RegRelMember reg %u, disp %u, termType %u\n" , reg, disp, terminationType); |
| 609 | |
| 610 | switch (terminationType) |
| 611 | { |
| 612 | int disOpcodeSize; |
| 613 | const char* var; |
| 614 | |
| 615 | case DISX86::trmtaFallThrough: |
| 616 | |
| 617 | /* some instructions like division have a TRAP termination type - ignore it */ |
| 618 | |
| 619 | case DISX86::trmtaTrap: |
| 620 | case DISX86::trmtaTrapCc: |
| 621 | |
| 622 | var = disComp->codeGen->siStackVarName((size_t)(pdis->Addr() - disStartAddr), pdis->Cb(), reg, disp); |
| 623 | if (var) |
| 624 | { |
| 625 | swprintf_s(wz, cchMax, W("%hs+%Xh '%hs'" ), getRegName(reg), disp, var); |
| 626 | *pdwDisp = 0; |
| 627 | |
| 628 | return 1; |
| 629 | } |
| 630 | |
| 631 | /* This case consists of non-static members */ |
| 632 | |
| 633 | /* find the emitter block and the offset for the fixup |
| 634 | * fixup is emited after the coding of the instruction - size = word (2 bytes) |
| 635 | * GRRRR!!! - for the 16 bit case we have to check for the address size prefix = 0x66 |
| 636 | */ |
| 637 | |
| 638 | if (*disGetLinearAddr(disCurOffset) == 0x66) |
| 639 | { |
| 640 | disOpcodeSize = 3; |
| 641 | } |
| 642 | else |
| 643 | { |
| 644 | disOpcodeSize = 2; |
| 645 | } |
| 646 | |
| 647 | return 0; |
| 648 | |
| 649 | case DISX86::trmtaCallNear16: |
| 650 | case DISX86::trmtaCallNear32: |
| 651 | case DISX86::trmtaJmpInd: |
| 652 | |
| 653 | break; |
| 654 | |
| 655 | case DISX86::trmtaCallInd: |
| 656 | |
| 657 | /* check if this is a one byte displacement */ |
| 658 | |
| 659 | if ((signed char)disp == (int)disp) |
| 660 | { |
| 661 | /* we have a one byte displacement -> there were no previous callbacks */ |
| 662 | |
| 663 | /* find the size of the call opcode - less the immediate */ |
| 664 | /* this is a call R/M indirect -> opcode size is 2 */ |
| 665 | |
| 666 | disOpcodeSize = 2; |
| 667 | |
| 668 | /* find the emitter block and the offset of the call fixup */ |
| 669 | |
| 670 | return 0; |
| 671 | } |
| 672 | else |
| 673 | { |
| 674 | /* check if we already have a symbol name as replacement */ |
| 675 | |
| 676 | if (disHasName) |
| 677 | { |
| 678 | /* CchFixup has been called before - we have a symbol name saved in global var disFuncTempBuf */ |
| 679 | |
| 680 | swprintf_s(wz, cchMax, W("%hs+%u '%hs'" ), getRegName(reg), disp, disFuncTempBuf); |
| 681 | *pdwDisp = 0; |
| 682 | disHasName = false; |
| 683 | return 1; |
| 684 | } |
| 685 | else |
| 686 | { |
| 687 | return 0; |
| 688 | } |
| 689 | } |
| 690 | |
| 691 | default: |
| 692 | |
| 693 | printf("Termination type is %d\n" , (int)terminationType); |
| 694 | assert(!"treat this case\n" ); |
| 695 | |
| 696 | break; |
| 697 | } |
| 698 | |
| 699 | #elif defined(_TARGET_ARM64_) |
| 700 | |
| 701 | DISARM64::TRMTA terminationType = DISARM64::TRMTA(pdis->Trmta()); |
| 702 | |
| 703 | DISASM_DUMP("RegRelMember reg %u, disp %u, termType %u\n" , reg, disp, terminationType); |
| 704 | |
| 705 | switch (terminationType) |
| 706 | { |
| 707 | int disOpcodeSize; |
| 708 | const char* var; |
| 709 | |
| 710 | case DISARM64::TRMTA::trmtaFallThrough: |
| 711 | |
| 712 | /* some instructions like division have a TRAP termination type - ignore it */ |
| 713 | |
| 714 | case DISARM64::TRMTA::trmtaTrap: |
| 715 | case DISARM64::TRMTA::trmtaTrapCc: |
| 716 | |
| 717 | var = disComp->codeGen->siStackVarName((size_t)(pdis->Addr() - disStartAddr), pdis->Cb(), reg, disp); |
| 718 | if (var) |
| 719 | { |
| 720 | swprintf_s(wz, cchMax, W("%hs+%Xh '%hs'" ), getRegName(reg), disp, var); |
| 721 | *pdwDisp = 0; |
| 722 | |
| 723 | return 1; |
| 724 | } |
| 725 | |
| 726 | /* This case consists of non-static members */ |
| 727 | |
| 728 | // TODO-ARM64-Bug?: Is this correct? |
| 729 | disOpcodeSize = 2; |
| 730 | return 0; |
| 731 | |
| 732 | case DISARM64::TRMTA::trmtaCall: |
| 733 | case DISARM64::TRMTA::trmtaCallCc: |
| 734 | case DISARM64::TRMTA::trmtaBraInd: |
| 735 | case DISARM64::TRMTA::trmtaBraCcInd: |
| 736 | break; |
| 737 | |
| 738 | case DISARM64::TRMTA::trmtaCallInd: |
| 739 | case DISARM64::TRMTA::trmtaCallCcInd: |
| 740 | |
| 741 | /* check if this is a one byte displacement */ |
| 742 | |
| 743 | if ((signed char)disp == (int)disp) |
| 744 | { |
| 745 | /* we have a one byte displacement -> there were no previous callbacks */ |
| 746 | |
| 747 | /* find the size of the call opcode - less the immediate */ |
| 748 | /* this is a call R/M indirect -> opcode size is 2 */ |
| 749 | |
| 750 | // TODO-ARM64-Bug?: Is this correct? |
| 751 | disOpcodeSize = 2; |
| 752 | |
| 753 | /* find the emitter block and the offset of the call fixup */ |
| 754 | |
| 755 | return 0; |
| 756 | } |
| 757 | else |
| 758 | { |
| 759 | /* check if we already have a symbol name as replacement */ |
| 760 | |
| 761 | if (disHasName) |
| 762 | { |
| 763 | /* CchFixup has been called before - we have a symbol name saved in global var disFuncTempBuf */ |
| 764 | |
| 765 | swprintf_s(wz, cchMax, W("%hs+%u '%hs'" ), getRegName(reg), disp, disFuncTempBuf); |
| 766 | *pdwDisp = 0; |
| 767 | disHasName = false; |
| 768 | return 1; |
| 769 | } |
| 770 | else |
| 771 | { |
| 772 | return 0; |
| 773 | } |
| 774 | } |
| 775 | |
| 776 | default: |
| 777 | |
| 778 | printf("Termination type is %d\n" , (int)terminationType); |
| 779 | assert(!"treat this case\n" ); |
| 780 | |
| 781 | break; |
| 782 | } |
| 783 | |
| 784 | #else // _TARGET_* |
| 785 | #error Unsupported or unset target architecture |
| 786 | #endif // _TARGET_* |
| 787 | |
| 788 | /* save displacement */ |
| 789 | |
| 790 | *pdwDisp = disp; |
| 791 | |
| 792 | return 1; |
| 793 | } |
| 794 | |
| 795 | /***************************************************************************** |
| 796 | * |
| 797 | * Callback for register operands. Most probably, this is a local variable or |
| 798 | * a parameter |
| 799 | * |
| 800 | * Return 1 if a name was written representing the register, 0 otherwise. |
| 801 | */ |
| 802 | |
| 803 | /* static */ |
| 804 | size_t __stdcall DisAssembler::disCchReg(const DIS* pdis, DIS::REGA reg, __in_ecount(cchMax) wchar_t* wz, size_t cchMax) |
| 805 | { |
| 806 | DisAssembler* pDisAsm = (DisAssembler*)pdis->PvClient(); |
| 807 | assert(pDisAsm); |
| 808 | |
| 809 | return pDisAsm->disCchRegMember(pdis, reg, wz, cchMax); |
| 810 | } |
| 811 | |
| 812 | size_t DisAssembler::disCchRegMember(const DIS* pdis, DIS::REGA reg, __in_ecount(cchMax) wchar_t* wz, size_t cchMax) |
| 813 | { |
| 814 | // TODO-Review: DIS::REGA does not directly map to our regNumber! E.g., look at DISARM64::REGA -- |
| 815 | // the Wt registers come first (and do map to our regNumber), but the Xt registers follow. |
| 816 | // Until this is fixed, don't use this function! |
| 817 | disHasName = false; |
| 818 | return 0; |
| 819 | |
| 820 | #if 0 |
| 821 | const char * var = disComp->codeGen->siRegVarName( |
| 822 | (size_t)(pdis->Addr() - disStartAddr), |
| 823 | pdis->Cb(), |
| 824 | reg); |
| 825 | |
| 826 | if (var) |
| 827 | { |
| 828 | if (disHasName) |
| 829 | { |
| 830 | /* CchRegRel has been called before - we have a symbol name saved in global var disFuncTempBuf */ |
| 831 | |
| 832 | swprintf_s(wz, cchMax, W("%hs'%hs.%hs'" ), getRegName(reg), var, disFuncTempBuf); |
| 833 | disHasName = false; |
| 834 | return 1; |
| 835 | } |
| 836 | else |
| 837 | { |
| 838 | swprintf_s(wz, cchMax, W("%hs'%hs'" ), getRegName(reg), var); |
| 839 | return 1; |
| 840 | } |
| 841 | } |
| 842 | else |
| 843 | { |
| 844 | if (disHasName) |
| 845 | { |
| 846 | /* this is the ugly case when a variable is incorrectly presumed dead */ |
| 847 | |
| 848 | swprintf_s(wz, cchMax, W("%hs'%hs.%hs'" ), getRegName(reg), "<InstVar>" , disFuncTempBuf); |
| 849 | disHasName = false; |
| 850 | return 1; |
| 851 | } |
| 852 | |
| 853 | /* just to make sure we didn't bungle if var returns NULL */ |
| 854 | disHasName = false; |
| 855 | return 0; |
| 856 | } |
| 857 | #endif // 0 |
| 858 | } |
| 859 | |
| 860 | /***************************************************************************** |
| 861 | * Helper function to lazily create a map from code address to CORINFO_METHOD_HANDLE. |
| 862 | */ |
| 863 | AddrToMethodHandleMap* DisAssembler::GetAddrToMethodHandleMap() |
| 864 | { |
| 865 | if (disAddrToMethodHandleMap == nullptr) |
| 866 | { |
| 867 | disAddrToMethodHandleMap = new (disComp->getAllocator()) AddrToMethodHandleMap(disComp->getAllocator()); |
| 868 | } |
| 869 | return disAddrToMethodHandleMap; |
| 870 | } |
| 871 | |
| 872 | /***************************************************************************** |
| 873 | * Helper function to lazily create a map from code address to CORINFO_METHOD_HANDLE. |
| 874 | */ |
| 875 | AddrToMethodHandleMap* DisAssembler::GetHelperAddrToMethodHandleMap() |
| 876 | { |
| 877 | if (disHelperAddrToMethodHandleMap == nullptr) |
| 878 | { |
| 879 | disHelperAddrToMethodHandleMap = new (disComp->getAllocator()) AddrToMethodHandleMap(disComp->getAllocator()); |
| 880 | } |
| 881 | return disHelperAddrToMethodHandleMap; |
| 882 | } |
| 883 | |
| 884 | /***************************************************************************** |
| 885 | * Helper function to lazily create a map from relocation address to relocation target address. |
| 886 | */ |
| 887 | AddrToAddrMap* DisAssembler::GetRelocationMap() |
| 888 | { |
| 889 | if (disRelocationMap == nullptr) |
| 890 | { |
| 891 | disRelocationMap = new (disComp->getAllocator()) AddrToAddrMap(disComp->getAllocator()); |
| 892 | } |
| 893 | return disRelocationMap; |
| 894 | } |
| 895 | |
| 896 | /***************************************************************************** |
| 897 | * Return the count of bytes disassembled. |
| 898 | */ |
| 899 | |
| 900 | size_t DisAssembler::CbDisassemble(DIS* pdis, |
| 901 | size_t offs, |
| 902 | DIS::ADDR addr, |
| 903 | const BYTE* pb, |
| 904 | size_t cbMax, |
| 905 | FILE* pfile, |
| 906 | bool findLabels, |
| 907 | bool printit /* = false */, |
| 908 | bool dispOffs /* = false */, |
| 909 | bool dispCodeBytes /* = false */) |
| 910 | { |
| 911 | assert(pdis); |
| 912 | |
| 913 | size_t cb = pdis->CbDisassemble(addr, pb, cbMax); |
| 914 | |
| 915 | if (cb == 0) |
| 916 | { |
| 917 | DISASM_DUMP("CbDisassemble offs %Iu addr %I64u\n" , offs, addr); |
| 918 | // assert(!"can't disassemble instruction!!!"); |
| 919 | fprintf(pfile, "MSVCDIS can't disassemble instruction @ offset %Iu (0x%02x)!!!\n" , offs, offs); |
| 920 | #if defined(_TARGET_ARM64_) |
| 921 | fprintf(pfile, "%08Xh\n" , *(unsigned int*)pb); |
| 922 | return 4; |
| 923 | #else |
| 924 | fprintf(pfile, "%02Xh\n" , *pb); |
| 925 | return 1; |
| 926 | #endif |
| 927 | } |
| 928 | |
| 929 | #if defined(_TARGET_ARM64_) |
| 930 | assert(cb == 4); // all instructions are 4 bytes! |
| 931 | #endif // _TARGET_ARM64_ |
| 932 | |
| 933 | /* remember current offset and instruction size */ |
| 934 | |
| 935 | disCurOffset = (size_t)addr; |
| 936 | disInstSize = cb; |
| 937 | |
| 938 | /* Set the disTarget address */ |
| 939 | |
| 940 | disTarget = (size_t)pdis->AddrTarget(); |
| 941 | |
| 942 | if (findLabels) |
| 943 | { |
| 944 | #if defined(_TARGET_XARCH_) |
| 945 | DISX86::TRMTA terminationType = DISX86::TRMTA(pdis->Trmta()); |
| 946 | |
| 947 | /* check the termination type of the instruction */ |
| 948 | |
| 949 | switch (terminationType) |
| 950 | { |
| 951 | case DISX86::trmtaCallNear16: |
| 952 | case DISX86::trmtaCallNear32: |
| 953 | case DISX86::trmtaCallFar: |
| 954 | |
| 955 | { |
| 956 | // Don't count addresses in the relocation table |
| 957 | size_t targetAddr; |
| 958 | size_t absoluteAddr = |
| 959 | (size_t)disGetLinearAddr((size_t)pdis->AddrAddress(1)); // Get the address in the instruction of the |
| 960 | // call target address (the address the |
| 961 | // reloc is applied to). |
| 962 | if (GetRelocationMap()->Lookup(absoluteAddr, &targetAddr)) |
| 963 | { |
| 964 | break; |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | __fallthrough; |
| 969 | |
| 970 | case DISX86::trmtaJmpShort: |
| 971 | case DISX86::trmtaJmpNear: |
| 972 | case DISX86::trmtaJmpFar: |
| 973 | case DISX86::trmtaJmpCcShort: |
| 974 | case DISX86::trmtaJmpCcNear: |
| 975 | |
| 976 | /* a CALL is local iff the disTarget is within the block boundary */ |
| 977 | |
| 978 | /* mark the jump label in the disTarget vector and return */ |
| 979 | |
| 980 | if (disTarget != DIS::addrNil) // There seems to be an assumption that you can't branch to the first |
| 981 | // address of the function (prolog). |
| 982 | { |
| 983 | if (0 <= disTarget && disTarget < disTotalCodeSize) |
| 984 | { |
| 985 | /* we're OK, disTarget within block boundary */ |
| 986 | |
| 987 | disLabels[disTarget] = 1; |
| 988 | } |
| 989 | } |
| 990 | break; |
| 991 | |
| 992 | case DISX86::trmtaFallThrough: |
| 993 | // We'd like to be able to get a label for code like "lea rcx, [4]" that we use for jump tables, but I |
| 994 | // can't figure out how. |
| 995 | break; |
| 996 | |
| 997 | default: |
| 998 | |
| 999 | /* jump is not in the current code block */ |
| 1000 | break; |
| 1001 | |
| 1002 | } // end switch |
| 1003 | #elif defined(_TARGET_ARM64_) |
| 1004 | DISARM64::TRMTA terminationType = DISARM64::TRMTA(pdis->Trmta()); |
| 1005 | |
| 1006 | /* check the termination type of the instruction */ |
| 1007 | |
| 1008 | switch (terminationType) |
| 1009 | { |
| 1010 | case DISARM64::TRMTA::trmtaCall: |
| 1011 | case DISARM64::TRMTA::trmtaCallCc: |
| 1012 | |
| 1013 | { |
| 1014 | // Don't count addresses in the relocation table |
| 1015 | size_t targetAddr; |
| 1016 | size_t absoluteAddr = |
| 1017 | (size_t)disGetLinearAddr((size_t)pdis->AddrAddress(1)); // Get the address in the instruction of the |
| 1018 | // call target address (the address the |
| 1019 | // reloc is applied to). |
| 1020 | if (GetRelocationMap()->Lookup(absoluteAddr, &targetAddr)) |
| 1021 | { |
| 1022 | break; |
| 1023 | } |
| 1024 | } |
| 1025 | |
| 1026 | __fallthrough; |
| 1027 | |
| 1028 | case DISARM64::TRMTA::trmtaBra: |
| 1029 | case DISARM64::TRMTA::trmtaBraCase: |
| 1030 | case DISARM64::TRMTA::trmtaBraCc: |
| 1031 | case DISARM64::TRMTA::trmtaBraCcCase: |
| 1032 | |
| 1033 | /* a CALL is local iff the disTarget is within the block boundary */ |
| 1034 | |
| 1035 | /* mark the jump label in the disTarget vector and return */ |
| 1036 | |
| 1037 | if (disTarget != DIS::addrNil) // There seems to be an assumption that you can't branch to the first |
| 1038 | // address of the function (prolog). |
| 1039 | { |
| 1040 | if (0 <= disTarget && disTarget < disTotalCodeSize) |
| 1041 | { |
| 1042 | /* we're OK, disTarget within block boundary */ |
| 1043 | |
| 1044 | disLabels[disTarget] = 1; |
| 1045 | } |
| 1046 | } |
| 1047 | break; |
| 1048 | |
| 1049 | case DISARM64::TRMTA::trmtaFallThrough: |
| 1050 | { |
| 1051 | DIS::INSTRUCTION instr; |
| 1052 | DIS::OPERAND ops[DISARM64::coperandMax]; |
| 1053 | bool ok = pdis->FDecode(&instr, ops, ArrLen(ops)); |
| 1054 | if (ok) |
| 1055 | { |
| 1056 | switch ((DISARM64::OPA)instr.opa) |
| 1057 | { |
| 1058 | case DISARM64::opaAdr: |
| 1059 | case DISARM64::opaAdrp: |
| 1060 | // operand 1 is an address |
| 1061 | assert(instr.coperand >= 2); |
| 1062 | assert(ops[1].opcls == DIS::opclsImmediate); |
| 1063 | assert(ops[1].imcls == DIS::imclsAddress); |
| 1064 | disTarget = ops[1].dwl; |
| 1065 | break; |
| 1066 | default: |
| 1067 | break; |
| 1068 | } |
| 1069 | |
| 1070 | if (0 <= disTarget && disTarget < disTotalCodeSize) |
| 1071 | { |
| 1072 | /* we're OK, disTarget within block boundary */ |
| 1073 | |
| 1074 | disLabels[disTarget] = 1; |
| 1075 | } |
| 1076 | } |
| 1077 | } |
| 1078 | break; |
| 1079 | |
| 1080 | default: |
| 1081 | |
| 1082 | /* jump is not in the current code block */ |
| 1083 | break; |
| 1084 | |
| 1085 | } // end switch |
| 1086 | #else // _TARGET_* |
| 1087 | #error Unsupported or unset target architecture |
| 1088 | #endif // _TARGET_* |
| 1089 | |
| 1090 | return cb; |
| 1091 | } // end if |
| 1092 | |
| 1093 | /* check if we have a label here */ |
| 1094 | |
| 1095 | if (printit) |
| 1096 | { |
| 1097 | if (disLabels[addr]) |
| 1098 | { |
| 1099 | /* print the label and the offset */ |
| 1100 | |
| 1101 | fprintf(pfile, "L_%02u:\n" , disLabels[addr]); |
| 1102 | } |
| 1103 | } |
| 1104 | |
| 1105 | wchar_t wz[MAX_CLASSNAME_LENGTH]; |
| 1106 | pdis->CchFormatInstr(wz, _countof(wz)); |
| 1107 | |
| 1108 | if (printit) |
| 1109 | { |
| 1110 | if (dispOffs) |
| 1111 | { |
| 1112 | fprintf(pfile, "%03X" , offs); |
| 1113 | } |
| 1114 | |
| 1115 | #ifdef _TARGET_ARM64_ |
| 1116 | #define CCH_INDENT 8 // fixed sized instructions, always 8 characters |
| 1117 | #elif defined(_TARGET_AMD64_) |
| 1118 | #define CCH_INDENT 30 // large constants sometimes |
| 1119 | #else |
| 1120 | #define CCH_INDENT 24 |
| 1121 | #endif |
| 1122 | |
| 1123 | size_t cchIndent = CCH_INDENT; |
| 1124 | |
| 1125 | if (dispCodeBytes) |
| 1126 | { |
| 1127 | static size_t cchBytesMax = -1; |
| 1128 | |
| 1129 | if (cchBytesMax == -1) |
| 1130 | { |
| 1131 | cchBytesMax = pdis->CchFormatBytesMax(); |
| 1132 | } |
| 1133 | |
| 1134 | wchar_t wzBytes[MAX_CLASSNAME_LENGTH]; |
| 1135 | assert(cchBytesMax < MAX_CLASSNAME_LENGTH); |
| 1136 | |
| 1137 | size_t cchBytes = pdis->CchFormatBytes(wzBytes, _countof(wzBytes)); |
| 1138 | |
| 1139 | if (cchBytes > CCH_INDENT) |
| 1140 | { |
| 1141 | // Truncate the bytes if they are too long |
| 1142 | |
| 1143 | static const wchar_t* elipses = W("...\0" ); |
| 1144 | const size_t cchElipses = 4; |
| 1145 | |
| 1146 | memcpy(&wzBytes[CCH_INDENT - cchElipses], elipses, cchElipses * sizeof(wchar_t)); |
| 1147 | |
| 1148 | cchBytes = CCH_INDENT; |
| 1149 | } |
| 1150 | |
| 1151 | fprintf(pfile, " %ls" , wzBytes); |
| 1152 | cchIndent = CCH_INDENT - cchBytes; |
| 1153 | } |
| 1154 | |
| 1155 | // print the dis-assembled instruction |
| 1156 | |
| 1157 | fprintf(pfile, "%*c %ls\n" , cchIndent, ' ', wz); |
| 1158 | } |
| 1159 | |
| 1160 | return cb; |
| 1161 | } |
| 1162 | |
| 1163 | // TODO-Cleanup: this is currently unused, unreferenced. |
| 1164 | size_t CbDisassembleWithBytes(DIS* pdis, DIS::ADDR addr, const BYTE* pb, size_t cbMax, FILE* pfile) |
| 1165 | { |
| 1166 | assert(pdis); |
| 1167 | DisAssembler* pDisAsm = (DisAssembler*)pdis->PvClient(); |
| 1168 | assert(pDisAsm); |
| 1169 | |
| 1170 | wchar_t wz[MAX_CLASSNAME_LENGTH]; |
| 1171 | |
| 1172 | pdis->CchFormatAddr(addr, wz, _countof(wz)); |
| 1173 | |
| 1174 | size_t cchIndent = (size_t)fprintf(pfile, " %ls: " , wz); |
| 1175 | |
| 1176 | size_t cb = pdis->CbDisassemble(addr, pb, cbMax); |
| 1177 | |
| 1178 | if (cb == 0) |
| 1179 | { |
| 1180 | fprintf(pfile, "%02Xh\n" , *pb); |
| 1181 | return (1); |
| 1182 | } |
| 1183 | |
| 1184 | size_t cchBytesMax = pdis->CchFormatBytesMax(); |
| 1185 | |
| 1186 | if (cchBytesMax > 18) |
| 1187 | { |
| 1188 | // Limit bytes coded to 18 characters |
| 1189 | |
| 1190 | cchBytesMax = 18; |
| 1191 | } |
| 1192 | |
| 1193 | wchar_t wzBytes[64]; |
| 1194 | size_t cchBytes = pdis->CchFormatBytes(wzBytes, _countof(wzBytes)); |
| 1195 | |
| 1196 | wchar_t* pwzBytes; |
| 1197 | wchar_t* pwzNext; |
| 1198 | |
| 1199 | for (pwzBytes = wzBytes; pwzBytes != NULL; pwzBytes = pwzNext) |
| 1200 | { |
| 1201 | BOOL fFirst = (pwzBytes == wzBytes); |
| 1202 | |
| 1203 | cchBytes = wcslen(pwzBytes); |
| 1204 | |
| 1205 | if (cchBytes <= cchBytesMax) |
| 1206 | { |
| 1207 | pwzNext = NULL; |
| 1208 | } |
| 1209 | |
| 1210 | else |
| 1211 | { |
| 1212 | wchar_t ch = pwzBytes[cchBytesMax]; |
| 1213 | pwzBytes[cchBytesMax] = '\0'; |
| 1214 | |
| 1215 | if (ch == W(' ')) |
| 1216 | { |
| 1217 | pwzNext = pwzBytes + cchBytesMax + 1; |
| 1218 | } |
| 1219 | |
| 1220 | else |
| 1221 | { |
| 1222 | pwzNext = wcsrchr(pwzBytes, W(' ')); |
| 1223 | assert(pwzNext); |
| 1224 | |
| 1225 | pwzBytes[cchBytesMax] = ch; |
| 1226 | *pwzNext++ = '\0'; |
| 1227 | } |
| 1228 | } |
| 1229 | |
| 1230 | if (fFirst) |
| 1231 | { |
| 1232 | pdis->CchFormatInstr(wz, _countof(wz)); |
| 1233 | fprintf(pfile, "%-*ls %ls\n" , cchBytesMax, pwzBytes, wz); |
| 1234 | } |
| 1235 | |
| 1236 | else |
| 1237 | { |
| 1238 | fprintf(pfile, "%*c%ls\n" , cchIndent, ' ', pwzBytes); |
| 1239 | } |
| 1240 | } |
| 1241 | |
| 1242 | return (cb); |
| 1243 | } |
| 1244 | |
| 1245 | void DisAssembler::DisasmBuffer(FILE* pfile, bool printit) |
| 1246 | { |
| 1247 | DIS* pdis = NULL; |
| 1248 | |
| 1249 | #ifdef _TARGET_X86_ |
| 1250 | pdis = DIS::PdisNew(DIS::distX86); |
| 1251 | #elif defined(_TARGET_AMD64_) |
| 1252 | pdis = DIS::PdisNew(DIS::distX8664); |
| 1253 | #elif defined(_TARGET_ARM64_) |
| 1254 | pdis = DIS::PdisNew(DIS::distArm64); |
| 1255 | #else // _TARGET_* |
| 1256 | #error Unsupported or unset target architecture |
| 1257 | #endif |
| 1258 | |
| 1259 | if (pdis == NULL) |
| 1260 | { |
| 1261 | assert(!"out of memory in disassembler?" ); |
| 1262 | return; |
| 1263 | } |
| 1264 | |
| 1265 | #ifdef _TARGET_64BIT_ |
| 1266 | pdis->SetAddr64(true); |
| 1267 | #endif |
| 1268 | |
| 1269 | // Store a pointer to the DisAssembler so that the callback functions |
| 1270 | // can get to it. |
| 1271 | |
| 1272 | pdis->PvClientSet((void*)this); |
| 1273 | |
| 1274 | /* Calculate addresses */ |
| 1275 | |
| 1276 | size_t ibCur = 0; |
| 1277 | DIS::ADDR addr = 0; // Always emit code with respect to a "0" base address. |
| 1278 | |
| 1279 | /* First walk the code to find all jump targets */ |
| 1280 | |
| 1281 | while (ibCur < disTotalCodeSize) |
| 1282 | { |
| 1283 | size_t cb; |
| 1284 | |
| 1285 | cb = CbDisassemble(pdis, ibCur, addr + ibCur, disGetLinearAddr(ibCur), disGetBufferSize(ibCur), pfile, |
| 1286 | true); // find labels |
| 1287 | |
| 1288 | // CbDisassemble returning > MAX_INT... give me a break. |
| 1289 | ibCur += cb; |
| 1290 | } |
| 1291 | |
| 1292 | /* reset the label counter and start assigning consecutive number labels to the label locations */ |
| 1293 | |
| 1294 | BYTE label = 0; |
| 1295 | for (unsigned i = 0; i < disTotalCodeSize; i++) |
| 1296 | { |
| 1297 | if (disLabels[i] != 0) |
| 1298 | { |
| 1299 | disLabels[i] = ++label; |
| 1300 | } |
| 1301 | } |
| 1302 | |
| 1303 | /* Re-initialize addresses for disassemble phase */ |
| 1304 | |
| 1305 | ibCur = 0; |
| 1306 | addr = 0; |
| 1307 | |
| 1308 | // Set callbacks only if we are displaying it. Else, the scheduler has called it |
| 1309 | |
| 1310 | if (printit) |
| 1311 | { |
| 1312 | /* Set the callback functions for symbol lookup */ |
| 1313 | |
| 1314 | pdis->PfncchaddrSet(disCchAddr); |
| 1315 | pdis->PfncchfixupSet(disCchFixup); |
| 1316 | pdis->PfncchregrelSet(disCchRegRel); |
| 1317 | pdis->PfncchregSet(disCchReg); |
| 1318 | } |
| 1319 | |
| 1320 | while (ibCur < disTotalCodeSize) |
| 1321 | { |
| 1322 | size_t cb; |
| 1323 | |
| 1324 | cb = CbDisassemble(pdis, ibCur, addr + ibCur, disGetLinearAddr(ibCur), disGetBufferSize(ibCur), pfile, |
| 1325 | false, // find labels |
| 1326 | printit, |
| 1327 | !disDiffable, // display relative offset |
| 1328 | #ifdef DEBUG |
| 1329 | !disDiffable // Display code bytes? |
| 1330 | #else |
| 1331 | false // Display code bytes? |
| 1332 | #endif |
| 1333 | ); |
| 1334 | |
| 1335 | ibCur += (unsigned)cb; |
| 1336 | } |
| 1337 | |
| 1338 | delete pdis; |
| 1339 | } |
| 1340 | |
| 1341 | /***************************************************************************** |
| 1342 | * Given a linear offset into the code, find a pointer to the actual code (either in the hot or cold section) |
| 1343 | * |
| 1344 | * Arguments: |
| 1345 | * offset - The linear offset into the code. It must point within the code. |
| 1346 | */ |
| 1347 | |
| 1348 | const BYTE* DisAssembler::disGetLinearAddr(size_t offset) |
| 1349 | { |
| 1350 | if (offset < disHotCodeSize) |
| 1351 | { |
| 1352 | return (const BYTE*)disHotCodeBlock + offset; |
| 1353 | } |
| 1354 | else |
| 1355 | { |
| 1356 | return (const BYTE*)disColdCodeBlock + offset - disHotCodeSize; |
| 1357 | } |
| 1358 | } |
| 1359 | |
| 1360 | /***************************************************************************** |
| 1361 | * Given a linear offset into the code, determine how many bytes are remaining in the buffer. |
| 1362 | * This will only return the number of bytes left in either the hot or cold buffer. This is used |
| 1363 | * to avoid walking off the end of the buffer. |
| 1364 | * |
| 1365 | * Arguments: |
| 1366 | * offset - The linear offset into the code. It must point within the code. |
| 1367 | */ |
| 1368 | |
| 1369 | size_t DisAssembler::disGetBufferSize(size_t offset) |
| 1370 | { |
| 1371 | if (offset < disHotCodeSize) |
| 1372 | { |
| 1373 | return disHotCodeSize - offset; |
| 1374 | } |
| 1375 | else |
| 1376 | { |
| 1377 | return disHotCodeSize + disColdCodeSize - offset; |
| 1378 | } |
| 1379 | } |
| 1380 | |
| 1381 | /***************************************************************************** |
| 1382 | * Get the function name for a given absolute address. |
| 1383 | */ |
| 1384 | |
| 1385 | const char* DisAssembler::disGetMethodFullName(size_t addr) |
| 1386 | { |
| 1387 | CORINFO_METHOD_HANDLE res; |
| 1388 | |
| 1389 | // First check the JIT helper table: they're very common. |
| 1390 | if (GetHelperAddrToMethodHandleMap()->Lookup(addr, &res)) |
| 1391 | { |
| 1392 | return disComp->eeGetMethodFullName(res); |
| 1393 | } |
| 1394 | |
| 1395 | // Next check the "normal" registered call targets |
| 1396 | if (GetAddrToMethodHandleMap()->Lookup(addr, &res)) |
| 1397 | { |
| 1398 | return disComp->eeGetMethodFullName(res); |
| 1399 | } |
| 1400 | |
| 1401 | return nullptr; |
| 1402 | } |
| 1403 | |
| 1404 | /***************************************************************************** |
| 1405 | * Register a called function address as associated with a CORINFO_METHOD_HANDLE. |
| 1406 | * |
| 1407 | * Arguments: |
| 1408 | * addr - The absolute address of the target function. |
| 1409 | * methHnd - The method handle associated with 'addr'. |
| 1410 | */ |
| 1411 | |
| 1412 | void DisAssembler::disSetMethod(size_t addr, CORINFO_METHOD_HANDLE methHnd) |
| 1413 | { |
| 1414 | if (!disComp->opts.doLateDisasm) |
| 1415 | { |
| 1416 | return; |
| 1417 | } |
| 1418 | |
| 1419 | if (disComp->eeGetHelperNum(methHnd)) |
| 1420 | { |
| 1421 | DISASM_DUMP("Helper function: %p => %p\n" , addr, methHnd); |
| 1422 | GetHelperAddrToMethodHandleMap()->Set(addr, methHnd); |
| 1423 | } |
| 1424 | else |
| 1425 | { |
| 1426 | DISASM_DUMP("Function: %p => %p\n" , addr, methHnd); |
| 1427 | GetAddrToMethodHandleMap()->Set(addr, methHnd); |
| 1428 | } |
| 1429 | } |
| 1430 | |
| 1431 | /***************************************************************************** |
| 1432 | * Register a relocation. |
| 1433 | * |
| 1434 | * Arguments: |
| 1435 | * relocAddr - The absolute address the relocation applies to. |
| 1436 | * targetAddr - The absolute address the relocation points to. |
| 1437 | */ |
| 1438 | |
| 1439 | void DisAssembler::disRecordRelocation(size_t relocAddr, size_t targetAddr) |
| 1440 | { |
| 1441 | if (!disComp->opts.doLateDisasm) |
| 1442 | { |
| 1443 | return; |
| 1444 | } |
| 1445 | |
| 1446 | DISASM_DUMP("Relocation %p => %p\n" , relocAddr, targetAddr); |
| 1447 | GetRelocationMap()->Set(relocAddr, targetAddr); |
| 1448 | } |
| 1449 | |
| 1450 | /***************************************************************************** |
| 1451 | * |
| 1452 | * Disassemble the code which has been generated |
| 1453 | */ |
| 1454 | |
| 1455 | void DisAssembler::disAsmCode(BYTE* hotCodePtr, size_t hotCodeSize, BYTE* coldCodePtr, size_t coldCodeSize) |
| 1456 | { |
| 1457 | if (!disComp->opts.doLateDisasm) |
| 1458 | { |
| 1459 | return; |
| 1460 | } |
| 1461 | |
| 1462 | #ifdef DEBUG |
| 1463 | // Should we make it diffable? |
| 1464 | disDiffable = disComp->opts.dspDiffable; |
| 1465 | #else // !DEBUG |
| 1466 | // NOTE: non-debug builds are always diffable! |
| 1467 | disDiffable = true; |
| 1468 | #endif // !DEBUG |
| 1469 | |
| 1470 | #ifdef DEBUG |
| 1471 | const wchar_t* fileName = JitConfig.JitLateDisasmTo(); |
| 1472 | if (fileName != nullptr) |
| 1473 | { |
| 1474 | errno_t ec = _wfopen_s(&disAsmFile, fileName, W("a+" )); |
| 1475 | if (ec != 0) |
| 1476 | { |
| 1477 | disAsmFile = nullptr; |
| 1478 | } |
| 1479 | } |
| 1480 | #else // !DEBUG |
| 1481 | // NOTE: non-DEBUG builds always use jitstdout currently! |
| 1482 | disAsmFile = jitstdout; |
| 1483 | #endif // !DEBUG |
| 1484 | |
| 1485 | if (disAsmFile == nullptr) |
| 1486 | { |
| 1487 | disAsmFile = jitstdout; |
| 1488 | } |
| 1489 | |
| 1490 | // As this writes to a common file, this is not reentrant. |
| 1491 | |
| 1492 | assert(hotCodeSize > 0); |
| 1493 | if (coldCodeSize == 0) |
| 1494 | { |
| 1495 | fprintf(disAsmFile, "************************** %hs:%hs size 0x%04IX **************************\n\n" , |
| 1496 | disCurClassName, disCurMethodName, hotCodeSize); |
| 1497 | |
| 1498 | fprintf(disAsmFile, "Base address : %ph\n" , dspAddr(hotCodePtr)); |
| 1499 | } |
| 1500 | else |
| 1501 | { |
| 1502 | fprintf(disAsmFile, |
| 1503 | "************************** %hs:%hs hot size 0x%04IX cold size 0x%04IX **************************\n\n" , |
| 1504 | disCurClassName, disCurMethodName, hotCodeSize, coldCodeSize); |
| 1505 | |
| 1506 | fprintf(disAsmFile, "Hot address : %ph\n" , dspAddr(hotCodePtr)); |
| 1507 | fprintf(disAsmFile, "Cold address : %ph\n" , dspAddr(coldCodePtr)); |
| 1508 | } |
| 1509 | |
| 1510 | disStartAddr = 0; |
| 1511 | disHotCodeBlock = (size_t)hotCodePtr; |
| 1512 | disHotCodeSize = hotCodeSize; |
| 1513 | disColdCodeBlock = (size_t)coldCodePtr; |
| 1514 | disColdCodeSize = coldCodeSize; |
| 1515 | |
| 1516 | disTotalCodeSize = disHotCodeSize + disColdCodeSize; |
| 1517 | |
| 1518 | disLabels = new (disComp, CMK_DebugOnly) BYTE[disTotalCodeSize](); |
| 1519 | |
| 1520 | DisasmBuffer(disAsmFile, /* printIt */ true); |
| 1521 | fprintf(disAsmFile, "\n" ); |
| 1522 | |
| 1523 | if (disAsmFile != jitstdout) |
| 1524 | { |
| 1525 | fclose(disAsmFile); |
| 1526 | } |
| 1527 | else |
| 1528 | { |
| 1529 | fflush(disAsmFile); |
| 1530 | } |
| 1531 | } |
| 1532 | |
| 1533 | /*****************************************************************************/ |
| 1534 | // This function is called for every method. Checks if we are supposed to disassemble |
| 1535 | // the method, and where to send the disassembly output. |
| 1536 | |
| 1537 | void DisAssembler::disOpenForLateDisAsm(const char* curMethodName, const char* curClassName, PCCOR_SIGNATURE sig) |
| 1538 | { |
| 1539 | if (!disComp->opts.doLateDisasm) |
| 1540 | { |
| 1541 | return; |
| 1542 | } |
| 1543 | |
| 1544 | disCurMethodName = curMethodName; |
| 1545 | disCurClassName = curClassName; |
| 1546 | } |
| 1547 | |
| 1548 | /*****************************************************************************/ |
| 1549 | |
| 1550 | void DisAssembler::disInit(Compiler* pComp) |
| 1551 | { |
| 1552 | assert(pComp); |
| 1553 | disComp = pComp; |
| 1554 | disHasName = false; |
| 1555 | disLabels = nullptr; |
| 1556 | disAddrToMethodHandleMap = nullptr; |
| 1557 | disHelperAddrToMethodHandleMap = nullptr; |
| 1558 | disRelocationMap = nullptr; |
| 1559 | disDiffable = false; |
| 1560 | disAsmFile = nullptr; |
| 1561 | } |
| 1562 | |
| 1563 | /*****************************************************************************/ |
| 1564 | #endif // LATE_DISASM |
| 1565 | /*****************************************************************************/ |
| 1566 | |