1 //===-- llvm-mc.cpp - Machine Code Hacking Driver ---------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This utility is a simple driver that allows command line hacking on machine
10 // code.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "Disassembler.h"
15 #include "llvm/MC/MCAsmBackend.h"
16 #include "llvm/MC/MCAsmInfo.h"
17 #include "llvm/MC/MCCodeEmitter.h"
18 #include "llvm/MC/MCContext.h"
19 #include "llvm/MC/MCInstPrinter.h"
20 #include "llvm/MC/MCInstrInfo.h"
21 #include "llvm/MC/MCObjectFileInfo.h"
22 #include "llvm/MC/MCObjectWriter.h"
23 #include "llvm/MC/MCParser/AsmLexer.h"
24 #include "llvm/MC/MCParser/MCTargetAsmParser.h"
25 #include "llvm/MC/MCRegisterInfo.h"
26 #include "llvm/MC/MCStreamer.h"
27 #include "llvm/MC/MCSubtargetInfo.h"
28 #include "llvm/MC/MCTargetOptionsCommandFlags.h"
29 #include "llvm/Support/CommandLine.h"
30 #include "llvm/Support/Compression.h"
31 #include "llvm/Support/FileUtilities.h"
32 #include "llvm/Support/FormattedStream.h"
33 #include "llvm/Support/Host.h"
34 #include "llvm/Support/InitLLVM.h"
35 #include "llvm/Support/MemoryBuffer.h"
36 #include "llvm/Support/SourceMgr.h"
37 #include "llvm/Support/TargetRegistry.h"
38 #include "llvm/Support/TargetSelect.h"
39 #include "llvm/Support/ToolOutputFile.h"
40 #include "llvm/Support/WithColor.h"
41 
42 using namespace llvm;
43 
44 static mc::RegisterMCTargetOptionsFlags MOF;
45 
46 static cl::opt<std::string>
47 InputFilename(cl::Positional, cl::desc("<input file>"), cl::init("-"));
48 
49 static cl::opt<std::string> OutputFilename("o", cl::desc("Output filename"),
50                                            cl::value_desc("filename"),
51                                            cl::init("-"));
52 
53 static cl::opt<std::string> SplitDwarfFile("split-dwarf-file",
54                                            cl::desc("DWO output filename"),
55                                            cl::value_desc("filename"));
56 
57 static cl::opt<bool>
58 ShowEncoding("show-encoding", cl::desc("Show instruction encodings"));
59 
60 static cl::opt<bool> RelaxELFRel(
61     "relax-relocations", cl::init(true),
62     cl::desc("Emit R_X86_64_GOTPCRELX instead of R_X86_64_GOTPCREL"));
63 
64 static cl::opt<DebugCompressionType> CompressDebugSections(
65     "compress-debug-sections", cl::ValueOptional,
66     cl::init(DebugCompressionType::None),
67     cl::desc("Choose DWARF debug sections compression:"),
68     cl::values(clEnumValN(DebugCompressionType::None, "none", "No compression"),
69                clEnumValN(DebugCompressionType::Z, "zlib",
70                           "Use zlib compression"),
71                clEnumValN(DebugCompressionType::GNU, "zlib-gnu",
72                           "Use zlib-gnu compression (deprecated)")));
73 
74 static cl::opt<bool>
75 ShowInst("show-inst", cl::desc("Show internal instruction representation"));
76 
77 static cl::opt<bool>
78 ShowInstOperands("show-inst-operands",
79                  cl::desc("Show instructions operands as parsed"));
80 
81 static cl::opt<unsigned>
82 OutputAsmVariant("output-asm-variant",
83                  cl::desc("Syntax variant to use for output printing"));
84 
85 static cl::opt<bool>
86 PrintImmHex("print-imm-hex", cl::init(false),
87             cl::desc("Prefer hex format for immediate values"));
88 
89 static cl::list<std::string>
90 DefineSymbol("defsym", cl::desc("Defines a symbol to be an integer constant"));
91 
92 static cl::opt<bool>
93     PreserveComments("preserve-comments",
94                      cl::desc("Preserve Comments in outputted assembly"));
95 
96 enum OutputFileType {
97   OFT_Null,
98   OFT_AssemblyFile,
99   OFT_ObjectFile
100 };
101 static cl::opt<OutputFileType>
102 FileType("filetype", cl::init(OFT_AssemblyFile),
103   cl::desc("Choose an output file type:"),
104   cl::values(
105        clEnumValN(OFT_AssemblyFile, "asm",
106                   "Emit an assembly ('.s') file"),
107        clEnumValN(OFT_Null, "null",
108                   "Don't emit anything (for timing purposes)"),
109        clEnumValN(OFT_ObjectFile, "obj",
110                   "Emit a native object ('.o') file")));
111 
112 static cl::list<std::string>
113 IncludeDirs("I", cl::desc("Directory of include files"),
114             cl::value_desc("directory"), cl::Prefix);
115 
116 static cl::opt<std::string>
117 ArchName("arch", cl::desc("Target arch to assemble for, "
118                           "see -version for available targets"));
119 
120 static cl::opt<std::string>
121 TripleName("triple", cl::desc("Target triple to assemble for, "
122                               "see -version for available targets"));
123 
124 static cl::opt<std::string>
125 MCPU("mcpu",
126      cl::desc("Target a specific cpu type (-mcpu=help for details)"),
127      cl::value_desc("cpu-name"),
128      cl::init(""));
129 
130 static cl::list<std::string>
131 MAttrs("mattr",
132   cl::CommaSeparated,
133   cl::desc("Target specific attributes (-mattr=help for details)"),
134   cl::value_desc("a1,+a2,-a3,..."));
135 
136 static cl::opt<bool> PIC("position-independent",
137                          cl::desc("Position independent"), cl::init(false));
138 
139 static cl::opt<bool>
140     LargeCodeModel("large-code-model",
141                    cl::desc("Create cfi directives that assume the code might "
142                             "be more than 2gb away"));
143 
144 static cl::opt<bool>
145 NoInitialTextSection("n", cl::desc("Don't assume assembly file starts "
146                                    "in the text section"));
147 
148 static cl::opt<bool>
149 GenDwarfForAssembly("g", cl::desc("Generate dwarf debugging info for assembly "
150                                   "source files"));
151 
152 static cl::opt<std::string>
153 DebugCompilationDir("fdebug-compilation-dir",
154                     cl::desc("Specifies the debug info's compilation dir"));
155 
156 static cl::list<std::string>
157 DebugPrefixMap("fdebug-prefix-map",
158                cl::desc("Map file source paths in debug info"),
159                cl::value_desc("= separated key-value pairs"));
160 
161 static cl::opt<std::string>
162 MainFileName("main-file-name",
163              cl::desc("Specifies the name we should consider the input file"));
164 
165 static cl::opt<bool> SaveTempLabels("save-temp-labels",
166                                     cl::desc("Don't discard temporary labels"));
167 
168 static cl::opt<bool> LexMasmIntegers(
169     "masm-integers",
170     cl::desc("Enable binary and hex masm integers (0b110 and 0ABCh)"));
171 
172 static cl::opt<bool> LexMasmHexFloats(
173     "masm-hexfloats",
174     cl::desc("Enable MASM-style hex float initializers (3F800000r)"));
175 
176 static cl::opt<bool> NoExecStack("no-exec-stack",
177                                  cl::desc("File doesn't need an exec stack"));
178 
179 enum ActionType {
180   AC_AsLex,
181   AC_Assemble,
182   AC_Disassemble,
183   AC_MDisassemble,
184 };
185 
186 static cl::opt<ActionType>
187 Action(cl::desc("Action to perform:"),
188        cl::init(AC_Assemble),
189        cl::values(clEnumValN(AC_AsLex, "as-lex",
190                              "Lex tokens from a .s file"),
191                   clEnumValN(AC_Assemble, "assemble",
192                              "Assemble a .s file (default)"),
193                   clEnumValN(AC_Disassemble, "disassemble",
194                              "Disassemble strings of hex bytes"),
195                   clEnumValN(AC_MDisassemble, "mdis",
196                              "Marked up disassembly of strings of hex bytes")));
197 
198 static const Target *GetTarget(const char *ProgName) {
199   // Figure out the target triple.
200   if (TripleName.empty())
201     TripleName = sys::getDefaultTargetTriple();
202   Triple TheTriple(Triple::normalize(TripleName));
203 
204   // Get the target specific parser.
205   std::string Error;
206   const Target *TheTarget = TargetRegistry::lookupTarget(ArchName, TheTriple,
207                                                          Error);
208   if (!TheTarget) {
209     WithColor::error(errs(), ProgName) << Error;
210     return nullptr;
211   }
212 
213   // Update the triple name and return the found target.
214   TripleName = TheTriple.getTriple();
215   return TheTarget;
216 }
217 
218 static std::unique_ptr<ToolOutputFile> GetOutputStream(StringRef Path,
219     sys::fs::OpenFlags Flags) {
220   std::error_code EC;
221   auto Out = std::make_unique<ToolOutputFile>(Path, EC, Flags);
222   if (EC) {
223     WithColor::error() << EC.message() << '\n';
224     return nullptr;
225   }
226 
227   return Out;
228 }
229 
230 static std::string DwarfDebugFlags;
231 static void setDwarfDebugFlags(int argc, char **argv) {
232   if (!getenv("RC_DEBUG_OPTIONS"))
233     return;
234   for (int i = 0; i < argc; i++) {
235     DwarfDebugFlags += argv[i];
236     if (i + 1 < argc)
237       DwarfDebugFlags += " ";
238   }
239 }
240 
241 static std::string DwarfDebugProducer;
242 static void setDwarfDebugProducer() {
243   if(!getenv("DEBUG_PRODUCER"))
244     return;
245   DwarfDebugProducer += getenv("DEBUG_PRODUCER");
246 }
247 
248 static int AsLexInput(SourceMgr &SrcMgr, MCAsmInfo &MAI,
249                       raw_ostream &OS) {
250 
251   AsmLexer Lexer(MAI);
252   Lexer.setBuffer(SrcMgr.getMemoryBuffer(SrcMgr.getMainFileID())->getBuffer());
253 
254   bool Error = false;
255   while (Lexer.Lex().isNot(AsmToken::Eof)) {
256     Lexer.getTok().dump(OS);
257     OS << "\n";
258     if (Lexer.getTok().getKind() == AsmToken::Error)
259       Error = true;
260   }
261 
262   return Error;
263 }
264 
265 static int fillCommandLineSymbols(MCAsmParser &Parser) {
266   for (auto &I: DefineSymbol) {
267     auto Pair = StringRef(I).split('=');
268     auto Sym = Pair.first;
269     auto Val = Pair.second;
270 
271     if (Sym.empty() || Val.empty()) {
272       WithColor::error() << "defsym must be of the form: sym=value: " << I
273                          << "\n";
274       return 1;
275     }
276     int64_t Value;
277     if (Val.getAsInteger(0, Value)) {
278       WithColor::error() << "value is not an integer: " << Val << "\n";
279       return 1;
280     }
281     Parser.getContext().setSymbolValue(Parser.getStreamer(), Sym, Value);
282   }
283   return 0;
284 }
285 
286 static int AssembleInput(const char *ProgName, const Target *TheTarget,
287                          SourceMgr &SrcMgr, MCContext &Ctx, MCStreamer &Str,
288                          MCAsmInfo &MAI, MCSubtargetInfo &STI,
289                          MCInstrInfo &MCII, MCTargetOptions const &MCOptions) {
290   std::unique_ptr<MCAsmParser> Parser(
291       createMCAsmParser(SrcMgr, Ctx, Str, MAI));
292   std::unique_ptr<MCTargetAsmParser> TAP(
293       TheTarget->createMCAsmParser(STI, *Parser, MCII, MCOptions));
294 
295   if (!TAP) {
296     WithColor::error(errs(), ProgName)
297         << "this target does not support assembly parsing.\n";
298     return 1;
299   }
300 
301   int SymbolResult = fillCommandLineSymbols(*Parser);
302   if(SymbolResult)
303     return SymbolResult;
304   Parser->setShowParsedOperands(ShowInstOperands);
305   Parser->setTargetParser(*TAP);
306   Parser->getLexer().setLexMasmIntegers(LexMasmIntegers);
307   Parser->getLexer().setLexMasmHexFloats(LexMasmHexFloats);
308 
309   int Res = Parser->Run(NoInitialTextSection);
310 
311   return Res;
312 }
313 
314 int main(int argc, char **argv) {
315   InitLLVM X(argc, argv);
316 
317   // Initialize targets and assembly printers/parsers.
318   llvm::InitializeAllTargetInfos();
319   llvm::InitializeAllTargetMCs();
320   llvm::InitializeAllAsmParsers();
321   llvm::InitializeAllDisassemblers();
322 
323   // Register the target printer for --version.
324   cl::AddExtraVersionPrinter(TargetRegistry::printRegisteredTargetsForVersion);
325 
326   cl::ParseCommandLineOptions(argc, argv, "llvm machine code playground\n");
327   const MCTargetOptions MCOptions = mc::InitMCTargetOptionsFromFlags();
328   setDwarfDebugFlags(argc, argv);
329 
330   setDwarfDebugProducer();
331 
332   const char *ProgName = argv[0];
333   const Target *TheTarget = GetTarget(ProgName);
334   if (!TheTarget)
335     return 1;
336   // Now that GetTarget() has (potentially) replaced TripleName, it's safe to
337   // construct the Triple object.
338   Triple TheTriple(TripleName);
339 
340   ErrorOr<std::unique_ptr<MemoryBuffer>> BufferPtr =
341       MemoryBuffer::getFileOrSTDIN(InputFilename);
342   if (std::error_code EC = BufferPtr.getError()) {
343     WithColor::error(errs(), ProgName)
344         << InputFilename << ": " << EC.message() << '\n';
345     return 1;
346   }
347   MemoryBuffer *Buffer = BufferPtr->get();
348 
349   SourceMgr SrcMgr;
350 
351   // Tell SrcMgr about this buffer, which is what the parser will pick up.
352   SrcMgr.AddNewSourceBuffer(std::move(*BufferPtr), SMLoc());
353 
354   // Record the location of the include directories so that the lexer can find
355   // it later.
356   SrcMgr.setIncludeDirs(IncludeDirs);
357 
358   std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TripleName));
359   assert(MRI && "Unable to create target register info!");
360 
361   std::unique_ptr<MCAsmInfo> MAI(
362       TheTarget->createMCAsmInfo(*MRI, TripleName, MCOptions));
363   assert(MAI && "Unable to create target asm info!");
364 
365   MAI->setRelaxELFRelocations(RelaxELFRel);
366 
367   if (CompressDebugSections != DebugCompressionType::None) {
368     if (!zlib::isAvailable()) {
369       WithColor::error(errs(), ProgName)
370           << "build tools with zlib to enable -compress-debug-sections";
371       return 1;
372     }
373     MAI->setCompressDebugSections(CompressDebugSections);
374   }
375   MAI->setPreserveAsmComments(PreserveComments);
376 
377   // FIXME: This is not pretty. MCContext has a ptr to MCObjectFileInfo and
378   // MCObjectFileInfo needs a MCContext reference in order to initialize itself.
379   MCObjectFileInfo MOFI;
380   MCContext Ctx(MAI.get(), MRI.get(), &MOFI, &SrcMgr, &MCOptions);
381   MOFI.InitMCObjectFileInfo(TheTriple, PIC, Ctx, LargeCodeModel);
382 
383   if (SaveTempLabels)
384     Ctx.setAllowTemporaryLabels(false);
385 
386   Ctx.setGenDwarfForAssembly(GenDwarfForAssembly);
387   // Default to 4 for dwarf version.
388   unsigned DwarfVersion = MCOptions.DwarfVersion ? MCOptions.DwarfVersion : 4;
389   if (DwarfVersion < 2 || DwarfVersion > 5) {
390     errs() << ProgName << ": Dwarf version " << DwarfVersion
391            << " is not supported." << '\n';
392     return 1;
393   }
394   Ctx.setDwarfVersion(DwarfVersion);
395   if (MCOptions.Dwarf64) {
396     // The 64-bit DWARF format was introduced in DWARFv3.
397     if (DwarfVersion < 3) {
398       errs() << ProgName
399              << ": the 64-bit DWARF format is not supported for DWARF versions "
400                 "prior to 3\n";
401       return 1;
402     }
403     // 32-bit targets don't support DWARF64, which requires 64-bit relocations.
404     if (MAI->getCodePointerSize() < 8) {
405       errs() << ProgName
406              << ": the 64-bit DWARF format is only supported for 64-bit "
407                 "targets\n";
408       return 1;
409     }
410     // If needsDwarfSectionOffsetDirective is true, we would eventually call
411     // MCStreamer::emitSymbolValue() with IsSectionRelative = true, but that
412     // is supported only for 4-byte long references.
413     if (MAI->needsDwarfSectionOffsetDirective()) {
414       errs() << ProgName << ": the 64-bit DWARF format is not supported for "
415              << TheTriple.normalize() << "\n";
416       return 1;
417     }
418     Ctx.setDwarfFormat(dwarf::DWARF64);
419   }
420   if (!DwarfDebugFlags.empty())
421     Ctx.setDwarfDebugFlags(StringRef(DwarfDebugFlags));
422   if (!DwarfDebugProducer.empty())
423     Ctx.setDwarfDebugProducer(StringRef(DwarfDebugProducer));
424   if (!DebugCompilationDir.empty())
425     Ctx.setCompilationDir(DebugCompilationDir);
426   else {
427     // If no compilation dir is set, try to use the current directory.
428     SmallString<128> CWD;
429     if (!sys::fs::current_path(CWD))
430       Ctx.setCompilationDir(CWD);
431   }
432   for (const auto &Arg : DebugPrefixMap) {
433     const auto &KV = StringRef(Arg).split('=');
434     Ctx.addDebugPrefixMapEntry(std::string(KV.first), std::string(KV.second));
435   }
436   if (!MainFileName.empty())
437     Ctx.setMainFileName(MainFileName);
438   if (GenDwarfForAssembly)
439     Ctx.setGenDwarfRootFile(InputFilename, Buffer->getBuffer());
440 
441   // Package up features to be passed to target/subtarget
442   std::string FeaturesStr;
443   if (MAttrs.size()) {
444     SubtargetFeatures Features;
445     for (unsigned i = 0; i != MAttrs.size(); ++i)
446       Features.AddFeature(MAttrs[i]);
447     FeaturesStr = Features.getString();
448   }
449 
450   sys::fs::OpenFlags Flags = (FileType == OFT_AssemblyFile) ? sys::fs::OF_Text
451                                                             : sys::fs::OF_None;
452   std::unique_ptr<ToolOutputFile> Out = GetOutputStream(OutputFilename, Flags);
453   if (!Out)
454     return 1;
455 
456   std::unique_ptr<ToolOutputFile> DwoOut;
457   if (!SplitDwarfFile.empty()) {
458     if (FileType != OFT_ObjectFile) {
459       WithColor::error() << "dwo output only supported with object files\n";
460       return 1;
461     }
462     DwoOut = GetOutputStream(SplitDwarfFile, sys::fs::OF_None);
463     if (!DwoOut)
464       return 1;
465   }
466 
467   std::unique_ptr<buffer_ostream> BOS;
468   raw_pwrite_stream *OS = &Out->os();
469   std::unique_ptr<MCStreamer> Str;
470 
471   std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
472   assert(MCII && "Unable to create instruction info!");
473 
474   std::unique_ptr<MCSubtargetInfo> STI(
475       TheTarget->createMCSubtargetInfo(TripleName, MCPU, FeaturesStr));
476   assert(STI && "Unable to create subtarget info!");
477 
478   MCInstPrinter *IP = nullptr;
479   if (FileType == OFT_AssemblyFile) {
480     IP = TheTarget->createMCInstPrinter(Triple(TripleName), OutputAsmVariant,
481                                         *MAI, *MCII, *MRI);
482 
483     if (!IP) {
484       WithColor::error()
485           << "unable to create instruction printer for target triple '"
486           << TheTriple.normalize() << "' with assembly variant "
487           << OutputAsmVariant << ".\n";
488       return 1;
489     }
490 
491     // Set the display preference for hex vs. decimal immediates.
492     IP->setPrintImmHex(PrintImmHex);
493 
494     // Set up the AsmStreamer.
495     std::unique_ptr<MCCodeEmitter> CE;
496     if (ShowEncoding)
497       CE.reset(TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx));
498 
499     std::unique_ptr<MCAsmBackend> MAB(
500         TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions));
501     auto FOut = std::make_unique<formatted_raw_ostream>(*OS);
502     Str.reset(
503         TheTarget->createAsmStreamer(Ctx, std::move(FOut), /*asmverbose*/ true,
504                                      /*useDwarfDirectory*/ true, IP,
505                                      std::move(CE), std::move(MAB), ShowInst));
506 
507   } else if (FileType == OFT_Null) {
508     Str.reset(TheTarget->createNullStreamer(Ctx));
509   } else {
510     assert(FileType == OFT_ObjectFile && "Invalid file type!");
511 
512     if (!Out->os().supportsSeeking()) {
513       BOS = std::make_unique<buffer_ostream>(Out->os());
514       OS = BOS.get();
515     }
516 
517     MCCodeEmitter *CE = TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx);
518     MCAsmBackend *MAB = TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions);
519     Str.reset(TheTarget->createMCObjectStreamer(
520         TheTriple, Ctx, std::unique_ptr<MCAsmBackend>(MAB),
521         DwoOut ? MAB->createDwoObjectWriter(*OS, DwoOut->os())
522                : MAB->createObjectWriter(*OS),
523         std::unique_ptr<MCCodeEmitter>(CE), *STI, MCOptions.MCRelaxAll,
524         MCOptions.MCIncrementalLinkerCompatible,
525         /*DWARFMustBeAtTheEnd*/ false));
526     if (NoExecStack)
527       Str->InitSections(true);
528   }
529 
530   // Use Assembler information for parsing.
531   Str->setUseAssemblerInfoForParsing(true);
532 
533   int Res = 1;
534   bool disassemble = false;
535   switch (Action) {
536   case AC_AsLex:
537     Res = AsLexInput(SrcMgr, *MAI, Out->os());
538     break;
539   case AC_Assemble:
540     Res = AssembleInput(ProgName, TheTarget, SrcMgr, Ctx, *Str, *MAI, *STI,
541                         *MCII, MCOptions);
542     break;
543   case AC_MDisassemble:
544     assert(IP && "Expected assembly output");
545     IP->setUseMarkup(true);
546     disassemble = true;
547     break;
548   case AC_Disassemble:
549     disassemble = true;
550     break;
551   }
552   if (disassemble)
553     Res = Disassembler::disassemble(*TheTarget, TripleName, *STI, *Str, *Buffer,
554                                     SrcMgr, Ctx, Out->os(), MCOptions);
555 
556   // Keep output if no errors.
557   if (Res == 0) {
558     Out->keep();
559     if (DwoOut)
560       DwoOut->keep();
561   }
562   return Res;
563 }
564