1 | //===- ELFAsmParser.cpp - ELF Assembly Parser -----------------------------===// |
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 | #include "llvm/ADT/StringExtras.h" |
10 | #include "llvm/ADT/StringRef.h" |
11 | #include "llvm/ADT/StringSwitch.h" |
12 | #include "llvm/BinaryFormat/ELF.h" |
13 | #include "llvm/MC/MCAsmInfo.h" |
14 | #include "llvm/MC/MCContext.h" |
15 | #include "llvm/MC/MCDirectives.h" |
16 | #include "llvm/MC/MCParser/MCAsmLexer.h" |
17 | #include "llvm/MC/MCParser/MCAsmParser.h" |
18 | #include "llvm/MC/MCParser/MCAsmParserExtension.h" |
19 | #include "llvm/MC/MCSectionELF.h" |
20 | #include "llvm/MC/MCStreamer.h" |
21 | #include "llvm/MC/MCSymbol.h" |
22 | #include "llvm/MC/MCSymbolELF.h" |
23 | #include "llvm/MC/SectionKind.h" |
24 | #include "llvm/Support/Casting.h" |
25 | #include "llvm/Support/MathExtras.h" |
26 | #include "llvm/Support/SMLoc.h" |
27 | #include <cassert> |
28 | #include <cstdint> |
29 | #include <utility> |
30 | |
31 | using namespace llvm; |
32 | |
33 | namespace { |
34 | |
35 | class ELFAsmParser : public MCAsmParserExtension { |
36 | template<bool (ELFAsmParser::*HandlerMethod)(StringRef, SMLoc)> |
37 | void addDirectiveHandler(StringRef Directive) { |
38 | MCAsmParser::ExtensionDirectiveHandler Handler = std::make_pair( |
39 | this, HandleDirective<ELFAsmParser, HandlerMethod>); |
40 | |
41 | getParser().addDirectiveHandler(Directive, Handler); |
42 | } |
43 | |
44 | bool ParseSectionSwitch(StringRef Section, unsigned Type, unsigned Flags, |
45 | SectionKind Kind); |
46 | |
47 | public: |
48 | ELFAsmParser() { BracketExpressionsSupported = true; } |
49 | |
50 | void Initialize(MCAsmParser &Parser) override { |
51 | // Call the base implementation. |
52 | this->MCAsmParserExtension::Initialize(Parser); |
53 | |
54 | addDirectiveHandler<&ELFAsmParser::ParseSectionDirectiveData>(Directive: ".data" ); |
55 | addDirectiveHandler<&ELFAsmParser::ParseSectionDirectiveText>(Directive: ".text" ); |
56 | addDirectiveHandler<&ELFAsmParser::ParseSectionDirectiveBSS>(Directive: ".bss" ); |
57 | addDirectiveHandler<&ELFAsmParser::ParseSectionDirectiveRoData>(Directive: ".rodata" ); |
58 | addDirectiveHandler<&ELFAsmParser::ParseSectionDirectiveTData>(Directive: ".tdata" ); |
59 | addDirectiveHandler<&ELFAsmParser::ParseSectionDirectiveTBSS>(Directive: ".tbss" ); |
60 | addDirectiveHandler< |
61 | &ELFAsmParser::ParseSectionDirectiveDataRel>(Directive: ".data.rel" ); |
62 | addDirectiveHandler< |
63 | &ELFAsmParser::ParseSectionDirectiveDataRelRo>(Directive: ".data.rel.ro" ); |
64 | addDirectiveHandler< |
65 | &ELFAsmParser::ParseSectionDirectiveEhFrame>(Directive: ".eh_frame" ); |
66 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveSection>(Directive: ".section" ); |
67 | addDirectiveHandler< |
68 | &ELFAsmParser::ParseDirectivePushSection>(Directive: ".pushsection" ); |
69 | addDirectiveHandler<&ELFAsmParser::ParseDirectivePopSection>(Directive: ".popsection" ); |
70 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveSize>(Directive: ".size" ); |
71 | addDirectiveHandler<&ELFAsmParser::ParseDirectivePrevious>(Directive: ".previous" ); |
72 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveType>(Directive: ".type" ); |
73 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveIdent>(Directive: ".ident" ); |
74 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveSymver>(Directive: ".symver" ); |
75 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveVersion>(Directive: ".version" ); |
76 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveWeakref>(Directive: ".weakref" ); |
77 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveSymbolAttribute>(Directive: ".weak" ); |
78 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveSymbolAttribute>(Directive: ".local" ); |
79 | addDirectiveHandler< |
80 | &ELFAsmParser::ParseDirectiveSymbolAttribute>(Directive: ".protected" ); |
81 | addDirectiveHandler< |
82 | &ELFAsmParser::ParseDirectiveSymbolAttribute>(Directive: ".internal" ); |
83 | addDirectiveHandler< |
84 | &ELFAsmParser::ParseDirectiveSymbolAttribute>(Directive: ".hidden" ); |
85 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveSubsection>(Directive: ".subsection" ); |
86 | addDirectiveHandler<&ELFAsmParser::ParseDirectiveCGProfile>(Directive: ".cg_profile" ); |
87 | } |
88 | |
89 | // FIXME: Part of this logic is duplicated in the MCELFStreamer. What is |
90 | // the best way for us to get access to it? |
91 | bool ParseSectionDirectiveData(StringRef, SMLoc) { |
92 | return ParseSectionSwitch(Section: ".data" , Type: ELF::SHT_PROGBITS, |
93 | Flags: ELF::SHF_WRITE | ELF::SHF_ALLOC, |
94 | Kind: SectionKind::getData()); |
95 | } |
96 | bool ParseSectionDirectiveText(StringRef, SMLoc) { |
97 | return ParseSectionSwitch(Section: ".text" , Type: ELF::SHT_PROGBITS, |
98 | Flags: ELF::SHF_EXECINSTR | |
99 | ELF::SHF_ALLOC, Kind: SectionKind::getText()); |
100 | } |
101 | bool ParseSectionDirectiveBSS(StringRef, SMLoc) { |
102 | return ParseSectionSwitch(Section: ".bss" , Type: ELF::SHT_NOBITS, |
103 | Flags: ELF::SHF_WRITE | |
104 | ELF::SHF_ALLOC, Kind: SectionKind::getBSS()); |
105 | } |
106 | bool ParseSectionDirectiveRoData(StringRef, SMLoc) { |
107 | return ParseSectionSwitch(Section: ".rodata" , Type: ELF::SHT_PROGBITS, |
108 | Flags: ELF::SHF_ALLOC, |
109 | Kind: SectionKind::getReadOnly()); |
110 | } |
111 | bool ParseSectionDirectiveTData(StringRef, SMLoc) { |
112 | return ParseSectionSwitch(Section: ".tdata" , Type: ELF::SHT_PROGBITS, |
113 | Flags: ELF::SHF_ALLOC | |
114 | ELF::SHF_TLS | ELF::SHF_WRITE, |
115 | Kind: SectionKind::getThreadData()); |
116 | } |
117 | bool ParseSectionDirectiveTBSS(StringRef, SMLoc) { |
118 | return ParseSectionSwitch(Section: ".tbss" , Type: ELF::SHT_NOBITS, |
119 | Flags: ELF::SHF_ALLOC | |
120 | ELF::SHF_TLS | ELF::SHF_WRITE, |
121 | Kind: SectionKind::getThreadBSS()); |
122 | } |
123 | bool ParseSectionDirectiveDataRel(StringRef, SMLoc) { |
124 | return ParseSectionSwitch(Section: ".data.rel" , Type: ELF::SHT_PROGBITS, |
125 | Flags: ELF::SHF_ALLOC | ELF::SHF_WRITE, |
126 | Kind: SectionKind::getData()); |
127 | } |
128 | bool ParseSectionDirectiveDataRelRo(StringRef, SMLoc) { |
129 | return ParseSectionSwitch(Section: ".data.rel.ro" , Type: ELF::SHT_PROGBITS, |
130 | Flags: ELF::SHF_ALLOC | |
131 | ELF::SHF_WRITE, |
132 | Kind: SectionKind::getReadOnlyWithRel()); |
133 | } |
134 | bool ParseSectionDirectiveEhFrame(StringRef, SMLoc) { |
135 | return ParseSectionSwitch(Section: ".eh_frame" , Type: ELF::SHT_PROGBITS, |
136 | Flags: ELF::SHF_ALLOC | ELF::SHF_WRITE, |
137 | Kind: SectionKind::getData()); |
138 | } |
139 | bool ParseDirectivePushSection(StringRef, SMLoc); |
140 | bool ParseDirectivePopSection(StringRef, SMLoc); |
141 | bool ParseDirectiveSection(StringRef, SMLoc); |
142 | bool ParseDirectiveSize(StringRef, SMLoc); |
143 | bool ParseDirectivePrevious(StringRef, SMLoc); |
144 | bool ParseDirectiveType(StringRef, SMLoc); |
145 | bool ParseDirectiveIdent(StringRef, SMLoc); |
146 | bool ParseDirectiveSymver(StringRef, SMLoc); |
147 | bool ParseDirectiveVersion(StringRef, SMLoc); |
148 | bool ParseDirectiveWeakref(StringRef, SMLoc); |
149 | bool ParseDirectiveSymbolAttribute(StringRef, SMLoc); |
150 | bool ParseDirectiveSubsection(StringRef, SMLoc); |
151 | bool ParseDirectiveCGProfile(StringRef, SMLoc); |
152 | |
153 | private: |
154 | bool ParseSectionName(StringRef &SectionName); |
155 | bool ParseSectionArguments(bool IsPush, SMLoc loc); |
156 | unsigned parseSunStyleSectionFlags(); |
157 | bool maybeParseSectionType(StringRef &TypeName); |
158 | bool parseMergeSize(int64_t &Size); |
159 | bool parseGroup(StringRef &GroupName, bool &IsComdat); |
160 | bool parseLinkedToSym(MCSymbolELF *&LinkedToSym); |
161 | bool maybeParseUniqueID(int64_t &UniqueID); |
162 | }; |
163 | |
164 | } // end anonymous namespace |
165 | |
166 | /// ParseDirectiveSymbolAttribute |
167 | /// ::= { ".local", ".weak", ... } [ identifier ( , identifier )* ] |
168 | bool ELFAsmParser::ParseDirectiveSymbolAttribute(StringRef Directive, SMLoc) { |
169 | MCSymbolAttr Attr = StringSwitch<MCSymbolAttr>(Directive) |
170 | .Case(S: ".weak" , Value: MCSA_Weak) |
171 | .Case(S: ".local" , Value: MCSA_Local) |
172 | .Case(S: ".hidden" , Value: MCSA_Hidden) |
173 | .Case(S: ".internal" , Value: MCSA_Internal) |
174 | .Case(S: ".protected" , Value: MCSA_Protected) |
175 | .Default(Value: MCSA_Invalid); |
176 | assert(Attr != MCSA_Invalid && "unexpected symbol attribute directive!" ); |
177 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) { |
178 | while (true) { |
179 | StringRef Name; |
180 | |
181 | if (getParser().parseIdentifier(Res&: Name)) |
182 | return TokError(Msg: "expected identifier" ); |
183 | |
184 | if (getParser().discardLTOSymbol(Name)) { |
185 | if (getLexer().is(K: AsmToken::EndOfStatement)) |
186 | break; |
187 | continue; |
188 | } |
189 | |
190 | MCSymbol *Sym = getContext().getOrCreateSymbol(Name); |
191 | |
192 | getStreamer().emitSymbolAttribute(Symbol: Sym, Attribute: Attr); |
193 | |
194 | if (getLexer().is(K: AsmToken::EndOfStatement)) |
195 | break; |
196 | |
197 | if (getLexer().isNot(K: AsmToken::Comma)) |
198 | return TokError(Msg: "expected comma" ); |
199 | Lex(); |
200 | } |
201 | } |
202 | |
203 | Lex(); |
204 | return false; |
205 | } |
206 | |
207 | bool ELFAsmParser::ParseSectionSwitch(StringRef Section, unsigned Type, |
208 | unsigned Flags, SectionKind Kind) { |
209 | const MCExpr *Subsection = nullptr; |
210 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) { |
211 | if (getParser().parseExpression(Res&: Subsection)) |
212 | return true; |
213 | } |
214 | Lex(); |
215 | |
216 | getStreamer().switchSection(Section: getContext().getELFSection(Section, Type, Flags), |
217 | Subsection); |
218 | |
219 | return false; |
220 | } |
221 | |
222 | bool ELFAsmParser::ParseDirectiveSize(StringRef, SMLoc) { |
223 | StringRef Name; |
224 | if (getParser().parseIdentifier(Res&: Name)) |
225 | return TokError(Msg: "expected identifier" ); |
226 | MCSymbolELF *Sym = cast<MCSymbolELF>(Val: getContext().getOrCreateSymbol(Name)); |
227 | |
228 | if (getLexer().isNot(K: AsmToken::Comma)) |
229 | return TokError(Msg: "expected comma" ); |
230 | Lex(); |
231 | |
232 | const MCExpr *Expr; |
233 | if (getParser().parseExpression(Res&: Expr)) |
234 | return true; |
235 | |
236 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) |
237 | return TokError(Msg: "unexpected token" ); |
238 | Lex(); |
239 | |
240 | getStreamer().emitELFSize(Symbol: Sym, Value: Expr); |
241 | return false; |
242 | } |
243 | |
244 | bool ELFAsmParser::ParseSectionName(StringRef &SectionName) { |
245 | // A section name can contain -, so we cannot just use |
246 | // parseIdentifier. |
247 | SMLoc FirstLoc = getLexer().getLoc(); |
248 | unsigned Size = 0; |
249 | |
250 | if (getLexer().is(K: AsmToken::String)) { |
251 | SectionName = getTok().getIdentifier(); |
252 | Lex(); |
253 | return false; |
254 | } |
255 | |
256 | while (!getParser().hasPendingError()) { |
257 | SMLoc PrevLoc = getLexer().getLoc(); |
258 | if (getLexer().is(K: AsmToken::Comma) || |
259 | getLexer().is(K: AsmToken::EndOfStatement)) |
260 | break; |
261 | |
262 | unsigned CurSize; |
263 | if (getLexer().is(K: AsmToken::String)) { |
264 | CurSize = getTok().getIdentifier().size() + 2; |
265 | Lex(); |
266 | } else if (getLexer().is(K: AsmToken::Identifier)) { |
267 | CurSize = getTok().getIdentifier().size(); |
268 | Lex(); |
269 | } else { |
270 | CurSize = getTok().getString().size(); |
271 | Lex(); |
272 | } |
273 | Size += CurSize; |
274 | SectionName = StringRef(FirstLoc.getPointer(), Size); |
275 | |
276 | // Make sure the following token is adjacent. |
277 | if (PrevLoc.getPointer() + CurSize != getTok().getLoc().getPointer()) |
278 | break; |
279 | } |
280 | if (Size == 0) |
281 | return true; |
282 | |
283 | return false; |
284 | } |
285 | |
286 | static unsigned parseSectionFlags(const Triple &TT, StringRef flagsStr, |
287 | bool *UseLastGroup) { |
288 | unsigned flags = 0; |
289 | |
290 | // If a valid numerical value is set for the section flag, use it verbatim |
291 | if (!flagsStr.getAsInteger(Radix: 0, Result&: flags)) |
292 | return flags; |
293 | |
294 | for (char i : flagsStr) { |
295 | switch (i) { |
296 | case 'a': |
297 | flags |= ELF::SHF_ALLOC; |
298 | break; |
299 | case 'e': |
300 | flags |= ELF::SHF_EXCLUDE; |
301 | break; |
302 | case 'x': |
303 | flags |= ELF::SHF_EXECINSTR; |
304 | break; |
305 | case 'w': |
306 | flags |= ELF::SHF_WRITE; |
307 | break; |
308 | case 'o': |
309 | flags |= ELF::SHF_LINK_ORDER; |
310 | break; |
311 | case 'M': |
312 | flags |= ELF::SHF_MERGE; |
313 | break; |
314 | case 'S': |
315 | flags |= ELF::SHF_STRINGS; |
316 | break; |
317 | case 'T': |
318 | flags |= ELF::SHF_TLS; |
319 | break; |
320 | case 'c': |
321 | if (TT.getArch() != Triple::xcore) |
322 | return -1U; |
323 | flags |= ELF::XCORE_SHF_CP_SECTION; |
324 | break; |
325 | case 'd': |
326 | if (TT.getArch() != Triple::xcore) |
327 | return -1U; |
328 | flags |= ELF::XCORE_SHF_DP_SECTION; |
329 | break; |
330 | case 'y': |
331 | if (!(TT.isARM() || TT.isThumb())) |
332 | return -1U; |
333 | flags |= ELF::SHF_ARM_PURECODE; |
334 | break; |
335 | case 's': |
336 | if (TT.getArch() != Triple::hexagon) |
337 | return -1U; |
338 | flags |= ELF::SHF_HEX_GPREL; |
339 | break; |
340 | case 'G': |
341 | flags |= ELF::SHF_GROUP; |
342 | break; |
343 | case 'l': |
344 | if (TT.getArch() != Triple::x86_64) |
345 | return -1U; |
346 | flags |= ELF::SHF_X86_64_LARGE; |
347 | break; |
348 | case 'R': |
349 | if (TT.isOSSolaris()) |
350 | flags |= ELF::SHF_SUNW_NODISCARD; |
351 | else |
352 | flags |= ELF::SHF_GNU_RETAIN; |
353 | break; |
354 | case '?': |
355 | *UseLastGroup = true; |
356 | break; |
357 | default: |
358 | return -1U; |
359 | } |
360 | } |
361 | |
362 | return flags; |
363 | } |
364 | |
365 | unsigned ELFAsmParser::parseSunStyleSectionFlags() { |
366 | unsigned flags = 0; |
367 | while (getLexer().is(K: AsmToken::Hash)) { |
368 | Lex(); // Eat the #. |
369 | |
370 | if (!getLexer().is(K: AsmToken::Identifier)) |
371 | return -1U; |
372 | |
373 | StringRef flagId = getTok().getIdentifier(); |
374 | if (flagId == "alloc" ) |
375 | flags |= ELF::SHF_ALLOC; |
376 | else if (flagId == "execinstr" ) |
377 | flags |= ELF::SHF_EXECINSTR; |
378 | else if (flagId == "write" ) |
379 | flags |= ELF::SHF_WRITE; |
380 | else if (flagId == "tls" ) |
381 | flags |= ELF::SHF_TLS; |
382 | else |
383 | return -1U; |
384 | |
385 | Lex(); // Eat the flag. |
386 | |
387 | if (!getLexer().is(K: AsmToken::Comma)) |
388 | break; |
389 | Lex(); // Eat the comma. |
390 | } |
391 | return flags; |
392 | } |
393 | |
394 | |
395 | bool ELFAsmParser::ParseDirectivePushSection(StringRef s, SMLoc loc) { |
396 | getStreamer().pushSection(); |
397 | |
398 | if (ParseSectionArguments(/*IsPush=*/true, loc)) { |
399 | getStreamer().popSection(); |
400 | return true; |
401 | } |
402 | |
403 | return false; |
404 | } |
405 | |
406 | bool ELFAsmParser::ParseDirectivePopSection(StringRef, SMLoc) { |
407 | if (!getStreamer().popSection()) |
408 | return TokError(Msg: ".popsection without corresponding .pushsection" ); |
409 | return false; |
410 | } |
411 | |
412 | bool ELFAsmParser::ParseDirectiveSection(StringRef, SMLoc loc) { |
413 | return ParseSectionArguments(/*IsPush=*/false, loc); |
414 | } |
415 | |
416 | bool ELFAsmParser::maybeParseSectionType(StringRef &TypeName) { |
417 | MCAsmLexer &L = getLexer(); |
418 | if (L.isNot(K: AsmToken::Comma)) |
419 | return false; |
420 | Lex(); |
421 | if (L.isNot(K: AsmToken::At) && L.isNot(K: AsmToken::Percent) && |
422 | L.isNot(K: AsmToken::String)) { |
423 | if (L.getAllowAtInIdentifier()) |
424 | return TokError(Msg: "expected '@<type>', '%<type>' or \"<type>\"" ); |
425 | else |
426 | return TokError(Msg: "expected '%<type>' or \"<type>\"" ); |
427 | } |
428 | if (!L.is(K: AsmToken::String)) |
429 | Lex(); |
430 | if (L.is(K: AsmToken::Integer)) { |
431 | TypeName = getTok().getString(); |
432 | Lex(); |
433 | } else if (getParser().parseIdentifier(Res&: TypeName)) |
434 | return TokError(Msg: "expected identifier" ); |
435 | return false; |
436 | } |
437 | |
438 | bool ELFAsmParser::parseMergeSize(int64_t &Size) { |
439 | if (getLexer().isNot(K: AsmToken::Comma)) |
440 | return TokError(Msg: "expected the entry size" ); |
441 | Lex(); |
442 | if (getParser().parseAbsoluteExpression(Res&: Size)) |
443 | return true; |
444 | if (Size <= 0) |
445 | return TokError(Msg: "entry size must be positive" ); |
446 | return false; |
447 | } |
448 | |
449 | bool ELFAsmParser::parseGroup(StringRef &GroupName, bool &IsComdat) { |
450 | MCAsmLexer &L = getLexer(); |
451 | if (L.isNot(K: AsmToken::Comma)) |
452 | return TokError(Msg: "expected group name" ); |
453 | Lex(); |
454 | if (L.is(K: AsmToken::Integer)) { |
455 | GroupName = getTok().getString(); |
456 | Lex(); |
457 | } else if (getParser().parseIdentifier(Res&: GroupName)) { |
458 | return TokError(Msg: "invalid group name" ); |
459 | } |
460 | if (L.is(K: AsmToken::Comma)) { |
461 | Lex(); |
462 | StringRef Linkage; |
463 | if (getParser().parseIdentifier(Res&: Linkage)) |
464 | return TokError(Msg: "invalid linkage" ); |
465 | if (Linkage != "comdat" ) |
466 | return TokError(Msg: "Linkage must be 'comdat'" ); |
467 | IsComdat = true; |
468 | } else { |
469 | IsComdat = false; |
470 | } |
471 | return false; |
472 | } |
473 | |
474 | bool ELFAsmParser::parseLinkedToSym(MCSymbolELF *&LinkedToSym) { |
475 | MCAsmLexer &L = getLexer(); |
476 | if (L.isNot(K: AsmToken::Comma)) |
477 | return TokError(Msg: "expected linked-to symbol" ); |
478 | Lex(); |
479 | StringRef Name; |
480 | SMLoc StartLoc = L.getLoc(); |
481 | if (getParser().parseIdentifier(Res&: Name)) { |
482 | if (getParser().getTok().getString() == "0" ) { |
483 | getParser().Lex(); |
484 | LinkedToSym = nullptr; |
485 | return false; |
486 | } |
487 | return TokError(Msg: "invalid linked-to symbol" ); |
488 | } |
489 | LinkedToSym = dyn_cast_or_null<MCSymbolELF>(Val: getContext().lookupSymbol(Name)); |
490 | if (!LinkedToSym || !LinkedToSym->isInSection()) |
491 | return Error(L: StartLoc, Msg: "linked-to symbol is not in a section: " + Name); |
492 | return false; |
493 | } |
494 | |
495 | bool ELFAsmParser::maybeParseUniqueID(int64_t &UniqueID) { |
496 | MCAsmLexer &L = getLexer(); |
497 | if (L.isNot(K: AsmToken::Comma)) |
498 | return false; |
499 | Lex(); |
500 | StringRef UniqueStr; |
501 | if (getParser().parseIdentifier(Res&: UniqueStr)) |
502 | return TokError(Msg: "expected identifier" ); |
503 | if (UniqueStr != "unique" ) |
504 | return TokError(Msg: "expected 'unique'" ); |
505 | if (L.isNot(K: AsmToken::Comma)) |
506 | return TokError(Msg: "expected commma" ); |
507 | Lex(); |
508 | if (getParser().parseAbsoluteExpression(Res&: UniqueID)) |
509 | return true; |
510 | if (UniqueID < 0) |
511 | return TokError(Msg: "unique id must be positive" ); |
512 | if (!isUInt<32>(x: UniqueID) || UniqueID == ~0U) |
513 | return TokError(Msg: "unique id is too large" ); |
514 | return false; |
515 | } |
516 | |
517 | static bool hasPrefix(StringRef SectionName, StringRef Prefix) { |
518 | return SectionName.consume_front(Prefix) && |
519 | (SectionName.empty() || SectionName[0] == '.'); |
520 | } |
521 | |
522 | static bool allowSectionTypeMismatch(const Triple &TT, StringRef SectionName, |
523 | unsigned Type) { |
524 | if (TT.getArch() == Triple::x86_64) { |
525 | // x86-64 psABI names SHT_X86_64_UNWIND as the canonical type for .eh_frame, |
526 | // but GNU as emits SHT_PROGBITS .eh_frame for .cfi_* directives. Don't |
527 | // error for SHT_PROGBITS .eh_frame |
528 | return SectionName == ".eh_frame" && Type == ELF::SHT_PROGBITS; |
529 | } |
530 | if (TT.isMIPS()) { |
531 | // MIPS .debug_* sections should have SHT_MIPS_DWARF section type to |
532 | // distinguish among sections contain DWARF and ECOFF debug formats, |
533 | // but in assembly files these sections have SHT_PROGBITS type. |
534 | return SectionName.starts_with(Prefix: ".debug_" ) && Type == ELF::SHT_PROGBITS; |
535 | } |
536 | return false; |
537 | } |
538 | |
539 | bool ELFAsmParser::ParseSectionArguments(bool IsPush, SMLoc loc) { |
540 | StringRef SectionName; |
541 | |
542 | if (ParseSectionName(SectionName)) |
543 | return TokError(Msg: "expected identifier" ); |
544 | |
545 | StringRef TypeName; |
546 | int64_t Size = 0; |
547 | StringRef GroupName; |
548 | bool IsComdat = false; |
549 | unsigned Flags = 0; |
550 | unsigned = 0; |
551 | const MCExpr *Subsection = nullptr; |
552 | bool UseLastGroup = false; |
553 | MCSymbolELF *LinkedToSym = nullptr; |
554 | int64_t UniqueID = ~0; |
555 | |
556 | // Set the defaults first. |
557 | if (hasPrefix(SectionName, Prefix: ".rodata" ) || SectionName == ".rodata1" ) |
558 | Flags |= ELF::SHF_ALLOC; |
559 | else if (SectionName == ".fini" || SectionName == ".init" || |
560 | hasPrefix(SectionName, Prefix: ".text" )) |
561 | Flags |= ELF::SHF_ALLOC | ELF::SHF_EXECINSTR; |
562 | else if (hasPrefix(SectionName, Prefix: ".data" ) || SectionName == ".data1" || |
563 | hasPrefix(SectionName, Prefix: ".bss" ) || |
564 | hasPrefix(SectionName, Prefix: ".init_array" ) || |
565 | hasPrefix(SectionName, Prefix: ".fini_array" ) || |
566 | hasPrefix(SectionName, Prefix: ".preinit_array" )) |
567 | Flags |= ELF::SHF_ALLOC | ELF::SHF_WRITE; |
568 | else if (hasPrefix(SectionName, Prefix: ".tdata" ) || hasPrefix(SectionName, Prefix: ".tbss" )) |
569 | Flags |= ELF::SHF_ALLOC | ELF::SHF_WRITE | ELF::SHF_TLS; |
570 | |
571 | if (getLexer().is(K: AsmToken::Comma)) { |
572 | Lex(); |
573 | |
574 | if (IsPush && getLexer().isNot(K: AsmToken::String)) { |
575 | if (getParser().parseExpression(Res&: Subsection)) |
576 | return true; |
577 | if (getLexer().isNot(K: AsmToken::Comma)) |
578 | goto EndStmt; |
579 | Lex(); |
580 | } |
581 | |
582 | if (getLexer().isNot(K: AsmToken::String)) { |
583 | if (getLexer().isNot(K: AsmToken::Hash)) |
584 | return TokError(Msg: "expected string" ); |
585 | extraFlags = parseSunStyleSectionFlags(); |
586 | } else { |
587 | StringRef FlagsStr = getTok().getStringContents(); |
588 | Lex(); |
589 | extraFlags = parseSectionFlags(TT: getContext().getTargetTriple(), flagsStr: FlagsStr, |
590 | UseLastGroup: &UseLastGroup); |
591 | } |
592 | |
593 | if (extraFlags == -1U) |
594 | return TokError(Msg: "unknown flag" ); |
595 | Flags |= extraFlags; |
596 | |
597 | bool Mergeable = Flags & ELF::SHF_MERGE; |
598 | bool Group = Flags & ELF::SHF_GROUP; |
599 | if (Group && UseLastGroup) |
600 | return TokError(Msg: "Section cannot specifiy a group name while also acting " |
601 | "as a member of the last group" ); |
602 | |
603 | if (maybeParseSectionType(TypeName)) |
604 | return true; |
605 | |
606 | MCAsmLexer &L = getLexer(); |
607 | if (TypeName.empty()) { |
608 | if (Mergeable) |
609 | return TokError(Msg: "Mergeable section must specify the type" ); |
610 | if (Group) |
611 | return TokError(Msg: "Group section must specify the type" ); |
612 | if (L.isNot(K: AsmToken::EndOfStatement)) |
613 | return TokError(Msg: "expected end of directive" ); |
614 | } |
615 | |
616 | if (Mergeable) |
617 | if (parseMergeSize(Size)) |
618 | return true; |
619 | if (Flags & ELF::SHF_LINK_ORDER) |
620 | if (parseLinkedToSym(LinkedToSym)) |
621 | return true; |
622 | if (Group) |
623 | if (parseGroup(GroupName, IsComdat)) |
624 | return true; |
625 | if (maybeParseUniqueID(UniqueID)) |
626 | return true; |
627 | } |
628 | |
629 | EndStmt: |
630 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) |
631 | return TokError(Msg: "expected end of directive" ); |
632 | Lex(); |
633 | |
634 | unsigned Type = ELF::SHT_PROGBITS; |
635 | |
636 | if (TypeName.empty()) { |
637 | if (SectionName.starts_with(Prefix: ".note" )) |
638 | Type = ELF::SHT_NOTE; |
639 | else if (hasPrefix(SectionName, Prefix: ".init_array" )) |
640 | Type = ELF::SHT_INIT_ARRAY; |
641 | else if (hasPrefix(SectionName, Prefix: ".bss" )) |
642 | Type = ELF::SHT_NOBITS; |
643 | else if (hasPrefix(SectionName, Prefix: ".tbss" )) |
644 | Type = ELF::SHT_NOBITS; |
645 | else if (hasPrefix(SectionName, Prefix: ".fini_array" )) |
646 | Type = ELF::SHT_FINI_ARRAY; |
647 | else if (hasPrefix(SectionName, Prefix: ".preinit_array" )) |
648 | Type = ELF::SHT_PREINIT_ARRAY; |
649 | } else { |
650 | if (TypeName == "init_array" ) |
651 | Type = ELF::SHT_INIT_ARRAY; |
652 | else if (TypeName == "fini_array" ) |
653 | Type = ELF::SHT_FINI_ARRAY; |
654 | else if (TypeName == "preinit_array" ) |
655 | Type = ELF::SHT_PREINIT_ARRAY; |
656 | else if (TypeName == "nobits" ) |
657 | Type = ELF::SHT_NOBITS; |
658 | else if (TypeName == "progbits" ) |
659 | Type = ELF::SHT_PROGBITS; |
660 | else if (TypeName == "note" ) |
661 | Type = ELF::SHT_NOTE; |
662 | else if (TypeName == "unwind" ) |
663 | Type = ELF::SHT_X86_64_UNWIND; |
664 | else if (TypeName == "llvm_odrtab" ) |
665 | Type = ELF::SHT_LLVM_ODRTAB; |
666 | else if (TypeName == "llvm_linker_options" ) |
667 | Type = ELF::SHT_LLVM_LINKER_OPTIONS; |
668 | else if (TypeName == "llvm_call_graph_profile" ) |
669 | Type = ELF::SHT_LLVM_CALL_GRAPH_PROFILE; |
670 | else if (TypeName == "llvm_dependent_libraries" ) |
671 | Type = ELF::SHT_LLVM_DEPENDENT_LIBRARIES; |
672 | else if (TypeName == "llvm_sympart" ) |
673 | Type = ELF::SHT_LLVM_SYMPART; |
674 | else if (TypeName == "llvm_bb_addr_map" ) |
675 | Type = ELF::SHT_LLVM_BB_ADDR_MAP; |
676 | else if (TypeName == "llvm_offloading" ) |
677 | Type = ELF::SHT_LLVM_OFFLOADING; |
678 | else if (TypeName == "llvm_lto" ) |
679 | Type = ELF::SHT_LLVM_LTO; |
680 | else if (TypeName.getAsInteger(Radix: 0, Result&: Type)) |
681 | return TokError(Msg: "unknown section type" ); |
682 | } |
683 | |
684 | if (UseLastGroup) { |
685 | if (const MCSectionELF *Section = |
686 | cast_or_null<MCSectionELF>(Val: getStreamer().getCurrentSectionOnly())) |
687 | if (const MCSymbol *Group = Section->getGroup()) { |
688 | GroupName = Group->getName(); |
689 | IsComdat = Section->isComdat(); |
690 | Flags |= ELF::SHF_GROUP; |
691 | } |
692 | } |
693 | |
694 | MCSectionELF *Section = |
695 | getContext().getELFSection(Section: SectionName, Type, Flags, EntrySize: Size, Group: GroupName, |
696 | IsComdat, UniqueID, LinkedToSym); |
697 | getStreamer().switchSection(Section, Subsection); |
698 | // Check that flags are used consistently. However, the GNU assembler permits |
699 | // to leave out in subsequent uses of the same sections; for compatibility, |
700 | // do likewise. |
701 | if (!TypeName.empty() && Section->getType() != Type && |
702 | !allowSectionTypeMismatch(TT: getContext().getTargetTriple(), SectionName, |
703 | Type)) |
704 | Error(L: loc, Msg: "changed section type for " + SectionName + ", expected: 0x" + |
705 | utohexstr(X: Section->getType())); |
706 | if ((extraFlags || Size || !TypeName.empty()) && Section->getFlags() != Flags) |
707 | Error(L: loc, Msg: "changed section flags for " + SectionName + ", expected: 0x" + |
708 | utohexstr(X: Section->getFlags())); |
709 | if ((extraFlags || Size || !TypeName.empty()) && |
710 | Section->getEntrySize() != Size) |
711 | Error(L: loc, Msg: "changed section entsize for " + SectionName + |
712 | ", expected: " + Twine(Section->getEntrySize())); |
713 | |
714 | if (getContext().getGenDwarfForAssembly() && |
715 | (Section->getFlags() & ELF::SHF_ALLOC) && |
716 | (Section->getFlags() & ELF::SHF_EXECINSTR)) { |
717 | bool InsertResult = getContext().addGenDwarfSection(Sec: Section); |
718 | if (InsertResult && getContext().getDwarfVersion() <= 2) |
719 | Warning(L: loc, Msg: "DWARF2 only supports one section per compilation unit" ); |
720 | } |
721 | |
722 | return false; |
723 | } |
724 | |
725 | bool ELFAsmParser::ParseDirectivePrevious(StringRef DirName, SMLoc) { |
726 | MCSectionSubPair PreviousSection = getStreamer().getPreviousSection(); |
727 | if (PreviousSection.first == nullptr) |
728 | return TokError(Msg: ".previous without corresponding .section" ); |
729 | getStreamer().switchSection(Section: PreviousSection.first, Subsec: PreviousSection.second); |
730 | |
731 | return false; |
732 | } |
733 | |
734 | static MCSymbolAttr MCAttrForString(StringRef Type) { |
735 | return StringSwitch<MCSymbolAttr>(Type) |
736 | .Cases(S0: "STT_FUNC" , S1: "function" , Value: MCSA_ELF_TypeFunction) |
737 | .Cases(S0: "STT_OBJECT" , S1: "object" , Value: MCSA_ELF_TypeObject) |
738 | .Cases(S0: "STT_TLS" , S1: "tls_object" , Value: MCSA_ELF_TypeTLS) |
739 | .Cases(S0: "STT_COMMON" , S1: "common" , Value: MCSA_ELF_TypeCommon) |
740 | .Cases(S0: "STT_NOTYPE" , S1: "notype" , Value: MCSA_ELF_TypeNoType) |
741 | .Cases(S0: "STT_GNU_IFUNC" , S1: "gnu_indirect_function" , |
742 | Value: MCSA_ELF_TypeIndFunction) |
743 | .Case(S: "gnu_unique_object" , Value: MCSA_ELF_TypeGnuUniqueObject) |
744 | .Default(Value: MCSA_Invalid); |
745 | } |
746 | |
747 | /// ParseDirectiveELFType |
748 | /// ::= .type identifier , STT_<TYPE_IN_UPPER_CASE> |
749 | /// ::= .type identifier , #attribute |
750 | /// ::= .type identifier , @attribute |
751 | /// ::= .type identifier , %attribute |
752 | /// ::= .type identifier , "attribute" |
753 | bool ELFAsmParser::ParseDirectiveType(StringRef, SMLoc) { |
754 | StringRef Name; |
755 | if (getParser().parseIdentifier(Res&: Name)) |
756 | return TokError(Msg: "expected identifier" ); |
757 | |
758 | // Handle the identifier as the key symbol. |
759 | MCSymbol *Sym = getContext().getOrCreateSymbol(Name); |
760 | |
761 | // NOTE the comma is optional in all cases. It is only documented as being |
762 | // optional in the first case, however, GAS will silently treat the comma as |
763 | // optional in all cases. Furthermore, although the documentation states that |
764 | // the first form only accepts STT_<TYPE_IN_UPPER_CASE>, in reality, GAS |
765 | // accepts both the upper case name as well as the lower case aliases. |
766 | if (getLexer().is(K: AsmToken::Comma)) |
767 | Lex(); |
768 | |
769 | if (getLexer().isNot(K: AsmToken::Identifier) && |
770 | getLexer().isNot(K: AsmToken::Hash) && |
771 | getLexer().isNot(K: AsmToken::Percent) && |
772 | getLexer().isNot(K: AsmToken::String)) { |
773 | if (!getLexer().getAllowAtInIdentifier()) |
774 | return TokError(Msg: "expected STT_<TYPE_IN_UPPER_CASE>, '#<type>', " |
775 | "'%<type>' or \"<type>\"" ); |
776 | else if (getLexer().isNot(K: AsmToken::At)) |
777 | return TokError(Msg: "expected STT_<TYPE_IN_UPPER_CASE>, '#<type>', '@<type>', " |
778 | "'%<type>' or \"<type>\"" ); |
779 | } |
780 | |
781 | if (getLexer().isNot(K: AsmToken::String) && |
782 | getLexer().isNot(K: AsmToken::Identifier)) |
783 | Lex(); |
784 | |
785 | SMLoc TypeLoc = getLexer().getLoc(); |
786 | |
787 | StringRef Type; |
788 | if (getParser().parseIdentifier(Res&: Type)) |
789 | return TokError(Msg: "expected symbol type" ); |
790 | |
791 | MCSymbolAttr Attr = MCAttrForString(Type); |
792 | if (Attr == MCSA_Invalid) |
793 | return Error(L: TypeLoc, Msg: "unsupported attribute" ); |
794 | |
795 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) |
796 | return TokError(Msg: "expected end of directive" ); |
797 | Lex(); |
798 | |
799 | getStreamer().emitSymbolAttribute(Symbol: Sym, Attribute: Attr); |
800 | |
801 | return false; |
802 | } |
803 | |
804 | /// ParseDirectiveIdent |
805 | /// ::= .ident string |
806 | bool ELFAsmParser::ParseDirectiveIdent(StringRef, SMLoc) { |
807 | if (getLexer().isNot(K: AsmToken::String)) |
808 | return TokError(Msg: "expected string" ); |
809 | |
810 | StringRef Data = getTok().getIdentifier(); |
811 | |
812 | Lex(); |
813 | |
814 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) |
815 | return TokError(Msg: "expected end of directive" ); |
816 | Lex(); |
817 | |
818 | getStreamer().emitIdent(IdentString: Data); |
819 | return false; |
820 | } |
821 | |
822 | /// ParseDirectiveSymver |
823 | /// ::= .symver foo, bar2@zed |
824 | bool ELFAsmParser::ParseDirectiveSymver(StringRef, SMLoc) { |
825 | StringRef OriginalName, Name, Action; |
826 | if (getParser().parseIdentifier(Res&: OriginalName)) |
827 | return TokError(Msg: "expected identifier" ); |
828 | |
829 | if (getLexer().isNot(K: AsmToken::Comma)) |
830 | return TokError(Msg: "expected a comma" ); |
831 | |
832 | // ARM assembly uses @ for a comment... |
833 | // except when parsing the second parameter of the .symver directive. |
834 | // Force the next symbol to allow @ in the identifier, which is |
835 | // required for this directive and then reset it to its initial state. |
836 | const bool AllowAtInIdentifier = getLexer().getAllowAtInIdentifier(); |
837 | getLexer().setAllowAtInIdentifier(true); |
838 | Lex(); |
839 | getLexer().setAllowAtInIdentifier(AllowAtInIdentifier); |
840 | |
841 | if (getParser().parseIdentifier(Res&: Name)) |
842 | return TokError(Msg: "expected identifier" ); |
843 | |
844 | if (!Name.contains(C: '@')) |
845 | return TokError(Msg: "expected a '@' in the name" ); |
846 | bool KeepOriginalSym = !Name.contains(Other: "@@@" ); |
847 | if (parseOptionalToken(T: AsmToken::Comma)) { |
848 | if (getParser().parseIdentifier(Res&: Action) || Action != "remove" ) |
849 | return TokError(Msg: "expected 'remove'" ); |
850 | KeepOriginalSym = false; |
851 | } |
852 | (void)parseOptionalToken(T: AsmToken::EndOfStatement); |
853 | |
854 | getStreamer().emitELFSymverDirective( |
855 | OriginalSym: getContext().getOrCreateSymbol(Name: OriginalName), Name, KeepOriginalSym); |
856 | return false; |
857 | } |
858 | |
859 | /// ParseDirectiveVersion |
860 | /// ::= .version string |
861 | bool ELFAsmParser::ParseDirectiveVersion(StringRef, SMLoc) { |
862 | if (getLexer().isNot(K: AsmToken::String)) |
863 | return TokError(Msg: "expected string" ); |
864 | |
865 | StringRef Data = getTok().getIdentifier(); |
866 | |
867 | Lex(); |
868 | |
869 | MCSection *Note = getContext().getELFSection(Section: ".note" , Type: ELF::SHT_NOTE, Flags: 0); |
870 | |
871 | getStreamer().pushSection(); |
872 | getStreamer().switchSection(Section: Note); |
873 | getStreamer().emitInt32(Value: Data.size() + 1); // namesz |
874 | getStreamer().emitInt32(Value: 0); // descsz = 0 (no description). |
875 | getStreamer().emitInt32(Value: 1); // type = NT_VERSION |
876 | getStreamer().emitBytes(Data); // name |
877 | getStreamer().emitInt8(Value: 0); // NUL |
878 | getStreamer().emitValueToAlignment(Alignment: Align(4)); |
879 | getStreamer().popSection(); |
880 | return false; |
881 | } |
882 | |
883 | /// ParseDirectiveWeakref |
884 | /// ::= .weakref foo, bar |
885 | bool ELFAsmParser::ParseDirectiveWeakref(StringRef, SMLoc) { |
886 | // FIXME: Share code with the other alias building directives. |
887 | |
888 | StringRef AliasName; |
889 | if (getParser().parseIdentifier(Res&: AliasName)) |
890 | return TokError(Msg: "expected identifier" ); |
891 | |
892 | if (getLexer().isNot(K: AsmToken::Comma)) |
893 | return TokError(Msg: "expected a comma" ); |
894 | |
895 | Lex(); |
896 | |
897 | StringRef Name; |
898 | if (getParser().parseIdentifier(Res&: Name)) |
899 | return TokError(Msg: "expected identifier" ); |
900 | |
901 | MCSymbol *Alias = getContext().getOrCreateSymbol(Name: AliasName); |
902 | |
903 | MCSymbol *Sym = getContext().getOrCreateSymbol(Name); |
904 | |
905 | getStreamer().emitWeakReference(Alias, Symbol: Sym); |
906 | return false; |
907 | } |
908 | |
909 | bool ELFAsmParser::ParseDirectiveSubsection(StringRef, SMLoc) { |
910 | const MCExpr *Subsection = MCConstantExpr::create(Value: 0, Ctx&: getContext()); |
911 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) { |
912 | if (getParser().parseExpression(Res&: Subsection)) |
913 | return true; |
914 | } |
915 | |
916 | if (getLexer().isNot(K: AsmToken::EndOfStatement)) |
917 | return TokError(Msg: "expected end of directive" ); |
918 | |
919 | Lex(); |
920 | |
921 | return getStreamer().switchSection(Section: getStreamer().getCurrentSectionOnly(), |
922 | Subsection); |
923 | } |
924 | |
925 | bool ELFAsmParser::ParseDirectiveCGProfile(StringRef S, SMLoc Loc) { |
926 | return MCAsmParserExtension::ParseDirectiveCGProfile(S, Loc); |
927 | } |
928 | |
929 | namespace llvm { |
930 | |
931 | MCAsmParserExtension *createELFAsmParser() { |
932 | return new ELFAsmParser; |
933 | } |
934 | |
935 | } // end namespace llvm |
936 | |