LLVM 19.0.0git
IRSymtab.cpp
Go to the documentation of this file.
1//===- IRSymtab.cpp - implementation of IR symbol tables ------------------===//
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
10#include "llvm/ADT/ArrayRef.h"
11#include "llvm/ADT/DenseMap.h"
15#include "llvm/ADT/StringRef.h"
17#include "llvm/Config/llvm-config.h"
18#include "llvm/IR/Comdat.h"
19#include "llvm/IR/DataLayout.h"
20#include "llvm/IR/GlobalAlias.h"
22#include "llvm/IR/Mangler.h"
23#include "llvm/IR/Metadata.h"
24#include "llvm/IR/Module.h"
31#include "llvm/Support/Error.h"
33#include "llvm/Support/VCSRevision.h"
36#include <cassert>
37#include <string>
38#include <utility>
39#include <vector>
40
41using namespace llvm;
42using namespace irsymtab;
43
45 "disable-bitcode-version-upgrade", cl::Hidden,
46 cl::desc("Disable automatic bitcode upgrade for version mismatch"));
47
48static const char *PreservedSymbols[] = {
49#define HANDLE_LIBCALL(code, name) name,
50#include "llvm/IR/RuntimeLibcalls.def"
51#undef HANDLE_LIBCALL
52 // There are global variables, so put it here instead of in
53 // RuntimeLibcalls.def.
54 // TODO: Are there similar such variables?
55 "__ssp_canary_word",
56 "__stack_chk_guard",
57};
58
59namespace {
60
61const char *getExpectedProducerName() {
62 static char DefaultName[] = LLVM_VERSION_STRING
63#ifdef LLVM_REVISION
64 " " LLVM_REVISION
65#endif
66 ;
67 // Allows for testing of the irsymtab writer and upgrade mechanism. This
68 // environment variable should not be set by users.
69 if (char *OverrideName = getenv("LLVM_OVERRIDE_PRODUCER"))
70 return OverrideName;
71 return DefaultName;
72}
73
74const char *kExpectedProducerName = getExpectedProducerName();
75
76/// Stores the temporary state that is required to build an IR symbol table.
77struct Builder {
79 StringTableBuilder &StrtabBuilder;
80 StringSaver Saver;
81
82 // This ctor initializes a StringSaver using the passed in BumpPtrAllocator.
83 // The StringTableBuilder does not create a copy of any strings added to it,
84 // so this provides somewhere to store any strings that we create.
85 Builder(SmallVector<char, 0> &Symtab, StringTableBuilder &StrtabBuilder,
86 BumpPtrAllocator &Alloc)
87 : Symtab(Symtab), StrtabBuilder(StrtabBuilder), Saver(Alloc) {}
88
90 Mangler Mang;
91 Triple TT;
92
93 std::vector<storage::Comdat> Comdats;
94 std::vector<storage::Module> Mods;
95 std::vector<storage::Symbol> Syms;
96 std::vector<storage::Uncommon> Uncommons;
97
98 std::string COFFLinkerOpts;
99 raw_string_ostream COFFLinkerOptsOS{COFFLinkerOpts};
100
101 std::vector<storage::Str> DependentLibraries;
102
103 void setStr(storage::Str &S, StringRef Value) {
104 S.Offset = StrtabBuilder.add(Value);
105 S.Size = Value.size();
106 }
107
108 template <typename T>
109 void writeRange(storage::Range<T> &R, const std::vector<T> &Objs) {
110 R.Offset = Symtab.size();
111 R.Size = Objs.size();
112 Symtab.insert(Symtab.end(), reinterpret_cast<const char *>(Objs.data()),
113 reinterpret_cast<const char *>(Objs.data() + Objs.size()));
114 }
115
116 Expected<int> getComdatIndex(const Comdat *C, const Module *M);
117
118 Error addModule(Module *M);
119 Error addSymbol(const ModuleSymbolTable &Msymtab,
122
124};
125
126Error Builder::addModule(Module *M) {
127 if (M->getDataLayoutStr().empty())
128 return make_error<StringError>("input module has no datalayout",
130
131 // Symbols in the llvm.used list will get the FB_Used bit and will not be
132 // internalized. We do this for llvm.compiler.used as well:
133 //
134 // IR symbol table tracks module-level asm symbol references but not inline
135 // asm. A symbol only referenced by inline asm is not in the IR symbol table,
136 // so we may not know that the definition (in another translation unit) is
137 // referenced. That definition may have __attribute__((used)) (which lowers to
138 // llvm.compiler.used on ELF targets) to communicate to the compiler that it
139 // may be used by inline asm. The usage is perfectly fine, so we treat
140 // llvm.compiler.used conservatively as llvm.used to work around our own
141 // limitation.
143 collectUsedGlobalVariables(*M, UsedV, /*CompilerUsed=*/false);
144 collectUsedGlobalVariables(*M, UsedV, /*CompilerUsed=*/true);
146
147 ModuleSymbolTable Msymtab;
148 Msymtab.addModule(M);
149
151 Mod.Begin = Syms.size();
152 Mod.End = Syms.size() + Msymtab.symbols().size();
153 Mod.UncBegin = Uncommons.size();
154 Mods.push_back(Mod);
155
156 if (TT.isOSBinFormatCOFF()) {
157 if (auto E = M->materializeMetadata())
158 return E;
159 if (NamedMDNode *LinkerOptions =
160 M->getNamedMetadata("llvm.linker.options")) {
161 for (MDNode *MDOptions : LinkerOptions->operands())
162 for (const MDOperand &MDOption : cast<MDNode>(MDOptions)->operands())
163 COFFLinkerOptsOS << " " << cast<MDString>(MDOption)->getString();
164 }
165 }
166
167 if (TT.isOSBinFormatELF()) {
168 if (auto E = M->materializeMetadata())
169 return E;
170 if (NamedMDNode *N = M->getNamedMetadata("llvm.dependent-libraries")) {
171 for (MDNode *MDOptions : N->operands()) {
172 const auto OperandStr =
173 cast<MDString>(cast<MDNode>(MDOptions)->getOperand(0))->getString();
174 storage::Str Specifier;
175 setStr(Specifier, OperandStr);
176 DependentLibraries.emplace_back(Specifier);
177 }
178 }
179 }
180
181 for (ModuleSymbolTable::Symbol Msym : Msymtab.symbols())
182 if (Error Err = addSymbol(Msymtab, Used, Msym))
183 return Err;
184
185 return Error::success();
186}
187
188Expected<int> Builder::getComdatIndex(const Comdat *C, const Module *M) {
189 auto P = ComdatMap.insert(std::make_pair(C, Comdats.size()));
190 if (P.second) {
191 std::string Name;
192 if (TT.isOSBinFormatCOFF()) {
193 const GlobalValue *GV = M->getNamedValue(C->getName());
194 if (!GV)
195 return make_error<StringError>("Could not find leader",
197 // Internal leaders do not affect symbol resolution, therefore they do not
198 // appear in the symbol table.
199 if (GV->hasLocalLinkage()) {
200 P.first->second = -1;
201 return -1;
202 }
204 Mang.getNameWithPrefix(OS, GV, false);
205 } else {
206 Name = std::string(C->getName());
207 }
208
210 setStr(Comdat.Name, Saver.save(Name));
211 Comdat.SelectionKind = C->getSelectionKind();
212 Comdats.push_back(Comdat);
213 }
214
215 return P.first->second;
216}
217
218static DenseSet<StringRef> buildPreservedSymbolsSet() {
219 return DenseSet<StringRef>(std::begin(PreservedSymbols),
220 std::end(PreservedSymbols));
221}
222
223Error Builder::addSymbol(const ModuleSymbolTable &Msymtab,
226 Syms.emplace_back();
227 storage::Symbol &Sym = Syms.back();
228 Sym = {};
229
230 storage::Uncommon *Unc = nullptr;
231 auto Uncommon = [&]() -> storage::Uncommon & {
232 if (Unc)
233 return *Unc;
235 Uncommons.emplace_back();
236 Unc = &Uncommons.back();
237 *Unc = {};
238 setStr(Unc->COFFWeakExternFallbackName, "");
239 setStr(Unc->SectionName, "");
240 return *Unc;
241 };
242
244 {
246 Msymtab.printSymbolName(OS, Msym);
247 }
248 setStr(Sym.Name, Saver.save(Name.str()));
249
250 auto Flags = Msymtab.getSymbolFlags(Msym);
265
266 Sym.ComdatIndex = -1;
267 auto *GV = dyn_cast_if_present<GlobalValue *>(Msym);
268 if (!GV) {
269 // Undefined module asm symbols act as GC roots and are implicitly used.
272 setStr(Sym.IRName, "");
273 return Error::success();
274 }
275
276 setStr(Sym.IRName, GV->getName());
277
278 static const DenseSet<StringRef> PreservedSymbolsSet =
279 buildPreservedSymbolsSet();
280 bool IsPreservedSymbol = PreservedSymbolsSet.contains(GV->getName());
281
282 if (Used.count(GV) || IsPreservedSymbol)
284 if (GV->isThreadLocal())
286 if (GV->hasGlobalUnnamedAddr())
291
293 auto *GVar = dyn_cast<GlobalVariable>(GV);
294 if (!GVar)
295 return make_error<StringError>("Only variables can have common linkage!",
297 Uncommon().CommonSize =
299 Uncommon().CommonAlign = GVar->getAlign() ? GVar->getAlign()->value() : 0;
300 }
301
302 const GlobalObject *GO = GV->getAliaseeObject();
303 if (!GO) {
304 if (isa<GlobalIFunc>(GV))
305 GO = cast<GlobalIFunc>(GV)->getResolverFunction();
306 if (!GO)
307 return make_error<StringError>("Unable to determine comdat of alias!",
309 }
310 if (const Comdat *C = GO->getComdat()) {
311 Expected<int> ComdatIndexOrErr = getComdatIndex(C, GV->getParent());
312 if (!ComdatIndexOrErr)
313 return ComdatIndexOrErr.takeError();
314 Sym.ComdatIndex = *ComdatIndexOrErr;
315 }
316
317 if (TT.isOSBinFormatCOFF()) {
318 emitLinkerFlagsForGlobalCOFF(COFFLinkerOptsOS, GV, TT, Mang);
319
320 if ((Flags & object::BasicSymbolRef::SF_Weak) &&
322 auto *Fallback = dyn_cast<GlobalValue>(
323 cast<GlobalAlias>(GV)->getAliasee()->stripPointerCasts());
324 if (!Fallback)
325 return make_error<StringError>("Invalid weak external",
327 std::string FallbackName;
328 raw_string_ostream OS(FallbackName);
329 Msymtab.printSymbolName(OS, Fallback);
330 OS.flush();
331 setStr(Uncommon().COFFWeakExternFallbackName, Saver.save(FallbackName));
332 }
333 }
334
335 if (!GO->getSection().empty())
336 setStr(Uncommon().SectionName, Saver.save(GO->getSection()));
337
338 return Error::success();
339}
340
341Error Builder::build(ArrayRef<Module *> IRMods) {
342 storage::Header Hdr;
343
344 assert(!IRMods.empty());
346 setStr(Hdr.Producer, kExpectedProducerName);
347 setStr(Hdr.TargetTriple, IRMods[0]->getTargetTriple());
348 setStr(Hdr.SourceFileName, IRMods[0]->getSourceFileName());
349 TT = Triple(IRMods[0]->getTargetTriple());
350
351 for (auto *M : IRMods)
352 if (Error Err = addModule(M))
353 return Err;
354
355 COFFLinkerOptsOS.flush();
356 setStr(Hdr.COFFLinkerOpts, Saver.save(COFFLinkerOpts));
357
358 // We are about to fill in the header's range fields, so reserve space for it
359 // and copy it in afterwards.
360 Symtab.resize(sizeof(storage::Header));
361 writeRange(Hdr.Modules, Mods);
362 writeRange(Hdr.Comdats, Comdats);
363 writeRange(Hdr.Symbols, Syms);
364 writeRange(Hdr.Uncommons, Uncommons);
365 writeRange(Hdr.DependentLibraries, DependentLibraries);
366 *reinterpret_cast<storage::Header *>(Symtab.data()) = Hdr;
367 return Error::success();
368}
369
370} // end anonymous namespace
371
373 StringTableBuilder &StrtabBuilder,
374 BumpPtrAllocator &Alloc) {
375 return Builder(Symtab, StrtabBuilder, Alloc).build(Mods);
376}
377
378// Upgrade a vector of bitcode modules created by an old version of LLVM by
379// creating an irsymtab for them in the current format.
381 FileContents FC;
382
383 LLVMContext Ctx;
384 std::vector<Module *> Mods;
385 std::vector<std::unique_ptr<Module>> OwnedMods;
386 for (auto BM : BMs) {
388 BM.getLazyModule(Ctx, /*ShouldLazyLoadMetadata*/ true,
389 /*IsImporting*/ false);
390 if (!MOrErr)
391 return MOrErr.takeError();
392
393 Mods.push_back(MOrErr->get());
394 OwnedMods.push_back(std::move(*MOrErr));
395 }
396
398 BumpPtrAllocator Alloc;
399 if (Error E = build(Mods, FC.Symtab, StrtabBuilder, Alloc))
400 return std::move(E);
401
402 StrtabBuilder.finalizeInOrder();
403 FC.Strtab.resize(StrtabBuilder.getSize());
404 StrtabBuilder.write((uint8_t *)FC.Strtab.data());
405
406 FC.TheReader = {{FC.Symtab.data(), FC.Symtab.size()},
407 {FC.Strtab.data(), FC.Strtab.size()}};
408 return std::move(FC);
409}
410
412 if (BFC.Mods.empty())
413 return make_error<StringError>("Bitcode file does not contain any modules",
415
417 if (BFC.StrtabForSymtab.empty() ||
418 BFC.Symtab.size() < sizeof(storage::Header))
419 return upgrade(BFC.Mods);
420
421 // We cannot use the regular reader to read the version and producer,
422 // because it will expect the header to be in the current format. The only
423 // thing we can rely on is that the version and producer will be present as
424 // the first struct elements.
425 auto *Hdr = reinterpret_cast<const storage::Header *>(BFC.Symtab.data());
426 unsigned Version = Hdr->Version;
427 StringRef Producer = Hdr->Producer.get(BFC.StrtabForSymtab);
428 if (Version != storage::Header::kCurrentVersion ||
429 Producer != kExpectedProducerName)
430 return upgrade(BFC.Mods);
431 }
432
433 FileContents FC;
434 FC.TheReader = {{BFC.Symtab.data(), BFC.Symtab.size()},
436
437 // Finally, make sure that the number of modules in the symbol table matches
438 // the number of modules in the bitcode file. If they differ, it may mean that
439 // the bitcode file was created by binary concatenation, so we need to create
440 // a new symbol table from scratch.
441 if (FC.TheReader.getNumModules() != BFC.Mods.size())
442 return upgrade(std::move(BFC.Mods));
443
444 return std::move(FC);
445}
This file defines the BumpPtrAllocator interface.
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This file defines the DenseMap class.
std::string Name
static void addSymbol(Object &Obj, const NewSymbolInfo &SymInfo, uint8_t DefaultVisibility)
Definition: ELFObjcopy.cpp:543
Symbol * Sym
Definition: ELF_riscv.cpp:479
static const char * PreservedSymbols[]
Definition: IRSymtab.cpp:48
static cl::opt< bool > DisableBitcodeVersionUpgrade("disable-bitcode-version-upgrade", cl::Hidden, cl::desc("Disable automatic bitcode upgrade for version mismatch"))
static Expected< FileContents > upgrade(ArrayRef< BitcodeModule > BMs)
Definition: IRSymtab.cpp:380
This file contains the declarations for metadata subclasses.
Module.h This file contains the declarations for the Module class.
#define P(N)
Module * Mod
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
raw_pwrite_stream & OS
This file defines the SmallPtrSet class.
This file defines the SmallString class.
This file defines the SmallVector class.
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
Definition: ArrayRef.h:41
bool empty() const
empty - Check if the array is empty.
Definition: ArrayRef.h:160
Allocate memory in an ever growing pool, as if by bump-pointer.
Definition: Allocator.h:66
TypeSize getTypeAllocSize(Type *Ty) const
Returns the offset in bytes between successive objects of the specified type, including alignment pad...
Definition: DataLayout.h:504
std::pair< iterator, bool > insert(const std::pair< KeyT, ValueT > &KV)
Definition: DenseMap.h:220
Implements a dense probed hash-table based set.
Definition: DenseSet.h:271
Lightweight error class with error context and mandatory checking.
Definition: Error.h:160
static ErrorSuccess success()
Create a success value.
Definition: Error.h:334
Tagged union holding either a T or a Error.
Definition: Error.h:474
Error takeError()
Take ownership of the stored error.
Definition: Error.h:601
reference get()
Returns a reference to the stored T value.
Definition: Error.h:571
StringRef getSection() const
Get the custom section of this global if it has one.
Definition: GlobalObject.h:118
const Comdat * getComdat() const
Definition: GlobalObject.h:129
bool isThreadLocal() const
If the value is "Thread Local", its value isn't shared by the threads.
Definition: GlobalValue.h:263
VisibilityTypes getVisibility() const
Definition: GlobalValue.h:248
bool hasLocalLinkage() const
Definition: GlobalValue.h:527
Module * getParent()
Get the module that this global value is contained inside of...
Definition: GlobalValue.h:655
const GlobalObject * getAliaseeObject() const
Definition: Globals.cpp:368
bool hasGlobalUnnamedAddr() const
Definition: GlobalValue.h:215
bool canBeOmittedFromSymbolTable() const
True if GV can be left out of the object symbol table.
Definition: Globals.cpp:393
Type * getValueType() const
Definition: GlobalValue.h:296
This is an important class for using LLVM in a threaded context.
Definition: LLVMContext.h:67
Metadata node.
Definition: Metadata.h:1067
ArrayRef< MDOperand > operands() const
Definition: Metadata.h:1426
Tracking metadata reference owned by Metadata.
Definition: Metadata.h:889
void getNameWithPrefix(raw_ostream &OS, const GlobalValue *GV, bool CannotUsePrivateLabel) const
Print the appropriate prefix and the specified global variable's name.
Definition: Mangler.cpp:119
void printSymbolName(raw_ostream &OS, Symbol S) const
uint32_t getSymbolFlags(Symbol S) const
ArrayRef< Symbol > symbols() const
A Module instance is used to store all the information related to an LLVM module.
Definition: Module.h:65
size_t size() const
Definition: Module.h:712
const DataLayout & getDataLayout() const
Get the data layout for the module's target platform.
Definition: Module.h:287
A tuple of MDNodes.
Definition: Metadata.h:1729
A discriminated union of two or more pointer types, with the discriminator in the low bit of the poin...
Definition: PointerUnion.h:118
SmallPtrSet - This class implements a set which is optimized for holding SmallSize or less elements.
Definition: SmallPtrSet.h:427
SmallString - A SmallString is just a SmallVector with methods and accessors that make it work better...
Definition: SmallString.h:26
size_t size() const
Definition: SmallVector.h:91
iterator insert(iterator I, T &&Elt)
Definition: SmallVector.h:818
void resize(size_type N)
Definition: SmallVector.h:651
pointer data()
Return a pointer to the vector's buffer, even if empty().
Definition: SmallVector.h:299
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
Definition: SmallVector.h:1209
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:50
constexpr bool empty() const
empty - Check if the string is empty.
Definition: StringRef.h:134
constexpr size_t size() const
size - Get the string size.
Definition: StringRef.h:137
constexpr const char * data() const
data - Get a pointer to the start of the string (which may not be null terminated).
Definition: StringRef.h:131
Saves strings in the provided stable storage and returns a StringRef with a stable character pointer.
Definition: StringSaver.h:21
StringRef save(const char *S)
Definition: StringSaver.h:30
Utility for building string tables with deduplicated suffixes.
void finalizeInOrder()
Finalize the string table without reording it.
void write(raw_ostream &OS) const
size_t add(CachedHashStringRef S)
Add a string to the builder.
Triple - Helper class for working with autoconf configuration names.
Definition: Triple.h:44
LLVM Value Representation.
Definition: Value.h:74
StringRef getName() const
Return a constant reference to the value's name.
Definition: Value.cpp:309
bool contains(const_arg_type_t< ValueT > V) const
Check if the set contains the given element.
Definition: DenseSet.h:185
A raw_ostream that writes to an std::string.
Definition: raw_ostream.h:660
A raw_ostream that writes to an SmallVector or SmallString.
Definition: raw_ostream.h:690
This file contains the declaration of the Comdat class, which represents a single COMDAT in LLVM.
@ C
The default llvm calling convention, compatible with C.
Definition: CallingConv.h:34
Expected< FileContents > readBitcode(const BitcodeFileContents &BFC)
Reads the contents of a bitcode file, creating its irsymtab if necessary.
Definition: IRSymtab.cpp:411
Error build(ArrayRef< Module * > Mods, SmallVector< char, 0 > &Symtab, StringTableBuilder &StrtabBuilder, BumpPtrAllocator &Alloc)
Fills in Symtab and StrtabBuilder with a valid symbol and string table for Mods.
Definition: IRSymtab.cpp:372
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
std::error_code inconvertibleErrorCode()
The value returned by this function can be returned from convertToErrorCode for Error values where no...
Definition: Error.cpp:90
void emitLinkerFlagsForGlobalCOFF(raw_ostream &OS, const GlobalValue *GV, const Triple &TT, Mangler &Mangler)
Definition: Mangler.cpp:212
GlobalVariable * collectUsedGlobalVariables(const Module &M, SmallVectorImpl< GlobalValue * > &Vec, bool CompilerUsed)
Given "llvm.used" or "llvm.compiler.used" as a global name, collect the initializer elements of that ...
Definition: Module.cpp:843
#define N
std::vector< BitcodeModule > Mods
The contents of the irsymtab in a bitcode file.
Definition: IRSymtab.h:369
This is equivalent to an IR comdat.
Definition: IRSymtab.h:82
Word Version
Version number of the symtab format.
Definition: IRSymtab.h:138
Str COFFLinkerOpts
COFF-specific: linker directives.
Definition: IRSymtab.h:155
Range< Str > DependentLibraries
Dependent Library Specifiers.
Definition: IRSymtab.h:158
Range< Uncommon > Uncommons
Definition: IRSymtab.h:150
Str Producer
The producer's version string (LLVM_VERSION_STRING " " LLVM_REVISION).
Definition: IRSymtab.h:145
Describes the range of a particular module's symbols within the symbol table.
Definition: IRSymtab.h:74
A reference to a range of objects in the symbol table.
Definition: IRSymtab.h:64
A reference to a string in the string table.
Definition: IRSymtab.h:55
StringRef get(StringRef Strtab) const
Definition: IRSymtab.h:58
Contains the information needed by linkers for symbol resolution, as well as by the LTO implementatio...
Definition: IRSymtab.h:91
This data structure contains rarely used symbol fields and is optionally referenced by a Symbol.
Definition: IRSymtab.h:122
Str SectionName
Specified section name, if any.
Definition: IRSymtab.h:130
Str COFFWeakExternFallbackName
COFF-specific: the name of the symbol that a weak external resolves to if not defined.
Definition: IRSymtab.h:127