1 //===- llvm-link.cpp - Low-level LLVM linker ------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This utility may be invoked in the following manner:
11 //  llvm-link a.bc b.bc c.bc -o x.bc
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/Bitcode/BitcodeReader.h"
17 #include "llvm/Bitcode/BitcodeWriter.h"
18 #include "llvm/IR/AutoUpgrade.h"
19 #include "llvm/IR/DiagnosticInfo.h"
20 #include "llvm/IR/DiagnosticPrinter.h"
21 #include "llvm/IR/LLVMContext.h"
22 #include "llvm/IR/Module.h"
23 #include "llvm/IR/ModuleSummaryIndex.h"
24 #include "llvm/IR/Verifier.h"
25 #include "llvm/IRReader/IRReader.h"
26 #include "llvm/Linker/Linker.h"
27 #include "llvm/Support/CommandLine.h"
28 #include "llvm/Support/FileSystem.h"
29 #include "llvm/Support/ManagedStatic.h"
30 #include "llvm/Support/Path.h"
31 #include "llvm/Support/PrettyStackTrace.h"
32 #include "llvm/Support/Signals.h"
33 #include "llvm/Support/SourceMgr.h"
34 #include "llvm/Support/SystemUtils.h"
35 #include "llvm/Support/ToolOutputFile.h"
36 #include "llvm/Transforms/IPO/FunctionImport.h"
37 #include "llvm/Transforms/IPO/Internalize.h"
38 #include "llvm/Transforms/Utils/FunctionImportUtils.h"
39 
40 #include <memory>
41 #include <utility>
42 using namespace llvm;
43 
44 static cl::list<std::string>
45 InputFilenames(cl::Positional, cl::OneOrMore,
46                cl::desc("<input bitcode files>"));
47 
48 static cl::list<std::string> OverridingInputs(
49     "override", cl::ZeroOrMore, cl::value_desc("filename"),
50     cl::desc(
51         "input bitcode file which can override previously defined symbol(s)"));
52 
53 // Option to simulate function importing for testing. This enables using
54 // llvm-link to simulate ThinLTO backend processes.
55 static cl::list<std::string> Imports(
56     "import", cl::ZeroOrMore, cl::value_desc("function:filename"),
57     cl::desc("Pair of function name and filename, where function should be "
58              "imported from bitcode in filename"));
59 
60 // Option to support testing of function importing. The module summary
61 // must be specified in the case were we request imports via the -import
62 // option, as well as when compiling any module with functions that may be
63 // exported (imported by a different llvm-link -import invocation), to ensure
64 // consistent promotion and renaming of locals.
65 static cl::opt<std::string>
66     SummaryIndex("summary-index", cl::desc("Module summary index filename"),
67                  cl::init(""), cl::value_desc("filename"));
68 
69 static cl::opt<std::string>
70 OutputFilename("o", cl::desc("Override output filename"), cl::init("-"),
71                cl::value_desc("filename"));
72 
73 static cl::opt<bool>
74 Internalize("internalize", cl::desc("Internalize linked symbols"));
75 
76 static cl::opt<bool>
77     DisableDITypeMap("disable-debug-info-type-map",
78                      cl::desc("Don't use a uniquing type map for debug info"));
79 
80 static cl::opt<bool>
81 OnlyNeeded("only-needed", cl::desc("Link only needed symbols"));
82 
83 static cl::opt<bool>
84 Force("f", cl::desc("Enable binary output on terminals"));
85 
86 static cl::opt<bool>
87     DisableLazyLoad("disable-lazy-loading",
88                     cl::desc("Disable lazy module loading"));
89 
90 static cl::opt<bool>
91     OutputAssembly("S", cl::desc("Write output as LLVM assembly"), cl::Hidden);
92 
93 static cl::opt<bool>
94 Verbose("v", cl::desc("Print information about actions taken"));
95 
96 static cl::opt<bool>
97 DumpAsm("d", cl::desc("Print assembly as linked"), cl::Hidden);
98 
99 static cl::opt<bool>
100 SuppressWarnings("suppress-warnings", cl::desc("Suppress all linking warnings"),
101                  cl::init(false));
102 
103 static cl::opt<bool> PreserveBitcodeUseListOrder(
104     "preserve-bc-uselistorder",
105     cl::desc("Preserve use-list order when writing LLVM bitcode."),
106     cl::init(true), cl::Hidden);
107 
108 static cl::opt<bool> PreserveAssemblyUseListOrder(
109     "preserve-ll-uselistorder",
110     cl::desc("Preserve use-list order when writing LLVM assembly."),
111     cl::init(false), cl::Hidden);
112 
113 static ExitOnError ExitOnErr;
114 
115 // Read the specified bitcode file in and return it. This routine searches the
116 // link path for the specified file to try to find it...
117 //
118 static std::unique_ptr<Module> loadFile(const char *argv0,
119                                         const std::string &FN,
120                                         LLVMContext &Context,
121                                         bool MaterializeMetadata = true) {
122   SMDiagnostic Err;
123   if (Verbose) errs() << "Loading '" << FN << "'\n";
124   std::unique_ptr<Module> Result;
125   if (DisableLazyLoad)
126     Result = parseIRFile(FN, Err, Context);
127   else
128     Result = getLazyIRFileModule(FN, Err, Context, !MaterializeMetadata);
129 
130   if (!Result) {
131     Err.print(argv0, errs());
132     return nullptr;
133   }
134 
135   if (MaterializeMetadata) {
136     ExitOnErr(Result->materializeMetadata());
137     UpgradeDebugInfo(*Result);
138   }
139 
140   return Result;
141 }
142 
143 namespace {
144 
145 /// Helper to load on demand a Module from file and cache it for subsequent
146 /// queries during function importing.
147 class ModuleLazyLoaderCache {
148   /// Cache of lazily loaded module for import.
149   StringMap<std::unique_ptr<Module>> ModuleMap;
150 
151   /// Retrieve a Module from the cache or lazily load it on demand.
152   std::function<std::unique_ptr<Module>(const char *argv0,
153                                         const std::string &FileName)>
154       createLazyModule;
155 
156 public:
157   /// Create the loader, Module will be initialized in \p Context.
158   ModuleLazyLoaderCache(std::function<std::unique_ptr<Module>(
159                             const char *argv0, const std::string &FileName)>
160                             createLazyModule)
161       : createLazyModule(std::move(createLazyModule)) {}
162 
163   /// Retrieve a Module from the cache or lazily load it on demand.
164   Module &operator()(const char *argv0, const std::string &FileName);
165 
166   std::unique_ptr<Module> takeModule(const std::string &FileName) {
167     auto I = ModuleMap.find(FileName);
168     assert(I != ModuleMap.end());
169     std::unique_ptr<Module> Ret = std::move(I->second);
170     ModuleMap.erase(I);
171     return Ret;
172   }
173 };
174 
175 // Get a Module for \p FileName from the cache, or load it lazily.
176 Module &ModuleLazyLoaderCache::operator()(const char *argv0,
177                                           const std::string &Identifier) {
178   auto &Module = ModuleMap[Identifier];
179   if (!Module)
180     Module = createLazyModule(argv0, Identifier);
181   return *Module;
182 }
183 } // anonymous namespace
184 
185 namespace {
186 struct LLVMLinkDiagnosticHandler : public DiagnosticHandler {
187   bool handleDiagnostics(const DiagnosticInfo &DI) override {
188     unsigned Severity = DI.getSeverity();
189     switch (Severity) {
190     case DS_Error:
191       errs() << "ERROR: ";
192       break;
193     case DS_Warning:
194       if (SuppressWarnings)
195         return true;
196       errs() << "WARNING: ";
197       break;
198     case DS_Remark:
199     case DS_Note:
200       llvm_unreachable("Only expecting warnings and errors");
201     }
202 
203     DiagnosticPrinterRawOStream DP(errs());
204     DI.print(DP);
205     errs() << '\n';
206     return true;
207   }
208 };
209 }
210 
211 /// Import any functions requested via the -import option.
212 static bool importFunctions(const char *argv0, Module &DestModule) {
213   if (SummaryIndex.empty())
214     return true;
215   std::unique_ptr<ModuleSummaryIndex> Index =
216       ExitOnErr(llvm::getModuleSummaryIndexForFile(SummaryIndex));
217 
218   // Map of Module -> List of globals to import from the Module
219   FunctionImporter::ImportMapTy ImportList;
220 
221   auto ModuleLoader = [&DestModule](const char *argv0,
222                                     const std::string &Identifier) {
223     return loadFile(argv0, Identifier, DestModule.getContext(), false);
224   };
225 
226   ModuleLazyLoaderCache ModuleLoaderCache(ModuleLoader);
227   for (const auto &Import : Imports) {
228     // Identify the requested function and its bitcode source file.
229     size_t Idx = Import.find(':');
230     if (Idx == std::string::npos) {
231       errs() << "Import parameter bad format: " << Import << "\n";
232       return false;
233     }
234     std::string FunctionName = Import.substr(0, Idx);
235     std::string FileName = Import.substr(Idx + 1, std::string::npos);
236 
237     // Load the specified source module.
238     auto &SrcModule = ModuleLoaderCache(argv0, FileName);
239 
240     if (verifyModule(SrcModule, &errs())) {
241       errs() << argv0 << ": " << FileName
242              << ": error: input module is broken!\n";
243       return false;
244     }
245 
246     Function *F = SrcModule.getFunction(FunctionName);
247     if (!F) {
248       errs() << "Ignoring import request for non-existent function "
249              << FunctionName << " from " << FileName << "\n";
250       continue;
251     }
252     // We cannot import weak_any functions without possibly affecting the
253     // order they are seen and selected by the linker, changing program
254     // semantics.
255     if (F->hasWeakAnyLinkage()) {
256       errs() << "Ignoring import request for weak-any function " << FunctionName
257              << " from " << FileName << "\n";
258       continue;
259     }
260 
261     if (Verbose)
262       errs() << "Importing " << FunctionName << " from " << FileName << "\n";
263 
264     auto &Entry = ImportList[FileName];
265     Entry.insert(std::make_pair(F->getGUID(), /* (Unused) threshold */ 1.0));
266   }
267   auto CachedModuleLoader = [&](StringRef Identifier) {
268     return ModuleLoaderCache.takeModule(Identifier);
269   };
270   FunctionImporter Importer(*Index, CachedModuleLoader);
271   ExitOnErr(Importer.importFunctions(DestModule, ImportList));
272 
273   return true;
274 }
275 
276 static bool linkFiles(const char *argv0, LLVMContext &Context, Linker &L,
277                       const cl::list<std::string> &Files,
278                       unsigned Flags) {
279   // Filter out flags that don't apply to the first file we load.
280   unsigned ApplicableFlags = Flags & Linker::Flags::OverrideFromSrc;
281   // Similar to some flags, internalization doesn't apply to the first file.
282   bool InternalizeLinkedSymbols = false;
283   for (const auto &File : Files) {
284     std::unique_ptr<Module> M = loadFile(argv0, File, Context);
285     if (!M.get()) {
286       errs() << argv0 << ": error loading file '" << File << "'\n";
287       return false;
288     }
289 
290     // Note that when ODR merging types cannot verify input files in here When
291     // doing that debug metadata in the src module might already be pointing to
292     // the destination.
293     if (DisableDITypeMap && verifyModule(*M, &errs())) {
294       errs() << argv0 << ": " << File << ": error: input module is broken!\n";
295       return false;
296     }
297 
298     // If a module summary index is supplied, load it so linkInModule can treat
299     // local functions/variables as exported and promote if necessary.
300     if (!SummaryIndex.empty()) {
301       std::unique_ptr<ModuleSummaryIndex> Index =
302           ExitOnErr(llvm::getModuleSummaryIndexForFile(SummaryIndex));
303 
304       // Conservatively mark all internal values as promoted, since this tool
305       // does not do the ThinLink that would normally determine what values to
306       // promote.
307       for (auto &I : *Index) {
308         for (auto &S : I.second.SummaryList) {
309           if (GlobalValue::isLocalLinkage(S->linkage()))
310             S->setLinkage(GlobalValue::ExternalLinkage);
311         }
312       }
313 
314       // Promotion
315       if (renameModuleForThinLTO(*M, *Index))
316         return true;
317     }
318 
319     if (Verbose)
320       errs() << "Linking in '" << File << "'\n";
321 
322     bool Err = false;
323     if (InternalizeLinkedSymbols) {
324       Err = L.linkInModule(
325           std::move(M), ApplicableFlags, [](Module &M, const StringSet<> &GVS) {
326             internalizeModule(M, [&GVS](const GlobalValue &GV) {
327               return !GV.hasName() || (GVS.count(GV.getName()) == 0);
328             });
329           });
330     } else {
331       Err = L.linkInModule(std::move(M), ApplicableFlags);
332     }
333 
334     if (Err)
335       return false;
336 
337     // Internalization applies to linking of subsequent files.
338     InternalizeLinkedSymbols = Internalize;
339 
340     // All linker flags apply to linking of subsequent files.
341     ApplicableFlags = Flags;
342   }
343 
344   return true;
345 }
346 
347 int main(int argc, char **argv) {
348   // Print a stack trace if we signal out.
349   sys::PrintStackTraceOnErrorSignal(argv[0]);
350   PrettyStackTraceProgram X(argc, argv);
351 
352   ExitOnErr.setBanner(std::string(argv[0]) + ": ");
353 
354   LLVMContext Context;
355   Context.setDiagnosticHandler(
356     llvm::make_unique<LLVMLinkDiagnosticHandler>(), true);
357   llvm_shutdown_obj Y;  // Call llvm_shutdown() on exit.
358   cl::ParseCommandLineOptions(argc, argv, "llvm linker\n");
359 
360   if (!DisableDITypeMap)
361     Context.enableDebugTypeODRUniquing();
362 
363   auto Composite = make_unique<Module>("llvm-link", Context);
364   Linker L(*Composite);
365 
366   unsigned Flags = Linker::Flags::None;
367   if (OnlyNeeded)
368     Flags |= Linker::Flags::LinkOnlyNeeded;
369 
370   // First add all the regular input files
371   if (!linkFiles(argv[0], Context, L, InputFilenames, Flags))
372     return 1;
373 
374   // Next the -override ones.
375   if (!linkFiles(argv[0], Context, L, OverridingInputs,
376                  Flags | Linker::Flags::OverrideFromSrc))
377     return 1;
378 
379   // Import any functions requested via -import
380   if (!importFunctions(argv[0], *Composite))
381     return 1;
382 
383   if (DumpAsm) errs() << "Here's the assembly:\n" << *Composite;
384 
385   std::error_code EC;
386   ToolOutputFile Out(OutputFilename, EC, sys::fs::F_None);
387   if (EC) {
388     errs() << EC.message() << '\n';
389     return 1;
390   }
391 
392   if (verifyModule(*Composite, &errs())) {
393     errs() << argv[0] << ": error: linked module is broken!\n";
394     return 1;
395   }
396 
397   if (Verbose) errs() << "Writing bitcode...\n";
398   if (OutputAssembly) {
399     Composite->print(Out.os(), nullptr, PreserveAssemblyUseListOrder);
400   } else if (Force || !CheckBitcodeOutputToConsole(Out.os(), true))
401     WriteBitcodeToFile(Composite.get(), Out.os(), PreserveBitcodeUseListOrder);
402 
403   // Declare success.
404   Out.keep();
405 
406   return 0;
407 }
408