Internalize.cpp revision 36b56886974eae4f9c5ebc96befd3e7bfe5de338
1//===-- Internalize.cpp - Mark functions internal -------------------------===// 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 pass loops over all of the functions and variables in the input module. 11// If the function or variable is not in the list of external names given to 12// the pass it is marked as internal. 13// 14// This transformation would not be legal in a regular compilation, but it gets 15// extra information from the linker about what is safe. 16// 17// For example: Internalizing a function with external linkage. Only if we are 18// told it is only used from within this module, it is safe to do it. 19// 20//===----------------------------------------------------------------------===// 21 22#define DEBUG_TYPE "internalize" 23#include "llvm/Transforms/IPO.h" 24#include "llvm/ADT/SmallPtrSet.h" 25#include "llvm/ADT/Statistic.h" 26#include "llvm/Analysis/CallGraph.h" 27#include "llvm/IR/Module.h" 28#include "llvm/Pass.h" 29#include "llvm/Support/CommandLine.h" 30#include "llvm/Support/Debug.h" 31#include "llvm/Support/raw_ostream.h" 32#include "llvm/Transforms/Utils/GlobalStatus.h" 33#include "llvm/Transforms/Utils/ModuleUtils.h" 34#include <fstream> 35#include <set> 36using namespace llvm; 37 38STATISTIC(NumAliases , "Number of aliases internalized"); 39STATISTIC(NumFunctions, "Number of functions internalized"); 40STATISTIC(NumGlobals , "Number of global vars internalized"); 41 42// APIFile - A file which contains a list of symbols that should not be marked 43// external. 44static cl::opt<std::string> 45APIFile("internalize-public-api-file", cl::value_desc("filename"), 46 cl::desc("A file containing list of symbol names to preserve")); 47 48// APIList - A list of symbols that should not be marked internal. 49static cl::list<std::string> 50APIList("internalize-public-api-list", cl::value_desc("list"), 51 cl::desc("A list of symbol names to preserve"), 52 cl::CommaSeparated); 53 54namespace { 55 class InternalizePass : public ModulePass { 56 std::set<std::string> ExternalNames; 57 public: 58 static char ID; // Pass identification, replacement for typeid 59 explicit InternalizePass(); 60 explicit InternalizePass(ArrayRef<const char *> ExportList); 61 void LoadFile(const char *Filename); 62 bool runOnModule(Module &M) override; 63 64 void getAnalysisUsage(AnalysisUsage &AU) const override { 65 AU.setPreservesCFG(); 66 AU.addPreserved<CallGraphWrapperPass>(); 67 } 68 }; 69} // end anonymous namespace 70 71char InternalizePass::ID = 0; 72INITIALIZE_PASS(InternalizePass, "internalize", 73 "Internalize Global Symbols", false, false) 74 75InternalizePass::InternalizePass() : ModulePass(ID) { 76 initializeInternalizePassPass(*PassRegistry::getPassRegistry()); 77 if (!APIFile.empty()) // If a filename is specified, use it. 78 LoadFile(APIFile.c_str()); 79 ExternalNames.insert(APIList.begin(), APIList.end()); 80} 81 82InternalizePass::InternalizePass(ArrayRef<const char *> ExportList) 83 : ModulePass(ID) { 84 initializeInternalizePassPass(*PassRegistry::getPassRegistry()); 85 for(ArrayRef<const char *>::const_iterator itr = ExportList.begin(); 86 itr != ExportList.end(); itr++) { 87 ExternalNames.insert(*itr); 88 } 89} 90 91void InternalizePass::LoadFile(const char *Filename) { 92 // Load the APIFile... 93 std::ifstream In(Filename); 94 if (!In.good()) { 95 errs() << "WARNING: Internalize couldn't load file '" << Filename 96 << "'! Continuing as if it's empty.\n"; 97 return; // Just continue as if the file were empty 98 } 99 while (In) { 100 std::string Symbol; 101 In >> Symbol; 102 if (!Symbol.empty()) 103 ExternalNames.insert(Symbol); 104 } 105} 106 107static bool shouldInternalize(const GlobalValue &GV, 108 const std::set<std::string> &ExternalNames) { 109 // Function must be defined here 110 if (GV.isDeclaration()) 111 return false; 112 113 // Available externally is really just a "declaration with a body". 114 if (GV.hasAvailableExternallyLinkage()) 115 return false; 116 117 // Assume that dllexported symbols are referenced elsewhere 118 if (GV.hasDLLExportStorageClass()) 119 return false; 120 121 // Already has internal linkage 122 if (GV.hasLocalLinkage()) 123 return false; 124 125 // Marked to keep external? 126 if (ExternalNames.count(GV.getName())) 127 return false; 128 129 return true; 130} 131 132bool InternalizePass::runOnModule(Module &M) { 133 CallGraphWrapperPass *CGPass = getAnalysisIfAvailable<CallGraphWrapperPass>(); 134 CallGraph *CG = CGPass ? &CGPass->getCallGraph() : 0; 135 CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0; 136 bool Changed = false; 137 138 SmallPtrSet<GlobalValue *, 8> Used; 139 collectUsedGlobalVariables(M, Used, false); 140 141 // We must assume that globals in llvm.used have a reference that not even 142 // the linker can see, so we don't internalize them. 143 // For llvm.compiler.used the situation is a bit fuzzy. The assembler and 144 // linker can drop those symbols. If this pass is running as part of LTO, 145 // one might think that it could just drop llvm.compiler.used. The problem 146 // is that even in LTO llvm doesn't see every reference. For example, 147 // we don't see references from function local inline assembly. To be 148 // conservative, we internalize symbols in llvm.compiler.used, but we 149 // keep llvm.compiler.used so that the symbol is not deleted by llvm. 150 for (SmallPtrSet<GlobalValue *, 8>::iterator I = Used.begin(), E = Used.end(); 151 I != E; ++I) { 152 GlobalValue *V = *I; 153 ExternalNames.insert(V->getName()); 154 } 155 156 // Mark all functions not in the api as internal. 157 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) { 158 if (!shouldInternalize(*I, ExternalNames)) 159 continue; 160 161 I->setLinkage(GlobalValue::InternalLinkage); 162 163 if (ExternalNode) 164 // Remove a callgraph edge from the external node to this function. 165 ExternalNode->removeOneAbstractEdgeTo((*CG)[I]); 166 167 Changed = true; 168 ++NumFunctions; 169 DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n"); 170 } 171 172 // Never internalize the llvm.used symbol. It is used to implement 173 // attribute((used)). 174 // FIXME: Shouldn't this just filter on llvm.metadata section?? 175 ExternalNames.insert("llvm.used"); 176 ExternalNames.insert("llvm.compiler.used"); 177 178 // Never internalize anchors used by the machine module info, else the info 179 // won't find them. (see MachineModuleInfo.) 180 ExternalNames.insert("llvm.global_ctors"); 181 ExternalNames.insert("llvm.global_dtors"); 182 ExternalNames.insert("llvm.global.annotations"); 183 184 // Never internalize symbols code-gen inserts. 185 // FIXME: We should probably add this (and the __stack_chk_guard) via some 186 // type of call-back in CodeGen. 187 ExternalNames.insert("__stack_chk_fail"); 188 ExternalNames.insert("__stack_chk_guard"); 189 190 // Mark all global variables with initializers that are not in the api as 191 // internal as well. 192 for (Module::global_iterator I = M.global_begin(), E = M.global_end(); 193 I != E; ++I) { 194 if (!shouldInternalize(*I, ExternalNames)) 195 continue; 196 197 I->setLinkage(GlobalValue::InternalLinkage); 198 Changed = true; 199 ++NumGlobals; 200 DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n"); 201 } 202 203 // Mark all aliases that are not in the api as internal as well. 204 for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end(); 205 I != E; ++I) { 206 if (!shouldInternalize(*I, ExternalNames)) 207 continue; 208 209 I->setLinkage(GlobalValue::InternalLinkage); 210 Changed = true; 211 ++NumAliases; 212 DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n"); 213 } 214 215 return Changed; 216} 217 218ModulePass *llvm::createInternalizePass() { return new InternalizePass(); } 219 220ModulePass *llvm::createInternalizePass(ArrayRef<const char *> ExportList) { 221 return new InternalizePass(ExportList); 222} 223