Internalize.cpp revision 0fb771667eb2f8a5cd623ce0ee3775b4d1416576
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//===----------------------------------------------------------------------===// 15 16#define DEBUG_TYPE "internalize" 17#include "llvm/Transforms/IPO.h" 18#include "llvm/ADT/SmallPtrSet.h" 19#include "llvm/ADT/Statistic.h" 20#include "llvm/Analysis/CallGraph.h" 21#include "llvm/IR/Module.h" 22#include "llvm/Pass.h" 23#include "llvm/Support/CommandLine.h" 24#include "llvm/Support/Debug.h" 25#include "llvm/Support/raw_ostream.h" 26#include "llvm/Transforms/Utils/ModuleUtils.h" 27#include <fstream> 28#include <set> 29using namespace llvm; 30 31STATISTIC(NumAliases , "Number of aliases internalized"); 32STATISTIC(NumFunctions, "Number of functions internalized"); 33STATISTIC(NumGlobals , "Number of global vars internalized"); 34 35// APIFile - A file which contains a list of symbols that should not be marked 36// external. 37static cl::opt<std::string> 38APIFile("internalize-public-api-file", cl::value_desc("filename"), 39 cl::desc("A file containing list of symbol names to preserve")); 40 41// APIList - A list of symbols that should not be marked internal. 42static cl::list<std::string> 43APIList("internalize-public-api-list", cl::value_desc("list"), 44 cl::desc("A list of symbol names to preserve"), 45 cl::CommaSeparated); 46 47namespace { 48 class InternalizePass : public ModulePass { 49 std::set<std::string> ExternalNames; 50 public: 51 static char ID; // Pass identification, replacement for typeid 52 explicit InternalizePass(); 53 explicit InternalizePass(ArrayRef<const char *> exportList); 54 void LoadFile(const char *Filename); 55 virtual bool runOnModule(Module &M); 56 57 virtual void getAnalysisUsage(AnalysisUsage &AU) const { 58 AU.setPreservesCFG(); 59 AU.addPreserved<CallGraph>(); 60 } 61 }; 62} // end anonymous namespace 63 64char InternalizePass::ID = 0; 65INITIALIZE_PASS(InternalizePass, "internalize", 66 "Internalize Global Symbols", false, false) 67 68InternalizePass::InternalizePass() 69 : ModulePass(ID) { 70 initializeInternalizePassPass(*PassRegistry::getPassRegistry()); 71 if (!APIFile.empty()) // If a filename is specified, use it. 72 LoadFile(APIFile.c_str()); 73 if (!APIList.empty()) // If a list is specified, use it as well. 74 ExternalNames.insert(APIList.begin(), APIList.end()); 75} 76 77InternalizePass::InternalizePass(ArrayRef<const char *> exportList) 78 : ModulePass(ID){ 79 initializeInternalizePassPass(*PassRegistry::getPassRegistry()); 80 for(ArrayRef<const char *>::const_iterator itr = exportList.begin(); 81 itr != exportList.end(); itr++) { 82 ExternalNames.insert(*itr); 83 } 84} 85 86void InternalizePass::LoadFile(const char *Filename) { 87 // Load the APIFile... 88 std::ifstream In(Filename); 89 if (!In.good()) { 90 errs() << "WARNING: Internalize couldn't load file '" << Filename 91 << "'! Continuing as if it's empty.\n"; 92 return; // Just continue as if the file were empty 93 } 94 while (In) { 95 std::string Symbol; 96 In >> Symbol; 97 if (!Symbol.empty()) 98 ExternalNames.insert(Symbol); 99 } 100} 101 102static bool shouldInternalize(const GlobalValue &GV, 103 const std::set<std::string> &ExternalNames) { 104 // Function must be defined here 105 if (GV.isDeclaration()) 106 return false; 107 108 // Available externally is really just a "declaration with a body". 109 if (GV.hasAvailableExternallyLinkage()) 110 return false; 111 112 // Already has internal linkage 113 if (GV.hasLocalLinkage()) 114 return false; 115 116 // Marked to keep external? 117 if (ExternalNames.count(GV.getName())) 118 return false; 119 120 return true; 121} 122 123bool InternalizePass::runOnModule(Module &M) { 124 CallGraph *CG = getAnalysisIfAvailable<CallGraph>(); 125 CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0; 126 bool Changed = false; 127 128 SmallPtrSet<GlobalValue *, 8> Used; 129 collectUsedGlobalVariables(M, Used, false); 130 131 // We must assume that globals in llvm.used have a reference that not even 132 // the linker can see, so we don't internalize them. 133 // For llvm.compiler.used the situation is a bit fuzzy. The assembler and 134 // linker can drop those symbols. If this pass is running as part of LTO, 135 // one might think that it could just drop llvm.compiler.used. The problem 136 // is that even in LTO llvm doesn't see every reference. For example, 137 // we don't see references from function local inline assembly. To be 138 // conservative, we internalize symbols in llvm.compiler.used, but we 139 // keep llvm.compiler.used so that the symbol is not deleted by llvm. 140 for (SmallPtrSet<GlobalValue *, 8>::iterator I = Used.begin(), E = Used.end(); 141 I != E; ++I) { 142 GlobalValue *V = *I; 143 ExternalNames.insert(V->getName()); 144 } 145 146 // Mark all functions not in the api as internal. 147 // FIXME: maybe use private linkage? 148 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) { 149 if (!shouldInternalize(*I, ExternalNames)) 150 continue; 151 152 I->setLinkage(GlobalValue::InternalLinkage); 153 154 if (ExternalNode) 155 // Remove a callgraph edge from the external node to this function. 156 ExternalNode->removeOneAbstractEdgeTo((*CG)[I]); 157 158 Changed = true; 159 ++NumFunctions; 160 DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n"); 161 } 162 163 // Never internalize the llvm.used symbol. It is used to implement 164 // attribute((used)). 165 // FIXME: Shouldn't this just filter on llvm.metadata section?? 166 ExternalNames.insert("llvm.used"); 167 ExternalNames.insert("llvm.compiler.used"); 168 169 // Never internalize anchors used by the machine module info, else the info 170 // won't find them. (see MachineModuleInfo.) 171 ExternalNames.insert("llvm.global_ctors"); 172 ExternalNames.insert("llvm.global_dtors"); 173 ExternalNames.insert("llvm.global.annotations"); 174 175 // Never internalize symbols code-gen inserts. 176 // FIXME: We should probably add this (and the __stack_chk_guard) via some 177 // type of call-back in CodeGen. 178 ExternalNames.insert("__stack_chk_fail"); 179 ExternalNames.insert("__stack_chk_guard"); 180 181 // Mark all global variables with initializers that are not in the api as 182 // internal as well. 183 // FIXME: maybe use private linkage? 184 for (Module::global_iterator I = M.global_begin(), E = M.global_end(); 185 I != E; ++I) { 186 if (!shouldInternalize(*I, ExternalNames)) 187 continue; 188 189 I->setLinkage(GlobalValue::InternalLinkage); 190 Changed = true; 191 ++NumGlobals; 192 DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n"); 193 } 194 195 // Mark all aliases that are not in the api as internal as well. 196 for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end(); 197 I != E; ++I) { 198 if (!shouldInternalize(*I, ExternalNames)) 199 continue; 200 201 I->setLinkage(GlobalValue::InternalLinkage); 202 Changed = true; 203 ++NumAliases; 204 DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n"); 205 } 206 207 return Changed; 208} 209 210ModulePass *llvm::createInternalizePass() { 211 return new InternalizePass(); 212} 213 214ModulePass *llvm::createInternalizePass(ArrayRef<const char *> el) { 215 return new InternalizePass(el); 216} 217