Small refactoring of LLVMBasedAliasAnalysis + allow disabling the CFL…

…-based alias analyses (#626)
secure-software-engineering · Jun 15, 2023 · 263ef58 · 263ef58
1 parent 2b4ec1c
commit 263ef58
Show file tree

Hide file tree

Showing 7 changed files with 77 additions and 47 deletions.
diff --git a/include/phasar/ControlFlow/CallGraphAnalysisType.def b/include/phasar/ControlFlow/CallGraphAnalysisType.def
@@ -16,6 +16,6 @@ CALL_GRAPH_ANALYSIS_TYPE(CHA, "cha", "Class hierarchy analysis")
 CALL_GRAPH_ANALYSIS_TYPE(RTA, "rta", "Rapid type analysis")
 CALL_GRAPH_ANALYSIS_TYPE(DTA, "dta", "Declared type analysis")
 CALL_GRAPH_ANALYSIS_TYPE(VTA, "vta", "Variable type analysis")
-CALL_GRAPH_ANALYSIS_TYPE(OTF, "otf", "On-the-fly analysis based on points-to info")
+CALL_GRAPH_ANALYSIS_TYPE(OTF, "otf", "On-the-fly analysis based on points-to info (default)")
 
 #undef CALL_GRAPH_ANALYSIS_TYPE
diff --git a/include/phasar/PhasarLLVM/Pointer/LLVMBasedAliasAnalysis.h b/include/phasar/PhasarLLVM/Pointer/LLVMBasedAliasAnalysis.h
@@ -13,8 +13,6 @@
 #include "phasar/Pointer/AliasAnalysisType.h"
 
 #include "llvm/Analysis/AliasAnalysis.h"
-#include "llvm/IR/PassManager.h"
-#include "llvm/Passes/PassBuilder.h"
 
 namespace llvm {
 class Value;
@@ -27,23 +25,19 @@ namespace psr {
 class LLVMProjectIRDB;
 
 class LLVMBasedAliasAnalysis {
-private:
-  llvm::PassBuilder PB;
-  llvm::AAManager AA;
-  llvm::FunctionAnalysisManager FAM;
-  llvm::FunctionPassManager FPM;
-  llvm::DenseMap<const llvm::Function *, llvm::AAResults *> AAInfos;
-  AliasAnalysisType PATy;
-
-  [[nodiscard]] bool hasAliasInfo(const llvm::Function &Fun) const;
-
-  void computeAliasInfo(llvm::Function &Fun);
 
 public:
-  LLVMBasedAliasAnalysis(LLVMProjectIRDB &IRDB, bool UseLazyEvaluation = true,
-                         AliasAnalysisType PATy = AliasAnalysisType::CFLAnders);
+  explicit LLVMBasedAliasAnalysis(
+      LLVMProjectIRDB &IRDB, bool UseLazyEvaluation,
+      AliasAnalysisType PATy = AliasAnalysisType::Basic);
 
-  ~LLVMBasedAliasAnalysis() = default;
+  LLVMBasedAliasAnalysis(LLVMBasedAliasAnalysis &&) noexcept = default;
+  LLVMBasedAliasAnalysis &
+  operator=(LLVMBasedAliasAnalysis &&) noexcept = default;
+
+  LLVMBasedAliasAnalysis(const LLVMBasedAliasAnalysis &) = delete;
+  LLVMBasedAliasAnalysis &operator=(const LLVMBasedAliasAnalysis &) = delete;
+  ~LLVMBasedAliasAnalysis();
 
   void print(llvm::raw_ostream &OS = llvm::outs()) const;
 
@@ -54,13 +48,26 @@ class LLVMBasedAliasAnalysis {
     return AAInfos.lookup(F);
   };
 
-  void erase(llvm::Function *F);
+  void erase(llvm::Function *F) noexcept;
 
-  void clear();
+  void clear() noexcept;
 
-  [[nodiscard]] inline AliasAnalysisType getPointerAnalysisType() const {
+  [[nodiscard]] inline AliasAnalysisType
+  getPointerAnalysisType() const noexcept {
     return PATy;
   };
+
+private:
+  [[nodiscard]] bool hasAliasInfo(const llvm::Function &Fun) const;
+
+  void computeAliasInfo(llvm::Function &Fun);
+
+  // -- data members
+
+  struct Impl;
+  std::unique_ptr<Impl> PImpl;
+  AliasAnalysisType PATy;
+  llvm::DenseMap<const llvm::Function *, llvm::AAResults *> AAInfos;
 };
 
 } // namespace psr

diff --git a/include/phasar/PhasarLLVM/Utils/BasicBlockOrdering.h b/include/phasar/PhasarLLVM/Utils/BasicBlockOrdering.h
@@ -12,6 +12,7 @@
 
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/FunctionExtras.h"
+#include "llvm/IR/Dominators.h"
 
 #include <memory>
 #include <type_traits>

diff --git a/include/phasar/Pointer/AliasAnalysisType.def b/include/phasar/Pointer/AliasAnalysisType.def
@@ -11,8 +11,9 @@
 #define ALIAS_ANALYSIS_TYPE(NAME, CMDFLAG, DESC)
 #endif
 
+ALIAS_ANALYSIS_TYPE(Basic, "basic", "Basic LLVM alias resolving based on simple, local properties")
 ALIAS_ANALYSIS_TYPE(CFLSteens, "cflsteens", "Steensgaard-style alias analysis (equality-based)")
-ALIAS_ANALYSIS_TYPE(CFLAnders, "cflanders", "Andersen-style alias analysis (subset-based)")
+ALIAS_ANALYSIS_TYPE(CFLAnders, "cflanders", "Andersen-style alias analysis (subset-based) (default)")
 ALIAS_ANALYSIS_TYPE(PointsTo, "points-to", "Alias-information based on (external) points-to information")
 
 #undef ALIAS_ANALYSIS_TYPE
diff --git a/lib/PhasarLLVM/Pointer/LLVMAliasSet.cpp b/lib/PhasarLLVM/Pointer/LLVMAliasSet.cpp
@@ -13,6 +13,7 @@
 #include "phasar/PhasarLLVM/Pointer/LLVMAliasInfo.h"
 #include "phasar/PhasarLLVM/Pointer/LLVMPointsToUtils.h"
 #include "phasar/PhasarLLVM/Utils/LLVMShorthands.h"
+#include "phasar/Pointer/AliasAnalysisType.h"
 #include "phasar/Utils/BoxedPointer.h"
 #include "phasar/Utils/Logger.h"
 #include "phasar/Utils/NlohmannLogging.h"
@@ -96,7 +97,7 @@ LLVMAliasSet::LLVMAliasSet(LLVMProjectIRDB *IRDB, bool UseLazyEvaluation,
 
 LLVMAliasSet::LLVMAliasSet(LLVMProjectIRDB *IRDB,
                            const nlohmann::json &SerializedPTS)
-    : PTA(*IRDB) {
+    : PTA(*IRDB, true) {
   assert(IRDB != nullptr);
   // Assume, we already have validated the json schema
 

diff --git a/lib/PhasarLLVM/Pointer/LLVMBasedAliasAnalysis.cpp b/lib/PhasarLLVM/Pointer/LLVMBasedAliasAnalysis.cpp
@@ -11,6 +11,7 @@
 
 #include "phasar/PhasarLLVM/DB/LLVMProjectIRDB.h"
 #include "phasar/PhasarLLVM/Pointer/LLVMPointsToUtils.h"
+#include "phasar/Pointer/AliasAnalysisType.h"
 
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallVector.h"
@@ -19,6 +20,7 @@
 #include "llvm/Analysis/BasicAliasAnalysis.h"
 #include "llvm/Analysis/CFLAndersAliasAnalysis.h"
 #include "llvm/Analysis/CFLSteensAliasAnalysis.h"
+#include "llvm/Analysis/ScopedNoAliasAA.h"
 #include "llvm/Analysis/TypeBasedAliasAnalysis.h"
 #include "llvm/IR/Argument.h"
 #include "llvm/IR/BasicBlock.h"
@@ -28,11 +30,18 @@
 #include "llvm/IR/PassManager.h"
 #include "llvm/IR/Value.h"
 #include "llvm/IR/Verifier.h"
+#include "llvm/Passes/PassBuilder.h"
 
 using namespace psr;
 
 namespace psr {
 
+struct LLVMBasedAliasAnalysis::Impl {
+  llvm::PassBuilder PB{};
+  llvm::FunctionAnalysisManager FAM{};
+  llvm::FunctionPassManager FPM{};
+};
+
 static void printResults(llvm::AliasResult AR, bool P, const llvm::Value *V1,
                          const llvm::Value *V2, const llvm::Module *M) {
   if (P) {
@@ -88,44 +97,54 @@ bool LLVMBasedAliasAnalysis::hasAliasInfo(const llvm::Function &Fun) const {
 }
 
 void LLVMBasedAliasAnalysis::computeAliasInfo(llvm::Function &Fun) {
-  llvm::PreservedAnalyses PA = FPM.run(Fun, FAM);
-  llvm::AAResults &AAR = FAM.getResult<llvm::AAManager>(Fun);
+  assert(PImpl != nullptr);
+  llvm::PreservedAnalyses PA = PImpl->FPM.run(Fun, PImpl->FAM);
+  llvm::AAResults &AAR = PImpl->FAM.getResult<llvm::AAManager>(Fun);
   AAInfos.insert(std::make_pair(&Fun, &AAR));
 }
 
-void LLVMBasedAliasAnalysis::erase(llvm::Function *F) {
+void LLVMBasedAliasAnalysis::erase(llvm::Function *F) noexcept {
   // after we clear all stuff, we need to set it up for the next function-wise
   // analysis
   AAInfos.erase(F);
-  FAM.clear(*F, F->getName());
+  PImpl->FAM.clear(*F, F->getName());
 }
 
-void LLVMBasedAliasAnalysis::clear() {
+void LLVMBasedAliasAnalysis::clear() noexcept {
   AAInfos.clear();
-  FAM.clear();
+  PImpl->FAM.clear();
 }
 
 LLVMBasedAliasAnalysis::LLVMBasedAliasAnalysis(LLVMProjectIRDB &IRDB,
                                                bool UseLazyEvaluation,
                                                AliasAnalysisType PATy)
-    : PATy(PATy) {
-  AA.registerFunctionAnalysis<llvm::BasicAA>();
-  switch (PATy) {
-  case AliasAnalysisType::CFLAnders:
-    AA.registerFunctionAnalysis<llvm::CFLAndersAA>();
-    break;
-  case AliasAnalysisType::CFLSteens:
-    AA.registerFunctionAnalysis<llvm::CFLSteensAA>();
-    break;
-  default:
-    break;
-  }
-  AA.registerFunctionAnalysis<llvm::TypeBasedAA>();
-  FAM.registerPass([&] { return std::move(AA); });
-  PB.registerFunctionAnalyses(FAM);
-  llvm::FunctionPassManager FPM;
-  // Always verify the input.
-  FPM.addPass(llvm::VerifierPass());
+    : PImpl(new Impl{}), PATy(PATy) {
+
+  PImpl->FAM.registerPass([&] {
+    llvm::AAManager AA;
+    switch (PATy) {
+    case AliasAnalysisType::CFLAnders:
+      AA.registerFunctionAnalysis<llvm::CFLAndersAA>();
+      break;
+    case AliasAnalysisType::CFLSteens:
+      AA.registerFunctionAnalysis<llvm::CFLSteensAA>();
+      break;
+    case AliasAnalysisType::Basic:
+      [[fallthrough]];
+    default:
+      break;
+    }
+    // Note: The order of the alias analyses is important. See LLVM's source
+    // code for reference (e.g. registerAAAnalyses() in
+    // llvm/CodeGen/CodeGenPassBuilder.h)
+    //
+    AA.registerFunctionAnalysis<llvm::TypeBasedAA>();
+    AA.registerFunctionAnalysis<llvm::ScopedNoAliasAA>();
+    AA.registerFunctionAnalysis<llvm::BasicAA>();
+    return AA;
+  });
+  PImpl->PB.registerFunctionAnalyses(PImpl->FAM);
+
   if (!UseLazyEvaluation) {
     for (auto &F : *IRDB.getModule()) {
       if (!F.isDeclaration()) {
@@ -135,6 +154,8 @@ LLVMBasedAliasAnalysis::LLVMBasedAliasAnalysis(LLVMProjectIRDB &IRDB,
   }
 }
 
+LLVMBasedAliasAnalysis::~LLVMBasedAliasAnalysis() = default;
+
 void LLVMBasedAliasAnalysis::print(llvm::raw_ostream &OS) const {
   OS << "Points-to Info:\n";
   for (const auto &[Fn, AA] : AAInfos) {

diff --git a/lib/PhasarLLVM/Utils/BasicBlockOrdering.cpp b/lib/PhasarLLVM/Utils/BasicBlockOrdering.cpp
@@ -2,7 +2,6 @@
 
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/IR/BasicBlock.h"
-#include "llvm/IR/Dominators.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Instruction.h"