Removes some deprecated functions, and removes the basic block functi… (

#569)

* Removes some deprecated functions, and removes the basic block function cloning apis, in favor of apis on Arch

* More tweaks for removing unnecessary APIs

* Goodbye dse

* Fix issue

* Tweaks

* Remove some annoying logs

bin/lift/Lift.cpp

@@ -270,7 +270,6 @@ int main(int argc, char *argv[]) {
   // Optimize the module, but with a particular focus on only the functions
   // that we actually lifted.
   remill::OptimizationGuide guide = {};
-  guide.eliminate_dead_stores = true;
   remill::OptimizeModule(arch, module, manager.traces, guide);
 
   // Create a new module in which we will move all the lifted functions. Prepare
@@ -422,7 +421,6 @@ int main(int argc, char *argv[]) {
 
     guide.slp_vectorize = true;
     guide.loop_vectorize = true;
-    guide.eliminate_dead_stores = false;
     remill::OptimizeBareModule(&dest_module, guide);
   }
 

include/remill/Arch/Arch.h

@@ -175,9 +175,27 @@ class Arch {
   // Returns the name of the program counter register.
   virtual std::string_view ProgramCounterRegisterName(void) const = 0;
 
+  // Create a lifted function declaration with name `name` inside of `module`.
+  //
+  // NOTE(pag): This should be called after `PrepareModule` and after the
+  //            semantics have been loaded.
+  llvm::Function *DeclareLiftedFunction(std::string_view name,
+                                        llvm::Module *module) const;
+
+  // Create a lifted function with name `name` inside of `module`.
+  //
+  // NOTE(pag): This should be called after `PrepareModule` and after the
+  //            semantics have been loaded.
+  llvm::Function *DefineLiftedFunction(std::string_view name,
+                                       llvm::Module *module) const;
+
+  // Initialize an empty lifted function with the default variables that it
+  // should contain.
+  void InitializeEmptyLiftedFunction(llvm::Function *func) const;
+
   // Converts an LLVM module object to have the right triple / data layout
-  // information for the target architecture and ensures remill required functions
-  // have the appropriate prototype and internal variables
+  // information for the target architecture and ensures remill required
+  // functions have the appropriate prototype and internal variables
   void PrepareModule(llvm::Module *mod) const;
 
   // Get the state pointer and various other types from the `llvm::LLVMContext`
@@ -293,9 +311,10 @@ class Arch {
   // Populate the table of register information.
   virtual void PopulateRegisterTable(void) const = 0;
 
-  // Populate the `__remill_basic_block` function with variables.
-  virtual void PopulateBasicBlockFunction(llvm::Module *module,
-                                          llvm::Function *bb_func) const = 0;
+  // Populate a just-initialized lifted function function with architecture-
+  // specific variables.
+  virtual void FinishLiftedFunctionInitialization(
+      llvm::Module *module, llvm::Function *bb_func) const = 0;
 
   llvm::Triple BasicTriple(void) const;
 

include/remill/Arch/Runtime/Intrinsics.h

@@ -21,10 +21,6 @@
 
 extern "C" {
 
-// The basic block "template".
-[[gnu::used]] Memory *__remill_basic_block(State &state, addr_t pc,
-                                           Memory *memory);
-
 // Memory read intrinsics.
 [[gnu::used, gnu::const]] extern uint8_t __remill_read_memory_8(Memory *,
                                                                 addr_t);

include/remill/Arch/X86/Runtime/State.h

@@ -689,9 +689,7 @@ static_assert(96 == sizeof(AddressSpace),
 
 // Named the same way as the 64-bit version to keep names the same
 // across architectures. All registers are here, even the 64-bit ones. The
-// 64-bit ones are inaccessible in lifted 32-bit code because they will
-// not be referenced by named variables in the `__remill_basic_block`
-// function.
+// 64-bit ones are not used in lifted 32-bit code.
 struct alignas(8) GPR final {
 
   // Prevents LLVM from casting a `GPR` into an `i64` to access `rax`.

include/remill/BC/Annotate.h

@@ -114,9 +114,6 @@ static bool Remove(llvm::Function *func) {
 template <typename OriginType>
 static void Annotate(llvm::Function *func) {
   auto &C = func->getContext();
-  DLOG(INFO) << "Annotating: " << func->getName().str() << ": "
-             << OriginType::metadata_kind << " -> "
-             << OriginType::metadata_value;
   auto node =
       llvm::MDNode::get(C, llvm::MDString::get(C, OriginType::metadata_value));
   func->setMetadata(OriginType::metadata_kind, node);

include/remill/BC/InstructionLifter.h

@@ -89,8 +89,6 @@ class InstructionLifter {
   void ClearCache(void) const;
 
  protected:
-  friend class TraceLifter;
-
   // Lift an operand to an instruction.
   virtual llvm::Value *LiftOperand(Instruction &inst, llvm::BasicBlock *block,
                                    llvm::Value *state_ptr, llvm::Argument *arg,
@@ -136,6 +134,8 @@ class InstructionLifter {
                        std::string_view reg_name, llvm::ConstantInt *zero);
 
  private:
+  friend class TraceLifter;
+
   InstructionLifter(const InstructionLifter &) = delete;
   InstructionLifter(InstructionLifter &&) noexcept = delete;
   InstructionLifter(void) = delete;

include/remill/BC/Lifter.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 202 Trail of Bits, Inc.
+ * Copyright (c) 2020 Trail of Bits, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.

include/remill/BC/Optimizer.h

@@ -46,7 +46,6 @@ struct OptimizationGuide {
   bool loop_vectorize;
   bool verify_input;
   bool verify_output;
-  bool eliminate_dead_stores;
 };
 
 template <typename T>
@@ -165,11 +164,8 @@ inline static void OptimizeModule(const remill::Arch *arch,
   return OptimizeModule(arch, module, trace_func_gen, guide);
 }
 
-// Optimize a normal module. This might not contain special functions
-// like `__remill_basic_block`.
-//
-// NOTE(pag): It is an error to specify `guide.eliminate_dead_stores` as
-//            `true`.
+// Optimize a normal module. This might not contain special Remill-specific
+// intrinsics functions like `__remill_jump`, etc.
 void OptimizeBareModule(llvm::Module *module, OptimizationGuide guide = {});
 
 inline static void

include/remill/BC/Util.h

@@ -177,15 +177,6 @@ std::string FindSemanticsBitcodeFile(std::string_view arch);
 // Return a pointer to the Nth argument (N=0 is the first argument).
 llvm::Argument *NthArgument(llvm::Function *func, size_t index);
 
-// Returns a pointer to the `__remill_basic_block` function.
-llvm::Function *BasicBlockFunction(llvm::Module *module);
-
-// Return the type of a lifted function.
-//
-// NOTE(pag): Deprecated. Use `remill::Arch::LiftedFunctionType()` instead.
-llvm::FunctionType *LiftedFunctionType(llvm::Module *module)
-    __attribute__((deprecated));
-
 // Return a vector of arguments to pass to a lifted function, where the
 // arguments are derived from `block`.
 std::array<llvm::Value *, kNumBlockArgs>
@@ -200,28 +191,6 @@ using ISelCallback =
     std::function<void(llvm::GlobalVariable *, llvm::Function *)>;
 void ForEachISel(llvm::Module *module, ISelCallback callback);
 
-// Declare a lifted function of the correct type.
-llvm::Function *DeclareLiftedFunction(llvm::Module *module,
-                                      std::string_view name);
-
-// Returns the type of a state pointer.
-//
-// NOTE(pag): Deprecated. Use `remill::Arch::StatePointerType()` instead.
-llvm::PointerType *StatePointerType(llvm::Module *module)
-    __attribute__((deprecated));
-
-// Returns the type of a state pointer.
-//
-// NOTE(pag): Deprecated. Use `remill::Arch::MemoryPointerType()` instead.
-llvm::PointerType *MemoryPointerType(llvm::Module *module)
-    __attribute__((deprecated));
-
-// Returns the type of an address (addr_t in the State.h).
-//
-// NOTE(pag): Deprecated. Use `remill::Arch::AddressType()` instead.
-llvm::IntegerType *AddressType(llvm::Module *module)
-    __attribute__((deprecated));
-
 using ValueMap = std::unordered_map<llvm::Value *, llvm::Value *>;
 using TypeMap = std::unordered_map<llvm::Type *, llvm::Type *>;
 using MDMap = std::unordered_map<llvm::Metadata *, llvm::Metadata *>;
@@ -242,9 +211,6 @@ void CloneFunctionInto(llvm::Function *source_func, llvm::Function *dest_func,
 //       `source_func` into the module of `dest_func`.
 void CloneFunctionInto(llvm::Function *source_func, llvm::Function *dest_func);
 
-// Make `func` a clone of the `__remill_basic_block` function.
-void CloneBlockFunctionInto(llvm::Function *func);
-
 // Returns a list of callers of a specific function.
 std::vector<llvm::CallInst *> CallersOf(llvm::Function *func);
 

lib/Arch/AArch32/Arch.cpp

@@ -162,9 +162,10 @@ void AArch32Arch::PopulateRegisterTable(void) const {
 }
 
 
-// Populate the `__remill_basic_block` function with variables.
-void AArch32Arch::PopulateBasicBlockFunction(llvm::Module *module,
-                                             llvm::Function *bb_func) const {
+// Populate a just-initialized lifted function function with architecture-
+// specific variables.
+void AArch32Arch::FinishLiftedFunctionInitialization(
+    llvm::Module *module, llvm::Function *bb_func) const {
   const auto &dl = module->getDataLayout();
   CHECK_EQ(sizeof(State), dl.getTypeAllocSize(StateStructType()))
       << "Mismatch between size of State type for x86/amd64 and what is in "

lib/Arch/AArch32/Arch.h

@@ -50,9 +50,10 @@ class AArch32Arch final : public Arch {
   // Populate the table of register information.
   void PopulateRegisterTable(void) const final;
 
-  // Populate the `__remill_basic_block` function with variables.
-  void PopulateBasicBlockFunction(llvm::Module *module,
-                                  llvm::Function *bb_func) const final;
+  // Populate a just-initialized lifted function function with architecture-
+  // specific variables.
+  void FinishLiftedFunctionInitialization(
+      llvm::Module *module, llvm::Function *bb_func) const final;
 
  private:
   AArch32Arch(void) = delete;

lib/Arch/AArch32/Runtime/BasicBlock.cpp

@@ -20,15 +20,3 @@
 
 #include "remill/Arch/AArch32/Runtime/State.h"
 #include "remill/Arch/Runtime/Float.h"
-
-extern "C" {
-
-#pragma clang diagnostic push
-#pragma clang diagnostic ignored "-Wunused-variable"
-
-// Instructions will be lifted into clones of this function.
-[[gnu::used]] Memory *__remill_basic_block(State &, addr_t, Memory *);
-
-#pragma clang diagnostic pop
-
-}  // extern C

lib/Arch/AArch64/Arch.cpp

@@ -137,9 +137,10 @@ class AArch64Arch final : public Arch {
   // Populate the table of register information.
   void PopulateRegisterTable(void) const final;
 
-  // Populate the `__remill_basic_block` function with variables.
-  void PopulateBasicBlockFunction(llvm::Module *module,
-                                  llvm::Function *bb_func) const final;
+  // Populate a just-initialized lifted function function with architecture-
+  // specific variables.
+  void FinishLiftedFunctionInitialization(
+      llvm::Module *module, llvm::Function *bb_func) const final;
 
  private:
   AArch64Arch(void) = delete;
@@ -458,9 +459,10 @@ void AArch64Arch::PopulateRegisterTable(void) const {
   REG(TPIDRRO_EL0, sr.tpidrro_el0.qword, u64);
 }
 
-// Populate the `__remill_basic_block` function with variables.
-void AArch64Arch::PopulateBasicBlockFunction(llvm::Module *module,
-                                             llvm::Function *bb_func) const {
+// Populate a just-initialized lifted function function with architecture-
+// specific variables.
+void AArch64Arch::FinishLiftedFunctionInitialization(
+    llvm::Module *module, llvm::Function *bb_func) const {
 
   auto &context = module->getContext();
   auto u32 = llvm::Type::getInt32Ty(context);

lib/Arch/AArch64/Runtime/BasicBlock.cpp

@@ -20,10 +20,3 @@
 
 #include "remill/Arch/AArch64/Runtime/State.h"
 #include "remill/Arch/Runtime/Float.h"
-
-extern "C" {
-
-// Instructions will be lifted into clones of this function.
-[[gnu::used]] Memory *__remill_basic_block(State &, addr_t, Memory *);
-
-}  // extern C