ShadowStackGC.cpp revision 9f5b2aa7fba203469386acc413c23dd41a713bc9 
1//===-- ShadowStackGC.cpp - GC support for uncooperative targets ----------===//
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 file implements lowering for the llvm.gc* intrinsics for targets that do
11// not natively support them (which includes the C backend). Note that the code
12// generated is not quite as efficient as algorithms which generate stack maps
13// to identify roots.
14//
15// This pass implements the code transformation described in this paper:
16//   "Accurate Garbage Collection in an Uncooperative Environment"
17//   Fergus Henderson, ISMM, 2002
18//
19// In runtime/GC/SemiSpace.cpp is a prototype runtime which is compatible with
20// ShadowStackGC.
21//
22// In order to support this particular transformation, all stack roots are
23// coallocated in the stack. This allows a fully target-independent stack map
24// while introducing only minor runtime overhead.
25//
26//===----------------------------------------------------------------------===//
27
28#define DEBUG_TYPE "shadowstackgc"
29#include "llvm/CodeGen/GCs.h"
30#include "llvm/ADT/StringExtras.h"
31#include "llvm/CodeGen/GCStrategy.h"
32#include "llvm/IntrinsicInst.h"
33#include "llvm/Module.h"
34#include "llvm/Support/Compiler.h"
35#include "llvm/Support/IRBuilder.h"
36
37using namespace llvm;
38
39namespace {
40
41  class VISIBILITY_HIDDEN ShadowStackGC : public GCStrategy {
42    /// RootChain - This is the global linked-list that contains the chain of GC
43    /// roots.
44    GlobalVariable *Head;
45
46    /// StackEntryTy - Abstract type of a link in the shadow stack.
47    ///
48    const StructType *StackEntryTy;
49
50    /// Roots - GC roots in the current function. Each is a pair of the
51    /// intrinsic call and its corresponding alloca.
52    std::vector<std::pair<CallInst*,AllocaInst*> > Roots;
53
54  public:
55    ShadowStackGC();
56
57    bool initializeCustomLowering(Module &M);
58    bool performCustomLowering(Function &F);
59
60  private:
61    bool IsNullValue(Value *V);
62    Constant *GetFrameMap(Function &F);
63    const Type* GetConcreteStackEntryType(Function &F);
64    void CollectRoots(Function &F);
65    static GetElementPtrInst *CreateGEP(LLVMContext *Context,
66                                        IRBuilder<> &B, Value *BasePtr,
67                                        int Idx1, const char *Name);
68    static GetElementPtrInst *CreateGEP(LLVMContext *Context,
69                                        IRBuilder<> &B, Value *BasePtr,
70                                        int Idx1, int Idx2, const char *Name);
71  };
72
73}
74
75static GCRegistry::Add<ShadowStackGC>
76X("shadow-stack", "Very portable GC for uncooperative code generators");
77
78namespace {
79  /// EscapeEnumerator - This is a little algorithm to find all escape points
80  /// from a function so that "finally"-style code can be inserted. In addition
81  /// to finding the existing return and unwind instructions, it also (if
82  /// necessary) transforms any call instructions into invokes and sends them to
83  /// a landing pad.
84  ///
85  /// It's wrapped up in a state machine using the same transform C# uses for
86  /// 'yield return' enumerators, This transform allows it to be non-allocating.
87  class VISIBILITY_HIDDEN EscapeEnumerator {
88    Function &F;
89    const char *CleanupBBName;
90
91    // State.
92    int State;
93    Function::iterator StateBB, StateE;
94    IRBuilder<> Builder;
95
96  public:
97    EscapeEnumerator(Function &F, const char *N = "cleanup")
98      : F(F), CleanupBBName(N), State(0), Builder(*F.getContext()) {}
99
100    IRBuilder<> *Next() {
101      switch (State) {
102      default:
103        return 0;
104
105      case 0:
106        StateBB = F.begin();
107        StateE = F.end();
108        State = 1;
109
110      case 1:
111        // Find all 'return' and 'unwind' instructions.
112        while (StateBB != StateE) {
113          BasicBlock *CurBB = StateBB++;
114
115          // Branches and invokes do not escape, only unwind and return do.
116          TerminatorInst *TI = CurBB->getTerminator();
117          if (!isa<UnwindInst>(TI) && !isa<ReturnInst>(TI))
118            continue;
119
120          Builder.SetInsertPoint(TI->getParent(), TI);
121          return &Builder;
122        }
123
124        State = 2;
125
126        // Find all 'call' instructions.
127        SmallVector<Instruction*,16> Calls;
128        for (Function::iterator BB = F.begin(),
129                                E = F.end(); BB != E; ++BB)
130          for (BasicBlock::iterator II = BB->begin(),
131                                    EE = BB->end(); II != EE; ++II)
132            if (CallInst *CI = dyn_cast<CallInst>(II))
133              if (!CI->getCalledFunction() ||
134                  !CI->getCalledFunction()->getIntrinsicID())
135                Calls.push_back(CI);
136
137        if (Calls.empty())
138          return 0;
139
140        // Create a cleanup block.
141        BasicBlock *CleanupBB = BasicBlock::Create(CleanupBBName, &F);
142        UnwindInst *UI = new UnwindInst(CleanupBB);
143
144        // Transform the 'call' instructions into 'invoke's branching to the
145        // cleanup block. Go in reverse order to make prettier BB names.
146        SmallVector<Value*,16> Args;
147        for (unsigned I = Calls.size(); I != 0; ) {
148          CallInst *CI = cast<CallInst>(Calls[--I]);
149
150          // Split the basic block containing the function call.
151          BasicBlock *CallBB = CI->getParent();
152          BasicBlock *NewBB =
153            CallBB->splitBasicBlock(CI, CallBB->getName() + ".cont");
154
155          // Remove the unconditional branch inserted at the end of CallBB.
156          CallBB->getInstList().pop_back();
157          NewBB->getInstList().remove(CI);
158
159          // Create a new invoke instruction.
160          Args.clear();
161          Args.append(CI->op_begin() + 1, CI->op_end());
162
163          InvokeInst *II = InvokeInst::Create(CI->getOperand(0),
164                                              NewBB, CleanupBB,
165                                              Args.begin(), Args.end(),
166                                              CI->getName(), CallBB);
167          II->setCallingConv(CI->getCallingConv());
168          II->setAttributes(CI->getAttributes());
169          CI->replaceAllUsesWith(II);
170          delete CI;
171        }
172
173        Builder.SetInsertPoint(UI->getParent(), UI);
174        return &Builder;
175      }
176    }
177  };
178}
179
180// -----------------------------------------------------------------------------
181
182void llvm::linkShadowStackGC() { }
183
184ShadowStackGC::ShadowStackGC() : Head(0), StackEntryTy(0) {
185  InitRoots = true;
186  CustomRoots = true;
187}
188
189Constant *ShadowStackGC::GetFrameMap(Function &F) {
190  // doInitialization creates the abstract type of this value.
191  LLVMContext *Context = F.getContext();
192
193  Type *VoidPtr = PointerType::getUnqual(Type::Int8Ty);
194
195  // Truncate the ShadowStackDescriptor if some metadata is null.
196  unsigned NumMeta = 0;
197  SmallVector<Constant*,16> Metadata;
198  for (unsigned I = 0; I != Roots.size(); ++I) {
199    Constant *C = cast<Constant>(Roots[I].first->getOperand(2));
200    if (!C->isNullValue())
201      NumMeta = I + 1;
202    Metadata.push_back(ConstantExpr::getBitCast(C, VoidPtr));
203  }
204
205  Constant *BaseElts[] = {
206    Context->getConstantInt(Type::Int32Ty, Roots.size(), false),
207    Context->getConstantInt(Type::Int32Ty, NumMeta, false),
208  };
209
210  Constant *DescriptorElts[] = {
211    Context->getConstantStruct(BaseElts, 2),
212    Context->getConstantArray(Context->getArrayType(VoidPtr, NumMeta),
213                       Metadata.begin(), NumMeta)
214  };
215
216  Constant *FrameMap = Context->getConstantStruct(DescriptorElts, 2);
217
218  std::string TypeName("gc_map.");
219  TypeName += utostr(NumMeta);
220  F.getParent()->addTypeName(TypeName, FrameMap->getType());
221
222  // FIXME: Is this actually dangerous as WritingAnLLVMPass.html claims? Seems
223  //        that, short of multithreaded LLVM, it should be safe; all that is
224  //        necessary is that a simple Module::iterator loop not be invalidated.
225  //        Appending to the GlobalVariable list is safe in that sense.
226  //
227  //        All of the output passes emit globals last. The ExecutionEngine
228  //        explicitly supports adding globals to the module after
229  //        initialization.
230  //
231  //        Still, if it isn't deemed acceptable, then this transformation needs
232  //        to be a ModulePass (which means it cannot be in the 'llc' pipeline
233  //        (which uses a FunctionPassManager (which segfaults (not asserts) if
234  //        provided a ModulePass))).
235  Constant *GV = new GlobalVariable(*F.getParent(), FrameMap->getType(), true,
236                                    GlobalVariable::InternalLinkage,
237                                    FrameMap, "__gc_" + F.getName());
238
239  Constant *GEPIndices[2] = { Context->getConstantInt(Type::Int32Ty, 0),
240                              Context->getConstantInt(Type::Int32Ty, 0) };
241  return Context->getConstantExprGetElementPtr(GV, GEPIndices, 2);
242}
243
244const Type* ShadowStackGC::GetConcreteStackEntryType(Function &F) {
245  // doInitialization creates the generic version of this type.
246  std::vector<const Type*> EltTys;
247  EltTys.push_back(StackEntryTy);
248  for (size_t I = 0; I != Roots.size(); I++)
249    EltTys.push_back(Roots[I].second->getAllocatedType());
250  Type *Ty = StructType::get(EltTys);
251
252  std::string TypeName("gc_stackentry.");
253  TypeName += F.getName();
254  F.getParent()->addTypeName(TypeName, Ty);
255
256  return Ty;
257}
258
259/// doInitialization - If this module uses the GC intrinsics, find them now. If
260/// not, exit fast.
261bool ShadowStackGC::initializeCustomLowering(Module &M) {
262  // struct FrameMap {
263  //   int32_t NumRoots; // Number of roots in stack frame.
264  //   int32_t NumMeta;  // Number of metadata descriptors. May be < NumRoots.
265  //   void *Meta[];     // May be absent for roots without metadata.
266  // };
267  std::vector<const Type*> EltTys;
268  EltTys.push_back(Type::Int32Ty); // 32 bits is ok up to a 32GB stack frame. :)
269  EltTys.push_back(Type::Int32Ty); // Specifies length of variable length array.
270  StructType *FrameMapTy = StructType::get(EltTys);
271  M.addTypeName("gc_map", FrameMapTy);
272  PointerType *FrameMapPtrTy = PointerType::getUnqual(FrameMapTy);
273
274  // struct StackEntry {
275  //   ShadowStackEntry *Next; // Caller's stack entry.
276  //   FrameMap *Map;          // Pointer to constant FrameMap.
277  //   void *Roots[];          // Stack roots (in-place array, so we pretend).
278  // };
279  OpaqueType *RecursiveTy = OpaqueType::get();
280
281  EltTys.clear();
282  EltTys.push_back(PointerType::getUnqual(RecursiveTy));
283  EltTys.push_back(FrameMapPtrTy);
284  PATypeHolder LinkTyH = StructType::get(EltTys);
285
286  RecursiveTy->refineAbstractTypeTo(LinkTyH.get());
287  StackEntryTy = cast<StructType>(LinkTyH.get());
288  const PointerType *StackEntryPtrTy = PointerType::getUnqual(StackEntryTy);
289  M.addTypeName("gc_stackentry", LinkTyH.get());  // FIXME: Is this safe from
290                                                  //        a FunctionPass?
291
292  // Get the root chain if it already exists.
293  Head = M.getGlobalVariable("llvm_gc_root_chain");
294  if (!Head) {
295    // If the root chain does not exist, insert a new one with linkonce
296    // linkage!
297    Head = new GlobalVariable(M, StackEntryPtrTy, false,
298                              GlobalValue::LinkOnceAnyLinkage,
299                              M.getContext().getNullValue(StackEntryPtrTy),
300                              "llvm_gc_root_chain");
301  } else if (Head->hasExternalLinkage() && Head->isDeclaration()) {
302    Head->setInitializer(M.getContext().getNullValue(StackEntryPtrTy));
303    Head->setLinkage(GlobalValue::LinkOnceAnyLinkage);
304  }
305
306  return true;
307}
308
309bool ShadowStackGC::IsNullValue(Value *V) {
310  if (Constant *C = dyn_cast<Constant>(V))
311    return C->isNullValue();
312  return false;
313}
314
315void ShadowStackGC::CollectRoots(Function &F) {
316  // FIXME: Account for original alignment. Could fragment the root array.
317  //   Approach 1: Null initialize empty slots at runtime. Yuck.
318  //   Approach 2: Emit a map of the array instead of just a count.
319
320  assert(Roots.empty() && "Not cleaned up?");
321
322  SmallVector<std::pair<CallInst*,AllocaInst*>,16> MetaRoots;
323
324  for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
325    for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E;)
326      if (IntrinsicInst *CI = dyn_cast<IntrinsicInst>(II++))
327        if (Function *F = CI->getCalledFunction())
328          if (F->getIntrinsicID() == Intrinsic::gcroot) {
329            std::pair<CallInst*,AllocaInst*> Pair = std::make_pair(
330              CI, cast<AllocaInst>(CI->getOperand(1)->stripPointerCasts()));
331            if (IsNullValue(CI->getOperand(2)))
332              Roots.push_back(Pair);
333            else
334              MetaRoots.push_back(Pair);
335          }
336
337  // Number roots with metadata (usually empty) at the beginning, so that the
338  // FrameMap::Meta array can be elided.
339  Roots.insert(Roots.begin(), MetaRoots.begin(), MetaRoots.end());
340}
341
342GetElementPtrInst *
343ShadowStackGC::CreateGEP(LLVMContext *Context, IRBuilder<> &B, Value *BasePtr,
344                         int Idx, int Idx2, const char *Name) {
345  Value *Indices[] = { Context->getConstantInt(Type::Int32Ty, 0),
346                       Context->getConstantInt(Type::Int32Ty, Idx),
347                       Context->getConstantInt(Type::Int32Ty, Idx2) };
348  Value* Val = B.CreateGEP(BasePtr, Indices, Indices + 3, Name);
349
350  assert(isa<GetElementPtrInst>(Val) && "Unexpected folded constant");
351
352  return dyn_cast<GetElementPtrInst>(Val);
353}
354
355GetElementPtrInst *
356ShadowStackGC::CreateGEP(LLVMContext *Context, IRBuilder<> &B, Value *BasePtr,
357                         int Idx, const char *Name) {
358  Value *Indices[] = { Context->getConstantInt(Type::Int32Ty, 0),
359                       Context->getConstantInt(Type::Int32Ty, Idx) };
360  Value *Val = B.CreateGEP(BasePtr, Indices, Indices + 2, Name);
361
362  assert(isa<GetElementPtrInst>(Val) && "Unexpected folded constant");
363
364  return dyn_cast<GetElementPtrInst>(Val);
365}
366
367/// runOnFunction - Insert code to maintain the shadow stack.
368bool ShadowStackGC::performCustomLowering(Function &F) {
369  LLVMContext *Context = F.getContext();
370
371  // Find calls to llvm.gcroot.
372  CollectRoots(F);
373
374  // If there are no roots in this function, then there is no need to add a
375  // stack map entry for it.
376  if (Roots.empty())
377    return false;
378
379  // Build the constant map and figure the type of the shadow stack entry.
380  Value *FrameMap = GetFrameMap(F);
381  const Type *ConcreteStackEntryTy = GetConcreteStackEntryType(F);
382
383  // Build the shadow stack entry at the very start of the function.
384  BasicBlock::iterator IP = F.getEntryBlock().begin();
385  IRBuilder<> AtEntry(IP->getParent(), IP);
386
387  Instruction *StackEntry   = AtEntry.CreateAlloca(ConcreteStackEntryTy, 0,
388                                                   "gc_frame");
389
390  while (isa<AllocaInst>(IP)) ++IP;
391  AtEntry.SetInsertPoint(IP->getParent(), IP);
392
393  // Initialize the map pointer and load the current head of the shadow stack.
394  Instruction *CurrentHead  = AtEntry.CreateLoad(Head, "gc_currhead");
395  Instruction *EntryMapPtr  = CreateGEP(Context, AtEntry, StackEntry,
396                                        0,1,"gc_frame.map");
397                              AtEntry.CreateStore(FrameMap, EntryMapPtr);
398
399  // After all the allocas...
400  for (unsigned I = 0, E = Roots.size(); I != E; ++I) {
401    // For each root, find the corresponding slot in the aggregate...
402    Value *SlotPtr = CreateGEP(Context, AtEntry, StackEntry, 1 + I, "gc_root");
403
404    // And use it in lieu of the alloca.
405    AllocaInst *OriginalAlloca = Roots[I].second;
406    SlotPtr->takeName(OriginalAlloca);
407    OriginalAlloca->replaceAllUsesWith(SlotPtr);
408  }
409
410  // Move past the original stores inserted by GCStrategy::InitRoots. This isn't
411  // really necessary (the collector would never see the intermediate state at
412  // runtime), but it's nicer not to push the half-initialized entry onto the
413  // shadow stack.
414  while (isa<StoreInst>(IP)) ++IP;
415  AtEntry.SetInsertPoint(IP->getParent(), IP);
416
417  // Push the entry onto the shadow stack.
418  Instruction *EntryNextPtr = CreateGEP(Context, AtEntry,
419                                        StackEntry,0,0,"gc_frame.next");
420  Instruction *NewHeadVal   = CreateGEP(Context, AtEntry,
421                                        StackEntry, 0, "gc_newhead");
422  AtEntry.CreateStore(CurrentHead, EntryNextPtr);
423  AtEntry.CreateStore(NewHeadVal, Head);
424
425  // For each instruction that escapes...
426  EscapeEnumerator EE(F, "gc_cleanup");
427  while (IRBuilder<> *AtExit = EE.Next()) {
428    // Pop the entry from the shadow stack. Don't reuse CurrentHead from
429    // AtEntry, since that would make the value live for the entire function.
430    Instruction *EntryNextPtr2 = CreateGEP(Context, *AtExit, StackEntry, 0, 0,
431                                           "gc_frame.next");
432    Value *SavedHead = AtExit->CreateLoad(EntryNextPtr2, "gc_savedhead");
433                       AtExit->CreateStore(SavedHead, Head);
434  }
435
436  // Delete the original allocas (which are no longer used) and the intrinsic
437  // calls (which are no longer valid). Doing this last avoids invalidating
438  // iterators.
439  for (unsigned I = 0, E = Roots.size(); I != E; ++I) {
440    Roots[I].first->eraseFromParent();
441    Roots[I].second->eraseFromParent();
442  }
443
444  Roots.clear();
445  return true;
446}
447