1 //===-- MemorySanitizer.cpp - detector of uninitialized reads -------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 /// This file is a part of MemorySanitizer, a detector of uninitialized
13 /// The algorithm of the tool is similar to Memcheck
14 /// (http://goo.gl/QKbem). We associate a few shadow bits with every
15 /// byte of the application memory, poison the shadow of the malloc-ed
16 /// or alloca-ed memory, load the shadow bits on every memory read,
17 /// propagate the shadow bits through some of the arithmetic
18 /// instruction (including MOV), store the shadow bits on every memory
19 /// write, report a bug on some other instructions (e.g. JMP) if the
20 /// associated shadow is poisoned.
22 /// But there are differences too. The first and the major one:
23 /// compiler instrumentation instead of binary instrumentation. This
24 /// gives us much better register allocation, possible compiler
25 /// optimizations and a fast start-up. But this brings the major issue
26 /// as well: msan needs to see all program events, including system
27 /// calls and reads/writes in system libraries, so we either need to
28 /// compile *everything* with msan or use a binary translation
29 /// component (e.g. DynamoRIO) to instrument pre-built libraries.
30 /// Another difference from Memcheck is that we use 8 shadow bits per
31 /// byte of application memory and use a direct shadow mapping. This
32 /// greatly simplifies the instrumentation code and avoids races on
33 /// shadow updates (Memcheck is single-threaded so races are not a
34 /// concern there. Memcheck uses 2 shadow bits per byte with a slow
35 /// path storage that uses 8 bits per byte).
37 /// The default value of shadow is 0, which means "clean" (not poisoned).
39 /// Every module initializer should call __msan_init to ensure that the
40 /// shadow memory is ready. On error, __msan_warning is called. Since
41 /// parameters and return values may be passed via registers, we have a
42 /// specialized thread-local shadow for return values
43 /// (__msan_retval_tls) and parameters (__msan_param_tls).
47 /// MemorySanitizer can track origins (allocation points) of all uninitialized
48 /// values. This behavior is controlled with a flag (msan-track-origins) and is
49 /// disabled by default.
51 /// Origins are 4-byte values created and interpreted by the runtime library.
52 /// They are stored in a second shadow mapping, one 4-byte value for 4 bytes
53 /// of application memory. Propagation of origins is basically a bunch of
54 /// "select" instructions that pick the origin of a dirty argument, if an
55 /// instruction has one.
57 /// Every 4 aligned, consecutive bytes of application memory have one origin
58 /// value associated with them. If these bytes contain uninitialized data
59 /// coming from 2 different allocations, the last store wins. Because of this,
60 /// MemorySanitizer reports can show unrelated origins, but this is unlikely in
63 /// Origins are meaningless for fully initialized values, so MemorySanitizer
64 /// avoids storing origin to memory when a fully initialized value is stored.
65 /// This way it avoids needless overwritting origin of the 4-byte region on
66 /// a short (i.e. 1 byte) clean store, and it is also good for performance.
70 /// Ideally, every atomic store of application value should update the
71 /// corresponding shadow location in an atomic way. Unfortunately, atomic store
72 /// of two disjoint locations can not be done without severe slowdown.
74 /// Therefore, we implement an approximation that may err on the safe side.
75 /// In this implementation, every atomically accessed location in the program
76 /// may only change from (partially) uninitialized to fully initialized, but
77 /// not the other way around. We load the shadow _after_ the application load,
78 /// and we store the shadow _before_ the app store. Also, we always store clean
79 /// shadow (if the application store is atomic). This way, if the store-load
80 /// pair constitutes a happens-before arc, shadow store and load are correctly
81 /// ordered such that the load will get either the value that was stored, or
82 /// some later value (which is always clean).
84 /// This does not work very well with Compare-And-Swap (CAS) and
85 /// Read-Modify-Write (RMW) operations. To follow the above logic, CAS and RMW
86 /// must store the new shadow before the app operation, and load the shadow
87 /// after the app operation. Computers don't work this way. Current
88 /// implementation ignores the load aspect of CAS/RMW, always returning a clean
89 /// value. It implements the store part as a simple atomic store by storing a
92 //===----------------------------------------------------------------------===//
94 #include "llvm/Transforms/Instrumentation.h"
95 #include "llvm/ADT/DepthFirstIterator.h"
96 #include "llvm/ADT/SmallString.h"
97 #include "llvm/ADT/SmallVector.h"
98 #include "llvm/ADT/StringExtras.h"
99 #include "llvm/ADT/Triple.h"
100 #include "llvm/IR/DataLayout.h"
101 #include "llvm/IR/Function.h"
102 #include "llvm/IR/IRBuilder.h"
103 #include "llvm/IR/InlineAsm.h"
104 #include "llvm/IR/InstVisitor.h"
105 #include "llvm/IR/IntrinsicInst.h"
106 #include "llvm/IR/LLVMContext.h"
107 #include "llvm/IR/MDBuilder.h"
108 #include "llvm/IR/Module.h"
109 #include "llvm/IR/Type.h"
110 #include "llvm/IR/ValueMap.h"
111 #include "llvm/Support/CommandLine.h"
112 #include "llvm/Support/Compiler.h"
113 #include "llvm/Support/Debug.h"
114 #include "llvm/Support/raw_ostream.h"
115 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
116 #include "llvm/Transforms/Utils/Local.h"
117 #include "llvm/Transforms/Utils/ModuleUtils.h"
119 using namespace llvm;
121 #define DEBUG_TYPE "msan"
123 static const uint64_t kShadowMask32 = 1ULL << 31;
124 static const uint64_t kShadowMask64 = 1ULL << 46;
125 static const uint64_t kOriginOffset32 = 1ULL << 30;
126 static const uint64_t kOriginOffset64 = 1ULL << 45;
127 static const unsigned kMinOriginAlignment = 4;
128 static const unsigned kShadowTLSAlignment = 8;
130 // These constants must be kept in sync with the ones in msan.h.
131 static const unsigned kParamTLSSize = 800;
132 static const unsigned kRetvalTLSSize = 800;
134 // Accesses sizes are powers of two: 1, 2, 4, 8.
135 static const size_t kNumberOfAccessSizes = 4;
137 /// \brief Track origins of uninitialized values.
139 /// Adds a section to MemorySanitizer report that points to the allocation
140 /// (stack or heap) the uninitialized bits came from originally.
141 static cl::opt<int> ClTrackOrigins("msan-track-origins",
142 cl::desc("Track origins (allocation sites) of poisoned memory"),
143 cl::Hidden, cl::init(0));
144 static cl::opt<bool> ClKeepGoing("msan-keep-going",
145 cl::desc("keep going after reporting a UMR"),
146 cl::Hidden, cl::init(false));
147 static cl::opt<bool> ClPoisonStack("msan-poison-stack",
148 cl::desc("poison uninitialized stack variables"),
149 cl::Hidden, cl::init(true));
150 static cl::opt<bool> ClPoisonStackWithCall("msan-poison-stack-with-call",
151 cl::desc("poison uninitialized stack variables with a call"),
152 cl::Hidden, cl::init(false));
153 static cl::opt<int> ClPoisonStackPattern("msan-poison-stack-pattern",
154 cl::desc("poison uninitialized stack variables with the given patter"),
155 cl::Hidden, cl::init(0xff));
156 static cl::opt<bool> ClPoisonUndef("msan-poison-undef",
157 cl::desc("poison undef temps"),
158 cl::Hidden, cl::init(true));
160 static cl::opt<bool> ClHandleICmp("msan-handle-icmp",
161 cl::desc("propagate shadow through ICmpEQ and ICmpNE"),
162 cl::Hidden, cl::init(true));
164 static cl::opt<bool> ClHandleICmpExact("msan-handle-icmp-exact",
165 cl::desc("exact handling of relational integer ICmp"),
166 cl::Hidden, cl::init(false));
168 // This flag controls whether we check the shadow of the address
169 // operand of load or store. Such bugs are very rare, since load from
170 // a garbage address typically results in SEGV, but still happen
171 // (e.g. only lower bits of address are garbage, or the access happens
172 // early at program startup where malloc-ed memory is more likely to
173 // be zeroed. As of 2012-08-28 this flag adds 20% slowdown.
174 static cl::opt<bool> ClCheckAccessAddress("msan-check-access-address",
175 cl::desc("report accesses through a pointer which has poisoned shadow"),
176 cl::Hidden, cl::init(true));
178 static cl::opt<bool> ClDumpStrictInstructions("msan-dump-strict-instructions",
179 cl::desc("print out instructions with default strict semantics"),
180 cl::Hidden, cl::init(false));
182 static cl::opt<int> ClInstrumentationWithCallThreshold(
183 "msan-instrumentation-with-call-threshold",
185 "If the function being instrumented requires more than "
186 "this number of checks and origin stores, use callbacks instead of "
187 "inline checks (-1 means never use callbacks)."),
188 cl::Hidden, cl::init(3500));
190 // Experimental. Wraps all indirect calls in the instrumented code with
191 // a call to the given function. This is needed to assist the dynamic
192 // helper tool (MSanDR) to regain control on transition between instrumented and
193 // non-instrumented code.
194 static cl::opt<std::string> ClWrapIndirectCalls("msan-wrap-indirect-calls",
195 cl::desc("Wrap indirect calls with a given function"),
198 static cl::opt<bool> ClWrapIndirectCallsFast("msan-wrap-indirect-calls-fast",
199 cl::desc("Do not wrap indirect calls with target in the same module"),
200 cl::Hidden, cl::init(true));
202 // This is an experiment to enable handling of cases where shadow is a non-zero
203 // compile-time constant. For some unexplainable reason they were silently
204 // ignored in the instrumentation.
205 static cl::opt<bool> ClCheckConstantShadow("msan-check-constant-shadow",
206 cl::desc("Insert checks for constant shadow values"),
207 cl::Hidden, cl::init(false));
211 /// \brief An instrumentation pass implementing detection of uninitialized
214 /// MemorySanitizer: instrument the code in module to find
215 /// uninitialized reads.
216 class MemorySanitizer : public FunctionPass {
218 MemorySanitizer(int TrackOrigins = 0)
220 TrackOrigins(std::max(TrackOrigins, (int)ClTrackOrigins)),
223 WrapIndirectCalls(!ClWrapIndirectCalls.empty()) {}
224 const char *getPassName() const override { return "MemorySanitizer"; }
225 bool runOnFunction(Function &F) override;
226 bool doInitialization(Module &M) override;
227 static char ID; // Pass identification, replacement for typeid.
230 void initializeCallbacks(Module &M);
232 /// \brief Track origins (allocation points) of uninitialized values.
235 const DataLayout *DL;
239 /// \brief Thread-local shadow storage for function parameters.
240 GlobalVariable *ParamTLS;
241 /// \brief Thread-local origin storage for function parameters.
242 GlobalVariable *ParamOriginTLS;
243 /// \brief Thread-local shadow storage for function return value.
244 GlobalVariable *RetvalTLS;
245 /// \brief Thread-local origin storage for function return value.
246 GlobalVariable *RetvalOriginTLS;
247 /// \brief Thread-local shadow storage for in-register va_arg function
248 /// parameters (x86_64-specific).
249 GlobalVariable *VAArgTLS;
250 /// \brief Thread-local shadow storage for va_arg overflow area
251 /// (x86_64-specific).
252 GlobalVariable *VAArgOverflowSizeTLS;
253 /// \brief Thread-local space used to pass origin value to the UMR reporting
255 GlobalVariable *OriginTLS;
257 GlobalVariable *MsandrModuleStart;
258 GlobalVariable *MsandrModuleEnd;
260 /// \brief The run-time callback to print a warning.
262 // These arrays are indexed by log2(AccessSize).
263 Value *MaybeWarningFn[kNumberOfAccessSizes];
264 Value *MaybeStoreOriginFn[kNumberOfAccessSizes];
266 /// \brief Run-time helper that generates a new origin value for a stack
268 Value *MsanSetAllocaOrigin4Fn;
269 /// \brief Run-time helper that poisons stack on function entry.
270 Value *MsanPoisonStackFn;
271 /// \brief Run-time helper that records a store (or any event) of an
272 /// uninitialized value and returns an updated origin id encoding this info.
273 Value *MsanChainOriginFn;
274 /// \brief MSan runtime replacements for memmove, memcpy and memset.
275 Value *MemmoveFn, *MemcpyFn, *MemsetFn;
277 /// \brief Address mask used in application-to-shadow address calculation.
278 /// ShadowAddr is computed as ApplicationAddr & ~ShadowMask.
280 /// \brief Offset of the origin shadow from the "normal" shadow.
281 /// OriginAddr is computed as (ShadowAddr + OriginOffset) & ~3ULL
282 uint64_t OriginOffset;
283 /// \brief Branch weights for error reporting.
284 MDNode *ColdCallWeights;
285 /// \brief Branch weights for origin store.
286 MDNode *OriginStoreWeights;
287 /// \brief An empty volatile inline asm that prevents callback merge.
290 bool WrapIndirectCalls;
291 /// \brief Run-time wrapper for indirect calls.
292 Value *IndirectCallWrapperFn;
293 // Argument and return type of IndirectCallWrapperFn: void (*f)(void).
294 Type *AnyFunctionPtrTy;
296 friend struct MemorySanitizerVisitor;
297 friend struct VarArgAMD64Helper;
301 char MemorySanitizer::ID = 0;
302 INITIALIZE_PASS(MemorySanitizer, "msan",
303 "MemorySanitizer: detects uninitialized reads.",
306 FunctionPass *llvm::createMemorySanitizerPass(int TrackOrigins) {
307 return new MemorySanitizer(TrackOrigins);
310 /// \brief Create a non-const global initialized with the given string.
312 /// Creates a writable global for Str so that we can pass it to the
313 /// run-time lib. Runtime uses first 4 bytes of the string to store the
314 /// frame ID, so the string needs to be mutable.
315 static GlobalVariable *createPrivateNonConstGlobalForString(Module &M,
317 Constant *StrConst = ConstantDataArray::getString(M.getContext(), Str);
318 return new GlobalVariable(M, StrConst->getType(), /*isConstant=*/false,
319 GlobalValue::PrivateLinkage, StrConst, "");
323 /// \brief Insert extern declaration of runtime-provided functions and globals.
324 void MemorySanitizer::initializeCallbacks(Module &M) {
325 // Only do this once.
330 // Create the callback.
331 // FIXME: this function should have "Cold" calling conv,
332 // which is not yet implemented.
333 StringRef WarningFnName = ClKeepGoing ? "__msan_warning"
334 : "__msan_warning_noreturn";
335 WarningFn = M.getOrInsertFunction(WarningFnName, IRB.getVoidTy(), nullptr);
337 for (size_t AccessSizeIndex = 0; AccessSizeIndex < kNumberOfAccessSizes;
339 unsigned AccessSize = 1 << AccessSizeIndex;
340 std::string FunctionName = "__msan_maybe_warning_" + itostr(AccessSize);
341 MaybeWarningFn[AccessSizeIndex] = M.getOrInsertFunction(
342 FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
343 IRB.getInt32Ty(), nullptr);
345 FunctionName = "__msan_maybe_store_origin_" + itostr(AccessSize);
346 MaybeStoreOriginFn[AccessSizeIndex] = M.getOrInsertFunction(
347 FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
348 IRB.getInt8PtrTy(), IRB.getInt32Ty(), nullptr);
351 MsanSetAllocaOrigin4Fn = M.getOrInsertFunction(
352 "__msan_set_alloca_origin4", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy,
353 IRB.getInt8PtrTy(), IntptrTy, nullptr);
355 M.getOrInsertFunction("__msan_poison_stack", IRB.getVoidTy(),
356 IRB.getInt8PtrTy(), IntptrTy, nullptr);
357 MsanChainOriginFn = M.getOrInsertFunction(
358 "__msan_chain_origin", IRB.getInt32Ty(), IRB.getInt32Ty(), nullptr);
359 MemmoveFn = M.getOrInsertFunction(
360 "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
361 IRB.getInt8PtrTy(), IntptrTy, nullptr);
362 MemcpyFn = M.getOrInsertFunction(
363 "__msan_memcpy", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
365 MemsetFn = M.getOrInsertFunction(
366 "__msan_memset", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt32Ty(),
370 RetvalTLS = new GlobalVariable(
371 M, ArrayType::get(IRB.getInt64Ty(), kRetvalTLSSize / 8), false,
372 GlobalVariable::ExternalLinkage, nullptr, "__msan_retval_tls", nullptr,
373 GlobalVariable::InitialExecTLSModel);
374 RetvalOriginTLS = new GlobalVariable(
375 M, OriginTy, false, GlobalVariable::ExternalLinkage, nullptr,
376 "__msan_retval_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);
378 ParamTLS = new GlobalVariable(
379 M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
380 GlobalVariable::ExternalLinkage, nullptr, "__msan_param_tls", nullptr,
381 GlobalVariable::InitialExecTLSModel);
382 ParamOriginTLS = new GlobalVariable(
383 M, ArrayType::get(OriginTy, kParamTLSSize / 4), false,
384 GlobalVariable::ExternalLinkage, nullptr, "__msan_param_origin_tls",
385 nullptr, GlobalVariable::InitialExecTLSModel);
387 VAArgTLS = new GlobalVariable(
388 M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
389 GlobalVariable::ExternalLinkage, nullptr, "__msan_va_arg_tls", nullptr,
390 GlobalVariable::InitialExecTLSModel);
391 VAArgOverflowSizeTLS = new GlobalVariable(
392 M, IRB.getInt64Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
393 "__msan_va_arg_overflow_size_tls", nullptr,
394 GlobalVariable::InitialExecTLSModel);
395 OriginTLS = new GlobalVariable(
396 M, IRB.getInt32Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
397 "__msan_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);
399 // We insert an empty inline asm after __msan_report* to avoid callback merge.
400 EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
401 StringRef(""), StringRef(""),
402 /*hasSideEffects=*/true);
404 if (WrapIndirectCalls) {
406 PointerType::getUnqual(FunctionType::get(IRB.getVoidTy(), false));
407 IndirectCallWrapperFn = M.getOrInsertFunction(
408 ClWrapIndirectCalls, AnyFunctionPtrTy, AnyFunctionPtrTy, nullptr);
411 if (WrapIndirectCalls && ClWrapIndirectCallsFast) {
412 MsandrModuleStart = new GlobalVariable(
413 M, IRB.getInt32Ty(), false, GlobalValue::ExternalLinkage,
414 nullptr, "__executable_start");
415 MsandrModuleStart->setVisibility(GlobalVariable::HiddenVisibility);
416 MsandrModuleEnd = new GlobalVariable(
417 M, IRB.getInt32Ty(), false, GlobalValue::ExternalLinkage,
419 MsandrModuleEnd->setVisibility(GlobalVariable::HiddenVisibility);
423 /// \brief Module-level initialization.
425 /// inserts a call to __msan_init to the module's constructor list.
426 bool MemorySanitizer::doInitialization(Module &M) {
427 DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
429 report_fatal_error("data layout missing");
430 DL = &DLP->getDataLayout();
432 C = &(M.getContext());
433 unsigned PtrSize = DL->getPointerSizeInBits(/* AddressSpace */0);
436 ShadowMask = kShadowMask64;
437 OriginOffset = kOriginOffset64;
440 ShadowMask = kShadowMask32;
441 OriginOffset = kOriginOffset32;
444 report_fatal_error("unsupported pointer size");
449 IntptrTy = IRB.getIntPtrTy(DL);
450 OriginTy = IRB.getInt32Ty();
452 ColdCallWeights = MDBuilder(*C).createBranchWeights(1, 1000);
453 OriginStoreWeights = MDBuilder(*C).createBranchWeights(1, 1000);
455 // Insert a call to __msan_init/__msan_track_origins into the module's CTORs.
456 appendToGlobalCtors(M, cast<Function>(M.getOrInsertFunction(
457 "__msan_init", IRB.getVoidTy(), nullptr)), 0);
460 new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
461 IRB.getInt32(TrackOrigins), "__msan_track_origins");
464 new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
465 IRB.getInt32(ClKeepGoing), "__msan_keep_going");
472 /// \brief A helper class that handles instrumentation of VarArg
473 /// functions on a particular platform.
475 /// Implementations are expected to insert the instrumentation
476 /// necessary to propagate argument shadow through VarArg function
477 /// calls. Visit* methods are called during an InstVisitor pass over
478 /// the function, and should avoid creating new basic blocks. A new
479 /// instance of this class is created for each instrumented function.
480 struct VarArgHelper {
481 /// \brief Visit a CallSite.
482 virtual void visitCallSite(CallSite &CS, IRBuilder<> &IRB) = 0;
484 /// \brief Visit a va_start call.
485 virtual void visitVAStartInst(VAStartInst &I) = 0;
487 /// \brief Visit a va_copy call.
488 virtual void visitVACopyInst(VACopyInst &I) = 0;
490 /// \brief Finalize function instrumentation.
492 /// This method is called after visiting all interesting (see above)
493 /// instructions in a function.
494 virtual void finalizeInstrumentation() = 0;
496 virtual ~VarArgHelper() {}
499 struct MemorySanitizerVisitor;
502 CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
503 MemorySanitizerVisitor &Visitor);
505 unsigned TypeSizeToSizeIndex(unsigned TypeSize) {
506 if (TypeSize <= 8) return 0;
507 return Log2_32_Ceil(TypeSize / 8);
510 /// This class does all the work for a given function. Store and Load
511 /// instructions store and load corresponding shadow and origin
512 /// values. Most instructions propagate shadow from arguments to their
513 /// return values. Certain instructions (most importantly, BranchInst)
514 /// test their argument shadow and print reports (with a runtime call) if it's
516 struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
519 SmallVector<PHINode *, 16> ShadowPHINodes, OriginPHINodes;
520 ValueMap<Value*, Value*> ShadowMap, OriginMap;
521 std::unique_ptr<VarArgHelper> VAHelper;
523 // The following flags disable parts of MSan instrumentation based on
524 // blacklist contents and command-line options.
526 bool PropagateShadow;
529 bool CheckReturnValue;
531 struct ShadowOriginAndInsertPoint {
534 Instruction *OrigIns;
535 ShadowOriginAndInsertPoint(Value *S, Value *O, Instruction *I)
536 : Shadow(S), Origin(O), OrigIns(I) { }
538 SmallVector<ShadowOriginAndInsertPoint, 16> InstrumentationList;
539 SmallVector<Instruction*, 16> StoreList;
540 SmallVector<CallSite, 16> IndirectCallList;
542 MemorySanitizerVisitor(Function &F, MemorySanitizer &MS)
543 : F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)) {
544 bool SanitizeFunction = F.getAttributes().hasAttribute(
545 AttributeSet::FunctionIndex, Attribute::SanitizeMemory);
546 InsertChecks = SanitizeFunction;
547 PropagateShadow = SanitizeFunction;
548 PoisonStack = SanitizeFunction && ClPoisonStack;
549 PoisonUndef = SanitizeFunction && ClPoisonUndef;
550 // FIXME: Consider using SpecialCaseList to specify a list of functions that
551 // must always return fully initialized values. For now, we hardcode "main".
552 CheckReturnValue = SanitizeFunction && (F.getName() == "main");
554 DEBUG(if (!InsertChecks)
555 dbgs() << "MemorySanitizer is not inserting checks into '"
556 << F.getName() << "'\n");
559 Value *updateOrigin(Value *V, IRBuilder<> &IRB) {
560 if (MS.TrackOrigins <= 1) return V;
561 return IRB.CreateCall(MS.MsanChainOriginFn, V);
564 void storeOrigin(IRBuilder<> &IRB, Value *Addr, Value *Shadow, Value *Origin,
565 unsigned Alignment, bool AsCall) {
566 if (isa<StructType>(Shadow->getType())) {
567 IRB.CreateAlignedStore(updateOrigin(Origin, IRB), getOriginPtr(Addr, IRB),
570 Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
571 // TODO(eugenis): handle non-zero constant shadow by inserting an
572 // unconditional check (can not simply fail compilation as this could
573 // be in the dead code).
574 if (!ClCheckConstantShadow)
575 if (isa<Constant>(ConvertedShadow)) return;
576 unsigned TypeSizeInBits =
577 MS.DL->getTypeSizeInBits(ConvertedShadow->getType());
578 unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
579 if (AsCall && SizeIndex < kNumberOfAccessSizes) {
580 Value *Fn = MS.MaybeStoreOriginFn[SizeIndex];
581 Value *ConvertedShadow2 = IRB.CreateZExt(
582 ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
583 IRB.CreateCall3(Fn, ConvertedShadow2,
584 IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()),
587 Value *Cmp = IRB.CreateICmpNE(
588 ConvertedShadow, getCleanShadow(ConvertedShadow), "_mscmp");
589 Instruction *CheckTerm = SplitBlockAndInsertIfThen(
590 Cmp, IRB.GetInsertPoint(), false, MS.OriginStoreWeights);
591 IRBuilder<> IRBNew(CheckTerm);
592 IRBNew.CreateAlignedStore(updateOrigin(Origin, IRBNew),
593 getOriginPtr(Addr, IRBNew), Alignment);
598 void materializeStores(bool InstrumentWithCalls) {
599 for (auto Inst : StoreList) {
600 StoreInst &SI = *dyn_cast<StoreInst>(Inst);
602 IRBuilder<> IRB(&SI);
603 Value *Val = SI.getValueOperand();
604 Value *Addr = SI.getPointerOperand();
605 Value *Shadow = SI.isAtomic() ? getCleanShadow(Val) : getShadow(Val);
606 Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);
609 IRB.CreateAlignedStore(Shadow, ShadowPtr, SI.getAlignment());
610 DEBUG(dbgs() << " STORE: " << *NewSI << "\n");
613 if (ClCheckAccessAddress) insertShadowCheck(Addr, &SI);
615 if (SI.isAtomic()) SI.setOrdering(addReleaseOrdering(SI.getOrdering()));
617 if (MS.TrackOrigins) {
618 unsigned Alignment = std::max(kMinOriginAlignment, SI.getAlignment());
619 storeOrigin(IRB, Addr, Shadow, getOrigin(Val), Alignment,
620 InstrumentWithCalls);
625 void materializeOneCheck(Instruction *OrigIns, Value *Shadow, Value *Origin,
627 IRBuilder<> IRB(OrigIns);
628 DEBUG(dbgs() << " SHAD0 : " << *Shadow << "\n");
629 Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
630 DEBUG(dbgs() << " SHAD1 : " << *ConvertedShadow << "\n");
631 // See the comment in storeOrigin().
632 if (!ClCheckConstantShadow)
633 if (isa<Constant>(ConvertedShadow)) return;
634 unsigned TypeSizeInBits =
635 MS.DL->getTypeSizeInBits(ConvertedShadow->getType());
636 unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
637 if (AsCall && SizeIndex < kNumberOfAccessSizes) {
638 Value *Fn = MS.MaybeWarningFn[SizeIndex];
639 Value *ConvertedShadow2 =
640 IRB.CreateZExt(ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
641 IRB.CreateCall2(Fn, ConvertedShadow2, MS.TrackOrigins && Origin
643 : (Value *)IRB.getInt32(0));
645 Value *Cmp = IRB.CreateICmpNE(ConvertedShadow,
646 getCleanShadow(ConvertedShadow), "_mscmp");
647 Instruction *CheckTerm = SplitBlockAndInsertIfThen(
649 /* Unreachable */ !ClKeepGoing, MS.ColdCallWeights);
651 IRB.SetInsertPoint(CheckTerm);
652 if (MS.TrackOrigins) {
653 IRB.CreateStore(Origin ? (Value *)Origin : (Value *)IRB.getInt32(0),
656 IRB.CreateCall(MS.WarningFn);
657 IRB.CreateCall(MS.EmptyAsm);
658 DEBUG(dbgs() << " CHECK: " << *Cmp << "\n");
662 void materializeChecks(bool InstrumentWithCalls) {
663 for (const auto &ShadowData : InstrumentationList) {
664 Instruction *OrigIns = ShadowData.OrigIns;
665 Value *Shadow = ShadowData.Shadow;
666 Value *Origin = ShadowData.Origin;
667 materializeOneCheck(OrigIns, Shadow, Origin, InstrumentWithCalls);
669 DEBUG(dbgs() << "DONE:\n" << F);
672 void materializeIndirectCalls() {
673 for (auto &CS : IndirectCallList) {
674 Instruction *I = CS.getInstruction();
675 BasicBlock *B = I->getParent();
677 Value *Fn0 = CS.getCalledValue();
678 Value *Fn = IRB.CreateBitCast(Fn0, MS.AnyFunctionPtrTy);
680 if (ClWrapIndirectCallsFast) {
681 // Check that call target is inside this module limits.
683 IRB.CreateBitCast(MS.MsandrModuleStart, MS.AnyFunctionPtrTy);
684 Value *End = IRB.CreateBitCast(MS.MsandrModuleEnd, MS.AnyFunctionPtrTy);
686 Value *NotInThisModule = IRB.CreateOr(IRB.CreateICmpULT(Fn, Start),
687 IRB.CreateICmpUGE(Fn, End));
690 IRB.CreatePHI(Fn0->getType(), 2, "msandr.indirect_target");
692 Instruction *CheckTerm = SplitBlockAndInsertIfThen(
693 NotInThisModule, NewFnPhi,
694 /* Unreachable */ false, MS.ColdCallWeights);
696 IRB.SetInsertPoint(CheckTerm);
697 // Slow path: call wrapper function to possibly transform the call
699 Value *NewFn = IRB.CreateBitCast(
700 IRB.CreateCall(MS.IndirectCallWrapperFn, Fn), Fn0->getType());
702 NewFnPhi->addIncoming(Fn0, B);
703 NewFnPhi->addIncoming(NewFn, dyn_cast<Instruction>(NewFn)->getParent());
704 CS.setCalledFunction(NewFnPhi);
706 Value *NewFn = IRB.CreateBitCast(
707 IRB.CreateCall(MS.IndirectCallWrapperFn, Fn), Fn0->getType());
708 CS.setCalledFunction(NewFn);
713 /// \brief Add MemorySanitizer instrumentation to a function.
714 bool runOnFunction() {
715 MS.initializeCallbacks(*F.getParent());
716 if (!MS.DL) return false;
718 // In the presence of unreachable blocks, we may see Phi nodes with
719 // incoming nodes from such blocks. Since InstVisitor skips unreachable
720 // blocks, such nodes will not have any shadow value associated with them.
721 // It's easier to remove unreachable blocks than deal with missing shadow.
722 removeUnreachableBlocks(F);
724 // Iterate all BBs in depth-first order and create shadow instructions
725 // for all instructions (where applicable).
726 // For PHI nodes we create dummy shadow PHIs which will be finalized later.
727 for (BasicBlock *BB : depth_first(&F.getEntryBlock()))
731 // Finalize PHI nodes.
732 for (PHINode *PN : ShadowPHINodes) {
733 PHINode *PNS = cast<PHINode>(getShadow(PN));
734 PHINode *PNO = MS.TrackOrigins ? cast<PHINode>(getOrigin(PN)) : nullptr;
735 size_t NumValues = PN->getNumIncomingValues();
736 for (size_t v = 0; v < NumValues; v++) {
737 PNS->addIncoming(getShadow(PN, v), PN->getIncomingBlock(v));
738 if (PNO) PNO->addIncoming(getOrigin(PN, v), PN->getIncomingBlock(v));
742 VAHelper->finalizeInstrumentation();
744 bool InstrumentWithCalls = ClInstrumentationWithCallThreshold >= 0 &&
745 InstrumentationList.size() + StoreList.size() >
746 (unsigned)ClInstrumentationWithCallThreshold;
748 // Delayed instrumentation of StoreInst.
749 // This may add new checks to be inserted later.
750 materializeStores(InstrumentWithCalls);
752 // Insert shadow value checks.
753 materializeChecks(InstrumentWithCalls);
755 // Wrap indirect calls.
756 materializeIndirectCalls();
761 /// \brief Compute the shadow type that corresponds to a given Value.
762 Type *getShadowTy(Value *V) {
763 return getShadowTy(V->getType());
766 /// \brief Compute the shadow type that corresponds to a given Type.
767 Type *getShadowTy(Type *OrigTy) {
768 if (!OrigTy->isSized()) {
771 // For integer type, shadow is the same as the original type.
772 // This may return weird-sized types like i1.
773 if (IntegerType *IT = dyn_cast<IntegerType>(OrigTy))
775 if (VectorType *VT = dyn_cast<VectorType>(OrigTy)) {
776 uint32_t EltSize = MS.DL->getTypeSizeInBits(VT->getElementType());
777 return VectorType::get(IntegerType::get(*MS.C, EltSize),
778 VT->getNumElements());
780 if (ArrayType *AT = dyn_cast<ArrayType>(OrigTy)) {
781 return ArrayType::get(getShadowTy(AT->getElementType()),
782 AT->getNumElements());
784 if (StructType *ST = dyn_cast<StructType>(OrigTy)) {
785 SmallVector<Type*, 4> Elements;
786 for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
787 Elements.push_back(getShadowTy(ST->getElementType(i)));
788 StructType *Res = StructType::get(*MS.C, Elements, ST->isPacked());
789 DEBUG(dbgs() << "getShadowTy: " << *ST << " ===> " << *Res << "\n");
792 uint32_t TypeSize = MS.DL->getTypeSizeInBits(OrigTy);
793 return IntegerType::get(*MS.C, TypeSize);
796 /// \brief Flatten a vector type.
797 Type *getShadowTyNoVec(Type *ty) {
798 if (VectorType *vt = dyn_cast<VectorType>(ty))
799 return IntegerType::get(*MS.C, vt->getBitWidth());
803 /// \brief Convert a shadow value to it's flattened variant.
804 Value *convertToShadowTyNoVec(Value *V, IRBuilder<> &IRB) {
805 Type *Ty = V->getType();
806 Type *NoVecTy = getShadowTyNoVec(Ty);
807 if (Ty == NoVecTy) return V;
808 return IRB.CreateBitCast(V, NoVecTy);
811 /// \brief Compute the shadow address that corresponds to a given application
814 /// Shadow = Addr & ~ShadowMask.
815 Value *getShadowPtr(Value *Addr, Type *ShadowTy,
818 IRB.CreateAnd(IRB.CreatePointerCast(Addr, MS.IntptrTy),
819 ConstantInt::get(MS.IntptrTy, ~MS.ShadowMask));
820 return IRB.CreateIntToPtr(ShadowLong, PointerType::get(ShadowTy, 0));
823 /// \brief Compute the origin address that corresponds to a given application
826 /// OriginAddr = (ShadowAddr + OriginOffset) & ~3ULL
827 Value *getOriginPtr(Value *Addr, IRBuilder<> &IRB) {
829 IRB.CreateAnd(IRB.CreatePointerCast(Addr, MS.IntptrTy),
830 ConstantInt::get(MS.IntptrTy, ~MS.ShadowMask));
832 IRB.CreateAdd(ShadowLong,
833 ConstantInt::get(MS.IntptrTy, MS.OriginOffset));
835 IRB.CreateAnd(Add, ConstantInt::get(MS.IntptrTy, ~3ULL));
836 return IRB.CreateIntToPtr(SecondAnd, PointerType::get(IRB.getInt32Ty(), 0));
839 /// \brief Compute the shadow address for a given function argument.
841 /// Shadow = ParamTLS+ArgOffset.
842 Value *getShadowPtrForArgument(Value *A, IRBuilder<> &IRB,
844 Value *Base = IRB.CreatePointerCast(MS.ParamTLS, MS.IntptrTy);
845 Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
846 return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
850 /// \brief Compute the origin address for a given function argument.
851 Value *getOriginPtrForArgument(Value *A, IRBuilder<> &IRB,
853 if (!MS.TrackOrigins) return nullptr;
854 Value *Base = IRB.CreatePointerCast(MS.ParamOriginTLS, MS.IntptrTy);
855 Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
856 return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
860 /// \brief Compute the shadow address for a retval.
861 Value *getShadowPtrForRetval(Value *A, IRBuilder<> &IRB) {
862 Value *Base = IRB.CreatePointerCast(MS.RetvalTLS, MS.IntptrTy);
863 return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
867 /// \brief Compute the origin address for a retval.
868 Value *getOriginPtrForRetval(IRBuilder<> &IRB) {
869 // We keep a single origin for the entire retval. Might be too optimistic.
870 return MS.RetvalOriginTLS;
873 /// \brief Set SV to be the shadow value for V.
874 void setShadow(Value *V, Value *SV) {
875 assert(!ShadowMap.count(V) && "Values may only have one shadow");
876 ShadowMap[V] = PropagateShadow ? SV : getCleanShadow(V);
879 /// \brief Set Origin to be the origin value for V.
880 void setOrigin(Value *V, Value *Origin) {
881 if (!MS.TrackOrigins) return;
882 assert(!OriginMap.count(V) && "Values may only have one origin");
883 DEBUG(dbgs() << "ORIGIN: " << *V << " ==> " << *Origin << "\n");
884 OriginMap[V] = Origin;
887 /// \brief Create a clean shadow value for a given value.
889 /// Clean shadow (all zeroes) means all bits of the value are defined
891 Constant *getCleanShadow(Value *V) {
892 Type *ShadowTy = getShadowTy(V);
895 return Constant::getNullValue(ShadowTy);
898 /// \brief Create a dirty shadow of a given shadow type.
899 Constant *getPoisonedShadow(Type *ShadowTy) {
901 if (isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy))
902 return Constant::getAllOnesValue(ShadowTy);
903 if (ArrayType *AT = dyn_cast<ArrayType>(ShadowTy)) {
904 SmallVector<Constant *, 4> Vals(AT->getNumElements(),
905 getPoisonedShadow(AT->getElementType()));
906 return ConstantArray::get(AT, Vals);
908 if (StructType *ST = dyn_cast<StructType>(ShadowTy)) {
909 SmallVector<Constant *, 4> Vals;
910 for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
911 Vals.push_back(getPoisonedShadow(ST->getElementType(i)));
912 return ConstantStruct::get(ST, Vals);
914 llvm_unreachable("Unexpected shadow type");
917 /// \brief Create a dirty shadow for a given value.
918 Constant *getPoisonedShadow(Value *V) {
919 Type *ShadowTy = getShadowTy(V);
922 return getPoisonedShadow(ShadowTy);
925 /// \brief Create a clean (zero) origin.
926 Value *getCleanOrigin() {
927 return Constant::getNullValue(MS.OriginTy);
930 /// \brief Get the shadow value for a given Value.
932 /// This function either returns the value set earlier with setShadow,
933 /// or extracts if from ParamTLS (for function arguments).
934 Value *getShadow(Value *V) {
935 if (!PropagateShadow) return getCleanShadow(V);
936 if (Instruction *I = dyn_cast<Instruction>(V)) {
937 // For instructions the shadow is already stored in the map.
938 Value *Shadow = ShadowMap[V];
940 DEBUG(dbgs() << "No shadow: " << *V << "\n" << *(I->getParent()));
942 assert(Shadow && "No shadow for a value");
946 if (UndefValue *U = dyn_cast<UndefValue>(V)) {
947 Value *AllOnes = PoisonUndef ? getPoisonedShadow(V) : getCleanShadow(V);
948 DEBUG(dbgs() << "Undef: " << *U << " ==> " << *AllOnes << "\n");
952 if (Argument *A = dyn_cast<Argument>(V)) {
953 // For arguments we compute the shadow on demand and store it in the map.
954 Value **ShadowPtr = &ShadowMap[V];
957 Function *F = A->getParent();
958 IRBuilder<> EntryIRB(F->getEntryBlock().getFirstNonPHI());
959 unsigned ArgOffset = 0;
960 for (auto &FArg : F->args()) {
961 if (!FArg.getType()->isSized()) {
962 DEBUG(dbgs() << "Arg is not sized\n");
965 unsigned Size = FArg.hasByValAttr()
966 ? MS.DL->getTypeAllocSize(FArg.getType()->getPointerElementType())
967 : MS.DL->getTypeAllocSize(FArg.getType());
969 bool Overflow = ArgOffset + Size > kParamTLSSize;
970 Value *Base = getShadowPtrForArgument(&FArg, EntryIRB, ArgOffset);
971 if (FArg.hasByValAttr()) {
972 // ByVal pointer itself has clean shadow. We copy the actual
973 // argument shadow to the underlying memory.
974 // Figure out maximal valid memcpy alignment.
975 unsigned ArgAlign = FArg.getParamAlignment();
977 Type *EltType = A->getType()->getPointerElementType();
978 ArgAlign = MS.DL->getABITypeAlignment(EltType);
981 // ParamTLS overflow.
982 EntryIRB.CreateMemSet(
983 getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB),
984 Constant::getNullValue(EntryIRB.getInt8Ty()), Size, ArgAlign);
986 unsigned CopyAlign = std::min(ArgAlign, kShadowTLSAlignment);
987 Value *Cpy = EntryIRB.CreateMemCpy(
988 getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB), Base, Size,
990 DEBUG(dbgs() << " ByValCpy: " << *Cpy << "\n");
993 *ShadowPtr = getCleanShadow(V);
996 // ParamTLS overflow.
997 *ShadowPtr = getCleanShadow(V);
1000 EntryIRB.CreateAlignedLoad(Base, kShadowTLSAlignment);
1003 DEBUG(dbgs() << " ARG: " << FArg << " ==> " <<
1004 **ShadowPtr << "\n");
1005 if (MS.TrackOrigins && !Overflow) {
1007 getOriginPtrForArgument(&FArg, EntryIRB, ArgOffset);
1008 setOrigin(A, EntryIRB.CreateLoad(OriginPtr));
1011 ArgOffset += RoundUpToAlignment(Size, kShadowTLSAlignment);
1013 assert(*ShadowPtr && "Could not find shadow for an argument");
1016 // For everything else the shadow is zero.
1017 return getCleanShadow(V);
1020 /// \brief Get the shadow for i-th argument of the instruction I.
1021 Value *getShadow(Instruction *I, int i) {
1022 return getShadow(I->getOperand(i));
1025 /// \brief Get the origin for a value.
1026 Value *getOrigin(Value *V) {
1027 if (!MS.TrackOrigins) return nullptr;
1028 if (isa<Instruction>(V) || isa<Argument>(V)) {
1029 Value *Origin = OriginMap[V];
1031 DEBUG(dbgs() << "NO ORIGIN: " << *V << "\n");
1032 Origin = getCleanOrigin();
1036 return getCleanOrigin();
1039 /// \brief Get the origin for i-th argument of the instruction I.
1040 Value *getOrigin(Instruction *I, int i) {
1041 return getOrigin(I->getOperand(i));
1044 /// \brief Remember the place where a shadow check should be inserted.
1046 /// This location will be later instrumented with a check that will print a
1047 /// UMR warning in runtime if the shadow value is not 0.
1048 void insertShadowCheck(Value *Shadow, Value *Origin, Instruction *OrigIns) {
1050 if (!InsertChecks) return;
1052 Type *ShadowTy = Shadow->getType();
1053 assert((isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy)) &&
1054 "Can only insert checks for integer and vector shadow types");
1056 InstrumentationList.push_back(
1057 ShadowOriginAndInsertPoint(Shadow, Origin, OrigIns));
1060 /// \brief Remember the place where a shadow check should be inserted.
1062 /// This location will be later instrumented with a check that will print a
1063 /// UMR warning in runtime if the value is not fully defined.
1064 void insertShadowCheck(Value *Val, Instruction *OrigIns) {
1066 Value *Shadow, *Origin;
1067 if (ClCheckConstantShadow) {
1068 Shadow = getShadow(Val);
1069 if (!Shadow) return;
1070 Origin = getOrigin(Val);
1072 Shadow = dyn_cast_or_null<Instruction>(getShadow(Val));
1073 if (!Shadow) return;
1074 Origin = dyn_cast_or_null<Instruction>(getOrigin(Val));
1076 insertShadowCheck(Shadow, Origin, OrigIns);
1079 AtomicOrdering addReleaseOrdering(AtomicOrdering a) {
1088 case AcquireRelease:
1089 return AcquireRelease;
1090 case SequentiallyConsistent:
1091 return SequentiallyConsistent;
1093 llvm_unreachable("Unknown ordering");
1096 AtomicOrdering addAcquireOrdering(AtomicOrdering a) {
1105 case AcquireRelease:
1106 return AcquireRelease;
1107 case SequentiallyConsistent:
1108 return SequentiallyConsistent;
1110 llvm_unreachable("Unknown ordering");
1113 // ------------------- Visitors.
1115 /// \brief Instrument LoadInst
1117 /// Loads the corresponding shadow and (optionally) origin.
1118 /// Optionally, checks that the load address is fully defined.
1119 void visitLoadInst(LoadInst &I) {
1120 assert(I.getType()->isSized() && "Load type must have size");
1121 IRBuilder<> IRB(I.getNextNode());
1122 Type *ShadowTy = getShadowTy(&I);
1123 Value *Addr = I.getPointerOperand();
1124 if (PropagateShadow) {
1125 Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
1127 IRB.CreateAlignedLoad(ShadowPtr, I.getAlignment(), "_msld"));
1129 setShadow(&I, getCleanShadow(&I));
1132 if (ClCheckAccessAddress)
1133 insertShadowCheck(I.getPointerOperand(), &I);
1136 I.setOrdering(addAcquireOrdering(I.getOrdering()));
1138 if (MS.TrackOrigins) {
1139 if (PropagateShadow) {
1140 unsigned Alignment = std::max(kMinOriginAlignment, I.getAlignment());
1142 IRB.CreateAlignedLoad(getOriginPtr(Addr, IRB), Alignment));
1144 setOrigin(&I, getCleanOrigin());
1149 /// \brief Instrument StoreInst
1151 /// Stores the corresponding shadow and (optionally) origin.
1152 /// Optionally, checks that the store address is fully defined.
1153 void visitStoreInst(StoreInst &I) {
1154 StoreList.push_back(&I);
1157 void handleCASOrRMW(Instruction &I) {
1158 assert(isa<AtomicRMWInst>(I) || isa<AtomicCmpXchgInst>(I));
1160 IRBuilder<> IRB(&I);
1161 Value *Addr = I.getOperand(0);
1162 Value *ShadowPtr = getShadowPtr(Addr, I.getType(), IRB);
1164 if (ClCheckAccessAddress)
1165 insertShadowCheck(Addr, &I);
1167 // Only test the conditional argument of cmpxchg instruction.
1168 // The other argument can potentially be uninitialized, but we can not
1169 // detect this situation reliably without possible false positives.
1170 if (isa<AtomicCmpXchgInst>(I))
1171 insertShadowCheck(I.getOperand(1), &I);
1173 IRB.CreateStore(getCleanShadow(&I), ShadowPtr);
1175 setShadow(&I, getCleanShadow(&I));
1178 void visitAtomicRMWInst(AtomicRMWInst &I) {
1180 I.setOrdering(addReleaseOrdering(I.getOrdering()));
1183 void visitAtomicCmpXchgInst(AtomicCmpXchgInst &I) {
1185 I.setSuccessOrdering(addReleaseOrdering(I.getSuccessOrdering()));
1188 // Vector manipulation.
1189 void visitExtractElementInst(ExtractElementInst &I) {
1190 insertShadowCheck(I.getOperand(1), &I);
1191 IRBuilder<> IRB(&I);
1192 setShadow(&I, IRB.CreateExtractElement(getShadow(&I, 0), I.getOperand(1),
1194 setOrigin(&I, getOrigin(&I, 0));
1197 void visitInsertElementInst(InsertElementInst &I) {
1198 insertShadowCheck(I.getOperand(2), &I);
1199 IRBuilder<> IRB(&I);
1200 setShadow(&I, IRB.CreateInsertElement(getShadow(&I, 0), getShadow(&I, 1),
1201 I.getOperand(2), "_msprop"));
1202 setOriginForNaryOp(I);
1205 void visitShuffleVectorInst(ShuffleVectorInst &I) {
1206 insertShadowCheck(I.getOperand(2), &I);
1207 IRBuilder<> IRB(&I);
1208 setShadow(&I, IRB.CreateShuffleVector(getShadow(&I, 0), getShadow(&I, 1),
1209 I.getOperand(2), "_msprop"));
1210 setOriginForNaryOp(I);
1214 void visitSExtInst(SExtInst &I) {
1215 IRBuilder<> IRB(&I);
1216 setShadow(&I, IRB.CreateSExt(getShadow(&I, 0), I.getType(), "_msprop"));
1217 setOrigin(&I, getOrigin(&I, 0));
1220 void visitZExtInst(ZExtInst &I) {
1221 IRBuilder<> IRB(&I);
1222 setShadow(&I, IRB.CreateZExt(getShadow(&I, 0), I.getType(), "_msprop"));
1223 setOrigin(&I, getOrigin(&I, 0));
1226 void visitTruncInst(TruncInst &I) {
1227 IRBuilder<> IRB(&I);
1228 setShadow(&I, IRB.CreateTrunc(getShadow(&I, 0), I.getType(), "_msprop"));
1229 setOrigin(&I, getOrigin(&I, 0));
1232 void visitBitCastInst(BitCastInst &I) {
1233 IRBuilder<> IRB(&I);
1234 setShadow(&I, IRB.CreateBitCast(getShadow(&I, 0), getShadowTy(&I)));
1235 setOrigin(&I, getOrigin(&I, 0));
1238 void visitPtrToIntInst(PtrToIntInst &I) {
1239 IRBuilder<> IRB(&I);
1240 setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1241 "_msprop_ptrtoint"));
1242 setOrigin(&I, getOrigin(&I, 0));
1245 void visitIntToPtrInst(IntToPtrInst &I) {
1246 IRBuilder<> IRB(&I);
1247 setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1248 "_msprop_inttoptr"));
1249 setOrigin(&I, getOrigin(&I, 0));
1252 void visitFPToSIInst(CastInst& I) { handleShadowOr(I); }
1253 void visitFPToUIInst(CastInst& I) { handleShadowOr(I); }
1254 void visitSIToFPInst(CastInst& I) { handleShadowOr(I); }
1255 void visitUIToFPInst(CastInst& I) { handleShadowOr(I); }
1256 void visitFPExtInst(CastInst& I) { handleShadowOr(I); }
1257 void visitFPTruncInst(CastInst& I) { handleShadowOr(I); }
1259 /// \brief Propagate shadow for bitwise AND.
1261 /// This code is exact, i.e. if, for example, a bit in the left argument
1262 /// is defined and 0, then neither the value not definedness of the
1263 /// corresponding bit in B don't affect the resulting shadow.
1264 void visitAnd(BinaryOperator &I) {
1265 IRBuilder<> IRB(&I);
1266 // "And" of 0 and a poisoned value results in unpoisoned value.
1267 // 1&1 => 1; 0&1 => 0; p&1 => p;
1268 // 1&0 => 0; 0&0 => 0; p&0 => 0;
1269 // 1&p => p; 0&p => 0; p&p => p;
1270 // S = (S1 & S2) | (V1 & S2) | (S1 & V2)
1271 Value *S1 = getShadow(&I, 0);
1272 Value *S2 = getShadow(&I, 1);
1273 Value *V1 = I.getOperand(0);
1274 Value *V2 = I.getOperand(1);
1275 if (V1->getType() != S1->getType()) {
1276 V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1277 V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1279 Value *S1S2 = IRB.CreateAnd(S1, S2);
1280 Value *V1S2 = IRB.CreateAnd(V1, S2);
1281 Value *S1V2 = IRB.CreateAnd(S1, V2);
1282 setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
1283 setOriginForNaryOp(I);
1286 void visitOr(BinaryOperator &I) {
1287 IRBuilder<> IRB(&I);
1288 // "Or" of 1 and a poisoned value results in unpoisoned value.
1289 // 1|1 => 1; 0|1 => 1; p|1 => 1;
1290 // 1|0 => 1; 0|0 => 0; p|0 => p;
1291 // 1|p => 1; 0|p => p; p|p => p;
1292 // S = (S1 & S2) | (~V1 & S2) | (S1 & ~V2)
1293 Value *S1 = getShadow(&I, 0);
1294 Value *S2 = getShadow(&I, 1);
1295 Value *V1 = IRB.CreateNot(I.getOperand(0));
1296 Value *V2 = IRB.CreateNot(I.getOperand(1));
1297 if (V1->getType() != S1->getType()) {
1298 V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1299 V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1301 Value *S1S2 = IRB.CreateAnd(S1, S2);
1302 Value *V1S2 = IRB.CreateAnd(V1, S2);
1303 Value *S1V2 = IRB.CreateAnd(S1, V2);
1304 setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
1305 setOriginForNaryOp(I);
1308 /// \brief Default propagation of shadow and/or origin.
1310 /// This class implements the general case of shadow propagation, used in all
1311 /// cases where we don't know and/or don't care about what the operation
1312 /// actually does. It converts all input shadow values to a common type
1313 /// (extending or truncating as necessary), and bitwise OR's them.
1315 /// This is much cheaper than inserting checks (i.e. requiring inputs to be
1316 /// fully initialized), and less prone to false positives.
1318 /// This class also implements the general case of origin propagation. For a
1319 /// Nary operation, result origin is set to the origin of an argument that is
1320 /// not entirely initialized. If there is more than one such arguments, the
1321 /// rightmost of them is picked. It does not matter which one is picked if all
1322 /// arguments are initialized.
1323 template <bool CombineShadow>
1328 MemorySanitizerVisitor *MSV;
1331 Combiner(MemorySanitizerVisitor *MSV, IRBuilder<> &IRB) :
1332 Shadow(nullptr), Origin(nullptr), IRB(IRB), MSV(MSV) {}
1334 /// \brief Add a pair of shadow and origin values to the mix.
1335 Combiner &Add(Value *OpShadow, Value *OpOrigin) {
1336 if (CombineShadow) {
1341 OpShadow = MSV->CreateShadowCast(IRB, OpShadow, Shadow->getType());
1342 Shadow = IRB.CreateOr(Shadow, OpShadow, "_msprop");
1346 if (MSV->MS.TrackOrigins) {
1351 Constant *ConstOrigin = dyn_cast<Constant>(OpOrigin);
1352 // No point in adding something that might result in 0 origin value.
1353 if (!ConstOrigin || !ConstOrigin->isNullValue()) {
1354 Value *FlatShadow = MSV->convertToShadowTyNoVec(OpShadow, IRB);
1356 IRB.CreateICmpNE(FlatShadow, MSV->getCleanShadow(FlatShadow));
1357 Origin = IRB.CreateSelect(Cond, OpOrigin, Origin);
1364 /// \brief Add an application value to the mix.
1365 Combiner &Add(Value *V) {
1366 Value *OpShadow = MSV->getShadow(V);
1367 Value *OpOrigin = MSV->MS.TrackOrigins ? MSV->getOrigin(V) : nullptr;
1368 return Add(OpShadow, OpOrigin);
1371 /// \brief Set the current combined values as the given instruction's shadow
1373 void Done(Instruction *I) {
1374 if (CombineShadow) {
1376 Shadow = MSV->CreateShadowCast(IRB, Shadow, MSV->getShadowTy(I));
1377 MSV->setShadow(I, Shadow);
1379 if (MSV->MS.TrackOrigins) {
1381 MSV->setOrigin(I, Origin);
1386 typedef Combiner<true> ShadowAndOriginCombiner;
1387 typedef Combiner<false> OriginCombiner;
1389 /// \brief Propagate origin for arbitrary operation.
1390 void setOriginForNaryOp(Instruction &I) {
1391 if (!MS.TrackOrigins) return;
1392 IRBuilder<> IRB(&I);
1393 OriginCombiner OC(this, IRB);
1394 for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
1399 size_t VectorOrPrimitiveTypeSizeInBits(Type *Ty) {
1400 assert(!(Ty->isVectorTy() && Ty->getScalarType()->isPointerTy()) &&
1401 "Vector of pointers is not a valid shadow type");
1402 return Ty->isVectorTy() ?
1403 Ty->getVectorNumElements() * Ty->getScalarSizeInBits() :
1404 Ty->getPrimitiveSizeInBits();
1407 /// \brief Cast between two shadow types, extending or truncating as
1409 Value *CreateShadowCast(IRBuilder<> &IRB, Value *V, Type *dstTy,
1410 bool Signed = false) {
1411 Type *srcTy = V->getType();
1412 if (dstTy->isIntegerTy() && srcTy->isIntegerTy())
1413 return IRB.CreateIntCast(V, dstTy, Signed);
1414 if (dstTy->isVectorTy() && srcTy->isVectorTy() &&
1415 dstTy->getVectorNumElements() == srcTy->getVectorNumElements())
1416 return IRB.CreateIntCast(V, dstTy, Signed);
1417 size_t srcSizeInBits = VectorOrPrimitiveTypeSizeInBits(srcTy);
1418 size_t dstSizeInBits = VectorOrPrimitiveTypeSizeInBits(dstTy);
1419 Value *V1 = IRB.CreateBitCast(V, Type::getIntNTy(*MS.C, srcSizeInBits));
1421 IRB.CreateIntCast(V1, Type::getIntNTy(*MS.C, dstSizeInBits), Signed);
1422 return IRB.CreateBitCast(V2, dstTy);
1423 // TODO: handle struct types.
1426 /// \brief Cast an application value to the type of its own shadow.
1427 Value *CreateAppToShadowCast(IRBuilder<> &IRB, Value *V) {
1428 Type *ShadowTy = getShadowTy(V);
1429 if (V->getType() == ShadowTy)
1431 if (V->getType()->isPtrOrPtrVectorTy())
1432 return IRB.CreatePtrToInt(V, ShadowTy);
1434 return IRB.CreateBitCast(V, ShadowTy);
1437 /// \brief Propagate shadow for arbitrary operation.
1438 void handleShadowOr(Instruction &I) {
1439 IRBuilder<> IRB(&I);
1440 ShadowAndOriginCombiner SC(this, IRB);
1441 for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
1446 // \brief Handle multiplication by constant.
1448 // Handle a special case of multiplication by constant that may have one or
1449 // more zeros in the lower bits. This makes corresponding number of lower bits
1450 // of the result zero as well. We model it by shifting the other operand
1451 // shadow left by the required number of bits. Effectively, we transform
1452 // (X * (A * 2**B)) to ((X << B) * A) and instrument (X << B) as (Sx << B).
1453 // We use multiplication by 2**N instead of shift to cover the case of
1454 // multiplication by 0, which may occur in some elements of a vector operand.
1455 void handleMulByConstant(BinaryOperator &I, Constant *ConstArg,
1457 Constant *ShadowMul;
1458 Type *Ty = ConstArg->getType();
1459 if (Ty->isVectorTy()) {
1460 unsigned NumElements = Ty->getVectorNumElements();
1461 Type *EltTy = Ty->getSequentialElementType();
1462 SmallVector<Constant *, 16> Elements;
1463 for (unsigned Idx = 0; Idx < NumElements; ++Idx) {
1465 dyn_cast<ConstantInt>(ConstArg->getAggregateElement(Idx));
1466 APInt V = Elt->getValue();
1467 APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
1468 Elements.push_back(ConstantInt::get(EltTy, V2));
1470 ShadowMul = ConstantVector::get(Elements);
1472 ConstantInt *Elt = dyn_cast<ConstantInt>(ConstArg);
1473 APInt V = Elt->getValue();
1474 APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
1475 ShadowMul = ConstantInt::get(Elt->getType(), V2);
1478 IRBuilder<> IRB(&I);
1480 IRB.CreateMul(getShadow(OtherArg), ShadowMul, "msprop_mul_cst"));
1481 setOrigin(&I, getOrigin(OtherArg));
1484 void visitMul(BinaryOperator &I) {
1485 Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
1486 Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
1487 if (constOp0 && !constOp1)
1488 handleMulByConstant(I, constOp0, I.getOperand(1));
1489 else if (constOp1 && !constOp0)
1490 handleMulByConstant(I, constOp1, I.getOperand(0));
1495 void visitFAdd(BinaryOperator &I) { handleShadowOr(I); }
1496 void visitFSub(BinaryOperator &I) { handleShadowOr(I); }
1497 void visitFMul(BinaryOperator &I) { handleShadowOr(I); }
1498 void visitAdd(BinaryOperator &I) { handleShadowOr(I); }
1499 void visitSub(BinaryOperator &I) { handleShadowOr(I); }
1500 void visitXor(BinaryOperator &I) { handleShadowOr(I); }
1502 void handleDiv(Instruction &I) {
1503 IRBuilder<> IRB(&I);
1504 // Strict on the second argument.
1505 insertShadowCheck(I.getOperand(1), &I);
1506 setShadow(&I, getShadow(&I, 0));
1507 setOrigin(&I, getOrigin(&I, 0));
1510 void visitUDiv(BinaryOperator &I) { handleDiv(I); }
1511 void visitSDiv(BinaryOperator &I) { handleDiv(I); }
1512 void visitFDiv(BinaryOperator &I) { handleDiv(I); }
1513 void visitURem(BinaryOperator &I) { handleDiv(I); }
1514 void visitSRem(BinaryOperator &I) { handleDiv(I); }
1515 void visitFRem(BinaryOperator &I) { handleDiv(I); }
1517 /// \brief Instrument == and != comparisons.
1519 /// Sometimes the comparison result is known even if some of the bits of the
1520 /// arguments are not.
1521 void handleEqualityComparison(ICmpInst &I) {
1522 IRBuilder<> IRB(&I);
1523 Value *A = I.getOperand(0);
1524 Value *B = I.getOperand(1);
1525 Value *Sa = getShadow(A);
1526 Value *Sb = getShadow(B);
1528 // Get rid of pointers and vectors of pointers.
1529 // For ints (and vectors of ints), types of A and Sa match,
1530 // and this is a no-op.
1531 A = IRB.CreatePointerCast(A, Sa->getType());
1532 B = IRB.CreatePointerCast(B, Sb->getType());
1534 // A == B <==> (C = A^B) == 0
1535 // A != B <==> (C = A^B) != 0
1537 Value *C = IRB.CreateXor(A, B);
1538 Value *Sc = IRB.CreateOr(Sa, Sb);
1539 // Now dealing with i = (C == 0) comparison (or C != 0, does not matter now)
1540 // Result is defined if one of the following is true
1541 // * there is a defined 1 bit in C
1542 // * C is fully defined
1543 // Si = !(C & ~Sc) && Sc
1544 Value *Zero = Constant::getNullValue(Sc->getType());
1545 Value *MinusOne = Constant::getAllOnesValue(Sc->getType());
1547 IRB.CreateAnd(IRB.CreateICmpNE(Sc, Zero),
1549 IRB.CreateAnd(IRB.CreateXor(Sc, MinusOne), C), Zero));
1550 Si->setName("_msprop_icmp");
1552 setOriginForNaryOp(I);
1555 /// \brief Build the lowest possible value of V, taking into account V's
1556 /// uninitialized bits.
1557 Value *getLowestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
1560 // Split shadow into sign bit and other bits.
1561 Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
1562 Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
1563 // Maximise the undefined shadow bit, minimize other undefined bits.
1565 IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaOtherBits)), SaSignBit);
1567 // Minimize undefined bits.
1568 return IRB.CreateAnd(A, IRB.CreateNot(Sa));
1572 /// \brief Build the highest possible value of V, taking into account V's
1573 /// uninitialized bits.
1574 Value *getHighestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
1577 // Split shadow into sign bit and other bits.
1578 Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
1579 Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
1580 // Minimise the undefined shadow bit, maximise other undefined bits.
1582 IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaSignBit)), SaOtherBits);
1584 // Maximize undefined bits.
1585 return IRB.CreateOr(A, Sa);
1589 /// \brief Instrument relational comparisons.
1591 /// This function does exact shadow propagation for all relational
1592 /// comparisons of integers, pointers and vectors of those.
1593 /// FIXME: output seems suboptimal when one of the operands is a constant
1594 void handleRelationalComparisonExact(ICmpInst &I) {
1595 IRBuilder<> IRB(&I);
1596 Value *A = I.getOperand(0);
1597 Value *B = I.getOperand(1);
1598 Value *Sa = getShadow(A);
1599 Value *Sb = getShadow(B);
1601 // Get rid of pointers and vectors of pointers.
1602 // For ints (and vectors of ints), types of A and Sa match,
1603 // and this is a no-op.
1604 A = IRB.CreatePointerCast(A, Sa->getType());
1605 B = IRB.CreatePointerCast(B, Sb->getType());
1607 // Let [a0, a1] be the interval of possible values of A, taking into account
1608 // its undefined bits. Let [b0, b1] be the interval of possible values of B.
1609 // Then (A cmp B) is defined iff (a0 cmp b1) == (a1 cmp b0).
1610 bool IsSigned = I.isSigned();
1611 Value *S1 = IRB.CreateICmp(I.getPredicate(),
1612 getLowestPossibleValue(IRB, A, Sa, IsSigned),
1613 getHighestPossibleValue(IRB, B, Sb, IsSigned));
1614 Value *S2 = IRB.CreateICmp(I.getPredicate(),
1615 getHighestPossibleValue(IRB, A, Sa, IsSigned),
1616 getLowestPossibleValue(IRB, B, Sb, IsSigned));
1617 Value *Si = IRB.CreateXor(S1, S2);
1619 setOriginForNaryOp(I);
1622 /// \brief Instrument signed relational comparisons.
1624 /// Handle (x<0) and (x>=0) comparisons (essentially, sign bit tests) by
1625 /// propagating the highest bit of the shadow. Everything else is delegated
1626 /// to handleShadowOr().
1627 void handleSignedRelationalComparison(ICmpInst &I) {
1628 Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
1629 Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
1630 Value* op = nullptr;
1631 CmpInst::Predicate pre = I.getPredicate();
1632 if (constOp0 && constOp0->isNullValue() &&
1633 (pre == CmpInst::ICMP_SGT || pre == CmpInst::ICMP_SLE)) {
1634 op = I.getOperand(1);
1635 } else if (constOp1 && constOp1->isNullValue() &&
1636 (pre == CmpInst::ICMP_SLT || pre == CmpInst::ICMP_SGE)) {
1637 op = I.getOperand(0);
1640 IRBuilder<> IRB(&I);
1642 IRB.CreateICmpSLT(getShadow(op), getCleanShadow(op), "_msprop_icmpslt");
1643 setShadow(&I, Shadow);
1644 setOrigin(&I, getOrigin(op));
1650 void visitICmpInst(ICmpInst &I) {
1651 if (!ClHandleICmp) {
1655 if (I.isEquality()) {
1656 handleEqualityComparison(I);
1660 assert(I.isRelational());
1661 if (ClHandleICmpExact) {
1662 handleRelationalComparisonExact(I);
1666 handleSignedRelationalComparison(I);
1670 assert(I.isUnsigned());
1671 if ((isa<Constant>(I.getOperand(0)) || isa<Constant>(I.getOperand(1)))) {
1672 handleRelationalComparisonExact(I);
1679 void visitFCmpInst(FCmpInst &I) {
1683 void handleShift(BinaryOperator &I) {
1684 IRBuilder<> IRB(&I);
1685 // If any of the S2 bits are poisoned, the whole thing is poisoned.
1686 // Otherwise perform the same shift on S1.
1687 Value *S1 = getShadow(&I, 0);
1688 Value *S2 = getShadow(&I, 1);
1689 Value *S2Conv = IRB.CreateSExt(IRB.CreateICmpNE(S2, getCleanShadow(S2)),
1691 Value *V2 = I.getOperand(1);
1692 Value *Shift = IRB.CreateBinOp(I.getOpcode(), S1, V2);
1693 setShadow(&I, IRB.CreateOr(Shift, S2Conv));
1694 setOriginForNaryOp(I);
1697 void visitShl(BinaryOperator &I) { handleShift(I); }
1698 void visitAShr(BinaryOperator &I) { handleShift(I); }
1699 void visitLShr(BinaryOperator &I) { handleShift(I); }
1701 /// \brief Instrument llvm.memmove
1703 /// At this point we don't know if llvm.memmove will be inlined or not.
1704 /// If we don't instrument it and it gets inlined,
1705 /// our interceptor will not kick in and we will lose the memmove.
1706 /// If we instrument the call here, but it does not get inlined,
1707 /// we will memove the shadow twice: which is bad in case
1708 /// of overlapping regions. So, we simply lower the intrinsic to a call.
1710 /// Similar situation exists for memcpy and memset.
1711 void visitMemMoveInst(MemMoveInst &I) {
1712 IRBuilder<> IRB(&I);
1715 IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
1716 IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
1717 IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
1718 I.eraseFromParent();
1721 // Similar to memmove: avoid copying shadow twice.
1722 // This is somewhat unfortunate as it may slowdown small constant memcpys.
1723 // FIXME: consider doing manual inline for small constant sizes and proper
1725 void visitMemCpyInst(MemCpyInst &I) {
1726 IRBuilder<> IRB(&I);
1729 IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
1730 IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
1731 IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
1732 I.eraseFromParent();
1736 void visitMemSetInst(MemSetInst &I) {
1737 IRBuilder<> IRB(&I);
1740 IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
1741 IRB.CreateIntCast(I.getArgOperand(1), IRB.getInt32Ty(), false),
1742 IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
1743 I.eraseFromParent();
1746 void visitVAStartInst(VAStartInst &I) {
1747 VAHelper->visitVAStartInst(I);
1750 void visitVACopyInst(VACopyInst &I) {
1751 VAHelper->visitVACopyInst(I);
1754 enum IntrinsicKind {
1755 IK_DoesNotAccessMemory,
1760 static IntrinsicKind getIntrinsicKind(Intrinsic::ID iid) {
1761 const int DoesNotAccessMemory = IK_DoesNotAccessMemory;
1762 const int OnlyReadsArgumentPointees = IK_OnlyReadsMemory;
1763 const int OnlyReadsMemory = IK_OnlyReadsMemory;
1764 const int OnlyAccessesArgumentPointees = IK_WritesMemory;
1765 const int UnknownModRefBehavior = IK_WritesMemory;
1766 #define GET_INTRINSIC_MODREF_BEHAVIOR
1767 #define ModRefBehavior IntrinsicKind
1768 #include "llvm/IR/Intrinsics.gen"
1769 #undef ModRefBehavior
1770 #undef GET_INTRINSIC_MODREF_BEHAVIOR
1773 /// \brief Handle vector store-like intrinsics.
1775 /// Instrument intrinsics that look like a simple SIMD store: writes memory,
1776 /// has 1 pointer argument and 1 vector argument, returns void.
1777 bool handleVectorStoreIntrinsic(IntrinsicInst &I) {
1778 IRBuilder<> IRB(&I);
1779 Value* Addr = I.getArgOperand(0);
1780 Value *Shadow = getShadow(&I, 1);
1781 Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);
1783 // We don't know the pointer alignment (could be unaligned SSE store!).
1784 // Have to assume to worst case.
1785 IRB.CreateAlignedStore(Shadow, ShadowPtr, 1);
1787 if (ClCheckAccessAddress)
1788 insertShadowCheck(Addr, &I);
1790 // FIXME: use ClStoreCleanOrigin
1791 // FIXME: factor out common code from materializeStores
1792 if (MS.TrackOrigins)
1793 IRB.CreateStore(getOrigin(&I, 1), getOriginPtr(Addr, IRB));
1797 /// \brief Handle vector load-like intrinsics.
1799 /// Instrument intrinsics that look like a simple SIMD load: reads memory,
1800 /// has 1 pointer argument, returns a vector.
1801 bool handleVectorLoadIntrinsic(IntrinsicInst &I) {
1802 IRBuilder<> IRB(&I);
1803 Value *Addr = I.getArgOperand(0);
1805 Type *ShadowTy = getShadowTy(&I);
1806 if (PropagateShadow) {
1807 Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
1808 // We don't know the pointer alignment (could be unaligned SSE load!).
1809 // Have to assume to worst case.
1810 setShadow(&I, IRB.CreateAlignedLoad(ShadowPtr, 1, "_msld"));
1812 setShadow(&I, getCleanShadow(&I));
1815 if (ClCheckAccessAddress)
1816 insertShadowCheck(Addr, &I);
1818 if (MS.TrackOrigins) {
1819 if (PropagateShadow)
1820 setOrigin(&I, IRB.CreateLoad(getOriginPtr(Addr, IRB)));
1822 setOrigin(&I, getCleanOrigin());
1827 /// \brief Handle (SIMD arithmetic)-like intrinsics.
1829 /// Instrument intrinsics with any number of arguments of the same type,
1830 /// equal to the return type. The type should be simple (no aggregates or
1831 /// pointers; vectors are fine).
1832 /// Caller guarantees that this intrinsic does not access memory.
1833 bool maybeHandleSimpleNomemIntrinsic(IntrinsicInst &I) {
1834 Type *RetTy = I.getType();
1835 if (!(RetTy->isIntOrIntVectorTy() ||
1836 RetTy->isFPOrFPVectorTy() ||
1837 RetTy->isX86_MMXTy()))
1840 unsigned NumArgOperands = I.getNumArgOperands();
1842 for (unsigned i = 0; i < NumArgOperands; ++i) {
1843 Type *Ty = I.getArgOperand(i)->getType();
1848 IRBuilder<> IRB(&I);
1849 ShadowAndOriginCombiner SC(this, IRB);
1850 for (unsigned i = 0; i < NumArgOperands; ++i)
1851 SC.Add(I.getArgOperand(i));
1857 /// \brief Heuristically instrument unknown intrinsics.
1859 /// The main purpose of this code is to do something reasonable with all
1860 /// random intrinsics we might encounter, most importantly - SIMD intrinsics.
1861 /// We recognize several classes of intrinsics by their argument types and
1862 /// ModRefBehaviour and apply special intrumentation when we are reasonably
1863 /// sure that we know what the intrinsic does.
1865 /// We special-case intrinsics where this approach fails. See llvm.bswap
1866 /// handling as an example of that.
1867 bool handleUnknownIntrinsic(IntrinsicInst &I) {
1868 unsigned NumArgOperands = I.getNumArgOperands();
1869 if (NumArgOperands == 0)
1872 Intrinsic::ID iid = I.getIntrinsicID();
1873 IntrinsicKind IK = getIntrinsicKind(iid);
1874 bool OnlyReadsMemory = IK == IK_OnlyReadsMemory;
1875 bool WritesMemory = IK == IK_WritesMemory;
1876 assert(!(OnlyReadsMemory && WritesMemory));
1878 if (NumArgOperands == 2 &&
1879 I.getArgOperand(0)->getType()->isPointerTy() &&
1880 I.getArgOperand(1)->getType()->isVectorTy() &&
1881 I.getType()->isVoidTy() &&
1883 // This looks like a vector store.
1884 return handleVectorStoreIntrinsic(I);
1887 if (NumArgOperands == 1 &&
1888 I.getArgOperand(0)->getType()->isPointerTy() &&
1889 I.getType()->isVectorTy() &&
1891 // This looks like a vector load.
1892 return handleVectorLoadIntrinsic(I);
1895 if (!OnlyReadsMemory && !WritesMemory)
1896 if (maybeHandleSimpleNomemIntrinsic(I))
1899 // FIXME: detect and handle SSE maskstore/maskload
1903 void handleBswap(IntrinsicInst &I) {
1904 IRBuilder<> IRB(&I);
1905 Value *Op = I.getArgOperand(0);
1906 Type *OpType = Op->getType();
1907 Function *BswapFunc = Intrinsic::getDeclaration(
1908 F.getParent(), Intrinsic::bswap, makeArrayRef(&OpType, 1));
1909 setShadow(&I, IRB.CreateCall(BswapFunc, getShadow(Op)));
1910 setOrigin(&I, getOrigin(Op));
1913 // \brief Instrument vector convert instrinsic.
1915 // This function instruments intrinsics like cvtsi2ss:
1916 // %Out = int_xxx_cvtyyy(%ConvertOp)
1918 // %Out = int_xxx_cvtyyy(%CopyOp, %ConvertOp)
1919 // Intrinsic converts \p NumUsedElements elements of \p ConvertOp to the same
1920 // number \p Out elements, and (if has 2 arguments) copies the rest of the
1921 // elements from \p CopyOp.
1922 // In most cases conversion involves floating-point value which may trigger a
1923 // hardware exception when not fully initialized. For this reason we require
1924 // \p ConvertOp[0:NumUsedElements] to be fully initialized and trap otherwise.
1925 // We copy the shadow of \p CopyOp[NumUsedElements:] to \p
1926 // Out[NumUsedElements:]. This means that intrinsics without \p CopyOp always
1927 // return a fully initialized value.
1928 void handleVectorConvertIntrinsic(IntrinsicInst &I, int NumUsedElements) {
1929 IRBuilder<> IRB(&I);
1930 Value *CopyOp, *ConvertOp;
1932 switch (I.getNumArgOperands()) {
1934 CopyOp = I.getArgOperand(0);
1935 ConvertOp = I.getArgOperand(1);
1938 ConvertOp = I.getArgOperand(0);
1942 llvm_unreachable("Cvt intrinsic with unsupported number of arguments.");
1945 // The first *NumUsedElements* elements of ConvertOp are converted to the
1946 // same number of output elements. The rest of the output is copied from
1947 // CopyOp, or (if not available) filled with zeroes.
1948 // Combine shadow for elements of ConvertOp that are used in this operation,
1949 // and insert a check.
1950 // FIXME: consider propagating shadow of ConvertOp, at least in the case of
1951 // int->any conversion.
1952 Value *ConvertShadow = getShadow(ConvertOp);
1953 Value *AggShadow = nullptr;
1954 if (ConvertOp->getType()->isVectorTy()) {
1955 AggShadow = IRB.CreateExtractElement(
1956 ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), 0));
1957 for (int i = 1; i < NumUsedElements; ++i) {
1958 Value *MoreShadow = IRB.CreateExtractElement(
1959 ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), i));
1960 AggShadow = IRB.CreateOr(AggShadow, MoreShadow);
1963 AggShadow = ConvertShadow;
1965 assert(AggShadow->getType()->isIntegerTy());
1966 insertShadowCheck(AggShadow, getOrigin(ConvertOp), &I);
1968 // Build result shadow by zero-filling parts of CopyOp shadow that come from
1971 assert(CopyOp->getType() == I.getType());
1972 assert(CopyOp->getType()->isVectorTy());
1973 Value *ResultShadow = getShadow(CopyOp);
1974 Type *EltTy = ResultShadow->getType()->getVectorElementType();
1975 for (int i = 0; i < NumUsedElements; ++i) {
1976 ResultShadow = IRB.CreateInsertElement(
1977 ResultShadow, ConstantInt::getNullValue(EltTy),
1978 ConstantInt::get(IRB.getInt32Ty(), i));
1980 setShadow(&I, ResultShadow);
1981 setOrigin(&I, getOrigin(CopyOp));
1983 setShadow(&I, getCleanShadow(&I));
1987 // Given a scalar or vector, extract lower 64 bits (or less), and return all
1988 // zeroes if it is zero, and all ones otherwise.
1989 Value *Lower64ShadowExtend(IRBuilder<> &IRB, Value *S, Type *T) {
1990 if (S->getType()->isVectorTy())
1991 S = CreateShadowCast(IRB, S, IRB.getInt64Ty(), /* Signed */ true);
1992 assert(S->getType()->getPrimitiveSizeInBits() <= 64);
1993 Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
1994 return CreateShadowCast(IRB, S2, T, /* Signed */ true);
1997 Value *VariableShadowExtend(IRBuilder<> &IRB, Value *S) {
1998 Type *T = S->getType();
1999 assert(T->isVectorTy());
2000 Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
2001 return IRB.CreateSExt(S2, T);
2004 // \brief Instrument vector shift instrinsic.
2006 // This function instruments intrinsics like int_x86_avx2_psll_w.
2007 // Intrinsic shifts %In by %ShiftSize bits.
2008 // %ShiftSize may be a vector. In that case the lower 64 bits determine shift
2009 // size, and the rest is ignored. Behavior is defined even if shift size is
2010 // greater than register (or field) width.
2011 void handleVectorShiftIntrinsic(IntrinsicInst &I, bool Variable) {
2012 assert(I.getNumArgOperands() == 2);
2013 IRBuilder<> IRB(&I);
2014 // If any of the S2 bits are poisoned, the whole thing is poisoned.
2015 // Otherwise perform the same shift on S1.
2016 Value *S1 = getShadow(&I, 0);
2017 Value *S2 = getShadow(&I, 1);
2018 Value *S2Conv = Variable ? VariableShadowExtend(IRB, S2)
2019 : Lower64ShadowExtend(IRB, S2, getShadowTy(&I));
2020 Value *V1 = I.getOperand(0);
2021 Value *V2 = I.getOperand(1);
2022 Value *Shift = IRB.CreateCall2(I.getCalledValue(),
2023 IRB.CreateBitCast(S1, V1->getType()), V2);
2024 Shift = IRB.CreateBitCast(Shift, getShadowTy(&I));
2025 setShadow(&I, IRB.CreateOr(Shift, S2Conv));
2026 setOriginForNaryOp(I);
2029 // \brief Get an X86_MMX-sized vector type.
2030 Type *getMMXVectorTy(unsigned EltSizeInBits) {
2031 const unsigned X86_MMXSizeInBits = 64;
2032 return VectorType::get(IntegerType::get(*MS.C, EltSizeInBits),
2033 X86_MMXSizeInBits / EltSizeInBits);
2036 // \brief Returns a signed counterpart for an (un)signed-saturate-and-pack
2038 Intrinsic::ID getSignedPackIntrinsic(Intrinsic::ID id) {
2040 case llvm::Intrinsic::x86_sse2_packsswb_128:
2041 case llvm::Intrinsic::x86_sse2_packuswb_128:
2042 return llvm::Intrinsic::x86_sse2_packsswb_128;
2044 case llvm::Intrinsic::x86_sse2_packssdw_128:
2045 case llvm::Intrinsic::x86_sse41_packusdw:
2046 return llvm::Intrinsic::x86_sse2_packssdw_128;
2048 case llvm::Intrinsic::x86_avx2_packsswb:
2049 case llvm::Intrinsic::x86_avx2_packuswb:
2050 return llvm::Intrinsic::x86_avx2_packsswb;
2052 case llvm::Intrinsic::x86_avx2_packssdw:
2053 case llvm::Intrinsic::x86_avx2_packusdw:
2054 return llvm::Intrinsic::x86_avx2_packssdw;
2056 case llvm::Intrinsic::x86_mmx_packsswb:
2057 case llvm::Intrinsic::x86_mmx_packuswb:
2058 return llvm::Intrinsic::x86_mmx_packsswb;
2060 case llvm::Intrinsic::x86_mmx_packssdw:
2061 return llvm::Intrinsic::x86_mmx_packssdw;
2063 llvm_unreachable("unexpected intrinsic id");
2067 // \brief Instrument vector pack instrinsic.
2069 // This function instruments intrinsics like x86_mmx_packsswb, that
2070 // packs elements of 2 input vectors into half as many bits with saturation.
2071 // Shadow is propagated with the signed variant of the same intrinsic applied
2072 // to sext(Sa != zeroinitializer), sext(Sb != zeroinitializer).
2073 // EltSizeInBits is used only for x86mmx arguments.
2074 void handleVectorPackIntrinsic(IntrinsicInst &I, unsigned EltSizeInBits = 0) {
2075 assert(I.getNumArgOperands() == 2);
2076 bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2077 IRBuilder<> IRB(&I);
2078 Value *S1 = getShadow(&I, 0);
2079 Value *S2 = getShadow(&I, 1);
2080 assert(isX86_MMX || S1->getType()->isVectorTy());
2082 // SExt and ICmpNE below must apply to individual elements of input vectors.
2083 // In case of x86mmx arguments, cast them to appropriate vector types and
2085 Type *T = isX86_MMX ? getMMXVectorTy(EltSizeInBits) : S1->getType();
2087 S1 = IRB.CreateBitCast(S1, T);
2088 S2 = IRB.CreateBitCast(S2, T);
2090 Value *S1_ext = IRB.CreateSExt(
2091 IRB.CreateICmpNE(S1, llvm::Constant::getNullValue(T)), T);
2092 Value *S2_ext = IRB.CreateSExt(
2093 IRB.CreateICmpNE(S2, llvm::Constant::getNullValue(T)), T);
2095 Type *X86_MMXTy = Type::getX86_MMXTy(*MS.C);
2096 S1_ext = IRB.CreateBitCast(S1_ext, X86_MMXTy);
2097 S2_ext = IRB.CreateBitCast(S2_ext, X86_MMXTy);
2100 Function *ShadowFn = Intrinsic::getDeclaration(
2101 F.getParent(), getSignedPackIntrinsic(I.getIntrinsicID()));
2103 Value *S = IRB.CreateCall2(ShadowFn, S1_ext, S2_ext, "_msprop_vector_pack");
2104 if (isX86_MMX) S = IRB.CreateBitCast(S, getShadowTy(&I));
2106 setOriginForNaryOp(I);
2109 // \brief Instrument sum-of-absolute-differencies intrinsic.
2110 void handleVectorSadIntrinsic(IntrinsicInst &I) {
2111 const unsigned SignificantBitsPerResultElement = 16;
2112 bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2113 Type *ResTy = isX86_MMX ? IntegerType::get(*MS.C, 64) : I.getType();
2114 unsigned ZeroBitsPerResultElement =
2115 ResTy->getScalarSizeInBits() - SignificantBitsPerResultElement;
2117 IRBuilder<> IRB(&I);
2118 Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2119 S = IRB.CreateBitCast(S, ResTy);
2120 S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2122 S = IRB.CreateLShr(S, ZeroBitsPerResultElement);
2123 S = IRB.CreateBitCast(S, getShadowTy(&I));
2125 setOriginForNaryOp(I);
2128 // \brief Instrument multiply-add intrinsic.
2129 void handleVectorPmaddIntrinsic(IntrinsicInst &I,
2130 unsigned EltSizeInBits = 0) {
2131 bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2132 Type *ResTy = isX86_MMX ? getMMXVectorTy(EltSizeInBits * 2) : I.getType();
2133 IRBuilder<> IRB(&I);
2134 Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2135 S = IRB.CreateBitCast(S, ResTy);
2136 S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2138 S = IRB.CreateBitCast(S, getShadowTy(&I));
2140 setOriginForNaryOp(I);
2143 void visitIntrinsicInst(IntrinsicInst &I) {
2144 switch (I.getIntrinsicID()) {
2145 case llvm::Intrinsic::bswap:
2148 case llvm::Intrinsic::x86_avx512_cvtsd2usi64:
2149 case llvm::Intrinsic::x86_avx512_cvtsd2usi:
2150 case llvm::Intrinsic::x86_avx512_cvtss2usi64:
2151 case llvm::Intrinsic::x86_avx512_cvtss2usi:
2152 case llvm::Intrinsic::x86_avx512_cvttss2usi64:
2153 case llvm::Intrinsic::x86_avx512_cvttss2usi:
2154 case llvm::Intrinsic::x86_avx512_cvttsd2usi64:
2155 case llvm::Intrinsic::x86_avx512_cvttsd2usi:
2156 case llvm::Intrinsic::x86_avx512_cvtusi2sd:
2157 case llvm::Intrinsic::x86_avx512_cvtusi2ss:
2158 case llvm::Intrinsic::x86_avx512_cvtusi642sd:
2159 case llvm::Intrinsic::x86_avx512_cvtusi642ss:
2160 case llvm::Intrinsic::x86_sse2_cvtsd2si64:
2161 case llvm::Intrinsic::x86_sse2_cvtsd2si:
2162 case llvm::Intrinsic::x86_sse2_cvtsd2ss:
2163 case llvm::Intrinsic::x86_sse2_cvtsi2sd:
2164 case llvm::Intrinsic::x86_sse2_cvtsi642sd:
2165 case llvm::Intrinsic::x86_sse2_cvtss2sd:
2166 case llvm::Intrinsic::x86_sse2_cvttsd2si64:
2167 case llvm::Intrinsic::x86_sse2_cvttsd2si:
2168 case llvm::Intrinsic::x86_sse_cvtsi2ss:
2169 case llvm::Intrinsic::x86_sse_cvtsi642ss:
2170 case llvm::Intrinsic::x86_sse_cvtss2si64:
2171 case llvm::Intrinsic::x86_sse_cvtss2si:
2172 case llvm::Intrinsic::x86_sse_cvttss2si64:
2173 case llvm::Intrinsic::x86_sse_cvttss2si:
2174 handleVectorConvertIntrinsic(I, 1);
2176 case llvm::Intrinsic::x86_sse2_cvtdq2pd:
2177 case llvm::Intrinsic::x86_sse2_cvtps2pd:
2178 case llvm::Intrinsic::x86_sse_cvtps2pi:
2179 case llvm::Intrinsic::x86_sse_cvttps2pi:
2180 handleVectorConvertIntrinsic(I, 2);
2182 case llvm::Intrinsic::x86_avx512_psll_dq:
2183 case llvm::Intrinsic::x86_avx512_psrl_dq:
2184 case llvm::Intrinsic::x86_avx2_psll_w:
2185 case llvm::Intrinsic::x86_avx2_psll_d:
2186 case llvm::Intrinsic::x86_avx2_psll_q:
2187 case llvm::Intrinsic::x86_avx2_pslli_w:
2188 case llvm::Intrinsic::x86_avx2_pslli_d:
2189 case llvm::Intrinsic::x86_avx2_pslli_q:
2190 case llvm::Intrinsic::x86_avx2_psll_dq:
2191 case llvm::Intrinsic::x86_avx2_psrl_w:
2192 case llvm::Intrinsic::x86_avx2_psrl_d:
2193 case llvm::Intrinsic::x86_avx2_psrl_q:
2194 case llvm::Intrinsic::x86_avx2_psra_w:
2195 case llvm::Intrinsic::x86_avx2_psra_d:
2196 case llvm::Intrinsic::x86_avx2_psrli_w:
2197 case llvm::Intrinsic::x86_avx2_psrli_d:
2198 case llvm::Intrinsic::x86_avx2_psrli_q:
2199 case llvm::Intrinsic::x86_avx2_psrai_w:
2200 case llvm::Intrinsic::x86_avx2_psrai_d:
2201 case llvm::Intrinsic::x86_avx2_psrl_dq:
2202 case llvm::Intrinsic::x86_sse2_psll_w:
2203 case llvm::Intrinsic::x86_sse2_psll_d:
2204 case llvm::Intrinsic::x86_sse2_psll_q:
2205 case llvm::Intrinsic::x86_sse2_pslli_w:
2206 case llvm::Intrinsic::x86_sse2_pslli_d:
2207 case llvm::Intrinsic::x86_sse2_pslli_q:
2208 case llvm::Intrinsic::x86_sse2_psll_dq:
2209 case llvm::Intrinsic::x86_sse2_psrl_w:
2210 case llvm::Intrinsic::x86_sse2_psrl_d:
2211 case llvm::Intrinsic::x86_sse2_psrl_q:
2212 case llvm::Intrinsic::x86_sse2_psra_w:
2213 case llvm::Intrinsic::x86_sse2_psra_d:
2214 case llvm::Intrinsic::x86_sse2_psrli_w:
2215 case llvm::Intrinsic::x86_sse2_psrli_d:
2216 case llvm::Intrinsic::x86_sse2_psrli_q:
2217 case llvm::Intrinsic::x86_sse2_psrai_w:
2218 case llvm::Intrinsic::x86_sse2_psrai_d:
2219 case llvm::Intrinsic::x86_sse2_psrl_dq:
2220 case llvm::Intrinsic::x86_mmx_psll_w:
2221 case llvm::Intrinsic::x86_mmx_psll_d:
2222 case llvm::Intrinsic::x86_mmx_psll_q:
2223 case llvm::Intrinsic::x86_mmx_pslli_w:
2224 case llvm::Intrinsic::x86_mmx_pslli_d:
2225 case llvm::Intrinsic::x86_mmx_pslli_q:
2226 case llvm::Intrinsic::x86_mmx_psrl_w:
2227 case llvm::Intrinsic::x86_mmx_psrl_d:
2228 case llvm::Intrinsic::x86_mmx_psrl_q:
2229 case llvm::Intrinsic::x86_mmx_psra_w:
2230 case llvm::Intrinsic::x86_mmx_psra_d:
2231 case llvm::Intrinsic::x86_mmx_psrli_w:
2232 case llvm::Intrinsic::x86_mmx_psrli_d:
2233 case llvm::Intrinsic::x86_mmx_psrli_q:
2234 case llvm::Intrinsic::x86_mmx_psrai_w:
2235 case llvm::Intrinsic::x86_mmx_psrai_d:
2236 handleVectorShiftIntrinsic(I, /* Variable */ false);
2238 case llvm::Intrinsic::x86_avx2_psllv_d:
2239 case llvm::Intrinsic::x86_avx2_psllv_d_256:
2240 case llvm::Intrinsic::x86_avx2_psllv_q:
2241 case llvm::Intrinsic::x86_avx2_psllv_q_256:
2242 case llvm::Intrinsic::x86_avx2_psrlv_d:
2243 case llvm::Intrinsic::x86_avx2_psrlv_d_256:
2244 case llvm::Intrinsic::x86_avx2_psrlv_q:
2245 case llvm::Intrinsic::x86_avx2_psrlv_q_256:
2246 case llvm::Intrinsic::x86_avx2_psrav_d:
2247 case llvm::Intrinsic::x86_avx2_psrav_d_256:
2248 handleVectorShiftIntrinsic(I, /* Variable */ true);
2251 // Byte shifts are not implemented.
2252 // case llvm::Intrinsic::x86_avx512_psll_dq_bs:
2253 // case llvm::Intrinsic::x86_avx512_psrl_dq_bs:
2254 // case llvm::Intrinsic::x86_avx2_psll_dq_bs:
2255 // case llvm::Intrinsic::x86_avx2_psrl_dq_bs:
2256 // case llvm::Intrinsic::x86_sse2_psll_dq_bs:
2257 // case llvm::Intrinsic::x86_sse2_psrl_dq_bs:
2259 case llvm::Intrinsic::x86_sse2_packsswb_128:
2260 case llvm::Intrinsic::x86_sse2_packssdw_128:
2261 case llvm::Intrinsic::x86_sse2_packuswb_128:
2262 case llvm::Intrinsic::x86_sse41_packusdw:
2263 case llvm::Intrinsic::x86_avx2_packsswb:
2264 case llvm::Intrinsic::x86_avx2_packssdw:
2265 case llvm::Intrinsic::x86_avx2_packuswb:
2266 case llvm::Intrinsic::x86_avx2_packusdw:
2267 handleVectorPackIntrinsic(I);
2270 case llvm::Intrinsic::x86_mmx_packsswb:
2271 case llvm::Intrinsic::x86_mmx_packuswb:
2272 handleVectorPackIntrinsic(I, 16);
2275 case llvm::Intrinsic::x86_mmx_packssdw:
2276 handleVectorPackIntrinsic(I, 32);
2279 case llvm::Intrinsic::x86_mmx_psad_bw:
2280 case llvm::Intrinsic::x86_sse2_psad_bw:
2281 case llvm::Intrinsic::x86_avx2_psad_bw:
2282 handleVectorSadIntrinsic(I);
2285 case llvm::Intrinsic::x86_sse2_pmadd_wd:
2286 case llvm::Intrinsic::x86_avx2_pmadd_wd:
2287 case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw_128:
2288 case llvm::Intrinsic::x86_avx2_pmadd_ub_sw:
2289 handleVectorPmaddIntrinsic(I);
2292 case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw:
2293 handleVectorPmaddIntrinsic(I, 8);
2296 case llvm::Intrinsic::x86_mmx_pmadd_wd:
2297 handleVectorPmaddIntrinsic(I, 16);
2301 if (!handleUnknownIntrinsic(I))
2302 visitInstruction(I);
2307 void visitCallSite(CallSite CS) {
2308 Instruction &I = *CS.getInstruction();
2309 assert((CS.isCall() || CS.isInvoke()) && "Unknown type of CallSite");
2311 CallInst *Call = cast<CallInst>(&I);
2313 // For inline asm, do the usual thing: check argument shadow and mark all
2314 // outputs as clean. Note that any side effects of the inline asm that are
2315 // not immediately visible in its constraints are not handled.
2316 if (Call->isInlineAsm()) {
2317 visitInstruction(I);
2321 assert(!isa<IntrinsicInst>(&I) && "intrinsics are handled elsewhere");
2323 // We are going to insert code that relies on the fact that the callee
2324 // will become a non-readonly function after it is instrumented by us. To
2325 // prevent this code from being optimized out, mark that function
2326 // non-readonly in advance.
2327 if (Function *Func = Call->getCalledFunction()) {
2328 // Clear out readonly/readnone attributes.
2330 B.addAttribute(Attribute::ReadOnly)
2331 .addAttribute(Attribute::ReadNone);
2332 Func->removeAttributes(AttributeSet::FunctionIndex,
2333 AttributeSet::get(Func->getContext(),
2334 AttributeSet::FunctionIndex,
2338 IRBuilder<> IRB(&I);
2340 if (MS.WrapIndirectCalls && !CS.getCalledFunction())
2341 IndirectCallList.push_back(CS);
2343 unsigned ArgOffset = 0;
2344 DEBUG(dbgs() << " CallSite: " << I << "\n");
2345 for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
2346 ArgIt != End; ++ArgIt) {
2348 unsigned i = ArgIt - CS.arg_begin();
2349 if (!A->getType()->isSized()) {
2350 DEBUG(dbgs() << "Arg " << i << " is not sized: " << I << "\n");
2354 Value *Store = nullptr;
2355 // Compute the Shadow for arg even if it is ByVal, because
2356 // in that case getShadow() will copy the actual arg shadow to
2357 // __msan_param_tls.
2358 Value *ArgShadow = getShadow(A);
2359 Value *ArgShadowBase = getShadowPtrForArgument(A, IRB, ArgOffset);
2360 DEBUG(dbgs() << " Arg#" << i << ": " << *A <<
2361 " Shadow: " << *ArgShadow << "\n");
2362 bool ArgIsInitialized = false;
2363 if (CS.paramHasAttr(i + 1, Attribute::ByVal)) {
2364 assert(A->getType()->isPointerTy() &&
2365 "ByVal argument is not a pointer!");
2366 Size = MS.DL->getTypeAllocSize(A->getType()->getPointerElementType());
2367 if (ArgOffset + Size > kParamTLSSize) break;
2368 unsigned ParamAlignment = CS.getParamAlignment(i + 1);
2369 unsigned Alignment = std::min(ParamAlignment, kShadowTLSAlignment);
2370 Store = IRB.CreateMemCpy(ArgShadowBase,
2371 getShadowPtr(A, Type::getInt8Ty(*MS.C), IRB),
2374 Size = MS.DL->getTypeAllocSize(A->getType());
2375 if (ArgOffset + Size > kParamTLSSize) break;
2376 Store = IRB.CreateAlignedStore(ArgShadow, ArgShadowBase,
2377 kShadowTLSAlignment);
2378 Constant *Cst = dyn_cast<Constant>(ArgShadow);
2379 if (Cst && Cst->isNullValue()) ArgIsInitialized = true;
2381 if (MS.TrackOrigins && !ArgIsInitialized)
2382 IRB.CreateStore(getOrigin(A),
2383 getOriginPtrForArgument(A, IRB, ArgOffset));
2385 assert(Size != 0 && Store != nullptr);
2386 DEBUG(dbgs() << " Param:" << *Store << "\n");
2387 ArgOffset += RoundUpToAlignment(Size, 8);
2389 DEBUG(dbgs() << " done with call args\n");
2392 cast<FunctionType>(CS.getCalledValue()->getType()->getContainedType(0));
2393 if (FT->isVarArg()) {
2394 VAHelper->visitCallSite(CS, IRB);
2397 // Now, get the shadow for the RetVal.
2398 if (!I.getType()->isSized()) return;
2399 IRBuilder<> IRBBefore(&I);
2400 // Until we have full dynamic coverage, make sure the retval shadow is 0.
2401 Value *Base = getShadowPtrForRetval(&I, IRBBefore);
2402 IRBBefore.CreateAlignedStore(getCleanShadow(&I), Base, kShadowTLSAlignment);
2403 Instruction *NextInsn = nullptr;
2405 NextInsn = I.getNextNode();
2407 BasicBlock *NormalDest = cast<InvokeInst>(&I)->getNormalDest();
2408 if (!NormalDest->getSinglePredecessor()) {
2409 // FIXME: this case is tricky, so we are just conservative here.
2410 // Perhaps we need to split the edge between this BB and NormalDest,
2411 // but a naive attempt to use SplitEdge leads to a crash.
2412 setShadow(&I, getCleanShadow(&I));
2413 setOrigin(&I, getCleanOrigin());
2416 NextInsn = NormalDest->getFirstInsertionPt();
2418 "Could not find insertion point for retval shadow load");
2420 IRBuilder<> IRBAfter(NextInsn);
2421 Value *RetvalShadow =
2422 IRBAfter.CreateAlignedLoad(getShadowPtrForRetval(&I, IRBAfter),
2423 kShadowTLSAlignment, "_msret");
2424 setShadow(&I, RetvalShadow);
2425 if (MS.TrackOrigins)
2426 setOrigin(&I, IRBAfter.CreateLoad(getOriginPtrForRetval(IRBAfter)));
2429 void visitReturnInst(ReturnInst &I) {
2430 IRBuilder<> IRB(&I);
2431 Value *RetVal = I.getReturnValue();
2432 if (!RetVal) return;
2433 Value *ShadowPtr = getShadowPtrForRetval(RetVal, IRB);
2434 if (CheckReturnValue) {
2435 insertShadowCheck(RetVal, &I);
2436 Value *Shadow = getCleanShadow(RetVal);
2437 IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
2439 Value *Shadow = getShadow(RetVal);
2440 IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
2441 // FIXME: make it conditional if ClStoreCleanOrigin==0
2442 if (MS.TrackOrigins)
2443 IRB.CreateStore(getOrigin(RetVal), getOriginPtrForRetval(IRB));
2447 void visitPHINode(PHINode &I) {
2448 IRBuilder<> IRB(&I);
2449 if (!PropagateShadow) {
2450 setShadow(&I, getCleanShadow(&I));
2454 ShadowPHINodes.push_back(&I);
2455 setShadow(&I, IRB.CreatePHI(getShadowTy(&I), I.getNumIncomingValues(),
2457 if (MS.TrackOrigins)
2458 setOrigin(&I, IRB.CreatePHI(MS.OriginTy, I.getNumIncomingValues(),
2462 void visitAllocaInst(AllocaInst &I) {
2463 setShadow(&I, getCleanShadow(&I));
2464 IRBuilder<> IRB(I.getNextNode());
2465 uint64_t Size = MS.DL->getTypeAllocSize(I.getAllocatedType());
2466 if (PoisonStack && ClPoisonStackWithCall) {
2467 IRB.CreateCall2(MS.MsanPoisonStackFn,
2468 IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()),
2469 ConstantInt::get(MS.IntptrTy, Size));
2471 Value *ShadowBase = getShadowPtr(&I, Type::getInt8PtrTy(*MS.C), IRB);
2472 Value *PoisonValue = IRB.getInt8(PoisonStack ? ClPoisonStackPattern : 0);
2473 IRB.CreateMemSet(ShadowBase, PoisonValue, Size, I.getAlignment());
2476 if (PoisonStack && MS.TrackOrigins) {
2477 setOrigin(&I, getCleanOrigin());
2478 SmallString<2048> StackDescriptionStorage;
2479 raw_svector_ostream StackDescription(StackDescriptionStorage);
2480 // We create a string with a description of the stack allocation and
2481 // pass it into __msan_set_alloca_origin.
2482 // It will be printed by the run-time if stack-originated UMR is found.
2483 // The first 4 bytes of the string are set to '----' and will be replaced
2484 // by __msan_va_arg_overflow_size_tls at the first call.
2485 StackDescription << "----" << I.getName() << "@" << F.getName();
2487 createPrivateNonConstGlobalForString(*F.getParent(),
2488 StackDescription.str());
2490 IRB.CreateCall4(MS.MsanSetAllocaOrigin4Fn,
2491 IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()),
2492 ConstantInt::get(MS.IntptrTy, Size),
2493 IRB.CreatePointerCast(Descr, IRB.getInt8PtrTy()),
2494 IRB.CreatePointerCast(&F, MS.IntptrTy));
2498 void visitSelectInst(SelectInst& I) {
2499 IRBuilder<> IRB(&I);
2500 // a = select b, c, d
2501 Value *B = I.getCondition();
2502 Value *C = I.getTrueValue();
2503 Value *D = I.getFalseValue();
2504 Value *Sb = getShadow(B);
2505 Value *Sc = getShadow(C);
2506 Value *Sd = getShadow(D);
2508 // Result shadow if condition shadow is 0.
2509 Value *Sa0 = IRB.CreateSelect(B, Sc, Sd);
2511 if (I.getType()->isAggregateType()) {
2512 // To avoid "sign extending" i1 to an arbitrary aggregate type, we just do
2513 // an extra "select". This results in much more compact IR.
2514 // Sa = select Sb, poisoned, (select b, Sc, Sd)
2515 Sa1 = getPoisonedShadow(getShadowTy(I.getType()));
2517 // Sa = select Sb, [ (c^d) | Sc | Sd ], [ b ? Sc : Sd ]
2518 // If Sb (condition is poisoned), look for bits in c and d that are equal
2519 // and both unpoisoned.
2520 // If !Sb (condition is unpoisoned), simply pick one of Sc and Sd.
2522 // Cast arguments to shadow-compatible type.
2523 C = CreateAppToShadowCast(IRB, C);
2524 D = CreateAppToShadowCast(IRB, D);
2526 // Result shadow if condition shadow is 1.
2527 Sa1 = IRB.CreateOr(IRB.CreateXor(C, D), IRB.CreateOr(Sc, Sd));
2529 Value *Sa = IRB.CreateSelect(Sb, Sa1, Sa0, "_msprop_select");
2531 if (MS.TrackOrigins) {
2532 // Origins are always i32, so any vector conditions must be flattened.
2533 // FIXME: consider tracking vector origins for app vectors?
2534 if (B->getType()->isVectorTy()) {
2535 Type *FlatTy = getShadowTyNoVec(B->getType());
2536 B = IRB.CreateICmpNE(IRB.CreateBitCast(B, FlatTy),
2537 ConstantInt::getNullValue(FlatTy));
2538 Sb = IRB.CreateICmpNE(IRB.CreateBitCast(Sb, FlatTy),
2539 ConstantInt::getNullValue(FlatTy));
2541 // a = select b, c, d
2542 // Oa = Sb ? Ob : (b ? Oc : Od)
2543 setOrigin(&I, IRB.CreateSelect(
2544 Sb, getOrigin(I.getCondition()),
2545 IRB.CreateSelect(B, getOrigin(C), getOrigin(D))));
2549 void visitLandingPadInst(LandingPadInst &I) {
2551 // See http://code.google.com/p/memory-sanitizer/issues/detail?id=1
2552 setShadow(&I, getCleanShadow(&I));
2553 setOrigin(&I, getCleanOrigin());
2556 void visitGetElementPtrInst(GetElementPtrInst &I) {
2560 void visitExtractValueInst(ExtractValueInst &I) {
2561 IRBuilder<> IRB(&I);
2562 Value *Agg = I.getAggregateOperand();
2563 DEBUG(dbgs() << "ExtractValue: " << I << "\n");
2564 Value *AggShadow = getShadow(Agg);
2565 DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
2566 Value *ResShadow = IRB.CreateExtractValue(AggShadow, I.getIndices());
2567 DEBUG(dbgs() << " ResShadow: " << *ResShadow << "\n");
2568 setShadow(&I, ResShadow);
2569 setOriginForNaryOp(I);
2572 void visitInsertValueInst(InsertValueInst &I) {
2573 IRBuilder<> IRB(&I);
2574 DEBUG(dbgs() << "InsertValue: " << I << "\n");
2575 Value *AggShadow = getShadow(I.getAggregateOperand());
2576 Value *InsShadow = getShadow(I.getInsertedValueOperand());
2577 DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
2578 DEBUG(dbgs() << " InsShadow: " << *InsShadow << "\n");
2579 Value *Res = IRB.CreateInsertValue(AggShadow, InsShadow, I.getIndices());
2580 DEBUG(dbgs() << " Res: " << *Res << "\n");
2582 setOriginForNaryOp(I);
2585 void dumpInst(Instruction &I) {
2586 if (CallInst *CI = dyn_cast<CallInst>(&I)) {
2587 errs() << "ZZZ call " << CI->getCalledFunction()->getName() << "\n";
2589 errs() << "ZZZ " << I.getOpcodeName() << "\n";
2591 errs() << "QQQ " << I << "\n";
2594 void visitResumeInst(ResumeInst &I) {
2595 DEBUG(dbgs() << "Resume: " << I << "\n");
2596 // Nothing to do here.
2599 void visitInstruction(Instruction &I) {
2600 // Everything else: stop propagating and check for poisoned shadow.
2601 if (ClDumpStrictInstructions)
2603 DEBUG(dbgs() << "DEFAULT: " << I << "\n");
2604 for (size_t i = 0, n = I.getNumOperands(); i < n; i++)
2605 insertShadowCheck(I.getOperand(i), &I);
2606 setShadow(&I, getCleanShadow(&I));
2607 setOrigin(&I, getCleanOrigin());
2611 /// \brief AMD64-specific implementation of VarArgHelper.
2612 struct VarArgAMD64Helper : public VarArgHelper {
2613 // An unfortunate workaround for asymmetric lowering of va_arg stuff.
2614 // See a comment in visitCallSite for more details.
2615 static const unsigned AMD64GpEndOffset = 48; // AMD64 ABI Draft 0.99.6 p3.5.7
2616 static const unsigned AMD64FpEndOffset = 176;
2619 MemorySanitizer &MS;
2620 MemorySanitizerVisitor &MSV;
2621 Value *VAArgTLSCopy;
2622 Value *VAArgOverflowSize;
2624 SmallVector<CallInst*, 16> VAStartInstrumentationList;
2626 VarArgAMD64Helper(Function &F, MemorySanitizer &MS,
2627 MemorySanitizerVisitor &MSV)
2628 : F(F), MS(MS), MSV(MSV), VAArgTLSCopy(nullptr),
2629 VAArgOverflowSize(nullptr) {}
2631 enum ArgKind { AK_GeneralPurpose, AK_FloatingPoint, AK_Memory };
2633 ArgKind classifyArgument(Value* arg) {
2634 // A very rough approximation of X86_64 argument classification rules.
2635 Type *T = arg->getType();
2636 if (T->isFPOrFPVectorTy() || T->isX86_MMXTy())
2637 return AK_FloatingPoint;
2638 if (T->isIntegerTy() && T->getPrimitiveSizeInBits() <= 64)
2639 return AK_GeneralPurpose;
2640 if (T->isPointerTy())
2641 return AK_GeneralPurpose;
2645 // For VarArg functions, store the argument shadow in an ABI-specific format
2646 // that corresponds to va_list layout.
2647 // We do this because Clang lowers va_arg in the frontend, and this pass
2648 // only sees the low level code that deals with va_list internals.
2649 // A much easier alternative (provided that Clang emits va_arg instructions)
2650 // would have been to associate each live instance of va_list with a copy of
2651 // MSanParamTLS, and extract shadow on va_arg() call in the argument list
2653 void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
2654 unsigned GpOffset = 0;
2655 unsigned FpOffset = AMD64GpEndOffset;
2656 unsigned OverflowOffset = AMD64FpEndOffset;
2657 for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
2658 ArgIt != End; ++ArgIt) {
2660 unsigned ArgNo = CS.getArgumentNo(ArgIt);
2661 bool IsByVal = CS.paramHasAttr(ArgNo + 1, Attribute::ByVal);
2663 // ByVal arguments always go to the overflow area.
2664 assert(A->getType()->isPointerTy());
2665 Type *RealTy = A->getType()->getPointerElementType();
2666 uint64_t ArgSize = MS.DL->getTypeAllocSize(RealTy);
2667 Value *Base = getShadowPtrForVAArgument(RealTy, IRB, OverflowOffset);
2668 OverflowOffset += RoundUpToAlignment(ArgSize, 8);
2669 IRB.CreateMemCpy(Base, MSV.getShadowPtr(A, IRB.getInt8Ty(), IRB),
2670 ArgSize, kShadowTLSAlignment);
2672 ArgKind AK = classifyArgument(A);
2673 if (AK == AK_GeneralPurpose && GpOffset >= AMD64GpEndOffset)
2675 if (AK == AK_FloatingPoint && FpOffset >= AMD64FpEndOffset)
2679 case AK_GeneralPurpose:
2680 Base = getShadowPtrForVAArgument(A->getType(), IRB, GpOffset);
2683 case AK_FloatingPoint:
2684 Base = getShadowPtrForVAArgument(A->getType(), IRB, FpOffset);
2688 uint64_t ArgSize = MS.DL->getTypeAllocSize(A->getType());
2689 Base = getShadowPtrForVAArgument(A->getType(), IRB, OverflowOffset);
2690 OverflowOffset += RoundUpToAlignment(ArgSize, 8);
2692 IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
2695 Constant *OverflowSize =
2696 ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AMD64FpEndOffset);
2697 IRB.CreateStore(OverflowSize, MS.VAArgOverflowSizeTLS);
2700 /// \brief Compute the shadow address for a given va_arg.
2701 Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
2703 Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
2704 Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
2705 return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
2709 void visitVAStartInst(VAStartInst &I) override {
2710 IRBuilder<> IRB(&I);
2711 VAStartInstrumentationList.push_back(&I);
2712 Value *VAListTag = I.getArgOperand(0);
2713 Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);
2715 // Unpoison the whole __va_list_tag.
2716 // FIXME: magic ABI constants.
2717 IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
2718 /* size */24, /* alignment */8, false);
2721 void visitVACopyInst(VACopyInst &I) override {
2722 IRBuilder<> IRB(&I);
2723 Value *VAListTag = I.getArgOperand(0);
2724 Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);
2726 // Unpoison the whole __va_list_tag.
2727 // FIXME: magic ABI constants.
2728 IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
2729 /* size */24, /* alignment */8, false);
2732 void finalizeInstrumentation() override {
2733 assert(!VAArgOverflowSize && !VAArgTLSCopy &&
2734 "finalizeInstrumentation called twice");
2735 if (!VAStartInstrumentationList.empty()) {
2736 // If there is a va_start in this function, make a backup copy of
2737 // va_arg_tls somewhere in the function entry block.
2738 IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI());
2739 VAArgOverflowSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
2741 IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, AMD64FpEndOffset),
2743 VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
2744 IRB.CreateMemCpy(VAArgTLSCopy, MS.VAArgTLS, CopySize, 8);
2747 // Instrument va_start.
2748 // Copy va_list shadow from the backup copy of the TLS contents.
2749 for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
2750 CallInst *OrigInst = VAStartInstrumentationList[i];
2751 IRBuilder<> IRB(OrigInst->getNextNode());
2752 Value *VAListTag = OrigInst->getArgOperand(0);
2754 Value *RegSaveAreaPtrPtr =
2756 IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
2757 ConstantInt::get(MS.IntptrTy, 16)),
2758 Type::getInt64PtrTy(*MS.C));
2759 Value *RegSaveAreaPtr = IRB.CreateLoad(RegSaveAreaPtrPtr);
2760 Value *RegSaveAreaShadowPtr =
2761 MSV.getShadowPtr(RegSaveAreaPtr, IRB.getInt8Ty(), IRB);
2762 IRB.CreateMemCpy(RegSaveAreaShadowPtr, VAArgTLSCopy,
2763 AMD64FpEndOffset, 16);
2765 Value *OverflowArgAreaPtrPtr =
2767 IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
2768 ConstantInt::get(MS.IntptrTy, 8)),
2769 Type::getInt64PtrTy(*MS.C));
2770 Value *OverflowArgAreaPtr = IRB.CreateLoad(OverflowArgAreaPtrPtr);
2771 Value *OverflowArgAreaShadowPtr =
2772 MSV.getShadowPtr(OverflowArgAreaPtr, IRB.getInt8Ty(), IRB);
2773 Value *SrcPtr = IRB.CreateConstGEP1_32(VAArgTLSCopy, AMD64FpEndOffset);
2774 IRB.CreateMemCpy(OverflowArgAreaShadowPtr, SrcPtr, VAArgOverflowSize, 16);
2779 /// \brief A no-op implementation of VarArgHelper.
2780 struct VarArgNoOpHelper : public VarArgHelper {
2781 VarArgNoOpHelper(Function &F, MemorySanitizer &MS,
2782 MemorySanitizerVisitor &MSV) {}
2784 void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {}
2786 void visitVAStartInst(VAStartInst &I) override {}
2788 void visitVACopyInst(VACopyInst &I) override {}
2790 void finalizeInstrumentation() override {}
2793 VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
2794 MemorySanitizerVisitor &Visitor) {
2795 // VarArg handling is only implemented on AMD64. False positives are possible
2796 // on other platforms.
2797 llvm::Triple TargetTriple(Func.getParent()->getTargetTriple());
2798 if (TargetTriple.getArch() == llvm::Triple::x86_64)
2799 return new VarArgAMD64Helper(Func, Msan, Visitor);
2801 return new VarArgNoOpHelper(Func, Msan, Visitor);
2806 bool MemorySanitizer::runOnFunction(Function &F) {
2807 MemorySanitizerVisitor Visitor(F, *this);
2809 // Clear out readonly/readnone attributes.
2811 B.addAttribute(Attribute::ReadOnly)
2812 .addAttribute(Attribute::ReadNone);
2813 F.removeAttributes(AttributeSet::FunctionIndex,
2814 AttributeSet::get(F.getContext(),
2815 AttributeSet::FunctionIndex, B));
2817 return Visitor.runOnFunction();