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 unsigned kMinOriginAlignment = 4;
124 static const unsigned kShadowTLSAlignment = 8;
126 // These constants must be kept in sync with the ones in msan.h.
127 static const unsigned kParamTLSSize = 800;
128 static const unsigned kRetvalTLSSize = 800;
130 // Accesses sizes are powers of two: 1, 2, 4, 8.
131 static const size_t kNumberOfAccessSizes = 4;
133 /// \brief Track origins of uninitialized values.
135 /// Adds a section to MemorySanitizer report that points to the allocation
136 /// (stack or heap) the uninitialized bits came from originally.
137 static cl::opt<int> ClTrackOrigins("msan-track-origins",
138 cl::desc("Track origins (allocation sites) of poisoned memory"),
139 cl::Hidden, cl::init(0));
140 static cl::opt<bool> ClKeepGoing("msan-keep-going",
141 cl::desc("keep going after reporting a UMR"),
142 cl::Hidden, cl::init(false));
143 static cl::opt<bool> ClPoisonStack("msan-poison-stack",
144 cl::desc("poison uninitialized stack variables"),
145 cl::Hidden, cl::init(true));
146 static cl::opt<bool> ClPoisonStackWithCall("msan-poison-stack-with-call",
147 cl::desc("poison uninitialized stack variables with a call"),
148 cl::Hidden, cl::init(false));
149 static cl::opt<int> ClPoisonStackPattern("msan-poison-stack-pattern",
150 cl::desc("poison uninitialized stack variables with the given patter"),
151 cl::Hidden, cl::init(0xff));
152 static cl::opt<bool> ClPoisonUndef("msan-poison-undef",
153 cl::desc("poison undef temps"),
154 cl::Hidden, cl::init(true));
156 static cl::opt<bool> ClHandleICmp("msan-handle-icmp",
157 cl::desc("propagate shadow through ICmpEQ and ICmpNE"),
158 cl::Hidden, cl::init(true));
160 static cl::opt<bool> ClHandleICmpExact("msan-handle-icmp-exact",
161 cl::desc("exact handling of relational integer ICmp"),
162 cl::Hidden, cl::init(false));
164 // This flag controls whether we check the shadow of the address
165 // operand of load or store. Such bugs are very rare, since load from
166 // a garbage address typically results in SEGV, but still happen
167 // (e.g. only lower bits of address are garbage, or the access happens
168 // early at program startup where malloc-ed memory is more likely to
169 // be zeroed. As of 2012-08-28 this flag adds 20% slowdown.
170 static cl::opt<bool> ClCheckAccessAddress("msan-check-access-address",
171 cl::desc("report accesses through a pointer which has poisoned shadow"),
172 cl::Hidden, cl::init(true));
174 static cl::opt<bool> ClDumpStrictInstructions("msan-dump-strict-instructions",
175 cl::desc("print out instructions with default strict semantics"),
176 cl::Hidden, cl::init(false));
178 static cl::opt<int> ClInstrumentationWithCallThreshold(
179 "msan-instrumentation-with-call-threshold",
181 "If the function being instrumented requires more than "
182 "this number of checks and origin stores, use callbacks instead of "
183 "inline checks (-1 means never use callbacks)."),
184 cl::Hidden, cl::init(3500));
186 // This is an experiment to enable handling of cases where shadow is a non-zero
187 // compile-time constant. For some unexplainable reason they were silently
188 // ignored in the instrumentation.
189 static cl::opt<bool> ClCheckConstantShadow("msan-check-constant-shadow",
190 cl::desc("Insert checks for constant shadow values"),
191 cl::Hidden, cl::init(false));
195 // Memory map parameters used in application-to-shadow address calculation.
196 // Offset = (Addr & ~AndMask) ^ XorMask
197 // Shadow = ShadowBase + Offset
198 // Origin = OriginBase + Offset
199 struct MemoryMapParams {
206 struct PlatformMemoryMapParams {
207 const MemoryMapParams *bits32;
208 const MemoryMapParams *bits64;
212 static const MemoryMapParams Linux_I386_MemoryMapParams = {
213 0x000080000000, // AndMask
214 0, // XorMask (not used)
215 0, // ShadowBase (not used)
216 0x000040000000, // OriginBase
220 static const MemoryMapParams Linux_X86_64_MemoryMapParams = {
221 0x400000000000, // AndMask
222 0, // XorMask (not used)
223 0, // ShadowBase (not used)
224 0x200000000000, // OriginBase
228 static const MemoryMapParams Linux_MIPS64_MemoryMapParams = {
229 0x004000000000, // AndMask
230 0, // XorMask (not used)
231 0, // ShadowBase (not used)
232 0x002000000000, // OriginBase
236 static const MemoryMapParams FreeBSD_I386_MemoryMapParams = {
237 0x000180000000, // AndMask
238 0x000040000000, // XorMask
239 0x000020000000, // ShadowBase
240 0x000700000000, // OriginBase
244 static const MemoryMapParams FreeBSD_X86_64_MemoryMapParams = {
245 0xc00000000000, // AndMask
246 0x200000000000, // XorMask
247 0x100000000000, // ShadowBase
248 0x380000000000, // OriginBase
251 static const PlatformMemoryMapParams Linux_X86_MemoryMapParams = {
252 &Linux_I386_MemoryMapParams,
253 &Linux_X86_64_MemoryMapParams,
256 static const PlatformMemoryMapParams Linux_MIPS_MemoryMapParams = {
258 &Linux_MIPS64_MemoryMapParams,
261 static const PlatformMemoryMapParams FreeBSD_X86_MemoryMapParams = {
262 &FreeBSD_I386_MemoryMapParams,
263 &FreeBSD_X86_64_MemoryMapParams,
266 /// \brief An instrumentation pass implementing detection of uninitialized
269 /// MemorySanitizer: instrument the code in module to find
270 /// uninitialized reads.
271 class MemorySanitizer : public FunctionPass {
273 MemorySanitizer(int TrackOrigins = 0)
275 TrackOrigins(std::max(TrackOrigins, (int)ClTrackOrigins)),
277 WarningFn(nullptr) {}
278 const char *getPassName() const override { return "MemorySanitizer"; }
279 bool runOnFunction(Function &F) override;
280 bool doInitialization(Module &M) override;
281 static char ID; // Pass identification, replacement for typeid.
284 void initializeCallbacks(Module &M);
286 /// \brief Track origins (allocation points) of uninitialized values.
289 const DataLayout *DL;
293 /// \brief Thread-local shadow storage for function parameters.
294 GlobalVariable *ParamTLS;
295 /// \brief Thread-local origin storage for function parameters.
296 GlobalVariable *ParamOriginTLS;
297 /// \brief Thread-local shadow storage for function return value.
298 GlobalVariable *RetvalTLS;
299 /// \brief Thread-local origin storage for function return value.
300 GlobalVariable *RetvalOriginTLS;
301 /// \brief Thread-local shadow storage for in-register va_arg function
302 /// parameters (x86_64-specific).
303 GlobalVariable *VAArgTLS;
304 /// \brief Thread-local shadow storage for va_arg overflow area
305 /// (x86_64-specific).
306 GlobalVariable *VAArgOverflowSizeTLS;
307 /// \brief Thread-local space used to pass origin value to the UMR reporting
309 GlobalVariable *OriginTLS;
311 /// \brief The run-time callback to print a warning.
313 // These arrays are indexed by log2(AccessSize).
314 Value *MaybeWarningFn[kNumberOfAccessSizes];
315 Value *MaybeStoreOriginFn[kNumberOfAccessSizes];
317 /// \brief Run-time helper that generates a new origin value for a stack
319 Value *MsanSetAllocaOrigin4Fn;
320 /// \brief Run-time helper that poisons stack on function entry.
321 Value *MsanPoisonStackFn;
322 /// \brief Run-time helper that records a store (or any event) of an
323 /// uninitialized value and returns an updated origin id encoding this info.
324 Value *MsanChainOriginFn;
325 /// \brief MSan runtime replacements for memmove, memcpy and memset.
326 Value *MemmoveFn, *MemcpyFn, *MemsetFn;
328 /// \brief Memory map parameters used in application-to-shadow calculation.
329 const MemoryMapParams *MapParams;
331 MDNode *ColdCallWeights;
332 /// \brief Branch weights for origin store.
333 MDNode *OriginStoreWeights;
334 /// \brief An empty volatile inline asm that prevents callback merge.
337 friend struct MemorySanitizerVisitor;
338 friend struct VarArgAMD64Helper;
342 char MemorySanitizer::ID = 0;
343 INITIALIZE_PASS(MemorySanitizer, "msan",
344 "MemorySanitizer: detects uninitialized reads.",
347 FunctionPass *llvm::createMemorySanitizerPass(int TrackOrigins) {
348 return new MemorySanitizer(TrackOrigins);
351 /// \brief Create a non-const global initialized with the given string.
353 /// Creates a writable global for Str so that we can pass it to the
354 /// run-time lib. Runtime uses first 4 bytes of the string to store the
355 /// frame ID, so the string needs to be mutable.
356 static GlobalVariable *createPrivateNonConstGlobalForString(Module &M,
358 Constant *StrConst = ConstantDataArray::getString(M.getContext(), Str);
359 return new GlobalVariable(M, StrConst->getType(), /*isConstant=*/false,
360 GlobalValue::PrivateLinkage, StrConst, "");
364 /// \brief Insert extern declaration of runtime-provided functions and globals.
365 void MemorySanitizer::initializeCallbacks(Module &M) {
366 // Only do this once.
371 // Create the callback.
372 // FIXME: this function should have "Cold" calling conv,
373 // which is not yet implemented.
374 StringRef WarningFnName = ClKeepGoing ? "__msan_warning"
375 : "__msan_warning_noreturn";
376 WarningFn = M.getOrInsertFunction(WarningFnName, IRB.getVoidTy(), nullptr);
378 for (size_t AccessSizeIndex = 0; AccessSizeIndex < kNumberOfAccessSizes;
380 unsigned AccessSize = 1 << AccessSizeIndex;
381 std::string FunctionName = "__msan_maybe_warning_" + itostr(AccessSize);
382 MaybeWarningFn[AccessSizeIndex] = M.getOrInsertFunction(
383 FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
384 IRB.getInt32Ty(), nullptr);
386 FunctionName = "__msan_maybe_store_origin_" + itostr(AccessSize);
387 MaybeStoreOriginFn[AccessSizeIndex] = M.getOrInsertFunction(
388 FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
389 IRB.getInt8PtrTy(), IRB.getInt32Ty(), nullptr);
392 MsanSetAllocaOrigin4Fn = M.getOrInsertFunction(
393 "__msan_set_alloca_origin4", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy,
394 IRB.getInt8PtrTy(), IntptrTy, nullptr);
396 M.getOrInsertFunction("__msan_poison_stack", IRB.getVoidTy(),
397 IRB.getInt8PtrTy(), IntptrTy, nullptr);
398 MsanChainOriginFn = M.getOrInsertFunction(
399 "__msan_chain_origin", IRB.getInt32Ty(), IRB.getInt32Ty(), nullptr);
400 MemmoveFn = M.getOrInsertFunction(
401 "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
402 IRB.getInt8PtrTy(), IntptrTy, nullptr);
403 MemcpyFn = M.getOrInsertFunction(
404 "__msan_memcpy", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
406 MemsetFn = M.getOrInsertFunction(
407 "__msan_memset", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt32Ty(),
411 RetvalTLS = new GlobalVariable(
412 M, ArrayType::get(IRB.getInt64Ty(), kRetvalTLSSize / 8), false,
413 GlobalVariable::ExternalLinkage, nullptr, "__msan_retval_tls", nullptr,
414 GlobalVariable::InitialExecTLSModel);
415 RetvalOriginTLS = new GlobalVariable(
416 M, OriginTy, false, GlobalVariable::ExternalLinkage, nullptr,
417 "__msan_retval_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);
419 ParamTLS = new GlobalVariable(
420 M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
421 GlobalVariable::ExternalLinkage, nullptr, "__msan_param_tls", nullptr,
422 GlobalVariable::InitialExecTLSModel);
423 ParamOriginTLS = new GlobalVariable(
424 M, ArrayType::get(OriginTy, kParamTLSSize / 4), false,
425 GlobalVariable::ExternalLinkage, nullptr, "__msan_param_origin_tls",
426 nullptr, GlobalVariable::InitialExecTLSModel);
428 VAArgTLS = new GlobalVariable(
429 M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
430 GlobalVariable::ExternalLinkage, nullptr, "__msan_va_arg_tls", nullptr,
431 GlobalVariable::InitialExecTLSModel);
432 VAArgOverflowSizeTLS = new GlobalVariable(
433 M, IRB.getInt64Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
434 "__msan_va_arg_overflow_size_tls", nullptr,
435 GlobalVariable::InitialExecTLSModel);
436 OriginTLS = new GlobalVariable(
437 M, IRB.getInt32Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
438 "__msan_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);
440 // We insert an empty inline asm after __msan_report* to avoid callback merge.
441 EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
442 StringRef(""), StringRef(""),
443 /*hasSideEffects=*/true);
446 /// \brief Module-level initialization.
448 /// inserts a call to __msan_init to the module's constructor list.
449 bool MemorySanitizer::doInitialization(Module &M) {
450 DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
452 report_fatal_error("data layout missing");
453 DL = &DLP->getDataLayout();
455 Triple TargetTriple(M.getTargetTriple());
456 switch (TargetTriple.getOS()) {
457 case Triple::FreeBSD:
458 switch (TargetTriple.getArch()) {
460 MapParams = FreeBSD_X86_MemoryMapParams.bits64;
463 MapParams = FreeBSD_X86_MemoryMapParams.bits32;
466 report_fatal_error("unsupported architecture");
470 switch (TargetTriple.getArch()) {
472 MapParams = Linux_X86_MemoryMapParams.bits64;
475 MapParams = Linux_X86_MemoryMapParams.bits32;
478 case Triple::mips64el:
479 MapParams = Linux_MIPS_MemoryMapParams.bits64;
482 report_fatal_error("unsupported architecture");
486 report_fatal_error("unsupported operating system");
489 C = &(M.getContext());
491 IntptrTy = IRB.getIntPtrTy(DL);
492 OriginTy = IRB.getInt32Ty();
494 ColdCallWeights = MDBuilder(*C).createBranchWeights(1, 1000);
495 OriginStoreWeights = MDBuilder(*C).createBranchWeights(1, 1000);
497 // Insert a call to __msan_init/__msan_track_origins into the module's CTORs.
498 appendToGlobalCtors(M, cast<Function>(M.getOrInsertFunction(
499 "__msan_init", IRB.getVoidTy(), nullptr)), 0);
502 new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
503 IRB.getInt32(TrackOrigins), "__msan_track_origins");
506 new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
507 IRB.getInt32(ClKeepGoing), "__msan_keep_going");
514 /// \brief A helper class that handles instrumentation of VarArg
515 /// functions on a particular platform.
517 /// Implementations are expected to insert the instrumentation
518 /// necessary to propagate argument shadow through VarArg function
519 /// calls. Visit* methods are called during an InstVisitor pass over
520 /// the function, and should avoid creating new basic blocks. A new
521 /// instance of this class is created for each instrumented function.
522 struct VarArgHelper {
523 /// \brief Visit a CallSite.
524 virtual void visitCallSite(CallSite &CS, IRBuilder<> &IRB) = 0;
526 /// \brief Visit a va_start call.
527 virtual void visitVAStartInst(VAStartInst &I) = 0;
529 /// \brief Visit a va_copy call.
530 virtual void visitVACopyInst(VACopyInst &I) = 0;
532 /// \brief Finalize function instrumentation.
534 /// This method is called after visiting all interesting (see above)
535 /// instructions in a function.
536 virtual void finalizeInstrumentation() = 0;
538 virtual ~VarArgHelper() {}
541 struct MemorySanitizerVisitor;
544 CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
545 MemorySanitizerVisitor &Visitor);
547 unsigned TypeSizeToSizeIndex(unsigned TypeSize) {
548 if (TypeSize <= 8) return 0;
549 return Log2_32_Ceil(TypeSize / 8);
552 /// This class does all the work for a given function. Store and Load
553 /// instructions store and load corresponding shadow and origin
554 /// values. Most instructions propagate shadow from arguments to their
555 /// return values. Certain instructions (most importantly, BranchInst)
556 /// test their argument shadow and print reports (with a runtime call) if it's
558 struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
561 SmallVector<PHINode *, 16> ShadowPHINodes, OriginPHINodes;
562 ValueMap<Value*, Value*> ShadowMap, OriginMap;
563 std::unique_ptr<VarArgHelper> VAHelper;
565 // The following flags disable parts of MSan instrumentation based on
566 // blacklist contents and command-line options.
568 bool PropagateShadow;
571 bool CheckReturnValue;
573 struct ShadowOriginAndInsertPoint {
576 Instruction *OrigIns;
577 ShadowOriginAndInsertPoint(Value *S, Value *O, Instruction *I)
578 : Shadow(S), Origin(O), OrigIns(I) { }
580 SmallVector<ShadowOriginAndInsertPoint, 16> InstrumentationList;
581 SmallVector<Instruction*, 16> StoreList;
583 MemorySanitizerVisitor(Function &F, MemorySanitizer &MS)
584 : F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)) {
585 bool SanitizeFunction = F.getAttributes().hasAttribute(
586 AttributeSet::FunctionIndex, Attribute::SanitizeMemory);
587 InsertChecks = SanitizeFunction;
588 PropagateShadow = SanitizeFunction;
589 PoisonStack = SanitizeFunction && ClPoisonStack;
590 PoisonUndef = SanitizeFunction && ClPoisonUndef;
591 // FIXME: Consider using SpecialCaseList to specify a list of functions that
592 // must always return fully initialized values. For now, we hardcode "main".
593 CheckReturnValue = SanitizeFunction && (F.getName() == "main");
595 DEBUG(if (!InsertChecks)
596 dbgs() << "MemorySanitizer is not inserting checks into '"
597 << F.getName() << "'\n");
600 Value *updateOrigin(Value *V, IRBuilder<> &IRB) {
601 if (MS.TrackOrigins <= 1) return V;
602 return IRB.CreateCall(MS.MsanChainOriginFn, V);
605 void storeOrigin(IRBuilder<> &IRB, Value *Addr, Value *Shadow, Value *Origin,
606 unsigned Alignment, bool AsCall) {
607 unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
608 if (isa<StructType>(Shadow->getType())) {
609 IRB.CreateAlignedStore(updateOrigin(Origin, IRB),
610 getOriginPtr(Addr, IRB, Alignment),
613 Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
614 // TODO(eugenis): handle non-zero constant shadow by inserting an
615 // unconditional check (can not simply fail compilation as this could
616 // be in the dead code).
617 if (!ClCheckConstantShadow)
618 if (isa<Constant>(ConvertedShadow)) return;
619 unsigned TypeSizeInBits =
620 MS.DL->getTypeSizeInBits(ConvertedShadow->getType());
621 unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
622 if (AsCall && SizeIndex < kNumberOfAccessSizes) {
623 Value *Fn = MS.MaybeStoreOriginFn[SizeIndex];
624 Value *ConvertedShadow2 = IRB.CreateZExt(
625 ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
626 IRB.CreateCall3(Fn, ConvertedShadow2,
627 IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()),
630 Value *Cmp = IRB.CreateICmpNE(
631 ConvertedShadow, getCleanShadow(ConvertedShadow), "_mscmp");
632 Instruction *CheckTerm = SplitBlockAndInsertIfThen(
633 Cmp, IRB.GetInsertPoint(), false, MS.OriginStoreWeights);
634 IRBuilder<> IRBNew(CheckTerm);
635 IRBNew.CreateAlignedStore(updateOrigin(Origin, IRBNew),
636 getOriginPtr(Addr, IRBNew, Alignment),
642 void materializeStores(bool InstrumentWithCalls) {
643 for (auto Inst : StoreList) {
644 StoreInst &SI = *dyn_cast<StoreInst>(Inst);
646 IRBuilder<> IRB(&SI);
647 Value *Val = SI.getValueOperand();
648 Value *Addr = SI.getPointerOperand();
649 Value *Shadow = SI.isAtomic() ? getCleanShadow(Val) : getShadow(Val);
650 Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);
653 IRB.CreateAlignedStore(Shadow, ShadowPtr, SI.getAlignment());
654 DEBUG(dbgs() << " STORE: " << *NewSI << "\n");
657 if (ClCheckAccessAddress) insertShadowCheck(Addr, &SI);
659 if (SI.isAtomic()) SI.setOrdering(addReleaseOrdering(SI.getOrdering()));
662 storeOrigin(IRB, Addr, Shadow, getOrigin(Val), SI.getAlignment(),
663 InstrumentWithCalls);
667 void materializeOneCheck(Instruction *OrigIns, Value *Shadow, Value *Origin,
669 IRBuilder<> IRB(OrigIns);
670 DEBUG(dbgs() << " SHAD0 : " << *Shadow << "\n");
671 Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
672 DEBUG(dbgs() << " SHAD1 : " << *ConvertedShadow << "\n");
673 // See the comment in storeOrigin().
674 if (!ClCheckConstantShadow)
675 if (isa<Constant>(ConvertedShadow)) return;
676 unsigned TypeSizeInBits =
677 MS.DL->getTypeSizeInBits(ConvertedShadow->getType());
678 unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
679 if (AsCall && SizeIndex < kNumberOfAccessSizes) {
680 Value *Fn = MS.MaybeWarningFn[SizeIndex];
681 Value *ConvertedShadow2 =
682 IRB.CreateZExt(ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
683 IRB.CreateCall2(Fn, ConvertedShadow2, MS.TrackOrigins && Origin
685 : (Value *)IRB.getInt32(0));
687 Value *Cmp = IRB.CreateICmpNE(ConvertedShadow,
688 getCleanShadow(ConvertedShadow), "_mscmp");
689 Instruction *CheckTerm = SplitBlockAndInsertIfThen(
691 /* Unreachable */ !ClKeepGoing, MS.ColdCallWeights);
693 IRB.SetInsertPoint(CheckTerm);
694 if (MS.TrackOrigins) {
695 IRB.CreateStore(Origin ? (Value *)Origin : (Value *)IRB.getInt32(0),
698 IRB.CreateCall(MS.WarningFn);
699 IRB.CreateCall(MS.EmptyAsm);
700 DEBUG(dbgs() << " CHECK: " << *Cmp << "\n");
704 void materializeChecks(bool InstrumentWithCalls) {
705 for (const auto &ShadowData : InstrumentationList) {
706 Instruction *OrigIns = ShadowData.OrigIns;
707 Value *Shadow = ShadowData.Shadow;
708 Value *Origin = ShadowData.Origin;
709 materializeOneCheck(OrigIns, Shadow, Origin, InstrumentWithCalls);
711 DEBUG(dbgs() << "DONE:\n" << F);
714 /// \brief Add MemorySanitizer instrumentation to a function.
715 bool runOnFunction() {
716 MS.initializeCallbacks(*F.getParent());
717 if (!MS.DL) return false;
719 // In the presence of unreachable blocks, we may see Phi nodes with
720 // incoming nodes from such blocks. Since InstVisitor skips unreachable
721 // blocks, such nodes will not have any shadow value associated with them.
722 // It's easier to remove unreachable blocks than deal with missing shadow.
723 removeUnreachableBlocks(F);
725 // Iterate all BBs in depth-first order and create shadow instructions
726 // for all instructions (where applicable).
727 // For PHI nodes we create dummy shadow PHIs which will be finalized later.
728 for (BasicBlock *BB : depth_first(&F.getEntryBlock()))
732 // Finalize PHI nodes.
733 for (PHINode *PN : ShadowPHINodes) {
734 PHINode *PNS = cast<PHINode>(getShadow(PN));
735 PHINode *PNO = MS.TrackOrigins ? cast<PHINode>(getOrigin(PN)) : nullptr;
736 size_t NumValues = PN->getNumIncomingValues();
737 for (size_t v = 0; v < NumValues; v++) {
738 PNS->addIncoming(getShadow(PN, v), PN->getIncomingBlock(v));
739 if (PNO) PNO->addIncoming(getOrigin(PN, v), PN->getIncomingBlock(v));
743 VAHelper->finalizeInstrumentation();
745 bool InstrumentWithCalls = ClInstrumentationWithCallThreshold >= 0 &&
746 InstrumentationList.size() + StoreList.size() >
747 (unsigned)ClInstrumentationWithCallThreshold;
749 // Delayed instrumentation of StoreInst.
750 // This may add new checks to be inserted later.
751 materializeStores(InstrumentWithCalls);
753 // Insert shadow value checks.
754 materializeChecks(InstrumentWithCalls);
759 /// \brief Compute the shadow type that corresponds to a given Value.
760 Type *getShadowTy(Value *V) {
761 return getShadowTy(V->getType());
764 /// \brief Compute the shadow type that corresponds to a given Type.
765 Type *getShadowTy(Type *OrigTy) {
766 if (!OrigTy->isSized()) {
769 // For integer type, shadow is the same as the original type.
770 // This may return weird-sized types like i1.
771 if (IntegerType *IT = dyn_cast<IntegerType>(OrigTy))
773 if (VectorType *VT = dyn_cast<VectorType>(OrigTy)) {
774 uint32_t EltSize = MS.DL->getTypeSizeInBits(VT->getElementType());
775 return VectorType::get(IntegerType::get(*MS.C, EltSize),
776 VT->getNumElements());
778 if (ArrayType *AT = dyn_cast<ArrayType>(OrigTy)) {
779 return ArrayType::get(getShadowTy(AT->getElementType()),
780 AT->getNumElements());
782 if (StructType *ST = dyn_cast<StructType>(OrigTy)) {
783 SmallVector<Type*, 4> Elements;
784 for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
785 Elements.push_back(getShadowTy(ST->getElementType(i)));
786 StructType *Res = StructType::get(*MS.C, Elements, ST->isPacked());
787 DEBUG(dbgs() << "getShadowTy: " << *ST << " ===> " << *Res << "\n");
790 uint32_t TypeSize = MS.DL->getTypeSizeInBits(OrigTy);
791 return IntegerType::get(*MS.C, TypeSize);
794 /// \brief Flatten a vector type.
795 Type *getShadowTyNoVec(Type *ty) {
796 if (VectorType *vt = dyn_cast<VectorType>(ty))
797 return IntegerType::get(*MS.C, vt->getBitWidth());
801 /// \brief Convert a shadow value to it's flattened variant.
802 Value *convertToShadowTyNoVec(Value *V, IRBuilder<> &IRB) {
803 Type *Ty = V->getType();
804 Type *NoVecTy = getShadowTyNoVec(Ty);
805 if (Ty == NoVecTy) return V;
806 return IRB.CreateBitCast(V, NoVecTy);
809 /// \brief Compute the integer shadow offset that corresponds to a given
810 /// application address.
812 /// Offset = (Addr & ~AndMask) ^ XorMask
813 Value *getShadowPtrOffset(Value *Addr, IRBuilder<> &IRB) {
814 uint64_t AndMask = MS.MapParams->AndMask;
815 assert(AndMask != 0 && "AndMask shall be specified");
817 IRB.CreateAnd(IRB.CreatePointerCast(Addr, MS.IntptrTy),
818 ConstantInt::get(MS.IntptrTy, ~AndMask));
820 uint64_t XorMask = MS.MapParams->XorMask;
822 OffsetLong = IRB.CreateXor(OffsetLong,
823 ConstantInt::get(MS.IntptrTy, XorMask));
827 /// \brief Compute the shadow address that corresponds to a given application
830 /// Shadow = ShadowBase + Offset
831 Value *getShadowPtr(Value *Addr, Type *ShadowTy,
833 Value *ShadowLong = getShadowPtrOffset(Addr, IRB);
834 uint64_t ShadowBase = MS.MapParams->ShadowBase;
837 IRB.CreateAdd(ShadowLong,
838 ConstantInt::get(MS.IntptrTy, ShadowBase));
839 return IRB.CreateIntToPtr(ShadowLong, PointerType::get(ShadowTy, 0));
842 /// \brief Compute the origin address that corresponds to a given application
845 /// OriginAddr = (OriginBase + Offset) & ~3ULL
846 Value *getOriginPtr(Value *Addr, IRBuilder<> &IRB, unsigned Alignment) {
847 Value *OriginLong = getShadowPtrOffset(Addr, IRB);
848 uint64_t OriginBase = MS.MapParams->OriginBase;
851 IRB.CreateAdd(OriginLong,
852 ConstantInt::get(MS.IntptrTy, OriginBase));
853 if (Alignment < kMinOriginAlignment) {
854 uint64_t Mask = kMinOriginAlignment - 1;
855 OriginLong = IRB.CreateAnd(OriginLong,
856 ConstantInt::get(MS.IntptrTy, ~Mask));
858 return IRB.CreateIntToPtr(OriginLong,
859 PointerType::get(IRB.getInt32Ty(), 0));
862 /// \brief Compute the shadow address for a given function argument.
864 /// Shadow = ParamTLS+ArgOffset.
865 Value *getShadowPtrForArgument(Value *A, IRBuilder<> &IRB,
867 Value *Base = IRB.CreatePointerCast(MS.ParamTLS, MS.IntptrTy);
868 Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
869 return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
873 /// \brief Compute the origin address for a given function argument.
874 Value *getOriginPtrForArgument(Value *A, IRBuilder<> &IRB,
876 if (!MS.TrackOrigins) return nullptr;
877 Value *Base = IRB.CreatePointerCast(MS.ParamOriginTLS, MS.IntptrTy);
878 Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
879 return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
883 /// \brief Compute the shadow address for a retval.
884 Value *getShadowPtrForRetval(Value *A, IRBuilder<> &IRB) {
885 Value *Base = IRB.CreatePointerCast(MS.RetvalTLS, MS.IntptrTy);
886 return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
890 /// \brief Compute the origin address for a retval.
891 Value *getOriginPtrForRetval(IRBuilder<> &IRB) {
892 // We keep a single origin for the entire retval. Might be too optimistic.
893 return MS.RetvalOriginTLS;
896 /// \brief Set SV to be the shadow value for V.
897 void setShadow(Value *V, Value *SV) {
898 assert(!ShadowMap.count(V) && "Values may only have one shadow");
899 ShadowMap[V] = PropagateShadow ? SV : getCleanShadow(V);
902 /// \brief Set Origin to be the origin value for V.
903 void setOrigin(Value *V, Value *Origin) {
904 if (!MS.TrackOrigins) return;
905 assert(!OriginMap.count(V) && "Values may only have one origin");
906 DEBUG(dbgs() << "ORIGIN: " << *V << " ==> " << *Origin << "\n");
907 OriginMap[V] = Origin;
910 /// \brief Create a clean shadow value for a given value.
912 /// Clean shadow (all zeroes) means all bits of the value are defined
914 Constant *getCleanShadow(Value *V) {
915 Type *ShadowTy = getShadowTy(V);
918 return Constant::getNullValue(ShadowTy);
921 /// \brief Create a dirty shadow of a given shadow type.
922 Constant *getPoisonedShadow(Type *ShadowTy) {
924 if (isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy))
925 return Constant::getAllOnesValue(ShadowTy);
926 if (ArrayType *AT = dyn_cast<ArrayType>(ShadowTy)) {
927 SmallVector<Constant *, 4> Vals(AT->getNumElements(),
928 getPoisonedShadow(AT->getElementType()));
929 return ConstantArray::get(AT, Vals);
931 if (StructType *ST = dyn_cast<StructType>(ShadowTy)) {
932 SmallVector<Constant *, 4> Vals;
933 for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
934 Vals.push_back(getPoisonedShadow(ST->getElementType(i)));
935 return ConstantStruct::get(ST, Vals);
937 llvm_unreachable("Unexpected shadow type");
940 /// \brief Create a dirty shadow for a given value.
941 Constant *getPoisonedShadow(Value *V) {
942 Type *ShadowTy = getShadowTy(V);
945 return getPoisonedShadow(ShadowTy);
948 /// \brief Create a clean (zero) origin.
949 Value *getCleanOrigin() {
950 return Constant::getNullValue(MS.OriginTy);
953 /// \brief Get the shadow value for a given Value.
955 /// This function either returns the value set earlier with setShadow,
956 /// or extracts if from ParamTLS (for function arguments).
957 Value *getShadow(Value *V) {
958 if (!PropagateShadow) return getCleanShadow(V);
959 if (Instruction *I = dyn_cast<Instruction>(V)) {
960 // For instructions the shadow is already stored in the map.
961 Value *Shadow = ShadowMap[V];
963 DEBUG(dbgs() << "No shadow: " << *V << "\n" << *(I->getParent()));
965 assert(Shadow && "No shadow for a value");
969 if (UndefValue *U = dyn_cast<UndefValue>(V)) {
970 Value *AllOnes = PoisonUndef ? getPoisonedShadow(V) : getCleanShadow(V);
971 DEBUG(dbgs() << "Undef: " << *U << " ==> " << *AllOnes << "\n");
975 if (Argument *A = dyn_cast<Argument>(V)) {
976 // For arguments we compute the shadow on demand and store it in the map.
977 Value **ShadowPtr = &ShadowMap[V];
980 Function *F = A->getParent();
981 IRBuilder<> EntryIRB(F->getEntryBlock().getFirstNonPHI());
982 unsigned ArgOffset = 0;
983 for (auto &FArg : F->args()) {
984 if (!FArg.getType()->isSized()) {
985 DEBUG(dbgs() << "Arg is not sized\n");
988 unsigned Size = FArg.hasByValAttr()
989 ? MS.DL->getTypeAllocSize(FArg.getType()->getPointerElementType())
990 : MS.DL->getTypeAllocSize(FArg.getType());
992 bool Overflow = ArgOffset + Size > kParamTLSSize;
993 Value *Base = getShadowPtrForArgument(&FArg, EntryIRB, ArgOffset);
994 if (FArg.hasByValAttr()) {
995 // ByVal pointer itself has clean shadow. We copy the actual
996 // argument shadow to the underlying memory.
997 // Figure out maximal valid memcpy alignment.
998 unsigned ArgAlign = FArg.getParamAlignment();
1000 Type *EltType = A->getType()->getPointerElementType();
1001 ArgAlign = MS.DL->getABITypeAlignment(EltType);
1004 // ParamTLS overflow.
1005 EntryIRB.CreateMemSet(
1006 getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB),
1007 Constant::getNullValue(EntryIRB.getInt8Ty()), Size, ArgAlign);
1009 unsigned CopyAlign = std::min(ArgAlign, kShadowTLSAlignment);
1010 Value *Cpy = EntryIRB.CreateMemCpy(
1011 getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB), Base, Size,
1013 DEBUG(dbgs() << " ByValCpy: " << *Cpy << "\n");
1016 *ShadowPtr = getCleanShadow(V);
1019 // ParamTLS overflow.
1020 *ShadowPtr = getCleanShadow(V);
1023 EntryIRB.CreateAlignedLoad(Base, kShadowTLSAlignment);
1026 DEBUG(dbgs() << " ARG: " << FArg << " ==> " <<
1027 **ShadowPtr << "\n");
1028 if (MS.TrackOrigins && !Overflow) {
1030 getOriginPtrForArgument(&FArg, EntryIRB, ArgOffset);
1031 setOrigin(A, EntryIRB.CreateLoad(OriginPtr));
1033 setOrigin(A, getCleanOrigin());
1036 ArgOffset += RoundUpToAlignment(Size, kShadowTLSAlignment);
1038 assert(*ShadowPtr && "Could not find shadow for an argument");
1041 // For everything else the shadow is zero.
1042 return getCleanShadow(V);
1045 /// \brief Get the shadow for i-th argument of the instruction I.
1046 Value *getShadow(Instruction *I, int i) {
1047 return getShadow(I->getOperand(i));
1050 /// \brief Get the origin for a value.
1051 Value *getOrigin(Value *V) {
1052 if (!MS.TrackOrigins) return nullptr;
1053 if (!PropagateShadow) return getCleanOrigin();
1054 if (isa<Constant>(V)) return getCleanOrigin();
1055 assert((isa<Instruction>(V) || isa<Argument>(V)) &&
1056 "Unexpected value type in getOrigin()");
1057 Value *Origin = OriginMap[V];
1058 assert(Origin && "Missing origin");
1062 /// \brief Get the origin for i-th argument of the instruction I.
1063 Value *getOrigin(Instruction *I, int i) {
1064 return getOrigin(I->getOperand(i));
1067 /// \brief Remember the place where a shadow check should be inserted.
1069 /// This location will be later instrumented with a check that will print a
1070 /// UMR warning in runtime if the shadow value is not 0.
1071 void insertShadowCheck(Value *Shadow, Value *Origin, Instruction *OrigIns) {
1073 if (!InsertChecks) return;
1075 Type *ShadowTy = Shadow->getType();
1076 assert((isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy)) &&
1077 "Can only insert checks for integer and vector shadow types");
1079 InstrumentationList.push_back(
1080 ShadowOriginAndInsertPoint(Shadow, Origin, OrigIns));
1083 /// \brief Remember the place where a shadow check should be inserted.
1085 /// This location will be later instrumented with a check that will print a
1086 /// UMR warning in runtime if the value is not fully defined.
1087 void insertShadowCheck(Value *Val, Instruction *OrigIns) {
1089 Value *Shadow, *Origin;
1090 if (ClCheckConstantShadow) {
1091 Shadow = getShadow(Val);
1092 if (!Shadow) return;
1093 Origin = getOrigin(Val);
1095 Shadow = dyn_cast_or_null<Instruction>(getShadow(Val));
1096 if (!Shadow) return;
1097 Origin = dyn_cast_or_null<Instruction>(getOrigin(Val));
1099 insertShadowCheck(Shadow, Origin, OrigIns);
1102 AtomicOrdering addReleaseOrdering(AtomicOrdering a) {
1111 case AcquireRelease:
1112 return AcquireRelease;
1113 case SequentiallyConsistent:
1114 return SequentiallyConsistent;
1116 llvm_unreachable("Unknown ordering");
1119 AtomicOrdering addAcquireOrdering(AtomicOrdering a) {
1128 case AcquireRelease:
1129 return AcquireRelease;
1130 case SequentiallyConsistent:
1131 return SequentiallyConsistent;
1133 llvm_unreachable("Unknown ordering");
1136 // ------------------- Visitors.
1138 /// \brief Instrument LoadInst
1140 /// Loads the corresponding shadow and (optionally) origin.
1141 /// Optionally, checks that the load address is fully defined.
1142 void visitLoadInst(LoadInst &I) {
1143 assert(I.getType()->isSized() && "Load type must have size");
1144 IRBuilder<> IRB(I.getNextNode());
1145 Type *ShadowTy = getShadowTy(&I);
1146 Value *Addr = I.getPointerOperand();
1147 if (PropagateShadow && !I.getMetadata("nosanitize")) {
1148 Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
1150 IRB.CreateAlignedLoad(ShadowPtr, I.getAlignment(), "_msld"));
1152 setShadow(&I, getCleanShadow(&I));
1155 if (ClCheckAccessAddress)
1156 insertShadowCheck(I.getPointerOperand(), &I);
1159 I.setOrdering(addAcquireOrdering(I.getOrdering()));
1161 if (MS.TrackOrigins) {
1162 if (PropagateShadow) {
1163 unsigned Alignment = I.getAlignment();
1164 unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
1165 setOrigin(&I, IRB.CreateAlignedLoad(getOriginPtr(Addr, IRB, Alignment),
1168 setOrigin(&I, getCleanOrigin());
1173 /// \brief Instrument StoreInst
1175 /// Stores the corresponding shadow and (optionally) origin.
1176 /// Optionally, checks that the store address is fully defined.
1177 void visitStoreInst(StoreInst &I) {
1178 StoreList.push_back(&I);
1181 void handleCASOrRMW(Instruction &I) {
1182 assert(isa<AtomicRMWInst>(I) || isa<AtomicCmpXchgInst>(I));
1184 IRBuilder<> IRB(&I);
1185 Value *Addr = I.getOperand(0);
1186 Value *ShadowPtr = getShadowPtr(Addr, I.getType(), IRB);
1188 if (ClCheckAccessAddress)
1189 insertShadowCheck(Addr, &I);
1191 // Only test the conditional argument of cmpxchg instruction.
1192 // The other argument can potentially be uninitialized, but we can not
1193 // detect this situation reliably without possible false positives.
1194 if (isa<AtomicCmpXchgInst>(I))
1195 insertShadowCheck(I.getOperand(1), &I);
1197 IRB.CreateStore(getCleanShadow(&I), ShadowPtr);
1199 setShadow(&I, getCleanShadow(&I));
1200 setOrigin(&I, getCleanOrigin());
1203 void visitAtomicRMWInst(AtomicRMWInst &I) {
1205 I.setOrdering(addReleaseOrdering(I.getOrdering()));
1208 void visitAtomicCmpXchgInst(AtomicCmpXchgInst &I) {
1210 I.setSuccessOrdering(addReleaseOrdering(I.getSuccessOrdering()));
1213 // Vector manipulation.
1214 void visitExtractElementInst(ExtractElementInst &I) {
1215 insertShadowCheck(I.getOperand(1), &I);
1216 IRBuilder<> IRB(&I);
1217 setShadow(&I, IRB.CreateExtractElement(getShadow(&I, 0), I.getOperand(1),
1219 setOrigin(&I, getOrigin(&I, 0));
1222 void visitInsertElementInst(InsertElementInst &I) {
1223 insertShadowCheck(I.getOperand(2), &I);
1224 IRBuilder<> IRB(&I);
1225 setShadow(&I, IRB.CreateInsertElement(getShadow(&I, 0), getShadow(&I, 1),
1226 I.getOperand(2), "_msprop"));
1227 setOriginForNaryOp(I);
1230 void visitShuffleVectorInst(ShuffleVectorInst &I) {
1231 insertShadowCheck(I.getOperand(2), &I);
1232 IRBuilder<> IRB(&I);
1233 setShadow(&I, IRB.CreateShuffleVector(getShadow(&I, 0), getShadow(&I, 1),
1234 I.getOperand(2), "_msprop"));
1235 setOriginForNaryOp(I);
1239 void visitSExtInst(SExtInst &I) {
1240 IRBuilder<> IRB(&I);
1241 setShadow(&I, IRB.CreateSExt(getShadow(&I, 0), I.getType(), "_msprop"));
1242 setOrigin(&I, getOrigin(&I, 0));
1245 void visitZExtInst(ZExtInst &I) {
1246 IRBuilder<> IRB(&I);
1247 setShadow(&I, IRB.CreateZExt(getShadow(&I, 0), I.getType(), "_msprop"));
1248 setOrigin(&I, getOrigin(&I, 0));
1251 void visitTruncInst(TruncInst &I) {
1252 IRBuilder<> IRB(&I);
1253 setShadow(&I, IRB.CreateTrunc(getShadow(&I, 0), I.getType(), "_msprop"));
1254 setOrigin(&I, getOrigin(&I, 0));
1257 void visitBitCastInst(BitCastInst &I) {
1258 IRBuilder<> IRB(&I);
1259 setShadow(&I, IRB.CreateBitCast(getShadow(&I, 0), getShadowTy(&I)));
1260 setOrigin(&I, getOrigin(&I, 0));
1263 void visitPtrToIntInst(PtrToIntInst &I) {
1264 IRBuilder<> IRB(&I);
1265 setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1266 "_msprop_ptrtoint"));
1267 setOrigin(&I, getOrigin(&I, 0));
1270 void visitIntToPtrInst(IntToPtrInst &I) {
1271 IRBuilder<> IRB(&I);
1272 setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1273 "_msprop_inttoptr"));
1274 setOrigin(&I, getOrigin(&I, 0));
1277 void visitFPToSIInst(CastInst& I) { handleShadowOr(I); }
1278 void visitFPToUIInst(CastInst& I) { handleShadowOr(I); }
1279 void visitSIToFPInst(CastInst& I) { handleShadowOr(I); }
1280 void visitUIToFPInst(CastInst& I) { handleShadowOr(I); }
1281 void visitFPExtInst(CastInst& I) { handleShadowOr(I); }
1282 void visitFPTruncInst(CastInst& I) { handleShadowOr(I); }
1284 /// \brief Propagate shadow for bitwise AND.
1286 /// This code is exact, i.e. if, for example, a bit in the left argument
1287 /// is defined and 0, then neither the value not definedness of the
1288 /// corresponding bit in B don't affect the resulting shadow.
1289 void visitAnd(BinaryOperator &I) {
1290 IRBuilder<> IRB(&I);
1291 // "And" of 0 and a poisoned value results in unpoisoned value.
1292 // 1&1 => 1; 0&1 => 0; p&1 => p;
1293 // 1&0 => 0; 0&0 => 0; p&0 => 0;
1294 // 1&p => p; 0&p => 0; p&p => p;
1295 // S = (S1 & S2) | (V1 & S2) | (S1 & V2)
1296 Value *S1 = getShadow(&I, 0);
1297 Value *S2 = getShadow(&I, 1);
1298 Value *V1 = I.getOperand(0);
1299 Value *V2 = I.getOperand(1);
1300 if (V1->getType() != S1->getType()) {
1301 V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1302 V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1304 Value *S1S2 = IRB.CreateAnd(S1, S2);
1305 Value *V1S2 = IRB.CreateAnd(V1, S2);
1306 Value *S1V2 = IRB.CreateAnd(S1, V2);
1307 setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
1308 setOriginForNaryOp(I);
1311 void visitOr(BinaryOperator &I) {
1312 IRBuilder<> IRB(&I);
1313 // "Or" of 1 and a poisoned value results in unpoisoned value.
1314 // 1|1 => 1; 0|1 => 1; p|1 => 1;
1315 // 1|0 => 1; 0|0 => 0; p|0 => p;
1316 // 1|p => 1; 0|p => p; p|p => p;
1317 // S = (S1 & S2) | (~V1 & S2) | (S1 & ~V2)
1318 Value *S1 = getShadow(&I, 0);
1319 Value *S2 = getShadow(&I, 1);
1320 Value *V1 = IRB.CreateNot(I.getOperand(0));
1321 Value *V2 = IRB.CreateNot(I.getOperand(1));
1322 if (V1->getType() != S1->getType()) {
1323 V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1324 V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1326 Value *S1S2 = IRB.CreateAnd(S1, S2);
1327 Value *V1S2 = IRB.CreateAnd(V1, S2);
1328 Value *S1V2 = IRB.CreateAnd(S1, V2);
1329 setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
1330 setOriginForNaryOp(I);
1333 /// \brief Default propagation of shadow and/or origin.
1335 /// This class implements the general case of shadow propagation, used in all
1336 /// cases where we don't know and/or don't care about what the operation
1337 /// actually does. It converts all input shadow values to a common type
1338 /// (extending or truncating as necessary), and bitwise OR's them.
1340 /// This is much cheaper than inserting checks (i.e. requiring inputs to be
1341 /// fully initialized), and less prone to false positives.
1343 /// This class also implements the general case of origin propagation. For a
1344 /// Nary operation, result origin is set to the origin of an argument that is
1345 /// not entirely initialized. If there is more than one such arguments, the
1346 /// rightmost of them is picked. It does not matter which one is picked if all
1347 /// arguments are initialized.
1348 template <bool CombineShadow>
1353 MemorySanitizerVisitor *MSV;
1356 Combiner(MemorySanitizerVisitor *MSV, IRBuilder<> &IRB) :
1357 Shadow(nullptr), Origin(nullptr), IRB(IRB), MSV(MSV) {}
1359 /// \brief Add a pair of shadow and origin values to the mix.
1360 Combiner &Add(Value *OpShadow, Value *OpOrigin) {
1361 if (CombineShadow) {
1366 OpShadow = MSV->CreateShadowCast(IRB, OpShadow, Shadow->getType());
1367 Shadow = IRB.CreateOr(Shadow, OpShadow, "_msprop");
1371 if (MSV->MS.TrackOrigins) {
1376 Constant *ConstOrigin = dyn_cast<Constant>(OpOrigin);
1377 // No point in adding something that might result in 0 origin value.
1378 if (!ConstOrigin || !ConstOrigin->isNullValue()) {
1379 Value *FlatShadow = MSV->convertToShadowTyNoVec(OpShadow, IRB);
1381 IRB.CreateICmpNE(FlatShadow, MSV->getCleanShadow(FlatShadow));
1382 Origin = IRB.CreateSelect(Cond, OpOrigin, Origin);
1389 /// \brief Add an application value to the mix.
1390 Combiner &Add(Value *V) {
1391 Value *OpShadow = MSV->getShadow(V);
1392 Value *OpOrigin = MSV->MS.TrackOrigins ? MSV->getOrigin(V) : nullptr;
1393 return Add(OpShadow, OpOrigin);
1396 /// \brief Set the current combined values as the given instruction's shadow
1398 void Done(Instruction *I) {
1399 if (CombineShadow) {
1401 Shadow = MSV->CreateShadowCast(IRB, Shadow, MSV->getShadowTy(I));
1402 MSV->setShadow(I, Shadow);
1404 if (MSV->MS.TrackOrigins) {
1406 MSV->setOrigin(I, Origin);
1411 typedef Combiner<true> ShadowAndOriginCombiner;
1412 typedef Combiner<false> OriginCombiner;
1414 /// \brief Propagate origin for arbitrary operation.
1415 void setOriginForNaryOp(Instruction &I) {
1416 if (!MS.TrackOrigins) return;
1417 IRBuilder<> IRB(&I);
1418 OriginCombiner OC(this, IRB);
1419 for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
1424 size_t VectorOrPrimitiveTypeSizeInBits(Type *Ty) {
1425 assert(!(Ty->isVectorTy() && Ty->getScalarType()->isPointerTy()) &&
1426 "Vector of pointers is not a valid shadow type");
1427 return Ty->isVectorTy() ?
1428 Ty->getVectorNumElements() * Ty->getScalarSizeInBits() :
1429 Ty->getPrimitiveSizeInBits();
1432 /// \brief Cast between two shadow types, extending or truncating as
1434 Value *CreateShadowCast(IRBuilder<> &IRB, Value *V, Type *dstTy,
1435 bool Signed = false) {
1436 Type *srcTy = V->getType();
1437 if (dstTy->isIntegerTy() && srcTy->isIntegerTy())
1438 return IRB.CreateIntCast(V, dstTy, Signed);
1439 if (dstTy->isVectorTy() && srcTy->isVectorTy() &&
1440 dstTy->getVectorNumElements() == srcTy->getVectorNumElements())
1441 return IRB.CreateIntCast(V, dstTy, Signed);
1442 size_t srcSizeInBits = VectorOrPrimitiveTypeSizeInBits(srcTy);
1443 size_t dstSizeInBits = VectorOrPrimitiveTypeSizeInBits(dstTy);
1444 Value *V1 = IRB.CreateBitCast(V, Type::getIntNTy(*MS.C, srcSizeInBits));
1446 IRB.CreateIntCast(V1, Type::getIntNTy(*MS.C, dstSizeInBits), Signed);
1447 return IRB.CreateBitCast(V2, dstTy);
1448 // TODO: handle struct types.
1451 /// \brief Cast an application value to the type of its own shadow.
1452 Value *CreateAppToShadowCast(IRBuilder<> &IRB, Value *V) {
1453 Type *ShadowTy = getShadowTy(V);
1454 if (V->getType() == ShadowTy)
1456 if (V->getType()->isPtrOrPtrVectorTy())
1457 return IRB.CreatePtrToInt(V, ShadowTy);
1459 return IRB.CreateBitCast(V, ShadowTy);
1462 /// \brief Propagate shadow for arbitrary operation.
1463 void handleShadowOr(Instruction &I) {
1464 IRBuilder<> IRB(&I);
1465 ShadowAndOriginCombiner SC(this, IRB);
1466 for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
1471 // \brief Handle multiplication by constant.
1473 // Handle a special case of multiplication by constant that may have one or
1474 // more zeros in the lower bits. This makes corresponding number of lower bits
1475 // of the result zero as well. We model it by shifting the other operand
1476 // shadow left by the required number of bits. Effectively, we transform
1477 // (X * (A * 2**B)) to ((X << B) * A) and instrument (X << B) as (Sx << B).
1478 // We use multiplication by 2**N instead of shift to cover the case of
1479 // multiplication by 0, which may occur in some elements of a vector operand.
1480 void handleMulByConstant(BinaryOperator &I, Constant *ConstArg,
1482 Constant *ShadowMul;
1483 Type *Ty = ConstArg->getType();
1484 if (Ty->isVectorTy()) {
1485 unsigned NumElements = Ty->getVectorNumElements();
1486 Type *EltTy = Ty->getSequentialElementType();
1487 SmallVector<Constant *, 16> Elements;
1488 for (unsigned Idx = 0; Idx < NumElements; ++Idx) {
1490 dyn_cast<ConstantInt>(ConstArg->getAggregateElement(Idx));
1491 APInt V = Elt->getValue();
1492 APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
1493 Elements.push_back(ConstantInt::get(EltTy, V2));
1495 ShadowMul = ConstantVector::get(Elements);
1497 ConstantInt *Elt = dyn_cast<ConstantInt>(ConstArg);
1498 APInt V = Elt->getValue();
1499 APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
1500 ShadowMul = ConstantInt::get(Elt->getType(), V2);
1503 IRBuilder<> IRB(&I);
1505 IRB.CreateMul(getShadow(OtherArg), ShadowMul, "msprop_mul_cst"));
1506 setOrigin(&I, getOrigin(OtherArg));
1509 void visitMul(BinaryOperator &I) {
1510 Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
1511 Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
1512 if (constOp0 && !constOp1)
1513 handleMulByConstant(I, constOp0, I.getOperand(1));
1514 else if (constOp1 && !constOp0)
1515 handleMulByConstant(I, constOp1, I.getOperand(0));
1520 void visitFAdd(BinaryOperator &I) { handleShadowOr(I); }
1521 void visitFSub(BinaryOperator &I) { handleShadowOr(I); }
1522 void visitFMul(BinaryOperator &I) { handleShadowOr(I); }
1523 void visitAdd(BinaryOperator &I) { handleShadowOr(I); }
1524 void visitSub(BinaryOperator &I) { handleShadowOr(I); }
1525 void visitXor(BinaryOperator &I) { handleShadowOr(I); }
1527 void handleDiv(Instruction &I) {
1528 IRBuilder<> IRB(&I);
1529 // Strict on the second argument.
1530 insertShadowCheck(I.getOperand(1), &I);
1531 setShadow(&I, getShadow(&I, 0));
1532 setOrigin(&I, getOrigin(&I, 0));
1535 void visitUDiv(BinaryOperator &I) { handleDiv(I); }
1536 void visitSDiv(BinaryOperator &I) { handleDiv(I); }
1537 void visitFDiv(BinaryOperator &I) { handleDiv(I); }
1538 void visitURem(BinaryOperator &I) { handleDiv(I); }
1539 void visitSRem(BinaryOperator &I) { handleDiv(I); }
1540 void visitFRem(BinaryOperator &I) { handleDiv(I); }
1542 /// \brief Instrument == and != comparisons.
1544 /// Sometimes the comparison result is known even if some of the bits of the
1545 /// arguments are not.
1546 void handleEqualityComparison(ICmpInst &I) {
1547 IRBuilder<> IRB(&I);
1548 Value *A = I.getOperand(0);
1549 Value *B = I.getOperand(1);
1550 Value *Sa = getShadow(A);
1551 Value *Sb = getShadow(B);
1553 // Get rid of pointers and vectors of pointers.
1554 // For ints (and vectors of ints), types of A and Sa match,
1555 // and this is a no-op.
1556 A = IRB.CreatePointerCast(A, Sa->getType());
1557 B = IRB.CreatePointerCast(B, Sb->getType());
1559 // A == B <==> (C = A^B) == 0
1560 // A != B <==> (C = A^B) != 0
1562 Value *C = IRB.CreateXor(A, B);
1563 Value *Sc = IRB.CreateOr(Sa, Sb);
1564 // Now dealing with i = (C == 0) comparison (or C != 0, does not matter now)
1565 // Result is defined if one of the following is true
1566 // * there is a defined 1 bit in C
1567 // * C is fully defined
1568 // Si = !(C & ~Sc) && Sc
1569 Value *Zero = Constant::getNullValue(Sc->getType());
1570 Value *MinusOne = Constant::getAllOnesValue(Sc->getType());
1572 IRB.CreateAnd(IRB.CreateICmpNE(Sc, Zero),
1574 IRB.CreateAnd(IRB.CreateXor(Sc, MinusOne), C), Zero));
1575 Si->setName("_msprop_icmp");
1577 setOriginForNaryOp(I);
1580 /// \brief Build the lowest possible value of V, taking into account V's
1581 /// uninitialized bits.
1582 Value *getLowestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
1585 // Split shadow into sign bit and other bits.
1586 Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
1587 Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
1588 // Maximise the undefined shadow bit, minimize other undefined bits.
1590 IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaOtherBits)), SaSignBit);
1592 // Minimize undefined bits.
1593 return IRB.CreateAnd(A, IRB.CreateNot(Sa));
1597 /// \brief Build the highest possible value of V, taking into account V's
1598 /// uninitialized bits.
1599 Value *getHighestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
1602 // Split shadow into sign bit and other bits.
1603 Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
1604 Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
1605 // Minimise the undefined shadow bit, maximise other undefined bits.
1607 IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaSignBit)), SaOtherBits);
1609 // Maximize undefined bits.
1610 return IRB.CreateOr(A, Sa);
1614 /// \brief Instrument relational comparisons.
1616 /// This function does exact shadow propagation for all relational
1617 /// comparisons of integers, pointers and vectors of those.
1618 /// FIXME: output seems suboptimal when one of the operands is a constant
1619 void handleRelationalComparisonExact(ICmpInst &I) {
1620 IRBuilder<> IRB(&I);
1621 Value *A = I.getOperand(0);
1622 Value *B = I.getOperand(1);
1623 Value *Sa = getShadow(A);
1624 Value *Sb = getShadow(B);
1626 // Get rid of pointers and vectors of pointers.
1627 // For ints (and vectors of ints), types of A and Sa match,
1628 // and this is a no-op.
1629 A = IRB.CreatePointerCast(A, Sa->getType());
1630 B = IRB.CreatePointerCast(B, Sb->getType());
1632 // Let [a0, a1] be the interval of possible values of A, taking into account
1633 // its undefined bits. Let [b0, b1] be the interval of possible values of B.
1634 // Then (A cmp B) is defined iff (a0 cmp b1) == (a1 cmp b0).
1635 bool IsSigned = I.isSigned();
1636 Value *S1 = IRB.CreateICmp(I.getPredicate(),
1637 getLowestPossibleValue(IRB, A, Sa, IsSigned),
1638 getHighestPossibleValue(IRB, B, Sb, IsSigned));
1639 Value *S2 = IRB.CreateICmp(I.getPredicate(),
1640 getHighestPossibleValue(IRB, A, Sa, IsSigned),
1641 getLowestPossibleValue(IRB, B, Sb, IsSigned));
1642 Value *Si = IRB.CreateXor(S1, S2);
1644 setOriginForNaryOp(I);
1647 /// \brief Instrument signed relational comparisons.
1649 /// Handle (x<0) and (x>=0) comparisons (essentially, sign bit tests) by
1650 /// propagating the highest bit of the shadow. Everything else is delegated
1651 /// to handleShadowOr().
1652 void handleSignedRelationalComparison(ICmpInst &I) {
1653 Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
1654 Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
1655 Value* op = nullptr;
1656 CmpInst::Predicate pre = I.getPredicate();
1657 if (constOp0 && constOp0->isNullValue() &&
1658 (pre == CmpInst::ICMP_SGT || pre == CmpInst::ICMP_SLE)) {
1659 op = I.getOperand(1);
1660 } else if (constOp1 && constOp1->isNullValue() &&
1661 (pre == CmpInst::ICMP_SLT || pre == CmpInst::ICMP_SGE)) {
1662 op = I.getOperand(0);
1665 IRBuilder<> IRB(&I);
1667 IRB.CreateICmpSLT(getShadow(op), getCleanShadow(op), "_msprop_icmpslt");
1668 setShadow(&I, Shadow);
1669 setOrigin(&I, getOrigin(op));
1675 void visitICmpInst(ICmpInst &I) {
1676 if (!ClHandleICmp) {
1680 if (I.isEquality()) {
1681 handleEqualityComparison(I);
1685 assert(I.isRelational());
1686 if (ClHandleICmpExact) {
1687 handleRelationalComparisonExact(I);
1691 handleSignedRelationalComparison(I);
1695 assert(I.isUnsigned());
1696 if ((isa<Constant>(I.getOperand(0)) || isa<Constant>(I.getOperand(1)))) {
1697 handleRelationalComparisonExact(I);
1704 void visitFCmpInst(FCmpInst &I) {
1708 void handleShift(BinaryOperator &I) {
1709 IRBuilder<> IRB(&I);
1710 // If any of the S2 bits are poisoned, the whole thing is poisoned.
1711 // Otherwise perform the same shift on S1.
1712 Value *S1 = getShadow(&I, 0);
1713 Value *S2 = getShadow(&I, 1);
1714 Value *S2Conv = IRB.CreateSExt(IRB.CreateICmpNE(S2, getCleanShadow(S2)),
1716 Value *V2 = I.getOperand(1);
1717 Value *Shift = IRB.CreateBinOp(I.getOpcode(), S1, V2);
1718 setShadow(&I, IRB.CreateOr(Shift, S2Conv));
1719 setOriginForNaryOp(I);
1722 void visitShl(BinaryOperator &I) { handleShift(I); }
1723 void visitAShr(BinaryOperator &I) { handleShift(I); }
1724 void visitLShr(BinaryOperator &I) { handleShift(I); }
1726 /// \brief Instrument llvm.memmove
1728 /// At this point we don't know if llvm.memmove will be inlined or not.
1729 /// If we don't instrument it and it gets inlined,
1730 /// our interceptor will not kick in and we will lose the memmove.
1731 /// If we instrument the call here, but it does not get inlined,
1732 /// we will memove the shadow twice: which is bad in case
1733 /// of overlapping regions. So, we simply lower the intrinsic to a call.
1735 /// Similar situation exists for memcpy and memset.
1736 void visitMemMoveInst(MemMoveInst &I) {
1737 IRBuilder<> IRB(&I);
1740 IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
1741 IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
1742 IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
1743 I.eraseFromParent();
1746 // Similar to memmove: avoid copying shadow twice.
1747 // This is somewhat unfortunate as it may slowdown small constant memcpys.
1748 // FIXME: consider doing manual inline for small constant sizes and proper
1750 void visitMemCpyInst(MemCpyInst &I) {
1751 IRBuilder<> IRB(&I);
1754 IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
1755 IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
1756 IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
1757 I.eraseFromParent();
1761 void visitMemSetInst(MemSetInst &I) {
1762 IRBuilder<> IRB(&I);
1765 IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
1766 IRB.CreateIntCast(I.getArgOperand(1), IRB.getInt32Ty(), false),
1767 IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false));
1768 I.eraseFromParent();
1771 void visitVAStartInst(VAStartInst &I) {
1772 VAHelper->visitVAStartInst(I);
1775 void visitVACopyInst(VACopyInst &I) {
1776 VAHelper->visitVACopyInst(I);
1779 enum IntrinsicKind {
1780 IK_DoesNotAccessMemory,
1785 static IntrinsicKind getIntrinsicKind(Intrinsic::ID iid) {
1786 const int DoesNotAccessMemory = IK_DoesNotAccessMemory;
1787 const int OnlyReadsArgumentPointees = IK_OnlyReadsMemory;
1788 const int OnlyReadsMemory = IK_OnlyReadsMemory;
1789 const int OnlyAccessesArgumentPointees = IK_WritesMemory;
1790 const int UnknownModRefBehavior = IK_WritesMemory;
1791 #define GET_INTRINSIC_MODREF_BEHAVIOR
1792 #define ModRefBehavior IntrinsicKind
1793 #include "llvm/IR/Intrinsics.gen"
1794 #undef ModRefBehavior
1795 #undef GET_INTRINSIC_MODREF_BEHAVIOR
1798 /// \brief Handle vector store-like intrinsics.
1800 /// Instrument intrinsics that look like a simple SIMD store: writes memory,
1801 /// has 1 pointer argument and 1 vector argument, returns void.
1802 bool handleVectorStoreIntrinsic(IntrinsicInst &I) {
1803 IRBuilder<> IRB(&I);
1804 Value* Addr = I.getArgOperand(0);
1805 Value *Shadow = getShadow(&I, 1);
1806 Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);
1808 // We don't know the pointer alignment (could be unaligned SSE store!).
1809 // Have to assume to worst case.
1810 IRB.CreateAlignedStore(Shadow, ShadowPtr, 1);
1812 if (ClCheckAccessAddress)
1813 insertShadowCheck(Addr, &I);
1815 // FIXME: use ClStoreCleanOrigin
1816 // FIXME: factor out common code from materializeStores
1817 if (MS.TrackOrigins)
1818 IRB.CreateStore(getOrigin(&I, 1), getOriginPtr(Addr, IRB, 1));
1822 /// \brief Handle vector load-like intrinsics.
1824 /// Instrument intrinsics that look like a simple SIMD load: reads memory,
1825 /// has 1 pointer argument, returns a vector.
1826 bool handleVectorLoadIntrinsic(IntrinsicInst &I) {
1827 IRBuilder<> IRB(&I);
1828 Value *Addr = I.getArgOperand(0);
1830 Type *ShadowTy = getShadowTy(&I);
1831 if (PropagateShadow) {
1832 Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
1833 // We don't know the pointer alignment (could be unaligned SSE load!).
1834 // Have to assume to worst case.
1835 setShadow(&I, IRB.CreateAlignedLoad(ShadowPtr, 1, "_msld"));
1837 setShadow(&I, getCleanShadow(&I));
1840 if (ClCheckAccessAddress)
1841 insertShadowCheck(Addr, &I);
1843 if (MS.TrackOrigins) {
1844 if (PropagateShadow)
1845 setOrigin(&I, IRB.CreateLoad(getOriginPtr(Addr, IRB, 1)));
1847 setOrigin(&I, getCleanOrigin());
1852 /// \brief Handle (SIMD arithmetic)-like intrinsics.
1854 /// Instrument intrinsics with any number of arguments of the same type,
1855 /// equal to the return type. The type should be simple (no aggregates or
1856 /// pointers; vectors are fine).
1857 /// Caller guarantees that this intrinsic does not access memory.
1858 bool maybeHandleSimpleNomemIntrinsic(IntrinsicInst &I) {
1859 Type *RetTy = I.getType();
1860 if (!(RetTy->isIntOrIntVectorTy() ||
1861 RetTy->isFPOrFPVectorTy() ||
1862 RetTy->isX86_MMXTy()))
1865 unsigned NumArgOperands = I.getNumArgOperands();
1867 for (unsigned i = 0; i < NumArgOperands; ++i) {
1868 Type *Ty = I.getArgOperand(i)->getType();
1873 IRBuilder<> IRB(&I);
1874 ShadowAndOriginCombiner SC(this, IRB);
1875 for (unsigned i = 0; i < NumArgOperands; ++i)
1876 SC.Add(I.getArgOperand(i));
1882 /// \brief Heuristically instrument unknown intrinsics.
1884 /// The main purpose of this code is to do something reasonable with all
1885 /// random intrinsics we might encounter, most importantly - SIMD intrinsics.
1886 /// We recognize several classes of intrinsics by their argument types and
1887 /// ModRefBehaviour and apply special intrumentation when we are reasonably
1888 /// sure that we know what the intrinsic does.
1890 /// We special-case intrinsics where this approach fails. See llvm.bswap
1891 /// handling as an example of that.
1892 bool handleUnknownIntrinsic(IntrinsicInst &I) {
1893 unsigned NumArgOperands = I.getNumArgOperands();
1894 if (NumArgOperands == 0)
1897 Intrinsic::ID iid = I.getIntrinsicID();
1898 IntrinsicKind IK = getIntrinsicKind(iid);
1899 bool OnlyReadsMemory = IK == IK_OnlyReadsMemory;
1900 bool WritesMemory = IK == IK_WritesMemory;
1901 assert(!(OnlyReadsMemory && WritesMemory));
1903 if (NumArgOperands == 2 &&
1904 I.getArgOperand(0)->getType()->isPointerTy() &&
1905 I.getArgOperand(1)->getType()->isVectorTy() &&
1906 I.getType()->isVoidTy() &&
1908 // This looks like a vector store.
1909 return handleVectorStoreIntrinsic(I);
1912 if (NumArgOperands == 1 &&
1913 I.getArgOperand(0)->getType()->isPointerTy() &&
1914 I.getType()->isVectorTy() &&
1916 // This looks like a vector load.
1917 return handleVectorLoadIntrinsic(I);
1920 if (!OnlyReadsMemory && !WritesMemory)
1921 if (maybeHandleSimpleNomemIntrinsic(I))
1924 // FIXME: detect and handle SSE maskstore/maskload
1928 void handleBswap(IntrinsicInst &I) {
1929 IRBuilder<> IRB(&I);
1930 Value *Op = I.getArgOperand(0);
1931 Type *OpType = Op->getType();
1932 Function *BswapFunc = Intrinsic::getDeclaration(
1933 F.getParent(), Intrinsic::bswap, makeArrayRef(&OpType, 1));
1934 setShadow(&I, IRB.CreateCall(BswapFunc, getShadow(Op)));
1935 setOrigin(&I, getOrigin(Op));
1938 // \brief Instrument vector convert instrinsic.
1940 // This function instruments intrinsics like cvtsi2ss:
1941 // %Out = int_xxx_cvtyyy(%ConvertOp)
1943 // %Out = int_xxx_cvtyyy(%CopyOp, %ConvertOp)
1944 // Intrinsic converts \p NumUsedElements elements of \p ConvertOp to the same
1945 // number \p Out elements, and (if has 2 arguments) copies the rest of the
1946 // elements from \p CopyOp.
1947 // In most cases conversion involves floating-point value which may trigger a
1948 // hardware exception when not fully initialized. For this reason we require
1949 // \p ConvertOp[0:NumUsedElements] to be fully initialized and trap otherwise.
1950 // We copy the shadow of \p CopyOp[NumUsedElements:] to \p
1951 // Out[NumUsedElements:]. This means that intrinsics without \p CopyOp always
1952 // return a fully initialized value.
1953 void handleVectorConvertIntrinsic(IntrinsicInst &I, int NumUsedElements) {
1954 IRBuilder<> IRB(&I);
1955 Value *CopyOp, *ConvertOp;
1957 switch (I.getNumArgOperands()) {
1959 CopyOp = I.getArgOperand(0);
1960 ConvertOp = I.getArgOperand(1);
1963 ConvertOp = I.getArgOperand(0);
1967 llvm_unreachable("Cvt intrinsic with unsupported number of arguments.");
1970 // The first *NumUsedElements* elements of ConvertOp are converted to the
1971 // same number of output elements. The rest of the output is copied from
1972 // CopyOp, or (if not available) filled with zeroes.
1973 // Combine shadow for elements of ConvertOp that are used in this operation,
1974 // and insert a check.
1975 // FIXME: consider propagating shadow of ConvertOp, at least in the case of
1976 // int->any conversion.
1977 Value *ConvertShadow = getShadow(ConvertOp);
1978 Value *AggShadow = nullptr;
1979 if (ConvertOp->getType()->isVectorTy()) {
1980 AggShadow = IRB.CreateExtractElement(
1981 ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), 0));
1982 for (int i = 1; i < NumUsedElements; ++i) {
1983 Value *MoreShadow = IRB.CreateExtractElement(
1984 ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), i));
1985 AggShadow = IRB.CreateOr(AggShadow, MoreShadow);
1988 AggShadow = ConvertShadow;
1990 assert(AggShadow->getType()->isIntegerTy());
1991 insertShadowCheck(AggShadow, getOrigin(ConvertOp), &I);
1993 // Build result shadow by zero-filling parts of CopyOp shadow that come from
1996 assert(CopyOp->getType() == I.getType());
1997 assert(CopyOp->getType()->isVectorTy());
1998 Value *ResultShadow = getShadow(CopyOp);
1999 Type *EltTy = ResultShadow->getType()->getVectorElementType();
2000 for (int i = 0; i < NumUsedElements; ++i) {
2001 ResultShadow = IRB.CreateInsertElement(
2002 ResultShadow, ConstantInt::getNullValue(EltTy),
2003 ConstantInt::get(IRB.getInt32Ty(), i));
2005 setShadow(&I, ResultShadow);
2006 setOrigin(&I, getOrigin(CopyOp));
2008 setShadow(&I, getCleanShadow(&I));
2009 setOrigin(&I, getCleanOrigin());
2013 // Given a scalar or vector, extract lower 64 bits (or less), and return all
2014 // zeroes if it is zero, and all ones otherwise.
2015 Value *Lower64ShadowExtend(IRBuilder<> &IRB, Value *S, Type *T) {
2016 if (S->getType()->isVectorTy())
2017 S = CreateShadowCast(IRB, S, IRB.getInt64Ty(), /* Signed */ true);
2018 assert(S->getType()->getPrimitiveSizeInBits() <= 64);
2019 Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
2020 return CreateShadowCast(IRB, S2, T, /* Signed */ true);
2023 Value *VariableShadowExtend(IRBuilder<> &IRB, Value *S) {
2024 Type *T = S->getType();
2025 assert(T->isVectorTy());
2026 Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
2027 return IRB.CreateSExt(S2, T);
2030 // \brief Instrument vector shift instrinsic.
2032 // This function instruments intrinsics like int_x86_avx2_psll_w.
2033 // Intrinsic shifts %In by %ShiftSize bits.
2034 // %ShiftSize may be a vector. In that case the lower 64 bits determine shift
2035 // size, and the rest is ignored. Behavior is defined even if shift size is
2036 // greater than register (or field) width.
2037 void handleVectorShiftIntrinsic(IntrinsicInst &I, bool Variable) {
2038 assert(I.getNumArgOperands() == 2);
2039 IRBuilder<> IRB(&I);
2040 // If any of the S2 bits are poisoned, the whole thing is poisoned.
2041 // Otherwise perform the same shift on S1.
2042 Value *S1 = getShadow(&I, 0);
2043 Value *S2 = getShadow(&I, 1);
2044 Value *S2Conv = Variable ? VariableShadowExtend(IRB, S2)
2045 : Lower64ShadowExtend(IRB, S2, getShadowTy(&I));
2046 Value *V1 = I.getOperand(0);
2047 Value *V2 = I.getOperand(1);
2048 Value *Shift = IRB.CreateCall2(I.getCalledValue(),
2049 IRB.CreateBitCast(S1, V1->getType()), V2);
2050 Shift = IRB.CreateBitCast(Shift, getShadowTy(&I));
2051 setShadow(&I, IRB.CreateOr(Shift, S2Conv));
2052 setOriginForNaryOp(I);
2055 // \brief Get an X86_MMX-sized vector type.
2056 Type *getMMXVectorTy(unsigned EltSizeInBits) {
2057 const unsigned X86_MMXSizeInBits = 64;
2058 return VectorType::get(IntegerType::get(*MS.C, EltSizeInBits),
2059 X86_MMXSizeInBits / EltSizeInBits);
2062 // \brief Returns a signed counterpart for an (un)signed-saturate-and-pack
2064 Intrinsic::ID getSignedPackIntrinsic(Intrinsic::ID id) {
2066 case llvm::Intrinsic::x86_sse2_packsswb_128:
2067 case llvm::Intrinsic::x86_sse2_packuswb_128:
2068 return llvm::Intrinsic::x86_sse2_packsswb_128;
2070 case llvm::Intrinsic::x86_sse2_packssdw_128:
2071 case llvm::Intrinsic::x86_sse41_packusdw:
2072 return llvm::Intrinsic::x86_sse2_packssdw_128;
2074 case llvm::Intrinsic::x86_avx2_packsswb:
2075 case llvm::Intrinsic::x86_avx2_packuswb:
2076 return llvm::Intrinsic::x86_avx2_packsswb;
2078 case llvm::Intrinsic::x86_avx2_packssdw:
2079 case llvm::Intrinsic::x86_avx2_packusdw:
2080 return llvm::Intrinsic::x86_avx2_packssdw;
2082 case llvm::Intrinsic::x86_mmx_packsswb:
2083 case llvm::Intrinsic::x86_mmx_packuswb:
2084 return llvm::Intrinsic::x86_mmx_packsswb;
2086 case llvm::Intrinsic::x86_mmx_packssdw:
2087 return llvm::Intrinsic::x86_mmx_packssdw;
2089 llvm_unreachable("unexpected intrinsic id");
2093 // \brief Instrument vector pack instrinsic.
2095 // This function instruments intrinsics like x86_mmx_packsswb, that
2096 // packs elements of 2 input vectors into half as many bits with saturation.
2097 // Shadow is propagated with the signed variant of the same intrinsic applied
2098 // to sext(Sa != zeroinitializer), sext(Sb != zeroinitializer).
2099 // EltSizeInBits is used only for x86mmx arguments.
2100 void handleVectorPackIntrinsic(IntrinsicInst &I, unsigned EltSizeInBits = 0) {
2101 assert(I.getNumArgOperands() == 2);
2102 bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2103 IRBuilder<> IRB(&I);
2104 Value *S1 = getShadow(&I, 0);
2105 Value *S2 = getShadow(&I, 1);
2106 assert(isX86_MMX || S1->getType()->isVectorTy());
2108 // SExt and ICmpNE below must apply to individual elements of input vectors.
2109 // In case of x86mmx arguments, cast them to appropriate vector types and
2111 Type *T = isX86_MMX ? getMMXVectorTy(EltSizeInBits) : S1->getType();
2113 S1 = IRB.CreateBitCast(S1, T);
2114 S2 = IRB.CreateBitCast(S2, T);
2116 Value *S1_ext = IRB.CreateSExt(
2117 IRB.CreateICmpNE(S1, llvm::Constant::getNullValue(T)), T);
2118 Value *S2_ext = IRB.CreateSExt(
2119 IRB.CreateICmpNE(S2, llvm::Constant::getNullValue(T)), T);
2121 Type *X86_MMXTy = Type::getX86_MMXTy(*MS.C);
2122 S1_ext = IRB.CreateBitCast(S1_ext, X86_MMXTy);
2123 S2_ext = IRB.CreateBitCast(S2_ext, X86_MMXTy);
2126 Function *ShadowFn = Intrinsic::getDeclaration(
2127 F.getParent(), getSignedPackIntrinsic(I.getIntrinsicID()));
2129 Value *S = IRB.CreateCall2(ShadowFn, S1_ext, S2_ext, "_msprop_vector_pack");
2130 if (isX86_MMX) S = IRB.CreateBitCast(S, getShadowTy(&I));
2132 setOriginForNaryOp(I);
2135 // \brief Instrument sum-of-absolute-differencies intrinsic.
2136 void handleVectorSadIntrinsic(IntrinsicInst &I) {
2137 const unsigned SignificantBitsPerResultElement = 16;
2138 bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2139 Type *ResTy = isX86_MMX ? IntegerType::get(*MS.C, 64) : I.getType();
2140 unsigned ZeroBitsPerResultElement =
2141 ResTy->getScalarSizeInBits() - SignificantBitsPerResultElement;
2143 IRBuilder<> IRB(&I);
2144 Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2145 S = IRB.CreateBitCast(S, ResTy);
2146 S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2148 S = IRB.CreateLShr(S, ZeroBitsPerResultElement);
2149 S = IRB.CreateBitCast(S, getShadowTy(&I));
2151 setOriginForNaryOp(I);
2154 // \brief Instrument multiply-add intrinsic.
2155 void handleVectorPmaddIntrinsic(IntrinsicInst &I,
2156 unsigned EltSizeInBits = 0) {
2157 bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2158 Type *ResTy = isX86_MMX ? getMMXVectorTy(EltSizeInBits * 2) : I.getType();
2159 IRBuilder<> IRB(&I);
2160 Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2161 S = IRB.CreateBitCast(S, ResTy);
2162 S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2164 S = IRB.CreateBitCast(S, getShadowTy(&I));
2166 setOriginForNaryOp(I);
2169 void visitIntrinsicInst(IntrinsicInst &I) {
2170 switch (I.getIntrinsicID()) {
2171 case llvm::Intrinsic::bswap:
2174 case llvm::Intrinsic::x86_avx512_cvtsd2usi64:
2175 case llvm::Intrinsic::x86_avx512_cvtsd2usi:
2176 case llvm::Intrinsic::x86_avx512_cvtss2usi64:
2177 case llvm::Intrinsic::x86_avx512_cvtss2usi:
2178 case llvm::Intrinsic::x86_avx512_cvttss2usi64:
2179 case llvm::Intrinsic::x86_avx512_cvttss2usi:
2180 case llvm::Intrinsic::x86_avx512_cvttsd2usi64:
2181 case llvm::Intrinsic::x86_avx512_cvttsd2usi:
2182 case llvm::Intrinsic::x86_avx512_cvtusi2sd:
2183 case llvm::Intrinsic::x86_avx512_cvtusi2ss:
2184 case llvm::Intrinsic::x86_avx512_cvtusi642sd:
2185 case llvm::Intrinsic::x86_avx512_cvtusi642ss:
2186 case llvm::Intrinsic::x86_sse2_cvtsd2si64:
2187 case llvm::Intrinsic::x86_sse2_cvtsd2si:
2188 case llvm::Intrinsic::x86_sse2_cvtsd2ss:
2189 case llvm::Intrinsic::x86_sse2_cvtsi2sd:
2190 case llvm::Intrinsic::x86_sse2_cvtsi642sd:
2191 case llvm::Intrinsic::x86_sse2_cvtss2sd:
2192 case llvm::Intrinsic::x86_sse2_cvttsd2si64:
2193 case llvm::Intrinsic::x86_sse2_cvttsd2si:
2194 case llvm::Intrinsic::x86_sse_cvtsi2ss:
2195 case llvm::Intrinsic::x86_sse_cvtsi642ss:
2196 case llvm::Intrinsic::x86_sse_cvtss2si64:
2197 case llvm::Intrinsic::x86_sse_cvtss2si:
2198 case llvm::Intrinsic::x86_sse_cvttss2si64:
2199 case llvm::Intrinsic::x86_sse_cvttss2si:
2200 handleVectorConvertIntrinsic(I, 1);
2202 case llvm::Intrinsic::x86_sse2_cvtdq2pd:
2203 case llvm::Intrinsic::x86_sse2_cvtps2pd:
2204 case llvm::Intrinsic::x86_sse_cvtps2pi:
2205 case llvm::Intrinsic::x86_sse_cvttps2pi:
2206 handleVectorConvertIntrinsic(I, 2);
2208 case llvm::Intrinsic::x86_avx512_psll_dq:
2209 case llvm::Intrinsic::x86_avx512_psrl_dq:
2210 case llvm::Intrinsic::x86_avx2_psll_w:
2211 case llvm::Intrinsic::x86_avx2_psll_d:
2212 case llvm::Intrinsic::x86_avx2_psll_q:
2213 case llvm::Intrinsic::x86_avx2_pslli_w:
2214 case llvm::Intrinsic::x86_avx2_pslli_d:
2215 case llvm::Intrinsic::x86_avx2_pslli_q:
2216 case llvm::Intrinsic::x86_avx2_psll_dq:
2217 case llvm::Intrinsic::x86_avx2_psrl_w:
2218 case llvm::Intrinsic::x86_avx2_psrl_d:
2219 case llvm::Intrinsic::x86_avx2_psrl_q:
2220 case llvm::Intrinsic::x86_avx2_psra_w:
2221 case llvm::Intrinsic::x86_avx2_psra_d:
2222 case llvm::Intrinsic::x86_avx2_psrli_w:
2223 case llvm::Intrinsic::x86_avx2_psrli_d:
2224 case llvm::Intrinsic::x86_avx2_psrli_q:
2225 case llvm::Intrinsic::x86_avx2_psrai_w:
2226 case llvm::Intrinsic::x86_avx2_psrai_d:
2227 case llvm::Intrinsic::x86_avx2_psrl_dq:
2228 case llvm::Intrinsic::x86_sse2_psll_w:
2229 case llvm::Intrinsic::x86_sse2_psll_d:
2230 case llvm::Intrinsic::x86_sse2_psll_q:
2231 case llvm::Intrinsic::x86_sse2_pslli_w:
2232 case llvm::Intrinsic::x86_sse2_pslli_d:
2233 case llvm::Intrinsic::x86_sse2_pslli_q:
2234 case llvm::Intrinsic::x86_sse2_psll_dq:
2235 case llvm::Intrinsic::x86_sse2_psrl_w:
2236 case llvm::Intrinsic::x86_sse2_psrl_d:
2237 case llvm::Intrinsic::x86_sse2_psrl_q:
2238 case llvm::Intrinsic::x86_sse2_psra_w:
2239 case llvm::Intrinsic::x86_sse2_psra_d:
2240 case llvm::Intrinsic::x86_sse2_psrli_w:
2241 case llvm::Intrinsic::x86_sse2_psrli_d:
2242 case llvm::Intrinsic::x86_sse2_psrli_q:
2243 case llvm::Intrinsic::x86_sse2_psrai_w:
2244 case llvm::Intrinsic::x86_sse2_psrai_d:
2245 case llvm::Intrinsic::x86_sse2_psrl_dq:
2246 case llvm::Intrinsic::x86_mmx_psll_w:
2247 case llvm::Intrinsic::x86_mmx_psll_d:
2248 case llvm::Intrinsic::x86_mmx_psll_q:
2249 case llvm::Intrinsic::x86_mmx_pslli_w:
2250 case llvm::Intrinsic::x86_mmx_pslli_d:
2251 case llvm::Intrinsic::x86_mmx_pslli_q:
2252 case llvm::Intrinsic::x86_mmx_psrl_w:
2253 case llvm::Intrinsic::x86_mmx_psrl_d:
2254 case llvm::Intrinsic::x86_mmx_psrl_q:
2255 case llvm::Intrinsic::x86_mmx_psra_w:
2256 case llvm::Intrinsic::x86_mmx_psra_d:
2257 case llvm::Intrinsic::x86_mmx_psrli_w:
2258 case llvm::Intrinsic::x86_mmx_psrli_d:
2259 case llvm::Intrinsic::x86_mmx_psrli_q:
2260 case llvm::Intrinsic::x86_mmx_psrai_w:
2261 case llvm::Intrinsic::x86_mmx_psrai_d:
2262 handleVectorShiftIntrinsic(I, /* Variable */ false);
2264 case llvm::Intrinsic::x86_avx2_psllv_d:
2265 case llvm::Intrinsic::x86_avx2_psllv_d_256:
2266 case llvm::Intrinsic::x86_avx2_psllv_q:
2267 case llvm::Intrinsic::x86_avx2_psllv_q_256:
2268 case llvm::Intrinsic::x86_avx2_psrlv_d:
2269 case llvm::Intrinsic::x86_avx2_psrlv_d_256:
2270 case llvm::Intrinsic::x86_avx2_psrlv_q:
2271 case llvm::Intrinsic::x86_avx2_psrlv_q_256:
2272 case llvm::Intrinsic::x86_avx2_psrav_d:
2273 case llvm::Intrinsic::x86_avx2_psrav_d_256:
2274 handleVectorShiftIntrinsic(I, /* Variable */ true);
2277 // Byte shifts are not implemented.
2278 // case llvm::Intrinsic::x86_avx512_psll_dq_bs:
2279 // case llvm::Intrinsic::x86_avx512_psrl_dq_bs:
2280 // case llvm::Intrinsic::x86_avx2_psll_dq_bs:
2281 // case llvm::Intrinsic::x86_avx2_psrl_dq_bs:
2282 // case llvm::Intrinsic::x86_sse2_psll_dq_bs:
2283 // case llvm::Intrinsic::x86_sse2_psrl_dq_bs:
2285 case llvm::Intrinsic::x86_sse2_packsswb_128:
2286 case llvm::Intrinsic::x86_sse2_packssdw_128:
2287 case llvm::Intrinsic::x86_sse2_packuswb_128:
2288 case llvm::Intrinsic::x86_sse41_packusdw:
2289 case llvm::Intrinsic::x86_avx2_packsswb:
2290 case llvm::Intrinsic::x86_avx2_packssdw:
2291 case llvm::Intrinsic::x86_avx2_packuswb:
2292 case llvm::Intrinsic::x86_avx2_packusdw:
2293 handleVectorPackIntrinsic(I);
2296 case llvm::Intrinsic::x86_mmx_packsswb:
2297 case llvm::Intrinsic::x86_mmx_packuswb:
2298 handleVectorPackIntrinsic(I, 16);
2301 case llvm::Intrinsic::x86_mmx_packssdw:
2302 handleVectorPackIntrinsic(I, 32);
2305 case llvm::Intrinsic::x86_mmx_psad_bw:
2306 case llvm::Intrinsic::x86_sse2_psad_bw:
2307 case llvm::Intrinsic::x86_avx2_psad_bw:
2308 handleVectorSadIntrinsic(I);
2311 case llvm::Intrinsic::x86_sse2_pmadd_wd:
2312 case llvm::Intrinsic::x86_avx2_pmadd_wd:
2313 case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw_128:
2314 case llvm::Intrinsic::x86_avx2_pmadd_ub_sw:
2315 handleVectorPmaddIntrinsic(I);
2318 case llvm::Intrinsic::x86_ssse3_pmadd_ub_sw:
2319 handleVectorPmaddIntrinsic(I, 8);
2322 case llvm::Intrinsic::x86_mmx_pmadd_wd:
2323 handleVectorPmaddIntrinsic(I, 16);
2327 if (!handleUnknownIntrinsic(I))
2328 visitInstruction(I);
2333 void visitCallSite(CallSite CS) {
2334 Instruction &I = *CS.getInstruction();
2335 assert((CS.isCall() || CS.isInvoke()) && "Unknown type of CallSite");
2337 CallInst *Call = cast<CallInst>(&I);
2339 // For inline asm, do the usual thing: check argument shadow and mark all
2340 // outputs as clean. Note that any side effects of the inline asm that are
2341 // not immediately visible in its constraints are not handled.
2342 if (Call->isInlineAsm()) {
2343 visitInstruction(I);
2347 assert(!isa<IntrinsicInst>(&I) && "intrinsics are handled elsewhere");
2349 // We are going to insert code that relies on the fact that the callee
2350 // will become a non-readonly function after it is instrumented by us. To
2351 // prevent this code from being optimized out, mark that function
2352 // non-readonly in advance.
2353 if (Function *Func = Call->getCalledFunction()) {
2354 // Clear out readonly/readnone attributes.
2356 B.addAttribute(Attribute::ReadOnly)
2357 .addAttribute(Attribute::ReadNone);
2358 Func->removeAttributes(AttributeSet::FunctionIndex,
2359 AttributeSet::get(Func->getContext(),
2360 AttributeSet::FunctionIndex,
2364 IRBuilder<> IRB(&I);
2366 unsigned ArgOffset = 0;
2367 DEBUG(dbgs() << " CallSite: " << I << "\n");
2368 for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
2369 ArgIt != End; ++ArgIt) {
2371 unsigned i = ArgIt - CS.arg_begin();
2372 if (!A->getType()->isSized()) {
2373 DEBUG(dbgs() << "Arg " << i << " is not sized: " << I << "\n");
2377 Value *Store = nullptr;
2378 // Compute the Shadow for arg even if it is ByVal, because
2379 // in that case getShadow() will copy the actual arg shadow to
2380 // __msan_param_tls.
2381 Value *ArgShadow = getShadow(A);
2382 Value *ArgShadowBase = getShadowPtrForArgument(A, IRB, ArgOffset);
2383 DEBUG(dbgs() << " Arg#" << i << ": " << *A <<
2384 " Shadow: " << *ArgShadow << "\n");
2385 bool ArgIsInitialized = false;
2386 if (CS.paramHasAttr(i + 1, Attribute::ByVal)) {
2387 assert(A->getType()->isPointerTy() &&
2388 "ByVal argument is not a pointer!");
2389 Size = MS.DL->getTypeAllocSize(A->getType()->getPointerElementType());
2390 if (ArgOffset + Size > kParamTLSSize) break;
2391 unsigned ParamAlignment = CS.getParamAlignment(i + 1);
2392 unsigned Alignment = std::min(ParamAlignment, kShadowTLSAlignment);
2393 Store = IRB.CreateMemCpy(ArgShadowBase,
2394 getShadowPtr(A, Type::getInt8Ty(*MS.C), IRB),
2397 Size = MS.DL->getTypeAllocSize(A->getType());
2398 if (ArgOffset + Size > kParamTLSSize) break;
2399 Store = IRB.CreateAlignedStore(ArgShadow, ArgShadowBase,
2400 kShadowTLSAlignment);
2401 Constant *Cst = dyn_cast<Constant>(ArgShadow);
2402 if (Cst && Cst->isNullValue()) ArgIsInitialized = true;
2404 if (MS.TrackOrigins && !ArgIsInitialized)
2405 IRB.CreateStore(getOrigin(A),
2406 getOriginPtrForArgument(A, IRB, ArgOffset));
2408 assert(Size != 0 && Store != nullptr);
2409 DEBUG(dbgs() << " Param:" << *Store << "\n");
2410 ArgOffset += RoundUpToAlignment(Size, 8);
2412 DEBUG(dbgs() << " done with call args\n");
2415 cast<FunctionType>(CS.getCalledValue()->getType()->getContainedType(0));
2416 if (FT->isVarArg()) {
2417 VAHelper->visitCallSite(CS, IRB);
2420 // Now, get the shadow for the RetVal.
2421 if (!I.getType()->isSized()) return;
2422 IRBuilder<> IRBBefore(&I);
2423 // Until we have full dynamic coverage, make sure the retval shadow is 0.
2424 Value *Base = getShadowPtrForRetval(&I, IRBBefore);
2425 IRBBefore.CreateAlignedStore(getCleanShadow(&I), Base, kShadowTLSAlignment);
2426 Instruction *NextInsn = nullptr;
2428 NextInsn = I.getNextNode();
2430 BasicBlock *NormalDest = cast<InvokeInst>(&I)->getNormalDest();
2431 if (!NormalDest->getSinglePredecessor()) {
2432 // FIXME: this case is tricky, so we are just conservative here.
2433 // Perhaps we need to split the edge between this BB and NormalDest,
2434 // but a naive attempt to use SplitEdge leads to a crash.
2435 setShadow(&I, getCleanShadow(&I));
2436 setOrigin(&I, getCleanOrigin());
2439 NextInsn = NormalDest->getFirstInsertionPt();
2441 "Could not find insertion point for retval shadow load");
2443 IRBuilder<> IRBAfter(NextInsn);
2444 Value *RetvalShadow =
2445 IRBAfter.CreateAlignedLoad(getShadowPtrForRetval(&I, IRBAfter),
2446 kShadowTLSAlignment, "_msret");
2447 setShadow(&I, RetvalShadow);
2448 if (MS.TrackOrigins)
2449 setOrigin(&I, IRBAfter.CreateLoad(getOriginPtrForRetval(IRBAfter)));
2452 void visitReturnInst(ReturnInst &I) {
2453 IRBuilder<> IRB(&I);
2454 Value *RetVal = I.getReturnValue();
2455 if (!RetVal) return;
2456 Value *ShadowPtr = getShadowPtrForRetval(RetVal, IRB);
2457 if (CheckReturnValue) {
2458 insertShadowCheck(RetVal, &I);
2459 Value *Shadow = getCleanShadow(RetVal);
2460 IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
2462 Value *Shadow = getShadow(RetVal);
2463 IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
2464 // FIXME: make it conditional if ClStoreCleanOrigin==0
2465 if (MS.TrackOrigins)
2466 IRB.CreateStore(getOrigin(RetVal), getOriginPtrForRetval(IRB));
2470 void visitPHINode(PHINode &I) {
2471 IRBuilder<> IRB(&I);
2472 if (!PropagateShadow) {
2473 setShadow(&I, getCleanShadow(&I));
2474 setOrigin(&I, getCleanOrigin());
2478 ShadowPHINodes.push_back(&I);
2479 setShadow(&I, IRB.CreatePHI(getShadowTy(&I), I.getNumIncomingValues(),
2481 if (MS.TrackOrigins)
2482 setOrigin(&I, IRB.CreatePHI(MS.OriginTy, I.getNumIncomingValues(),
2486 void visitAllocaInst(AllocaInst &I) {
2487 setShadow(&I, getCleanShadow(&I));
2488 setOrigin(&I, getCleanOrigin());
2489 IRBuilder<> IRB(I.getNextNode());
2490 uint64_t Size = MS.DL->getTypeAllocSize(I.getAllocatedType());
2491 if (PoisonStack && ClPoisonStackWithCall) {
2492 IRB.CreateCall2(MS.MsanPoisonStackFn,
2493 IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()),
2494 ConstantInt::get(MS.IntptrTy, Size));
2496 Value *ShadowBase = getShadowPtr(&I, Type::getInt8PtrTy(*MS.C), IRB);
2497 Value *PoisonValue = IRB.getInt8(PoisonStack ? ClPoisonStackPattern : 0);
2498 IRB.CreateMemSet(ShadowBase, PoisonValue, Size, I.getAlignment());
2501 if (PoisonStack && MS.TrackOrigins) {
2502 SmallString<2048> StackDescriptionStorage;
2503 raw_svector_ostream StackDescription(StackDescriptionStorage);
2504 // We create a string with a description of the stack allocation and
2505 // pass it into __msan_set_alloca_origin.
2506 // It will be printed by the run-time if stack-originated UMR is found.
2507 // The first 4 bytes of the string are set to '----' and will be replaced
2508 // by __msan_va_arg_overflow_size_tls at the first call.
2509 StackDescription << "----" << I.getName() << "@" << F.getName();
2511 createPrivateNonConstGlobalForString(*F.getParent(),
2512 StackDescription.str());
2514 IRB.CreateCall4(MS.MsanSetAllocaOrigin4Fn,
2515 IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()),
2516 ConstantInt::get(MS.IntptrTy, Size),
2517 IRB.CreatePointerCast(Descr, IRB.getInt8PtrTy()),
2518 IRB.CreatePointerCast(&F, MS.IntptrTy));
2522 void visitSelectInst(SelectInst& I) {
2523 IRBuilder<> IRB(&I);
2524 // a = select b, c, d
2525 Value *B = I.getCondition();
2526 Value *C = I.getTrueValue();
2527 Value *D = I.getFalseValue();
2528 Value *Sb = getShadow(B);
2529 Value *Sc = getShadow(C);
2530 Value *Sd = getShadow(D);
2532 // Result shadow if condition shadow is 0.
2533 Value *Sa0 = IRB.CreateSelect(B, Sc, Sd);
2535 if (I.getType()->isAggregateType()) {
2536 // To avoid "sign extending" i1 to an arbitrary aggregate type, we just do
2537 // an extra "select". This results in much more compact IR.
2538 // Sa = select Sb, poisoned, (select b, Sc, Sd)
2539 Sa1 = getPoisonedShadow(getShadowTy(I.getType()));
2541 // Sa = select Sb, [ (c^d) | Sc | Sd ], [ b ? Sc : Sd ]
2542 // If Sb (condition is poisoned), look for bits in c and d that are equal
2543 // and both unpoisoned.
2544 // If !Sb (condition is unpoisoned), simply pick one of Sc and Sd.
2546 // Cast arguments to shadow-compatible type.
2547 C = CreateAppToShadowCast(IRB, C);
2548 D = CreateAppToShadowCast(IRB, D);
2550 // Result shadow if condition shadow is 1.
2551 Sa1 = IRB.CreateOr(IRB.CreateXor(C, D), IRB.CreateOr(Sc, Sd));
2553 Value *Sa = IRB.CreateSelect(Sb, Sa1, Sa0, "_msprop_select");
2555 if (MS.TrackOrigins) {
2556 // Origins are always i32, so any vector conditions must be flattened.
2557 // FIXME: consider tracking vector origins for app vectors?
2558 if (B->getType()->isVectorTy()) {
2559 Type *FlatTy = getShadowTyNoVec(B->getType());
2560 B = IRB.CreateICmpNE(IRB.CreateBitCast(B, FlatTy),
2561 ConstantInt::getNullValue(FlatTy));
2562 Sb = IRB.CreateICmpNE(IRB.CreateBitCast(Sb, FlatTy),
2563 ConstantInt::getNullValue(FlatTy));
2565 // a = select b, c, d
2566 // Oa = Sb ? Ob : (b ? Oc : Od)
2568 &I, IRB.CreateSelect(Sb, getOrigin(I.getCondition()),
2569 IRB.CreateSelect(B, getOrigin(I.getTrueValue()),
2570 getOrigin(I.getFalseValue()))));
2574 void visitLandingPadInst(LandingPadInst &I) {
2576 // See http://code.google.com/p/memory-sanitizer/issues/detail?id=1
2577 setShadow(&I, getCleanShadow(&I));
2578 setOrigin(&I, getCleanOrigin());
2581 void visitGetElementPtrInst(GetElementPtrInst &I) {
2585 void visitExtractValueInst(ExtractValueInst &I) {
2586 IRBuilder<> IRB(&I);
2587 Value *Agg = I.getAggregateOperand();
2588 DEBUG(dbgs() << "ExtractValue: " << I << "\n");
2589 Value *AggShadow = getShadow(Agg);
2590 DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
2591 Value *ResShadow = IRB.CreateExtractValue(AggShadow, I.getIndices());
2592 DEBUG(dbgs() << " ResShadow: " << *ResShadow << "\n");
2593 setShadow(&I, ResShadow);
2594 setOriginForNaryOp(I);
2597 void visitInsertValueInst(InsertValueInst &I) {
2598 IRBuilder<> IRB(&I);
2599 DEBUG(dbgs() << "InsertValue: " << I << "\n");
2600 Value *AggShadow = getShadow(I.getAggregateOperand());
2601 Value *InsShadow = getShadow(I.getInsertedValueOperand());
2602 DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
2603 DEBUG(dbgs() << " InsShadow: " << *InsShadow << "\n");
2604 Value *Res = IRB.CreateInsertValue(AggShadow, InsShadow, I.getIndices());
2605 DEBUG(dbgs() << " Res: " << *Res << "\n");
2607 setOriginForNaryOp(I);
2610 void dumpInst(Instruction &I) {
2611 if (CallInst *CI = dyn_cast<CallInst>(&I)) {
2612 errs() << "ZZZ call " << CI->getCalledFunction()->getName() << "\n";
2614 errs() << "ZZZ " << I.getOpcodeName() << "\n";
2616 errs() << "QQQ " << I << "\n";
2619 void visitResumeInst(ResumeInst &I) {
2620 DEBUG(dbgs() << "Resume: " << I << "\n");
2621 // Nothing to do here.
2624 void visitInstruction(Instruction &I) {
2625 // Everything else: stop propagating and check for poisoned shadow.
2626 if (ClDumpStrictInstructions)
2628 DEBUG(dbgs() << "DEFAULT: " << I << "\n");
2629 for (size_t i = 0, n = I.getNumOperands(); i < n; i++)
2630 insertShadowCheck(I.getOperand(i), &I);
2631 setShadow(&I, getCleanShadow(&I));
2632 setOrigin(&I, getCleanOrigin());
2636 /// \brief AMD64-specific implementation of VarArgHelper.
2637 struct VarArgAMD64Helper : public VarArgHelper {
2638 // An unfortunate workaround for asymmetric lowering of va_arg stuff.
2639 // See a comment in visitCallSite for more details.
2640 static const unsigned AMD64GpEndOffset = 48; // AMD64 ABI Draft 0.99.6 p3.5.7
2641 static const unsigned AMD64FpEndOffset = 176;
2644 MemorySanitizer &MS;
2645 MemorySanitizerVisitor &MSV;
2646 Value *VAArgTLSCopy;
2647 Value *VAArgOverflowSize;
2649 SmallVector<CallInst*, 16> VAStartInstrumentationList;
2651 VarArgAMD64Helper(Function &F, MemorySanitizer &MS,
2652 MemorySanitizerVisitor &MSV)
2653 : F(F), MS(MS), MSV(MSV), VAArgTLSCopy(nullptr),
2654 VAArgOverflowSize(nullptr) {}
2656 enum ArgKind { AK_GeneralPurpose, AK_FloatingPoint, AK_Memory };
2658 ArgKind classifyArgument(Value* arg) {
2659 // A very rough approximation of X86_64 argument classification rules.
2660 Type *T = arg->getType();
2661 if (T->isFPOrFPVectorTy() || T->isX86_MMXTy())
2662 return AK_FloatingPoint;
2663 if (T->isIntegerTy() && T->getPrimitiveSizeInBits() <= 64)
2664 return AK_GeneralPurpose;
2665 if (T->isPointerTy())
2666 return AK_GeneralPurpose;
2670 // For VarArg functions, store the argument shadow in an ABI-specific format
2671 // that corresponds to va_list layout.
2672 // We do this because Clang lowers va_arg in the frontend, and this pass
2673 // only sees the low level code that deals with va_list internals.
2674 // A much easier alternative (provided that Clang emits va_arg instructions)
2675 // would have been to associate each live instance of va_list with a copy of
2676 // MSanParamTLS, and extract shadow on va_arg() call in the argument list
2678 void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
2679 unsigned GpOffset = 0;
2680 unsigned FpOffset = AMD64GpEndOffset;
2681 unsigned OverflowOffset = AMD64FpEndOffset;
2682 for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
2683 ArgIt != End; ++ArgIt) {
2685 unsigned ArgNo = CS.getArgumentNo(ArgIt);
2686 bool IsByVal = CS.paramHasAttr(ArgNo + 1, Attribute::ByVal);
2688 // ByVal arguments always go to the overflow area.
2689 assert(A->getType()->isPointerTy());
2690 Type *RealTy = A->getType()->getPointerElementType();
2691 uint64_t ArgSize = MS.DL->getTypeAllocSize(RealTy);
2692 Value *Base = getShadowPtrForVAArgument(RealTy, IRB, OverflowOffset);
2693 OverflowOffset += RoundUpToAlignment(ArgSize, 8);
2694 IRB.CreateMemCpy(Base, MSV.getShadowPtr(A, IRB.getInt8Ty(), IRB),
2695 ArgSize, kShadowTLSAlignment);
2697 ArgKind AK = classifyArgument(A);
2698 if (AK == AK_GeneralPurpose && GpOffset >= AMD64GpEndOffset)
2700 if (AK == AK_FloatingPoint && FpOffset >= AMD64FpEndOffset)
2704 case AK_GeneralPurpose:
2705 Base = getShadowPtrForVAArgument(A->getType(), IRB, GpOffset);
2708 case AK_FloatingPoint:
2709 Base = getShadowPtrForVAArgument(A->getType(), IRB, FpOffset);
2713 uint64_t ArgSize = MS.DL->getTypeAllocSize(A->getType());
2714 Base = getShadowPtrForVAArgument(A->getType(), IRB, OverflowOffset);
2715 OverflowOffset += RoundUpToAlignment(ArgSize, 8);
2717 IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
2720 Constant *OverflowSize =
2721 ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AMD64FpEndOffset);
2722 IRB.CreateStore(OverflowSize, MS.VAArgOverflowSizeTLS);
2725 /// \brief Compute the shadow address for a given va_arg.
2726 Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
2728 Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
2729 Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
2730 return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
2734 void visitVAStartInst(VAStartInst &I) override {
2735 IRBuilder<> IRB(&I);
2736 VAStartInstrumentationList.push_back(&I);
2737 Value *VAListTag = I.getArgOperand(0);
2738 Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);
2740 // Unpoison the whole __va_list_tag.
2741 // FIXME: magic ABI constants.
2742 IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
2743 /* size */24, /* alignment */8, false);
2746 void visitVACopyInst(VACopyInst &I) override {
2747 IRBuilder<> IRB(&I);
2748 Value *VAListTag = I.getArgOperand(0);
2749 Value *ShadowPtr = MSV.getShadowPtr(VAListTag, IRB.getInt8Ty(), IRB);
2751 // Unpoison the whole __va_list_tag.
2752 // FIXME: magic ABI constants.
2753 IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
2754 /* size */24, /* alignment */8, false);
2757 void finalizeInstrumentation() override {
2758 assert(!VAArgOverflowSize && !VAArgTLSCopy &&
2759 "finalizeInstrumentation called twice");
2760 if (!VAStartInstrumentationList.empty()) {
2761 // If there is a va_start in this function, make a backup copy of
2762 // va_arg_tls somewhere in the function entry block.
2763 IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI());
2764 VAArgOverflowSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
2766 IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, AMD64FpEndOffset),
2768 VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
2769 IRB.CreateMemCpy(VAArgTLSCopy, MS.VAArgTLS, CopySize, 8);
2772 // Instrument va_start.
2773 // Copy va_list shadow from the backup copy of the TLS contents.
2774 for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
2775 CallInst *OrigInst = VAStartInstrumentationList[i];
2776 IRBuilder<> IRB(OrigInst->getNextNode());
2777 Value *VAListTag = OrigInst->getArgOperand(0);
2779 Value *RegSaveAreaPtrPtr =
2781 IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
2782 ConstantInt::get(MS.IntptrTy, 16)),
2783 Type::getInt64PtrTy(*MS.C));
2784 Value *RegSaveAreaPtr = IRB.CreateLoad(RegSaveAreaPtrPtr);
2785 Value *RegSaveAreaShadowPtr =
2786 MSV.getShadowPtr(RegSaveAreaPtr, IRB.getInt8Ty(), IRB);
2787 IRB.CreateMemCpy(RegSaveAreaShadowPtr, VAArgTLSCopy,
2788 AMD64FpEndOffset, 16);
2790 Value *OverflowArgAreaPtrPtr =
2792 IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
2793 ConstantInt::get(MS.IntptrTy, 8)),
2794 Type::getInt64PtrTy(*MS.C));
2795 Value *OverflowArgAreaPtr = IRB.CreateLoad(OverflowArgAreaPtrPtr);
2796 Value *OverflowArgAreaShadowPtr =
2797 MSV.getShadowPtr(OverflowArgAreaPtr, IRB.getInt8Ty(), IRB);
2798 Value *SrcPtr = IRB.CreateConstGEP1_32(VAArgTLSCopy, AMD64FpEndOffset);
2799 IRB.CreateMemCpy(OverflowArgAreaShadowPtr, SrcPtr, VAArgOverflowSize, 16);
2804 /// \brief A no-op implementation of VarArgHelper.
2805 struct VarArgNoOpHelper : public VarArgHelper {
2806 VarArgNoOpHelper(Function &F, MemorySanitizer &MS,
2807 MemorySanitizerVisitor &MSV) {}
2809 void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {}
2811 void visitVAStartInst(VAStartInst &I) override {}
2813 void visitVACopyInst(VACopyInst &I) override {}
2815 void finalizeInstrumentation() override {}
2818 VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
2819 MemorySanitizerVisitor &Visitor) {
2820 // VarArg handling is only implemented on AMD64. False positives are possible
2821 // on other platforms.
2822 llvm::Triple TargetTriple(Func.getParent()->getTargetTriple());
2823 if (TargetTriple.getArch() == llvm::Triple::x86_64)
2824 return new VarArgAMD64Helper(Func, Msan, Visitor);
2826 return new VarArgNoOpHelper(Func, Msan, Visitor);
2831 bool MemorySanitizer::runOnFunction(Function &F) {
2832 MemorySanitizerVisitor Visitor(F, *this);
2834 // Clear out readonly/readnone attributes.
2836 B.addAttribute(Attribute::ReadOnly)
2837 .addAttribute(Attribute::ReadNone);
2838 F.removeAttributes(AttributeSet::FunctionIndex,
2839 AttributeSet::get(F.getContext(),
2840 AttributeSet::FunctionIndex, B));
2842 return Visitor.runOnFunction();