LLVM  8.0.0svn
MemorySanitizer.cpp
Go to the documentation of this file.
1 //===- MemorySanitizer.cpp - detector of uninitialized reads --------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 /// \file
11 /// This file is a part of MemorySanitizer, a detector of uninitialized
12 /// reads.
13 ///
14 /// The algorithm of the tool is similar to Memcheck
15 /// (http://goo.gl/QKbem). We associate a few shadow bits with every
16 /// byte of the application memory, poison the shadow of the malloc-ed
17 /// or alloca-ed memory, load the shadow bits on every memory read,
18 /// propagate the shadow bits through some of the arithmetic
19 /// instruction (including MOV), store the shadow bits on every memory
20 /// write, report a bug on some other instructions (e.g. JMP) if the
21 /// associated shadow is poisoned.
22 ///
23 /// But there are differences too. The first and the major one:
24 /// compiler instrumentation instead of binary instrumentation. This
25 /// gives us much better register allocation, possible compiler
26 /// optimizations and a fast start-up. But this brings the major issue
27 /// as well: msan needs to see all program events, including system
28 /// calls and reads/writes in system libraries, so we either need to
29 /// compile *everything* with msan or use a binary translation
30 /// component (e.g. DynamoRIO) to instrument pre-built libraries.
31 /// Another difference from Memcheck is that we use 8 shadow bits per
32 /// byte of application memory and use a direct shadow mapping. This
33 /// greatly simplifies the instrumentation code and avoids races on
34 /// shadow updates (Memcheck is single-threaded so races are not a
35 /// concern there. Memcheck uses 2 shadow bits per byte with a slow
36 /// path storage that uses 8 bits per byte).
37 ///
38 /// The default value of shadow is 0, which means "clean" (not poisoned).
39 ///
40 /// Every module initializer should call __msan_init to ensure that the
41 /// shadow memory is ready. On error, __msan_warning is called. Since
42 /// parameters and return values may be passed via registers, we have a
43 /// specialized thread-local shadow for return values
44 /// (__msan_retval_tls) and parameters (__msan_param_tls).
45 ///
46 /// Origin tracking.
47 ///
48 /// MemorySanitizer can track origins (allocation points) of all uninitialized
49 /// values. This behavior is controlled with a flag (msan-track-origins) and is
50 /// disabled by default.
51 ///
52 /// Origins are 4-byte values created and interpreted by the runtime library.
53 /// They are stored in a second shadow mapping, one 4-byte value for 4 bytes
54 /// of application memory. Propagation of origins is basically a bunch of
55 /// "select" instructions that pick the origin of a dirty argument, if an
56 /// instruction has one.
57 ///
58 /// Every 4 aligned, consecutive bytes of application memory have one origin
59 /// value associated with them. If these bytes contain uninitialized data
60 /// coming from 2 different allocations, the last store wins. Because of this,
61 /// MemorySanitizer reports can show unrelated origins, but this is unlikely in
62 /// practice.
63 ///
64 /// Origins are meaningless for fully initialized values, so MemorySanitizer
65 /// avoids storing origin to memory when a fully initialized value is stored.
66 /// This way it avoids needless overwritting origin of the 4-byte region on
67 /// a short (i.e. 1 byte) clean store, and it is also good for performance.
68 ///
69 /// Atomic handling.
70 ///
71 /// Ideally, every atomic store of application value should update the
72 /// corresponding shadow location in an atomic way. Unfortunately, atomic store
73 /// of two disjoint locations can not be done without severe slowdown.
74 ///
75 /// Therefore, we implement an approximation that may err on the safe side.
76 /// In this implementation, every atomically accessed location in the program
77 /// may only change from (partially) uninitialized to fully initialized, but
78 /// not the other way around. We load the shadow _after_ the application load,
79 /// and we store the shadow _before_ the app store. Also, we always store clean
80 /// shadow (if the application store is atomic). This way, if the store-load
81 /// pair constitutes a happens-before arc, shadow store and load are correctly
82 /// ordered such that the load will get either the value that was stored, or
83 /// some later value (which is always clean).
84 ///
85 /// This does not work very well with Compare-And-Swap (CAS) and
86 /// Read-Modify-Write (RMW) operations. To follow the above logic, CAS and RMW
87 /// must store the new shadow before the app operation, and load the shadow
88 /// after the app operation. Computers don't work this way. Current
89 /// implementation ignores the load aspect of CAS/RMW, always returning a clean
90 /// value. It implements the store part as a simple atomic store by storing a
91 /// clean shadow.
92 ///
93 /// Instrumenting inline assembly.
94 ///
95 /// For inline assembly code LLVM has little idea about which memory locations
96 /// become initialized depending on the arguments. It can be possible to figure
97 /// out which arguments are meant to point to inputs and outputs, but the
98 /// actual semantics can be only visible at runtime. In the Linux kernel it's
99 /// also possible that the arguments only indicate the offset for a base taken
100 /// from a segment register, so it's dangerous to treat any asm() arguments as
101 /// pointers. We take a conservative approach generating calls to
102 /// __msan_instrument_asm_load(ptr, size) and
103 /// __msan_instrument_asm_store(ptr, size)
104 /// , which defer the memory checking/unpoisoning to the runtime library.
105 /// The latter can perform more complex address checks to figure out whether
106 /// it's safe to touch the shadow memory.
107 /// Like with atomic operations, we call __msan_instrument_asm_store() before
108 /// the assembly call, so that changes to the shadow memory will be seen by
109 /// other threads together with main memory initialization.
110 ///
111 /// KernelMemorySanitizer (KMSAN) implementation.
112 ///
113 /// The major differences between KMSAN and MSan instrumentation are:
114 /// - KMSAN always tracks the origins and implies msan-keep-going=true;
115 /// - KMSAN allocates shadow and origin memory for each page separately, so
116 /// there are no explicit accesses to shadow and origin in the
117 /// instrumentation.
118 /// Shadow and origin values for a particular X-byte memory location
119 /// (X=1,2,4,8) are accessed through pointers obtained via the
120 /// __msan_metadata_ptr_for_load_X(ptr)
121 /// __msan_metadata_ptr_for_store_X(ptr)
122 /// functions. The corresponding functions check that the X-byte accesses
123 /// are possible and returns the pointers to shadow and origin memory.
124 /// Arbitrary sized accesses are handled with:
125 /// __msan_metadata_ptr_for_load_n(ptr, size)
126 /// __msan_metadata_ptr_for_store_n(ptr, size);
127 /// - TLS variables are stored in a single per-task struct. A call to a
128 /// function __msan_get_context_state() returning a pointer to that struct
129 /// is inserted into every instrumented function before the entry block;
130 /// - __msan_warning() takes a 32-bit origin parameter;
131 /// - local variables are poisoned with __msan_poison_alloca() upon function
132 /// entry and unpoisoned with __msan_unpoison_alloca() before leaving the
133 /// function;
134 /// - the pass doesn't declare any global variables or add global constructors
135 /// to the translation unit.
136 ///
137 /// Also, KMSAN currently ignores uninitialized memory passed into inline asm
138 /// calls, making sure we're on the safe side wrt. possible false positives.
139 ///
140 /// KernelMemorySanitizer only supports X86_64 at the moment.
141 ///
142 //===----------------------------------------------------------------------===//
143 
144 #include "llvm/ADT/APInt.h"
145 #include "llvm/ADT/ArrayRef.h"
147 #include "llvm/ADT/SmallString.h"
148 #include "llvm/ADT/SmallVector.h"
149 #include "llvm/ADT/StringExtras.h"
150 #include "llvm/ADT/StringRef.h"
151 #include "llvm/ADT/Triple.h"
154 #include "llvm/IR/Argument.h"
155 #include "llvm/IR/Attributes.h"
156 #include "llvm/IR/BasicBlock.h"
157 #include "llvm/IR/CallSite.h"
158 #include "llvm/IR/CallingConv.h"
159 #include "llvm/IR/Constant.h"
160 #include "llvm/IR/Constants.h"
161 #include "llvm/IR/DataLayout.h"
162 #include "llvm/IR/DerivedTypes.h"
163 #include "llvm/IR/Function.h"
164 #include "llvm/IR/GlobalValue.h"
165 #include "llvm/IR/GlobalVariable.h"
166 #include "llvm/IR/IRBuilder.h"
167 #include "llvm/IR/InlineAsm.h"
168 #include "llvm/IR/InstVisitor.h"
169 #include "llvm/IR/InstrTypes.h"
170 #include "llvm/IR/Instruction.h"
171 #include "llvm/IR/Instructions.h"
172 #include "llvm/IR/IntrinsicInst.h"
173 #include "llvm/IR/Intrinsics.h"
174 #include "llvm/IR/LLVMContext.h"
175 #include "llvm/IR/MDBuilder.h"
176 #include "llvm/IR/Module.h"
177 #include "llvm/IR/Type.h"
178 #include "llvm/IR/Value.h"
179 #include "llvm/IR/ValueMap.h"
180 #include "llvm/Pass.h"
182 #include "llvm/Support/Casting.h"
184 #include "llvm/Support/Compiler.h"
185 #include "llvm/Support/Debug.h"
187 #include "llvm/Support/MathExtras.h"
192 #include <algorithm>
193 #include <cassert>
194 #include <cstddef>
195 #include <cstdint>
196 #include <memory>
197 #include <string>
198 #include <tuple>
199 
200 using namespace llvm;
201 
202 #define DEBUG_TYPE "msan"
203 
204 static const unsigned kOriginSize = 4;
205 static const unsigned kMinOriginAlignment = 4;
206 static const unsigned kShadowTLSAlignment = 8;
207 
208 // These constants must be kept in sync with the ones in msan.h.
209 static const unsigned kParamTLSSize = 800;
210 static const unsigned kRetvalTLSSize = 800;
211 
212 // Accesses sizes are powers of two: 1, 2, 4, 8.
213 static const size_t kNumberOfAccessSizes = 4;
214 
215 /// Track origins of uninitialized values.
216 ///
217 /// Adds a section to MemorySanitizer report that points to the allocation
218 /// (stack or heap) the uninitialized bits came from originally.
219 static cl::opt<int> ClTrackOrigins("msan-track-origins",
220  cl::desc("Track origins (allocation sites) of poisoned memory"),
221  cl::Hidden, cl::init(0));
222 
223 static cl::opt<bool> ClKeepGoing("msan-keep-going",
224  cl::desc("keep going after reporting a UMR"),
225  cl::Hidden, cl::init(false));
226 
227 static cl::opt<bool> ClPoisonStack("msan-poison-stack",
228  cl::desc("poison uninitialized stack variables"),
229  cl::Hidden, cl::init(true));
230 
231 static cl::opt<bool> ClPoisonStackWithCall("msan-poison-stack-with-call",
232  cl::desc("poison uninitialized stack variables with a call"),
233  cl::Hidden, cl::init(false));
234 
235 static cl::opt<int> ClPoisonStackPattern("msan-poison-stack-pattern",
236  cl::desc("poison uninitialized stack variables with the given pattern"),
237  cl::Hidden, cl::init(0xff));
238 
239 static cl::opt<bool> ClPoisonUndef("msan-poison-undef",
240  cl::desc("poison undef temps"),
241  cl::Hidden, cl::init(true));
242 
243 static cl::opt<bool> ClHandleICmp("msan-handle-icmp",
244  cl::desc("propagate shadow through ICmpEQ and ICmpNE"),
245  cl::Hidden, cl::init(true));
246 
247 static cl::opt<bool> ClHandleICmpExact("msan-handle-icmp-exact",
248  cl::desc("exact handling of relational integer ICmp"),
249  cl::Hidden, cl::init(false));
250 
251 // When compiling the Linux kernel, we sometimes see false positives related to
252 // MSan being unable to understand that inline assembly calls may initialize
253 // local variables.
254 // This flag makes the compiler conservatively unpoison every memory location
255 // passed into an assembly call. Note that this may cause false positives.
256 // Because it's impossible to figure out the array sizes, we can only unpoison
257 // the first sizeof(type) bytes for each type* pointer.
259  "msan-handle-asm-conservative",
260  cl::desc("conservative handling of inline assembly"), cl::Hidden,
261  cl::init(false));
262 
263 // This flag controls whether we check the shadow of the address
264 // operand of load or store. Such bugs are very rare, since load from
265 // a garbage address typically results in SEGV, but still happen
266 // (e.g. only lower bits of address are garbage, or the access happens
267 // early at program startup where malloc-ed memory is more likely to
268 // be zeroed. As of 2012-08-28 this flag adds 20% slowdown.
269 static cl::opt<bool> ClCheckAccessAddress("msan-check-access-address",
270  cl::desc("report accesses through a pointer which has poisoned shadow"),
271  cl::Hidden, cl::init(true));
272 
273 static cl::opt<bool> ClDumpStrictInstructions("msan-dump-strict-instructions",
274  cl::desc("print out instructions with default strict semantics"),
275  cl::Hidden, cl::init(false));
276 
278  "msan-instrumentation-with-call-threshold",
279  cl::desc(
280  "If the function being instrumented requires more than "
281  "this number of checks and origin stores, use callbacks instead of "
282  "inline checks (-1 means never use callbacks)."),
283  cl::Hidden, cl::init(3500));
284 
285 static cl::opt<bool>
286  ClEnableKmsan("msan-kernel",
287  cl::desc("Enable KernelMemorySanitizer instrumentation"),
288  cl::Hidden, cl::init(false));
289 
290 // This is an experiment to enable handling of cases where shadow is a non-zero
291 // compile-time constant. For some unexplainable reason they were silently
292 // ignored in the instrumentation.
293 static cl::opt<bool> ClCheckConstantShadow("msan-check-constant-shadow",
294  cl::desc("Insert checks for constant shadow values"),
295  cl::Hidden, cl::init(false));
296 
297 // This is off by default because of a bug in gold:
298 // https://sourceware.org/bugzilla/show_bug.cgi?id=19002
299 static cl::opt<bool> ClWithComdat("msan-with-comdat",
300  cl::desc("Place MSan constructors in comdat sections"),
301  cl::Hidden, cl::init(false));
302 
303 // These options allow to specify custom memory map parameters
304 // See MemoryMapParams for details.
305 static cl::opt<unsigned long long> ClAndMask("msan-and-mask",
306  cl::desc("Define custom MSan AndMask"),
307  cl::Hidden, cl::init(0));
308 
309 static cl::opt<unsigned long long> ClXorMask("msan-xor-mask",
310  cl::desc("Define custom MSan XorMask"),
311  cl::Hidden, cl::init(0));
312 
313 static cl::opt<unsigned long long> ClShadowBase("msan-shadow-base",
314  cl::desc("Define custom MSan ShadowBase"),
315  cl::Hidden, cl::init(0));
316 
317 static cl::opt<unsigned long long> ClOriginBase("msan-origin-base",
318  cl::desc("Define custom MSan OriginBase"),
319  cl::Hidden, cl::init(0));
320 
321 static const char *const kMsanModuleCtorName = "msan.module_ctor";
322 static const char *const kMsanInitName = "__msan_init";
323 
324 namespace {
325 
326 // Memory map parameters used in application-to-shadow address calculation.
327 // Offset = (Addr & ~AndMask) ^ XorMask
328 // Shadow = ShadowBase + Offset
329 // Origin = OriginBase + Offset
330 struct MemoryMapParams {
331  uint64_t AndMask;
332  uint64_t XorMask;
333  uint64_t ShadowBase;
334  uint64_t OriginBase;
335 };
336 
337 struct PlatformMemoryMapParams {
338  const MemoryMapParams *bits32;
339  const MemoryMapParams *bits64;
340 };
341 
342 } // end anonymous namespace
343 
344 // i386 Linux
345 static const MemoryMapParams Linux_I386_MemoryMapParams = {
346  0x000080000000, // AndMask
347  0, // XorMask (not used)
348  0, // ShadowBase (not used)
349  0x000040000000, // OriginBase
350 };
351 
352 // x86_64 Linux
353 static const MemoryMapParams Linux_X86_64_MemoryMapParams = {
354 #ifdef MSAN_LINUX_X86_64_OLD_MAPPING
355  0x400000000000, // AndMask
356  0, // XorMask (not used)
357  0, // ShadowBase (not used)
358  0x200000000000, // OriginBase
359 #else
360  0, // AndMask (not used)
361  0x500000000000, // XorMask
362  0, // ShadowBase (not used)
363  0x100000000000, // OriginBase
364 #endif
365 };
366 
367 // mips64 Linux
368 static const MemoryMapParams Linux_MIPS64_MemoryMapParams = {
369  0, // AndMask (not used)
370  0x008000000000, // XorMask
371  0, // ShadowBase (not used)
372  0x002000000000, // OriginBase
373 };
374 
375 // ppc64 Linux
376 static const MemoryMapParams Linux_PowerPC64_MemoryMapParams = {
377  0xE00000000000, // AndMask
378  0x100000000000, // XorMask
379  0x080000000000, // ShadowBase
380  0x1C0000000000, // OriginBase
381 };
382 
383 // aarch64 Linux
384 static const MemoryMapParams Linux_AArch64_MemoryMapParams = {
385  0, // AndMask (not used)
386  0x06000000000, // XorMask
387  0, // ShadowBase (not used)
388  0x01000000000, // OriginBase
389 };
390 
391 // i386 FreeBSD
392 static const MemoryMapParams FreeBSD_I386_MemoryMapParams = {
393  0x000180000000, // AndMask
394  0x000040000000, // XorMask
395  0x000020000000, // ShadowBase
396  0x000700000000, // OriginBase
397 };
398 
399 // x86_64 FreeBSD
400 static const MemoryMapParams FreeBSD_X86_64_MemoryMapParams = {
401  0xc00000000000, // AndMask
402  0x200000000000, // XorMask
403  0x100000000000, // ShadowBase
404  0x380000000000, // OriginBase
405 };
406 
407 // x86_64 NetBSD
408 static const MemoryMapParams NetBSD_X86_64_MemoryMapParams = {
409  0, // AndMask
410  0x500000000000, // XorMask
411  0, // ShadowBase
412  0x100000000000, // OriginBase
413 };
414 
415 static const PlatformMemoryMapParams Linux_X86_MemoryMapParams = {
418 };
419 
420 static const PlatformMemoryMapParams Linux_MIPS_MemoryMapParams = {
421  nullptr,
423 };
424 
425 static const PlatformMemoryMapParams Linux_PowerPC_MemoryMapParams = {
426  nullptr,
428 };
429 
430 static const PlatformMemoryMapParams Linux_ARM_MemoryMapParams = {
431  nullptr,
433 };
434 
435 static const PlatformMemoryMapParams FreeBSD_X86_MemoryMapParams = {
438 };
439 
440 static const PlatformMemoryMapParams NetBSD_X86_MemoryMapParams = {
441  nullptr,
443 };
444 
445 namespace {
446 
447 /// An instrumentation pass implementing detection of uninitialized
448 /// reads.
449 ///
450 /// MemorySanitizer: instrument the code in module to find
451 /// uninitialized reads.
452 class MemorySanitizer : public FunctionPass {
453 public:
454  // Pass identification, replacement for typeid.
455  static char ID;
456 
457  MemorySanitizer(int TrackOrigins = 0, bool Recover = false,
458  bool EnableKmsan = false)
459  : FunctionPass(ID) {
460  this->CompileKernel =
461  ClEnableKmsan.getNumOccurrences() > 0 ? ClEnableKmsan : EnableKmsan;
462  if (ClTrackOrigins.getNumOccurrences() > 0)
463  this->TrackOrigins = ClTrackOrigins;
464  else
465  this->TrackOrigins = this->CompileKernel ? 2 : TrackOrigins;
466  this->Recover = ClKeepGoing.getNumOccurrences() > 0
467  ? ClKeepGoing
468  : (this->CompileKernel | Recover);
469  }
470  StringRef getPassName() const override { return "MemorySanitizer"; }
471 
472  void getAnalysisUsage(AnalysisUsage &AU) const override {
474  }
475 
476  bool runOnFunction(Function &F) override;
477  bool doInitialization(Module &M) override;
478 
479 private:
480  friend struct MemorySanitizerVisitor;
481  friend struct VarArgAMD64Helper;
482  friend struct VarArgMIPS64Helper;
483  friend struct VarArgAArch64Helper;
484  friend struct VarArgPowerPC64Helper;
485 
486  void initializeCallbacks(Module &M);
487  void createKernelApi(Module &M);
488  void createUserspaceApi(Module &M);
489 
490  /// True if we're compiling the Linux kernel.
491  bool CompileKernel;
492 
493  /// Track origins (allocation points) of uninitialized values.
494  int TrackOrigins;
495  bool Recover;
496 
497  LLVMContext *C;
498  Type *IntptrTy;
499  Type *OriginTy;
500 
501  // XxxTLS variables represent the per-thread state in MSan and per-task state
502  // in KMSAN.
503  // For the userspace these point to thread-local globals. In the kernel land
504  // they point to the members of a per-task struct obtained via a call to
505  // __msan_get_context_state().
506 
507  /// Thread-local shadow storage for function parameters.
508  Value *ParamTLS;
509 
510  /// Thread-local origin storage for function parameters.
511  Value *ParamOriginTLS;
512 
513  /// Thread-local shadow storage for function return value.
514  Value *RetvalTLS;
515 
516  /// Thread-local origin storage for function return value.
517  Value *RetvalOriginTLS;
518 
519  /// Thread-local shadow storage for in-register va_arg function
520  /// parameters (x86_64-specific).
521  Value *VAArgTLS;
522 
523  /// Thread-local shadow storage for in-register va_arg function
524  /// parameters (x86_64-specific).
525  Value *VAArgOriginTLS;
526 
527  /// Thread-local shadow storage for va_arg overflow area
528  /// (x86_64-specific).
529  Value *VAArgOverflowSizeTLS;
530 
531  /// Thread-local space used to pass origin value to the UMR reporting
532  /// function.
533  Value *OriginTLS;
534 
535  /// Are the instrumentation callbacks set up?
536  bool CallbacksInitialized = false;
537 
538  /// The run-time callback to print a warning.
539  Value *WarningFn;
540 
541  // These arrays are indexed by log2(AccessSize).
542  Value *MaybeWarningFn[kNumberOfAccessSizes];
543  Value *MaybeStoreOriginFn[kNumberOfAccessSizes];
544 
545  /// Run-time helper that generates a new origin value for a stack
546  /// allocation.
547  Value *MsanSetAllocaOrigin4Fn;
548 
549  /// Run-time helper that poisons stack on function entry.
550  Value *MsanPoisonStackFn;
551 
552  /// Run-time helper that records a store (or any event) of an
553  /// uninitialized value and returns an updated origin id encoding this info.
554  Value *MsanChainOriginFn;
555 
556  /// MSan runtime replacements for memmove, memcpy and memset.
557  Value *MemmoveFn, *MemcpyFn, *MemsetFn;
558 
559  /// KMSAN callback for task-local function argument shadow.
560  Value *MsanGetContextStateFn;
561 
562  /// Functions for poisoning/unpoisoning local variables
563  Value *MsanPoisonAllocaFn, *MsanUnpoisonAllocaFn;
564 
565  /// Each of the MsanMetadataPtrXxx functions returns a pair of shadow/origin
566  /// pointers.
567  Value *MsanMetadataPtrForLoadN, *MsanMetadataPtrForStoreN;
568  Value *MsanMetadataPtrForLoad_1_8[4];
569  Value *MsanMetadataPtrForStore_1_8[4];
570  Value *MsanInstrumentAsmStoreFn, *MsanInstrumentAsmLoadFn;
571 
572  /// Helper to choose between different MsanMetadataPtrXxx().
573  Value *getKmsanShadowOriginAccessFn(bool isStore, int size);
574 
575  /// Memory map parameters used in application-to-shadow calculation.
576  const MemoryMapParams *MapParams;
577 
578  /// Custom memory map parameters used when -msan-shadow-base or
579  // -msan-origin-base is provided.
580  MemoryMapParams CustomMapParams;
581 
582  MDNode *ColdCallWeights;
583 
584  /// Branch weights for origin store.
585  MDNode *OriginStoreWeights;
586 
587  /// An empty volatile inline asm that prevents callback merge.
588  InlineAsm *EmptyAsm;
589 
590  Function *MsanCtorFunction;
591 };
592 
593 } // end anonymous namespace
594 
595 char MemorySanitizer::ID = 0;
596 
598  MemorySanitizer, "msan",
599  "MemorySanitizer: detects uninitialized reads.", false, false)
602  MemorySanitizer, "msan",
603  "MemorySanitizer: detects uninitialized reads.", false, false)
604 
605 FunctionPass *llvm::createMemorySanitizerPass(int TrackOrigins, bool Recover,
606  bool CompileKernel) {
607  return new MemorySanitizer(TrackOrigins, Recover, CompileKernel);
608 }
609 
610 /// Create a non-const global initialized with the given string.
611 ///
612 /// Creates a writable global for Str so that we can pass it to the
613 /// run-time lib. Runtime uses first 4 bytes of the string to store the
614 /// frame ID, so the string needs to be mutable.
616  StringRef Str) {
617  Constant *StrConst = ConstantDataArray::getString(M.getContext(), Str);
618  return new GlobalVariable(M, StrConst->getType(), /*isConstant=*/false,
619  GlobalValue::PrivateLinkage, StrConst, "");
620 }
621 
622 /// Create KMSAN API callbacks.
623 void MemorySanitizer::createKernelApi(Module &M) {
624  IRBuilder<> IRB(*C);
625 
626  // These will be initialized in insertKmsanPrologue().
627  RetvalTLS = nullptr;
628  RetvalOriginTLS = nullptr;
629  ParamTLS = nullptr;
630  ParamOriginTLS = nullptr;
631  VAArgTLS = nullptr;
632  VAArgOriginTLS = nullptr;
633  VAArgOverflowSizeTLS = nullptr;
634  // OriginTLS is unused in the kernel.
635  OriginTLS = nullptr;
636 
637  // __msan_warning() in the kernel takes an origin.
638  WarningFn = M.getOrInsertFunction("__msan_warning", IRB.getVoidTy(),
639  IRB.getInt32Ty());
640  // Requests the per-task context state (kmsan_context_state*) from the
641  // runtime library.
642  MsanGetContextStateFn = M.getOrInsertFunction(
643  "__msan_get_context_state",
645  StructType::get(ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8),
646  ArrayType::get(IRB.getInt64Ty(), kRetvalTLSSize / 8),
647  ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8),
649  kParamTLSSize / 8), /* va_arg_origin */
650  IRB.getInt64Ty(),
651  ArrayType::get(OriginTy, kParamTLSSize / 4), OriginTy,
652  OriginTy),
653  0));
654 
655  Type *RetTy = StructType::get(PointerType::get(IRB.getInt8Ty(), 0),
656  PointerType::get(IRB.getInt32Ty(), 0));
657 
658  for (int ind = 0, size = 1; ind < 4; ind++, size <<= 1) {
659  std::string name_load =
660  "__msan_metadata_ptr_for_load_" + std::to_string(size);
661  std::string name_store =
662  "__msan_metadata_ptr_for_store_" + std::to_string(size);
663  MsanMetadataPtrForLoad_1_8[ind] = M.getOrInsertFunction(
664  name_load, RetTy, PointerType::get(IRB.getInt8Ty(), 0));
665  MsanMetadataPtrForStore_1_8[ind] = M.getOrInsertFunction(
666  name_store, RetTy, PointerType::get(IRB.getInt8Ty(), 0));
667  }
668 
669  MsanMetadataPtrForLoadN = M.getOrInsertFunction(
670  "__msan_metadata_ptr_for_load_n", RetTy,
671  PointerType::get(IRB.getInt8Ty(), 0), IRB.getInt64Ty());
672  MsanMetadataPtrForStoreN = M.getOrInsertFunction(
673  "__msan_metadata_ptr_for_store_n", RetTy,
674  PointerType::get(IRB.getInt8Ty(), 0), IRB.getInt64Ty());
675 
676  // Functions for poisoning and unpoisoning memory.
677  MsanPoisonAllocaFn =
678  M.getOrInsertFunction("__msan_poison_alloca", IRB.getVoidTy(),
679  IRB.getInt8PtrTy(), IntptrTy, IRB.getInt8PtrTy());
680  MsanUnpoisonAllocaFn = M.getOrInsertFunction(
681  "__msan_unpoison_alloca", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy);
682 }
683 
684 /// Insert declarations for userspace-specific functions and globals.
685 void MemorySanitizer::createUserspaceApi(Module &M) {
686  IRBuilder<> IRB(*C);
687  // Create the callback.
688  // FIXME: this function should have "Cold" calling conv,
689  // which is not yet implemented.
690  StringRef WarningFnName = Recover ? "__msan_warning"
691  : "__msan_warning_noreturn";
692  WarningFn = M.getOrInsertFunction(WarningFnName, IRB.getVoidTy());
693 
694  // Create the global TLS variables.
695  RetvalTLS = new GlobalVariable(
696  M, ArrayType::get(IRB.getInt64Ty(), kRetvalTLSSize / 8), false,
697  GlobalVariable::ExternalLinkage, nullptr, "__msan_retval_tls", nullptr,
699 
700  RetvalOriginTLS = new GlobalVariable(
701  M, OriginTy, false, GlobalVariable::ExternalLinkage, nullptr,
702  "__msan_retval_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);
703 
704  ParamTLS = new GlobalVariable(
705  M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
706  GlobalVariable::ExternalLinkage, nullptr, "__msan_param_tls", nullptr,
708 
709  ParamOriginTLS = new GlobalVariable(
710  M, ArrayType::get(OriginTy, kParamTLSSize / 4), false,
711  GlobalVariable::ExternalLinkage, nullptr, "__msan_param_origin_tls",
713 
714  VAArgTLS = new GlobalVariable(
715  M, ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), false,
716  GlobalVariable::ExternalLinkage, nullptr, "__msan_va_arg_tls", nullptr,
718 
719  VAArgOriginTLS = new GlobalVariable(
720  M, ArrayType::get(OriginTy, kParamTLSSize / 4), false,
721  GlobalVariable::ExternalLinkage, nullptr, "__msan_va_arg_origin_tls",
723 
724  VAArgOverflowSizeTLS = new GlobalVariable(
725  M, IRB.getInt64Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
726  "__msan_va_arg_overflow_size_tls", nullptr,
728  OriginTLS = new GlobalVariable(
729  M, IRB.getInt32Ty(), false, GlobalVariable::ExternalLinkage, nullptr,
730  "__msan_origin_tls", nullptr, GlobalVariable::InitialExecTLSModel);
731 
732  for (size_t AccessSizeIndex = 0; AccessSizeIndex < kNumberOfAccessSizes;
733  AccessSizeIndex++) {
734  unsigned AccessSize = 1 << AccessSizeIndex;
735  std::string FunctionName = "__msan_maybe_warning_" + itostr(AccessSize);
736  MaybeWarningFn[AccessSizeIndex] = M.getOrInsertFunction(
737  FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
738  IRB.getInt32Ty());
739 
740  FunctionName = "__msan_maybe_store_origin_" + itostr(AccessSize);
741  MaybeStoreOriginFn[AccessSizeIndex] = M.getOrInsertFunction(
742  FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
743  IRB.getInt8PtrTy(), IRB.getInt32Ty());
744  }
745 
746  MsanSetAllocaOrigin4Fn = M.getOrInsertFunction(
747  "__msan_set_alloca_origin4", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy,
748  IRB.getInt8PtrTy(), IntptrTy);
749  MsanPoisonStackFn =
750  M.getOrInsertFunction("__msan_poison_stack", IRB.getVoidTy(),
751  IRB.getInt8PtrTy(), IntptrTy);
752 }
753 
754 /// Insert extern declaration of runtime-provided functions and globals.
755 void MemorySanitizer::initializeCallbacks(Module &M) {
756  // Only do this once.
757  if (CallbacksInitialized)
758  return;
759 
760  IRBuilder<> IRB(*C);
761  // Initialize callbacks that are common for kernel and userspace
762  // instrumentation.
763  MsanChainOriginFn = M.getOrInsertFunction(
764  "__msan_chain_origin", IRB.getInt32Ty(), IRB.getInt32Ty());
765  MemmoveFn = M.getOrInsertFunction(
766  "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
767  IRB.getInt8PtrTy(), IntptrTy);
768  MemcpyFn = M.getOrInsertFunction(
769  "__msan_memcpy", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
770  IntptrTy);
771  MemsetFn = M.getOrInsertFunction(
772  "__msan_memset", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt32Ty(),
773  IntptrTy);
774  // We insert an empty inline asm after __msan_report* to avoid callback merge.
775  EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
776  StringRef(""), StringRef(""),
777  /*hasSideEffects=*/true);
778 
779  MsanInstrumentAsmLoadFn =
780  M.getOrInsertFunction("__msan_instrument_asm_load", IRB.getVoidTy(),
781  PointerType::get(IRB.getInt8Ty(), 0), IntptrTy);
782  MsanInstrumentAsmStoreFn =
783  M.getOrInsertFunction("__msan_instrument_asm_store", IRB.getVoidTy(),
784  PointerType::get(IRB.getInt8Ty(), 0), IntptrTy);
785 
786  if (CompileKernel) {
787  createKernelApi(M);
788  } else {
789  createUserspaceApi(M);
790  }
791  CallbacksInitialized = true;
792 }
793 
794 Value *MemorySanitizer::getKmsanShadowOriginAccessFn(bool isStore, int size) {
795  Value **Fns =
796  isStore ? MsanMetadataPtrForStore_1_8 : MsanMetadataPtrForLoad_1_8;
797  switch (size) {
798  case 1:
799  return Fns[0];
800  case 2:
801  return Fns[1];
802  case 4:
803  return Fns[2];
804  case 8:
805  return Fns[3];
806  default:
807  return nullptr;
808  }
809 }
810 
811 /// Module-level initialization.
812 ///
813 /// inserts a call to __msan_init to the module's constructor list.
814 bool MemorySanitizer::doInitialization(Module &M) {
815  auto &DL = M.getDataLayout();
816 
817  bool ShadowPassed = ClShadowBase.getNumOccurrences() > 0;
818  bool OriginPassed = ClOriginBase.getNumOccurrences() > 0;
819  // Check the overrides first
820  if (ShadowPassed || OriginPassed) {
821  CustomMapParams.AndMask = ClAndMask;
822  CustomMapParams.XorMask = ClXorMask;
823  CustomMapParams.ShadowBase = ClShadowBase;
824  CustomMapParams.OriginBase = ClOriginBase;
825  MapParams = &CustomMapParams;
826  } else {
827  Triple TargetTriple(M.getTargetTriple());
828  switch (TargetTriple.getOS()) {
829  case Triple::FreeBSD:
830  switch (TargetTriple.getArch()) {
831  case Triple::x86_64:
832  MapParams = FreeBSD_X86_MemoryMapParams.bits64;
833  break;
834  case Triple::x86:
835  MapParams = FreeBSD_X86_MemoryMapParams.bits32;
836  break;
837  default:
838  report_fatal_error("unsupported architecture");
839  }
840  break;
841  case Triple::NetBSD:
842  switch (TargetTriple.getArch()) {
843  case Triple::x86_64:
844  MapParams = NetBSD_X86_MemoryMapParams.bits64;
845  break;
846  default:
847  report_fatal_error("unsupported architecture");
848  }
849  break;
850  case Triple::Linux:
851  switch (TargetTriple.getArch()) {
852  case Triple::x86_64:
853  MapParams = Linux_X86_MemoryMapParams.bits64;
854  break;
855  case Triple::x86:
856  MapParams = Linux_X86_MemoryMapParams.bits32;
857  break;
858  case Triple::mips64:
859  case Triple::mips64el:
860  MapParams = Linux_MIPS_MemoryMapParams.bits64;
861  break;
862  case Triple::ppc64:
863  case Triple::ppc64le:
864  MapParams = Linux_PowerPC_MemoryMapParams.bits64;
865  break;
866  case Triple::aarch64:
867  case Triple::aarch64_be:
868  MapParams = Linux_ARM_MemoryMapParams.bits64;
869  break;
870  default:
871  report_fatal_error("unsupported architecture");
872  }
873  break;
874  default:
875  report_fatal_error("unsupported operating system");
876  }
877  }
878 
879  C = &(M.getContext());
880  IRBuilder<> IRB(*C);
881  IntptrTy = IRB.getIntPtrTy(DL);
882  OriginTy = IRB.getInt32Ty();
883 
884  ColdCallWeights = MDBuilder(*C).createBranchWeights(1, 1000);
885  OriginStoreWeights = MDBuilder(*C).createBranchWeights(1, 1000);
886 
887  if (!CompileKernel) {
888  std::tie(MsanCtorFunction, std::ignore) =
891  /*InitArgTypes=*/{},
892  /*InitArgs=*/{});
893  if (ClWithComdat) {
894  Comdat *MsanCtorComdat = M.getOrInsertComdat(kMsanModuleCtorName);
895  MsanCtorFunction->setComdat(MsanCtorComdat);
896  appendToGlobalCtors(M, MsanCtorFunction, 0, MsanCtorFunction);
897  } else {
898  appendToGlobalCtors(M, MsanCtorFunction, 0);
899  }
900 
901  if (TrackOrigins)
903  IRB.getInt32(TrackOrigins), "__msan_track_origins");
904 
905  if (Recover)
907  IRB.getInt32(Recover), "__msan_keep_going");
908  }
909  return true;
910 }
911 
912 namespace {
913 
914 /// A helper class that handles instrumentation of VarArg
915 /// functions on a particular platform.
916 ///
917 /// Implementations are expected to insert the instrumentation
918 /// necessary to propagate argument shadow through VarArg function
919 /// calls. Visit* methods are called during an InstVisitor pass over
920 /// the function, and should avoid creating new basic blocks. A new
921 /// instance of this class is created for each instrumented function.
922 struct VarArgHelper {
923  virtual ~VarArgHelper() = default;
924 
925  /// Visit a CallSite.
926  virtual void visitCallSite(CallSite &CS, IRBuilder<> &IRB) = 0;
927 
928  /// Visit a va_start call.
929  virtual void visitVAStartInst(VAStartInst &I) = 0;
930 
931  /// Visit a va_copy call.
932  virtual void visitVACopyInst(VACopyInst &I) = 0;
933 
934  /// Finalize function instrumentation.
935  ///
936  /// This method is called after visiting all interesting (see above)
937  /// instructions in a function.
938  virtual void finalizeInstrumentation() = 0;
939 };
940 
941 struct MemorySanitizerVisitor;
942 
943 } // end anonymous namespace
944 
945 static VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
946  MemorySanitizerVisitor &Visitor);
947 
948 static unsigned TypeSizeToSizeIndex(unsigned TypeSize) {
949  if (TypeSize <= 8) return 0;
950  return Log2_32_Ceil((TypeSize + 7) / 8);
951 }
952 
953 namespace {
954 
955 /// This class does all the work for a given function. Store and Load
956 /// instructions store and load corresponding shadow and origin
957 /// values. Most instructions propagate shadow from arguments to their
958 /// return values. Certain instructions (most importantly, BranchInst)
959 /// test their argument shadow and print reports (with a runtime call) if it's
960 /// non-zero.
961 struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
962  Function &F;
963  MemorySanitizer &MS;
964  SmallVector<PHINode *, 16> ShadowPHINodes, OriginPHINodes;
965  ValueMap<Value*, Value*> ShadowMap, OriginMap;
966  std::unique_ptr<VarArgHelper> VAHelper;
967  const TargetLibraryInfo *TLI;
968  BasicBlock *ActualFnStart;
969 
970  // The following flags disable parts of MSan instrumentation based on
971  // blacklist contents and command-line options.
972  bool InsertChecks;
973  bool PropagateShadow;
974  bool PoisonStack;
975  bool PoisonUndef;
976  bool CheckReturnValue;
977 
978  struct ShadowOriginAndInsertPoint {
979  Value *Shadow;
980  Value *Origin;
981  Instruction *OrigIns;
982 
983  ShadowOriginAndInsertPoint(Value *S, Value *O, Instruction *I)
984  : Shadow(S), Origin(O), OrigIns(I) {}
985  };
988 
989  MemorySanitizerVisitor(Function &F, MemorySanitizer &MS)
990  : F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)) {
991  bool SanitizeFunction = F.hasFnAttribute(Attribute::SanitizeMemory);
992  InsertChecks = SanitizeFunction;
993  PropagateShadow = SanitizeFunction;
994  PoisonStack = SanitizeFunction && ClPoisonStack;
995  PoisonUndef = SanitizeFunction && ClPoisonUndef;
996  // FIXME: Consider using SpecialCaseList to specify a list of functions that
997  // must always return fully initialized values. For now, we hardcode "main".
998  CheckReturnValue = SanitizeFunction && (F.getName() == "main");
999  TLI = &MS.getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
1000 
1001  MS.initializeCallbacks(*F.getParent());
1002  if (MS.CompileKernel)
1003  ActualFnStart = insertKmsanPrologue(F);
1004  else
1005  ActualFnStart = &F.getEntryBlock();
1006 
1007  LLVM_DEBUG(if (!InsertChecks) dbgs()
1008  << "MemorySanitizer is not inserting checks into '"
1009  << F.getName() << "'\n");
1010  }
1011 
1012  Value *updateOrigin(Value *V, IRBuilder<> &IRB) {
1013  if (MS.TrackOrigins <= 1) return V;
1014  return IRB.CreateCall(MS.MsanChainOriginFn, V);
1015  }
1016 
1017  Value *originToIntptr(IRBuilder<> &IRB, Value *Origin) {
1018  const DataLayout &DL = F.getParent()->getDataLayout();
1019  unsigned IntptrSize = DL.getTypeStoreSize(MS.IntptrTy);
1020  if (IntptrSize == kOriginSize) return Origin;
1021  assert(IntptrSize == kOriginSize * 2);
1022  Origin = IRB.CreateIntCast(Origin, MS.IntptrTy, /* isSigned */ false);
1023  return IRB.CreateOr(Origin, IRB.CreateShl(Origin, kOriginSize * 8));
1024  }
1025 
1026  /// Fill memory range with the given origin value.
1027  void paintOrigin(IRBuilder<> &IRB, Value *Origin, Value *OriginPtr,
1028  unsigned Size, unsigned Alignment) {
1029  const DataLayout &DL = F.getParent()->getDataLayout();
1030  unsigned IntptrAlignment = DL.getABITypeAlignment(MS.IntptrTy);
1031  unsigned IntptrSize = DL.getTypeStoreSize(MS.IntptrTy);
1032  assert(IntptrAlignment >= kMinOriginAlignment);
1033  assert(IntptrSize >= kOriginSize);
1034 
1035  unsigned Ofs = 0;
1036  unsigned CurrentAlignment = Alignment;
1037  if (Alignment >= IntptrAlignment && IntptrSize > kOriginSize) {
1038  Value *IntptrOrigin = originToIntptr(IRB, Origin);
1039  Value *IntptrOriginPtr =
1040  IRB.CreatePointerCast(OriginPtr, PointerType::get(MS.IntptrTy, 0));
1041  for (unsigned i = 0; i < Size / IntptrSize; ++i) {
1042  Value *Ptr = i ? IRB.CreateConstGEP1_32(MS.IntptrTy, IntptrOriginPtr, i)
1043  : IntptrOriginPtr;
1044  IRB.CreateAlignedStore(IntptrOrigin, Ptr, CurrentAlignment);
1045  Ofs += IntptrSize / kOriginSize;
1046  CurrentAlignment = IntptrAlignment;
1047  }
1048  }
1049 
1050  for (unsigned i = Ofs; i < (Size + kOriginSize - 1) / kOriginSize; ++i) {
1051  Value *GEP =
1052  i ? IRB.CreateConstGEP1_32(nullptr, OriginPtr, i) : OriginPtr;
1053  IRB.CreateAlignedStore(Origin, GEP, CurrentAlignment);
1054  CurrentAlignment = kMinOriginAlignment;
1055  }
1056  }
1057 
1058  void storeOrigin(IRBuilder<> &IRB, Value *Addr, Value *Shadow, Value *Origin,
1059  Value *OriginPtr, unsigned Alignment, bool AsCall) {
1060  const DataLayout &DL = F.getParent()->getDataLayout();
1061  unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
1062  unsigned StoreSize = DL.getTypeStoreSize(Shadow->getType());
1063  if (Shadow->getType()->isAggregateType()) {
1064  paintOrigin(IRB, updateOrigin(Origin, IRB), OriginPtr, StoreSize,
1065  OriginAlignment);
1066  } else {
1067  Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
1068  Constant *ConstantShadow = dyn_cast_or_null<Constant>(ConvertedShadow);
1069  if (ConstantShadow) {
1070  if (ClCheckConstantShadow && !ConstantShadow->isZeroValue())
1071  paintOrigin(IRB, updateOrigin(Origin, IRB), OriginPtr, StoreSize,
1072  OriginAlignment);
1073  return;
1074  }
1075 
1076  unsigned TypeSizeInBits =
1077  DL.getTypeSizeInBits(ConvertedShadow->getType());
1078  unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
1079  if (AsCall && SizeIndex < kNumberOfAccessSizes && !MS.CompileKernel) {
1080  Value *Fn = MS.MaybeStoreOriginFn[SizeIndex];
1081  Value *ConvertedShadow2 = IRB.CreateZExt(
1082  ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
1083  IRB.CreateCall(Fn, {ConvertedShadow2,
1084  IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()),
1085  Origin});
1086  } else {
1087  Value *Cmp = IRB.CreateICmpNE(
1088  ConvertedShadow, getCleanShadow(ConvertedShadow), "_mscmp");
1090  Cmp, &*IRB.GetInsertPoint(), false, MS.OriginStoreWeights);
1091  IRBuilder<> IRBNew(CheckTerm);
1092  paintOrigin(IRBNew, updateOrigin(Origin, IRBNew), OriginPtr, StoreSize,
1093  OriginAlignment);
1094  }
1095  }
1096  }
1097 
1098  void materializeStores(bool InstrumentWithCalls) {
1099  for (StoreInst *SI : StoreList) {
1100  IRBuilder<> IRB(SI);
1101  Value *Val = SI->getValueOperand();
1102  Value *Addr = SI->getPointerOperand();
1103  Value *Shadow = SI->isAtomic() ? getCleanShadow(Val) : getShadow(Val);
1104  Value *ShadowPtr, *OriginPtr;
1105  Type *ShadowTy = Shadow->getType();
1106  unsigned Alignment = SI->getAlignment();
1107  unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
1108  std::tie(ShadowPtr, OriginPtr) =
1109  getShadowOriginPtr(Addr, IRB, ShadowTy, Alignment, /*isStore*/ true);
1110 
1111  StoreInst *NewSI = IRB.CreateAlignedStore(Shadow, ShadowPtr, Alignment);
1112  LLVM_DEBUG(dbgs() << " STORE: " << *NewSI << "\n");
1113  (void)NewSI;
1114 
1115  if (SI->isAtomic())
1116  SI->setOrdering(addReleaseOrdering(SI->getOrdering()));
1117 
1118  if (MS.TrackOrigins && !SI->isAtomic())
1119  storeOrigin(IRB, Addr, Shadow, getOrigin(Val), OriginPtr,
1120  OriginAlignment, InstrumentWithCalls);
1121  }
1122  }
1123 
1124  /// Helper function to insert a warning at IRB's current insert point.
1125  void insertWarningFn(IRBuilder<> &IRB, Value *Origin) {
1126  if (!Origin)
1127  Origin = (Value *)IRB.getInt32(0);
1128  if (MS.CompileKernel) {
1129  IRB.CreateCall(MS.WarningFn, Origin);
1130  } else {
1131  if (MS.TrackOrigins) {
1132  IRB.CreateStore(Origin, MS.OriginTLS);
1133  }
1134  IRB.CreateCall(MS.WarningFn, {});
1135  }
1136  IRB.CreateCall(MS.EmptyAsm, {});
1137  // FIXME: Insert UnreachableInst if !MS.Recover?
1138  // This may invalidate some of the following checks and needs to be done
1139  // at the very end.
1140  }
1141 
1142  void materializeOneCheck(Instruction *OrigIns, Value *Shadow, Value *Origin,
1143  bool AsCall) {
1144  IRBuilder<> IRB(OrigIns);
1145  LLVM_DEBUG(dbgs() << " SHAD0 : " << *Shadow << "\n");
1146  Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
1147  LLVM_DEBUG(dbgs() << " SHAD1 : " << *ConvertedShadow << "\n");
1148 
1149  Constant *ConstantShadow = dyn_cast_or_null<Constant>(ConvertedShadow);
1150  if (ConstantShadow) {
1151  if (ClCheckConstantShadow && !ConstantShadow->isZeroValue()) {
1152  insertWarningFn(IRB, Origin);
1153  }
1154  return;
1155  }
1156 
1157  const DataLayout &DL = OrigIns->getModule()->getDataLayout();
1158 
1159  unsigned TypeSizeInBits = DL.getTypeSizeInBits(ConvertedShadow->getType());
1160  unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
1161  if (AsCall && SizeIndex < kNumberOfAccessSizes && !MS.CompileKernel) {
1162  Value *Fn = MS.MaybeWarningFn[SizeIndex];
1163  Value *ConvertedShadow2 =
1164  IRB.CreateZExt(ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
1165  IRB.CreateCall(Fn, {ConvertedShadow2, MS.TrackOrigins && Origin
1166  ? Origin
1167  : (Value *)IRB.getInt32(0)});
1168  } else {
1169  Value *Cmp = IRB.CreateICmpNE(ConvertedShadow,
1170  getCleanShadow(ConvertedShadow), "_mscmp");
1172  Cmp, OrigIns,
1173  /* Unreachable */ !MS.Recover, MS.ColdCallWeights);
1174 
1175  IRB.SetInsertPoint(CheckTerm);
1176  insertWarningFn(IRB, Origin);
1177  LLVM_DEBUG(dbgs() << " CHECK: " << *Cmp << "\n");
1178  }
1179  }
1180 
1181  void materializeChecks(bool InstrumentWithCalls) {
1182  for (const auto &ShadowData : InstrumentationList) {
1183  Instruction *OrigIns = ShadowData.OrigIns;
1184  Value *Shadow = ShadowData.Shadow;
1185  Value *Origin = ShadowData.Origin;
1186  materializeOneCheck(OrigIns, Shadow, Origin, InstrumentWithCalls);
1187  }
1188  LLVM_DEBUG(dbgs() << "DONE:\n" << F);
1189  }
1190 
1191  BasicBlock *insertKmsanPrologue(Function &F) {
1192  BasicBlock *ret =
1195  Value *ContextState = IRB.CreateCall(MS.MsanGetContextStateFn, {});
1196  Constant *Zero = IRB.getInt32(0);
1197  MS.ParamTLS =
1198  IRB.CreateGEP(ContextState, {Zero, IRB.getInt32(0)}, "param_shadow");
1199  MS.RetvalTLS =
1200  IRB.CreateGEP(ContextState, {Zero, IRB.getInt32(1)}, "retval_shadow");
1201  MS.VAArgTLS =
1202  IRB.CreateGEP(ContextState, {Zero, IRB.getInt32(2)}, "va_arg_shadow");
1203  MS.VAArgOriginTLS =
1204  IRB.CreateGEP(ContextState, {Zero, IRB.getInt32(3)}, "va_arg_origin");
1205  MS.VAArgOverflowSizeTLS = IRB.CreateGEP(
1206  ContextState, {Zero, IRB.getInt32(4)}, "va_arg_overflow_size");
1207  MS.ParamOriginTLS =
1208  IRB.CreateGEP(ContextState, {Zero, IRB.getInt32(5)}, "param_origin");
1209  MS.RetvalOriginTLS =
1210  IRB.CreateGEP(ContextState, {Zero, IRB.getInt32(6)}, "retval_origin");
1211  return ret;
1212  }
1213 
1214  /// Add MemorySanitizer instrumentation to a function.
1215  bool runOnFunction() {
1216  // In the presence of unreachable blocks, we may see Phi nodes with
1217  // incoming nodes from such blocks. Since InstVisitor skips unreachable
1218  // blocks, such nodes will not have any shadow value associated with them.
1219  // It's easier to remove unreachable blocks than deal with missing shadow.
1221 
1222  // Iterate all BBs in depth-first order and create shadow instructions
1223  // for all instructions (where applicable).
1224  // For PHI nodes we create dummy shadow PHIs which will be finalized later.
1225  for (BasicBlock *BB : depth_first(ActualFnStart))
1226  visit(*BB);
1227 
1228  // Finalize PHI nodes.
1229  for (PHINode *PN : ShadowPHINodes) {
1230  PHINode *PNS = cast<PHINode>(getShadow(PN));
1231  PHINode *PNO = MS.TrackOrigins ? cast<PHINode>(getOrigin(PN)) : nullptr;
1232  size_t NumValues = PN->getNumIncomingValues();
1233  for (size_t v = 0; v < NumValues; v++) {
1234  PNS->addIncoming(getShadow(PN, v), PN->getIncomingBlock(v));
1235  if (PNO) PNO->addIncoming(getOrigin(PN, v), PN->getIncomingBlock(v));
1236  }
1237  }
1238 
1239  VAHelper->finalizeInstrumentation();
1240 
1241  bool InstrumentWithCalls = ClInstrumentationWithCallThreshold >= 0 &&
1242  InstrumentationList.size() + StoreList.size() >
1244 
1245  // Insert shadow value checks.
1246  materializeChecks(InstrumentWithCalls);
1247 
1248  // Delayed instrumentation of StoreInst.
1249  // This may not add new address checks.
1250  materializeStores(InstrumentWithCalls);
1251 
1252  return true;
1253  }
1254 
1255  /// Compute the shadow type that corresponds to a given Value.
1256  Type *getShadowTy(Value *V) {
1257  return getShadowTy(V->getType());
1258  }
1259 
1260  /// Compute the shadow type that corresponds to a given Type.
1261  Type *getShadowTy(Type *OrigTy) {
1262  if (!OrigTy->isSized()) {
1263  return nullptr;
1264  }
1265  // For integer type, shadow is the same as the original type.
1266  // This may return weird-sized types like i1.
1267  if (IntegerType *IT = dyn_cast<IntegerType>(OrigTy))
1268  return IT;
1269  const DataLayout &DL = F.getParent()->getDataLayout();
1270  if (VectorType *VT = dyn_cast<VectorType>(OrigTy)) {
1271  uint32_t EltSize = DL.getTypeSizeInBits(VT->getElementType());
1272  return VectorType::get(IntegerType::get(*MS.C, EltSize),
1273  VT->getNumElements());
1274  }
1275  if (ArrayType *AT = dyn_cast<ArrayType>(OrigTy)) {
1276  return ArrayType::get(getShadowTy(AT->getElementType()),
1277  AT->getNumElements());
1278  }
1279  if (StructType *ST = dyn_cast<StructType>(OrigTy)) {
1280  SmallVector<Type*, 4> Elements;
1281  for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
1282  Elements.push_back(getShadowTy(ST->getElementType(i)));
1283  StructType *Res = StructType::get(*MS.C, Elements, ST->isPacked());
1284  LLVM_DEBUG(dbgs() << "getShadowTy: " << *ST << " ===> " << *Res << "\n");
1285  return Res;
1286  }
1287  uint32_t TypeSize = DL.getTypeSizeInBits(OrigTy);
1288  return IntegerType::get(*MS.C, TypeSize);
1289  }
1290 
1291  /// Flatten a vector type.
1292  Type *getShadowTyNoVec(Type *ty) {
1293  if (VectorType *vt = dyn_cast<VectorType>(ty))
1294  return IntegerType::get(*MS.C, vt->getBitWidth());
1295  return ty;
1296  }
1297 
1298  /// Convert a shadow value to it's flattened variant.
1299  Value *convertToShadowTyNoVec(Value *V, IRBuilder<> &IRB) {
1300  Type *Ty = V->getType();
1301  Type *NoVecTy = getShadowTyNoVec(Ty);
1302  if (Ty == NoVecTy) return V;
1303  return IRB.CreateBitCast(V, NoVecTy);
1304  }
1305 
1306  /// Compute the integer shadow offset that corresponds to a given
1307  /// application address.
1308  ///
1309  /// Offset = (Addr & ~AndMask) ^ XorMask
1310  Value *getShadowPtrOffset(Value *Addr, IRBuilder<> &IRB) {
1311  Value *OffsetLong = IRB.CreatePointerCast(Addr, MS.IntptrTy);
1312 
1313  uint64_t AndMask = MS.MapParams->AndMask;
1314  if (AndMask)
1315  OffsetLong =
1316  IRB.CreateAnd(OffsetLong, ConstantInt::get(MS.IntptrTy, ~AndMask));
1317 
1318  uint64_t XorMask = MS.MapParams->XorMask;
1319  if (XorMask)
1320  OffsetLong =
1321  IRB.CreateXor(OffsetLong, ConstantInt::get(MS.IntptrTy, XorMask));
1322  return OffsetLong;
1323  }
1324 
1325  /// Compute the shadow and origin addresses corresponding to a given
1326  /// application address.
1327  ///
1328  /// Shadow = ShadowBase + Offset
1329  /// Origin = (OriginBase + Offset) & ~3ULL
1330  std::pair<Value *, Value *> getShadowOriginPtrUserspace(Value *Addr,
1331  IRBuilder<> &IRB,
1332  Type *ShadowTy,
1333  unsigned Alignment) {
1334  Value *ShadowOffset = getShadowPtrOffset(Addr, IRB);
1335  Value *ShadowLong = ShadowOffset;
1336  uint64_t ShadowBase = MS.MapParams->ShadowBase;
1337  if (ShadowBase != 0) {
1338  ShadowLong =
1339  IRB.CreateAdd(ShadowLong,
1340  ConstantInt::get(MS.IntptrTy, ShadowBase));
1341  }
1342  Value *ShadowPtr =
1343  IRB.CreateIntToPtr(ShadowLong, PointerType::get(ShadowTy, 0));
1344  Value *OriginPtr = nullptr;
1345  if (MS.TrackOrigins) {
1346  Value *OriginLong = ShadowOffset;
1347  uint64_t OriginBase = MS.MapParams->OriginBase;
1348  if (OriginBase != 0)
1349  OriginLong = IRB.CreateAdd(OriginLong,
1350  ConstantInt::get(MS.IntptrTy, OriginBase));
1351  if (Alignment < kMinOriginAlignment) {
1352  uint64_t Mask = kMinOriginAlignment - 1;
1353  OriginLong =
1354  IRB.CreateAnd(OriginLong, ConstantInt::get(MS.IntptrTy, ~Mask));
1355  }
1356  OriginPtr =
1357  IRB.CreateIntToPtr(OriginLong, PointerType::get(IRB.getInt32Ty(), 0));
1358  }
1359  return std::make_pair(ShadowPtr, OriginPtr);
1360  }
1361 
1362  std::pair<Value *, Value *>
1363  getShadowOriginPtrKernel(Value *Addr, IRBuilder<> &IRB, Type *ShadowTy,
1364  unsigned Alignment, bool isStore) {
1365  Value *ShadowOriginPtrs;
1366  const DataLayout &DL = F.getParent()->getDataLayout();
1367  int Size = DL.getTypeStoreSize(ShadowTy);
1368 
1369  Value *Getter = MS.getKmsanShadowOriginAccessFn(isStore, Size);
1370  Value *AddrCast =
1371  IRB.CreatePointerCast(Addr, PointerType::get(IRB.getInt8Ty(), 0));
1372  if (Getter) {
1373  ShadowOriginPtrs = IRB.CreateCall(Getter, AddrCast);
1374  } else {
1375  Value *SizeVal = ConstantInt::get(MS.IntptrTy, Size);
1376  ShadowOriginPtrs = IRB.CreateCall(isStore ? MS.MsanMetadataPtrForStoreN
1377  : MS.MsanMetadataPtrForLoadN,
1378  {AddrCast, SizeVal});
1379  }
1380  Value *ShadowPtr = IRB.CreateExtractValue(ShadowOriginPtrs, 0);
1381  ShadowPtr = IRB.CreatePointerCast(ShadowPtr, PointerType::get(ShadowTy, 0));
1382  Value *OriginPtr = IRB.CreateExtractValue(ShadowOriginPtrs, 1);
1383 
1384  return std::make_pair(ShadowPtr, OriginPtr);
1385  }
1386 
1387  std::pair<Value *, Value *> getShadowOriginPtr(Value *Addr, IRBuilder<> &IRB,
1388  Type *ShadowTy,
1389  unsigned Alignment,
1390  bool isStore) {
1391  std::pair<Value *, Value *> ret;
1392  if (MS.CompileKernel)
1393  ret = getShadowOriginPtrKernel(Addr, IRB, ShadowTy, Alignment, isStore);
1394  else
1395  ret = getShadowOriginPtrUserspace(Addr, IRB, ShadowTy, Alignment);
1396  return ret;
1397  }
1398 
1399  /// Compute the shadow address for a given function argument.
1400  ///
1401  /// Shadow = ParamTLS+ArgOffset.
1402  Value *getShadowPtrForArgument(Value *A, IRBuilder<> &IRB,
1403  int ArgOffset) {
1404  Value *Base = IRB.CreatePointerCast(MS.ParamTLS, MS.IntptrTy);
1405  if (ArgOffset)
1406  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
1407  return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
1408  "_msarg");
1409  }
1410 
1411  /// Compute the origin address for a given function argument.
1412  Value *getOriginPtrForArgument(Value *A, IRBuilder<> &IRB,
1413  int ArgOffset) {
1414  if (!MS.TrackOrigins)
1415  return nullptr;
1416  Value *Base = IRB.CreatePointerCast(MS.ParamOriginTLS, MS.IntptrTy);
1417  if (ArgOffset)
1418  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
1419  return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
1420  "_msarg_o");
1421  }
1422 
1423  /// Compute the shadow address for a retval.
1424  Value *getShadowPtrForRetval(Value *A, IRBuilder<> &IRB) {
1425  return IRB.CreatePointerCast(MS.RetvalTLS,
1426  PointerType::get(getShadowTy(A), 0),
1427  "_msret");
1428  }
1429 
1430  /// Compute the origin address for a retval.
1431  Value *getOriginPtrForRetval(IRBuilder<> &IRB) {
1432  // We keep a single origin for the entire retval. Might be too optimistic.
1433  return MS.RetvalOriginTLS;
1434  }
1435 
1436  /// Set SV to be the shadow value for V.
1437  void setShadow(Value *V, Value *SV) {
1438  assert(!ShadowMap.count(V) && "Values may only have one shadow");
1439  ShadowMap[V] = PropagateShadow ? SV : getCleanShadow(V);
1440  }
1441 
1442  /// Set Origin to be the origin value for V.
1443  void setOrigin(Value *V, Value *Origin) {
1444  if (!MS.TrackOrigins) return;
1445  assert(!OriginMap.count(V) && "Values may only have one origin");
1446  LLVM_DEBUG(dbgs() << "ORIGIN: " << *V << " ==> " << *Origin << "\n");
1447  OriginMap[V] = Origin;
1448  }
1449 
1450  Constant *getCleanShadow(Type *OrigTy) {
1451  Type *ShadowTy = getShadowTy(OrigTy);
1452  if (!ShadowTy)
1453  return nullptr;
1454  return Constant::getNullValue(ShadowTy);
1455  }
1456 
1457  /// Create a clean shadow value for a given value.
1458  ///
1459  /// Clean shadow (all zeroes) means all bits of the value are defined
1460  /// (initialized).
1461  Constant *getCleanShadow(Value *V) {
1462  return getCleanShadow(V->getType());
1463  }
1464 
1465  /// Create a dirty shadow of a given shadow type.
1466  Constant *getPoisonedShadow(Type *ShadowTy) {
1467  assert(ShadowTy);
1468  if (isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy))
1469  return Constant::getAllOnesValue(ShadowTy);
1470  if (ArrayType *AT = dyn_cast<ArrayType>(ShadowTy)) {
1471  SmallVector<Constant *, 4> Vals(AT->getNumElements(),
1472  getPoisonedShadow(AT->getElementType()));
1473  return ConstantArray::get(AT, Vals);
1474  }
1475  if (StructType *ST = dyn_cast<StructType>(ShadowTy)) {
1477  for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
1478  Vals.push_back(getPoisonedShadow(ST->getElementType(i)));
1479  return ConstantStruct::get(ST, Vals);
1480  }
1481  llvm_unreachable("Unexpected shadow type");
1482  }
1483 
1484  /// Create a dirty shadow for a given value.
1485  Constant *getPoisonedShadow(Value *V) {
1486  Type *ShadowTy = getShadowTy(V);
1487  if (!ShadowTy)
1488  return nullptr;
1489  return getPoisonedShadow(ShadowTy);
1490  }
1491 
1492  /// Create a clean (zero) origin.
1493  Value *getCleanOrigin() {
1494  return Constant::getNullValue(MS.OriginTy);
1495  }
1496 
1497  /// Get the shadow value for a given Value.
1498  ///
1499  /// This function either returns the value set earlier with setShadow,
1500  /// or extracts if from ParamTLS (for function arguments).
1501  Value *getShadow(Value *V) {
1502  if (!PropagateShadow) return getCleanShadow(V);
1503  if (Instruction *I = dyn_cast<Instruction>(V)) {
1504  if (I->getMetadata("nosanitize"))
1505  return getCleanShadow(V);
1506  // For instructions the shadow is already stored in the map.
1507  Value *Shadow = ShadowMap[V];
1508  if (!Shadow) {
1509  LLVM_DEBUG(dbgs() << "No shadow: " << *V << "\n" << *(I->getParent()));
1510  (void)I;
1511  assert(Shadow && "No shadow for a value");
1512  }
1513  return Shadow;
1514  }
1515  if (UndefValue *U = dyn_cast<UndefValue>(V)) {
1516  Value *AllOnes = PoisonUndef ? getPoisonedShadow(V) : getCleanShadow(V);
1517  LLVM_DEBUG(dbgs() << "Undef: " << *U << " ==> " << *AllOnes << "\n");
1518  (void)U;
1519  return AllOnes;
1520  }
1521  if (Argument *A = dyn_cast<Argument>(V)) {
1522  // For arguments we compute the shadow on demand and store it in the map.
1523  Value **ShadowPtr = &ShadowMap[V];
1524  if (*ShadowPtr)
1525  return *ShadowPtr;
1526  Function *F = A->getParent();
1527  IRBuilder<> EntryIRB(ActualFnStart->getFirstNonPHI());
1528  unsigned ArgOffset = 0;
1529  const DataLayout &DL = F->getParent()->getDataLayout();
1530  for (auto &FArg : F->args()) {
1531  if (!FArg.getType()->isSized()) {
1532  LLVM_DEBUG(dbgs() << "Arg is not sized\n");
1533  continue;
1534  }
1535  unsigned Size =
1536  FArg.hasByValAttr()
1537  ? DL.getTypeAllocSize(FArg.getType()->getPointerElementType())
1538  : DL.getTypeAllocSize(FArg.getType());
1539  if (A == &FArg) {
1540  bool Overflow = ArgOffset + Size > kParamTLSSize;
1541  Value *Base = getShadowPtrForArgument(&FArg, EntryIRB, ArgOffset);
1542  if (FArg.hasByValAttr()) {
1543  // ByVal pointer itself has clean shadow. We copy the actual
1544  // argument shadow to the underlying memory.
1545  // Figure out maximal valid memcpy alignment.
1546  unsigned ArgAlign = FArg.getParamAlignment();
1547  if (ArgAlign == 0) {
1548  Type *EltType = A->getType()->getPointerElementType();
1549  ArgAlign = DL.getABITypeAlignment(EltType);
1550  }
1551  Value *CpShadowPtr =
1552  getShadowOriginPtr(V, EntryIRB, EntryIRB.getInt8Ty(), ArgAlign,
1553  /*isStore*/ true)
1554  .first;
1555  // TODO(glider): need to copy origins.
1556  if (Overflow) {
1557  // ParamTLS overflow.
1558  EntryIRB.CreateMemSet(
1559  CpShadowPtr, Constant::getNullValue(EntryIRB.getInt8Ty()),
1560  Size, ArgAlign);
1561  } else {
1562  unsigned CopyAlign = std::min(ArgAlign, kShadowTLSAlignment);
1563  Value *Cpy = EntryIRB.CreateMemCpy(CpShadowPtr, CopyAlign, Base,
1564  CopyAlign, Size);
1565  LLVM_DEBUG(dbgs() << " ByValCpy: " << *Cpy << "\n");
1566  (void)Cpy;
1567  }
1568  *ShadowPtr = getCleanShadow(V);
1569  } else {
1570  if (Overflow) {
1571  // ParamTLS overflow.
1572  *ShadowPtr = getCleanShadow(V);
1573  } else {
1574  *ShadowPtr =
1575  EntryIRB.CreateAlignedLoad(Base, kShadowTLSAlignment);
1576  }
1577  }
1578  LLVM_DEBUG(dbgs()
1579  << " ARG: " << FArg << " ==> " << **ShadowPtr << "\n");
1580  if (MS.TrackOrigins && !Overflow) {
1581  Value *OriginPtr =
1582  getOriginPtrForArgument(&FArg, EntryIRB, ArgOffset);
1583  setOrigin(A, EntryIRB.CreateLoad(OriginPtr));
1584  } else {
1585  setOrigin(A, getCleanOrigin());
1586  }
1587  }
1588  ArgOffset += alignTo(Size, kShadowTLSAlignment);
1589  }
1590  assert(*ShadowPtr && "Could not find shadow for an argument");
1591  return *ShadowPtr;
1592  }
1593  // For everything else the shadow is zero.
1594  return getCleanShadow(V);
1595  }
1596 
1597  /// Get the shadow for i-th argument of the instruction I.
1598  Value *getShadow(Instruction *I, int i) {
1599  return getShadow(I->getOperand(i));
1600  }
1601 
1602  /// Get the origin for a value.
1603  Value *getOrigin(Value *V) {
1604  if (!MS.TrackOrigins) return nullptr;
1605  if (!PropagateShadow) return getCleanOrigin();
1606  if (isa<Constant>(V)) return getCleanOrigin();
1607  assert((isa<Instruction>(V) || isa<Argument>(V)) &&
1608  "Unexpected value type in getOrigin()");
1609  if (Instruction *I = dyn_cast<Instruction>(V)) {
1610  if (I->getMetadata("nosanitize"))
1611  return getCleanOrigin();
1612  }
1613  Value *Origin = OriginMap[V];
1614  assert(Origin && "Missing origin");
1615  return Origin;
1616  }
1617 
1618  /// Get the origin for i-th argument of the instruction I.
1619  Value *getOrigin(Instruction *I, int i) {
1620  return getOrigin(I->getOperand(i));
1621  }
1622 
1623  /// Remember the place where a shadow check should be inserted.
1624  ///
1625  /// This location will be later instrumented with a check that will print a
1626  /// UMR warning in runtime if the shadow value is not 0.
1627  void insertShadowCheck(Value *Shadow, Value *Origin, Instruction *OrigIns) {
1628  assert(Shadow);
1629  if (!InsertChecks) return;
1630 #ifndef NDEBUG
1631  Type *ShadowTy = Shadow->getType();
1632  assert((isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy)) &&
1633  "Can only insert checks for integer and vector shadow types");
1634 #endif
1635  InstrumentationList.push_back(
1636  ShadowOriginAndInsertPoint(Shadow, Origin, OrigIns));
1637  }
1638 
1639  /// Remember the place where a shadow check should be inserted.
1640  ///
1641  /// This location will be later instrumented with a check that will print a
1642  /// UMR warning in runtime if the value is not fully defined.
1643  void insertShadowCheck(Value *Val, Instruction *OrigIns) {
1644  assert(Val);
1645  Value *Shadow, *Origin;
1646  if (ClCheckConstantShadow) {
1647  Shadow = getShadow(Val);
1648  if (!Shadow) return;
1649  Origin = getOrigin(Val);
1650  } else {
1651  Shadow = dyn_cast_or_null<Instruction>(getShadow(Val));
1652  if (!Shadow) return;
1653  Origin = dyn_cast_or_null<Instruction>(getOrigin(Val));
1654  }
1655  insertShadowCheck(Shadow, Origin, OrigIns);
1656  }
1657 
1658  AtomicOrdering addReleaseOrdering(AtomicOrdering a) {
1659  switch (a) {
1665  return AtomicOrdering::Release;
1671  }
1672  llvm_unreachable("Unknown ordering");
1673  }
1674 
1675  AtomicOrdering addAcquireOrdering(AtomicOrdering a) {
1676  switch (a) {
1682  return AtomicOrdering::Acquire;
1688  }
1689  llvm_unreachable("Unknown ordering");
1690  }
1691 
1692  // ------------------- Visitors.
1694  void visit(Instruction &I) {
1695  if (!I.getMetadata("nosanitize"))
1697  }
1698 
1699  /// Instrument LoadInst
1700  ///
1701  /// Loads the corresponding shadow and (optionally) origin.
1702  /// Optionally, checks that the load address is fully defined.
1703  void visitLoadInst(LoadInst &I) {
1704  assert(I.getType()->isSized() && "Load type must have size");
1705  assert(!I.getMetadata("nosanitize"));
1706  IRBuilder<> IRB(I.getNextNode());
1707  Type *ShadowTy = getShadowTy(&I);
1708  Value *Addr = I.getPointerOperand();
1709  Value *ShadowPtr, *OriginPtr;
1710  unsigned Alignment = I.getAlignment();
1711  if (PropagateShadow) {
1712  std::tie(ShadowPtr, OriginPtr) =
1713  getShadowOriginPtr(Addr, IRB, ShadowTy, Alignment, /*isStore*/ false);
1714  setShadow(&I, IRB.CreateAlignedLoad(ShadowPtr, Alignment, "_msld"));
1715  } else {
1716  setShadow(&I, getCleanShadow(&I));
1717  }
1718 
1720  insertShadowCheck(I.getPointerOperand(), &I);
1721 
1722  if (I.isAtomic())
1723  I.setOrdering(addAcquireOrdering(I.getOrdering()));
1724 
1725  if (MS.TrackOrigins) {
1726  if (PropagateShadow) {
1727  unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
1728  setOrigin(&I, IRB.CreateAlignedLoad(OriginPtr, OriginAlignment));
1729  } else {
1730  setOrigin(&I, getCleanOrigin());
1731  }
1732  }
1733  }
1734 
1735  /// Instrument StoreInst
1736  ///
1737  /// Stores the corresponding shadow and (optionally) origin.
1738  /// Optionally, checks that the store address is fully defined.
1739  void visitStoreInst(StoreInst &I) {
1740  StoreList.push_back(&I);
1742  insertShadowCheck(I.getPointerOperand(), &I);
1743  }
1744 
1745  void handleCASOrRMW(Instruction &I) {
1746  assert(isa<AtomicRMWInst>(I) || isa<AtomicCmpXchgInst>(I));
1747 
1748  IRBuilder<> IRB(&I);
1749  Value *Addr = I.getOperand(0);
1750  Value *ShadowPtr = getShadowOriginPtr(Addr, IRB, I.getType(),
1751  /*Alignment*/ 1, /*isStore*/ true)
1752  .first;
1753 
1755  insertShadowCheck(Addr, &I);
1756 
1757  // Only test the conditional argument of cmpxchg instruction.
1758  // The other argument can potentially be uninitialized, but we can not
1759  // detect this situation reliably without possible false positives.
1760  if (isa<AtomicCmpXchgInst>(I))
1761  insertShadowCheck(I.getOperand(1), &I);
1762 
1763  IRB.CreateStore(getCleanShadow(&I), ShadowPtr);
1764 
1765  setShadow(&I, getCleanShadow(&I));
1766  setOrigin(&I, getCleanOrigin());
1767  }
1768 
1769  void visitAtomicRMWInst(AtomicRMWInst &I) {
1770  handleCASOrRMW(I);
1771  I.setOrdering(addReleaseOrdering(I.getOrdering()));
1772  }
1773 
1774  void visitAtomicCmpXchgInst(AtomicCmpXchgInst &I) {
1775  handleCASOrRMW(I);
1776  I.setSuccessOrdering(addReleaseOrdering(I.getSuccessOrdering()));
1777  }
1778 
1779  // Vector manipulation.
1780  void visitExtractElementInst(ExtractElementInst &I) {
1781  insertShadowCheck(I.getOperand(1), &I);
1782  IRBuilder<> IRB(&I);
1783  setShadow(&I, IRB.CreateExtractElement(getShadow(&I, 0), I.getOperand(1),
1784  "_msprop"));
1785  setOrigin(&I, getOrigin(&I, 0));
1786  }
1787 
1788  void visitInsertElementInst(InsertElementInst &I) {
1789  insertShadowCheck(I.getOperand(2), &I);
1790  IRBuilder<> IRB(&I);
1791  setShadow(&I, IRB.CreateInsertElement(getShadow(&I, 0), getShadow(&I, 1),
1792  I.getOperand(2), "_msprop"));
1793  setOriginForNaryOp(I);
1794  }
1795 
1796  void visitShuffleVectorInst(ShuffleVectorInst &I) {
1797  insertShadowCheck(I.getOperand(2), &I);
1798  IRBuilder<> IRB(&I);
1799  setShadow(&I, IRB.CreateShuffleVector(getShadow(&I, 0), getShadow(&I, 1),
1800  I.getOperand(2), "_msprop"));
1801  setOriginForNaryOp(I);
1802  }
1803 
1804  // Casts.
1805  void visitSExtInst(SExtInst &I) {
1806  IRBuilder<> IRB(&I);
1807  setShadow(&I, IRB.CreateSExt(getShadow(&I, 0), I.getType(), "_msprop"));
1808  setOrigin(&I, getOrigin(&I, 0));
1809  }
1810 
1811  void visitZExtInst(ZExtInst &I) {
1812  IRBuilder<> IRB(&I);
1813  setShadow(&I, IRB.CreateZExt(getShadow(&I, 0), I.getType(), "_msprop"));
1814  setOrigin(&I, getOrigin(&I, 0));
1815  }
1816 
1817  void visitTruncInst(TruncInst &I) {
1818  IRBuilder<> IRB(&I);
1819  setShadow(&I, IRB.CreateTrunc(getShadow(&I, 0), I.getType(), "_msprop"));
1820  setOrigin(&I, getOrigin(&I, 0));
1821  }
1822 
1823  void visitBitCastInst(BitCastInst &I) {
1824  // Special case: if this is the bitcast (there is exactly 1 allowed) between
1825  // a musttail call and a ret, don't instrument. New instructions are not
1826  // allowed after a musttail call.
1827  if (auto *CI = dyn_cast<CallInst>(I.getOperand(0)))
1828  if (CI->isMustTailCall())
1829  return;
1830  IRBuilder<> IRB(&I);
1831  setShadow(&I, IRB.CreateBitCast(getShadow(&I, 0), getShadowTy(&I)));
1832  setOrigin(&I, getOrigin(&I, 0));
1833  }
1834 
1835  void visitPtrToIntInst(PtrToIntInst &I) {
1836  IRBuilder<> IRB(&I);
1837  setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1838  "_msprop_ptrtoint"));
1839  setOrigin(&I, getOrigin(&I, 0));
1840  }
1841 
1842  void visitIntToPtrInst(IntToPtrInst &I) {
1843  IRBuilder<> IRB(&I);
1844  setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1845  "_msprop_inttoptr"));
1846  setOrigin(&I, getOrigin(&I, 0));
1847  }
1848 
1849  void visitFPToSIInst(CastInst& I) { handleShadowOr(I); }
1850  void visitFPToUIInst(CastInst& I) { handleShadowOr(I); }
1851  void visitSIToFPInst(CastInst& I) { handleShadowOr(I); }
1852  void visitUIToFPInst(CastInst& I) { handleShadowOr(I); }
1853  void visitFPExtInst(CastInst& I) { handleShadowOr(I); }
1854  void visitFPTruncInst(CastInst& I) { handleShadowOr(I); }
1855 
1856  /// Propagate shadow for bitwise AND.
1857  ///
1858  /// This code is exact, i.e. if, for example, a bit in the left argument
1859  /// is defined and 0, then neither the value not definedness of the
1860  /// corresponding bit in B don't affect the resulting shadow.
1861  void visitAnd(BinaryOperator &I) {
1862  IRBuilder<> IRB(&I);
1863  // "And" of 0 and a poisoned value results in unpoisoned value.
1864  // 1&1 => 1; 0&1 => 0; p&1 => p;
1865  // 1&0 => 0; 0&0 => 0; p&0 => 0;
1866  // 1&p => p; 0&p => 0; p&p => p;
1867  // S = (S1 & S2) | (V1 & S2) | (S1 & V2)
1868  Value *S1 = getShadow(&I, 0);
1869  Value *S2 = getShadow(&I, 1);
1870  Value *V1 = I.getOperand(0);
1871  Value *V2 = I.getOperand(1);
1872  if (V1->getType() != S1->getType()) {
1873  V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1874  V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1875  }
1876  Value *S1S2 = IRB.CreateAnd(S1, S2);
1877  Value *V1S2 = IRB.CreateAnd(V1, S2);
1878  Value *S1V2 = IRB.CreateAnd(S1, V2);
1879  setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
1880  setOriginForNaryOp(I);
1881  }
1882 
1883  void visitOr(BinaryOperator &I) {
1884  IRBuilder<> IRB(&I);
1885  // "Or" of 1 and a poisoned value results in unpoisoned value.
1886  // 1|1 => 1; 0|1 => 1; p|1 => 1;
1887  // 1|0 => 1; 0|0 => 0; p|0 => p;
1888  // 1|p => 1; 0|p => p; p|p => p;
1889  // S = (S1 & S2) | (~V1 & S2) | (S1 & ~V2)
1890  Value *S1 = getShadow(&I, 0);
1891  Value *S2 = getShadow(&I, 1);
1892  Value *V1 = IRB.CreateNot(I.getOperand(0));
1893  Value *V2 = IRB.CreateNot(I.getOperand(1));
1894  if (V1->getType() != S1->getType()) {
1895  V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1896  V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1897  }
1898  Value *S1S2 = IRB.CreateAnd(S1, S2);
1899  Value *V1S2 = IRB.CreateAnd(V1, S2);
1900  Value *S1V2 = IRB.CreateAnd(S1, V2);
1901  setShadow(&I, IRB.CreateOr(S1S2, IRB.CreateOr(V1S2, S1V2)));
1902  setOriginForNaryOp(I);
1903  }
1904 
1905  /// Default propagation of shadow and/or origin.
1906  ///
1907  /// This class implements the general case of shadow propagation, used in all
1908  /// cases where we don't know and/or don't care about what the operation
1909  /// actually does. It converts all input shadow values to a common type
1910  /// (extending or truncating as necessary), and bitwise OR's them.
1911  ///
1912  /// This is much cheaper than inserting checks (i.e. requiring inputs to be
1913  /// fully initialized), and less prone to false positives.
1914  ///
1915  /// This class also implements the general case of origin propagation. For a
1916  /// Nary operation, result origin is set to the origin of an argument that is
1917  /// not entirely initialized. If there is more than one such arguments, the
1918  /// rightmost of them is picked. It does not matter which one is picked if all
1919  /// arguments are initialized.
1920  template <bool CombineShadow>
1921  class Combiner {
1922  Value *Shadow = nullptr;
1923  Value *Origin = nullptr;
1924  IRBuilder<> &IRB;
1925  MemorySanitizerVisitor *MSV;
1926 
1927  public:
1928  Combiner(MemorySanitizerVisitor *MSV, IRBuilder<> &IRB)
1929  : IRB(IRB), MSV(MSV) {}
1930 
1931  /// Add a pair of shadow and origin values to the mix.
1932  Combiner &Add(Value *OpShadow, Value *OpOrigin) {
1933  if (CombineShadow) {
1934  assert(OpShadow);
1935  if (!Shadow)
1936  Shadow = OpShadow;
1937  else {
1938  OpShadow = MSV->CreateShadowCast(IRB, OpShadow, Shadow->getType());
1939  Shadow = IRB.CreateOr(Shadow, OpShadow, "_msprop");
1940  }
1941  }
1942 
1943  if (MSV->MS.TrackOrigins) {
1944  assert(OpOrigin);
1945  if (!Origin) {
1946  Origin = OpOrigin;
1947  } else {
1948  Constant *ConstOrigin = dyn_cast<Constant>(OpOrigin);
1949  // No point in adding something that might result in 0 origin value.
1950  if (!ConstOrigin || !ConstOrigin->isNullValue()) {
1951  Value *FlatShadow = MSV->convertToShadowTyNoVec(OpShadow, IRB);
1952  Value *Cond =
1953  IRB.CreateICmpNE(FlatShadow, MSV->getCleanShadow(FlatShadow));
1954  Origin = IRB.CreateSelect(Cond, OpOrigin, Origin);
1955  }
1956  }
1957  }
1958  return *this;
1959  }
1960 
1961  /// Add an application value to the mix.
1962  Combiner &Add(Value *V) {
1963  Value *OpShadow = MSV->getShadow(V);
1964  Value *OpOrigin = MSV->MS.TrackOrigins ? MSV->getOrigin(V) : nullptr;
1965  return Add(OpShadow, OpOrigin);
1966  }
1967 
1968  /// Set the current combined values as the given instruction's shadow
1969  /// and origin.
1970  void Done(Instruction *I) {
1971  if (CombineShadow) {
1972  assert(Shadow);
1973  Shadow = MSV->CreateShadowCast(IRB, Shadow, MSV->getShadowTy(I));
1974  MSV->setShadow(I, Shadow);
1975  }
1976  if (MSV->MS.TrackOrigins) {
1977  assert(Origin);
1978  MSV->setOrigin(I, Origin);
1979  }
1980  }
1981  };
1982 
1983  using ShadowAndOriginCombiner = Combiner<true>;
1984  using OriginCombiner = Combiner<false>;
1985 
1986  /// Propagate origin for arbitrary operation.
1987  void setOriginForNaryOp(Instruction &I) {
1988  if (!MS.TrackOrigins) return;
1989  IRBuilder<> IRB(&I);
1990  OriginCombiner OC(this, IRB);
1991  for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
1992  OC.Add(OI->get());
1993  OC.Done(&I);
1994  }
1995 
1996  size_t VectorOrPrimitiveTypeSizeInBits(Type *Ty) {
1997  assert(!(Ty->isVectorTy() && Ty->getScalarType()->isPointerTy()) &&
1998  "Vector of pointers is not a valid shadow type");
1999  return Ty->isVectorTy() ?
2001  Ty->getPrimitiveSizeInBits();
2002  }
2003 
2004  /// Cast between two shadow types, extending or truncating as
2005  /// necessary.
2006  Value *CreateShadowCast(IRBuilder<> &IRB, Value *V, Type *dstTy,
2007  bool Signed = false) {
2008  Type *srcTy = V->getType();
2009  size_t srcSizeInBits = VectorOrPrimitiveTypeSizeInBits(srcTy);
2010  size_t dstSizeInBits = VectorOrPrimitiveTypeSizeInBits(dstTy);
2011  if (srcSizeInBits > 1 && dstSizeInBits == 1)
2012  return IRB.CreateICmpNE(V, getCleanShadow(V));
2013 
2014  if (dstTy->isIntegerTy() && srcTy->isIntegerTy())
2015  return IRB.CreateIntCast(V, dstTy, Signed);
2016  if (dstTy->isVectorTy() && srcTy->isVectorTy() &&
2017  dstTy->getVectorNumElements() == srcTy->getVectorNumElements())
2018  return IRB.CreateIntCast(V, dstTy, Signed);
2019  Value *V1 = IRB.CreateBitCast(V, Type::getIntNTy(*MS.C, srcSizeInBits));
2020  Value *V2 =
2021  IRB.CreateIntCast(V1, Type::getIntNTy(*MS.C, dstSizeInBits), Signed);
2022  return IRB.CreateBitCast(V2, dstTy);
2023  // TODO: handle struct types.
2024  }
2025 
2026  /// Cast an application value to the type of its own shadow.
2027  Value *CreateAppToShadowCast(IRBuilder<> &IRB, Value *V) {
2028  Type *ShadowTy = getShadowTy(V);
2029  if (V->getType() == ShadowTy)
2030  return V;
2031  if (V->getType()->isPtrOrPtrVectorTy())
2032  return IRB.CreatePtrToInt(V, ShadowTy);
2033  else
2034  return IRB.CreateBitCast(V, ShadowTy);
2035  }
2036 
2037  /// Propagate shadow for arbitrary operation.
2038  void handleShadowOr(Instruction &I) {
2039  IRBuilder<> IRB(&I);
2040  ShadowAndOriginCombiner SC(this, IRB);
2041  for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
2042  SC.Add(OI->get());
2043  SC.Done(&I);
2044  }
2045 
2046  // Handle multiplication by constant.
2047  //
2048  // Handle a special case of multiplication by constant that may have one or
2049  // more zeros in the lower bits. This makes corresponding number of lower bits
2050  // of the result zero as well. We model it by shifting the other operand
2051  // shadow left by the required number of bits. Effectively, we transform
2052  // (X * (A * 2**B)) to ((X << B) * A) and instrument (X << B) as (Sx << B).
2053  // We use multiplication by 2**N instead of shift to cover the case of
2054  // multiplication by 0, which may occur in some elements of a vector operand.
2055  void handleMulByConstant(BinaryOperator &I, Constant *ConstArg,
2056  Value *OtherArg) {
2057  Constant *ShadowMul;
2058  Type *Ty = ConstArg->getType();
2059  if (Ty->isVectorTy()) {
2060  unsigned NumElements = Ty->getVectorNumElements();
2061  Type *EltTy = Ty->getSequentialElementType();
2062  SmallVector<Constant *, 16> Elements;
2063  for (unsigned Idx = 0; Idx < NumElements; ++Idx) {
2064  if (ConstantInt *Elt =
2065  dyn_cast<ConstantInt>(ConstArg->getAggregateElement(Idx))) {
2066  const APInt &V = Elt->getValue();
2067  APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
2068  Elements.push_back(ConstantInt::get(EltTy, V2));
2069  } else {
2070  Elements.push_back(ConstantInt::get(EltTy, 1));
2071  }
2072  }
2073  ShadowMul = ConstantVector::get(Elements);
2074  } else {
2075  if (ConstantInt *Elt = dyn_cast<ConstantInt>(ConstArg)) {
2076  const APInt &V = Elt->getValue();
2077  APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
2078  ShadowMul = ConstantInt::get(Ty, V2);
2079  } else {
2080  ShadowMul = ConstantInt::get(Ty, 1);
2081  }
2082  }
2083 
2084  IRBuilder<> IRB(&I);
2085  setShadow(&I,
2086  IRB.CreateMul(getShadow(OtherArg), ShadowMul, "msprop_mul_cst"));
2087  setOrigin(&I, getOrigin(OtherArg));
2088  }
2089 
2090  void visitMul(BinaryOperator &I) {
2091  Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
2092  Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
2093  if (constOp0 && !constOp1)
2094  handleMulByConstant(I, constOp0, I.getOperand(1));
2095  else if (constOp1 && !constOp0)
2096  handleMulByConstant(I, constOp1, I.getOperand(0));
2097  else
2098  handleShadowOr(I);
2099  }
2100 
2101  void visitFAdd(BinaryOperator &I) { handleShadowOr(I); }
2102  void visitFSub(BinaryOperator &I) { handleShadowOr(I); }
2103  void visitFMul(BinaryOperator &I) { handleShadowOr(I); }
2104  void visitAdd(BinaryOperator &I) { handleShadowOr(I); }
2105  void visitSub(BinaryOperator &I) { handleShadowOr(I); }
2106  void visitXor(BinaryOperator &I) { handleShadowOr(I); }
2107 
2108  void handleIntegerDiv(Instruction &I) {
2109  IRBuilder<> IRB(&I);
2110  // Strict on the second argument.
2111  insertShadowCheck(I.getOperand(1), &I);
2112  setShadow(&I, getShadow(&I, 0));
2113  setOrigin(&I, getOrigin(&I, 0));
2114  }
2115 
2116  void visitUDiv(BinaryOperator &I) { handleIntegerDiv(I); }
2117  void visitSDiv(BinaryOperator &I) { handleIntegerDiv(I); }
2118  void visitURem(BinaryOperator &I) { handleIntegerDiv(I); }
2119  void visitSRem(BinaryOperator &I) { handleIntegerDiv(I); }
2120 
2121  // Floating point division is side-effect free. We can not require that the
2122  // divisor is fully initialized and must propagate shadow. See PR37523.
2123  void visitFDiv(BinaryOperator &I) { handleShadowOr(I); }
2124  void visitFRem(BinaryOperator &I) { handleShadowOr(I); }
2125 
2126  /// Instrument == and != comparisons.
2127  ///
2128  /// Sometimes the comparison result is known even if some of the bits of the
2129  /// arguments are not.
2130  void handleEqualityComparison(ICmpInst &I) {
2131  IRBuilder<> IRB(&I);
2132  Value *A = I.getOperand(0);
2133  Value *B = I.getOperand(1);
2134  Value *Sa = getShadow(A);
2135  Value *Sb = getShadow(B);
2136 
2137  // Get rid of pointers and vectors of pointers.
2138  // For ints (and vectors of ints), types of A and Sa match,
2139  // and this is a no-op.
2140  A = IRB.CreatePointerCast(A, Sa->getType());
2141  B = IRB.CreatePointerCast(B, Sb->getType());
2142 
2143  // A == B <==> (C = A^B) == 0
2144  // A != B <==> (C = A^B) != 0
2145  // Sc = Sa | Sb
2146  Value *C = IRB.CreateXor(A, B);
2147  Value *Sc = IRB.CreateOr(Sa, Sb);
2148  // Now dealing with i = (C == 0) comparison (or C != 0, does not matter now)
2149  // Result is defined if one of the following is true
2150  // * there is a defined 1 bit in C
2151  // * C is fully defined
2152  // Si = !(C & ~Sc) && Sc
2153  Value *Zero = Constant::getNullValue(Sc->getType());
2154  Value *MinusOne = Constant::getAllOnesValue(Sc->getType());
2155  Value *Si =
2156  IRB.CreateAnd(IRB.CreateICmpNE(Sc, Zero),
2157  IRB.CreateICmpEQ(
2158  IRB.CreateAnd(IRB.CreateXor(Sc, MinusOne), C), Zero));
2159  Si->setName("_msprop_icmp");
2160  setShadow(&I, Si);
2161  setOriginForNaryOp(I);
2162  }
2163 
2164  /// Build the lowest possible value of V, taking into account V's
2165  /// uninitialized bits.
2166  Value *getLowestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
2167  bool isSigned) {
2168  if (isSigned) {
2169  // Split shadow into sign bit and other bits.
2170  Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
2171  Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
2172  // Maximise the undefined shadow bit, minimize other undefined bits.
2173  return
2174  IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaOtherBits)), SaSignBit);
2175  } else {
2176  // Minimize undefined bits.
2177  return IRB.CreateAnd(A, IRB.CreateNot(Sa));
2178  }
2179  }
2180 
2181  /// Build the highest possible value of V, taking into account V's
2182  /// uninitialized bits.
2183  Value *getHighestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
2184  bool isSigned) {
2185  if (isSigned) {
2186  // Split shadow into sign bit and other bits.
2187  Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
2188  Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
2189  // Minimise the undefined shadow bit, maximise other undefined bits.
2190  return
2191  IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaSignBit)), SaOtherBits);
2192  } else {
2193  // Maximize undefined bits.
2194  return IRB.CreateOr(A, Sa);
2195  }
2196  }
2197 
2198  /// Instrument relational comparisons.
2199  ///
2200  /// This function does exact shadow propagation for all relational
2201  /// comparisons of integers, pointers and vectors of those.
2202  /// FIXME: output seems suboptimal when one of the operands is a constant
2203  void handleRelationalComparisonExact(ICmpInst &I) {
2204  IRBuilder<> IRB(&I);
2205  Value *A = I.getOperand(0);
2206  Value *B = I.getOperand(1);
2207  Value *Sa = getShadow(A);
2208  Value *Sb = getShadow(B);
2209 
2210  // Get rid of pointers and vectors of pointers.
2211  // For ints (and vectors of ints), types of A and Sa match,
2212  // and this is a no-op.
2213  A = IRB.CreatePointerCast(A, Sa->getType());
2214  B = IRB.CreatePointerCast(B, Sb->getType());
2215 
2216  // Let [a0, a1] be the interval of possible values of A, taking into account
2217  // its undefined bits. Let [b0, b1] be the interval of possible values of B.
2218  // Then (A cmp B) is defined iff (a0 cmp b1) == (a1 cmp b0).
2219  bool IsSigned = I.isSigned();
2220  Value *S1 = IRB.CreateICmp(I.getPredicate(),
2221  getLowestPossibleValue(IRB, A, Sa, IsSigned),
2222  getHighestPossibleValue(IRB, B, Sb, IsSigned));
2223  Value *S2 = IRB.CreateICmp(I.getPredicate(),
2224  getHighestPossibleValue(IRB, A, Sa, IsSigned),
2225  getLowestPossibleValue(IRB, B, Sb, IsSigned));
2226  Value *Si = IRB.CreateXor(S1, S2);
2227  setShadow(&I, Si);
2228  setOriginForNaryOp(I);
2229  }
2230 
2231  /// Instrument signed relational comparisons.
2232  ///
2233  /// Handle sign bit tests: x<0, x>=0, x<=-1, x>-1 by propagating the highest
2234  /// bit of the shadow. Everything else is delegated to handleShadowOr().
2235  void handleSignedRelationalComparison(ICmpInst &I) {
2236  Constant *constOp;
2237  Value *op = nullptr;
2238  CmpInst::Predicate pre;
2239  if ((constOp = dyn_cast<Constant>(I.getOperand(1)))) {
2240  op = I.getOperand(0);
2241  pre = I.getPredicate();
2242  } else if ((constOp = dyn_cast<Constant>(I.getOperand(0)))) {
2243  op = I.getOperand(1);
2244  pre = I.getSwappedPredicate();
2245  } else {
2246  handleShadowOr(I);
2247  return;
2248  }
2249 
2250  if ((constOp->isNullValue() &&
2251  (pre == CmpInst::ICMP_SLT || pre == CmpInst::ICMP_SGE)) ||
2252  (constOp->isAllOnesValue() &&
2253  (pre == CmpInst::ICMP_SGT || pre == CmpInst::ICMP_SLE))) {
2254  IRBuilder<> IRB(&I);
2255  Value *Shadow = IRB.CreateICmpSLT(getShadow(op), getCleanShadow(op),
2256  "_msprop_icmp_s");
2257  setShadow(&I, Shadow);
2258  setOrigin(&I, getOrigin(op));
2259  } else {
2260  handleShadowOr(I);
2261  }
2262  }
2263 
2264  void visitICmpInst(ICmpInst &I) {
2265  if (!ClHandleICmp) {
2266  handleShadowOr(I);
2267  return;
2268  }
2269  if (I.isEquality()) {
2270  handleEqualityComparison(I);
2271  return;
2272  }
2273 
2274  assert(I.isRelational());
2275  if (ClHandleICmpExact) {
2276  handleRelationalComparisonExact(I);
2277  return;
2278  }
2279  if (I.isSigned()) {
2280  handleSignedRelationalComparison(I);
2281  return;
2282  }
2283 
2284  assert(I.isUnsigned());
2285  if ((isa<Constant>(I.getOperand(0)) || isa<Constant>(I.getOperand(1)))) {
2286  handleRelationalComparisonExact(I);
2287  return;
2288  }
2289 
2290  handleShadowOr(I);
2291  }
2292 
2293  void visitFCmpInst(FCmpInst &I) {
2294  handleShadowOr(I);
2295  }
2296 
2297  void handleShift(BinaryOperator &I) {
2298  IRBuilder<> IRB(&I);
2299  // If any of the S2 bits are poisoned, the whole thing is poisoned.
2300  // Otherwise perform the same shift on S1.
2301  Value *S1 = getShadow(&I, 0);
2302  Value *S2 = getShadow(&I, 1);
2303  Value *S2Conv = IRB.CreateSExt(IRB.CreateICmpNE(S2, getCleanShadow(S2)),
2304  S2->getType());
2305  Value *V2 = I.getOperand(1);
2306  Value *Shift = IRB.CreateBinOp(I.getOpcode(), S1, V2);
2307  setShadow(&I, IRB.CreateOr(Shift, S2Conv));
2308  setOriginForNaryOp(I);
2309  }
2310 
2311  void visitShl(BinaryOperator &I) { handleShift(I); }
2312  void visitAShr(BinaryOperator &I) { handleShift(I); }
2313  void visitLShr(BinaryOperator &I) { handleShift(I); }
2314 
2315  /// Instrument llvm.memmove
2316  ///
2317  /// At this point we don't know if llvm.memmove will be inlined or not.
2318  /// If we don't instrument it and it gets inlined,
2319  /// our interceptor will not kick in and we will lose the memmove.
2320  /// If we instrument the call here, but it does not get inlined,
2321  /// we will memove the shadow twice: which is bad in case
2322  /// of overlapping regions. So, we simply lower the intrinsic to a call.
2323  ///
2324  /// Similar situation exists for memcpy and memset.
2325  void visitMemMoveInst(MemMoveInst &I) {
2326  IRBuilder<> IRB(&I);
2327  IRB.CreateCall(
2328  MS.MemmoveFn,
2329  {IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
2330  IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
2331  IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false)});
2332  I.eraseFromParent();
2333  }
2334 
2335  // Similar to memmove: avoid copying shadow twice.
2336  // This is somewhat unfortunate as it may slowdown small constant memcpys.
2337  // FIXME: consider doing manual inline for small constant sizes and proper
2338  // alignment.
2339  void visitMemCpyInst(MemCpyInst &I) {
2340  IRBuilder<> IRB(&I);
2341  IRB.CreateCall(
2342  MS.MemcpyFn,
2343  {IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
2344  IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
2345  IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false)});
2346  I.eraseFromParent();
2347  }
2348 
2349  // Same as memcpy.
2350  void visitMemSetInst(MemSetInst &I) {
2351  IRBuilder<> IRB(&I);
2352  IRB.CreateCall(
2353  MS.MemsetFn,
2354  {IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
2355  IRB.CreateIntCast(I.getArgOperand(1), IRB.getInt32Ty(), false),
2356  IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false)});
2357  I.eraseFromParent();
2358  }
2359 
2360  void visitVAStartInst(VAStartInst &I) {
2361  VAHelper->visitVAStartInst(I);
2362  }
2363 
2364  void visitVACopyInst(VACopyInst &I) {
2365  VAHelper->visitVACopyInst(I);
2366  }
2367 
2368  /// Handle vector store-like intrinsics.
2369  ///
2370  /// Instrument intrinsics that look like a simple SIMD store: writes memory,
2371  /// has 1 pointer argument and 1 vector argument, returns void.
2372  bool handleVectorStoreIntrinsic(IntrinsicInst &I) {
2373  IRBuilder<> IRB(&I);
2374  Value* Addr = I.getArgOperand(0);
2375  Value *Shadow = getShadow(&I, 1);
2376  Value *ShadowPtr, *OriginPtr;
2377 
2378  // We don't know the pointer alignment (could be unaligned SSE store!).
2379  // Have to assume to worst case.
2380  std::tie(ShadowPtr, OriginPtr) = getShadowOriginPtr(
2381  Addr, IRB, Shadow->getType(), /*Alignment*/ 1, /*isStore*/ true);
2382  IRB.CreateAlignedStore(Shadow, ShadowPtr, 1);
2383 
2385  insertShadowCheck(Addr, &I);
2386 
2387  // FIXME: factor out common code from materializeStores
2388  if (MS.TrackOrigins) IRB.CreateStore(getOrigin(&I, 1), OriginPtr);
2389  return true;
2390  }
2391 
2392  /// Handle vector load-like intrinsics.
2393  ///
2394  /// Instrument intrinsics that look like a simple SIMD load: reads memory,
2395  /// has 1 pointer argument, returns a vector.
2396  bool handleVectorLoadIntrinsic(IntrinsicInst &I) {
2397  IRBuilder<> IRB(&I);
2398  Value *Addr = I.getArgOperand(0);
2399 
2400  Type *ShadowTy = getShadowTy(&I);
2401  Value *ShadowPtr, *OriginPtr;
2402  if (PropagateShadow) {
2403  // We don't know the pointer alignment (could be unaligned SSE load!).
2404  // Have to assume to worst case.
2405  unsigned Alignment = 1;
2406  std::tie(ShadowPtr, OriginPtr) =
2407  getShadowOriginPtr(Addr, IRB, ShadowTy, Alignment, /*isStore*/ false);
2408  setShadow(&I, IRB.CreateAlignedLoad(ShadowPtr, Alignment, "_msld"));
2409  } else {
2410  setShadow(&I, getCleanShadow(&I));
2411  }
2412 
2414  insertShadowCheck(Addr, &I);
2415 
2416  if (MS.TrackOrigins) {
2417  if (PropagateShadow)
2418  setOrigin(&I, IRB.CreateLoad(OriginPtr));
2419  else
2420  setOrigin(&I, getCleanOrigin());
2421  }
2422  return true;
2423  }
2424 
2425  /// Handle (SIMD arithmetic)-like intrinsics.
2426  ///
2427  /// Instrument intrinsics with any number of arguments of the same type,
2428  /// equal to the return type. The type should be simple (no aggregates or
2429  /// pointers; vectors are fine).
2430  /// Caller guarantees that this intrinsic does not access memory.
2431  bool maybeHandleSimpleNomemIntrinsic(IntrinsicInst &I) {
2432  Type *RetTy = I.getType();
2433  if (!(RetTy->isIntOrIntVectorTy() ||
2434  RetTy->isFPOrFPVectorTy() ||
2435  RetTy->isX86_MMXTy()))
2436  return false;
2437 
2438  unsigned NumArgOperands = I.getNumArgOperands();
2439 
2440  for (unsigned i = 0; i < NumArgOperands; ++i) {
2441  Type *Ty = I.getArgOperand(i)->getType();
2442  if (Ty != RetTy)
2443  return false;
2444  }
2445 
2446  IRBuilder<> IRB(&I);
2447  ShadowAndOriginCombiner SC(this, IRB);
2448  for (unsigned i = 0; i < NumArgOperands; ++i)
2449  SC.Add(I.getArgOperand(i));
2450  SC.Done(&I);
2451 
2452  return true;
2453  }
2454 
2455  /// Heuristically instrument unknown intrinsics.
2456  ///
2457  /// The main purpose of this code is to do something reasonable with all
2458  /// random intrinsics we might encounter, most importantly - SIMD intrinsics.
2459  /// We recognize several classes of intrinsics by their argument types and
2460  /// ModRefBehaviour and apply special intrumentation when we are reasonably
2461  /// sure that we know what the intrinsic does.
2462  ///
2463  /// We special-case intrinsics where this approach fails. See llvm.bswap
2464  /// handling as an example of that.
2465  bool handleUnknownIntrinsic(IntrinsicInst &I) {
2466  unsigned NumArgOperands = I.getNumArgOperands();
2467  if (NumArgOperands == 0)
2468  return false;
2469 
2470  if (NumArgOperands == 2 &&
2471  I.getArgOperand(0)->getType()->isPointerTy() &&
2472  I.getArgOperand(1)->getType()->isVectorTy() &&
2473  I.getType()->isVoidTy() &&
2474  !I.onlyReadsMemory()) {
2475  // This looks like a vector store.
2476  return handleVectorStoreIntrinsic(I);
2477  }
2478 
2479  if (NumArgOperands == 1 &&
2480  I.getArgOperand(0)->getType()->isPointerTy() &&
2481  I.getType()->isVectorTy() &&
2482  I.onlyReadsMemory()) {
2483  // This looks like a vector load.
2484  return handleVectorLoadIntrinsic(I);
2485  }
2486 
2487  if (I.doesNotAccessMemory())
2488  if (maybeHandleSimpleNomemIntrinsic(I))
2489  return true;
2490 
2491  // FIXME: detect and handle SSE maskstore/maskload
2492  return false;
2493  }
2494 
2495  void handleBswap(IntrinsicInst &I) {
2496  IRBuilder<> IRB(&I);
2497  Value *Op = I.getArgOperand(0);
2498  Type *OpType = Op->getType();
2499  Function *BswapFunc = Intrinsic::getDeclaration(
2500  F.getParent(), Intrinsic::bswap, makeArrayRef(&OpType, 1));
2501  setShadow(&I, IRB.CreateCall(BswapFunc, getShadow(Op)));
2502  setOrigin(&I, getOrigin(Op));
2503  }
2504 
2505  // Instrument vector convert instrinsic.
2506  //
2507  // This function instruments intrinsics like cvtsi2ss:
2508  // %Out = int_xxx_cvtyyy(%ConvertOp)
2509  // or
2510  // %Out = int_xxx_cvtyyy(%CopyOp, %ConvertOp)
2511  // Intrinsic converts \p NumUsedElements elements of \p ConvertOp to the same
2512  // number \p Out elements, and (if has 2 arguments) copies the rest of the
2513  // elements from \p CopyOp.
2514  // In most cases conversion involves floating-point value which may trigger a
2515  // hardware exception when not fully initialized. For this reason we require
2516  // \p ConvertOp[0:NumUsedElements] to be fully initialized and trap otherwise.
2517  // We copy the shadow of \p CopyOp[NumUsedElements:] to \p
2518  // Out[NumUsedElements:]. This means that intrinsics without \p CopyOp always
2519  // return a fully initialized value.
2520  void handleVectorConvertIntrinsic(IntrinsicInst &I, int NumUsedElements) {
2521  IRBuilder<> IRB(&I);
2522  Value *CopyOp, *ConvertOp;
2523 
2524  switch (I.getNumArgOperands()) {
2525  case 3:
2526  assert(isa<ConstantInt>(I.getArgOperand(2)) && "Invalid rounding mode");
2528  case 2:
2529  CopyOp = I.getArgOperand(0);
2530  ConvertOp = I.getArgOperand(1);
2531  break;
2532  case 1:
2533  ConvertOp = I.getArgOperand(0);
2534  CopyOp = nullptr;
2535  break;
2536  default:
2537  llvm_unreachable("Cvt intrinsic with unsupported number of arguments.");
2538  }
2539 
2540  // The first *NumUsedElements* elements of ConvertOp are converted to the
2541  // same number of output elements. The rest of the output is copied from
2542  // CopyOp, or (if not available) filled with zeroes.
2543  // Combine shadow for elements of ConvertOp that are used in this operation,
2544  // and insert a check.
2545  // FIXME: consider propagating shadow of ConvertOp, at least in the case of
2546  // int->any conversion.
2547  Value *ConvertShadow = getShadow(ConvertOp);
2548  Value *AggShadow = nullptr;
2549  if (ConvertOp->getType()->isVectorTy()) {
2550  AggShadow = IRB.CreateExtractElement(
2551  ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), 0));
2552  for (int i = 1; i < NumUsedElements; ++i) {
2553  Value *MoreShadow = IRB.CreateExtractElement(
2554  ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), i));
2555  AggShadow = IRB.CreateOr(AggShadow, MoreShadow);
2556  }
2557  } else {
2558  AggShadow = ConvertShadow;
2559  }
2560  assert(AggShadow->getType()->isIntegerTy());
2561  insertShadowCheck(AggShadow, getOrigin(ConvertOp), &I);
2562 
2563  // Build result shadow by zero-filling parts of CopyOp shadow that come from
2564  // ConvertOp.
2565  if (CopyOp) {
2566  assert(CopyOp->getType() == I.getType());
2567  assert(CopyOp->getType()->isVectorTy());
2568  Value *ResultShadow = getShadow(CopyOp);
2569  Type *EltTy = ResultShadow->getType()->getVectorElementType();
2570  for (int i = 0; i < NumUsedElements; ++i) {
2571  ResultShadow = IRB.CreateInsertElement(
2572  ResultShadow, ConstantInt::getNullValue(EltTy),
2573  ConstantInt::get(IRB.getInt32Ty(), i));
2574  }
2575  setShadow(&I, ResultShadow);
2576  setOrigin(&I, getOrigin(CopyOp));
2577  } else {
2578  setShadow(&I, getCleanShadow(&I));
2579  setOrigin(&I, getCleanOrigin());
2580  }
2581  }
2582 
2583  // Given a scalar or vector, extract lower 64 bits (or less), and return all
2584  // zeroes if it is zero, and all ones otherwise.
2585  Value *Lower64ShadowExtend(IRBuilder<> &IRB, Value *S, Type *T) {
2586  if (S->getType()->isVectorTy())
2587  S = CreateShadowCast(IRB, S, IRB.getInt64Ty(), /* Signed */ true);
2588  assert(S->getType()->getPrimitiveSizeInBits() <= 64);
2589  Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
2590  return CreateShadowCast(IRB, S2, T, /* Signed */ true);
2591  }
2592 
2593  // Given a vector, extract its first element, and return all
2594  // zeroes if it is zero, and all ones otherwise.
2595  Value *LowerElementShadowExtend(IRBuilder<> &IRB, Value *S, Type *T) {
2596  Value *S1 = IRB.CreateExtractElement(S, (uint64_t)0);
2597  Value *S2 = IRB.CreateICmpNE(S1, getCleanShadow(S1));
2598  return CreateShadowCast(IRB, S2, T, /* Signed */ true);
2599  }
2600 
2601  Value *VariableShadowExtend(IRBuilder<> &IRB, Value *S) {
2602  Type *T = S->getType();
2603  assert(T->isVectorTy());
2604  Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
2605  return IRB.CreateSExt(S2, T);
2606  }
2607 
2608  // Instrument vector shift instrinsic.
2609  //
2610  // This function instruments intrinsics like int_x86_avx2_psll_w.
2611  // Intrinsic shifts %In by %ShiftSize bits.
2612  // %ShiftSize may be a vector. In that case the lower 64 bits determine shift
2613  // size, and the rest is ignored. Behavior is defined even if shift size is
2614  // greater than register (or field) width.
2615  void handleVectorShiftIntrinsic(IntrinsicInst &I, bool Variable) {
2616  assert(I.getNumArgOperands() == 2);
2617  IRBuilder<> IRB(&I);
2618  // If any of the S2 bits are poisoned, the whole thing is poisoned.
2619  // Otherwise perform the same shift on S1.
2620  Value *S1 = getShadow(&I, 0);
2621  Value *S2 = getShadow(&I, 1);
2622  Value *S2Conv = Variable ? VariableShadowExtend(IRB, S2)
2623  : Lower64ShadowExtend(IRB, S2, getShadowTy(&I));
2624  Value *V1 = I.getOperand(0);
2625  Value *V2 = I.getOperand(1);
2626  Value *Shift = IRB.CreateCall(I.getCalledValue(),
2627  {IRB.CreateBitCast(S1, V1->getType()), V2});
2628  Shift = IRB.CreateBitCast(Shift, getShadowTy(&I));
2629  setShadow(&I, IRB.CreateOr(Shift, S2Conv));
2630  setOriginForNaryOp(I);
2631  }
2632 
2633  // Get an X86_MMX-sized vector type.
2634  Type *getMMXVectorTy(unsigned EltSizeInBits) {
2635  const unsigned X86_MMXSizeInBits = 64;
2636  return VectorType::get(IntegerType::get(*MS.C, EltSizeInBits),
2637  X86_MMXSizeInBits / EltSizeInBits);
2638  }
2639 
2640  // Returns a signed counterpart for an (un)signed-saturate-and-pack
2641  // intrinsic.
2642  Intrinsic::ID getSignedPackIntrinsic(Intrinsic::ID id) {
2643  switch (id) {
2644  case Intrinsic::x86_sse2_packsswb_128:
2645  case Intrinsic::x86_sse2_packuswb_128:
2646  return Intrinsic::x86_sse2_packsswb_128;
2647 
2648  case Intrinsic::x86_sse2_packssdw_128:
2649  case Intrinsic::x86_sse41_packusdw:
2650  return Intrinsic::x86_sse2_packssdw_128;
2651 
2652  case Intrinsic::x86_avx2_packsswb:
2653  case Intrinsic::x86_avx2_packuswb:
2654  return Intrinsic::x86_avx2_packsswb;
2655 
2656  case Intrinsic::x86_avx2_packssdw:
2657  case Intrinsic::x86_avx2_packusdw:
2658  return Intrinsic::x86_avx2_packssdw;
2659 
2660  case Intrinsic::x86_mmx_packsswb:
2661  case Intrinsic::x86_mmx_packuswb:
2662  return Intrinsic::x86_mmx_packsswb;
2663 
2664  case Intrinsic::x86_mmx_packssdw:
2665  return Intrinsic::x86_mmx_packssdw;
2666  default:
2667  llvm_unreachable("unexpected intrinsic id");
2668  }
2669  }
2670 
2671  // Instrument vector pack instrinsic.
2672  //
2673  // This function instruments intrinsics like x86_mmx_packsswb, that
2674  // packs elements of 2 input vectors into half as many bits with saturation.
2675  // Shadow is propagated with the signed variant of the same intrinsic applied
2676  // to sext(Sa != zeroinitializer), sext(Sb != zeroinitializer).
2677  // EltSizeInBits is used only for x86mmx arguments.
2678  void handleVectorPackIntrinsic(IntrinsicInst &I, unsigned EltSizeInBits = 0) {
2679  assert(I.getNumArgOperands() == 2);
2680  bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2681  IRBuilder<> IRB(&I);
2682  Value *S1 = getShadow(&I, 0);
2683  Value *S2 = getShadow(&I, 1);
2684  assert(isX86_MMX || S1->getType()->isVectorTy());
2685 
2686  // SExt and ICmpNE below must apply to individual elements of input vectors.
2687  // In case of x86mmx arguments, cast them to appropriate vector types and
2688  // back.
2689  Type *T = isX86_MMX ? getMMXVectorTy(EltSizeInBits) : S1->getType();
2690  if (isX86_MMX) {
2691  S1 = IRB.CreateBitCast(S1, T);
2692  S2 = IRB.CreateBitCast(S2, T);
2693  }
2694  Value *S1_ext = IRB.CreateSExt(
2695  IRB.CreateICmpNE(S1, Constant::getNullValue(T)), T);
2696  Value *S2_ext = IRB.CreateSExt(
2697  IRB.CreateICmpNE(S2, Constant::getNullValue(T)), T);
2698  if (isX86_MMX) {
2699  Type *X86_MMXTy = Type::getX86_MMXTy(*MS.C);
2700  S1_ext = IRB.CreateBitCast(S1_ext, X86_MMXTy);
2701  S2_ext = IRB.CreateBitCast(S2_ext, X86_MMXTy);
2702  }
2703 
2704  Function *ShadowFn = Intrinsic::getDeclaration(
2705  F.getParent(), getSignedPackIntrinsic(I.getIntrinsicID()));
2706 
2707  Value *S =
2708  IRB.CreateCall(ShadowFn, {S1_ext, S2_ext}, "_msprop_vector_pack");
2709  if (isX86_MMX) S = IRB.CreateBitCast(S, getShadowTy(&I));
2710  setShadow(&I, S);
2711  setOriginForNaryOp(I);
2712  }
2713 
2714  // Instrument sum-of-absolute-differencies intrinsic.
2715  void handleVectorSadIntrinsic(IntrinsicInst &I) {
2716  const unsigned SignificantBitsPerResultElement = 16;
2717  bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2718  Type *ResTy = isX86_MMX ? IntegerType::get(*MS.C, 64) : I.getType();
2719  unsigned ZeroBitsPerResultElement =
2720  ResTy->getScalarSizeInBits() - SignificantBitsPerResultElement;
2721 
2722  IRBuilder<> IRB(&I);
2723  Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2724  S = IRB.CreateBitCast(S, ResTy);
2725  S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2726  ResTy);
2727  S = IRB.CreateLShr(S, ZeroBitsPerResultElement);
2728  S = IRB.CreateBitCast(S, getShadowTy(&I));
2729  setShadow(&I, S);
2730  setOriginForNaryOp(I);
2731  }
2732 
2733  // Instrument multiply-add intrinsic.
2734  void handleVectorPmaddIntrinsic(IntrinsicInst &I,
2735  unsigned EltSizeInBits = 0) {
2736  bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2737  Type *ResTy = isX86_MMX ? getMMXVectorTy(EltSizeInBits * 2) : I.getType();
2738  IRBuilder<> IRB(&I);
2739  Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2740  S = IRB.CreateBitCast(S, ResTy);
2741  S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2742  ResTy);
2743  S = IRB.CreateBitCast(S, getShadowTy(&I));
2744  setShadow(&I, S);
2745  setOriginForNaryOp(I);
2746  }
2747 
2748  // Instrument compare-packed intrinsic.
2749  // Basically, an or followed by sext(icmp ne 0) to end up with all-zeros or
2750  // all-ones shadow.
2751  void handleVectorComparePackedIntrinsic(IntrinsicInst &I) {
2752  IRBuilder<> IRB(&I);
2753  Type *ResTy = getShadowTy(&I);
2754  Value *S0 = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2755  Value *S = IRB.CreateSExt(
2756  IRB.CreateICmpNE(S0, Constant::getNullValue(ResTy)), ResTy);
2757  setShadow(&I, S);
2758  setOriginForNaryOp(I);
2759  }
2760 
2761  // Instrument compare-scalar intrinsic.
2762  // This handles both cmp* intrinsics which return the result in the first
2763  // element of a vector, and comi* which return the result as i32.
2764  void handleVectorCompareScalarIntrinsic(IntrinsicInst &I) {
2765  IRBuilder<> IRB(&I);
2766  Value *S0 = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2767  Value *S = LowerElementShadowExtend(IRB, S0, getShadowTy(&I));
2768  setShadow(&I, S);
2769  setOriginForNaryOp(I);
2770  }
2771 
2772  void handleStmxcsr(IntrinsicInst &I) {
2773  IRBuilder<> IRB(&I);
2774  Value* Addr = I.getArgOperand(0);
2775  Type *Ty = IRB.getInt32Ty();
2776  Value *ShadowPtr =
2777  getShadowOriginPtr(Addr, IRB, Ty, /*Alignment*/ 1, /*isStore*/ true)
2778  .first;
2779 
2780  IRB.CreateStore(getCleanShadow(Ty),
2781  IRB.CreatePointerCast(ShadowPtr, Ty->getPointerTo()));
2782 
2784  insertShadowCheck(Addr, &I);
2785  }
2786 
2787  void handleLdmxcsr(IntrinsicInst &I) {
2788  if (!InsertChecks) return;
2789 
2790  IRBuilder<> IRB(&I);
2791  Value *Addr = I.getArgOperand(0);
2792  Type *Ty = IRB.getInt32Ty();
2793  unsigned Alignment = 1;
2794  Value *ShadowPtr, *OriginPtr;
2795  std::tie(ShadowPtr, OriginPtr) =
2796  getShadowOriginPtr(Addr, IRB, Ty, Alignment, /*isStore*/ false);
2797 
2799  insertShadowCheck(Addr, &I);
2800 
2801  Value *Shadow = IRB.CreateAlignedLoad(ShadowPtr, Alignment, "_ldmxcsr");
2802  Value *Origin =
2803  MS.TrackOrigins ? IRB.CreateLoad(OriginPtr) : getCleanOrigin();
2804  insertShadowCheck(Shadow, Origin, &I);
2805  }
2806 
2807  void handleMaskedStore(IntrinsicInst &I) {
2808  IRBuilder<> IRB(&I);
2809  Value *V = I.getArgOperand(0);
2810  Value *Addr = I.getArgOperand(1);
2811  unsigned Align = cast<ConstantInt>(I.getArgOperand(2))->getZExtValue();
2812  Value *Mask = I.getArgOperand(3);
2813  Value *Shadow = getShadow(V);
2814 
2815  Value *ShadowPtr;
2816  Value *OriginPtr;
2817  std::tie(ShadowPtr, OriginPtr) = getShadowOriginPtr(
2818  Addr, IRB, Shadow->getType(), Align, /*isStore*/ true);
2819 
2820  if (ClCheckAccessAddress) {
2821  insertShadowCheck(Addr, &I);
2822  // Uninitialized mask is kind of like uninitialized address, but not as
2823  // scary.
2824  insertShadowCheck(Mask, &I);
2825  }
2826 
2827  IRB.CreateMaskedStore(Shadow, ShadowPtr, Align, Mask);
2828 
2829  if (MS.TrackOrigins) {
2830  auto &DL = F.getParent()->getDataLayout();
2831  paintOrigin(IRB, getOrigin(V), OriginPtr,
2832  DL.getTypeStoreSize(Shadow->getType()),
2833  std::max(Align, kMinOriginAlignment));
2834  }
2835  }
2836 
2837  bool handleMaskedLoad(IntrinsicInst &I) {
2838  IRBuilder<> IRB(&I);
2839  Value *Addr = I.getArgOperand(0);
2840  unsigned Align = cast<ConstantInt>(I.getArgOperand(1))->getZExtValue();
2841  Value *Mask = I.getArgOperand(2);
2842  Value *PassThru = I.getArgOperand(3);
2843 
2844  Type *ShadowTy = getShadowTy(&I);
2845  Value *ShadowPtr, *OriginPtr;
2846  if (PropagateShadow) {
2847  std::tie(ShadowPtr, OriginPtr) =
2848  getShadowOriginPtr(Addr, IRB, ShadowTy, Align, /*isStore*/ false);
2849  setShadow(&I, IRB.CreateMaskedLoad(ShadowPtr, Align, Mask,
2850  getShadow(PassThru), "_msmaskedld"));
2851  } else {
2852  setShadow(&I, getCleanShadow(&I));
2853  }
2854 
2855  if (ClCheckAccessAddress) {
2856  insertShadowCheck(Addr, &I);
2857  insertShadowCheck(Mask, &I);
2858  }
2859 
2860  if (MS.TrackOrigins) {
2861  if (PropagateShadow) {
2862  // Choose between PassThru's and the loaded value's origins.
2863  Value *MaskedPassThruShadow = IRB.CreateAnd(
2864  getShadow(PassThru), IRB.CreateSExt(IRB.CreateNeg(Mask), ShadowTy));
2865 
2866  Value *Acc = IRB.CreateExtractElement(
2867  MaskedPassThruShadow, ConstantInt::get(IRB.getInt32Ty(), 0));
2868  for (int i = 1, N = PassThru->getType()->getVectorNumElements(); i < N;
2869  ++i) {
2870  Value *More = IRB.CreateExtractElement(
2871  MaskedPassThruShadow, ConstantInt::get(IRB.getInt32Ty(), i));
2872  Acc = IRB.CreateOr(Acc, More);
2873  }
2874 
2875  Value *Origin = IRB.CreateSelect(
2876  IRB.CreateICmpNE(Acc, Constant::getNullValue(Acc->getType())),
2877  getOrigin(PassThru), IRB.CreateLoad(OriginPtr));
2878 
2879  setOrigin(&I, Origin);
2880  } else {
2881  setOrigin(&I, getCleanOrigin());
2882  }
2883  }
2884  return true;
2885  }
2886 
2887 
2888  void visitIntrinsicInst(IntrinsicInst &I) {
2889  switch (I.getIntrinsicID()) {
2890  case Intrinsic::bswap:
2891  handleBswap(I);
2892  break;
2893  case Intrinsic::masked_store:
2894  handleMaskedStore(I);
2895  break;
2896  case Intrinsic::masked_load:
2897  handleMaskedLoad(I);
2898  break;
2899  case Intrinsic::x86_sse_stmxcsr:
2900  handleStmxcsr(I);
2901  break;
2902  case Intrinsic::x86_sse_ldmxcsr:
2903  handleLdmxcsr(I);
2904  break;
2905  case Intrinsic::x86_avx512_vcvtsd2usi64:
2906  case Intrinsic::x86_avx512_vcvtsd2usi32:
2907  case Intrinsic::x86_avx512_vcvtss2usi64:
2908  case Intrinsic::x86_avx512_vcvtss2usi32:
2909  case Intrinsic::x86_avx512_cvttss2usi64:
2910  case Intrinsic::x86_avx512_cvttss2usi:
2911  case Intrinsic::x86_avx512_cvttsd2usi64:
2912  case Intrinsic::x86_avx512_cvttsd2usi:
2913  case Intrinsic::x86_avx512_cvtusi2ss:
2914  case Intrinsic::x86_avx512_cvtusi642sd:
2915  case Intrinsic::x86_avx512_cvtusi642ss:
2916  case Intrinsic::x86_sse2_cvtsd2si64:
2917  case Intrinsic::x86_sse2_cvtsd2si:
2918  case Intrinsic::x86_sse2_cvtsd2ss:
2919  case Intrinsic::x86_sse2_cvttsd2si64:
2920  case Intrinsic::x86_sse2_cvttsd2si:
2921  case Intrinsic::x86_sse_cvtss2si64:
2922  case Intrinsic::x86_sse_cvtss2si:
2923  case Intrinsic::x86_sse_cvttss2si64:
2924  case Intrinsic::x86_sse_cvttss2si:
2925  handleVectorConvertIntrinsic(I, 1);
2926  break;
2927  case Intrinsic::x86_sse_cvtps2pi:
2928  case Intrinsic::x86_sse_cvttps2pi:
2929  handleVectorConvertIntrinsic(I, 2);
2930  break;
2931 
2932  case Intrinsic::x86_avx512_psll_w_512:
2933  case Intrinsic::x86_avx512_psll_d_512:
2934  case Intrinsic::x86_avx512_psll_q_512:
2935  case Intrinsic::x86_avx512_pslli_w_512:
2936  case Intrinsic::x86_avx512_pslli_d_512:
2937  case Intrinsic::x86_avx512_pslli_q_512:
2938  case Intrinsic::x86_avx512_psrl_w_512:
2939  case Intrinsic::x86_avx512_psrl_d_512:
2940  case Intrinsic::x86_avx512_psrl_q_512:
2941  case Intrinsic::x86_avx512_psra_w_512:
2942  case Intrinsic::x86_avx512_psra_d_512:
2943  case Intrinsic::x86_avx512_psra_q_512:
2944  case Intrinsic::x86_avx512_psrli_w_512:
2945  case Intrinsic::x86_avx512_psrli_d_512:
2946  case Intrinsic::x86_avx512_psrli_q_512:
2947  case Intrinsic::x86_avx512_psrai_w_512:
2948  case Intrinsic::x86_avx512_psrai_d_512:
2949  case Intrinsic::x86_avx512_psrai_q_512:
2950  case Intrinsic::x86_avx512_psra_q_256:
2951  case Intrinsic::x86_avx512_psra_q_128:
2952  case Intrinsic::x86_avx512_psrai_q_256:
2953  case Intrinsic::x86_avx512_psrai_q_128:
2954  case Intrinsic::x86_avx2_psll_w:
2955  case Intrinsic::x86_avx2_psll_d:
2956  case Intrinsic::x86_avx2_psll_q:
2957  case Intrinsic::x86_avx2_pslli_w:
2958  case Intrinsic::x86_avx2_pslli_d:
2959  case Intrinsic::x86_avx2_pslli_q:
2960  case Intrinsic::x86_avx2_psrl_w:
2961  case Intrinsic::x86_avx2_psrl_d:
2962  case Intrinsic::x86_avx2_psrl_q:
2963  case Intrinsic::x86_avx2_psra_w:
2964  case Intrinsic::x86_avx2_psra_d:
2965  case Intrinsic::x86_avx2_psrli_w:
2966  case Intrinsic::x86_avx2_psrli_d:
2967  case Intrinsic::x86_avx2_psrli_q:
2968  case Intrinsic::x86_avx2_psrai_w:
2969  case Intrinsic::x86_avx2_psrai_d:
2970  case Intrinsic::x86_sse2_psll_w:
2971  case Intrinsic::x86_sse2_psll_d:
2972  case Intrinsic::x86_sse2_psll_q:
2973  case Intrinsic::x86_sse2_pslli_w:
2974  case Intrinsic::x86_sse2_pslli_d:
2975  case Intrinsic::x86_sse2_pslli_q:
2976  case Intrinsic::x86_sse2_psrl_w:
2977  case Intrinsic::x86_sse2_psrl_d:
2978  case Intrinsic::x86_sse2_psrl_q:
2979  case Intrinsic::x86_sse2_psra_w:
2980  case Intrinsic::x86_sse2_psra_d:
2981  case Intrinsic::x86_sse2_psrli_w:
2982  case Intrinsic::x86_sse2_psrli_d:
2983  case Intrinsic::x86_sse2_psrli_q:
2984  case Intrinsic::x86_sse2_psrai_w:
2985  case Intrinsic::x86_sse2_psrai_d:
2986  case Intrinsic::x86_mmx_psll_w:
2987  case Intrinsic::x86_mmx_psll_d:
2988  case Intrinsic::x86_mmx_psll_q:
2989  case Intrinsic::x86_mmx_pslli_w:
2990  case Intrinsic::x86_mmx_pslli_d:
2991  case Intrinsic::x86_mmx_pslli_q:
2992  case Intrinsic::x86_mmx_psrl_w:
2993  case Intrinsic::x86_mmx_psrl_d:
2994  case Intrinsic::x86_mmx_psrl_q:
2995  case Intrinsic::x86_mmx_psra_w:
2996  case Intrinsic::x86_mmx_psra_d:
2997  case Intrinsic::x86_mmx_psrli_w:
2998  case Intrinsic::x86_mmx_psrli_d:
2999  case Intrinsic::x86_mmx_psrli_q:
3000  case Intrinsic::x86_mmx_psrai_w:
3001  case Intrinsic::x86_mmx_psrai_d:
3002  handleVectorShiftIntrinsic(I, /* Variable */ false);
3003  break;
3004  case Intrinsic::x86_avx2_psllv_d:
3005  case Intrinsic::x86_avx2_psllv_d_256:
3006  case Intrinsic::x86_avx512_psllv_d_512:
3007  case Intrinsic::x86_avx2_psllv_q:
3008  case Intrinsic::x86_avx2_psllv_q_256:
3009  case Intrinsic::x86_avx512_psllv_q_512:
3010  case Intrinsic::x86_avx2_psrlv_d:
3011  case Intrinsic::x86_avx2_psrlv_d_256:
3012  case Intrinsic::x86_avx512_psrlv_d_512:
3013  case Intrinsic::x86_avx2_psrlv_q:
3014  case Intrinsic::x86_avx2_psrlv_q_256:
3015  case Intrinsic::x86_avx512_psrlv_q_512:
3016  case Intrinsic::x86_avx2_psrav_d:
3017  case Intrinsic::x86_avx2_psrav_d_256:
3018  case Intrinsic::x86_avx512_psrav_d_512:
3019  case Intrinsic::x86_avx512_psrav_q_128:
3020  case Intrinsic::x86_avx512_psrav_q_256:
3021  case Intrinsic::x86_avx512_psrav_q_512:
3022  handleVectorShiftIntrinsic(I, /* Variable */ true);
3023  break;
3024 
3025  case Intrinsic::x86_sse2_packsswb_128:
3026  case Intrinsic::x86_sse2_packssdw_128:
3027  case Intrinsic::x86_sse2_packuswb_128:
3028  case Intrinsic::x86_sse41_packusdw:
3029  case Intrinsic::x86_avx2_packsswb:
3030  case Intrinsic::x86_avx2_packssdw:
3031  case Intrinsic::x86_avx2_packuswb:
3032  case Intrinsic::x86_avx2_packusdw:
3033  handleVectorPackIntrinsic(I);
3034  break;
3035 
3036  case Intrinsic::x86_mmx_packsswb:
3037  case Intrinsic::x86_mmx_packuswb:
3038  handleVectorPackIntrinsic(I, 16);
3039  break;
3040 
3041  case Intrinsic::x86_mmx_packssdw:
3042  handleVectorPackIntrinsic(I, 32);
3043  break;
3044 
3045  case Intrinsic::x86_mmx_psad_bw:
3046  case Intrinsic::x86_sse2_psad_bw:
3047  case Intrinsic::x86_avx2_psad_bw:
3048  handleVectorSadIntrinsic(I);
3049  break;
3050 
3051  case Intrinsic::x86_sse2_pmadd_wd:
3052  case Intrinsic::x86_avx2_pmadd_wd:
3053  case Intrinsic::x86_ssse3_pmadd_ub_sw_128:
3054  case Intrinsic::x86_avx2_pmadd_ub_sw:
3055  handleVectorPmaddIntrinsic(I);
3056  break;
3057 
3058  case Intrinsic::x86_ssse3_pmadd_ub_sw:
3059  handleVectorPmaddIntrinsic(I, 8);
3060  break;
3061 
3062  case Intrinsic::x86_mmx_pmadd_wd:
3063  handleVectorPmaddIntrinsic(I, 16);
3064  break;
3065 
3066  case Intrinsic::x86_sse_cmp_ss:
3067  case Intrinsic::x86_sse2_cmp_sd:
3068  case Intrinsic::x86_sse_comieq_ss:
3069  case Intrinsic::x86_sse_comilt_ss:
3070  case Intrinsic::x86_sse_comile_ss:
3071  case Intrinsic::x86_sse_comigt_ss:
3072  case Intrinsic::x86_sse_comige_ss:
3073  case Intrinsic::x86_sse_comineq_ss:
3074  case Intrinsic::x86_sse_ucomieq_ss:
3075  case Intrinsic::x86_sse_ucomilt_ss:
3076  case Intrinsic::x86_sse_ucomile_ss:
3077  case Intrinsic::x86_sse_ucomigt_ss:
3078  case Intrinsic::x86_sse_ucomige_ss:
3079  case Intrinsic::x86_sse_ucomineq_ss:
3080  case Intrinsic::x86_sse2_comieq_sd:
3081  case Intrinsic::x86_sse2_comilt_sd:
3082  case Intrinsic::x86_sse2_comile_sd:
3083  case Intrinsic::x86_sse2_comigt_sd:
3084  case Intrinsic::x86_sse2_comige_sd:
3085  case Intrinsic::x86_sse2_comineq_sd:
3086  case Intrinsic::x86_sse2_ucomieq_sd:
3087  case Intrinsic::x86_sse2_ucomilt_sd:
3088  case Intrinsic::x86_sse2_ucomile_sd:
3089  case Intrinsic::x86_sse2_ucomigt_sd:
3090  case Intrinsic::x86_sse2_ucomige_sd:
3091  case Intrinsic::x86_sse2_ucomineq_sd:
3092  handleVectorCompareScalarIntrinsic(I);
3093  break;
3094 
3095  case Intrinsic::x86_sse_cmp_ps:
3096  case Intrinsic::x86_sse2_cmp_pd:
3097  // FIXME: For x86_avx_cmp_pd_256 and x86_avx_cmp_ps_256 this function
3098  // generates reasonably looking IR that fails in the backend with "Do not
3099  // know how to split the result of this operator!".
3100  handleVectorComparePackedIntrinsic(I);
3101  break;
3102 
3103  default:
3104  if (!handleUnknownIntrinsic(I))
3105  visitInstruction(I);
3106  break;
3107  }
3108  }
3109 
3110  void visitCallSite(CallSite CS) {
3111  Instruction &I = *CS.getInstruction();
3112  assert(!I.getMetadata("nosanitize"));
3113  assert((CS.isCall() || CS.isInvoke()) && "Unknown type of CallSite");
3114  if (CS.isCall()) {
3115  CallInst *Call = cast<CallInst>(&I);
3116 
3117  // For inline asm, do the usual thing: check argument shadow and mark all
3118  // outputs as clean. Note that any side effects of the inline asm that are
3119  // not immediately visible in its constraints are not handled.
3120  if (Call->isInlineAsm()) {
3122  visitAsmInstruction(I);
3123  else
3124  visitInstruction(I);
3125  return;
3126  }
3127 
3128  assert(!isa<IntrinsicInst>(&I) && "intrinsics are handled elsewhere");
3129 
3130  // We are going to insert code that relies on the fact that the callee
3131  // will become a non-readonly function after it is instrumented by us. To
3132  // prevent this code from being optimized out, mark that function
3133  // non-readonly in advance.
3134  if (Function *Func = Call->getCalledFunction()) {
3135  // Clear out readonly/readnone attributes.
3136  AttrBuilder B;
3137  B.addAttribute(Attribute::ReadOnly)
3138  .addAttribute(Attribute::ReadNone);
3140  }
3141 
3143  }
3144  IRBuilder<> IRB(&I);
3145 
3146  unsigned ArgOffset = 0;
3147  LLVM_DEBUG(dbgs() << " CallSite: " << I << "\n");
3148  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
3149  ArgIt != End; ++ArgIt) {
3150  Value *A = *ArgIt;
3151  unsigned i = ArgIt - CS.arg_begin();
3152  if (!A->getType()->isSized()) {
3153  LLVM_DEBUG(dbgs() << "Arg " << i << " is not sized: " << I << "\n");
3154  continue;
3155  }
3156  unsigned Size = 0;
3157  Value *Store = nullptr;
3158  // Compute the Shadow for arg even if it is ByVal, because
3159  // in that case getShadow() will copy the actual arg shadow to
3160  // __msan_param_tls.
3161  Value *ArgShadow = getShadow(A);
3162  Value *ArgShadowBase = getShadowPtrForArgument(A, IRB, ArgOffset);
3163  LLVM_DEBUG(dbgs() << " Arg#" << i << ": " << *A
3164  << " Shadow: " << *ArgShadow << "\n");
3165  bool ArgIsInitialized = false;
3166  const DataLayout &DL = F.getParent()->getDataLayout();
3167  if (CS.paramHasAttr(i, Attribute::ByVal)) {
3168  assert(A->getType()->isPointerTy() &&
3169  "ByVal argument is not a pointer!");
3170  Size = DL.getTypeAllocSize(A->getType()->getPointerElementType());
3171  if (ArgOffset + Size > kParamTLSSize) break;
3172  unsigned ParamAlignment = CS.getParamAlignment(i);
3173  unsigned Alignment = std::min(ParamAlignment, kShadowTLSAlignment);
3174  Value *AShadowPtr =
3175  getShadowOriginPtr(A, IRB, IRB.getInt8Ty(), Alignment,
3176  /*isStore*/ false)
3177  .first;
3178 
3179  Store = IRB.CreateMemCpy(ArgShadowBase, Alignment, AShadowPtr,
3180  Alignment, Size);
3181  // TODO(glider): need to copy origins.
3182  } else {
3183  Size = DL.getTypeAllocSize(A->getType());
3184  if (ArgOffset + Size > kParamTLSSize) break;
3185  Store = IRB.CreateAlignedStore(ArgShadow, ArgShadowBase,
3186  kShadowTLSAlignment);
3187  Constant *Cst = dyn_cast<Constant>(ArgShadow);
3188  if (Cst && Cst->isNullValue()) ArgIsInitialized = true;
3189  }
3190  if (MS.TrackOrigins && !ArgIsInitialized)
3191  IRB.CreateStore(getOrigin(A),
3192  getOriginPtrForArgument(A, IRB, ArgOffset));
3193  (void)Store;
3194  assert(Size != 0 && Store != nullptr);
3195  LLVM_DEBUG(dbgs() << " Param:" << *Store << "\n");
3196  ArgOffset += alignTo(Size, 8);
3197  }
3198  LLVM_DEBUG(dbgs() << " done with call args\n");
3199 
3200  FunctionType *FT =
3201  cast<FunctionType>(CS.getCalledValue()->getType()->getContainedType(0));
3202  if (FT->isVarArg()) {
3203  VAHelper->visitCallSite(CS, IRB);
3204  }
3205 
3206  // Now, get the shadow for the RetVal.
3207  if (!I.getType()->isSized()) return;
3208  // Don't emit the epilogue for musttail call returns.
3209  if (CS.isCall() && cast<CallInst>(&I)->isMustTailCall()) return;
3210  IRBuilder<> IRBBefore(&I);
3211  // Until we have full dynamic coverage, make sure the retval shadow is 0.
3212  Value *Base = getShadowPtrForRetval(&I, IRBBefore);
3213  IRBBefore.CreateAlignedStore(getCleanShadow(&I), Base, kShadowTLSAlignment);
3214  BasicBlock::iterator NextInsn;
3215  if (CS.isCall()) {
3216  NextInsn = ++I.getIterator();
3217  assert(NextInsn != I.getParent()->end());
3218  } else {
3219  BasicBlock *NormalDest = cast<InvokeInst>(&I)->getNormalDest();
3220  if (!NormalDest->getSinglePredecessor()) {
3221  // FIXME: this case is tricky, so we are just conservative here.
3222  // Perhaps we need to split the edge between this BB and NormalDest,
3223  // but a naive attempt to use SplitEdge leads to a crash.
3224  setShadow(&I, getCleanShadow(&I));
3225  setOrigin(&I, getCleanOrigin());
3226  return;
3227  }
3228  // FIXME: NextInsn is likely in a basic block that has not been visited yet.
3229  // Anything inserted there will be instrumented by MSan later!
3230  NextInsn = NormalDest->getFirstInsertionPt();
3231  assert(NextInsn != NormalDest->end() &&
3232  "Could not find insertion point for retval shadow load");
3233  }
3234  IRBuilder<> IRBAfter(&*NextInsn);
3235  Value *RetvalShadow =
3236  IRBAfter.CreateAlignedLoad(getShadowPtrForRetval(&I, IRBAfter),
3237  kShadowTLSAlignment, "_msret");
3238  setShadow(&I, RetvalShadow);
3239  if (MS.TrackOrigins)
3240  setOrigin(&I, IRBAfter.CreateLoad(getOriginPtrForRetval(IRBAfter)));
3241  }
3242 
3243  bool isAMustTailRetVal(Value *RetVal) {
3244  if (auto *I = dyn_cast<BitCastInst>(RetVal)) {
3245  RetVal = I->getOperand(0);
3246  }
3247  if (auto *I = dyn_cast<CallInst>(RetVal)) {
3248  return I->isMustTailCall();
3249  }
3250  return false;
3251  }
3252 
3253  void visitReturnInst(ReturnInst &I) {
3254  IRBuilder<> IRB(&I);
3255  Value *RetVal = I.getReturnValue();
3256  if (!RetVal) return;
3257  // Don't emit the epilogue for musttail call returns.
3258  if (isAMustTailRetVal(RetVal)) return;
3259  Value *ShadowPtr = getShadowPtrForRetval(RetVal, IRB);
3260  if (CheckReturnValue) {
3261  insertShadowCheck(RetVal, &I);
3262  Value *Shadow = getCleanShadow(RetVal);
3263  IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
3264  } else {
3265  Value *Shadow = getShadow(RetVal);
3266  IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
3267  if (MS.TrackOrigins)
3268  IRB.CreateStore(getOrigin(RetVal), getOriginPtrForRetval(IRB));
3269  }
3270  }
3271 
3272  void visitPHINode(PHINode &I) {
3273  IRBuilder<> IRB(&I);
3274  if (!PropagateShadow) {
3275  setShadow(&I, getCleanShadow(&I));
3276  setOrigin(&I, getCleanOrigin());
3277  return;
3278  }
3279 
3280  ShadowPHINodes.push_back(&I);
3281  setShadow(&I, IRB.CreatePHI(getShadowTy(&I), I.getNumIncomingValues(),
3282  "_msphi_s"));
3283  if (MS.TrackOrigins)
3284  setOrigin(&I, IRB.CreatePHI(MS.OriginTy, I.getNumIncomingValues(),
3285  "_msphi_o"));
3286  }
3287 
3288  Value *getLocalVarDescription(AllocaInst &I) {
3289  SmallString<2048> StackDescriptionStorage;
3290  raw_svector_ostream StackDescription(StackDescriptionStorage);
3291  // We create a string with a description of the stack allocation and
3292  // pass it into __msan_set_alloca_origin.
3293  // It will be printed by the run-time if stack-originated UMR is found.
3294  // The first 4 bytes of the string are set to '----' and will be replaced
3295  // by __msan_va_arg_overflow_size_tls at the first call.
3296  StackDescription << "----" << I.getName() << "@" << F.getName();
3298  StackDescription.str());
3299  }
3300 
3301  void instrumentAllocaUserspace(AllocaInst &I, IRBuilder<> &IRB, Value *Len) {
3302  if (PoisonStack && ClPoisonStackWithCall) {
3303  IRB.CreateCall(MS.MsanPoisonStackFn,
3304  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len});
3305  } else {
3306  Value *ShadowBase, *OriginBase;
3307  std::tie(ShadowBase, OriginBase) =
3308  getShadowOriginPtr(&I, IRB, IRB.getInt8Ty(), 1, /*isStore*/ true);
3309 
3310  Value *PoisonValue = IRB.getInt8(PoisonStack ? ClPoisonStackPattern : 0);
3311  IRB.CreateMemSet(ShadowBase, PoisonValue, Len, I.getAlignment());
3312  }
3313 
3314  if (PoisonStack && MS.TrackOrigins) {
3315  Value *Descr = getLocalVarDescription(I);
3316  IRB.CreateCall(MS.MsanSetAllocaOrigin4Fn,
3317  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len,
3318  IRB.CreatePointerCast(Descr, IRB.getInt8PtrTy()),
3319  IRB.CreatePointerCast(&F, MS.IntptrTy)});
3320  }
3321  }
3322 
3323  void instrumentAllocaKmsan(AllocaInst &I, IRBuilder<> &IRB, Value *Len) {
3324  Value *Descr = getLocalVarDescription(I);
3325  if (PoisonStack) {
3326  IRB.CreateCall(MS.MsanPoisonAllocaFn,
3327  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len,
3328  IRB.CreatePointerCast(Descr, IRB.getInt8PtrTy())});
3329  } else {
3330  IRB.CreateCall(MS.MsanUnpoisonAllocaFn,
3331  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len});
3332  }
3333  }
3334 
3335  void visitAllocaInst(AllocaInst &I) {
3336  setShadow(&I, getCleanShadow(&I));
3337  setOrigin(&I, getCleanOrigin());
3338  IRBuilder<> IRB(I.getNextNode());
3339  const DataLayout &DL = F.getParent()->getDataLayout();
3340  uint64_t TypeSize = DL.getTypeAllocSize(I.getAllocatedType());
3341  Value *Len = ConstantInt::get(MS.IntptrTy, TypeSize);
3342  if (I.isArrayAllocation())
3343  Len = IRB.CreateMul(Len, I.getArraySize());
3344 
3345  if (MS.CompileKernel)
3346  instrumentAllocaKmsan(I, IRB, Len);
3347  else
3348  instrumentAllocaUserspace(I, IRB, Len);
3349  }
3350 
3351  void visitSelectInst(SelectInst& I) {
3352  IRBuilder<> IRB(&I);
3353  // a = select b, c, d
3354  Value *B = I.getCondition();
3355  Value *C = I.getTrueValue();
3356  Value *D = I.getFalseValue();
3357  Value *Sb = getShadow(B);
3358  Value *Sc = getShadow(C);
3359  Value *Sd = getShadow(D);
3360 
3361  // Result shadow if condition shadow is 0.
3362  Value *Sa0 = IRB.CreateSelect(B, Sc, Sd);
3363  Value *Sa1;
3364  if (I.getType()->isAggregateType()) {
3365  // To avoid "sign extending" i1 to an arbitrary aggregate type, we just do
3366  // an extra "select". This results in much more compact IR.
3367  // Sa = select Sb, poisoned, (select b, Sc, Sd)
3368  Sa1 = getPoisonedShadow(getShadowTy(I.getType()));
3369  } else {
3370  // Sa = select Sb, [ (c^d) | Sc | Sd ], [ b ? Sc : Sd ]
3371  // If Sb (condition is poisoned), look for bits in c and d that are equal
3372  // and both unpoisoned.
3373  // If !Sb (condition is unpoisoned), simply pick one of Sc and Sd.
3374 
3375  // Cast arguments to shadow-compatible type.
3376  C = CreateAppToShadowCast(IRB, C);
3377  D = CreateAppToShadowCast(IRB, D);
3378 
3379  // Result shadow if condition shadow is 1.
3380  Sa1 = IRB.CreateOr(IRB.CreateXor(C, D), IRB.CreateOr(Sc, Sd));
3381  }
3382  Value *Sa = IRB.CreateSelect(Sb, Sa1, Sa0, "_msprop_select");
3383  setShadow(&I, Sa);
3384  if (MS.TrackOrigins) {
3385  // Origins are always i32, so any vector conditions must be flattened.
3386  // FIXME: consider tracking vector origins for app vectors?
3387  if (B->getType()->isVectorTy()) {
3388  Type *FlatTy = getShadowTyNoVec(B->getType());
3389  B = IRB.CreateICmpNE(IRB.CreateBitCast(B, FlatTy),
3390  ConstantInt::getNullValue(FlatTy));
3391  Sb = IRB.CreateICmpNE(IRB.CreateBitCast(Sb, FlatTy),
3392  ConstantInt::getNullValue(FlatTy));
3393  }
3394  // a = select b, c, d
3395  // Oa = Sb ? Ob : (b ? Oc : Od)
3396  setOrigin(
3397  &I, IRB.CreateSelect(Sb, getOrigin(I.getCondition()),
3398  IRB.CreateSelect(B, getOrigin(I.getTrueValue()),
3399  getOrigin(I.getFalseValue()))));
3400  }
3401  }
3402 
3403  void visitLandingPadInst(LandingPadInst &I) {
3404  // Do nothing.
3405  // See https://github.com/google/sanitizers/issues/504
3406  setShadow(&I, getCleanShadow(&I));
3407  setOrigin(&I, getCleanOrigin());
3408  }
3409 
3410  void visitCatchSwitchInst(CatchSwitchInst &I) {
3411  setShadow(&I, getCleanShadow(&I));
3412  setOrigin(&I, getCleanOrigin());
3413  }
3414 
3415  void visitFuncletPadInst(FuncletPadInst &I) {
3416  setShadow(&I, getCleanShadow(&I));
3417  setOrigin(&I, getCleanOrigin());
3418  }
3419 
3420  void visitGetElementPtrInst(GetElementPtrInst &I) {
3421  handleShadowOr(I);
3422  }
3423 
3424  void visitExtractValueInst(ExtractValueInst &I) {
3425  IRBuilder<> IRB(&I);
3426  Value *Agg = I.getAggregateOperand();
3427  LLVM_DEBUG(dbgs() << "ExtractValue: " << I << "\n");
3428  Value *AggShadow = getShadow(Agg);
3429  LLVM_DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
3430  Value *ResShadow = IRB.CreateExtractValue(AggShadow, I.getIndices());
3431  LLVM_DEBUG(dbgs() << " ResShadow: " << *ResShadow << "\n");
3432  setShadow(&I, ResShadow);
3433  setOriginForNaryOp(I);
3434  }
3435 
3436  void visitInsertValueInst(InsertValueInst &I) {
3437  IRBuilder<> IRB(&I);
3438  LLVM_DEBUG(dbgs() << "InsertValue: " << I << "\n");
3439  Value *AggShadow = getShadow(I.getAggregateOperand());
3440  Value *InsShadow = getShadow(I.getInsertedValueOperand());
3441  LLVM_DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
3442  LLVM_DEBUG(dbgs() << " InsShadow: " << *InsShadow << "\n");
3443  Value *Res = IRB.CreateInsertValue(AggShadow, InsShadow, I.getIndices());
3444  LLVM_DEBUG(dbgs() << " Res: " << *Res << "\n");
3445  setShadow(&I, Res);
3446  setOriginForNaryOp(I);
3447  }
3448 
3449  void dumpInst(Instruction &I) {
3450  if (CallInst *CI = dyn_cast<CallInst>(&I)) {
3451  errs() << "ZZZ call " << CI->getCalledFunction()->getName() << "\n";
3452  } else {
3453  errs() << "ZZZ " << I.getOpcodeName() << "\n";
3454  }
3455  errs() << "QQQ " << I << "\n";
3456  }
3457 
3458  void visitResumeInst(ResumeInst &I) {
3459  LLVM_DEBUG(dbgs() << "Resume: " << I << "\n");
3460  // Nothing to do here.
3461  }
3462 
3463  void visitCleanupReturnInst(CleanupReturnInst &CRI) {
3464  LLVM_DEBUG(dbgs() << "CleanupReturn: " << CRI << "\n");
3465  // Nothing to do here.
3466  }
3467 
3468  void visitCatchReturnInst(CatchReturnInst &CRI) {
3469  LLVM_DEBUG(dbgs() << "CatchReturn: " << CRI << "\n");
3470  // Nothing to do here.
3471  }
3472 
3473  void instrumentAsmArgument(Value *Operand, Instruction &I, IRBuilder<> &IRB,
3474  const DataLayout &DL, bool isOutput) {
3475  // For each assembly argument, we check its value for being initialized.
3476  // If the argument is a pointer, we assume it points to a single element
3477  // of the corresponding type (or to a 8-byte word, if the type is unsized).
3478  // Each such pointer is instrumented with a call to the runtime library.
3479  Type *OpType = Operand->getType();
3480  // Check the operand value itself.
3481  insertShadowCheck(Operand, &I);
3482  if (!OpType->isPointerTy()) {
3483  assert(!isOutput);
3484  return;
3485  }
3486  Value *Hook =
3487  isOutput ? MS.MsanInstrumentAsmStoreFn : MS.MsanInstrumentAsmLoadFn;
3488  Type *ElType = OpType->getPointerElementType();
3489  if (!ElType->isSized())
3490  return;
3491  int Size = DL.getTypeStoreSize(ElType);
3492  Value *Ptr = IRB.CreatePointerCast(Operand, IRB.getInt8PtrTy());
3493  Value *SizeVal = ConstantInt::get(MS.IntptrTy, Size);
3494  IRB.CreateCall(Hook, {Ptr, SizeVal});
3495  }
3496 
3497  /// Get the number of output arguments returned by pointers.
3498  int getNumOutputArgs(InlineAsm *IA, CallInst *CI) {
3499  int NumRetOutputs = 0;
3500  int NumOutputs = 0;
3501  Type *RetTy = dyn_cast<Value>(CI)->getType();
3502  if (!RetTy->isVoidTy()) {
3503  // Register outputs are returned via the CallInst return value.
3504  StructType *ST = dyn_cast_or_null<StructType>(RetTy);
3505  if (ST)
3506  NumRetOutputs = ST->getNumElements();
3507  else
3508  NumRetOutputs = 1;
3509  }
3511  for (size_t i = 0, n = Constraints.size(); i < n; i++) {
3512  InlineAsm::ConstraintInfo Info = Constraints[i];
3513  switch (Info.Type) {
3514  case InlineAsm::isOutput:
3515  NumOutputs++;
3516  break;
3517  default:
3518  break;
3519  }
3520  }
3521  return NumOutputs - NumRetOutputs;
3522  }
3523 
3524  void visitAsmInstruction(Instruction &I) {
3525  // Conservative inline assembly handling: check for poisoned shadow of
3526  // asm() arguments, then unpoison the result and all the memory locations
3527  // pointed to by those arguments.
3528  // An inline asm() statement in C++ contains lists of input and output
3529  // arguments used by the assembly code. These are mapped to operands of the
3530  // CallInst as follows:
3531  // - nR register outputs ("=r) are returned by value in a single structure
3532  // (SSA value of the CallInst);
3533  // - nO other outputs ("=m" and others) are returned by pointer as first
3534  // nO operands of the CallInst;
3535  // - nI inputs ("r", "m" and others) are passed to CallInst as the
3536  // remaining nI operands.
3537  // The total number of asm() arguments in the source is nR+nO+nI, and the
3538  // corresponding CallInst has nO+nI+1 operands (the last operand is the
3539  // function to be called).
3540  const DataLayout &DL = F.getParent()->getDataLayout();
3541  CallInst *CI = dyn_cast<CallInst>(&I);
3542  IRBuilder<> IRB(&I);
3543  InlineAsm *IA = cast<InlineAsm>(CI->getCalledValue());
3544  int OutputArgs = getNumOutputArgs(IA, CI);
3545  // The last operand of a CallInst is the function itself.
3546  int NumOperands = CI->getNumOperands() - 1;
3547 
3548  // Check input arguments. Doing so before unpoisoning output arguments, so
3549  // that we won't overwrite uninit values before checking them.
3550  for (int i = OutputArgs; i < NumOperands; i++) {
3551  Value *Operand = CI->getOperand(i);
3552  instrumentAsmArgument(Operand, I, IRB, DL, /*isOutput*/ false);
3553  }
3554  // Unpoison output arguments. This must happen before the actual InlineAsm
3555  // call, so that the shadow for memory published in the asm() statement
3556  // remains valid.
3557  for (int i = 0; i < OutputArgs; i++) {
3558  Value *Operand = CI->getOperand(i);
3559  instrumentAsmArgument(Operand, I, IRB, DL, /*isOutput*/ true);
3560  }
3561 
3562  setShadow(&I, getCleanShadow(&I));
3563  setOrigin(&I, getCleanOrigin());
3564  }
3565 
3566  void visitInstruction(Instruction &I) {
3567  // Everything else: stop propagating and check for poisoned shadow.
3569  dumpInst(I);
3570  LLVM_DEBUG(dbgs() << "DEFAULT: " << I << "\n");
3571  for (size_t i = 0, n = I.getNumOperands(); i < n; i++) {
3572  Value *Operand = I.getOperand(i);
3573  if (Operand->getType()->isSized())
3574  insertShadowCheck(Operand, &I);
3575  }
3576  setShadow(&I, getCleanShadow(&I));
3577  setOrigin(&I, getCleanOrigin());
3578  }
3579 };
3580 
3581 /// AMD64-specific implementation of VarArgHelper.
3582 struct VarArgAMD64Helper : public VarArgHelper {
3583  // An unfortunate workaround for asymmetric lowering of va_arg stuff.
3584  // See a comment in visitCallSite for more details.
3585  static const unsigned AMD64GpEndOffset = 48; // AMD64 ABI Draft 0.99.6 p3.5.7
3586  static const unsigned AMD64FpEndOffsetSSE = 176;
3587  // If SSE is disabled, fp_offset in va_list is zero.
3588  static const unsigned AMD64FpEndOffsetNoSSE = AMD64GpEndOffset;
3589 
3590  unsigned AMD64FpEndOffset;
3591  Function &F;
3592  MemorySanitizer &MS;
3593  MemorySanitizerVisitor &MSV;
3594  Value *VAArgTLSCopy = nullptr;
3595  Value *VAArgTLSOriginCopy = nullptr;
3596  Value *VAArgOverflowSize = nullptr;
3597 
3598  SmallVector<CallInst*, 16> VAStartInstrumentationList;
3599 
3600  enum ArgKind { AK_GeneralPurpose, AK_FloatingPoint, AK_Memory };
3601 
3602  VarArgAMD64Helper(Function &F, MemorySanitizer &MS,
3603  MemorySanitizerVisitor &MSV)
3604  : F(F), MS(MS), MSV(MSV) {
3605  AMD64FpEndOffset = AMD64FpEndOffsetSSE;
3606  for (const auto &Attr : F.getAttributes().getFnAttributes()) {
3607  if (Attr.isStringAttribute() &&
3608  (Attr.getKindAsString() == "target-features")) {
3609  if (Attr.getValueAsString().contains("-sse"))
3610  AMD64FpEndOffset = AMD64FpEndOffsetNoSSE;
3611  break;
3612  }
3613  }
3614  }
3615 
3616  ArgKind classifyArgument(Value* arg) {
3617  // A very rough approximation of X86_64 argument classification rules.
3618  Type *T = arg->getType();
3619  if (T->isFPOrFPVectorTy() || T->isX86_MMXTy())
3620  return AK_FloatingPoint;
3621  if (T->isIntegerTy() && T->getPrimitiveSizeInBits() <= 64)
3622  return AK_GeneralPurpose;
3623  if (T->isPointerTy())
3624  return AK_GeneralPurpose;
3625  return AK_Memory;
3626  }
3627 
3628  // For VarArg functions, store the argument shadow in an ABI-specific format
3629  // that corresponds to va_list layout.
3630  // We do this because Clang lowers va_arg in the frontend, and this pass
3631  // only sees the low level code that deals with va_list internals.
3632  // A much easier alternative (provided that Clang emits va_arg instructions)
3633  // would have been to associate each live instance of va_list with a copy of
3634  // MSanParamTLS, and extract shadow on va_arg() call in the argument list
3635  // order.
3636  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
3637  unsigned GpOffset = 0;
3638  unsigned FpOffset = AMD64GpEndOffset;
3639  unsigned OverflowOffset = AMD64FpEndOffset;
3640  const DataLayout &DL = F.getParent()->getDataLayout();
3641  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
3642  ArgIt != End; ++ArgIt) {
3643  Value *A = *ArgIt;
3644  unsigned ArgNo = CS.getArgumentNo(ArgIt);
3645  bool IsFixed = ArgNo < CS.getFunctionType()->getNumParams();
3646  bool IsByVal = CS.paramHasAttr(ArgNo, Attribute::ByVal);
3647  if (IsByVal) {
3648  // ByVal arguments always go to the overflow area.
3649  // Fixed arguments passed through the overflow area will be stepped
3650  // over by va_start, so don't count them towards the offset.
3651  if (IsFixed)
3652  continue;
3653  assert(A->getType()->isPointerTy());
3654  Type *RealTy = A->getType()->getPointerElementType();
3655  uint64_t ArgSize = DL.getTypeAllocSize(RealTy);
3656  Value *ShadowBase = getShadowPtrForVAArgument(
3657  RealTy, IRB, OverflowOffset, alignTo(ArgSize, 8));
3658  Value *OriginBase = nullptr;
3659  if (MS.TrackOrigins)
3660  OriginBase = getOriginPtrForVAArgument(RealTy, IRB, OverflowOffset);
3661  OverflowOffset += alignTo(ArgSize, 8);
3662  if (!ShadowBase)
3663  continue;
3664  Value *ShadowPtr, *OriginPtr;
3665  std::tie(ShadowPtr, OriginPtr) =
3666  MSV.getShadowOriginPtr(A, IRB, IRB.getInt8Ty(), kShadowTLSAlignment,
3667  /*isStore*/ false);
3668 
3669  IRB.CreateMemCpy(ShadowBase, kShadowTLSAlignment, ShadowPtr,
3670  kShadowTLSAlignment, ArgSize);
3671  if (MS.TrackOrigins)
3672  IRB.CreateMemCpy(OriginBase, kShadowTLSAlignment, OriginPtr,
3673  kShadowTLSAlignment, ArgSize);
3674  } else {
3675  ArgKind AK = classifyArgument(A);
3676  if (AK == AK_GeneralPurpose && GpOffset >= AMD64GpEndOffset)
3677  AK = AK_Memory;
3678  if (AK == AK_FloatingPoint && FpOffset >= AMD64FpEndOffset)
3679  AK = AK_Memory;
3680  Value *ShadowBase, *OriginBase = nullptr;
3681  switch (AK) {
3682  case AK_GeneralPurpose:
3683  ShadowBase =
3684  getShadowPtrForVAArgument(A->getType(), IRB, GpOffset, 8);
3685  if (MS.TrackOrigins)
3686  OriginBase =
3687  getOriginPtrForVAArgument(A->getType(), IRB, GpOffset);
3688  GpOffset += 8;
3689  break;
3690  case AK_FloatingPoint:
3691  ShadowBase =
3692  getShadowPtrForVAArgument(A->getType(), IRB, FpOffset, 16);
3693  if (MS.TrackOrigins)
3694  OriginBase =
3695  getOriginPtrForVAArgument(A->getType(), IRB, FpOffset);
3696  FpOffset += 16;
3697  break;
3698  case AK_Memory:
3699  if (IsFixed)
3700  continue;
3701  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
3702  ShadowBase =
3703  getShadowPtrForVAArgument(A->getType(), IRB, OverflowOffset, 8);
3704  if (MS.TrackOrigins)
3705  OriginBase =
3706  getOriginPtrForVAArgument(A->getType(), IRB, OverflowOffset);
3707  OverflowOffset += alignTo(ArgSize, 8);
3708  }
3709  // Take fixed arguments into account for GpOffset and FpOffset,
3710  // but don't actually store shadows for them.
3711  // TODO(glider): don't call get*PtrForVAArgument() for them.
3712  if (IsFixed)
3713  continue;
3714  if (!ShadowBase)
3715  continue;
3716  Value *Shadow = MSV.getShadow(A);
3717  IRB.CreateAlignedStore(Shadow, ShadowBase, kShadowTLSAlignment);
3718  if (MS.TrackOrigins) {
3719  Value *Origin = MSV.getOrigin(A);
3720  unsigned StoreSize = DL.getTypeStoreSize(Shadow->getType());
3721  MSV.paintOrigin(IRB, Origin, OriginBase, StoreSize,
3722  std::max(kShadowTLSAlignment, kMinOriginAlignment));
3723  }
3724  }
3725  }
3726  Constant *OverflowSize =
3727  ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AMD64FpEndOffset);
3728  IRB.CreateStore(OverflowSize, MS.VAArgOverflowSizeTLS);
3729  }
3730 
3731  /// Compute the shadow address for a given va_arg.
3732  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
3733  unsigned ArgOffset, unsigned ArgSize) {
3734  // Make sure we don't overflow __msan_va_arg_tls.
3735  if (ArgOffset + ArgSize > kParamTLSSize)
3736  return nullptr;
3737  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
3738  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
3739  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
3740  "_msarg_va_s");
3741  }
3742 
3743  /// Compute the origin address for a given va_arg.
3744  Value *getOriginPtrForVAArgument(Type *Ty, IRBuilder<> &IRB, int ArgOffset) {
3745  Value *Base = IRB.CreatePointerCast(MS.VAArgOriginTLS, MS.IntptrTy);
3746  // getOriginPtrForVAArgument() is always called after
3747  // getShadowPtrForVAArgument(), so __msan_va_arg_origin_tls can never
3748  // overflow.
3749  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
3750  return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
3751  "_msarg_va_o");
3752  }
3753 
3754  void unpoisonVAListTagForInst(IntrinsicInst &I) {
3755  IRBuilder<> IRB(&I);
3756  Value *VAListTag = I.getArgOperand(0);
3757  Value *ShadowPtr, *OriginPtr;
3758  unsigned Alignment = 8;
3759  std::tie(ShadowPtr, OriginPtr) =
3760  MSV.getShadowOriginPtr(VAListTag, IRB, IRB.getInt8Ty(), Alignment,
3761  /*isStore*/ true);
3762 
3763  // Unpoison the whole __va_list_tag.
3764  // FIXME: magic ABI constants.
3765  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
3766  /* size */ 24, Alignment, false);
3767  // We shouldn't need to zero out the origins, as they're only checked for
3768  // nonzero shadow.
3769  }
3770 
3771  void visitVAStartInst(VAStartInst &I) override {
3773  return;
3774  VAStartInstrumentationList.push_back(&I);
3775  unpoisonVAListTagForInst(I);
3776  }
3777 
3778  void visitVACopyInst(VACopyInst &I) override {
3779  if (F.getCallingConv() == CallingConv::Win64) return;
3780  unpoisonVAListTagForInst(I);
3781  }
3782 
3783  void finalizeInstrumentation() override {
3784  assert(!VAArgOverflowSize && !VAArgTLSCopy &&
3785  "finalizeInstrumentation called twice");
3786  if (!VAStartInstrumentationList.empty()) {
3787  // If there is a va_start in this function, make a backup copy of
3788  // va_arg_tls somewhere in the function entry block.
3789  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
3790  VAArgOverflowSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
3791  Value *CopySize =
3792  IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, AMD64FpEndOffset),
3793  VAArgOverflowSize);
3794  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
3795  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
3796  if (MS.TrackOrigins) {
3797  VAArgTLSOriginCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
3798  IRB.CreateMemCpy(VAArgTLSOriginCopy, 8, MS.VAArgOriginTLS, 8, CopySize);
3799  }
3800  }
3801 
3802  // Instrument va_start.
3803  // Copy va_list shadow from the backup copy of the TLS contents.
3804  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
3805  CallInst *OrigInst = VAStartInstrumentationList[i];
3806  IRBuilder<> IRB(OrigInst->getNextNode());
3807  Value *VAListTag = OrigInst->getArgOperand(0);
3808 
3809  Value *RegSaveAreaPtrPtr = IRB.CreateIntToPtr(
3810  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
3811  ConstantInt::get(MS.IntptrTy, 16)),
3813  Value *RegSaveAreaPtr = IRB.CreateLoad(RegSaveAreaPtrPtr);
3814  Value *RegSaveAreaShadowPtr, *RegSaveAreaOriginPtr;
3815  unsigned Alignment = 16;
3816  std::tie(RegSaveAreaShadowPtr, RegSaveAreaOriginPtr) =
3817  MSV.getShadowOriginPtr(RegSaveAreaPtr, IRB, IRB.getInt8Ty(),
3818  Alignment, /*isStore*/ true);
3819  IRB.CreateMemCpy(RegSaveAreaShadowPtr, Alignment, VAArgTLSCopy, Alignment,
3820  AMD64FpEndOffset);
3821  if (MS.TrackOrigins)
3822  IRB.CreateMemCpy(RegSaveAreaOriginPtr, Alignment, VAArgTLSOriginCopy,
3823  Alignment, AMD64FpEndOffset);
3824  Value *OverflowArgAreaPtrPtr = IRB.CreateIntToPtr(
3825  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
3826  ConstantInt::get(MS.IntptrTy, 8)),
3828  Value *OverflowArgAreaPtr = IRB.CreateLoad(OverflowArgAreaPtrPtr);
3829  Value *OverflowArgAreaShadowPtr, *OverflowArgAreaOriginPtr;
3830  std::tie(OverflowArgAreaShadowPtr, OverflowArgAreaOriginPtr) =
3831  MSV.getShadowOriginPtr(OverflowArgAreaPtr, IRB, IRB.getInt8Ty(),
3832  Alignment, /*isStore*/ true);
3833  Value *SrcPtr = IRB.CreateConstGEP1_32(IRB.getInt8Ty(), VAArgTLSCopy,
3834  AMD64FpEndOffset);
3835  IRB.CreateMemCpy(OverflowArgAreaShadowPtr, Alignment, SrcPtr, Alignment,
3836  VAArgOverflowSize);
3837  if (MS.TrackOrigins) {
3838  SrcPtr = IRB.CreateConstGEP1_32(IRB.getInt8Ty(), VAArgTLSOriginCopy,
3839  AMD64FpEndOffset);
3840  IRB.CreateMemCpy(OverflowArgAreaOriginPtr, Alignment, SrcPtr, Alignment,
3841  VAArgOverflowSize);
3842  }
3843  }
3844  }
3845 };
3846 
3847 /// MIPS64-specific implementation of VarArgHelper.
3848 struct VarArgMIPS64Helper : public VarArgHelper {
3849  Function &F;
3850  MemorySanitizer &MS;
3851  MemorySanitizerVisitor &MSV;
3852  Value *VAArgTLSCopy = nullptr;
3853  Value *VAArgSize = nullptr;
3854 
3855  SmallVector<CallInst*, 16> VAStartInstrumentationList;
3856 
3857  VarArgMIPS64Helper(Function &F, MemorySanitizer &MS,
3858  MemorySanitizerVisitor &MSV) : F(F), MS(MS), MSV(MSV) {}
3859 
3860  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
3861  unsigned VAArgOffset = 0;
3862  const DataLayout &DL = F.getParent()->getDataLayout();
3863  for (CallSite::arg_iterator ArgIt = CS.arg_begin() +
3864  CS.getFunctionType()->getNumParams(), End = CS.arg_end();
3865  ArgIt != End; ++ArgIt) {
3866  Triple TargetTriple(F.getParent()->getTargetTriple());
3867  Value *A = *ArgIt;
3868  Value *Base;
3869  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
3870  if (TargetTriple.getArch() == Triple::mips64) {
3871  // Adjusting the shadow for argument with size < 8 to match the placement
3872  // of bits in big endian system
3873  if (ArgSize < 8)
3874  VAArgOffset += (8 - ArgSize);
3875  }
3876  Base = getShadowPtrForVAArgument(A->getType(), IRB, VAArgOffset, ArgSize);
3877  VAArgOffset += ArgSize;
3878  VAArgOffset = alignTo(VAArgOffset, 8);
3879  if (!Base)
3880  continue;
3881  IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
3882  }
3883 
3884  Constant *TotalVAArgSize = ConstantInt::get(IRB.getInt64Ty(), VAArgOffset);
3885  // Here using VAArgOverflowSizeTLS as VAArgSizeTLS to avoid creation of
3886  // a new class member i.e. it is the total size of all VarArgs.
3887  IRB.CreateStore(TotalVAArgSize, MS.VAArgOverflowSizeTLS);
3888  }
3889 
3890  /// Compute the shadow address for a given va_arg.
3891  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
3892  unsigned ArgOffset, unsigned ArgSize) {
3893  // Make sure we don't overflow __msan_va_arg_tls.
3894  if (ArgOffset + ArgSize > kParamTLSSize)
3895  return nullptr;
3896  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
3897  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
3898  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
3899  "_msarg");
3900  }
3901 
3902  void visitVAStartInst(VAStartInst &I) override {
3903  IRBuilder<> IRB(&I);
3904  VAStartInstrumentationList.push_back(&I);
3905  Value *VAListTag = I.getArgOperand(0);
3906  Value *ShadowPtr, *OriginPtr;
3907  unsigned Alignment = 8;
3908  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
3909  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
3910  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
3911  /* size */ 8, Alignment, false);
3912  }
3913 
3914  void visitVACopyInst(VACopyInst &I) override {
3915  IRBuilder<> IRB(&I);
3916  VAStartInstrumentationList.push_back(&I);
3917  Value *VAListTag = I.getArgOperand(0);
3918  Value *ShadowPtr, *OriginPtr;
3919  unsigned Alignment = 8;
3920  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
3921  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
3922  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
3923  /* size */ 8, Alignment, false);
3924  }
3925 
3926  void finalizeInstrumentation() override {
3927  assert(!VAArgSize && !VAArgTLSCopy &&
3928  "finalizeInstrumentation called twice");
3929  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
3930  VAArgSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
3931  Value *CopySize = IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, 0),
3932  VAArgSize);
3933 
3934  if (!VAStartInstrumentationList.empty()) {
3935  // If there is a va_start in this function, make a backup copy of
3936  // va_arg_tls somewhere in the function entry block.
3937  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
3938  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
3939  }
3940 
3941  // Instrument va_start.
3942  // Copy va_list shadow from the backup copy of the TLS contents.
3943  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
3944  CallInst *OrigInst = VAStartInstrumentationList[i];
3945  IRBuilder<> IRB(OrigInst->getNextNode());
3946  Value *VAListTag = OrigInst->getArgOperand(0);
3947  Value *RegSaveAreaPtrPtr =
3948  IRB.CreateIntToPtr(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
3950  Value *RegSaveAreaPtr = IRB.CreateLoad(RegSaveAreaPtrPtr);
3951  Value *RegSaveAreaShadowPtr, *RegSaveAreaOriginPtr;
3952  unsigned Alignment = 8;
3953  std::tie(RegSaveAreaShadowPtr, RegSaveAreaOriginPtr) =
3954  MSV.getShadowOriginPtr(RegSaveAreaPtr, IRB, IRB.getInt8Ty(),
3955  Alignment, /*isStore*/ true);
3956  IRB.CreateMemCpy(RegSaveAreaShadowPtr, Alignment, VAArgTLSCopy, Alignment,
3957  CopySize);
3958  }
3959  }
3960 };
3961 
3962 /// AArch64-specific implementation of VarArgHelper.
3963 struct VarArgAArch64Helper : public VarArgHelper {
3964  static const unsigned kAArch64GrArgSize = 64;
3965  static const unsigned kAArch64VrArgSize = 128;
3966 
3967  static const unsigned AArch64GrBegOffset = 0;
3968  static const unsigned AArch64GrEndOffset = kAArch64GrArgSize;
3969  // Make VR space aligned to 16 bytes.
3970  static const unsigned AArch64VrBegOffset = AArch64GrEndOffset;
3971  static const unsigned AArch64VrEndOffset = AArch64VrBegOffset
3972  + kAArch64VrArgSize;
3973  static const unsigned AArch64VAEndOffset = AArch64VrEndOffset;
3974 
3975  Function &F;
3976  MemorySanitizer &MS;
3977  MemorySanitizerVisitor &MSV;
3978  Value *VAArgTLSCopy = nullptr;
3979  Value *VAArgOverflowSize = nullptr;
3980 
3981  SmallVector<CallInst*, 16> VAStartInstrumentationList;
3982 
3983  enum ArgKind { AK_GeneralPurpose, AK_FloatingPoint, AK_Memory };
3984 
3985  VarArgAArch64Helper(Function &F, MemorySanitizer &MS,
3986  MemorySanitizerVisitor &MSV) : F(F), MS(MS), MSV(MSV) {}
3987 
3988  ArgKind classifyArgument(Value* arg) {
3989  Type *T = arg->getType();
3990  if (T->isFPOrFPVectorTy())
3991  return AK_FloatingPoint;
3992  if ((T->isIntegerTy() && T->getPrimitiveSizeInBits() <= 64)
3993  || (T->isPointerTy()))
3994  return AK_GeneralPurpose;
3995  return AK_Memory;
3996  }
3997 
3998  // The instrumentation stores the argument shadow in a non ABI-specific
3999  // format because it does not know which argument is named (since Clang,
4000  // like x86_64 case, lowers the va_args in the frontend and this pass only
4001  // sees the low level code that deals with va_list internals).
4002  // The first seven GR registers are saved in the first 56 bytes of the
4003  // va_arg tls arra, followers by the first 8 FP/SIMD registers, and then
4004  // the remaining arguments.
4005  // Using constant offset within the va_arg TLS array allows fast copy
4006  // in the finalize instrumentation.
4007  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
4008  unsigned GrOffset = AArch64GrBegOffset;
4009  unsigned VrOffset = AArch64VrBegOffset;
4010  unsigned OverflowOffset = AArch64VAEndOffset;
4011 
4012  const DataLayout &DL = F.getParent()->getDataLayout();
4013  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
4014  ArgIt != End; ++ArgIt) {
4015  Value *A = *ArgIt;
4016  unsigned ArgNo = CS.getArgumentNo(ArgIt);
4017  bool IsFixed = ArgNo < CS.getFunctionType()->getNumParams();
4018  ArgKind AK = classifyArgument(A);
4019  if (AK == AK_GeneralPurpose && GrOffset >= AArch64GrEndOffset)
4020  AK = AK_Memory;
4021  if (AK == AK_FloatingPoint && VrOffset >= AArch64VrEndOffset)
4022  AK = AK_Memory;
4023  Value *Base;
4024  switch (AK) {
4025  case AK_GeneralPurpose:
4026  Base = getShadowPtrForVAArgument(A->getType(), IRB, GrOffset, 8);
4027  GrOffset += 8;
4028  break;
4029  case AK_FloatingPoint:
4030  Base = getShadowPtrForVAArgument(A->getType(), IRB, VrOffset, 8);
4031  VrOffset += 16;
4032  break;
4033  case AK_Memory:
4034  // Don't count fixed arguments in the overflow area - va_start will
4035  // skip right over them.
4036  if (IsFixed)
4037  continue;
4038  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
4039  Base = getShadowPtrForVAArgument(A->getType(), IRB, OverflowOffset,
4040  alignTo(ArgSize, 8));
4041  OverflowOffset += alignTo(ArgSize, 8);
4042  break;
4043  }
4044  // Count Gp/Vr fixed arguments to their respective offsets, but don't
4045  // bother to actually store a shadow.
4046  if (IsFixed)
4047  continue;
4048  if (!Base)
4049  continue;
4050  IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
4051  }
4052  Constant *OverflowSize =
4053  ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AArch64VAEndOffset);
4054  IRB.CreateStore(OverflowSize, MS.VAArgOverflowSizeTLS);
4055  }
4056 
4057  /// Compute the shadow address for a given va_arg.
4058  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
4059  unsigned ArgOffset, unsigned ArgSize) {
4060  // Make sure we don't overflow __msan_va_arg_tls.
4061  if (ArgOffset + ArgSize > kParamTLSSize)
4062  return nullptr;
4063  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
4064  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
4065  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
4066  "_msarg");
4067  }
4068 
4069  void visitVAStartInst(VAStartInst &I) override {
4070  IRBuilder<> IRB(&I);
4071  VAStartInstrumentationList.push_back(&I);
4072  Value *VAListTag = I.getArgOperand(0);
4073  Value *ShadowPtr, *OriginPtr;
4074  unsigned Alignment = 8;
4075  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4076  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4077  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4078  /* size */ 32, Alignment, false);
4079  }
4080 
4081  void visitVACopyInst(VACopyInst &I) override {
4082  IRBuilder<> IRB(&I);
4083  VAStartInstrumentationList.push_back(&I);
4084  Value *VAListTag = I.getArgOperand(0);
4085  Value *ShadowPtr, *OriginPtr;
4086  unsigned Alignment = 8;
4087  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4088  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4089  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4090  /* size */ 32, Alignment, false);
4091  }
4092 
4093  // Retrieve a va_list field of 'void*' size.
4094  Value* getVAField64(IRBuilder<> &IRB, Value *VAListTag, int offset) {
4095  Value *SaveAreaPtrPtr =
4096  IRB.CreateIntToPtr(
4097  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
4098  ConstantInt::get(MS.IntptrTy, offset)),
4099  Type::getInt64PtrTy(*MS.C));
4100  return IRB.CreateLoad(SaveAreaPtrPtr);
4101  }
4102 
4103  // Retrieve a va_list field of 'int' size.
4104  Value* getVAField32(IRBuilder<> &IRB, Value *VAListTag, int offset) {
4105  Value *SaveAreaPtr =
4106  IRB.CreateIntToPtr(
4107  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
4108  ConstantInt::get(MS.IntptrTy, offset)),
4109  Type::getInt32PtrTy(*MS.C));
4110  Value *SaveArea32 = IRB.CreateLoad(SaveAreaPtr);
4111  return IRB.CreateSExt(SaveArea32, MS.IntptrTy);
4112  }
4113 
4114  void finalizeInstrumentation() override {
4115  assert(!VAArgOverflowSize && !VAArgTLSCopy &&
4116  "finalizeInstrumentation called twice");
4117  if (!VAStartInstrumentationList.empty()) {
4118  // If there is a va_start in this function, make a backup copy of
4119  // va_arg_tls somewhere in the function entry block.
4120  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
4121  VAArgOverflowSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
4122  Value *CopySize =
4123  IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, AArch64VAEndOffset),
4124  VAArgOverflowSize);
4125  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
4126  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
4127  }
4128 
4129  Value *GrArgSize = ConstantInt::get(MS.IntptrTy, kAArch64GrArgSize);
4130  Value *VrArgSize = ConstantInt::get(MS.IntptrTy, kAArch64VrArgSize);
4131 
4132  // Instrument va_start, copy va_list shadow from the backup copy of
4133  // the TLS contents.
4134  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
4135  CallInst *OrigInst = VAStartInstrumentationList[i];
4136  IRBuilder<> IRB(OrigInst->getNextNode());
4137 
4138  Value *VAListTag = OrigInst->getArgOperand(0);
4139 
4140  // The variadic ABI for AArch64 creates two areas to save the incoming
4141  // argument registers (one for 64-bit general register xn-x7 and another
4142  // for 128-bit FP/SIMD vn-v7).
4143  // We need then to propagate the shadow arguments on both regions
4144  // 'va::__gr_top + va::__gr_offs' and 'va::__vr_top + va::__vr_offs'.
4145  // The remaning arguments are saved on shadow for 'va::stack'.
4146  // One caveat is it requires only to propagate the non-named arguments,
4147  // however on the call site instrumentation 'all' the arguments are
4148  // saved. So to copy the shadow values from the va_arg TLS array
4149  // we need to adjust the offset for both GR and VR fields based on
4150  // the __{gr,vr}_offs value (since they are stores based on incoming
4151  // named arguments).
4152 
4153  // Read the stack pointer from the va_list.
4154  Value *StackSaveAreaPtr = getVAField64(IRB, VAListTag, 0);
4155 
4156  // Read both the __gr_top and __gr_off and add them up.
4157  Value *GrTopSaveAreaPtr = getVAField64(IRB, VAListTag, 8);
4158  Value *GrOffSaveArea = getVAField32(IRB, VAListTag, 24);
4159 
4160  Value *GrRegSaveAreaPtr = IRB.CreateAdd(GrTopSaveAreaPtr, GrOffSaveArea);
4161 
4162  // Read both the __vr_top and __vr_off and add them up.
4163  Value *VrTopSaveAreaPtr = getVAField64(IRB, VAListTag, 16);
4164  Value *VrOffSaveArea = getVAField32(IRB, VAListTag, 28);
4165 
4166  Value *VrRegSaveAreaPtr = IRB.CreateAdd(VrTopSaveAreaPtr, VrOffSaveArea);
4167 
4168  // It does not know how many named arguments is being used and, on the
4169  // callsite all the arguments were saved. Since __gr_off is defined as
4170  // '0 - ((8 - named_gr) * 8)', the idea is to just propagate the variadic
4171  // argument by ignoring the bytes of shadow from named arguments.
4172  Value *GrRegSaveAreaShadowPtrOff =
4173  IRB.CreateAdd(GrArgSize, GrOffSaveArea);
4174 
4175  Value *GrRegSaveAreaShadowPtr =
4176  MSV.getShadowOriginPtr(GrRegSaveAreaPtr, IRB, IRB.getInt8Ty(),
4177  /*Alignment*/ 8, /*isStore*/ true)
4178  .first;
4179 
4180  Value *GrSrcPtr = IRB.CreateInBoundsGEP(IRB.getInt8Ty(), VAArgTLSCopy,
4181  GrRegSaveAreaShadowPtrOff);
4182  Value *GrCopySize = IRB.CreateSub(GrArgSize, GrRegSaveAreaShadowPtrOff);
4183 
4184  IRB.CreateMemCpy(GrRegSaveAreaShadowPtr, 8, GrSrcPtr, 8, GrCopySize);
4185 
4186  // Again, but for FP/SIMD values.
4187  Value *VrRegSaveAreaShadowPtrOff =
4188  IRB.CreateAdd(VrArgSize, VrOffSaveArea);
4189 
4190  Value *VrRegSaveAreaShadowPtr =
4191  MSV.getShadowOriginPtr(VrRegSaveAreaPtr, IRB, IRB.getInt8Ty(),
4192  /*Alignment*/ 8, /*isStore*/ true)
4193  .first;
4194 
4195  Value *VrSrcPtr = IRB.CreateInBoundsGEP(
4196  IRB.getInt8Ty(),
4197  IRB.CreateInBoundsGEP(IRB.getInt8Ty(), VAArgTLSCopy,
4198  IRB.getInt32(AArch64VrBegOffset)),
4199  VrRegSaveAreaShadowPtrOff);
4200  Value *VrCopySize = IRB.CreateSub(VrArgSize, VrRegSaveAreaShadowPtrOff);
4201 
4202  IRB.CreateMemCpy(VrRegSaveAreaShadowPtr, 8, VrSrcPtr, 8, VrCopySize);
4203 
4204  // And finally for remaining arguments.
4205  Value *StackSaveAreaShadowPtr =
4206  MSV.getShadowOriginPtr(StackSaveAreaPtr, IRB, IRB.getInt8Ty(),
4207  /*Alignment*/ 16, /*isStore*/ true)
4208  .first;
4209 
4210  Value *StackSrcPtr =
4211  IRB.CreateInBoundsGEP(IRB.getInt8Ty(), VAArgTLSCopy,
4212  IRB.getInt32(AArch64VAEndOffset));
4213 
4214  IRB.CreateMemCpy(StackSaveAreaShadowPtr, 16, StackSrcPtr, 16,
4215  VAArgOverflowSize);
4216  }
4217  }
4218 };
4219 
4220 /// PowerPC64-specific implementation of VarArgHelper.
4221 struct VarArgPowerPC64Helper : public VarArgHelper {
4222  Function &F;
4223  MemorySanitizer &MS;
4224  MemorySanitizerVisitor &MSV;
4225  Value *VAArgTLSCopy = nullptr;
4226  Value *VAArgSize = nullptr;
4227 
4228  SmallVector<CallInst*, 16> VAStartInstrumentationList;
4229 
4230  VarArgPowerPC64Helper(Function &F, MemorySanitizer &MS,
4231  MemorySanitizerVisitor &MSV) : F(F), MS(MS), MSV(MSV) {}
4232 
4233  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
4234  // For PowerPC, we need to deal with alignment of stack arguments -
4235  // they are mostly aligned to 8 bytes, but vectors and i128 arrays
4236  // are aligned to 16 bytes, byvals can be aligned to 8 or 16 bytes,
4237  // and QPX vectors are aligned to 32 bytes. For that reason, we
4238  // compute current offset from stack pointer (which is always properly
4239  // aligned), and offset for the first vararg, then subtract them.
4240  unsigned VAArgBase;
4241  Triple TargetTriple(F.getParent()->getTargetTriple());
4242  // Parameter save area starts at 48 bytes from frame pointer for ABIv1,
4243  // and 32 bytes for ABIv2. This is usually determined by target
4244  // endianness, but in theory could be overriden by function attribute.
4245  // For simplicity, we ignore it here (it'd only matter for QPX vectors).
4246  if (TargetTriple.getArch() == Triple::ppc64)
4247  VAArgBase = 48;
4248  else
4249  VAArgBase = 32;
4250  unsigned VAArgOffset = VAArgBase;
4251  const DataLayout &DL = F.getParent()->getDataLayout();
4252  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
4253  ArgIt != End; ++ArgIt) {
4254  Value *A = *ArgIt;
4255  unsigned ArgNo = CS.getArgumentNo(ArgIt);
4256  bool IsFixed = ArgNo < CS.getFunctionType()->getNumParams();
4257  bool IsByVal = CS.paramHasAttr(ArgNo, Attribute::ByVal);
4258  if (IsByVal) {
4259  assert(A->getType()->isPointerTy());
4260  Type *RealTy = A->getType()->getPointerElementType();
4261  uint64_t ArgSize = DL.getTypeAllocSize(RealTy);
4262  uint64_t ArgAlign = CS.getParamAlignment(ArgNo);
4263  if (ArgAlign < 8)
4264  ArgAlign = 8;
4265  VAArgOffset = alignTo(VAArgOffset, ArgAlign);
4266  if (!IsFixed) {
4267  Value *Base = getShadowPtrForVAArgument(
4268  RealTy, IRB, VAArgOffset - VAArgBase, ArgSize);
4269  if (Base) {
4270  Value *AShadowPtr, *AOriginPtr;
4271  std::tie(AShadowPtr, AOriginPtr) =
4272  MSV.getShadowOriginPtr(A, IRB, IRB.getInt8Ty(),
4273  kShadowTLSAlignment, /*isStore*/ false);
4274 
4275  IRB.CreateMemCpy(Base, kShadowTLSAlignment, AShadowPtr,
4276  kShadowTLSAlignment, ArgSize);
4277  }
4278  }
4279  VAArgOffset += alignTo(ArgSize, 8);
4280  } else {
4281  Value *Base;
4282  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
4283  uint64_t ArgAlign = 8;
4284  if (A->getType()->isArrayTy()) {
4285  // Arrays are aligned to element size, except for long double
4286  // arrays, which are aligned to 8 bytes.
4287  Type *ElementTy = A->getType()->getArrayElementType();
4288  if (!ElementTy->isPPC_FP128Ty())
4289  ArgAlign = DL.getTypeAllocSize(ElementTy);
4290  } else if (A->getType()->isVectorTy()) {
4291  // Vectors are naturally aligned.
4292  ArgAlign = DL.getTypeAllocSize(A->getType());
4293  }
4294  if (ArgAlign < 8)
4295  ArgAlign = 8;
4296  VAArgOffset = alignTo(VAArgOffset, ArgAlign);
4297  if (DL.isBigEndian()) {
4298  // Adjusting the shadow for argument with size < 8 to match the placement
4299  // of bits in big endian system
4300  if (ArgSize < 8)
4301  VAArgOffset += (8 - ArgSize);
4302  }
4303  if (!IsFixed) {
4304  Base = getShadowPtrForVAArgument(A->getType(), IRB,
4305  VAArgOffset - VAArgBase, ArgSize);
4306  if (Base)
4307  IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
4308  }
4309  VAArgOffset += ArgSize;
4310  VAArgOffset = alignTo(VAArgOffset, 8);
4311  }
4312  if (IsFixed)
4313  VAArgBase = VAArgOffset;
4314  }
4315 
4316  Constant *TotalVAArgSize = ConstantInt::get(IRB.getInt64Ty(),
4317  VAArgOffset - VAArgBase);
4318  // Here using VAArgOverflowSizeTLS as VAArgSizeTLS to avoid creation of
4319  // a new class member i.e. it is the total size of all VarArgs.
4320  IRB.CreateStore(TotalVAArgSize, MS.VAArgOverflowSizeTLS);
4321  }
4322 
4323  /// Compute the shadow address for a given va_arg.
4324  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
4325  unsigned ArgOffset, unsigned ArgSize) {
4326  // Make sure we don't overflow __msan_va_arg_tls.
4327  if (ArgOffset + ArgSize > kParamTLSSize)
4328  return nullptr;
4329  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
4330  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
4331  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
4332  "_msarg");
4333  }
4334 
4335  void visitVAStartInst(VAStartInst &I) override {
4336  IRBuilder<> IRB(&I);
4337  VAStartInstrumentationList.push_back(&I);
4338  Value *VAListTag = I.getArgOperand(0);
4339  Value *ShadowPtr, *OriginPtr;
4340  unsigned Alignment = 8;
4341  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4342  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4343  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4344  /* size */ 8, Alignment, false);
4345  }
4346 
4347  void visitVACopyInst(VACopyInst &I) override {
4348  IRBuilder<> IRB(&I);
4349  Value *VAListTag = I.getArgOperand(0);
4350  Value *ShadowPtr, *OriginPtr;
4351  unsigned Alignment = 8;
4352  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4353  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4354  // Unpoison the whole __va_list_tag.
4355  // FIXME: magic ABI constants.
4356  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4357  /* size */ 8, Alignment, false);
4358  }
4359 
4360  void finalizeInstrumentation() override {
4361  assert(!VAArgSize && !VAArgTLSCopy &&
4362  "finalizeInstrumentation called twice");
4363  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
4364  VAArgSize = IRB.CreateLoad(MS.VAArgOverflowSizeTLS);
4365  Value *CopySize = IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, 0),
4366  VAArgSize);
4367 
4368  if (!VAStartInstrumentationList.empty()) {
4369  // If there is a va_start in this function, make a backup copy of
4370  // va_arg_tls somewhere in the function entry block.
4371  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
4372  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
4373  }
4374 
4375  // Instrument va_start.
4376  // Copy va_list shadow from the backup copy of the TLS contents.
4377  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
4378  CallInst *OrigInst = VAStartInstrumentationList[i];
4379  IRBuilder<> IRB(OrigInst->getNextNode());
4380  Value *VAListTag = OrigInst->getArgOperand(0);
4381  Value *RegSaveAreaPtrPtr =
4382  IRB.CreateIntToPtr(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
4384  Value *RegSaveAreaPtr = IRB.CreateLoad(RegSaveAreaPtrPtr);
4385  Value *RegSaveAreaShadowPtr, *RegSaveAreaOriginPtr;
4386  unsigned Alignment = 8;
4387  std::tie(RegSaveAreaShadowPtr, RegSaveAreaOriginPtr) =
4388  MSV.getShadowOriginPtr(RegSaveAreaPtr, IRB, IRB.getInt8Ty(),
4389  Alignment, /*isStore*/ true);
4390  IRB.CreateMemCpy(RegSaveAreaShadowPtr, Alignment, VAArgTLSCopy, Alignment,
4391  CopySize);
4392  }
4393  }
4394 };
4395 
4396 /// A no-op implementation of VarArgHelper.
4397 struct VarArgNoOpHelper : public VarArgHelper {
4398  VarArgNoOpHelper(Function &F, MemorySanitizer &MS,
4399  MemorySanitizerVisitor &MSV) {}
4400 
4401  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {}
4402 
4403  void visitVAStartInst(VAStartInst &I) override {}
4404 
4405  void visitVACopyInst(VACopyInst &I) override {}
4406 
4407  void finalizeInstrumentation() override {}
4408 };
4409 
4410 } // end anonymous namespace
4411 
4412 static VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
4413  MemorySanitizerVisitor &Visitor) {
4414  // VarArg handling is only implemented on AMD64. False positives are possible
4415  // on other platforms.
4416  Triple TargetTriple(Func.getParent()->getTargetTriple());
4417  if (TargetTriple.getArch() == Triple::x86_64)
4418  return new VarArgAMD64Helper(Func, Msan, Visitor);
4419  else if (TargetTriple.isMIPS64())
4420  return new VarArgMIPS64Helper(Func, Msan, Visitor);
4421  else if (TargetTriple.getArch() == Triple::aarch64)
4422  return new VarArgAArch64Helper(Func, Msan, Visitor);
4423  else if (TargetTriple.getArch() == Triple::ppc64 ||
4424  TargetTriple.getArch() == Triple::ppc64le)
4425  return new VarArgPowerPC64Helper(Func, Msan, Visitor);
4426  else
4427  return new VarArgNoOpHelper(Func, Msan, Visitor);
4428 }
4429 
4431  if (!CompileKernel && (&F == MsanCtorFunction))
4432  return false;
4433  MemorySanitizerVisitor Visitor(F, *this);
4434 
4435  // Clear out readonly/readnone attributes.
4436  AttrBuilder B;
4437  B.addAttribute(Attribute::ReadOnly)
4438  .addAttribute(Attribute::ReadNone);
4440 
4441  return Visitor.runOnFunction();
4442 }
Value * CreateInBoundsGEP(Value *Ptr, ArrayRef< Value *> IdxList, const Twine &Name="")
Definition: IRBuilder.h:1405
Type * getVectorElementType() const
Definition: Type.h:371
uint64_t CallInst * C
Return a value (possibly void), from a function.
User::op_iterator arg_iterator
The type of iterator to use when looping over actual arguments at this call site. ...
Definition: CallSite.h:213
SymbolTableList< Instruction >::iterator eraseFromParent()
This method unlinks &#39;this&#39; from the containing basic block and deletes it.
Definition: Instruction.cpp:68
unsigned Log2_32_Ceil(uint32_t Value)
Return the ceil log base 2 of the specified value, 32 if the value is zero.
Definition: MathExtras.h:552
A parsed version of the target data layout string in and methods for querying it. ...
Definition: DataLayout.h:111
constexpr char Align[]
Key for Kernel::Arg::Metadata::mAlign.
const std::string & getTargetTriple() const
Get the target triple which is a string describing the target host.
Definition: Module.h:239
static const MemoryMapParams Linux_PowerPC64_MemoryMapParams
Function * getCalledFunction() const
Return the function called, or null if this is an indirect function invocation.
Value * CreateICmp(CmpInst::Predicate P, Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1858
Value * CreateConstGEP1_32(Value *Ptr, unsigned Idx0, const Twine &Name="")
Definition: IRBuilder.h:1444
static Constant * getString(LLVMContext &Context, StringRef Initializer, bool AddNull=true)
This method constructs a CDS and initializes it with a text string.
Definition: Constants.cpp:2545
bool isAllOnesValue() const
Return true if this is the value that would be returned by getAllOnesValue.
Definition: Constants.cpp:100
raw_ostream & errs()
This returns a reference to a raw_ostream for standard error.
void addIncoming(Value *V, BasicBlock *BB)
Add an incoming value to the end of the PHI list.
This instruction extracts a struct member or array element value from an aggregate value...
Value * CreateBinOp(Instruction::BinaryOps Opc, Value *LHS, Value *RHS, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1258
GCNRegPressure max(const GCNRegPressure &P1, const GCNRegPressure &P2)
This class represents an incoming formal argument to a Function.
Definition: Argument.h:30
static const PlatformMemoryMapParams Linux_PowerPC_MemoryMapParams
AllocaInst * CreateAlloca(Type *Ty, unsigned AddrSpace, Value *ArraySize=nullptr, const Twine &Name="")
Definition: IRBuilder.h:1304
Base class for instruction visitors.
Definition: InstVisitor.h:81
Value * getAggregateOperand()
Value * CreateICmpNE(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1752
Atomic ordering constants.
NodeTy * getNextNode()
Get the next node, or nullptr for the list tail.
Definition: ilist_node.h:289
LLVM_ATTRIBUTE_NORETURN void report_fatal_error(Error Err, bool gen_crash_diag=true)
Report a serious error, calling any installed error handler.
Definition: Error.cpp:139
Compute iterated dominance frontiers using a linear time algorithm.
Definition: AllocatorList.h:24
BinaryOps getOpcode() const
Definition: InstrTypes.h:311
unsigned getParamAlignment(unsigned ArgNo) const
Extract the alignment for a call or parameter (0=unknown).
Definition: CallSite.h:406
bool isAtomic() const
Return true if this instruction has an AtomicOrdering of unordered or higher.
Constant * getOrInsertFunction(StringRef Name, FunctionType *T, AttributeList AttributeList)
Look up the specified function in the module symbol table.
Definition: Module.cpp:143
Value * CreateXor(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1160
A Module instance is used to store all the information related to an LLVM module. ...
Definition: Module.h:64
bool isSized(SmallPtrSetImpl< Type *> *Visited=nullptr) const
Return true if it makes sense to take the size of this type.
Definition: Type.h:265
#define LLVM_FALLTHROUGH
Definition: Compiler.h:86
void setOrdering(AtomicOrdering Ordering)
Sets the ordering constraint of this rmw instruction.
Definition: Instructions.h:768
static const MemoryMapParams Linux_I386_MemoryMapParams
Same, but only replaced by something equivalent.
Definition: GlobalValue.h:54
an instruction that atomically checks whether a specified value is in a memory location, and, if it is, stores a new value there.
Definition: Instructions.h:518
static const MemoryMapParams NetBSD_X86_64_MemoryMapParams
static cl::opt< bool > ClPoisonStackWithCall("msan-poison-stack-with-call", cl::desc("poison uninitialized stack variables with a call"), cl::Hidden, cl::init(false))
This class represents zero extension of integer types.
static const unsigned kRetvalTLSSize
unsigned getNumElements() const
Random access to the elements.
Definition: DerivedTypes.h:313
bool removeUnreachableBlocks(Function &F, LazyValueInfo *LVI=nullptr, DomTreeUpdater *DTU=nullptr, MemorySSAUpdater *MSSAU=nullptr)
Remove all blocks that can not be reached from the function&#39;s entry.
Definition: Local.cpp:2218
Value * CreateICmpSLT(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1780
This class represents a function call, abstracting a target machine&#39;s calling convention.
static PointerType * getInt32PtrTy(LLVMContext &C, unsigned AS=0)
Definition: Type.cpp:228
void setOrdering(AtomicOrdering Ordering)
Sets the ordering constraint of this load instruction.
Definition: Instructions.h:243
static PointerType * get(Type *ElementType, unsigned AddressSpace)
This constructs a pointer to an object of the specified type in a numbered address space...
Definition: Type.cpp:630
const Value * getTrueValue() const
AtomicOrdering getOrdering() const
Returns the ordering constraint of this load instruction.
Definition: Instructions.h:237
Like Internal, but omit from symbol table.
Definition: GlobalValue.h:57
static VarArgHelper * CreateVarArgHelper(Function &Func, MemorySanitizer &Msan, MemorySanitizerVisitor &Visitor)
This instruction constructs a fixed permutation of two input vectors.
static cl::opt< bool > ClWithComdat("msan-with-comdat", cl::desc("Place MSan constructors in comdat sections"), cl::Hidden, cl::init(false))
Externally visible function.
Definition: GlobalValue.h:49
bool hasFnAttribute(Attribute::AttrKind Kind) const
Return true if the function has the attribute.
Definition: Function.h:321
A raw_ostream that writes to an SmallVector or SmallString.
Definition: raw_ostream.h:510
This class wraps the llvm.memset intrinsic.
static bool isEquality(Predicate P)
Return true if this predicate is either EQ or NE.
Value * CreateSExt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1572
Metadata node.
Definition: Metadata.h:864
static unsigned TypeSizeToSizeIndex(unsigned TypeSize)
F(f)
This class represents a sign extension of integer types.
uint64_t alignTo(uint64_t Value, uint64_t Align, uint64_t Skew=0)
Returns the next integer (mod 2**64) that is greater than or equal to Value and is a multiple of Alig...
Definition: MathExtras.h:685
CallInst * CreateMemSet(Value *Ptr, Value *Val, uint64_t Size, unsigned Align, bool isVolatile=false, MDNode *TBAATag=nullptr, MDNode *ScopeTag=nullptr, MDNode *NoAliasTag=nullptr)
Create and insert a memset to the specified pointer and the specified value.
Definition: IRBuilder.h:404
An instruction for reading from memory.
Definition: Instructions.h:168
AttrBuilder & addAttribute(Attribute::AttrKind Val)
Add an attribute to the builder.
an instruction that atomically reads a memory location, combines it with another value, and then stores the result back.
Definition: Instructions.h:681
Hexagon Common GEP
bool isVectorTy() const
True if this is an instance of VectorType.
Definition: Type.h:230
#define op(i)
bool isMustTailCall() const
static Type * getX86_MMXTy(LLVMContext &C)
Definition: Type.cpp:171
static cl::opt< unsigned long long > ClXorMask("msan-xor-mask", cl::desc("Define custom MSan XorMask"), cl::Hidden, cl::init(0))
Use * op_iterator
Definition: User.h:225
bool isPPC_FP128Ty() const
Return true if this is powerpc long double.
Definition: Type.h:159
static const MemoryMapParams Linux_AArch64_MemoryMapParams
static cl::opt< bool > ClHandleICmp("msan-handle-icmp", cl::desc("propagate shadow through ICmpEQ and ICmpNE"), cl::Hidden, cl::init(true))
op_iterator op_begin()
Definition: User.h:230
static PointerType * getInt64PtrTy(LLVMContext &C, unsigned AS=0)
Definition: Type.cpp:232
static Constant * get(ArrayType *T, ArrayRef< Constant *> V)
Definition: Constants.cpp:960
unsigned getBitWidth() const
Return the number of bits in the APInt.
Definition: APInt.h:1503
bool onlyReadsMemory() const
Determine if the call does not access or only reads memory.
static Constant * getNullValue(Type *Ty)
Constructor to create a &#39;0&#39; constant of arbitrary type.
Definition: Constants.cpp:268
Value * CreateNot(Value *V, const Twine &Name="")
Definition: IRBuilder.h:1294
StoreInst * CreateAlignedStore(Value *Val, Value *Ptr, unsigned Align, bool isVolatile=false)
Definition: IRBuilder.h:1358
IntegerType * getInt32Ty()
Fetch the type representing a 32-bit integer.
Definition: IRBuilder.h:347
unsigned countTrailingZeros() const
Count the number of trailing zero bits.
Definition: APInt.h:1626
static cl::opt< int > ClPoisonStackPattern("msan-poison-stack-pattern", cl::desc("poison uninitialized stack variables with the given pattern"), cl::Hidden, cl::init(0xff))
ArrayRef< unsigned > getIndices() const
AnalysisUsage & addRequired()
#define INITIALIZE_PASS_DEPENDENCY(depName)
Definition: PassSupport.h:51
bool isSigned() const
Definition: InstrTypes.h:811
static cl::opt< bool > ClDumpStrictInstructions("msan-dump-strict-instructions", cl::desc("print out instructions with default strict semantics"), cl::Hidden, cl::init(false))
This class represents the LLVM &#39;select&#39; instruction.
Type * getPointerElementType() const
Definition: Type.h:376
const DataLayout & getDataLayout() const
Get the data layout for the module&#39;s target platform.
Definition: Module.cpp:364
unsigned getAlignment() const
Return the alignment of the memory that is being allocated by the instruction.
Definition: Instructions.h:113
IntegerType * getInt64Ty()
Fetch the type representing a 64-bit integer.
Definition: IRBuilder.h:352
This is the base class for all instructions that perform data casts.
Definition: InstrTypes.h:348
The C convention as implemented on Windows/x86-64 and AArch64.
Definition: CallingConv.h:154
ArrayRef< T > makeArrayRef(const T &OneElt)
Construct an ArrayRef from a single element.
Definition: ArrayRef.h:451
&#39;undef&#39; values are things that do not have specified contents.
Definition: Constants.h:1282
This class wraps the llvm.memmove intrinsic.
Class to represent struct types.
Definition: DerivedTypes.h:201
LLVMContext & getContext() const
Get the global data context.
Definition: Module.h:243
static cl::opt< bool > ClCheckAccessAddress("msan-check-access-address", cl::desc("report accesses through a pointer which has poisoned shadow"), cl::Hidden, cl::init(true))
PointerType * getPointerTo(unsigned AddrSpace=0) const
Return a pointer to the current type.
Definition: Type.cpp:652
IterTy arg_end() const
Definition: CallSite.h:575
bool isUnsigned() const
Definition: InstrTypes.h:817
static cl::opt< unsigned long long > ClAndMask("msan-and-mask", cl::desc("Define custom MSan AndMask"), cl::Hidden, cl::init(0))
bool isIntegerTy() const
True if this is an instance of IntegerType.
Definition: Type.h:197
This provides a uniform API for creating instructions and inserting them into a basic block: either a...
Definition: IRBuilder.h:743
IntegerType * getIntPtrTy(const DataLayout &DL, unsigned AddrSpace=0)
Fetch the type representing a pointer to an integer value.
Definition: IRBuilder.h:390
This file contains the simple types necessary to represent the attributes associated with functions a...
static cl::opt< unsigned long long > ClOriginBase("msan-origin-base", cl::desc("Define custom MSan OriginBase"), cl::Hidden, cl::init(0))
InstrTy * getInstruction() const
Definition: CallSite.h:92
Value * CreateAdd(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:974
void setName(const Twine &Name)
Change the name of the value.
Definition: Value.cpp:295
static cl::opt< bool > ClHandleAsmConservative("msan-handle-asm-conservative", cl::desc("conservative handling of inline assembly"), cl::Hidden, cl::init(false))
static StructType * get(LLVMContext &Context, ArrayRef< Type *> Elements, bool isPacked=false)
This static method is the primary way to create a literal StructType.
Definition: Type.cpp:342
This file implements a class to represent arbitrary precision integral constant values and operations...
Type * getVoidTy()
Fetch the type representing void.
Definition: IRBuilder.h:380
This class represents a cast from a pointer to an integer.
AtomicOrdering
Atomic ordering for LLVM&#39;s memory model.
StoreInst * CreateStore(Value *Val, Value *Ptr, bool isVolatile=false)
Definition: IRBuilder.h:1333
Value * CreateIntToPtr(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1636
ValTy * getCalledValue() const
Return the pointer to function that is being called.
Definition: CallSite.h:100
bool isNullValue() const
Return true if this is the value that would be returned by getNullValue.
Definition: Constants.cpp:85
Class to represent function types.
Definition: DerivedTypes.h:103
Value * CreateBitCast(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1641
Type * getType() const
All values are typed, get the type of this value.
Definition: Value.h:245
This represents the llvm.va_start intrinsic.
static const char *const kMsanInitName
std::string itostr(int64_t X)
Definition: StringExtras.h:239
AtomicOrdering getSuccessOrdering() const
Returns the success ordering constraint of this cmpxchg instruction.
Definition: Instructions.h:572
#define T
static cl::opt< int > ClTrackOrigins("msan-track-origins", cl::desc("Track origins (allocation sites) of poisoned memory"), cl::Hidden, cl::init(0))
Track origins of uninitialized values.
Class to represent array types.
Definition: DerivedTypes.h:369
This instruction compares its operands according to the predicate given to the constructor.
static bool isStore(int Opcode)
bool isVarArg() const
Definition: DerivedTypes.h:123
This class represents a no-op cast from one type to another.
bool paramHasAttr(unsigned ArgNo, Attribute::AttrKind Kind) const
Return true if the call or the callee has the given attribute.
Definition: CallSite.h:377
MDNode * getMetadata(unsigned KindID) const
Get the metadata of given kind attached to this Instruction.
Definition: Instruction.h:221
Value * getInsertedValueOperand()
SmallString - A SmallString is just a SmallVector with methods and accessors that make it work better...
Definition: SmallString.h:26
Value * CreateSub(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:991
AttributeList getAttributes() const
Return the attribute list for this Function.
Definition: Function.h:224
An instruction for storing to memory.
Definition: Instructions.h:310
bool isIntOrIntVectorTy() const
Return true if this is an integer type or a vector of integer types.
Definition: Type.h:203
static const unsigned kParamTLSSize
Value * CreateZExt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1568
ConstraintPrefix Type
Type - The basic type of the constraint: input/output/clobber.
Definition: InlineAsm.h:121
static cl::opt< bool > ClPoisonStack("msan-poison-stack", cl::desc("poison uninitialized stack variables"), cl::Hidden, cl::init(true))
Function * getDeclaration(Module *M, ID id, ArrayRef< Type *> Tys=None)
Create or insert an LLVM Function declaration for an intrinsic, and return it.
Definition: Function.cpp:1021
This class represents a truncation of integer types.
void SetInsertPoint(BasicBlock *TheBB)
This specifies that created instructions should be appended to the end of the specified block...
Definition: IRBuilder.h:127
Value * getOperand(unsigned i) const
Definition: User.h:170
static const PlatformMemoryMapParams Linux_MIPS_MemoryMapParams
bool isRelational() const
Return true if the predicate is relational (not EQ or NE).
bool isCall() const
Return true if a CallInst is enclosed.
Definition: CallSite.h:87
Value * CreateOr(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1142
static const unsigned kMinOriginAlignment
Constant * getAggregateElement(unsigned Elt) const
For aggregates (struct/array/vector) return the constant that corresponds to the specified element if...
Definition: Constants.cpp:338
bool doesNotAccessMemory() const
Determine if the call does not access memory.
Type * getScalarType() const
If this is a vector type, return the element type, otherwise return &#39;this&#39;.
Definition: Type.h:304
Value * getOperand(unsigned i_nocapture) const
bool isZeroValue() const
Return true if the value is negative zero or null value.
Definition: Constants.cpp:65
bool isVoidTy() const
Return true if this is &#39;void&#39;.
Definition: Type.h:141
const BasicBlock & getEntryBlock() const
Definition: Function.h:640
an instruction for type-safe pointer arithmetic to access elements of arrays and structs ...
Definition: Instructions.h:843
LoadInst * CreateLoad(Value *Ptr, const char *Name)
Provided to resolve &#39;CreateLoad(Ptr, "...")&#39; correctly, instead of converting the string to &#39;bool&#39; fo...
Definition: IRBuilder.h:1317
static cl::opt< unsigned long long > ClShadowBase("msan-shadow-base", cl::desc("Define custom MSan ShadowBase"), cl::Hidden, cl::init(0))
static bool runOnFunction(Function &F, bool PostInlining)
initializer< Ty > init(const Ty &Val)
Definition: CommandLine.h:419
unsigned getNumOperands() const
This instruction inserts a single (scalar) element into a VectorType value.
The landingpad instruction holds all of the information necessary to generate correct exception handl...
const Instruction * getFirstNonPHI() const
Returns a pointer to the first instruction in this block that is not a PHINode instruction.
Definition: BasicBlock.cpp:190
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
const_iterator getFirstInsertionPt() const
Returns an iterator to the first instruction in this block that is suitable for inserting a non-PHI i...
Definition: BasicBlock.cpp:219
const BasicBlock * getSinglePredecessor() const
Return the predecessor of this block if it has a single predecessor block.
Definition: BasicBlock.cpp:236
INITIALIZE_PASS_BEGIN(MemorySanitizer, "msan", "MemorySanitizer: detects uninitialized reads.", false, false) INITIALIZE_PASS_END(MemorySanitizer
LLVM Basic Block Representation.
Definition: BasicBlock.h:58
The instances of the Type class are immutable: once they are created, they are never changed...
Definition: Type.h:46
std::pair< Function *, Function * > createSanitizerCtorAndInitFunctions(Module &M, StringRef CtorName, StringRef InitName, ArrayRef< Type *> InitArgTypes, ArrayRef< Value *> InitArgs, StringRef VersionCheckName=StringRef())
Creates sanitizer constructor function, and calls sanitizer&#39;s init function from it.
This is an important class for using LLVM in a threaded context.
Definition: LLVMContext.h:69
const char * getOpcodeName() const
Definition: Instruction.h:128
This is an important base class in LLVM.
Definition: Constant.h:42
Resume the propagation of an exception.
This file contains the declarations for the subclasses of Constant, which represent the different fla...
Value * CreateSelect(Value *C, Value *True, Value *False, const Twine &Name="", Instruction *MDFrom=nullptr)
Definition: IRBuilder.h:1913
bool isPointerTy() const
True if this is an instance of PointerType.
Definition: Type.h:224
unsigned getNumParams() const
Return the number of fixed parameters this function type requires.
Definition: DerivedTypes.h:139
Represent the analysis usage information of a pass.
op_iterator op_end()
Definition: User.h:232
Value * CreateNeg(Value *V, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1268
static const PlatformMemoryMapParams Linux_X86_MemoryMapParams
This instruction compares its operands according to the predicate given to the constructor.
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
Definition: InstrTypes.h:641
FunctionPass class - This class is used to implement most global optimizations.
Definition: Pass.h:285
static const unsigned kShadowTLSAlignment
static FunctionType * get(Type *Result, ArrayRef< Type *> Params, bool isVarArg)
This static method is the primary way of constructing a FunctionType.
Definition: Type.cpp:297
static Constant * get(StructType *T, ArrayRef< Constant *> V)
Definition: Constants.cpp:1021
Value * getPointerOperand()
Definition: Instructions.h:274
bool isX86_MMXTy() const
Return true if this is X86 MMX.
Definition: Type.h:182
Value * CreateICmpEQ(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1748
self_iterator getIterator()
Definition: ilist_node.h:82
Class to represent integer types.
Definition: DerivedTypes.h:40
IntegerType * getIntNTy(unsigned N)
Fetch the type representing an N-bit integer.
Definition: IRBuilder.h:360
Value * CreateExtractElement(Value *Vec, Value *Idx, const Twine &Name="")
Definition: IRBuilder.h:1933
This class represents a cast from an integer to a pointer.
const Value * getCondition() const
static Constant * getAllOnesValue(Type *Ty)
Definition: Constants.cpp:322
static const PlatformMemoryMapParams Linux_ARM_MemoryMapParams
static const MemoryMapParams FreeBSD_X86_64_MemoryMapParams
Comdat * getOrInsertComdat(StringRef Name)
Return the Comdat in the module with the specified name.
Definition: Module.cpp:477
const Value * getArraySize() const
Get the number of elements allocated.
Definition: Instructions.h:93
Value * CreateExtractValue(Value *Agg, ArrayRef< unsigned > Idxs, const Twine &Name="")
Definition: IRBuilder.h:1975
size_t size() const
Definition: SmallVector.h:53
static wasm::ValType getType(const TargetRegisterClass *RC)
PointerType * getInt8PtrTy(unsigned AddrSpace=0)
Fetch the type representing a pointer to an 8-bit integer value.
Definition: IRBuilder.h:385
MDNode * createBranchWeights(uint32_t TrueWeight, uint32_t FalseWeight)
Return metadata containing two branch weights.
Definition: MDBuilder.cpp:38
AtomicOrdering getOrdering() const
Returns the ordering constraint of this rmw instruction.
Definition: Instructions.h:763
INITIALIZE_PASS_END(RegBankSelect, DEBUG_TYPE, "Assign register bank of generic virtual registers", false, false) RegBankSelect
#define llvm_unreachable(msg)
Marks that the current location is not supposed to be reachable.
Value * CreateMul(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1008
Value * CreateTrunc(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1564
Type * getAllocatedType() const
Return the type that is being allocated by the instruction.
Definition: Instructions.h:106
Triple - Helper class for working with autoconf configuration names.
Definition: Triple.h:44
signed greater than
Definition: InstrTypes.h:668
std::vector< ConstraintInfo > ConstraintInfoVector
Definition: InlineAsm.h:116
unsigned first
bool isInvoke() const
Return true if a InvokeInst is enclosed.
Definition: CallSite.h:90
Intrinsic::ID getIntrinsicID() const
Return the intrinsic ID of this intrinsic.
Definition: IntrinsicInst.h:51
PHINode * CreatePHI(Type *Ty, unsigned NumReservedValues, const Twine &Name="")
Definition: IRBuilder.h:1878
Value * CreateGEP(Value *Ptr, ArrayRef< Value *> IdxList, const Twine &Name="")
Definition: IRBuilder.h:1386
bool isPtrOrPtrVectorTy() const
Return true if this is a pointer type or a vector of pointer types.
Definition: Type.h:227
static IntegerType * get(LLVMContext &C, unsigned NumBits)
This static method is the primary way of constructing an IntegerType.
Definition: Type.cpp:240
Type * getSequentialElementType() const
Definition: Type.h:358
Iterator for intrusive lists based on ilist_node.
unsigned getNumOperands() const
Definition: User.h:192
See the file comment.
Definition: ValueMap.h:86
This is the shared class of boolean and integer constants.
Definition: Constants.h:84
auto size(R &&Range, typename std::enable_if< std::is_same< typename std::iterator_traits< decltype(Range.begin())>::iterator_category, std::random_access_iterator_tag >::value, void >::type *=nullptr) -> decltype(std::distance(Range.begin(), Range.end()))
Get the size of a range.
Definition: STLExtras.h:1029
iterator end()
Definition: BasicBlock.h:265
unsigned getScalarSizeInBits() const LLVM_READONLY
If this is a vector type, return the getPrimitiveSizeInBits value for the element type...
Definition: Type.cpp:130
CallingConv::ID getCallingConv() const
getCallingConv()/setCallingConv(CC) - These method get and set the calling convention of this functio...
Definition: Function.h:213
IterTy arg_begin() const
Definition: CallSite.h:571
Value * CreateIntCast(Value *V, Type *DestTy, bool isSigned, const Twine &Name="")
Definition: IRBuilder.h:1710
This is a &#39;vector&#39; (really, a variable-sized array), optimized for the case when the array is small...
Definition: SmallVector.h:847
Module.h This file contains the declarations for the Module class.
Value * CreateInsertElement(Value *Vec, Value *NewElt, Value *Idx, const Twine &Name="")
Definition: IRBuilder.h:1946
Provides information about what library functions are available for the current target.
unsigned getABITypeAlignment(Type *Ty) const
Returns the minimum ABI-required alignment for the specified type.
Definition: DataLayout.cpp:722
bool isAggregateType() const
Return true if the type is an aggregate type.
Definition: Type.h:258
signed less than
Definition: InstrTypes.h:670
CallInst * CreateMaskedStore(Value *Val, Value *Ptr, unsigned Align, Value *Mask)
Create a call to Masked Store intrinsic.
Definition: IRBuilder.cpp:492
CHAIN = SC CHAIN, Imm128 - System call.
ConstantInt * getInt32(uint32_t C)
Get a constant 32-bit value.
Definition: IRBuilder.h:307
static IntegerType * getIntNTy(LLVMContext &C, unsigned N)
Definition: Type.cpp:180
StringRef str()
Return a StringRef for the vector contents.
Definition: raw_ostream.h:535
static GCRegistry::Add< StatepointGC > D("statepoint-example", "an example strategy for statepoint")
CallInst * CreateMemCpy(Value *Dst, unsigned DstAlign, Value *Src, unsigned SrcAlign, uint64_t Size, bool isVolatile=false, MDNode *TBAATag=nullptr, MDNode *TBAAStructTag=nullptr, MDNode *ScopeTag=nullptr, MDNode *NoAliasTag=nullptr)
Create and insert a memcpy between the specified pointers.
Definition: IRBuilder.h:446
This class wraps the llvm.memcpy intrinsic.
static Constant * get(Type *Ty, uint64_t V, bool isSigned=false)
If Ty is a vector type, return a Constant with a splat of the given value.
Definition: Constants.cpp:621
void appendToGlobalCtors(Module &M, Function *F, int Priority, Constant *Data=nullptr)
Append F to the list of global ctors of module M with the given Priority.
Definition: ModuleUtils.cpp:84
CallInst * CreateMaskedLoad(Value *Ptr, unsigned Align, Value *Mask, Value *PassThru=nullptr, const Twine &Name="")
Create a call to Masked Load intrinsic.
Definition: IRBuilder.cpp:471
unsigned getNumIncomingValues() const
Return the number of incoming edges.
static const size_t kNumberOfAccessSizes
static GlobalVariable * createPrivateNonConstGlobalForString(Module &M, StringRef Str)
Create a non-const global initialized with the given string.
static cl::opt< bool > ClKeepGoing("msan-keep-going", cl::desc("keep going after reporting a UMR"), cl::Hidden, cl::init(false))
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
Definition: Debug.cpp:133
unsigned getVectorNumElements() const
Definition: DerivedTypes.h:462
signed less or equal
Definition: InstrTypes.h:671
Class to represent vector types.
Definition: DerivedTypes.h:393
const Module * getModule() const
Return the module owning the function this instruction belongs to or nullptr it the function does not...
Definition: Instruction.cpp:56
Class for arbitrary precision integers.
Definition: APInt.h:70
IntegerType * getInt8Ty()
Fetch the type representing an 8-bit integer.
Definition: IRBuilder.h:337
Value * CreateShl(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1063
const Value * getFalseValue() const
Value * CreatePointerCast(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1687
Instruction * SplitBlockAndInsertIfThen(Value *Cond, Instruction *SplitBefore, bool Unreachable, MDNode *BranchWeights=nullptr, DominatorTree *DT=nullptr, LoopInfo *LI=nullptr)
Split the containing block at the specified instruction - everything before SplitBefore stays in the ...
uint64_t getTypeSizeInBits(Type *Ty) const
Size examples:
Definition: DataLayout.h:560
static cl::opt< ITMode > IT(cl::desc("IT block support"), cl::Hidden, cl::init(DefaultIT), cl::ZeroOrMore, cl::values(clEnumValN(DefaultIT, "arm-default-it", "Generate IT block based on arch"), clEnumValN(RestrictedIT, "arm-restrict-it", "Disallow deprecated IT based on ARMv8"), clEnumValN(NoRestrictedIT, "arm-no-restrict-it", "Allow IT blocks based on ARMv7")))
uint64_t getTypeAllocSize(Type *Ty) const
Returns the offset in bytes between successive objects of the specified type, including alignment pad...
Definition: DataLayout.h:428
void removeAttributes(unsigned i, const AttrBuilder &Attrs)
removes the attributes from the list of attributes.
Definition: Function.cpp:417
Predicate getPredicate() const
Return the predicate for this instruction.
Definition: InstrTypes.h:716
static cl::opt< bool > ClEnableKmsan("msan-kernel", cl::desc("Enable KernelMemorySanitizer instrumentation"), cl::Hidden, cl::init(false))
bool isInlineAsm() const
Check if this call is an inline asm statement.
static cl::opt< bool > ClHandleICmpExact("msan-handle-icmp-exact", cl::desc("exact handling of relational integer ICmp"), cl::Hidden, cl::init(false))
unsigned getAlignment() const
Return the alignment of the access that is being performed.
Definition: Instructions.h:230
LLVM_NODISCARD bool empty() const
Definition: SmallVector.h:56
unsigned getNumArgOperands() const
Return the number of call arguments.
StringRef getName() const
Return a constant reference to the value&#39;s name.
Definition: Value.cpp:224
#define I(x, y, z)
Definition: MD5.cpp:58
#define N
static const MemoryMapParams Linux_X86_64_MemoryMapParams
static ArrayType * get(Type *ElementType, uint64_t NumElements)
This static method is the primary way to construct an ArrayType.
Definition: Type.cpp:581
LLVM_NODISCARD std::enable_if<!is_simple_type< Y >::value, typename cast_retty< X, const Y >::ret_type >::type dyn_cast(const Y &Val)
Definition: Casting.h:323
This instruction extracts a single (scalar) element from a VectorType value.
uint32_t Size
Definition: Profile.cpp:47
void maybeMarkSanitizerLibraryCallNoBuiltin(CallInst *CI, const TargetLibraryInfo *TLI)
Given a CallInst, check if it calls a string function known to CodeGen, and mark it with NoBuiltin if...
Definition: Local.cpp:2834
static const PlatformMemoryMapParams NetBSD_X86_MemoryMapParams
Value * getReturnValue() const
Convenience accessor. Returns null if there is no return value.
static InlineAsm * get(FunctionType *Ty, StringRef AsmString, StringRef Constraints, bool hasSideEffects, bool isAlignStack=false, AsmDialect asmDialect=AD_ATT)
InlineAsm::get - Return the specified uniqued inline asm string.
Definition: InlineAsm.cpp:43
static cl::opt< bool > ClCheckConstantShadow("msan-check-constant-shadow", cl::desc("Insert checks for constant shadow values"), cl::Hidden, cl::init(false))
static const MemoryMapParams Linux_MIPS64_MemoryMapParams
Value * CreateAnd(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1124
Value * CreatePtrToInt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1631
static const unsigned kOriginSize
const std::string to_string(const T &Value)
Definition: ScopedPrinter.h:62
static const MemoryMapParams FreeBSD_I386_MemoryMapParams
bool isFPOrFPVectorTy() const
Return true if this is a FP type or a vector of FP.
Definition: Type.h:185
iterator_range< df_iterator< T > > depth_first(const T &G)
This represents the llvm.va_copy intrinsic.
bool isArrayAllocation() const
Return true if there is an allocation size parameter to the allocation instruction that is not 1...
size_type count(const KeyT &Val) const
Return 1 if the specified key is in the map, 0 otherwise.
Definition: ValueMap.h:158
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
Value * getArgOperand(unsigned i) const
getArgOperand/setArgOperand - Return/set the i-th call argument.
LoadInst * CreateAlignedLoad(Value *Ptr, unsigned Align, const char *Name)
Provided to resolve &#39;CreateAlignedLoad(Ptr, Align, "...")&#39; correctly, instead of converting the strin...
Definition: IRBuilder.h:1340
static cl::opt< int > ClInstrumentationWithCallThreshold("msan-instrumentation-with-call-threshold", cl::desc("If the function being instrumented requires more than " "this number of checks and origin stores, use callbacks instead of " "inline checks (-1 means never use callbacks)."), cl::Hidden, cl::init(3500))
unsigned getPrimitiveSizeInBits() const LLVM_READONLY
Return the basic size of this type if it is a primitive type.
Definition: Type.cpp:115
void setSuccessOrdering(AtomicOrdering Ordering)
Sets the success ordering constraint of this cmpxchg instruction.
Definition: Instructions.h:577
ArrayRef< unsigned > getIndices() const
Module * getParent()
Get the module that this global value is contained inside of...
Definition: GlobalValue.h:566
LLVM Value Representation.
Definition: Value.h:73
uint64_t getTypeStoreSize(Type *Ty) const
Returns the maximum number of bytes that may be overwritten by storing the specified type...
Definition: DataLayout.h:411
FunctionType * getFunctionType() const
Definition: CallSite.h:320
static VectorType * get(Type *ElementType, unsigned NumElements)
This static method is the primary way to construct an VectorType.
Definition: Type.cpp:606
std::underlying_type< E >::type Mask()
Get a bitmask with 1s in all places up to the high-order bit of E&#39;s largest value.
Definition: BitmaskEnum.h:81
unsigned getArgumentNo(Value::const_user_iterator I) const
Given a value use iterator, returns the argument that corresponds to it.
Definition: CallSite.h:199
AttributeSet getFnAttributes() const
The function attributes are returned.
BasicBlock::iterator GetInsertPoint() const
Definition: IRBuilder.h:122
BasicBlock * SplitBlock(BasicBlock *Old, Instruction *SplitPt, DominatorTree *DT=nullptr, LoopInfo *LI=nullptr, MemorySSAUpdater *MSSAU=nullptr)
Split the specified block at the specified instruction - everything before SplitPt stays in Old and e...
Value * CreateLShr(Value *LHS, Value *RHS, const Twine &Name="", bool isExact=false)
Definition: IRBuilder.h:1084
const Value * getCalledValue() const
Get a pointer to the function that is invoked by this instruction.
static cl::opt< bool > ClPoisonUndef("msan-poison-undef", cl::desc("poison undef temps"), cl::Hidden, cl::init(true))
ConstantInt * getInt8(uint8_t C)
Get a constant 8-bit value.
Definition: IRBuilder.h:297
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:49
Predicate getSwappedPredicate() const
For example, EQ->EQ, SLE->SGE, ULT->UGT, OEQ->OEQ, ULE->UGE, OLT->OGT, etc.
Definition: InstrTypes.h:756
static const PlatformMemoryMapParams FreeBSD_X86_MemoryMapParams
Type * getArrayElementType() const
Definition: Type.h:365
FunctionPass * createMemorySanitizerPass(int TrackOrigins=0, bool Recover=false, bool EnableKmsan=false)
Value * CreateInsertValue(Value *Agg, Value *Val, ArrayRef< unsigned > Idxs, const Twine &Name="")
Definition: IRBuilder.h:1983
bool isBigEndian() const
Definition: DataLayout.h:222
#define LLVM_DEBUG(X)
Definition: Debug.h:123
Value * getPointerOperand()
Definition: Instructions.h:402
static IntegerType * getInt8Ty(LLVMContext &C)
Definition: Type.cpp:174
static const char *const kMsanModuleCtorName
static Constant * get(ArrayRef< Constant *> V)
Definition: Constants.cpp:1056
iterator_range< arg_iterator > args()
Definition: Function.h:689
signed greater or equal
Definition: InstrTypes.h:669
static ConstraintInfoVector ParseConstraints(StringRef ConstraintString)
ParseConstraints - Split up the constraint string into the specific constraints and their prefixes...
Definition: InlineAsm.cpp:208
bool isArrayTy() const
True if this is an instance of ArrayType.
Definition: Type.h:221
Type * getContainedType(unsigned i) const
This method is used to implement the type iterator (defined at the end of the file).
Definition: Type.h:333
A wrapper class for inspecting calls to intrinsic functions.
Definition: IntrinsicInst.h:44
const BasicBlock * getParent() const
Definition: Instruction.h:67
an instruction to allocate memory on the stack
Definition: Instructions.h:60
This instruction inserts a struct field of array element value into an aggregate value.
CallInst * CreateCall(Value *Callee, ArrayRef< Value *> Args=None, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:1883