LLVM  10.0.0svn
MemorySanitizer.cpp
Go to the documentation of this file.
1 //===- MemorySanitizer.cpp - detector of uninitialized reads --------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 /// \file
10 /// This file is a part of MemorySanitizer, a detector of uninitialized
11 /// reads.
12 ///
13 /// The algorithm of the tool is similar to Memcheck
14 /// (http://goo.gl/QKbem). We associate a few shadow bits with every
15 /// byte of the application memory, poison the shadow of the malloc-ed
16 /// or alloca-ed memory, load the shadow bits on every memory read,
17 /// propagate the shadow bits through some of the arithmetic
18 /// instruction (including MOV), store the shadow bits on every memory
19 /// write, report a bug on some other instructions (e.g. JMP) if the
20 /// associated shadow is poisoned.
21 ///
22 /// But there are differences too. The first and the major one:
23 /// compiler instrumentation instead of binary instrumentation. This
24 /// gives us much better register allocation, possible compiler
25 /// optimizations and a fast start-up. But this brings the major issue
26 /// as well: msan needs to see all program events, including system
27 /// calls and reads/writes in system libraries, so we either need to
28 /// compile *everything* with msan or use a binary translation
29 /// component (e.g. DynamoRIO) to instrument pre-built libraries.
30 /// Another difference from Memcheck is that we use 8 shadow bits per
31 /// byte of application memory and use a direct shadow mapping. This
32 /// greatly simplifies the instrumentation code and avoids races on
33 /// shadow updates (Memcheck is single-threaded so races are not a
34 /// concern there. Memcheck uses 2 shadow bits per byte with a slow
35 /// path storage that uses 8 bits per byte).
36 ///
37 /// The default value of shadow is 0, which means "clean" (not poisoned).
38 ///
39 /// Every module initializer should call __msan_init to ensure that the
40 /// shadow memory is ready. On error, __msan_warning is called. Since
41 /// parameters and return values may be passed via registers, we have a
42 /// specialized thread-local shadow for return values
43 /// (__msan_retval_tls) and parameters (__msan_param_tls).
44 ///
45 /// Origin tracking.
46 ///
47 /// MemorySanitizer can track origins (allocation points) of all uninitialized
48 /// values. This behavior is controlled with a flag (msan-track-origins) and is
49 /// disabled by default.
50 ///
51 /// Origins are 4-byte values created and interpreted by the runtime library.
52 /// They are stored in a second shadow mapping, one 4-byte value for 4 bytes
53 /// of application memory. Propagation of origins is basically a bunch of
54 /// "select" instructions that pick the origin of a dirty argument, if an
55 /// instruction has one.
56 ///
57 /// Every 4 aligned, consecutive bytes of application memory have one origin
58 /// value associated with them. If these bytes contain uninitialized data
59 /// coming from 2 different allocations, the last store wins. Because of this,
60 /// MemorySanitizer reports can show unrelated origins, but this is unlikely in
61 /// practice.
62 ///
63 /// Origins are meaningless for fully initialized values, so MemorySanitizer
64 /// avoids storing origin to memory when a fully initialized value is stored.
65 /// This way it avoids needless overwritting origin of the 4-byte region on
66 /// a short (i.e. 1 byte) clean store, and it is also good for performance.
67 ///
68 /// Atomic handling.
69 ///
70 /// Ideally, every atomic store of application value should update the
71 /// corresponding shadow location in an atomic way. Unfortunately, atomic store
72 /// of two disjoint locations can not be done without severe slowdown.
73 ///
74 /// Therefore, we implement an approximation that may err on the safe side.
75 /// In this implementation, every atomically accessed location in the program
76 /// may only change from (partially) uninitialized to fully initialized, but
77 /// not the other way around. We load the shadow _after_ the application load,
78 /// and we store the shadow _before_ the app store. Also, we always store clean
79 /// shadow (if the application store is atomic). This way, if the store-load
80 /// pair constitutes a happens-before arc, shadow store and load are correctly
81 /// ordered such that the load will get either the value that was stored, or
82 /// some later value (which is always clean).
83 ///
84 /// This does not work very well with Compare-And-Swap (CAS) and
85 /// Read-Modify-Write (RMW) operations. To follow the above logic, CAS and RMW
86 /// must store the new shadow before the app operation, and load the shadow
87 /// after the app operation. Computers don't work this way. Current
88 /// implementation ignores the load aspect of CAS/RMW, always returning a clean
89 /// value. It implements the store part as a simple atomic store by storing a
90 /// clean shadow.
91 ///
92 /// Instrumenting inline assembly.
93 ///
94 /// For inline assembly code LLVM has little idea about which memory locations
95 /// become initialized depending on the arguments. It can be possible to figure
96 /// out which arguments are meant to point to inputs and outputs, but the
97 /// actual semantics can be only visible at runtime. In the Linux kernel it's
98 /// also possible that the arguments only indicate the offset for a base taken
99 /// from a segment register, so it's dangerous to treat any asm() arguments as
100 /// pointers. We take a conservative approach generating calls to
101 /// __msan_instrument_asm_store(ptr, size)
102 /// , which defer the memory unpoisoning to the runtime library.
103 /// The latter can perform more complex address checks to figure out whether
104 /// it's safe to touch the shadow memory.
105 /// Like with atomic operations, we call __msan_instrument_asm_store() before
106 /// the assembly call, so that changes to the shadow memory will be seen by
107 /// other threads together with main memory initialization.
108 ///
109 /// KernelMemorySanitizer (KMSAN) implementation.
110 ///
111 /// The major differences between KMSAN and MSan instrumentation are:
112 /// - KMSAN always tracks the origins and implies msan-keep-going=true;
113 /// - KMSAN allocates shadow and origin memory for each page separately, so
114 /// there are no explicit accesses to shadow and origin in the
115 /// instrumentation.
116 /// Shadow and origin values for a particular X-byte memory location
117 /// (X=1,2,4,8) are accessed through pointers obtained via the
118 /// __msan_metadata_ptr_for_load_X(ptr)
119 /// __msan_metadata_ptr_for_store_X(ptr)
120 /// functions. The corresponding functions check that the X-byte accesses
121 /// are possible and returns the pointers to shadow and origin memory.
122 /// Arbitrary sized accesses are handled with:
123 /// __msan_metadata_ptr_for_load_n(ptr, size)
124 /// __msan_metadata_ptr_for_store_n(ptr, size);
125 /// - TLS variables are stored in a single per-task struct. A call to a
126 /// function __msan_get_context_state() returning a pointer to that struct
127 /// is inserted into every instrumented function before the entry block;
128 /// - __msan_warning() takes a 32-bit origin parameter;
129 /// - local variables are poisoned with __msan_poison_alloca() upon function
130 /// entry and unpoisoned with __msan_unpoison_alloca() before leaving the
131 /// function;
132 /// - the pass doesn't declare any global variables or add global constructors
133 /// to the translation unit.
134 ///
135 /// Also, KMSAN currently ignores uninitialized memory passed into inline asm
136 /// calls, making sure we're on the safe side wrt. possible false positives.
137 ///
138 /// KernelMemorySanitizer only supports X86_64 at the moment.
139 ///
140 //===----------------------------------------------------------------------===//
141 
143 #include "llvm/ADT/APInt.h"
144 #include "llvm/ADT/ArrayRef.h"
146 #include "llvm/ADT/SmallSet.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"
153 #include "llvm/IR/Argument.h"
154 #include "llvm/IR/Attributes.h"
155 #include "llvm/IR/BasicBlock.h"
156 #include "llvm/IR/CallSite.h"
157 #include "llvm/IR/CallingConv.h"
158 #include "llvm/IR/Constant.h"
159 #include "llvm/IR/Constants.h"
160 #include "llvm/IR/DataLayout.h"
161 #include "llvm/IR/DerivedTypes.h"
162 #include "llvm/IR/Function.h"
163 #include "llvm/IR/GlobalValue.h"
164 #include "llvm/IR/GlobalVariable.h"
165 #include "llvm/IR/IRBuilder.h"
166 #include "llvm/IR/InlineAsm.h"
167 #include "llvm/IR/InstVisitor.h"
168 #include "llvm/IR/InstrTypes.h"
169 #include "llvm/IR/Instruction.h"
170 #include "llvm/IR/Instructions.h"
171 #include "llvm/IR/IntrinsicInst.h"
172 #include "llvm/IR/Intrinsics.h"
173 #include "llvm/IR/LLVMContext.h"
174 #include "llvm/IR/MDBuilder.h"
175 #include "llvm/IR/Module.h"
176 #include "llvm/IR/Type.h"
177 #include "llvm/IR/Value.h"
178 #include "llvm/IR/ValueMap.h"
179 #include "llvm/Pass.h"
181 #include "llvm/Support/Casting.h"
183 #include "llvm/Support/Compiler.h"
184 #include "llvm/Support/Debug.h"
186 #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 
252  "msan-handle-lifetime-intrinsics",
253  cl::desc(
254  "when possible, poison scoped variables at the beginning of the scope "
255  "(slower, but more precise)"),
256  cl::Hidden, cl::init(true));
257 
258 // When compiling the Linux kernel, we sometimes see false positives related to
259 // MSan being unable to understand that inline assembly calls may initialize
260 // local variables.
261 // This flag makes the compiler conservatively unpoison every memory location
262 // passed into an assembly call. Note that this may cause false positives.
263 // Because it's impossible to figure out the array sizes, we can only unpoison
264 // the first sizeof(type) bytes for each type* pointer.
265 // The instrumentation is only enabled in KMSAN builds, and only if
266 // -msan-handle-asm-conservative is on. This is done because we may want to
267 // quickly disable assembly instrumentation when it breaks.
269  "msan-handle-asm-conservative",
270  cl::desc("conservative handling of inline assembly"), cl::Hidden,
271  cl::init(true));
272 
273 // This flag controls whether we check the shadow of the address
274 // operand of load or store. Such bugs are very rare, since load from
275 // a garbage address typically results in SEGV, but still happen
276 // (e.g. only lower bits of address are garbage, or the access happens
277 // early at program startup where malloc-ed memory is more likely to
278 // be zeroed. As of 2012-08-28 this flag adds 20% slowdown.
279 static cl::opt<bool> ClCheckAccessAddress("msan-check-access-address",
280  cl::desc("report accesses through a pointer which has poisoned shadow"),
281  cl::Hidden, cl::init(true));
282 
283 static cl::opt<bool> ClDumpStrictInstructions("msan-dump-strict-instructions",
284  cl::desc("print out instructions with default strict semantics"),
285  cl::Hidden, cl::init(false));
286 
288  "msan-instrumentation-with-call-threshold",
289  cl::desc(
290  "If the function being instrumented requires more than "
291  "this number of checks and origin stores, use callbacks instead of "
292  "inline checks (-1 means never use callbacks)."),
293  cl::Hidden, cl::init(3500));
294 
295 static cl::opt<bool>
296  ClEnableKmsan("msan-kernel",
297  cl::desc("Enable KernelMemorySanitizer instrumentation"),
298  cl::Hidden, cl::init(false));
299 
300 // This is an experiment to enable handling of cases where shadow is a non-zero
301 // compile-time constant. For some unexplainable reason they were silently
302 // ignored in the instrumentation.
303 static cl::opt<bool> ClCheckConstantShadow("msan-check-constant-shadow",
304  cl::desc("Insert checks for constant shadow values"),
305  cl::Hidden, cl::init(false));
306 
307 // This is off by default because of a bug in gold:
308 // https://sourceware.org/bugzilla/show_bug.cgi?id=19002
309 static cl::opt<bool> ClWithComdat("msan-with-comdat",
310  cl::desc("Place MSan constructors in comdat sections"),
311  cl::Hidden, cl::init(false));
312 
313 // These options allow to specify custom memory map parameters
314 // See MemoryMapParams for details.
315 static cl::opt<uint64_t> ClAndMask("msan-and-mask",
316  cl::desc("Define custom MSan AndMask"),
317  cl::Hidden, cl::init(0));
318 
319 static cl::opt<uint64_t> ClXorMask("msan-xor-mask",
320  cl::desc("Define custom MSan XorMask"),
321  cl::Hidden, cl::init(0));
322 
323 static cl::opt<uint64_t> ClShadowBase("msan-shadow-base",
324  cl::desc("Define custom MSan ShadowBase"),
325  cl::Hidden, cl::init(0));
326 
327 static cl::opt<uint64_t> ClOriginBase("msan-origin-base",
328  cl::desc("Define custom MSan OriginBase"),
329  cl::Hidden, cl::init(0));
330 
331 static const char *const kMsanModuleCtorName = "msan.module_ctor";
332 static const char *const kMsanInitName = "__msan_init";
333 
334 namespace {
335 
336 // Memory map parameters used in application-to-shadow address calculation.
337 // Offset = (Addr & ~AndMask) ^ XorMask
338 // Shadow = ShadowBase + Offset
339 // Origin = OriginBase + Offset
340 struct MemoryMapParams {
341  uint64_t AndMask;
342  uint64_t XorMask;
343  uint64_t ShadowBase;
344  uint64_t OriginBase;
345 };
346 
347 struct PlatformMemoryMapParams {
348  const MemoryMapParams *bits32;
349  const MemoryMapParams *bits64;
350 };
351 
352 } // end anonymous namespace
353 
354 // i386 Linux
355 static const MemoryMapParams Linux_I386_MemoryMapParams = {
356  0x000080000000, // AndMask
357  0, // XorMask (not used)
358  0, // ShadowBase (not used)
359  0x000040000000, // OriginBase
360 };
361 
362 // x86_64 Linux
363 static const MemoryMapParams Linux_X86_64_MemoryMapParams = {
364 #ifdef MSAN_LINUX_X86_64_OLD_MAPPING
365  0x400000000000, // AndMask
366  0, // XorMask (not used)
367  0, // ShadowBase (not used)
368  0x200000000000, // OriginBase
369 #else
370  0, // AndMask (not used)
371  0x500000000000, // XorMask
372  0, // ShadowBase (not used)
373  0x100000000000, // OriginBase
374 #endif
375 };
376 
377 // mips64 Linux
378 static const MemoryMapParams Linux_MIPS64_MemoryMapParams = {
379  0, // AndMask (not used)
380  0x008000000000, // XorMask
381  0, // ShadowBase (not used)
382  0x002000000000, // OriginBase
383 };
384 
385 // ppc64 Linux
386 static const MemoryMapParams Linux_PowerPC64_MemoryMapParams = {
387  0xE00000000000, // AndMask
388  0x100000000000, // XorMask
389  0x080000000000, // ShadowBase
390  0x1C0000000000, // OriginBase
391 };
392 
393 // aarch64 Linux
394 static const MemoryMapParams Linux_AArch64_MemoryMapParams = {
395  0, // AndMask (not used)
396  0x06000000000, // XorMask
397  0, // ShadowBase (not used)
398  0x01000000000, // OriginBase
399 };
400 
401 // i386 FreeBSD
402 static const MemoryMapParams FreeBSD_I386_MemoryMapParams = {
403  0x000180000000, // AndMask
404  0x000040000000, // XorMask
405  0x000020000000, // ShadowBase
406  0x000700000000, // OriginBase
407 };
408 
409 // x86_64 FreeBSD
410 static const MemoryMapParams FreeBSD_X86_64_MemoryMapParams = {
411  0xc00000000000, // AndMask
412  0x200000000000, // XorMask
413  0x100000000000, // ShadowBase
414  0x380000000000, // OriginBase
415 };
416 
417 // x86_64 NetBSD
418 static const MemoryMapParams NetBSD_X86_64_MemoryMapParams = {
419  0, // AndMask
420  0x500000000000, // XorMask
421  0, // ShadowBase
422  0x100000000000, // OriginBase
423 };
424 
425 static const PlatformMemoryMapParams Linux_X86_MemoryMapParams = {
428 };
429 
430 static const PlatformMemoryMapParams Linux_MIPS_MemoryMapParams = {
431  nullptr,
433 };
434 
435 static const PlatformMemoryMapParams Linux_PowerPC_MemoryMapParams = {
436  nullptr,
438 };
439 
440 static const PlatformMemoryMapParams Linux_ARM_MemoryMapParams = {
441  nullptr,
443 };
444 
445 static const PlatformMemoryMapParams FreeBSD_X86_MemoryMapParams = {
448 };
449 
450 static const PlatformMemoryMapParams NetBSD_X86_MemoryMapParams = {
451  nullptr,
453 };
454 
455 namespace {
456 
457 /// Instrument functions of a module to detect uninitialized reads.
458 ///
459 /// Instantiating MemorySanitizer inserts the msan runtime library API function
460 /// declarations into the module if they don't exist already. Instantiating
461 /// ensures the __msan_init function is in the list of global constructors for
462 /// the module.
463 class MemorySanitizer {
464 public:
465  MemorySanitizer(Module &M, MemorySanitizerOptions Options)
466  : CompileKernel(Options.Kernel), TrackOrigins(Options.TrackOrigins),
467  Recover(Options.Recover) {
468  initializeModule(M);
469  }
470 
471  // MSan cannot be moved or copied because of MapParams.
472  MemorySanitizer(MemorySanitizer &&) = delete;
473  MemorySanitizer &operator=(MemorySanitizer &&) = delete;
474  MemorySanitizer(const MemorySanitizer &) = delete;
475  MemorySanitizer &operator=(const MemorySanitizer &) = delete;
476 
477  bool sanitizeFunction(Function &F, TargetLibraryInfo &TLI);
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 initializeModule(Module &M);
487  void initializeCallbacks(Module &M);
488  void createKernelApi(Module &M);
489  void createUserspaceApi(Module &M);
490 
491  /// True if we're compiling the Linux kernel.
492  bool CompileKernel;
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  FunctionCallee WarningFn;
540 
541  // These arrays are indexed by log2(AccessSize).
542  FunctionCallee MaybeWarningFn[kNumberOfAccessSizes];
543  FunctionCallee MaybeStoreOriginFn[kNumberOfAccessSizes];
544 
545  /// Run-time helper that generates a new origin value for a stack
546  /// allocation.
547  FunctionCallee MsanSetAllocaOrigin4Fn;
548 
549  /// Run-time helper that poisons stack on function entry.
550  FunctionCallee 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  FunctionCallee MsanChainOriginFn;
555 
556  /// MSan runtime replacements for memmove, memcpy and memset.
557  FunctionCallee MemmoveFn, MemcpyFn, MemsetFn;
558 
559  /// KMSAN callback for task-local function argument shadow.
560  StructType *MsanContextStateTy;
561  FunctionCallee MsanGetContextStateFn;
562 
563  /// Functions for poisoning/unpoisoning local variables
564  FunctionCallee MsanPoisonAllocaFn, MsanUnpoisonAllocaFn;
565 
566  /// Each of the MsanMetadataPtrXxx functions returns a pair of shadow/origin
567  /// pointers.
568  FunctionCallee MsanMetadataPtrForLoadN, MsanMetadataPtrForStoreN;
569  FunctionCallee MsanMetadataPtrForLoad_1_8[4];
570  FunctionCallee MsanMetadataPtrForStore_1_8[4];
571  FunctionCallee MsanInstrumentAsmStoreFn;
572 
573  /// Helper to choose between different MsanMetadataPtrXxx().
574  FunctionCallee getKmsanShadowOriginAccessFn(bool isStore, int size);
575 
576  /// Memory map parameters used in application-to-shadow calculation.
577  const MemoryMapParams *MapParams;
578 
579  /// Custom memory map parameters used when -msan-shadow-base or
580  // -msan-origin-base is provided.
581  MemoryMapParams CustomMapParams;
582 
583  MDNode *ColdCallWeights;
584 
585  /// Branch weights for origin store.
586  MDNode *OriginStoreWeights;
587 
588  /// An empty volatile inline asm that prevents callback merge.
589  InlineAsm *EmptyAsm;
590 };
591 
592 void insertModuleCtor(Module &M) {
594  M, kMsanModuleCtorName, kMsanInitName,
595  /*InitArgTypes=*/{},
596  /*InitArgs=*/{},
597  // This callback is invoked when the functions are created the first
598  // time. Hook them into the global ctors list in that case:
599  [&](Function *Ctor, FunctionCallee) {
600  if (!ClWithComdat) {
601  appendToGlobalCtors(M, Ctor, 0);
602  return;
603  }
604  Comdat *MsanCtorComdat = M.getOrInsertComdat(kMsanModuleCtorName);
605  Ctor->setComdat(MsanCtorComdat);
606  appendToGlobalCtors(M, Ctor, 0, Ctor);
607  });
608 }
609 
610 /// A legacy function pass for msan instrumentation.
611 ///
612 /// Instruments functions to detect unitialized reads.
613 struct MemorySanitizerLegacyPass : public FunctionPass {
614  // Pass identification, replacement for typeid.
615  static char ID;
616 
617  MemorySanitizerLegacyPass(MemorySanitizerOptions Options = {})
618  : FunctionPass(ID), Options(Options) {}
619  StringRef getPassName() const override { return "MemorySanitizerLegacyPass"; }
620 
621  void getAnalysisUsage(AnalysisUsage &AU) const override {
623  }
624 
625  bool runOnFunction(Function &F) override {
626  return MSan->sanitizeFunction(
627  F, getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F));
628  }
629  bool doInitialization(Module &M) override;
630 
632  MemorySanitizerOptions Options;
633 };
634 
635 template <class T> T getOptOrDefault(const cl::opt<T> &Opt, T Default) {
636  return (Opt.getNumOccurrences() > 0) ? Opt : Default;
637 }
638 
639 } // end anonymous namespace
640 
642  : Kernel(getOptOrDefault(ClEnableKmsan, K)),
643  TrackOrigins(getOptOrDefault(ClTrackOrigins, Kernel ? 2 : TO)),
644  Recover(getOptOrDefault(ClKeepGoing, Kernel || R)) {}
645 
648  MemorySanitizer Msan(*F.getParent(), Options);
649  if (Msan.sanitizeFunction(F, FAM.getResult<TargetLibraryAnalysis>(F)))
650  return PreservedAnalyses::none();
651  return PreservedAnalyses::all();
652 }
653 
655  ModuleAnalysisManager &AM) {
656  if (Options.Kernel)
657  return PreservedAnalyses::all();
658  insertModuleCtor(M);
659  return PreservedAnalyses::none();
660 }
661 
663 
664 INITIALIZE_PASS_BEGIN(MemorySanitizerLegacyPass, "msan",
665  "MemorySanitizer: detects uninitialized reads.", false,
666  false)
668 INITIALIZE_PASS_END(MemorySanitizerLegacyPass, "msan",
669  "MemorySanitizer: detects uninitialized reads.", false,
670  false)
671 
672 FunctionPass *
674  return new MemorySanitizerLegacyPass(Options);
675 }
676 
677 /// Create a non-const global initialized with the given string.
678 ///
679 /// Creates a writable global for Str so that we can pass it to the
680 /// run-time lib. Runtime uses first 4 bytes of the string to store the
681 /// frame ID, so the string needs to be mutable.
683  StringRef Str) {
684  Constant *StrConst = ConstantDataArray::getString(M.getContext(), Str);
685  return new GlobalVariable(M, StrConst->getType(), /*isConstant=*/false,
686  GlobalValue::PrivateLinkage, StrConst, "");
687 }
688 
689 /// Create KMSAN API callbacks.
690 void MemorySanitizer::createKernelApi(Module &M) {
691  IRBuilder<> IRB(*C);
692 
693  // These will be initialized in insertKmsanPrologue().
694  RetvalTLS = nullptr;
695  RetvalOriginTLS = nullptr;
696  ParamTLS = nullptr;
697  ParamOriginTLS = nullptr;
698  VAArgTLS = nullptr;
699  VAArgOriginTLS = nullptr;
700  VAArgOverflowSizeTLS = nullptr;
701  // OriginTLS is unused in the kernel.
702  OriginTLS = nullptr;
703 
704  // __msan_warning() in the kernel takes an origin.
705  WarningFn = M.getOrInsertFunction("__msan_warning", IRB.getVoidTy(),
706  IRB.getInt32Ty());
707  // Requests the per-task context state (kmsan_context_state*) from the
708  // runtime library.
709  MsanContextStateTy = StructType::get(
710  ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8),
711  ArrayType::get(IRB.getInt64Ty(), kRetvalTLSSize / 8),
712  ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8),
713  ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8), /* va_arg_origin */
714  IRB.getInt64Ty(), ArrayType::get(OriginTy, kParamTLSSize / 4), OriginTy,
715  OriginTy);
716  MsanGetContextStateFn = M.getOrInsertFunction(
717  "__msan_get_context_state", PointerType::get(MsanContextStateTy, 0));
718 
719  Type *RetTy = StructType::get(PointerType::get(IRB.getInt8Ty(), 0),
720  PointerType::get(IRB.getInt32Ty(), 0));
721 
722  for (int ind = 0, size = 1; ind < 4; ind++, size <<= 1) {
723  std::string name_load =
724  "__msan_metadata_ptr_for_load_" + std::to_string(size);
725  std::string name_store =
726  "__msan_metadata_ptr_for_store_" + std::to_string(size);
727  MsanMetadataPtrForLoad_1_8[ind] = M.getOrInsertFunction(
728  name_load, RetTy, PointerType::get(IRB.getInt8Ty(), 0));
729  MsanMetadataPtrForStore_1_8[ind] = M.getOrInsertFunction(
730  name_store, RetTy, PointerType::get(IRB.getInt8Ty(), 0));
731  }
732 
733  MsanMetadataPtrForLoadN = M.getOrInsertFunction(
734  "__msan_metadata_ptr_for_load_n", RetTy,
735  PointerType::get(IRB.getInt8Ty(), 0), IRB.getInt64Ty());
736  MsanMetadataPtrForStoreN = M.getOrInsertFunction(
737  "__msan_metadata_ptr_for_store_n", RetTy,
738  PointerType::get(IRB.getInt8Ty(), 0), IRB.getInt64Ty());
739 
740  // Functions for poisoning and unpoisoning memory.
741  MsanPoisonAllocaFn =
742  M.getOrInsertFunction("__msan_poison_alloca", IRB.getVoidTy(),
743  IRB.getInt8PtrTy(), IntptrTy, IRB.getInt8PtrTy());
744  MsanUnpoisonAllocaFn = M.getOrInsertFunction(
745  "__msan_unpoison_alloca", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy);
746 }
747 
749  return M.getOrInsertGlobal(Name, Ty, [&] {
750  return new GlobalVariable(M, Ty, false, GlobalVariable::ExternalLinkage,
751  nullptr, Name, nullptr,
753  });
754 }
755 
756 /// Insert declarations for userspace-specific functions and globals.
757 void MemorySanitizer::createUserspaceApi(Module &M) {
758  IRBuilder<> IRB(*C);
759  // Create the callback.
760  // FIXME: this function should have "Cold" calling conv,
761  // which is not yet implemented.
762  StringRef WarningFnName = Recover ? "__msan_warning"
763  : "__msan_warning_noreturn";
764  WarningFn = M.getOrInsertFunction(WarningFnName, IRB.getVoidTy());
765 
766  // Create the global TLS variables.
767  RetvalTLS =
768  getOrInsertGlobal(M, "__msan_retval_tls",
769  ArrayType::get(IRB.getInt64Ty(), kRetvalTLSSize / 8));
770 
771  RetvalOriginTLS = getOrInsertGlobal(M, "__msan_retval_origin_tls", OriginTy);
772 
773  ParamTLS =
774  getOrInsertGlobal(M, "__msan_param_tls",
775  ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8));
776 
777  ParamOriginTLS =
778  getOrInsertGlobal(M, "__msan_param_origin_tls",
779  ArrayType::get(OriginTy, kParamTLSSize / 4));
780 
781  VAArgTLS =
782  getOrInsertGlobal(M, "__msan_va_arg_tls",
783  ArrayType::get(IRB.getInt64Ty(), kParamTLSSize / 8));
784 
785  VAArgOriginTLS =
786  getOrInsertGlobal(M, "__msan_va_arg_origin_tls",
787  ArrayType::get(OriginTy, kParamTLSSize / 4));
788 
789  VAArgOverflowSizeTLS =
790  getOrInsertGlobal(M, "__msan_va_arg_overflow_size_tls", IRB.getInt64Ty());
791  OriginTLS = getOrInsertGlobal(M, "__msan_origin_tls", IRB.getInt32Ty());
792 
793  for (size_t AccessSizeIndex = 0; AccessSizeIndex < kNumberOfAccessSizes;
794  AccessSizeIndex++) {
795  unsigned AccessSize = 1 << AccessSizeIndex;
796  std::string FunctionName = "__msan_maybe_warning_" + itostr(AccessSize);
797  MaybeWarningFn[AccessSizeIndex] = M.getOrInsertFunction(
798  FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
799  IRB.getInt32Ty());
800 
801  FunctionName = "__msan_maybe_store_origin_" + itostr(AccessSize);
802  MaybeStoreOriginFn[AccessSizeIndex] = M.getOrInsertFunction(
803  FunctionName, IRB.getVoidTy(), IRB.getIntNTy(AccessSize * 8),
804  IRB.getInt8PtrTy(), IRB.getInt32Ty());
805  }
806 
807  MsanSetAllocaOrigin4Fn = M.getOrInsertFunction(
808  "__msan_set_alloca_origin4", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy,
809  IRB.getInt8PtrTy(), IntptrTy);
810  MsanPoisonStackFn =
811  M.getOrInsertFunction("__msan_poison_stack", IRB.getVoidTy(),
812  IRB.getInt8PtrTy(), IntptrTy);
813 }
814 
815 /// Insert extern declaration of runtime-provided functions and globals.
816 void MemorySanitizer::initializeCallbacks(Module &M) {
817  // Only do this once.
818  if (CallbacksInitialized)
819  return;
820 
821  IRBuilder<> IRB(*C);
822  // Initialize callbacks that are common for kernel and userspace
823  // instrumentation.
824  MsanChainOriginFn = M.getOrInsertFunction(
825  "__msan_chain_origin", IRB.getInt32Ty(), IRB.getInt32Ty());
826  MemmoveFn = M.getOrInsertFunction(
827  "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
828  IRB.getInt8PtrTy(), IntptrTy);
829  MemcpyFn = M.getOrInsertFunction(
830  "__msan_memcpy", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
831  IntptrTy);
832  MemsetFn = M.getOrInsertFunction(
833  "__msan_memset", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt32Ty(),
834  IntptrTy);
835  // We insert an empty inline asm after __msan_report* to avoid callback merge.
836  EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
837  StringRef(""), StringRef(""),
838  /*hasSideEffects=*/true);
839 
840  MsanInstrumentAsmStoreFn =
841  M.getOrInsertFunction("__msan_instrument_asm_store", IRB.getVoidTy(),
842  PointerType::get(IRB.getInt8Ty(), 0), IntptrTy);
843 
844  if (CompileKernel) {
845  createKernelApi(M);
846  } else {
847  createUserspaceApi(M);
848  }
849  CallbacksInitialized = true;
850 }
851 
852 FunctionCallee MemorySanitizer::getKmsanShadowOriginAccessFn(bool isStore,
853  int size) {
854  FunctionCallee *Fns =
855  isStore ? MsanMetadataPtrForStore_1_8 : MsanMetadataPtrForLoad_1_8;
856  switch (size) {
857  case 1:
858  return Fns[0];
859  case 2:
860  return Fns[1];
861  case 4:
862  return Fns[2];
863  case 8:
864  return Fns[3];
865  default:
866  return nullptr;
867  }
868 }
869 
870 /// Module-level initialization.
871 ///
872 /// inserts a call to __msan_init to the module's constructor list.
873 void MemorySanitizer::initializeModule(Module &M) {
874  auto &DL = M.getDataLayout();
875 
876  bool ShadowPassed = ClShadowBase.getNumOccurrences() > 0;
877  bool OriginPassed = ClOriginBase.getNumOccurrences() > 0;
878  // Check the overrides first
879  if (ShadowPassed || OriginPassed) {
880  CustomMapParams.AndMask = ClAndMask;
881  CustomMapParams.XorMask = ClXorMask;
882  CustomMapParams.ShadowBase = ClShadowBase;
883  CustomMapParams.OriginBase = ClOriginBase;
884  MapParams = &CustomMapParams;
885  } else {
886  Triple TargetTriple(M.getTargetTriple());
887  switch (TargetTriple.getOS()) {
888  case Triple::FreeBSD:
889  switch (TargetTriple.getArch()) {
890  case Triple::x86_64:
891  MapParams = FreeBSD_X86_MemoryMapParams.bits64;
892  break;
893  case Triple::x86:
894  MapParams = FreeBSD_X86_MemoryMapParams.bits32;
895  break;
896  default:
897  report_fatal_error("unsupported architecture");
898  }
899  break;
900  case Triple::NetBSD:
901  switch (TargetTriple.getArch()) {
902  case Triple::x86_64:
903  MapParams = NetBSD_X86_MemoryMapParams.bits64;
904  break;
905  default:
906  report_fatal_error("unsupported architecture");
907  }
908  break;
909  case Triple::Linux:
910  switch (TargetTriple.getArch()) {
911  case Triple::x86_64:
912  MapParams = Linux_X86_MemoryMapParams.bits64;
913  break;
914  case Triple::x86:
915  MapParams = Linux_X86_MemoryMapParams.bits32;
916  break;
917  case Triple::mips64:
918  case Triple::mips64el:
919  MapParams = Linux_MIPS_MemoryMapParams.bits64;
920  break;
921  case Triple::ppc64:
922  case Triple::ppc64le:
923  MapParams = Linux_PowerPC_MemoryMapParams.bits64;
924  break;
925  case Triple::aarch64:
926  case Triple::aarch64_be:
927  MapParams = Linux_ARM_MemoryMapParams.bits64;
928  break;
929  default:
930  report_fatal_error("unsupported architecture");
931  }
932  break;
933  default:
934  report_fatal_error("unsupported operating system");
935  }
936  }
937 
938  C = &(M.getContext());
939  IRBuilder<> IRB(*C);
940  IntptrTy = IRB.getIntPtrTy(DL);
941  OriginTy = IRB.getInt32Ty();
942 
943  ColdCallWeights = MDBuilder(*C).createBranchWeights(1, 1000);
944  OriginStoreWeights = MDBuilder(*C).createBranchWeights(1, 1000);
945 
946  if (!CompileKernel) {
947  if (TrackOrigins)
948  M.getOrInsertGlobal("__msan_track_origins", IRB.getInt32Ty(), [&] {
949  return new GlobalVariable(
950  M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
951  IRB.getInt32(TrackOrigins), "__msan_track_origins");
952  });
953 
954  if (Recover)
955  M.getOrInsertGlobal("__msan_keep_going", IRB.getInt32Ty(), [&] {
956  return new GlobalVariable(M, IRB.getInt32Ty(), true,
958  IRB.getInt32(Recover), "__msan_keep_going");
959  });
960 }
961 }
962 
963 bool MemorySanitizerLegacyPass::doInitialization(Module &M) {
964  if (!Options.Kernel)
965  insertModuleCtor(M);
966  MSan.emplace(M, Options);
967  return true;
968 }
969 
970 namespace {
971 
972 /// A helper class that handles instrumentation of VarArg
973 /// functions on a particular platform.
974 ///
975 /// Implementations are expected to insert the instrumentation
976 /// necessary to propagate argument shadow through VarArg function
977 /// calls. Visit* methods are called during an InstVisitor pass over
978 /// the function, and should avoid creating new basic blocks. A new
979 /// instance of this class is created for each instrumented function.
980 struct VarArgHelper {
981  virtual ~VarArgHelper() = default;
982 
983  /// Visit a CallSite.
984  virtual void visitCallSite(CallSite &CS, IRBuilder<> &IRB) = 0;
985 
986  /// Visit a va_start call.
987  virtual void visitVAStartInst(VAStartInst &I) = 0;
988 
989  /// Visit a va_copy call.
990  virtual void visitVACopyInst(VACopyInst &I) = 0;
991 
992  /// Finalize function instrumentation.
993  ///
994  /// This method is called after visiting all interesting (see above)
995  /// instructions in a function.
996  virtual void finalizeInstrumentation() = 0;
997 };
998 
999 struct MemorySanitizerVisitor;
1000 
1001 } // end anonymous namespace
1002 
1003 static VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
1004  MemorySanitizerVisitor &Visitor);
1005 
1006 static unsigned TypeSizeToSizeIndex(unsigned TypeSize) {
1007  if (TypeSize <= 8) return 0;
1008  return Log2_32_Ceil((TypeSize + 7) / 8);
1009 }
1010 
1011 namespace {
1012 
1013 /// This class does all the work for a given function. Store and Load
1014 /// instructions store and load corresponding shadow and origin
1015 /// values. Most instructions propagate shadow from arguments to their
1016 /// return values. Certain instructions (most importantly, BranchInst)
1017 /// test their argument shadow and print reports (with a runtime call) if it's
1018 /// non-zero.
1019 struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
1020  Function &F;
1021  MemorySanitizer &MS;
1022  SmallVector<PHINode *, 16> ShadowPHINodes, OriginPHINodes;
1023  ValueMap<Value*, Value*> ShadowMap, OriginMap;
1024  std::unique_ptr<VarArgHelper> VAHelper;
1025  const TargetLibraryInfo *TLI;
1026  BasicBlock *ActualFnStart;
1027 
1028  // The following flags disable parts of MSan instrumentation based on
1029  // blacklist contents and command-line options.
1030  bool InsertChecks;
1031  bool PropagateShadow;
1032  bool PoisonStack;
1033  bool PoisonUndef;
1034  bool CheckReturnValue;
1035 
1036  struct ShadowOriginAndInsertPoint {
1037  Value *Shadow;
1038  Value *Origin;
1039  Instruction *OrigIns;
1040 
1041  ShadowOriginAndInsertPoint(Value *S, Value *O, Instruction *I)
1042  : Shadow(S), Origin(O), OrigIns(I) {}
1043  };
1044  SmallVector<ShadowOriginAndInsertPoint, 16> InstrumentationList;
1045  bool InstrumentLifetimeStart = ClHandleLifetimeIntrinsics;
1046  SmallSet<AllocaInst *, 16> AllocaSet;
1048  SmallVector<StoreInst *, 16> StoreList;
1049 
1050  MemorySanitizerVisitor(Function &F, MemorySanitizer &MS,
1051  const TargetLibraryInfo &TLI)
1052  : F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)), TLI(&TLI) {
1053  bool SanitizeFunction = F.hasFnAttribute(Attribute::SanitizeMemory);
1054  InsertChecks = SanitizeFunction;
1055  PropagateShadow = SanitizeFunction;
1056  PoisonStack = SanitizeFunction && ClPoisonStack;
1057  PoisonUndef = SanitizeFunction && ClPoisonUndef;
1058  // FIXME: Consider using SpecialCaseList to specify a list of functions that
1059  // must always return fully initialized values. For now, we hardcode "main".
1060  CheckReturnValue = SanitizeFunction && (F.getName() == "main");
1061 
1062  MS.initializeCallbacks(*F.getParent());
1063  if (MS.CompileKernel)
1064  ActualFnStart = insertKmsanPrologue(F);
1065  else
1066  ActualFnStart = &F.getEntryBlock();
1067 
1068  LLVM_DEBUG(if (!InsertChecks) dbgs()
1069  << "MemorySanitizer is not inserting checks into '"
1070  << F.getName() << "'\n");
1071  }
1072 
1073  Value *updateOrigin(Value *V, IRBuilder<> &IRB) {
1074  if (MS.TrackOrigins <= 1) return V;
1075  return IRB.CreateCall(MS.MsanChainOriginFn, V);
1076  }
1077 
1078  Value *originToIntptr(IRBuilder<> &IRB, Value *Origin) {
1079  const DataLayout &DL = F.getParent()->getDataLayout();
1080  unsigned IntptrSize = DL.getTypeStoreSize(MS.IntptrTy);
1081  if (IntptrSize == kOriginSize) return Origin;
1082  assert(IntptrSize == kOriginSize * 2);
1083  Origin = IRB.CreateIntCast(Origin, MS.IntptrTy, /* isSigned */ false);
1084  return IRB.CreateOr(Origin, IRB.CreateShl(Origin, kOriginSize * 8));
1085  }
1086 
1087  /// Fill memory range with the given origin value.
1088  void paintOrigin(IRBuilder<> &IRB, Value *Origin, Value *OriginPtr,
1089  unsigned Size, unsigned Alignment) {
1090  const DataLayout &DL = F.getParent()->getDataLayout();
1091  unsigned IntptrAlignment = DL.getABITypeAlignment(MS.IntptrTy);
1092  unsigned IntptrSize = DL.getTypeStoreSize(MS.IntptrTy);
1093  assert(IntptrAlignment >= kMinOriginAlignment);
1094  assert(IntptrSize >= kOriginSize);
1095 
1096  unsigned Ofs = 0;
1097  unsigned CurrentAlignment = Alignment;
1098  if (Alignment >= IntptrAlignment && IntptrSize > kOriginSize) {
1099  Value *IntptrOrigin = originToIntptr(IRB, Origin);
1100  Value *IntptrOriginPtr =
1101  IRB.CreatePointerCast(OriginPtr, PointerType::get(MS.IntptrTy, 0));
1102  for (unsigned i = 0; i < Size / IntptrSize; ++i) {
1103  Value *Ptr = i ? IRB.CreateConstGEP1_32(MS.IntptrTy, IntptrOriginPtr, i)
1104  : IntptrOriginPtr;
1105  IRB.CreateAlignedStore(IntptrOrigin, Ptr, CurrentAlignment);
1106  Ofs += IntptrSize / kOriginSize;
1107  CurrentAlignment = IntptrAlignment;
1108  }
1109  }
1110 
1111  for (unsigned i = Ofs; i < (Size + kOriginSize - 1) / kOriginSize; ++i) {
1112  Value *GEP =
1113  i ? IRB.CreateConstGEP1_32(MS.OriginTy, OriginPtr, i) : OriginPtr;
1114  IRB.CreateAlignedStore(Origin, GEP, CurrentAlignment);
1115  CurrentAlignment = kMinOriginAlignment;
1116  }
1117  }
1118 
1119  void storeOrigin(IRBuilder<> &IRB, Value *Addr, Value *Shadow, Value *Origin,
1120  Value *OriginPtr, unsigned Alignment, bool AsCall) {
1121  const DataLayout &DL = F.getParent()->getDataLayout();
1122  unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
1123  unsigned StoreSize = DL.getTypeStoreSize(Shadow->getType());
1124  if (Shadow->getType()->isAggregateType()) {
1125  paintOrigin(IRB, updateOrigin(Origin, IRB), OriginPtr, StoreSize,
1126  OriginAlignment);
1127  } else {
1128  Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
1129  Constant *ConstantShadow = dyn_cast_or_null<Constant>(ConvertedShadow);
1130  if (ConstantShadow) {
1131  if (ClCheckConstantShadow && !ConstantShadow->isZeroValue())
1132  paintOrigin(IRB, updateOrigin(Origin, IRB), OriginPtr, StoreSize,
1133  OriginAlignment);
1134  return;
1135  }
1136 
1137  unsigned TypeSizeInBits =
1138  DL.getTypeSizeInBits(ConvertedShadow->getType());
1139  unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
1140  if (AsCall && SizeIndex < kNumberOfAccessSizes && !MS.CompileKernel) {
1141  FunctionCallee Fn = MS.MaybeStoreOriginFn[SizeIndex];
1142  Value *ConvertedShadow2 = IRB.CreateZExt(
1143  ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
1144  IRB.CreateCall(Fn, {ConvertedShadow2,
1145  IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()),
1146  Origin});
1147  } else {
1148  Value *Cmp = IRB.CreateICmpNE(
1149  ConvertedShadow, getCleanShadow(ConvertedShadow), "_mscmp");
1151  Cmp, &*IRB.GetInsertPoint(), false, MS.OriginStoreWeights);
1152  IRBuilder<> IRBNew(CheckTerm);
1153  paintOrigin(IRBNew, updateOrigin(Origin, IRBNew), OriginPtr, StoreSize,
1154  OriginAlignment);
1155  }
1156  }
1157  }
1158 
1159  void materializeStores(bool InstrumentWithCalls) {
1160  for (StoreInst *SI : StoreList) {
1161  IRBuilder<> IRB(SI);
1162  Value *Val = SI->getValueOperand();
1163  Value *Addr = SI->getPointerOperand();
1164  Value *Shadow = SI->isAtomic() ? getCleanShadow(Val) : getShadow(Val);
1165  Value *ShadowPtr, *OriginPtr;
1166  Type *ShadowTy = Shadow->getType();
1167  unsigned Alignment = SI->getAlignment();
1168  unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
1169  std::tie(ShadowPtr, OriginPtr) =
1170  getShadowOriginPtr(Addr, IRB, ShadowTy, Alignment, /*isStore*/ true);
1171 
1172  StoreInst *NewSI = IRB.CreateAlignedStore(Shadow, ShadowPtr, Alignment);
1173  LLVM_DEBUG(dbgs() << " STORE: " << *NewSI << "\n");
1174  (void)NewSI;
1175 
1176  if (SI->isAtomic())
1177  SI->setOrdering(addReleaseOrdering(SI->getOrdering()));
1178 
1179  if (MS.TrackOrigins && !SI->isAtomic())
1180  storeOrigin(IRB, Addr, Shadow, getOrigin(Val), OriginPtr,
1181  OriginAlignment, InstrumentWithCalls);
1182  }
1183  }
1184 
1185  /// Helper function to insert a warning at IRB's current insert point.
1186  void insertWarningFn(IRBuilder<> &IRB, Value *Origin) {
1187  if (!Origin)
1188  Origin = (Value *)IRB.getInt32(0);
1189  if (MS.CompileKernel) {
1190  IRB.CreateCall(MS.WarningFn, Origin);
1191  } else {
1192  if (MS.TrackOrigins) {
1193  IRB.CreateStore(Origin, MS.OriginTLS);
1194  }
1195  IRB.CreateCall(MS.WarningFn, {});
1196  }
1197  IRB.CreateCall(MS.EmptyAsm, {});
1198  // FIXME: Insert UnreachableInst if !MS.Recover?
1199  // This may invalidate some of the following checks and needs to be done
1200  // at the very end.
1201  }
1202 
1203  void materializeOneCheck(Instruction *OrigIns, Value *Shadow, Value *Origin,
1204  bool AsCall) {
1205  IRBuilder<> IRB(OrigIns);
1206  LLVM_DEBUG(dbgs() << " SHAD0 : " << *Shadow << "\n");
1207  Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
1208  LLVM_DEBUG(dbgs() << " SHAD1 : " << *ConvertedShadow << "\n");
1209 
1210  Constant *ConstantShadow = dyn_cast_or_null<Constant>(ConvertedShadow);
1211  if (ConstantShadow) {
1212  if (ClCheckConstantShadow && !ConstantShadow->isZeroValue()) {
1213  insertWarningFn(IRB, Origin);
1214  }
1215  return;
1216  }
1217 
1218  const DataLayout &DL = OrigIns->getModule()->getDataLayout();
1219 
1220  unsigned TypeSizeInBits = DL.getTypeSizeInBits(ConvertedShadow->getType());
1221  unsigned SizeIndex = TypeSizeToSizeIndex(TypeSizeInBits);
1222  if (AsCall && SizeIndex < kNumberOfAccessSizes && !MS.CompileKernel) {
1223  FunctionCallee Fn = MS.MaybeWarningFn[SizeIndex];
1224  Value *ConvertedShadow2 =
1225  IRB.CreateZExt(ConvertedShadow, IRB.getIntNTy(8 * (1 << SizeIndex)));
1226  IRB.CreateCall(Fn, {ConvertedShadow2, MS.TrackOrigins && Origin
1227  ? Origin
1228  : (Value *)IRB.getInt32(0)});
1229  } else {
1230  Value *Cmp = IRB.CreateICmpNE(ConvertedShadow,
1231  getCleanShadow(ConvertedShadow), "_mscmp");
1233  Cmp, OrigIns,
1234  /* Unreachable */ !MS.Recover, MS.ColdCallWeights);
1235 
1236  IRB.SetInsertPoint(CheckTerm);
1237  insertWarningFn(IRB, Origin);
1238  LLVM_DEBUG(dbgs() << " CHECK: " << *Cmp << "\n");
1239  }
1240  }
1241 
1242  void materializeChecks(bool InstrumentWithCalls) {
1243  for (const auto &ShadowData : InstrumentationList) {
1244  Instruction *OrigIns = ShadowData.OrigIns;
1245  Value *Shadow = ShadowData.Shadow;
1246  Value *Origin = ShadowData.Origin;
1247  materializeOneCheck(OrigIns, Shadow, Origin, InstrumentWithCalls);
1248  }
1249  LLVM_DEBUG(dbgs() << "DONE:\n" << F);
1250  }
1251 
1252  BasicBlock *insertKmsanPrologue(Function &F) {
1253  BasicBlock *ret =
1256  Value *ContextState = IRB.CreateCall(MS.MsanGetContextStateFn, {});
1257  Constant *Zero = IRB.getInt32(0);
1258  MS.ParamTLS = IRB.CreateGEP(MS.MsanContextStateTy, ContextState,
1259  {Zero, IRB.getInt32(0)}, "param_shadow");
1260  MS.RetvalTLS = IRB.CreateGEP(MS.MsanContextStateTy, ContextState,
1261  {Zero, IRB.getInt32(1)}, "retval_shadow");
1262  MS.VAArgTLS = IRB.CreateGEP(MS.MsanContextStateTy, ContextState,
1263  {Zero, IRB.getInt32(2)}, "va_arg_shadow");
1264  MS.VAArgOriginTLS = IRB.CreateGEP(MS.MsanContextStateTy, ContextState,
1265  {Zero, IRB.getInt32(3)}, "va_arg_origin");
1266  MS.VAArgOverflowSizeTLS =
1267  IRB.CreateGEP(MS.MsanContextStateTy, ContextState,
1268  {Zero, IRB.getInt32(4)}, "va_arg_overflow_size");
1269  MS.ParamOriginTLS = IRB.CreateGEP(MS.MsanContextStateTy, ContextState,
1270  {Zero, IRB.getInt32(5)}, "param_origin");
1271  MS.RetvalOriginTLS =
1272  IRB.CreateGEP(MS.MsanContextStateTy, ContextState,
1273  {Zero, IRB.getInt32(6)}, "retval_origin");
1274  return ret;
1275  }
1276 
1277  /// Add MemorySanitizer instrumentation to a function.
1278  bool runOnFunction() {
1279  // In the presence of unreachable blocks, we may see Phi nodes with
1280  // incoming nodes from such blocks. Since InstVisitor skips unreachable
1281  // blocks, such nodes will not have any shadow value associated with them.
1282  // It's easier to remove unreachable blocks than deal with missing shadow.
1284 
1285  // Iterate all BBs in depth-first order and create shadow instructions
1286  // for all instructions (where applicable).
1287  // For PHI nodes we create dummy shadow PHIs which will be finalized later.
1288  for (BasicBlock *BB : depth_first(ActualFnStart))
1289  visit(*BB);
1290 
1291  // Finalize PHI nodes.
1292  for (PHINode *PN : ShadowPHINodes) {
1293  PHINode *PNS = cast<PHINode>(getShadow(PN));
1294  PHINode *PNO = MS.TrackOrigins ? cast<PHINode>(getOrigin(PN)) : nullptr;
1295  size_t NumValues = PN->getNumIncomingValues();
1296  for (size_t v = 0; v < NumValues; v++) {
1297  PNS->addIncoming(getShadow(PN, v), PN->getIncomingBlock(v));
1298  if (PNO) PNO->addIncoming(getOrigin(PN, v), PN->getIncomingBlock(v));
1299  }
1300  }
1301 
1302  VAHelper->finalizeInstrumentation();
1303 
1304  // Poison llvm.lifetime.start intrinsics, if we haven't fallen back to
1305  // instrumenting only allocas.
1306  if (InstrumentLifetimeStart) {
1307  for (auto Item : LifetimeStartList) {
1308  instrumentAlloca(*Item.second, Item.first);
1309  AllocaSet.erase(Item.second);
1310  }
1311  }
1312  // Poison the allocas for which we didn't instrument the corresponding
1313  // lifetime intrinsics.
1314  for (AllocaInst *AI : AllocaSet)
1315  instrumentAlloca(*AI);
1316 
1317  bool InstrumentWithCalls = ClInstrumentationWithCallThreshold >= 0 &&
1318  InstrumentationList.size() + StoreList.size() >
1320 
1321  // Insert shadow value checks.
1322  materializeChecks(InstrumentWithCalls);
1323 
1324  // Delayed instrumentation of StoreInst.
1325  // This may not add new address checks.
1326  materializeStores(InstrumentWithCalls);
1327 
1328  return true;
1329  }
1330 
1331  /// Compute the shadow type that corresponds to a given Value.
1332  Type *getShadowTy(Value *V) {
1333  return getShadowTy(V->getType());
1334  }
1335 
1336  /// Compute the shadow type that corresponds to a given Type.
1337  Type *getShadowTy(Type *OrigTy) {
1338  if (!OrigTy->isSized()) {
1339  return nullptr;
1340  }
1341  // For integer type, shadow is the same as the original type.
1342  // This may return weird-sized types like i1.
1343  if (IntegerType *IT = dyn_cast<IntegerType>(OrigTy))
1344  return IT;
1345  const DataLayout &DL = F.getParent()->getDataLayout();
1346  if (VectorType *VT = dyn_cast<VectorType>(OrigTy)) {
1347  uint32_t EltSize = DL.getTypeSizeInBits(VT->getElementType());
1348  return VectorType::get(IntegerType::get(*MS.C, EltSize),
1349  VT->getNumElements());
1350  }
1351  if (ArrayType *AT = dyn_cast<ArrayType>(OrigTy)) {
1352  return ArrayType::get(getShadowTy(AT->getElementType()),
1353  AT->getNumElements());
1354  }
1355  if (StructType *ST = dyn_cast<StructType>(OrigTy)) {
1356  SmallVector<Type*, 4> Elements;
1357  for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
1358  Elements.push_back(getShadowTy(ST->getElementType(i)));
1359  StructType *Res = StructType::get(*MS.C, Elements, ST->isPacked());
1360  LLVM_DEBUG(dbgs() << "getShadowTy: " << *ST << " ===> " << *Res << "\n");
1361  return Res;
1362  }
1363  uint32_t TypeSize = DL.getTypeSizeInBits(OrigTy);
1364  return IntegerType::get(*MS.C, TypeSize);
1365  }
1366 
1367  /// Flatten a vector type.
1368  Type *getShadowTyNoVec(Type *ty) {
1369  if (VectorType *vt = dyn_cast<VectorType>(ty))
1370  return IntegerType::get(*MS.C, vt->getBitWidth());
1371  return ty;
1372  }
1373 
1374  /// Convert a shadow value to it's flattened variant.
1375  Value *convertToShadowTyNoVec(Value *V, IRBuilder<> &IRB) {
1376  Type *Ty = V->getType();
1377  Type *NoVecTy = getShadowTyNoVec(Ty);
1378  if (Ty == NoVecTy) return V;
1379  return IRB.CreateBitCast(V, NoVecTy);
1380  }
1381 
1382  /// Compute the integer shadow offset that corresponds to a given
1383  /// application address.
1384  ///
1385  /// Offset = (Addr & ~AndMask) ^ XorMask
1386  Value *getShadowPtrOffset(Value *Addr, IRBuilder<> &IRB) {
1387  Value *OffsetLong = IRB.CreatePointerCast(Addr, MS.IntptrTy);
1388 
1389  uint64_t AndMask = MS.MapParams->AndMask;
1390  if (AndMask)
1391  OffsetLong =
1392  IRB.CreateAnd(OffsetLong, ConstantInt::get(MS.IntptrTy, ~AndMask));
1393 
1394  uint64_t XorMask = MS.MapParams->XorMask;
1395  if (XorMask)
1396  OffsetLong =
1397  IRB.CreateXor(OffsetLong, ConstantInt::get(MS.IntptrTy, XorMask));
1398  return OffsetLong;
1399  }
1400 
1401  /// Compute the shadow and origin addresses corresponding to a given
1402  /// application address.
1403  ///
1404  /// Shadow = ShadowBase + Offset
1405  /// Origin = (OriginBase + Offset) & ~3ULL
1406  std::pair<Value *, Value *> getShadowOriginPtrUserspace(Value *Addr,
1407  IRBuilder<> &IRB,
1408  Type *ShadowTy,
1409  unsigned Alignment) {
1410  Value *ShadowOffset = getShadowPtrOffset(Addr, IRB);
1411  Value *ShadowLong = ShadowOffset;
1412  uint64_t ShadowBase = MS.MapParams->ShadowBase;
1413  if (ShadowBase != 0) {
1414  ShadowLong =
1415  IRB.CreateAdd(ShadowLong,
1416  ConstantInt::get(MS.IntptrTy, ShadowBase));
1417  }
1418  Value *ShadowPtr =
1419  IRB.CreateIntToPtr(ShadowLong, PointerType::get(ShadowTy, 0));
1420  Value *OriginPtr = nullptr;
1421  if (MS.TrackOrigins) {
1422  Value *OriginLong = ShadowOffset;
1423  uint64_t OriginBase = MS.MapParams->OriginBase;
1424  if (OriginBase != 0)
1425  OriginLong = IRB.CreateAdd(OriginLong,
1426  ConstantInt::get(MS.IntptrTy, OriginBase));
1427  if (Alignment < kMinOriginAlignment) {
1428  uint64_t Mask = kMinOriginAlignment - 1;
1429  OriginLong =
1430  IRB.CreateAnd(OriginLong, ConstantInt::get(MS.IntptrTy, ~Mask));
1431  }
1432  OriginPtr =
1433  IRB.CreateIntToPtr(OriginLong, PointerType::get(MS.OriginTy, 0));
1434  }
1435  return std::make_pair(ShadowPtr, OriginPtr);
1436  }
1437 
1438  std::pair<Value *, Value *>
1439  getShadowOriginPtrKernel(Value *Addr, IRBuilder<> &IRB, Type *ShadowTy,
1440  unsigned Alignment, bool isStore) {
1441  Value *ShadowOriginPtrs;
1442  const DataLayout &DL = F.getParent()->getDataLayout();
1443  int Size = DL.getTypeStoreSize(ShadowTy);
1444 
1445  FunctionCallee Getter = MS.getKmsanShadowOriginAccessFn(isStore, Size);
1446  Value *AddrCast =
1447  IRB.CreatePointerCast(Addr, PointerType::get(IRB.getInt8Ty(), 0));
1448  if (Getter) {
1449  ShadowOriginPtrs = IRB.CreateCall(Getter, AddrCast);
1450  } else {
1451  Value *SizeVal = ConstantInt::get(MS.IntptrTy, Size);
1452  ShadowOriginPtrs = IRB.CreateCall(isStore ? MS.MsanMetadataPtrForStoreN
1453  : MS.MsanMetadataPtrForLoadN,
1454  {AddrCast, SizeVal});
1455  }
1456  Value *ShadowPtr = IRB.CreateExtractValue(ShadowOriginPtrs, 0);
1457  ShadowPtr = IRB.CreatePointerCast(ShadowPtr, PointerType::get(ShadowTy, 0));
1458  Value *OriginPtr = IRB.CreateExtractValue(ShadowOriginPtrs, 1);
1459 
1460  return std::make_pair(ShadowPtr, OriginPtr);
1461  }
1462 
1463  std::pair<Value *, Value *> getShadowOriginPtr(Value *Addr, IRBuilder<> &IRB,
1464  Type *ShadowTy,
1465  unsigned Alignment,
1466  bool isStore) {
1467  std::pair<Value *, Value *> ret;
1468  if (MS.CompileKernel)
1469  ret = getShadowOriginPtrKernel(Addr, IRB, ShadowTy, Alignment, isStore);
1470  else
1471  ret = getShadowOriginPtrUserspace(Addr, IRB, ShadowTy, Alignment);
1472  return ret;
1473  }
1474 
1475  /// Compute the shadow address for a given function argument.
1476  ///
1477  /// Shadow = ParamTLS+ArgOffset.
1478  Value *getShadowPtrForArgument(Value *A, IRBuilder<> &IRB,
1479  int ArgOffset) {
1480  Value *Base = IRB.CreatePointerCast(MS.ParamTLS, MS.IntptrTy);
1481  if (ArgOffset)
1482  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
1483  return IRB.CreateIntToPtr(Base, PointerType::get(getShadowTy(A), 0),
1484  "_msarg");
1485  }
1486 
1487  /// Compute the origin address for a given function argument.
1488  Value *getOriginPtrForArgument(Value *A, IRBuilder<> &IRB,
1489  int ArgOffset) {
1490  if (!MS.TrackOrigins)
1491  return nullptr;
1492  Value *Base = IRB.CreatePointerCast(MS.ParamOriginTLS, MS.IntptrTy);
1493  if (ArgOffset)
1494  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
1495  return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
1496  "_msarg_o");
1497  }
1498 
1499  /// Compute the shadow address for a retval.
1500  Value *getShadowPtrForRetval(Value *A, IRBuilder<> &IRB) {
1501  return IRB.CreatePointerCast(MS.RetvalTLS,
1502  PointerType::get(getShadowTy(A), 0),
1503  "_msret");
1504  }
1505 
1506  /// Compute the origin address for a retval.
1507  Value *getOriginPtrForRetval(IRBuilder<> &IRB) {
1508  // We keep a single origin for the entire retval. Might be too optimistic.
1509  return MS.RetvalOriginTLS;
1510  }
1511 
1512  /// Set SV to be the shadow value for V.
1513  void setShadow(Value *V, Value *SV) {
1514  assert(!ShadowMap.count(V) && "Values may only have one shadow");
1515  ShadowMap[V] = PropagateShadow ? SV : getCleanShadow(V);
1516  }
1517 
1518  /// Set Origin to be the origin value for V.
1519  void setOrigin(Value *V, Value *Origin) {
1520  if (!MS.TrackOrigins) return;
1521  assert(!OriginMap.count(V) && "Values may only have one origin");
1522  LLVM_DEBUG(dbgs() << "ORIGIN: " << *V << " ==> " << *Origin << "\n");
1523  OriginMap[V] = Origin;
1524  }
1525 
1526  Constant *getCleanShadow(Type *OrigTy) {
1527  Type *ShadowTy = getShadowTy(OrigTy);
1528  if (!ShadowTy)
1529  return nullptr;
1530  return Constant::getNullValue(ShadowTy);
1531  }
1532 
1533  /// Create a clean shadow value for a given value.
1534  ///
1535  /// Clean shadow (all zeroes) means all bits of the value are defined
1536  /// (initialized).
1537  Constant *getCleanShadow(Value *V) {
1538  return getCleanShadow(V->getType());
1539  }
1540 
1541  /// Create a dirty shadow of a given shadow type.
1542  Constant *getPoisonedShadow(Type *ShadowTy) {
1543  assert(ShadowTy);
1544  if (isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy))
1545  return Constant::getAllOnesValue(ShadowTy);
1546  if (ArrayType *AT = dyn_cast<ArrayType>(ShadowTy)) {
1547  SmallVector<Constant *, 4> Vals(AT->getNumElements(),
1548  getPoisonedShadow(AT->getElementType()));
1549  return ConstantArray::get(AT, Vals);
1550  }
1551  if (StructType *ST = dyn_cast<StructType>(ShadowTy)) {
1553  for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
1554  Vals.push_back(getPoisonedShadow(ST->getElementType(i)));
1555  return ConstantStruct::get(ST, Vals);
1556  }
1557  llvm_unreachable("Unexpected shadow type");
1558  }
1559 
1560  /// Create a dirty shadow for a given value.
1561  Constant *getPoisonedShadow(Value *V) {
1562  Type *ShadowTy = getShadowTy(V);
1563  if (!ShadowTy)
1564  return nullptr;
1565  return getPoisonedShadow(ShadowTy);
1566  }
1567 
1568  /// Create a clean (zero) origin.
1569  Value *getCleanOrigin() {
1570  return Constant::getNullValue(MS.OriginTy);
1571  }
1572 
1573  /// Get the shadow value for a given Value.
1574  ///
1575  /// This function either returns the value set earlier with setShadow,
1576  /// or extracts if from ParamTLS (for function arguments).
1577  Value *getShadow(Value *V) {
1578  if (!PropagateShadow) return getCleanShadow(V);
1579  if (Instruction *I = dyn_cast<Instruction>(V)) {
1580  if (I->getMetadata("nosanitize"))
1581  return getCleanShadow(V);
1582  // For instructions the shadow is already stored in the map.
1583  Value *Shadow = ShadowMap[V];
1584  if (!Shadow) {
1585  LLVM_DEBUG(dbgs() << "No shadow: " << *V << "\n" << *(I->getParent()));
1586  (void)I;
1587  assert(Shadow && "No shadow for a value");
1588  }
1589  return Shadow;
1590  }
1591  if (UndefValue *U = dyn_cast<UndefValue>(V)) {
1592  Value *AllOnes = PoisonUndef ? getPoisonedShadow(V) : getCleanShadow(V);
1593  LLVM_DEBUG(dbgs() << "Undef: " << *U << " ==> " << *AllOnes << "\n");
1594  (void)U;
1595  return AllOnes;
1596  }
1597  if (Argument *A = dyn_cast<Argument>(V)) {
1598  // For arguments we compute the shadow on demand and store it in the map.
1599  Value **ShadowPtr = &ShadowMap[V];
1600  if (*ShadowPtr)
1601  return *ShadowPtr;
1602  Function *F = A->getParent();
1603  IRBuilder<> EntryIRB(ActualFnStart->getFirstNonPHI());
1604  unsigned ArgOffset = 0;
1605  const DataLayout &DL = F->getParent()->getDataLayout();
1606  for (auto &FArg : F->args()) {
1607  if (!FArg.getType()->isSized()) {
1608  LLVM_DEBUG(dbgs() << "Arg is not sized\n");
1609  continue;
1610  }
1611  unsigned Size =
1612  FArg.hasByValAttr()
1613  ? DL.getTypeAllocSize(FArg.getType()->getPointerElementType())
1614  : DL.getTypeAllocSize(FArg.getType());
1615  if (A == &FArg) {
1616  bool Overflow = ArgOffset + Size > kParamTLSSize;
1617  Value *Base = getShadowPtrForArgument(&FArg, EntryIRB, ArgOffset);
1618  if (FArg.hasByValAttr()) {
1619  // ByVal pointer itself has clean shadow. We copy the actual
1620  // argument shadow to the underlying memory.
1621  // Figure out maximal valid memcpy alignment.
1622  unsigned ArgAlign = FArg.getParamAlignment();
1623  if (ArgAlign == 0) {
1624  Type *EltType = A->getType()->getPointerElementType();
1625  ArgAlign = DL.getABITypeAlignment(EltType);
1626  }
1627  Value *CpShadowPtr =
1628  getShadowOriginPtr(V, EntryIRB, EntryIRB.getInt8Ty(), ArgAlign,
1629  /*isStore*/ true)
1630  .first;
1631  // TODO(glider): need to copy origins.
1632  if (Overflow) {
1633  // ParamTLS overflow.
1634  EntryIRB.CreateMemSet(
1635  CpShadowPtr, Constant::getNullValue(EntryIRB.getInt8Ty()),
1636  Size, ArgAlign);
1637  } else {
1638  unsigned CopyAlign = std::min(ArgAlign, kShadowTLSAlignment);
1639  Value *Cpy = EntryIRB.CreateMemCpy(CpShadowPtr, CopyAlign, Base,
1640  CopyAlign, Size);
1641  LLVM_DEBUG(dbgs() << " ByValCpy: " << *Cpy << "\n");
1642  (void)Cpy;
1643  }
1644  *ShadowPtr = getCleanShadow(V);
1645  } else {
1646  if (Overflow) {
1647  // ParamTLS overflow.
1648  *ShadowPtr = getCleanShadow(V);
1649  } else {
1650  *ShadowPtr = EntryIRB.CreateAlignedLoad(getShadowTy(&FArg), Base,
1651  kShadowTLSAlignment);
1652  }
1653  }
1654  LLVM_DEBUG(dbgs()
1655  << " ARG: " << FArg << " ==> " << **ShadowPtr << "\n");
1656  if (MS.TrackOrigins && !Overflow) {
1657  Value *OriginPtr =
1658  getOriginPtrForArgument(&FArg, EntryIRB, ArgOffset);
1659  setOrigin(A, EntryIRB.CreateLoad(MS.OriginTy, OriginPtr));
1660  } else {
1661  setOrigin(A, getCleanOrigin());
1662  }
1663  }
1664  ArgOffset += alignTo(Size, kShadowTLSAlignment);
1665  }
1666  assert(*ShadowPtr && "Could not find shadow for an argument");
1667  return *ShadowPtr;
1668  }
1669  // For everything else the shadow is zero.
1670  return getCleanShadow(V);
1671  }
1672 
1673  /// Get the shadow for i-th argument of the instruction I.
1674  Value *getShadow(Instruction *I, int i) {
1675  return getShadow(I->getOperand(i));
1676  }
1677 
1678  /// Get the origin for a value.
1679  Value *getOrigin(Value *V) {
1680  if (!MS.TrackOrigins) return nullptr;
1681  if (!PropagateShadow) return getCleanOrigin();
1682  if (isa<Constant>(V)) return getCleanOrigin();
1683  assert((isa<Instruction>(V) || isa<Argument>(V)) &&
1684  "Unexpected value type in getOrigin()");
1685  if (Instruction *I = dyn_cast<Instruction>(V)) {
1686  if (I->getMetadata("nosanitize"))
1687  return getCleanOrigin();
1688  }
1689  Value *Origin = OriginMap[V];
1690  assert(Origin && "Missing origin");
1691  return Origin;
1692  }
1693 
1694  /// Get the origin for i-th argument of the instruction I.
1695  Value *getOrigin(Instruction *I, int i) {
1696  return getOrigin(I->getOperand(i));
1697  }
1698 
1699  /// Remember the place where a shadow check should be inserted.
1700  ///
1701  /// This location will be later instrumented with a check that will print a
1702  /// UMR warning in runtime if the shadow value is not 0.
1703  void insertShadowCheck(Value *Shadow, Value *Origin, Instruction *OrigIns) {
1704  assert(Shadow);
1705  if (!InsertChecks) return;
1706 #ifndef NDEBUG
1707  Type *ShadowTy = Shadow->getType();
1708  assert((isa<IntegerType>(ShadowTy) || isa<VectorType>(ShadowTy)) &&
1709  "Can only insert checks for integer and vector shadow types");
1710 #endif
1711  InstrumentationList.push_back(
1712  ShadowOriginAndInsertPoint(Shadow, Origin, OrigIns));
1713  }
1714 
1715  /// Remember the place where a shadow check should be inserted.
1716  ///
1717  /// This location will be later instrumented with a check that will print a
1718  /// UMR warning in runtime if the value is not fully defined.
1719  void insertShadowCheck(Value *Val, Instruction *OrigIns) {
1720  assert(Val);
1721  Value *Shadow, *Origin;
1722  if (ClCheckConstantShadow) {
1723  Shadow = getShadow(Val);
1724  if (!Shadow) return;
1725  Origin = getOrigin(Val);
1726  } else {
1727  Shadow = dyn_cast_or_null<Instruction>(getShadow(Val));
1728  if (!Shadow) return;
1729  Origin = dyn_cast_or_null<Instruction>(getOrigin(Val));
1730  }
1731  insertShadowCheck(Shadow, Origin, OrigIns);
1732  }
1733 
1734  AtomicOrdering addReleaseOrdering(AtomicOrdering a) {
1735  switch (a) {
1741  return AtomicOrdering::Release;
1747  }
1748  llvm_unreachable("Unknown ordering");
1749  }
1750 
1751  AtomicOrdering addAcquireOrdering(AtomicOrdering a) {
1752  switch (a) {
1758  return AtomicOrdering::Acquire;
1764  }
1765  llvm_unreachable("Unknown ordering");
1766  }
1767 
1768  // ------------------- Visitors.
1770  void visit(Instruction &I) {
1771  if (!I.getMetadata("nosanitize"))
1773  }
1774 
1775  /// Instrument LoadInst
1776  ///
1777  /// Loads the corresponding shadow and (optionally) origin.
1778  /// Optionally, checks that the load address is fully defined.
1779  void visitLoadInst(LoadInst &I) {
1780  assert(I.getType()->isSized() && "Load type must have size");
1781  assert(!I.getMetadata("nosanitize"));
1782  IRBuilder<> IRB(I.getNextNode());
1783  Type *ShadowTy = getShadowTy(&I);
1784  Value *Addr = I.getPointerOperand();
1785  Value *ShadowPtr, *OriginPtr;
1786  unsigned Alignment = I.getAlignment();
1787  if (PropagateShadow) {
1788  std::tie(ShadowPtr, OriginPtr) =
1789  getShadowOriginPtr(Addr, IRB, ShadowTy, Alignment, /*isStore*/ false);
1790  setShadow(&I,
1791  IRB.CreateAlignedLoad(ShadowTy, ShadowPtr, Alignment, "_msld"));
1792  } else {
1793  setShadow(&I, getCleanShadow(&I));
1794  }
1795 
1797  insertShadowCheck(I.getPointerOperand(), &I);
1798 
1799  if (I.isAtomic())
1800  I.setOrdering(addAcquireOrdering(I.getOrdering()));
1801 
1802  if (MS.TrackOrigins) {
1803  if (PropagateShadow) {
1804  unsigned OriginAlignment = std::max(kMinOriginAlignment, Alignment);
1805  setOrigin(
1806  &I, IRB.CreateAlignedLoad(MS.OriginTy, OriginPtr, OriginAlignment));
1807  } else {
1808  setOrigin(&I, getCleanOrigin());
1809  }
1810  }
1811  }
1812 
1813  /// Instrument StoreInst
1814  ///
1815  /// Stores the corresponding shadow and (optionally) origin.
1816  /// Optionally, checks that the store address is fully defined.
1817  void visitStoreInst(StoreInst &I) {
1818  StoreList.push_back(&I);
1820  insertShadowCheck(I.getPointerOperand(), &I);
1821  }
1822 
1823  void handleCASOrRMW(Instruction &I) {
1824  assert(isa<AtomicRMWInst>(I) || isa<AtomicCmpXchgInst>(I));
1825 
1826  IRBuilder<> IRB(&I);
1827  Value *Addr = I.getOperand(0);
1828  Value *ShadowPtr = getShadowOriginPtr(Addr, IRB, I.getType(),
1829  /*Alignment*/ 1, /*isStore*/ true)
1830  .first;
1831 
1833  insertShadowCheck(Addr, &I);
1834 
1835  // Only test the conditional argument of cmpxchg instruction.
1836  // The other argument can potentially be uninitialized, but we can not
1837  // detect this situation reliably without possible false positives.
1838  if (isa<AtomicCmpXchgInst>(I))
1839  insertShadowCheck(I.getOperand(1), &I);
1840 
1841  IRB.CreateStore(getCleanShadow(&I), ShadowPtr);
1842 
1843  setShadow(&I, getCleanShadow(&I));
1844  setOrigin(&I, getCleanOrigin());
1845  }
1846 
1847  void visitAtomicRMWInst(AtomicRMWInst &I) {
1848  handleCASOrRMW(I);
1849  I.setOrdering(addReleaseOrdering(I.getOrdering()));
1850  }
1851 
1852  void visitAtomicCmpXchgInst(AtomicCmpXchgInst &I) {
1853  handleCASOrRMW(I);
1854  I.setSuccessOrdering(addReleaseOrdering(I.getSuccessOrdering()));
1855  }
1856 
1857  // Vector manipulation.
1858  void visitExtractElementInst(ExtractElementInst &I) {
1859  insertShadowCheck(I.getOperand(1), &I);
1860  IRBuilder<> IRB(&I);
1861  setShadow(&I, IRB.CreateExtractElement(getShadow(&I, 0), I.getOperand(1),
1862  "_msprop"));
1863  setOrigin(&I, getOrigin(&I, 0));
1864  }
1865 
1866  void visitInsertElementInst(InsertElementInst &I) {
1867  insertShadowCheck(I.getOperand(2), &I);
1868  IRBuilder<> IRB(&I);
1869  setShadow(&I, IRB.CreateInsertElement(getShadow(&I, 0), getShadow(&I, 1),
1870  I.getOperand(2), "_msprop"));
1871  setOriginForNaryOp(I);
1872  }
1873 
1874  void visitShuffleVectorInst(ShuffleVectorInst &I) {
1875  insertShadowCheck(I.getOperand(2), &I);
1876  IRBuilder<> IRB(&I);
1877  setShadow(&I, IRB.CreateShuffleVector(getShadow(&I, 0), getShadow(&I, 1),
1878  I.getOperand(2), "_msprop"));
1879  setOriginForNaryOp(I);
1880  }
1881 
1882  // Casts.
1883  void visitSExtInst(SExtInst &I) {
1884  IRBuilder<> IRB(&I);
1885  setShadow(&I, IRB.CreateSExt(getShadow(&I, 0), I.getType(), "_msprop"));
1886  setOrigin(&I, getOrigin(&I, 0));
1887  }
1888 
1889  void visitZExtInst(ZExtInst &I) {
1890  IRBuilder<> IRB(&I);
1891  setShadow(&I, IRB.CreateZExt(getShadow(&I, 0), I.getType(), "_msprop"));
1892  setOrigin(&I, getOrigin(&I, 0));
1893  }
1894 
1895  void visitTruncInst(TruncInst &I) {
1896  IRBuilder<> IRB(&I);
1897  setShadow(&I, IRB.CreateTrunc(getShadow(&I, 0), I.getType(), "_msprop"));
1898  setOrigin(&I, getOrigin(&I, 0));
1899  }
1900 
1901  void visitBitCastInst(BitCastInst &I) {
1902  // Special case: if this is the bitcast (there is exactly 1 allowed) between
1903  // a musttail call and a ret, don't instrument. New instructions are not
1904  // allowed after a musttail call.
1905  if (auto *CI = dyn_cast<CallInst>(I.getOperand(0)))
1906  if (CI->isMustTailCall())
1907  return;
1908  IRBuilder<> IRB(&I);
1909  setShadow(&I, IRB.CreateBitCast(getShadow(&I, 0), getShadowTy(&I)));
1910  setOrigin(&I, getOrigin(&I, 0));
1911  }
1912 
1913  void visitPtrToIntInst(PtrToIntInst &I) {
1914  IRBuilder<> IRB(&I);
1915  setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1916  "_msprop_ptrtoint"));
1917  setOrigin(&I, getOrigin(&I, 0));
1918  }
1919 
1920  void visitIntToPtrInst(IntToPtrInst &I) {
1921  IRBuilder<> IRB(&I);
1922  setShadow(&I, IRB.CreateIntCast(getShadow(&I, 0), getShadowTy(&I), false,
1923  "_msprop_inttoptr"));
1924  setOrigin(&I, getOrigin(&I, 0));
1925  }
1926 
1927  void visitFPToSIInst(CastInst& I) { handleShadowOr(I); }
1928  void visitFPToUIInst(CastInst& I) { handleShadowOr(I); }
1929  void visitSIToFPInst(CastInst& I) { handleShadowOr(I); }
1930  void visitUIToFPInst(CastInst& I) { handleShadowOr(I); }
1931  void visitFPExtInst(CastInst& I) { handleShadowOr(I); }
1932  void visitFPTruncInst(CastInst& I) { handleShadowOr(I); }
1933 
1934  /// Propagate shadow for bitwise AND.
1935  ///
1936  /// This code is exact, i.e. if, for example, a bit in the left argument
1937  /// is defined and 0, then neither the value not definedness of the
1938  /// corresponding bit in B don't affect the resulting shadow.
1939  void visitAnd(BinaryOperator &I) {
1940  IRBuilder<> IRB(&I);
1941  // "And" of 0 and a poisoned value results in unpoisoned value.
1942  // 1&1 => 1; 0&1 => 0; p&1 => p;
1943  // 1&0 => 0; 0&0 => 0; p&0 => 0;
1944  // 1&p => p; 0&p => 0; p&p => p;
1945  // S = (S1 & S2) | (V1 & S2) | (S1 & V2)
1946  Value *S1 = getShadow(&I, 0);
1947  Value *S2 = getShadow(&I, 1);
1948  Value *V1 = I.getOperand(0);
1949  Value *V2 = I.getOperand(1);
1950  if (V1->getType() != S1->getType()) {
1951  V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1952  V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1953  }
1954  Value *S1S2 = IRB.CreateAnd(S1, S2);
1955  Value *V1S2 = IRB.CreateAnd(V1, S2);
1956  Value *S1V2 = IRB.CreateAnd(S1, V2);
1957  setShadow(&I, IRB.CreateOr({S1S2, V1S2, S1V2}));
1958  setOriginForNaryOp(I);
1959  }
1960 
1961  void visitOr(BinaryOperator &I) {
1962  IRBuilder<> IRB(&I);
1963  // "Or" of 1 and a poisoned value results in unpoisoned value.
1964  // 1|1 => 1; 0|1 => 1; p|1 => 1;
1965  // 1|0 => 1; 0|0 => 0; p|0 => p;
1966  // 1|p => 1; 0|p => p; p|p => p;
1967  // S = (S1 & S2) | (~V1 & S2) | (S1 & ~V2)
1968  Value *S1 = getShadow(&I, 0);
1969  Value *S2 = getShadow(&I, 1);
1970  Value *V1 = IRB.CreateNot(I.getOperand(0));
1971  Value *V2 = IRB.CreateNot(I.getOperand(1));
1972  if (V1->getType() != S1->getType()) {
1973  V1 = IRB.CreateIntCast(V1, S1->getType(), false);
1974  V2 = IRB.CreateIntCast(V2, S2->getType(), false);
1975  }
1976  Value *S1S2 = IRB.CreateAnd(S1, S2);
1977  Value *V1S2 = IRB.CreateAnd(V1, S2);
1978  Value *S1V2 = IRB.CreateAnd(S1, V2);
1979  setShadow(&I, IRB.CreateOr({S1S2, V1S2, S1V2}));
1980  setOriginForNaryOp(I);
1981  }
1982 
1983  /// Default propagation of shadow and/or origin.
1984  ///
1985  /// This class implements the general case of shadow propagation, used in all
1986  /// cases where we don't know and/or don't care about what the operation
1987  /// actually does. It converts all input shadow values to a common type
1988  /// (extending or truncating as necessary), and bitwise OR's them.
1989  ///
1990  /// This is much cheaper than inserting checks (i.e. requiring inputs to be
1991  /// fully initialized), and less prone to false positives.
1992  ///
1993  /// This class also implements the general case of origin propagation. For a
1994  /// Nary operation, result origin is set to the origin of an argument that is
1995  /// not entirely initialized. If there is more than one such arguments, the
1996  /// rightmost of them is picked. It does not matter which one is picked if all
1997  /// arguments are initialized.
1998  template <bool CombineShadow>
1999  class Combiner {
2000  Value *Shadow = nullptr;
2001  Value *Origin = nullptr;
2002  IRBuilder<> &IRB;
2003  MemorySanitizerVisitor *MSV;
2004 
2005  public:
2006  Combiner(MemorySanitizerVisitor *MSV, IRBuilder<> &IRB)
2007  : IRB(IRB), MSV(MSV) {}
2008 
2009  /// Add a pair of shadow and origin values to the mix.
2010  Combiner &Add(Value *OpShadow, Value *OpOrigin) {
2011  if (CombineShadow) {
2012  assert(OpShadow);
2013  if (!Shadow)
2014  Shadow = OpShadow;
2015  else {
2016  OpShadow = MSV->CreateShadowCast(IRB, OpShadow, Shadow->getType());
2017  Shadow = IRB.CreateOr(Shadow, OpShadow, "_msprop");
2018  }
2019  }
2020 
2021  if (MSV->MS.TrackOrigins) {
2022  assert(OpOrigin);
2023  if (!Origin) {
2024  Origin = OpOrigin;
2025  } else {
2026  Constant *ConstOrigin = dyn_cast<Constant>(OpOrigin);
2027  // No point in adding something that might result in 0 origin value.
2028  if (!ConstOrigin || !ConstOrigin->isNullValue()) {
2029  Value *FlatShadow = MSV->convertToShadowTyNoVec(OpShadow, IRB);
2030  Value *Cond =
2031  IRB.CreateICmpNE(FlatShadow, MSV->getCleanShadow(FlatShadow));
2032  Origin = IRB.CreateSelect(Cond, OpOrigin, Origin);
2033  }
2034  }
2035  }
2036  return *this;
2037  }
2038 
2039  /// Add an application value to the mix.
2040  Combiner &Add(Value *V) {
2041  Value *OpShadow = MSV->getShadow(V);
2042  Value *OpOrigin = MSV->MS.TrackOrigins ? MSV->getOrigin(V) : nullptr;
2043  return Add(OpShadow, OpOrigin);
2044  }
2045 
2046  /// Set the current combined values as the given instruction's shadow
2047  /// and origin.
2048  void Done(Instruction *I) {
2049  if (CombineShadow) {
2050  assert(Shadow);
2051  Shadow = MSV->CreateShadowCast(IRB, Shadow, MSV->getShadowTy(I));
2052  MSV->setShadow(I, Shadow);
2053  }
2054  if (MSV->MS.TrackOrigins) {
2055  assert(Origin);
2056  MSV->setOrigin(I, Origin);
2057  }
2058  }
2059  };
2060 
2061  using ShadowAndOriginCombiner = Combiner<true>;
2062  using OriginCombiner = Combiner<false>;
2063 
2064  /// Propagate origin for arbitrary operation.
2065  void setOriginForNaryOp(Instruction &I) {
2066  if (!MS.TrackOrigins) return;
2067  IRBuilder<> IRB(&I);
2068  OriginCombiner OC(this, IRB);
2069  for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
2070  OC.Add(OI->get());
2071  OC.Done(&I);
2072  }
2073 
2074  size_t VectorOrPrimitiveTypeSizeInBits(Type *Ty) {
2075  assert(!(Ty->isVectorTy() && Ty->getScalarType()->isPointerTy()) &&
2076  "Vector of pointers is not a valid shadow type");
2077  return Ty->isVectorTy() ?
2079  Ty->getPrimitiveSizeInBits();
2080  }
2081 
2082  /// Cast between two shadow types, extending or truncating as
2083  /// necessary.
2084  Value *CreateShadowCast(IRBuilder<> &IRB, Value *V, Type *dstTy,
2085  bool Signed = false) {
2086  Type *srcTy = V->getType();
2087  size_t srcSizeInBits = VectorOrPrimitiveTypeSizeInBits(srcTy);
2088  size_t dstSizeInBits = VectorOrPrimitiveTypeSizeInBits(dstTy);
2089  if (srcSizeInBits > 1 && dstSizeInBits == 1)
2090  return IRB.CreateICmpNE(V, getCleanShadow(V));
2091 
2092  if (dstTy->isIntegerTy() && srcTy->isIntegerTy())
2093  return IRB.CreateIntCast(V, dstTy, Signed);
2094  if (dstTy->isVectorTy() && srcTy->isVectorTy() &&
2095  dstTy->getVectorNumElements() == srcTy->getVectorNumElements())
2096  return IRB.CreateIntCast(V, dstTy, Signed);
2097  Value *V1 = IRB.CreateBitCast(V, Type::getIntNTy(*MS.C, srcSizeInBits));
2098  Value *V2 =
2099  IRB.CreateIntCast(V1, Type::getIntNTy(*MS.C, dstSizeInBits), Signed);
2100  return IRB.CreateBitCast(V2, dstTy);
2101  // TODO: handle struct types.
2102  }
2103 
2104  /// Cast an application value to the type of its own shadow.
2105  Value *CreateAppToShadowCast(IRBuilder<> &IRB, Value *V) {
2106  Type *ShadowTy = getShadowTy(V);
2107  if (V->getType() == ShadowTy)
2108  return V;
2109  if (V->getType()->isPtrOrPtrVectorTy())
2110  return IRB.CreatePtrToInt(V, ShadowTy);
2111  else
2112  return IRB.CreateBitCast(V, ShadowTy);
2113  }
2114 
2115  /// Propagate shadow for arbitrary operation.
2116  void handleShadowOr(Instruction &I) {
2117  IRBuilder<> IRB(&I);
2118  ShadowAndOriginCombiner SC(this, IRB);
2119  for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
2120  SC.Add(OI->get());
2121  SC.Done(&I);
2122  }
2123 
2124  void visitFNeg(UnaryOperator &I) { handleShadowOr(I); }
2125 
2126  // Handle multiplication by constant.
2127  //
2128  // Handle a special case of multiplication by constant that may have one or
2129  // more zeros in the lower bits. This makes corresponding number of lower bits
2130  // of the result zero as well. We model it by shifting the other operand
2131  // shadow left by the required number of bits. Effectively, we transform
2132  // (X * (A * 2**B)) to ((X << B) * A) and instrument (X << B) as (Sx << B).
2133  // We use multiplication by 2**N instead of shift to cover the case of
2134  // multiplication by 0, which may occur in some elements of a vector operand.
2135  void handleMulByConstant(BinaryOperator &I, Constant *ConstArg,
2136  Value *OtherArg) {
2137  Constant *ShadowMul;
2138  Type *Ty = ConstArg->getType();
2139  if (Ty->isVectorTy()) {
2140  unsigned NumElements = Ty->getVectorNumElements();
2141  Type *EltTy = Ty->getSequentialElementType();
2142  SmallVector<Constant *, 16> Elements;
2143  for (unsigned Idx = 0; Idx < NumElements; ++Idx) {
2144  if (ConstantInt *Elt =
2145  dyn_cast<ConstantInt>(ConstArg->getAggregateElement(Idx))) {
2146  const APInt &V = Elt->getValue();
2147  APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
2148  Elements.push_back(ConstantInt::get(EltTy, V2));
2149  } else {
2150  Elements.push_back(ConstantInt::get(EltTy, 1));
2151  }
2152  }
2153  ShadowMul = ConstantVector::get(Elements);
2154  } else {
2155  if (ConstantInt *Elt = dyn_cast<ConstantInt>(ConstArg)) {
2156  const APInt &V = Elt->getValue();
2157  APInt V2 = APInt(V.getBitWidth(), 1) << V.countTrailingZeros();
2158  ShadowMul = ConstantInt::get(Ty, V2);
2159  } else {
2160  ShadowMul = ConstantInt::get(Ty, 1);
2161  }
2162  }
2163 
2164  IRBuilder<> IRB(&I);
2165  setShadow(&I,
2166  IRB.CreateMul(getShadow(OtherArg), ShadowMul, "msprop_mul_cst"));
2167  setOrigin(&I, getOrigin(OtherArg));
2168  }
2169 
2170  void visitMul(BinaryOperator &I) {
2171  Constant *constOp0 = dyn_cast<Constant>(I.getOperand(0));
2172  Constant *constOp1 = dyn_cast<Constant>(I.getOperand(1));
2173  if (constOp0 && !constOp1)
2174  handleMulByConstant(I, constOp0, I.getOperand(1));
2175  else if (constOp1 && !constOp0)
2176  handleMulByConstant(I, constOp1, I.getOperand(0));
2177  else
2178  handleShadowOr(I);
2179  }
2180 
2181  void visitFAdd(BinaryOperator &I) { handleShadowOr(I); }
2182  void visitFSub(BinaryOperator &I) { handleShadowOr(I); }
2183  void visitFMul(BinaryOperator &I) { handleShadowOr(I); }
2184  void visitAdd(BinaryOperator &I) { handleShadowOr(I); }
2185  void visitSub(BinaryOperator &I) { handleShadowOr(I); }
2186  void visitXor(BinaryOperator &I) { handleShadowOr(I); }
2187 
2188  void handleIntegerDiv(Instruction &I) {
2189  IRBuilder<> IRB(&I);
2190  // Strict on the second argument.
2191  insertShadowCheck(I.getOperand(1), &I);
2192  setShadow(&I, getShadow(&I, 0));
2193  setOrigin(&I, getOrigin(&I, 0));
2194  }
2195 
2196  void visitUDiv(BinaryOperator &I) { handleIntegerDiv(I); }
2197  void visitSDiv(BinaryOperator &I) { handleIntegerDiv(I); }
2198  void visitURem(BinaryOperator &I) { handleIntegerDiv(I); }
2199  void visitSRem(BinaryOperator &I) { handleIntegerDiv(I); }
2200 
2201  // Floating point division is side-effect free. We can not require that the
2202  // divisor is fully initialized and must propagate shadow. See PR37523.
2203  void visitFDiv(BinaryOperator &I) { handleShadowOr(I); }
2204  void visitFRem(BinaryOperator &I) { handleShadowOr(I); }
2205 
2206  /// Instrument == and != comparisons.
2207  ///
2208  /// Sometimes the comparison result is known even if some of the bits of the
2209  /// arguments are not.
2210  void handleEqualityComparison(ICmpInst &I) {
2211  IRBuilder<> IRB(&I);
2212  Value *A = I.getOperand(0);
2213  Value *B = I.getOperand(1);
2214  Value *Sa = getShadow(A);
2215  Value *Sb = getShadow(B);
2216 
2217  // Get rid of pointers and vectors of pointers.
2218  // For ints (and vectors of ints), types of A and Sa match,
2219  // and this is a no-op.
2220  A = IRB.CreatePointerCast(A, Sa->getType());
2221  B = IRB.CreatePointerCast(B, Sb->getType());
2222 
2223  // A == B <==> (C = A^B) == 0
2224  // A != B <==> (C = A^B) != 0
2225  // Sc = Sa | Sb
2226  Value *C = IRB.CreateXor(A, B);
2227  Value *Sc = IRB.CreateOr(Sa, Sb);
2228  // Now dealing with i = (C == 0) comparison (or C != 0, does not matter now)
2229  // Result is defined if one of the following is true
2230  // * there is a defined 1 bit in C
2231  // * C is fully defined
2232  // Si = !(C & ~Sc) && Sc
2233  Value *Zero = Constant::getNullValue(Sc->getType());
2234  Value *MinusOne = Constant::getAllOnesValue(Sc->getType());
2235  Value *Si =
2236  IRB.CreateAnd(IRB.CreateICmpNE(Sc, Zero),
2237  IRB.CreateICmpEQ(
2238  IRB.CreateAnd(IRB.CreateXor(Sc, MinusOne), C), Zero));
2239  Si->setName("_msprop_icmp");
2240  setShadow(&I, Si);
2241  setOriginForNaryOp(I);
2242  }
2243 
2244  /// Build the lowest possible value of V, taking into account V's
2245  /// uninitialized bits.
2246  Value *getLowestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
2247  bool isSigned) {
2248  if (isSigned) {
2249  // Split shadow into sign bit and other bits.
2250  Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
2251  Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
2252  // Maximise the undefined shadow bit, minimize other undefined bits.
2253  return
2254  IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaOtherBits)), SaSignBit);
2255  } else {
2256  // Minimize undefined bits.
2257  return IRB.CreateAnd(A, IRB.CreateNot(Sa));
2258  }
2259  }
2260 
2261  /// Build the highest possible value of V, taking into account V's
2262  /// uninitialized bits.
2263  Value *getHighestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
2264  bool isSigned) {
2265  if (isSigned) {
2266  // Split shadow into sign bit and other bits.
2267  Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
2268  Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
2269  // Minimise the undefined shadow bit, maximise other undefined bits.
2270  return
2271  IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaSignBit)), SaOtherBits);
2272  } else {
2273  // Maximize undefined bits.
2274  return IRB.CreateOr(A, Sa);
2275  }
2276  }
2277 
2278  /// Instrument relational comparisons.
2279  ///
2280  /// This function does exact shadow propagation for all relational
2281  /// comparisons of integers, pointers and vectors of those.
2282  /// FIXME: output seems suboptimal when one of the operands is a constant
2283  void handleRelationalComparisonExact(ICmpInst &I) {
2284  IRBuilder<> IRB(&I);
2285  Value *A = I.getOperand(0);
2286  Value *B = I.getOperand(1);
2287  Value *Sa = getShadow(A);
2288  Value *Sb = getShadow(B);
2289 
2290  // Get rid of pointers and vectors of pointers.
2291  // For ints (and vectors of ints), types of A and Sa match,
2292  // and this is a no-op.
2293  A = IRB.CreatePointerCast(A, Sa->getType());
2294  B = IRB.CreatePointerCast(B, Sb->getType());
2295 
2296  // Let [a0, a1] be the interval of possible values of A, taking into account
2297  // its undefined bits. Let [b0, b1] be the interval of possible values of B.
2298  // Then (A cmp B) is defined iff (a0 cmp b1) == (a1 cmp b0).
2299  bool IsSigned = I.isSigned();
2300  Value *S1 = IRB.CreateICmp(I.getPredicate(),
2301  getLowestPossibleValue(IRB, A, Sa, IsSigned),
2302  getHighestPossibleValue(IRB, B, Sb, IsSigned));
2303  Value *S2 = IRB.CreateICmp(I.getPredicate(),
2304  getHighestPossibleValue(IRB, A, Sa, IsSigned),
2305  getLowestPossibleValue(IRB, B, Sb, IsSigned));
2306  Value *Si = IRB.CreateXor(S1, S2);
2307  setShadow(&I, Si);
2308  setOriginForNaryOp(I);
2309  }
2310 
2311  /// Instrument signed relational comparisons.
2312  ///
2313  /// Handle sign bit tests: x<0, x>=0, x<=-1, x>-1 by propagating the highest
2314  /// bit of the shadow. Everything else is delegated to handleShadowOr().
2315  void handleSignedRelationalComparison(ICmpInst &I) {
2316  Constant *constOp;
2317  Value *op = nullptr;
2318  CmpInst::Predicate pre;
2319  if ((constOp = dyn_cast<Constant>(I.getOperand(1)))) {
2320  op = I.getOperand(0);
2321  pre = I.getPredicate();
2322  } else if ((constOp = dyn_cast<Constant>(I.getOperand(0)))) {
2323  op = I.getOperand(1);
2324  pre = I.getSwappedPredicate();
2325  } else {
2326  handleShadowOr(I);
2327  return;
2328  }
2329 
2330  if ((constOp->isNullValue() &&
2331  (pre == CmpInst::ICMP_SLT || pre == CmpInst::ICMP_SGE)) ||
2332  (constOp->isAllOnesValue() &&
2333  (pre == CmpInst::ICMP_SGT || pre == CmpInst::ICMP_SLE))) {
2334  IRBuilder<> IRB(&I);
2335  Value *Shadow = IRB.CreateICmpSLT(getShadow(op), getCleanShadow(op),
2336  "_msprop_icmp_s");
2337  setShadow(&I, Shadow);
2338  setOrigin(&I, getOrigin(op));
2339  } else {
2340  handleShadowOr(I);
2341  }
2342  }
2343 
2344  void visitICmpInst(ICmpInst &I) {
2345  if (!ClHandleICmp) {
2346  handleShadowOr(I);
2347  return;
2348  }
2349  if (I.isEquality()) {
2350  handleEqualityComparison(I);
2351  return;
2352  }
2353 
2354  assert(I.isRelational());
2355  if (ClHandleICmpExact) {
2356  handleRelationalComparisonExact(I);
2357  return;
2358  }
2359  if (I.isSigned()) {
2360  handleSignedRelationalComparison(I);
2361  return;
2362  }
2363 
2364  assert(I.isUnsigned());
2365  if ((isa<Constant>(I.getOperand(0)) || isa<Constant>(I.getOperand(1)))) {
2366  handleRelationalComparisonExact(I);
2367  return;
2368  }
2369 
2370  handleShadowOr(I);
2371  }
2372 
2373  void visitFCmpInst(FCmpInst &I) {
2374  handleShadowOr(I);
2375  }
2376 
2377  void handleShift(BinaryOperator &I) {
2378  IRBuilder<> IRB(&I);
2379  // If any of the S2 bits are poisoned, the whole thing is poisoned.
2380  // Otherwise perform the same shift on S1.
2381  Value *S1 = getShadow(&I, 0);
2382  Value *S2 = getShadow(&I, 1);
2383  Value *S2Conv = IRB.CreateSExt(IRB.CreateICmpNE(S2, getCleanShadow(S2)),
2384  S2->getType());
2385  Value *V2 = I.getOperand(1);
2386  Value *Shift = IRB.CreateBinOp(I.getOpcode(), S1, V2);
2387  setShadow(&I, IRB.CreateOr(Shift, S2Conv));
2388  setOriginForNaryOp(I);
2389  }
2390 
2391  void visitShl(BinaryOperator &I) { handleShift(I); }
2392  void visitAShr(BinaryOperator &I) { handleShift(I); }
2393  void visitLShr(BinaryOperator &I) { handleShift(I); }
2394 
2395  /// Instrument llvm.memmove
2396  ///
2397  /// At this point we don't know if llvm.memmove will be inlined or not.
2398  /// If we don't instrument it and it gets inlined,
2399  /// our interceptor will not kick in and we will lose the memmove.
2400  /// If we instrument the call here, but it does not get inlined,
2401  /// we will memove the shadow twice: which is bad in case
2402  /// of overlapping regions. So, we simply lower the intrinsic to a call.
2403  ///
2404  /// Similar situation exists for memcpy and memset.
2405  void visitMemMoveInst(MemMoveInst &I) {
2406  IRBuilder<> IRB(&I);
2407  IRB.CreateCall(
2408  MS.MemmoveFn,
2409  {IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
2410  IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
2411  IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false)});
2412  I.eraseFromParent();
2413  }
2414 
2415  // Similar to memmove: avoid copying shadow twice.
2416  // This is somewhat unfortunate as it may slowdown small constant memcpys.
2417  // FIXME: consider doing manual inline for small constant sizes and proper
2418  // alignment.
2419  void visitMemCpyInst(MemCpyInst &I) {
2420  IRBuilder<> IRB(&I);
2421  IRB.CreateCall(
2422  MS.MemcpyFn,
2423  {IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
2424  IRB.CreatePointerCast(I.getArgOperand(1), IRB.getInt8PtrTy()),
2425  IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false)});
2426  I.eraseFromParent();
2427  }
2428 
2429  // Same as memcpy.
2430  void visitMemSetInst(MemSetInst &I) {
2431  IRBuilder<> IRB(&I);
2432  IRB.CreateCall(
2433  MS.MemsetFn,
2434  {IRB.CreatePointerCast(I.getArgOperand(0), IRB.getInt8PtrTy()),
2435  IRB.CreateIntCast(I.getArgOperand(1), IRB.getInt32Ty(), false),
2436  IRB.CreateIntCast(I.getArgOperand(2), MS.IntptrTy, false)});
2437  I.eraseFromParent();
2438  }
2439 
2440  void visitVAStartInst(VAStartInst &I) {
2441  VAHelper->visitVAStartInst(I);
2442  }
2443 
2444  void visitVACopyInst(VACopyInst &I) {
2445  VAHelper->visitVACopyInst(I);
2446  }
2447 
2448  /// Handle vector store-like intrinsics.
2449  ///
2450  /// Instrument intrinsics that look like a simple SIMD store: writes memory,
2451  /// has 1 pointer argument and 1 vector argument, returns void.
2452  bool handleVectorStoreIntrinsic(IntrinsicInst &I) {
2453  IRBuilder<> IRB(&I);
2454  Value* Addr = I.getArgOperand(0);
2455  Value *Shadow = getShadow(&I, 1);
2456  Value *ShadowPtr, *OriginPtr;
2457 
2458  // We don't know the pointer alignment (could be unaligned SSE store!).
2459  // Have to assume to worst case.
2460  std::tie(ShadowPtr, OriginPtr) = getShadowOriginPtr(
2461  Addr, IRB, Shadow->getType(), /*Alignment*/ 1, /*isStore*/ true);
2462  IRB.CreateAlignedStore(Shadow, ShadowPtr, 1);
2463 
2465  insertShadowCheck(Addr, &I);
2466 
2467  // FIXME: factor out common code from materializeStores
2468  if (MS.TrackOrigins) IRB.CreateStore(getOrigin(&I, 1), OriginPtr);
2469  return true;
2470  }
2471 
2472  /// Handle vector load-like intrinsics.
2473  ///
2474  /// Instrument intrinsics that look like a simple SIMD load: reads memory,
2475  /// has 1 pointer argument, returns a vector.
2476  bool handleVectorLoadIntrinsic(IntrinsicInst &I) {
2477  IRBuilder<> IRB(&I);
2478  Value *Addr = I.getArgOperand(0);
2479 
2480  Type *ShadowTy = getShadowTy(&I);
2481  Value *ShadowPtr, *OriginPtr;
2482  if (PropagateShadow) {
2483  // We don't know the pointer alignment (could be unaligned SSE load!).
2484  // Have to assume to worst case.
2485  unsigned Alignment = 1;
2486  std::tie(ShadowPtr, OriginPtr) =
2487  getShadowOriginPtr(Addr, IRB, ShadowTy, Alignment, /*isStore*/ false);
2488  setShadow(&I,
2489  IRB.CreateAlignedLoad(ShadowTy, ShadowPtr, Alignment, "_msld"));
2490  } else {
2491  setShadow(&I, getCleanShadow(&I));
2492  }
2493 
2495  insertShadowCheck(Addr, &I);
2496 
2497  if (MS.TrackOrigins) {
2498  if (PropagateShadow)
2499  setOrigin(&I, IRB.CreateLoad(MS.OriginTy, OriginPtr));
2500  else
2501  setOrigin(&I, getCleanOrigin());
2502  }
2503  return true;
2504  }
2505 
2506  /// Handle (SIMD arithmetic)-like intrinsics.
2507  ///
2508  /// Instrument intrinsics with any number of arguments of the same type,
2509  /// equal to the return type. The type should be simple (no aggregates or
2510  /// pointers; vectors are fine).
2511  /// Caller guarantees that this intrinsic does not access memory.
2512  bool maybeHandleSimpleNomemIntrinsic(IntrinsicInst &I) {
2513  Type *RetTy = I.getType();
2514  if (!(RetTy->isIntOrIntVectorTy() ||
2515  RetTy->isFPOrFPVectorTy() ||
2516  RetTy->isX86_MMXTy()))
2517  return false;
2518 
2519  unsigned NumArgOperands = I.getNumArgOperands();
2520 
2521  for (unsigned i = 0; i < NumArgOperands; ++i) {
2522  Type *Ty = I.getArgOperand(i)->getType();
2523  if (Ty != RetTy)
2524  return false;
2525  }
2526 
2527  IRBuilder<> IRB(&I);
2528  ShadowAndOriginCombiner SC(this, IRB);
2529  for (unsigned i = 0; i < NumArgOperands; ++i)
2530  SC.Add(I.getArgOperand(i));
2531  SC.Done(&I);
2532 
2533  return true;
2534  }
2535 
2536  /// Heuristically instrument unknown intrinsics.
2537  ///
2538  /// The main purpose of this code is to do something reasonable with all
2539  /// random intrinsics we might encounter, most importantly - SIMD intrinsics.
2540  /// We recognize several classes of intrinsics by their argument types and
2541  /// ModRefBehaviour and apply special intrumentation when we are reasonably
2542  /// sure that we know what the intrinsic does.
2543  ///
2544  /// We special-case intrinsics where this approach fails. See llvm.bswap
2545  /// handling as an example of that.
2546  bool handleUnknownIntrinsic(IntrinsicInst &I) {
2547  unsigned NumArgOperands = I.getNumArgOperands();
2548  if (NumArgOperands == 0)
2549  return false;
2550 
2551  if (NumArgOperands == 2 &&
2552  I.getArgOperand(0)->getType()->isPointerTy() &&
2553  I.getArgOperand(1)->getType()->isVectorTy() &&
2554  I.getType()->isVoidTy() &&
2555  !I.onlyReadsMemory()) {
2556  // This looks like a vector store.
2557  return handleVectorStoreIntrinsic(I);
2558  }
2559 
2560  if (NumArgOperands == 1 &&
2561  I.getArgOperand(0)->getType()->isPointerTy() &&
2562  I.getType()->isVectorTy() &&
2563  I.onlyReadsMemory()) {
2564  // This looks like a vector load.
2565  return handleVectorLoadIntrinsic(I);
2566  }
2567 
2568  if (I.doesNotAccessMemory())
2569  if (maybeHandleSimpleNomemIntrinsic(I))
2570  return true;
2571 
2572  // FIXME: detect and handle SSE maskstore/maskload
2573  return false;
2574  }
2575 
2576  void handleInvariantGroup(IntrinsicInst &I) {
2577  setShadow(&I, getShadow(&I, 0));
2578  setOrigin(&I, getOrigin(&I, 0));
2579  }
2580 
2581  void handleLifetimeStart(IntrinsicInst &I) {
2582  if (!PoisonStack)
2583  return;
2584  DenseMap<Value *, AllocaInst *> AllocaForValue;
2585  AllocaInst *AI =
2586  llvm::findAllocaForValue(I.getArgOperand(1), AllocaForValue);
2587  if (!AI)
2588  InstrumentLifetimeStart = false;
2589  LifetimeStartList.push_back(std::make_pair(&I, AI));
2590  }
2591 
2592  void handleBswap(IntrinsicInst &I) {
2593  IRBuilder<> IRB(&I);
2594  Value *Op = I.getArgOperand(0);
2595  Type *OpType = Op->getType();
2596  Function *BswapFunc = Intrinsic::getDeclaration(
2597  F.getParent(), Intrinsic::bswap, makeArrayRef(&OpType, 1));
2598  setShadow(&I, IRB.CreateCall(BswapFunc, getShadow(Op)));
2599  setOrigin(&I, getOrigin(Op));
2600  }
2601 
2602  // Instrument vector convert instrinsic.
2603  //
2604  // This function instruments intrinsics like cvtsi2ss:
2605  // %Out = int_xxx_cvtyyy(%ConvertOp)
2606  // or
2607  // %Out = int_xxx_cvtyyy(%CopyOp, %ConvertOp)
2608  // Intrinsic converts \p NumUsedElements elements of \p ConvertOp to the same
2609  // number \p Out elements, and (if has 2 arguments) copies the rest of the
2610  // elements from \p CopyOp.
2611  // In most cases conversion involves floating-point value which may trigger a
2612  // hardware exception when not fully initialized. For this reason we require
2613  // \p ConvertOp[0:NumUsedElements] to be fully initialized and trap otherwise.
2614  // We copy the shadow of \p CopyOp[NumUsedElements:] to \p
2615  // Out[NumUsedElements:]. This means that intrinsics without \p CopyOp always
2616  // return a fully initialized value.
2617  void handleVectorConvertIntrinsic(IntrinsicInst &I, int NumUsedElements) {
2618  IRBuilder<> IRB(&I);
2619  Value *CopyOp, *ConvertOp;
2620 
2621  switch (I.getNumArgOperands()) {
2622  case 3:
2623  assert(isa<ConstantInt>(I.getArgOperand(2)) && "Invalid rounding mode");
2625  case 2:
2626  CopyOp = I.getArgOperand(0);
2627  ConvertOp = I.getArgOperand(1);
2628  break;
2629  case 1:
2630  ConvertOp = I.getArgOperand(0);
2631  CopyOp = nullptr;
2632  break;
2633  default:
2634  llvm_unreachable("Cvt intrinsic with unsupported number of arguments.");
2635  }
2636 
2637  // The first *NumUsedElements* elements of ConvertOp are converted to the
2638  // same number of output elements. The rest of the output is copied from
2639  // CopyOp, or (if not available) filled with zeroes.
2640  // Combine shadow for elements of ConvertOp that are used in this operation,
2641  // and insert a check.
2642  // FIXME: consider propagating shadow of ConvertOp, at least in the case of
2643  // int->any conversion.
2644  Value *ConvertShadow = getShadow(ConvertOp);
2645  Value *AggShadow = nullptr;
2646  if (ConvertOp->getType()->isVectorTy()) {
2647  AggShadow = IRB.CreateExtractElement(
2648  ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), 0));
2649  for (int i = 1; i < NumUsedElements; ++i) {
2650  Value *MoreShadow = IRB.CreateExtractElement(
2651  ConvertShadow, ConstantInt::get(IRB.getInt32Ty(), i));
2652  AggShadow = IRB.CreateOr(AggShadow, MoreShadow);
2653  }
2654  } else {
2655  AggShadow = ConvertShadow;
2656  }
2657  assert(AggShadow->getType()->isIntegerTy());
2658  insertShadowCheck(AggShadow, getOrigin(ConvertOp), &I);
2659 
2660  // Build result shadow by zero-filling parts of CopyOp shadow that come from
2661  // ConvertOp.
2662  if (CopyOp) {
2663  assert(CopyOp->getType() == I.getType());
2664  assert(CopyOp->getType()->isVectorTy());
2665  Value *ResultShadow = getShadow(CopyOp);
2666  Type *EltTy = ResultShadow->getType()->getVectorElementType();
2667  for (int i = 0; i < NumUsedElements; ++i) {
2668  ResultShadow = IRB.CreateInsertElement(
2669  ResultShadow, ConstantInt::getNullValue(EltTy),
2670  ConstantInt::get(IRB.getInt32Ty(), i));
2671  }
2672  setShadow(&I, ResultShadow);
2673  setOrigin(&I, getOrigin(CopyOp));
2674  } else {
2675  setShadow(&I, getCleanShadow(&I));
2676  setOrigin(&I, getCleanOrigin());
2677  }
2678  }
2679 
2680  // Given a scalar or vector, extract lower 64 bits (or less), and return all
2681  // zeroes if it is zero, and all ones otherwise.
2682  Value *Lower64ShadowExtend(IRBuilder<> &IRB, Value *S, Type *T) {
2683  if (S->getType()->isVectorTy())
2684  S = CreateShadowCast(IRB, S, IRB.getInt64Ty(), /* Signed */ true);
2685  assert(S->getType()->getPrimitiveSizeInBits() <= 64);
2686  Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
2687  return CreateShadowCast(IRB, S2, T, /* Signed */ true);
2688  }
2689 
2690  // Given a vector, extract its first element, and return all
2691  // zeroes if it is zero, and all ones otherwise.
2692  Value *LowerElementShadowExtend(IRBuilder<> &IRB, Value *S, Type *T) {
2693  Value *S1 = IRB.CreateExtractElement(S, (uint64_t)0);
2694  Value *S2 = IRB.CreateICmpNE(S1, getCleanShadow(S1));
2695  return CreateShadowCast(IRB, S2, T, /* Signed */ true);
2696  }
2697 
2698  Value *VariableShadowExtend(IRBuilder<> &IRB, Value *S) {
2699  Type *T = S->getType();
2700  assert(T->isVectorTy());
2701  Value *S2 = IRB.CreateICmpNE(S, getCleanShadow(S));
2702  return IRB.CreateSExt(S2, T);
2703  }
2704 
2705  // Instrument vector shift instrinsic.
2706  //
2707  // This function instruments intrinsics like int_x86_avx2_psll_w.
2708  // Intrinsic shifts %In by %ShiftSize bits.
2709  // %ShiftSize may be a vector. In that case the lower 64 bits determine shift
2710  // size, and the rest is ignored. Behavior is defined even if shift size is
2711  // greater than register (or field) width.
2712  void handleVectorShiftIntrinsic(IntrinsicInst &I, bool Variable) {
2713  assert(I.getNumArgOperands() == 2);
2714  IRBuilder<> IRB(&I);
2715  // If any of the S2 bits are poisoned, the whole thing is poisoned.
2716  // Otherwise perform the same shift on S1.
2717  Value *S1 = getShadow(&I, 0);
2718  Value *S2 = getShadow(&I, 1);
2719  Value *S2Conv = Variable ? VariableShadowExtend(IRB, S2)
2720  : Lower64ShadowExtend(IRB, S2, getShadowTy(&I));
2721  Value *V1 = I.getOperand(0);
2722  Value *V2 = I.getOperand(1);
2723  Value *Shift = IRB.CreateCall(I.getFunctionType(), I.getCalledValue(),
2724  {IRB.CreateBitCast(S1, V1->getType()), V2});
2725  Shift = IRB.CreateBitCast(Shift, getShadowTy(&I));
2726  setShadow(&I, IRB.CreateOr(Shift, S2Conv));
2727  setOriginForNaryOp(I);
2728  }
2729 
2730  // Get an X86_MMX-sized vector type.
2731  Type *getMMXVectorTy(unsigned EltSizeInBits) {
2732  const unsigned X86_MMXSizeInBits = 64;
2733  assert(EltSizeInBits != 0 && (X86_MMXSizeInBits % EltSizeInBits) == 0 &&
2734  "Illegal MMX vector element size");
2735  return VectorType::get(IntegerType::get(*MS.C, EltSizeInBits),
2736  X86_MMXSizeInBits / EltSizeInBits);
2737  }
2738 
2739  // Returns a signed counterpart for an (un)signed-saturate-and-pack
2740  // intrinsic.
2741  Intrinsic::ID getSignedPackIntrinsic(Intrinsic::ID id) {
2742  switch (id) {
2743  case Intrinsic::x86_sse2_packsswb_128:
2744  case Intrinsic::x86_sse2_packuswb_128:
2745  return Intrinsic::x86_sse2_packsswb_128;
2746 
2747  case Intrinsic::x86_sse2_packssdw_128:
2748  case Intrinsic::x86_sse41_packusdw:
2749  return Intrinsic::x86_sse2_packssdw_128;
2750 
2751  case Intrinsic::x86_avx2_packsswb:
2752  case Intrinsic::x86_avx2_packuswb:
2753  return Intrinsic::x86_avx2_packsswb;
2754 
2755  case Intrinsic::x86_avx2_packssdw:
2756  case Intrinsic::x86_avx2_packusdw:
2757  return Intrinsic::x86_avx2_packssdw;
2758 
2759  case Intrinsic::x86_mmx_packsswb:
2760  case Intrinsic::x86_mmx_packuswb:
2761  return Intrinsic::x86_mmx_packsswb;
2762 
2763  case Intrinsic::x86_mmx_packssdw:
2764  return Intrinsic::x86_mmx_packssdw;
2765  default:
2766  llvm_unreachable("unexpected intrinsic id");
2767  }
2768  }
2769 
2770  // Instrument vector pack instrinsic.
2771  //
2772  // This function instruments intrinsics like x86_mmx_packsswb, that
2773  // packs elements of 2 input vectors into half as many bits with saturation.
2774  // Shadow is propagated with the signed variant of the same intrinsic applied
2775  // to sext(Sa != zeroinitializer), sext(Sb != zeroinitializer).
2776  // EltSizeInBits is used only for x86mmx arguments.
2777  void handleVectorPackIntrinsic(IntrinsicInst &I, unsigned EltSizeInBits = 0) {
2778  assert(I.getNumArgOperands() == 2);
2779  bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2780  IRBuilder<> IRB(&I);
2781  Value *S1 = getShadow(&I, 0);
2782  Value *S2 = getShadow(&I, 1);
2783  assert(isX86_MMX || S1->getType()->isVectorTy());
2784 
2785  // SExt and ICmpNE below must apply to individual elements of input vectors.
2786  // In case of x86mmx arguments, cast them to appropriate vector types and
2787  // back.
2788  Type *T = isX86_MMX ? getMMXVectorTy(EltSizeInBits) : S1->getType();
2789  if (isX86_MMX) {
2790  S1 = IRB.CreateBitCast(S1, T);
2791  S2 = IRB.CreateBitCast(S2, T);
2792  }
2793  Value *S1_ext = IRB.CreateSExt(
2794  IRB.CreateICmpNE(S1, Constant::getNullValue(T)), T);
2795  Value *S2_ext = IRB.CreateSExt(
2796  IRB.CreateICmpNE(S2, Constant::getNullValue(T)), T);
2797  if (isX86_MMX) {
2798  Type *X86_MMXTy = Type::getX86_MMXTy(*MS.C);
2799  S1_ext = IRB.CreateBitCast(S1_ext, X86_MMXTy);
2800  S2_ext = IRB.CreateBitCast(S2_ext, X86_MMXTy);
2801  }
2802 
2803  Function *ShadowFn = Intrinsic::getDeclaration(
2804  F.getParent(), getSignedPackIntrinsic(I.getIntrinsicID()));
2805 
2806  Value *S =
2807  IRB.CreateCall(ShadowFn, {S1_ext, S2_ext}, "_msprop_vector_pack");
2808  if (isX86_MMX) S = IRB.CreateBitCast(S, getShadowTy(&I));
2809  setShadow(&I, S);
2810  setOriginForNaryOp(I);
2811  }
2812 
2813  // Instrument sum-of-absolute-differencies intrinsic.
2814  void handleVectorSadIntrinsic(IntrinsicInst &I) {
2815  const unsigned SignificantBitsPerResultElement = 16;
2816  bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2817  Type *ResTy = isX86_MMX ? IntegerType::get(*MS.C, 64) : I.getType();
2818  unsigned ZeroBitsPerResultElement =
2819  ResTy->getScalarSizeInBits() - SignificantBitsPerResultElement;
2820 
2821  IRBuilder<> IRB(&I);
2822  Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2823  S = IRB.CreateBitCast(S, ResTy);
2824  S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2825  ResTy);
2826  S = IRB.CreateLShr(S, ZeroBitsPerResultElement);
2827  S = IRB.CreateBitCast(S, getShadowTy(&I));
2828  setShadow(&I, S);
2829  setOriginForNaryOp(I);
2830  }
2831 
2832  // Instrument multiply-add intrinsic.
2833  void handleVectorPmaddIntrinsic(IntrinsicInst &I,
2834  unsigned EltSizeInBits = 0) {
2835  bool isX86_MMX = I.getOperand(0)->getType()->isX86_MMXTy();
2836  Type *ResTy = isX86_MMX ? getMMXVectorTy(EltSizeInBits * 2) : I.getType();
2837  IRBuilder<> IRB(&I);
2838  Value *S = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2839  S = IRB.CreateBitCast(S, ResTy);
2840  S = IRB.CreateSExt(IRB.CreateICmpNE(S, Constant::getNullValue(ResTy)),
2841  ResTy);
2842  S = IRB.CreateBitCast(S, getShadowTy(&I));
2843  setShadow(&I, S);
2844  setOriginForNaryOp(I);
2845  }
2846 
2847  // Instrument compare-packed intrinsic.
2848  // Basically, an or followed by sext(icmp ne 0) to end up with all-zeros or
2849  // all-ones shadow.
2850  void handleVectorComparePackedIntrinsic(IntrinsicInst &I) {
2851  IRBuilder<> IRB(&I);
2852  Type *ResTy = getShadowTy(&I);
2853  Value *S0 = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2854  Value *S = IRB.CreateSExt(
2855  IRB.CreateICmpNE(S0, Constant::getNullValue(ResTy)), ResTy);
2856  setShadow(&I, S);
2857  setOriginForNaryOp(I);
2858  }
2859 
2860  // Instrument compare-scalar intrinsic.
2861  // This handles both cmp* intrinsics which return the result in the first
2862  // element of a vector, and comi* which return the result as i32.
2863  void handleVectorCompareScalarIntrinsic(IntrinsicInst &I) {
2864  IRBuilder<> IRB(&I);
2865  Value *S0 = IRB.CreateOr(getShadow(&I, 0), getShadow(&I, 1));
2866  Value *S = LowerElementShadowExtend(IRB, S0, getShadowTy(&I));
2867  setShadow(&I, S);
2868  setOriginForNaryOp(I);
2869  }
2870 
2871  void handleStmxcsr(IntrinsicInst &I) {
2872  IRBuilder<> IRB(&I);
2873  Value* Addr = I.getArgOperand(0);
2874  Type *Ty = IRB.getInt32Ty();
2875  Value *ShadowPtr =
2876  getShadowOriginPtr(Addr, IRB, Ty, /*Alignment*/ 1, /*isStore*/ true)
2877  .first;
2878 
2879  IRB.CreateStore(getCleanShadow(Ty),
2880  IRB.CreatePointerCast(ShadowPtr, Ty->getPointerTo()));
2881 
2883  insertShadowCheck(Addr, &I);
2884  }
2885 
2886  void handleLdmxcsr(IntrinsicInst &I) {
2887  if (!InsertChecks) return;
2888 
2889  IRBuilder<> IRB(&I);
2890  Value *Addr = I.getArgOperand(0);
2891  Type *Ty = IRB.getInt32Ty();
2892  unsigned Alignment = 1;
2893  Value *ShadowPtr, *OriginPtr;
2894  std::tie(ShadowPtr, OriginPtr) =
2895  getShadowOriginPtr(Addr, IRB, Ty, Alignment, /*isStore*/ false);
2896 
2898  insertShadowCheck(Addr, &I);
2899 
2900  Value *Shadow = IRB.CreateAlignedLoad(Ty, ShadowPtr, Alignment, "_ldmxcsr");
2901  Value *Origin = MS.TrackOrigins ? IRB.CreateLoad(MS.OriginTy, OriginPtr)
2902  : getCleanOrigin();
2903  insertShadowCheck(Shadow, Origin, &I);
2904  }
2905 
2906  void handleMaskedStore(IntrinsicInst &I) {
2907  IRBuilder<> IRB(&I);
2908  Value *V = I.getArgOperand(0);
2909  Value *Addr = I.getArgOperand(1);
2910  unsigned Align = cast<ConstantInt>(I.getArgOperand(2))->getZExtValue();
2911  Value *Mask = I.getArgOperand(3);
2912  Value *Shadow = getShadow(V);
2913 
2914  Value *ShadowPtr;
2915  Value *OriginPtr;
2916  std::tie(ShadowPtr, OriginPtr) = getShadowOriginPtr(
2917  Addr, IRB, Shadow->getType(), Align, /*isStore*/ true);
2918 
2919  if (ClCheckAccessAddress) {
2920  insertShadowCheck(Addr, &I);
2921  // Uninitialized mask is kind of like uninitialized address, but not as
2922  // scary.
2923  insertShadowCheck(Mask, &I);
2924  }
2925 
2926  IRB.CreateMaskedStore(Shadow, ShadowPtr, Align, Mask);
2927 
2928  if (MS.TrackOrigins) {
2929  auto &DL = F.getParent()->getDataLayout();
2930  paintOrigin(IRB, getOrigin(V), OriginPtr,
2931  DL.getTypeStoreSize(Shadow->getType()),
2932  std::max(Align, kMinOriginAlignment));
2933  }
2934  }
2935 
2936  bool handleMaskedLoad(IntrinsicInst &I) {
2937  IRBuilder<> IRB(&I);
2938  Value *Addr = I.getArgOperand(0);
2939  unsigned Align = cast<ConstantInt>(I.getArgOperand(1))->getZExtValue();
2940  Value *Mask = I.getArgOperand(2);
2941  Value *PassThru = I.getArgOperand(3);
2942 
2943  Type *ShadowTy = getShadowTy(&I);
2944  Value *ShadowPtr, *OriginPtr;
2945  if (PropagateShadow) {
2946  std::tie(ShadowPtr, OriginPtr) =
2947  getShadowOriginPtr(Addr, IRB, ShadowTy, Align, /*isStore*/ false);
2948  setShadow(&I, IRB.CreateMaskedLoad(ShadowPtr, Align, Mask,
2949  getShadow(PassThru), "_msmaskedld"));
2950  } else {
2951  setShadow(&I, getCleanShadow(&I));
2952  }
2953 
2954  if (ClCheckAccessAddress) {
2955  insertShadowCheck(Addr, &I);
2956  insertShadowCheck(Mask, &I);
2957  }
2958 
2959  if (MS.TrackOrigins) {
2960  if (PropagateShadow) {
2961  // Choose between PassThru's and the loaded value's origins.
2962  Value *MaskedPassThruShadow = IRB.CreateAnd(
2963  getShadow(PassThru), IRB.CreateSExt(IRB.CreateNeg(Mask), ShadowTy));
2964 
2965  Value *Acc = IRB.CreateExtractElement(
2966  MaskedPassThruShadow, ConstantInt::get(IRB.getInt32Ty(), 0));
2967  for (int i = 1, N = PassThru->getType()->getVectorNumElements(); i < N;
2968  ++i) {
2969  Value *More = IRB.CreateExtractElement(
2970  MaskedPassThruShadow, ConstantInt::get(IRB.getInt32Ty(), i));
2971  Acc = IRB.CreateOr(Acc, More);
2972  }
2973 
2974  Value *Origin = IRB.CreateSelect(
2975  IRB.CreateICmpNE(Acc, Constant::getNullValue(Acc->getType())),
2976  getOrigin(PassThru), IRB.CreateLoad(MS.OriginTy, OriginPtr));
2977 
2978  setOrigin(&I, Origin);
2979  } else {
2980  setOrigin(&I, getCleanOrigin());
2981  }
2982  }
2983  return true;
2984  }
2985 
2986  // Instrument BMI / BMI2 intrinsics.
2987  // All of these intrinsics are Z = I(X, Y)
2988  // where the types of all operands and the result match, and are either i32 or i64.
2989  // The following instrumentation happens to work for all of them:
2990  // Sz = I(Sx, Y) | (sext (Sy != 0))
2991  void handleBmiIntrinsic(IntrinsicInst &I) {
2992  IRBuilder<> IRB(&I);
2993  Type *ShadowTy = getShadowTy(&I);
2994 
2995  // If any bit of the mask operand is poisoned, then the whole thing is.
2996  Value *SMask = getShadow(&I, 1);
2997  SMask = IRB.CreateSExt(IRB.CreateICmpNE(SMask, getCleanShadow(ShadowTy)),
2998  ShadowTy);
2999  // Apply the same intrinsic to the shadow of the first operand.
3000  Value *S = IRB.CreateCall(I.getCalledFunction(),
3001  {getShadow(&I, 0), I.getOperand(1)});
3002  S = IRB.CreateOr(SMask, S);
3003  setShadow(&I, S);
3004  setOriginForNaryOp(I);
3005  }
3006 
3007  void visitIntrinsicInst(IntrinsicInst &I) {
3008  switch (I.getIntrinsicID()) {
3009  case Intrinsic::lifetime_start:
3010  handleLifetimeStart(I);
3011  break;
3012  case Intrinsic::launder_invariant_group:
3013  case Intrinsic::strip_invariant_group:
3014  handleInvariantGroup(I);
3015  break;
3016  case Intrinsic::bswap:
3017  handleBswap(I);
3018  break;
3019  case Intrinsic::masked_store:
3020  handleMaskedStore(I);
3021  break;
3022  case Intrinsic::masked_load:
3023  handleMaskedLoad(I);
3024  break;
3025  case Intrinsic::x86_sse_stmxcsr:
3026  handleStmxcsr(I);
3027  break;
3028  case Intrinsic::x86_sse_ldmxcsr:
3029  handleLdmxcsr(I);
3030  break;
3031  case Intrinsic::x86_avx512_vcvtsd2usi64:
3032  case Intrinsic::x86_avx512_vcvtsd2usi32:
3033  case Intrinsic::x86_avx512_vcvtss2usi64:
3034  case Intrinsic::x86_avx512_vcvtss2usi32:
3035  case Intrinsic::x86_avx512_cvttss2usi64:
3036  case Intrinsic::x86_avx512_cvttss2usi:
3037  case Intrinsic::x86_avx512_cvttsd2usi64:
3038  case Intrinsic::x86_avx512_cvttsd2usi:
3039  case Intrinsic::x86_avx512_cvtusi2ss:
3040  case Intrinsic::x86_avx512_cvtusi642sd:
3041  case Intrinsic::x86_avx512_cvtusi642ss:
3042  case Intrinsic::x86_sse2_cvtsd2si64:
3043  case Intrinsic::x86_sse2_cvtsd2si:
3044  case Intrinsic::x86_sse2_cvtsd2ss:
3045  case Intrinsic::x86_sse2_cvttsd2si64:
3046  case Intrinsic::x86_sse2_cvttsd2si:
3047  case Intrinsic::x86_sse_cvtss2si64:
3048  case Intrinsic::x86_sse_cvtss2si:
3049  case Intrinsic::x86_sse_cvttss2si64:
3050  case Intrinsic::x86_sse_cvttss2si:
3051  handleVectorConvertIntrinsic(I, 1);
3052  break;
3053  case Intrinsic::x86_sse_cvtps2pi:
3054  case Intrinsic::x86_sse_cvttps2pi:
3055  handleVectorConvertIntrinsic(I, 2);
3056  break;
3057 
3058  case Intrinsic::x86_avx512_psll_w_512:
3059  case Intrinsic::x86_avx512_psll_d_512:
3060  case Intrinsic::x86_avx512_psll_q_512:
3061  case Intrinsic::x86_avx512_pslli_w_512:
3062  case Intrinsic::x86_avx512_pslli_d_512:
3063  case Intrinsic::x86_avx512_pslli_q_512:
3064  case Intrinsic::x86_avx512_psrl_w_512:
3065  case Intrinsic::x86_avx512_psrl_d_512:
3066  case Intrinsic::x86_avx512_psrl_q_512:
3067  case Intrinsic::x86_avx512_psra_w_512:
3068  case Intrinsic::x86_avx512_psra_d_512:
3069  case Intrinsic::x86_avx512_psra_q_512:
3070  case Intrinsic::x86_avx512_psrli_w_512:
3071  case Intrinsic::x86_avx512_psrli_d_512:
3072  case Intrinsic::x86_avx512_psrli_q_512:
3073  case Intrinsic::x86_avx512_psrai_w_512:
3074  case Intrinsic::x86_avx512_psrai_d_512:
3075  case Intrinsic::x86_avx512_psrai_q_512:
3076  case Intrinsic::x86_avx512_psra_q_256:
3077  case Intrinsic::x86_avx512_psra_q_128:
3078  case Intrinsic::x86_avx512_psrai_q_256:
3079  case Intrinsic::x86_avx512_psrai_q_128:
3080  case Intrinsic::x86_avx2_psll_w:
3081  case Intrinsic::x86_avx2_psll_d:
3082  case Intrinsic::x86_avx2_psll_q:
3083  case Intrinsic::x86_avx2_pslli_w:
3084  case Intrinsic::x86_avx2_pslli_d:
3085  case Intrinsic::x86_avx2_pslli_q:
3086  case Intrinsic::x86_avx2_psrl_w:
3087  case Intrinsic::x86_avx2_psrl_d:
3088  case Intrinsic::x86_avx2_psrl_q:
3089  case Intrinsic::x86_avx2_psra_w:
3090  case Intrinsic::x86_avx2_psra_d:
3091  case Intrinsic::x86_avx2_psrli_w:
3092  case Intrinsic::x86_avx2_psrli_d:
3093  case Intrinsic::x86_avx2_psrli_q:
3094  case Intrinsic::x86_avx2_psrai_w:
3095  case Intrinsic::x86_avx2_psrai_d:
3096  case Intrinsic::x86_sse2_psll_w:
3097  case Intrinsic::x86_sse2_psll_d:
3098  case Intrinsic::x86_sse2_psll_q:
3099  case Intrinsic::x86_sse2_pslli_w:
3100  case Intrinsic::x86_sse2_pslli_d:
3101  case Intrinsic::x86_sse2_pslli_q:
3102  case Intrinsic::x86_sse2_psrl_w:
3103  case Intrinsic::x86_sse2_psrl_d:
3104  case Intrinsic::x86_sse2_psrl_q:
3105  case Intrinsic::x86_sse2_psra_w:
3106  case Intrinsic::x86_sse2_psra_d:
3107  case Intrinsic::x86_sse2_psrli_w:
3108  case Intrinsic::x86_sse2_psrli_d:
3109  case Intrinsic::x86_sse2_psrli_q:
3110  case Intrinsic::x86_sse2_psrai_w:
3111  case Intrinsic::x86_sse2_psrai_d:
3112  case Intrinsic::x86_mmx_psll_w:
3113  case Intrinsic::x86_mmx_psll_d:
3114  case Intrinsic::x86_mmx_psll_q:
3115  case Intrinsic::x86_mmx_pslli_w:
3116  case Intrinsic::x86_mmx_pslli_d:
3117  case Intrinsic::x86_mmx_pslli_q:
3118  case Intrinsic::x86_mmx_psrl_w:
3119  case Intrinsic::x86_mmx_psrl_d:
3120  case Intrinsic::x86_mmx_psrl_q:
3121  case Intrinsic::x86_mmx_psra_w:
3122  case Intrinsic::x86_mmx_psra_d:
3123  case Intrinsic::x86_mmx_psrli_w:
3124  case Intrinsic::x86_mmx_psrli_d:
3125  case Intrinsic::x86_mmx_psrli_q:
3126  case Intrinsic::x86_mmx_psrai_w:
3127  case Intrinsic::x86_mmx_psrai_d:
3128  handleVectorShiftIntrinsic(I, /* Variable */ false);
3129  break;
3130  case Intrinsic::x86_avx2_psllv_d:
3131  case Intrinsic::x86_avx2_psllv_d_256:
3132  case Intrinsic::x86_avx512_psllv_d_512:
3133  case Intrinsic::x86_avx2_psllv_q:
3134  case Intrinsic::x86_avx2_psllv_q_256:
3135  case Intrinsic::x86_avx512_psllv_q_512:
3136  case Intrinsic::x86_avx2_psrlv_d:
3137  case Intrinsic::x86_avx2_psrlv_d_256:
3138  case Intrinsic::x86_avx512_psrlv_d_512:
3139  case Intrinsic::x86_avx2_psrlv_q:
3140  case Intrinsic::x86_avx2_psrlv_q_256:
3141  case Intrinsic::x86_avx512_psrlv_q_512:
3142  case Intrinsic::x86_avx2_psrav_d:
3143  case Intrinsic::x86_avx2_psrav_d_256:
3144  case Intrinsic::x86_avx512_psrav_d_512:
3145  case Intrinsic::x86_avx512_psrav_q_128:
3146  case Intrinsic::x86_avx512_psrav_q_256:
3147  case Intrinsic::x86_avx512_psrav_q_512:
3148  handleVectorShiftIntrinsic(I, /* Variable */ true);
3149  break;
3150 
3151  case Intrinsic::x86_sse2_packsswb_128:
3152  case Intrinsic::x86_sse2_packssdw_128:
3153  case Intrinsic::x86_sse2_packuswb_128:
3154  case Intrinsic::x86_sse41_packusdw:
3155  case Intrinsic::x86_avx2_packsswb:
3156  case Intrinsic::x86_avx2_packssdw:
3157  case Intrinsic::x86_avx2_packuswb:
3158  case Intrinsic::x86_avx2_packusdw:
3159  handleVectorPackIntrinsic(I);
3160  break;
3161 
3162  case Intrinsic::x86_mmx_packsswb:
3163  case Intrinsic::x86_mmx_packuswb:
3164  handleVectorPackIntrinsic(I, 16);
3165  break;
3166 
3167  case Intrinsic::x86_mmx_packssdw:
3168  handleVectorPackIntrinsic(I, 32);
3169  break;
3170 
3171  case Intrinsic::x86_mmx_psad_bw:
3172  case Intrinsic::x86_sse2_psad_bw:
3173  case Intrinsic::x86_avx2_psad_bw:
3174  handleVectorSadIntrinsic(I);
3175  break;
3176 
3177  case Intrinsic::x86_sse2_pmadd_wd:
3178  case Intrinsic::x86_avx2_pmadd_wd:
3179  case Intrinsic::x86_ssse3_pmadd_ub_sw_128:
3180  case Intrinsic::x86_avx2_pmadd_ub_sw:
3181  handleVectorPmaddIntrinsic(I);
3182  break;
3183 
3184  case Intrinsic::x86_ssse3_pmadd_ub_sw:
3185  handleVectorPmaddIntrinsic(I, 8);
3186  break;
3187 
3188  case Intrinsic::x86_mmx_pmadd_wd:
3189  handleVectorPmaddIntrinsic(I, 16);
3190  break;
3191 
3192  case Intrinsic::x86_sse_cmp_ss:
3193  case Intrinsic::x86_sse2_cmp_sd:
3194  case Intrinsic::x86_sse_comieq_ss:
3195  case Intrinsic::x86_sse_comilt_ss:
3196  case Intrinsic::x86_sse_comile_ss:
3197  case Intrinsic::x86_sse_comigt_ss:
3198  case Intrinsic::x86_sse_comige_ss:
3199  case Intrinsic::x86_sse_comineq_ss:
3200  case Intrinsic::x86_sse_ucomieq_ss:
3201  case Intrinsic::x86_sse_ucomilt_ss:
3202  case Intrinsic::x86_sse_ucomile_ss:
3203  case Intrinsic::x86_sse_ucomigt_ss:
3204  case Intrinsic::x86_sse_ucomige_ss:
3205  case Intrinsic::x86_sse_ucomineq_ss:
3206  case Intrinsic::x86_sse2_comieq_sd:
3207  case Intrinsic::x86_sse2_comilt_sd:
3208  case Intrinsic::x86_sse2_comile_sd:
3209  case Intrinsic::x86_sse2_comigt_sd:
3210  case Intrinsic::x86_sse2_comige_sd:
3211  case Intrinsic::x86_sse2_comineq_sd:
3212  case Intrinsic::x86_sse2_ucomieq_sd:
3213  case Intrinsic::x86_sse2_ucomilt_sd:
3214  case Intrinsic::x86_sse2_ucomile_sd:
3215  case Intrinsic::x86_sse2_ucomigt_sd:
3216  case Intrinsic::x86_sse2_ucomige_sd:
3217  case Intrinsic::x86_sse2_ucomineq_sd:
3218  handleVectorCompareScalarIntrinsic(I);
3219  break;
3220 
3221  case Intrinsic::x86_sse_cmp_ps:
3222  case Intrinsic::x86_sse2_cmp_pd:
3223  // FIXME: For x86_avx_cmp_pd_256 and x86_avx_cmp_ps_256 this function
3224  // generates reasonably looking IR that fails in the backend with "Do not
3225  // know how to split the result of this operator!".
3226  handleVectorComparePackedIntrinsic(I);
3227  break;
3228 
3229  case Intrinsic::x86_bmi_bextr_32:
3230  case Intrinsic::x86_bmi_bextr_64:
3231  case Intrinsic::x86_bmi_bzhi_32:
3232  case Intrinsic::x86_bmi_bzhi_64:
3233  case Intrinsic::x86_bmi_pdep_32:
3234  case Intrinsic::x86_bmi_pdep_64:
3235  case Intrinsic::x86_bmi_pext_32:
3236  case Intrinsic::x86_bmi_pext_64:
3237  handleBmiIntrinsic(I);
3238  break;
3239 
3240  case Intrinsic::is_constant:
3241  // The result of llvm.is.constant() is always defined.
3242  setShadow(&I, getCleanShadow(&I));
3243  setOrigin(&I, getCleanOrigin());
3244  break;
3245 
3246  default:
3247  if (!handleUnknownIntrinsic(I))
3248  visitInstruction(I);
3249  break;
3250  }
3251  }
3252 
3253  void visitCallSite(CallSite CS) {
3254  Instruction &I = *CS.getInstruction();
3255  assert(!I.getMetadata("nosanitize"));
3256  assert((CS.isCall() || CS.isInvoke() || CS.isCallBr()) &&
3257  "Unknown type of CallSite");
3258  if (CS.isCallBr() || (CS.isCall() && cast<CallInst>(&I)->isInlineAsm())) {
3259  // For inline asm (either a call to asm function, or callbr instruction),
3260  // do the usual thing: check argument shadow and mark all outputs as
3261  // clean. Note that any side effects of the inline asm that are not
3262  // immediately visible in its constraints are not handled.
3263  if (ClHandleAsmConservative && MS.CompileKernel)
3264  visitAsmInstruction(I);
3265  else
3266  visitInstruction(I);
3267  return;
3268  }
3269  if (CS.isCall()) {
3270  CallInst *Call = cast<CallInst>(&I);
3271  assert(!isa<IntrinsicInst>(&I) && "intrinsics are handled elsewhere");
3272 
3273  // We are going to insert code that relies on the fact that the callee
3274  // will become a non-readonly function after it is instrumented by us. To
3275  // prevent this code from being optimized out, mark that function
3276  // non-readonly in advance.
3277  if (Function *Func = Call->getCalledFunction()) {
3278  // Clear out readonly/readnone attributes.
3279  AttrBuilder B;
3280  B.addAttribute(Attribute::ReadOnly)
3281  .addAttribute(Attribute::ReadNone);
3283  }
3284 
3286  }
3287  IRBuilder<> IRB(&I);
3288 
3289  unsigned ArgOffset = 0;
3290  LLVM_DEBUG(dbgs() << " CallSite: " << I << "\n");
3291  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
3292  ArgIt != End; ++ArgIt) {
3293  Value *A = *ArgIt;
3294  unsigned i = ArgIt - CS.arg_begin();
3295  if (!A->getType()->isSized()) {
3296  LLVM_DEBUG(dbgs() << "Arg " << i << " is not sized: " << I << "\n");
3297  continue;
3298  }
3299  unsigned Size = 0;
3300  Value *Store = nullptr;
3301  // Compute the Shadow for arg even if it is ByVal, because
3302  // in that case getShadow() will copy the actual arg shadow to
3303  // __msan_param_tls.
3304  Value *ArgShadow = getShadow(A);
3305  Value *ArgShadowBase = getShadowPtrForArgument(A, IRB, ArgOffset);
3306  LLVM_DEBUG(dbgs() << " Arg#" << i << ": " << *A
3307  << " Shadow: " << *ArgShadow << "\n");
3308  bool ArgIsInitialized = false;
3309  const DataLayout &DL = F.getParent()->getDataLayout();
3310  if (CS.paramHasAttr(i, Attribute::ByVal)) {
3311  assert(A->getType()->isPointerTy() &&
3312  "ByVal argument is not a pointer!");
3313  Size = DL.getTypeAllocSize(A->getType()->getPointerElementType());
3314  if (ArgOffset + Size > kParamTLSSize) break;
3315  unsigned ParamAlignment = CS.getParamAlignment(i);
3316  unsigned Alignment = std::min(ParamAlignment, kShadowTLSAlignment);
3317  Value *AShadowPtr =
3318  getShadowOriginPtr(A, IRB, IRB.getInt8Ty(), Alignment,
3319  /*isStore*/ false)
3320  .first;
3321 
3322  Store = IRB.CreateMemCpy(ArgShadowBase, Alignment, AShadowPtr,
3323  Alignment, Size);
3324  // TODO(glider): need to copy origins.
3325  } else {
3326  Size = DL.getTypeAllocSize(A->getType());
3327  if (ArgOffset + Size > kParamTLSSize) break;
3328  Store = IRB.CreateAlignedStore(ArgShadow, ArgShadowBase,
3329  kShadowTLSAlignment);
3330  Constant *Cst = dyn_cast<Constant>(ArgShadow);
3331  if (Cst && Cst->isNullValue()) ArgIsInitialized = true;
3332  }
3333  if (MS.TrackOrigins && !ArgIsInitialized)
3334  IRB.CreateStore(getOrigin(A),
3335  getOriginPtrForArgument(A, IRB, ArgOffset));
3336  (void)Store;
3337  assert(Size != 0 && Store != nullptr);
3338  LLVM_DEBUG(dbgs() << " Param:" << *Store << "\n");
3339  ArgOffset += alignTo(Size, 8);
3340  }
3341  LLVM_DEBUG(dbgs() << " done with call args\n");
3342 
3343  FunctionType *FT = CS.getFunctionType();
3344  if (FT->isVarArg()) {
3345  VAHelper->visitCallSite(CS, IRB);
3346  }
3347 
3348  // Now, get the shadow for the RetVal.
3349  if (!I.getType()->isSized()) return;
3350  // Don't emit the epilogue for musttail call returns.
3351  if (CS.isCall() && cast<CallInst>(&I)->isMustTailCall()) return;
3352  IRBuilder<> IRBBefore(&I);
3353  // Until we have full dynamic coverage, make sure the retval shadow is 0.
3354  Value *Base = getShadowPtrForRetval(&I, IRBBefore);
3355  IRBBefore.CreateAlignedStore(getCleanShadow(&I), Base, kShadowTLSAlignment);
3356  BasicBlock::iterator NextInsn;
3357  if (CS.isCall()) {
3358  NextInsn = ++I.getIterator();
3359  assert(NextInsn != I.getParent()->end());
3360  } else {
3361  BasicBlock *NormalDest = cast<InvokeInst>(&I)->getNormalDest();
3362  if (!NormalDest->getSinglePredecessor()) {
3363  // FIXME: this case is tricky, so we are just conservative here.
3364  // Perhaps we need to split the edge between this BB and NormalDest,
3365  // but a naive attempt to use SplitEdge leads to a crash.
3366  setShadow(&I, getCleanShadow(&I));
3367  setOrigin(&I, getCleanOrigin());
3368  return;
3369  }
3370  // FIXME: NextInsn is likely in a basic block that has not been visited yet.
3371  // Anything inserted there will be instrumented by MSan later!
3372  NextInsn = NormalDest->getFirstInsertionPt();
3373  assert(NextInsn != NormalDest->end() &&
3374  "Could not find insertion point for retval shadow load");
3375  }
3376  IRBuilder<> IRBAfter(&*NextInsn);
3377  Value *RetvalShadow = IRBAfter.CreateAlignedLoad(
3378  getShadowTy(&I), getShadowPtrForRetval(&I, IRBAfter),
3379  kShadowTLSAlignment, "_msret");
3380  setShadow(&I, RetvalShadow);
3381  if (MS.TrackOrigins)
3382  setOrigin(&I, IRBAfter.CreateLoad(MS.OriginTy,
3383  getOriginPtrForRetval(IRBAfter)));
3384  }
3385 
3386  bool isAMustTailRetVal(Value *RetVal) {
3387  if (auto *I = dyn_cast<BitCastInst>(RetVal)) {
3388  RetVal = I->getOperand(0);
3389  }
3390  if (auto *I = dyn_cast<CallInst>(RetVal)) {
3391  return I->isMustTailCall();
3392  }
3393  return false;
3394  }
3395 
3396  void visitReturnInst(ReturnInst &I) {
3397  IRBuilder<> IRB(&I);
3398  Value *RetVal = I.getReturnValue();
3399  if (!RetVal) return;
3400  // Don't emit the epilogue for musttail call returns.
3401  if (isAMustTailRetVal(RetVal)) return;
3402  Value *ShadowPtr = getShadowPtrForRetval(RetVal, IRB);
3403  if (CheckReturnValue) {
3404  insertShadowCheck(RetVal, &I);
3405  Value *Shadow = getCleanShadow(RetVal);
3406  IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
3407  } else {
3408  Value *Shadow = getShadow(RetVal);
3409  IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
3410  if (MS.TrackOrigins)
3411  IRB.CreateStore(getOrigin(RetVal), getOriginPtrForRetval(IRB));
3412  }
3413  }
3414 
3415  void visitPHINode(PHINode &I) {
3416  IRBuilder<> IRB(&I);
3417  if (!PropagateShadow) {
3418  setShadow(&I, getCleanShadow(&I));
3419  setOrigin(&I, getCleanOrigin());
3420  return;
3421  }
3422 
3423  ShadowPHINodes.push_back(&I);
3424  setShadow(&I, IRB.CreatePHI(getShadowTy(&I), I.getNumIncomingValues(),
3425  "_msphi_s"));
3426  if (MS.TrackOrigins)
3427  setOrigin(&I, IRB.CreatePHI(MS.OriginTy, I.getNumIncomingValues(),
3428  "_msphi_o"));
3429  }
3430 
3431  Value *getLocalVarDescription(AllocaInst &I) {
3432  SmallString<2048> StackDescriptionStorage;
3433  raw_svector_ostream StackDescription(StackDescriptionStorage);
3434  // We create a string with a description of the stack allocation and
3435  // pass it into __msan_set_alloca_origin.
3436  // It will be printed by the run-time if stack-originated UMR is found.
3437  // The first 4 bytes of the string are set to '----' and will be replaced
3438  // by __msan_va_arg_overflow_size_tls at the first call.
3439  StackDescription << "----" << I.getName() << "@" << F.getName();
3441  StackDescription.str());
3442  }
3443 
3444  void poisonAllocaUserspace(AllocaInst &I, IRBuilder<> &IRB, Value *Len) {
3445  if (PoisonStack && ClPoisonStackWithCall) {
3446  IRB.CreateCall(MS.MsanPoisonStackFn,
3447  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len});
3448  } else {
3449  Value *ShadowBase, *OriginBase;
3450  std::tie(ShadowBase, OriginBase) =
3451  getShadowOriginPtr(&I, IRB, IRB.getInt8Ty(), 1, /*isStore*/ true);
3452 
3453  Value *PoisonValue = IRB.getInt8(PoisonStack ? ClPoisonStackPattern : 0);
3454  IRB.CreateMemSet(ShadowBase, PoisonValue, Len, I.getAlignment());
3455  }
3456 
3457  if (PoisonStack && MS.TrackOrigins) {
3458  Value *Descr = getLocalVarDescription(I);
3459  IRB.CreateCall(MS.MsanSetAllocaOrigin4Fn,
3460  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len,
3461  IRB.CreatePointerCast(Descr, IRB.getInt8PtrTy()),
3462  IRB.CreatePointerCast(&F, MS.IntptrTy)});
3463  }
3464  }
3465 
3466  void poisonAllocaKmsan(AllocaInst &I, IRBuilder<> &IRB, Value *Len) {
3467  Value *Descr = getLocalVarDescription(I);
3468  if (PoisonStack) {
3469  IRB.CreateCall(MS.MsanPoisonAllocaFn,
3470  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len,
3471  IRB.CreatePointerCast(Descr, IRB.getInt8PtrTy())});
3472  } else {
3473  IRB.CreateCall(MS.MsanUnpoisonAllocaFn,
3474  {IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()), Len});
3475  }
3476  }
3477 
3478  void instrumentAlloca(AllocaInst &I, Instruction *InsPoint = nullptr) {
3479  if (!InsPoint)
3480  InsPoint = &I;
3481  IRBuilder<> IRB(InsPoint->getNextNode());
3482  const DataLayout &DL = F.getParent()->getDataLayout();
3483  uint64_t TypeSize = DL.getTypeAllocSize(I.getAllocatedType());
3484  Value *Len = ConstantInt::get(MS.IntptrTy, TypeSize);
3485  if (I.isArrayAllocation())
3486  Len = IRB.CreateMul(Len, I.getArraySize());
3487 
3488  if (MS.CompileKernel)
3489  poisonAllocaKmsan(I, IRB, Len);
3490  else
3491  poisonAllocaUserspace(I, IRB, Len);
3492  }
3493 
3494  void visitAllocaInst(AllocaInst &I) {
3495  setShadow(&I, getCleanShadow(&I));
3496  setOrigin(&I, getCleanOrigin());
3497  // We'll get to this alloca later unless it's poisoned at the corresponding
3498  // llvm.lifetime.start.
3499  AllocaSet.insert(&I);
3500  }
3501 
3502  void visitSelectInst(SelectInst& I) {
3503  IRBuilder<> IRB(&I);
3504  // a = select b, c, d
3505  Value *B = I.getCondition();
3506  Value *C = I.getTrueValue();
3507  Value *D = I.getFalseValue();
3508  Value *Sb = getShadow(B);
3509  Value *Sc = getShadow(C);
3510  Value *Sd = getShadow(D);
3511 
3512  // Result shadow if condition shadow is 0.
3513  Value *Sa0 = IRB.CreateSelect(B, Sc, Sd);
3514  Value *Sa1;
3515  if (I.getType()->isAggregateType()) {
3516  // To avoid "sign extending" i1 to an arbitrary aggregate type, we just do
3517  // an extra "select". This results in much more compact IR.
3518  // Sa = select Sb, poisoned, (select b, Sc, Sd)
3519  Sa1 = getPoisonedShadow(getShadowTy(I.getType()));
3520  } else {
3521  // Sa = select Sb, [ (c^d) | Sc | Sd ], [ b ? Sc : Sd ]
3522  // If Sb (condition is poisoned), look for bits in c and d that are equal
3523  // and both unpoisoned.
3524  // If !Sb (condition is unpoisoned), simply pick one of Sc and Sd.
3525 
3526  // Cast arguments to shadow-compatible type.
3527  C = CreateAppToShadowCast(IRB, C);
3528  D = CreateAppToShadowCast(IRB, D);
3529 
3530  // Result shadow if condition shadow is 1.
3531  Sa1 = IRB.CreateOr({IRB.CreateXor(C, D), Sc, Sd});
3532  }
3533  Value *Sa = IRB.CreateSelect(Sb, Sa1, Sa0, "_msprop_select");
3534  setShadow(&I, Sa);
3535  if (MS.TrackOrigins) {
3536  // Origins are always i32, so any vector conditions must be flattened.
3537  // FIXME: consider tracking vector origins for app vectors?
3538  if (B->getType()->isVectorTy()) {
3539  Type *FlatTy = getShadowTyNoVec(B->getType());
3540  B = IRB.CreateICmpNE(IRB.CreateBitCast(B, FlatTy),
3541  ConstantInt::getNullValue(FlatTy));
3542  Sb = IRB.CreateICmpNE(IRB.CreateBitCast(Sb, FlatTy),
3543  ConstantInt::getNullValue(FlatTy));
3544  }
3545  // a = select b, c, d
3546  // Oa = Sb ? Ob : (b ? Oc : Od)
3547  setOrigin(
3548  &I, IRB.CreateSelect(Sb, getOrigin(I.getCondition()),
3549  IRB.CreateSelect(B, getOrigin(I.getTrueValue()),
3550  getOrigin(I.getFalseValue()))));
3551  }
3552  }
3553 
3554  void visitLandingPadInst(LandingPadInst &I) {
3555  // Do nothing.
3556  // See https://github.com/google/sanitizers/issues/504
3557  setShadow(&I, getCleanShadow(&I));
3558  setOrigin(&I, getCleanOrigin());
3559  }
3560 
3561  void visitCatchSwitchInst(CatchSwitchInst &I) {
3562  setShadow(&I, getCleanShadow(&I));
3563  setOrigin(&I, getCleanOrigin());
3564  }
3565 
3566  void visitFuncletPadInst(FuncletPadInst &I) {
3567  setShadow(&I, getCleanShadow(&I));
3568  setOrigin(&I, getCleanOrigin());
3569  }
3570 
3571  void visitGetElementPtrInst(GetElementPtrInst &I) {
3572  handleShadowOr(I);
3573  }
3574 
3575  void visitExtractValueInst(ExtractValueInst &I) {
3576  IRBuilder<> IRB(&I);
3577  Value *Agg = I.getAggregateOperand();
3578  LLVM_DEBUG(dbgs() << "ExtractValue: " << I << "\n");
3579  Value *AggShadow = getShadow(Agg);
3580  LLVM_DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
3581  Value *ResShadow = IRB.CreateExtractValue(AggShadow, I.getIndices());
3582  LLVM_DEBUG(dbgs() << " ResShadow: " << *ResShadow << "\n");
3583  setShadow(&I, ResShadow);
3584  setOriginForNaryOp(I);
3585  }
3586 
3587  void visitInsertValueInst(InsertValueInst &I) {
3588  IRBuilder<> IRB(&I);
3589  LLVM_DEBUG(dbgs() << "InsertValue: " << I << "\n");
3590  Value *AggShadow = getShadow(I.getAggregateOperand());
3591  Value *InsShadow = getShadow(I.getInsertedValueOperand());
3592  LLVM_DEBUG(dbgs() << " AggShadow: " << *AggShadow << "\n");
3593  LLVM_DEBUG(dbgs() << " InsShadow: " << *InsShadow << "\n");
3594  Value *Res = IRB.CreateInsertValue(AggShadow, InsShadow, I.getIndices());
3595  LLVM_DEBUG(dbgs() << " Res: " << *Res << "\n");
3596  setShadow(&I, Res);
3597  setOriginForNaryOp(I);
3598  }
3599 
3600  void dumpInst(Instruction &I) {
3601  if (CallInst *CI = dyn_cast<CallInst>(&I)) {
3602  errs() << "ZZZ call " << CI->getCalledFunction()->getName() << "\n";
3603  } else {
3604  errs() << "ZZZ " << I.getOpcodeName() << "\n";
3605  }
3606  errs() << "QQQ " << I << "\n";
3607  }
3608 
3609  void visitResumeInst(ResumeInst &I) {
3610  LLVM_DEBUG(dbgs() << "Resume: " << I << "\n");
3611  // Nothing to do here.
3612  }
3613 
3614  void visitCleanupReturnInst(CleanupReturnInst &CRI) {
3615  LLVM_DEBUG(dbgs() << "CleanupReturn: " << CRI << "\n");
3616  // Nothing to do here.
3617  }
3618 
3619  void visitCatchReturnInst(CatchReturnInst &CRI) {
3620  LLVM_DEBUG(dbgs() << "CatchReturn: " << CRI << "\n");
3621  // Nothing to do here.
3622  }
3623 
3624  void instrumentAsmArgument(Value *Operand, Instruction &I, IRBuilder<> &IRB,
3625  const DataLayout &DL, bool isOutput) {
3626  // For each assembly argument, we check its value for being initialized.
3627  // If the argument is a pointer, we assume it points to a single element
3628  // of the corresponding type (or to a 8-byte word, if the type is unsized).
3629  // Each such pointer is instrumented with a call to the runtime library.
3630  Type *OpType = Operand->getType();
3631  // Check the operand value itself.
3632  insertShadowCheck(Operand, &I);
3633  if (!OpType->isPointerTy() || !isOutput) {
3634  assert(!isOutput);
3635  return;
3636  }
3637  Type *ElType = OpType->getPointerElementType();
3638  if (!ElType->isSized())
3639  return;
3640  int Size = DL.getTypeStoreSize(ElType);
3641  Value *Ptr = IRB.CreatePointerCast(Operand, IRB.getInt8PtrTy());
3642  Value *SizeVal = ConstantInt::get(MS.IntptrTy, Size);
3643  IRB.CreateCall(MS.MsanInstrumentAsmStoreFn, {Ptr, SizeVal});
3644  }
3645 
3646  /// Get the number of output arguments returned by pointers.
3647  int getNumOutputArgs(InlineAsm *IA, CallBase *CB) {
3648  int NumRetOutputs = 0;
3649  int NumOutputs = 0;
3650  Type *RetTy = cast<Value>(CB)->getType();
3651  if (!RetTy->isVoidTy()) {
3652  // Register outputs are returned via the CallInst return value.
3653  auto *ST = dyn_cast<StructType>(RetTy);
3654  if (ST)
3655  NumRetOutputs = ST->getNumElements();
3656  else
3657  NumRetOutputs = 1;
3658  }
3660  for (size_t i = 0, n = Constraints.size(); i < n; i++) {
3661  InlineAsm::ConstraintInfo Info = Constraints[i];
3662  switch (Info.Type) {
3663  case InlineAsm::isOutput:
3664  NumOutputs++;
3665  break;
3666  default:
3667  break;
3668  }
3669  }
3670  return NumOutputs - NumRetOutputs;
3671  }
3672 
3673  void visitAsmInstruction(Instruction &I) {
3674  // Conservative inline assembly handling: check for poisoned shadow of
3675  // asm() arguments, then unpoison the result and all the memory locations
3676  // pointed to by those arguments.
3677  // An inline asm() statement in C++ contains lists of input and output
3678  // arguments used by the assembly code. These are mapped to operands of the
3679  // CallInst as follows:
3680  // - nR register outputs ("=r) are returned by value in a single structure
3681  // (SSA value of the CallInst);
3682  // - nO other outputs ("=m" and others) are returned by pointer as first
3683  // nO operands of the CallInst;
3684  // - nI inputs ("r", "m" and others) are passed to CallInst as the
3685  // remaining nI operands.
3686  // The total number of asm() arguments in the source is nR+nO+nI, and the
3687  // corresponding CallInst has nO+nI+1 operands (the last operand is the
3688  // function to be called).
3689  const DataLayout &DL = F.getParent()->getDataLayout();
3690  CallBase *CB = cast<CallBase>(&I);
3691  IRBuilder<> IRB(&I);
3692  InlineAsm *IA = cast<InlineAsm>(CB->getCalledValue());
3693  int OutputArgs = getNumOutputArgs(IA, CB);
3694  // The last operand of a CallInst is the function itself.
3695  int NumOperands = CB->getNumOperands() - 1;
3696 
3697  // Check input arguments. Doing so before unpoisoning output arguments, so
3698  // that we won't overwrite uninit values before checking them.
3699  for (int i = OutputArgs; i < NumOperands; i++) {
3700  Value *Operand = CB->getOperand(i);
3701  instrumentAsmArgument(Operand, I, IRB, DL, /*isOutput*/ false);
3702  }
3703  // Unpoison output arguments. This must happen before the actual InlineAsm
3704  // call, so that the shadow for memory published in the asm() statement
3705  // remains valid.
3706  for (int i = 0; i < OutputArgs; i++) {
3707  Value *Operand = CB->getOperand(i);
3708  instrumentAsmArgument(Operand, I, IRB, DL, /*isOutput*/ true);
3709  }
3710 
3711  setShadow(&I, getCleanShadow(&I));
3712  setOrigin(&I, getCleanOrigin());
3713  }
3714 
3715  void visitInstruction(Instruction &I) {
3716  // Everything else: stop propagating and check for poisoned shadow.
3718  dumpInst(I);
3719  LLVM_DEBUG(dbgs() << "DEFAULT: " << I << "\n");
3720  for (size_t i = 0, n = I.getNumOperands(); i < n; i++) {
3721  Value *Operand = I.getOperand(i);
3722  if (Operand->getType()->isSized())
3723  insertShadowCheck(Operand, &I);
3724  }
3725  setShadow(&I, getCleanShadow(&I));
3726  setOrigin(&I, getCleanOrigin());
3727  }
3728 };
3729 
3730 /// AMD64-specific implementation of VarArgHelper.
3731 struct VarArgAMD64Helper : public VarArgHelper {
3732  // An unfortunate workaround for asymmetric lowering of va_arg stuff.
3733  // See a comment in visitCallSite for more details.
3734  static const unsigned AMD64GpEndOffset = 48; // AMD64 ABI Draft 0.99.6 p3.5.7
3735  static const unsigned AMD64FpEndOffsetSSE = 176;
3736  // If SSE is disabled, fp_offset in va_list is zero.
3737  static const unsigned AMD64FpEndOffsetNoSSE = AMD64GpEndOffset;
3738 
3739  unsigned AMD64FpEndOffset;
3740  Function &F;
3741  MemorySanitizer &MS;
3742  MemorySanitizerVisitor &MSV;
3743  Value *VAArgTLSCopy = nullptr;
3744  Value *VAArgTLSOriginCopy = nullptr;
3745  Value *VAArgOverflowSize = nullptr;
3746 
3747  SmallVector<CallInst*, 16> VAStartInstrumentationList;
3748 
3749  enum ArgKind { AK_GeneralPurpose, AK_FloatingPoint, AK_Memory };
3750 
3751  VarArgAMD64Helper(Function &F, MemorySanitizer &MS,
3752  MemorySanitizerVisitor &MSV)
3753  : F(F), MS(MS), MSV(MSV) {
3754  AMD64FpEndOffset = AMD64FpEndOffsetSSE;
3755  for (const auto &Attr : F.getAttributes().getFnAttributes()) {
3756  if (Attr.isStringAttribute() &&
3757  (Attr.getKindAsString() == "target-features")) {
3758  if (Attr.getValueAsString().contains("-sse"))
3759  AMD64FpEndOffset = AMD64FpEndOffsetNoSSE;
3760  break;
3761  }
3762  }
3763  }
3764 
3765  ArgKind classifyArgument(Value* arg) {
3766  // A very rough approximation of X86_64 argument classification rules.
3767  Type *T = arg->getType();
3768  if (T->isFPOrFPVectorTy() || T->isX86_MMXTy())
3769  return AK_FloatingPoint;
3770  if (T->isIntegerTy() && T->getPrimitiveSizeInBits() <= 64)
3771  return AK_GeneralPurpose;
3772  if (T->isPointerTy())
3773  return AK_GeneralPurpose;
3774  return AK_Memory;
3775  }
3776 
3777  // For VarArg functions, store the argument shadow in an ABI-specific format
3778  // that corresponds to va_list layout.
3779  // We do this because Clang lowers va_arg in the frontend, and this pass
3780  // only sees the low level code that deals with va_list internals.
3781  // A much easier alternative (provided that Clang emits va_arg instructions)
3782  // would have been to associate each live instance of va_list with a copy of
3783  // MSanParamTLS, and extract shadow on va_arg() call in the argument list
3784  // order.
3785  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
3786  unsigned GpOffset = 0;
3787  unsigned FpOffset = AMD64GpEndOffset;
3788  unsigned OverflowOffset = AMD64FpEndOffset;
3789  const DataLayout &DL = F.getParent()->getDataLayout();
3790  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
3791  ArgIt != End; ++ArgIt) {
3792  Value *A = *ArgIt;
3793  unsigned ArgNo = CS.getArgumentNo(ArgIt);
3794  bool IsFixed = ArgNo < CS.getFunctionType()->getNumParams();
3795  bool IsByVal = CS.paramHasAttr(ArgNo, Attribute::ByVal);
3796  if (IsByVal) {
3797  // ByVal arguments always go to the overflow area.
3798  // Fixed arguments passed through the overflow area will be stepped
3799  // over by va_start, so don't count them towards the offset.
3800  if (IsFixed)
3801  continue;
3802  assert(A->getType()->isPointerTy());
3803  Type *RealTy = A->getType()->getPointerElementType();
3804  uint64_t ArgSize = DL.getTypeAllocSize(RealTy);
3805  Value *ShadowBase = getShadowPtrForVAArgument(
3806  RealTy, IRB, OverflowOffset, alignTo(ArgSize, 8));
3807  Value *OriginBase = nullptr;
3808  if (MS.TrackOrigins)
3809  OriginBase = getOriginPtrForVAArgument(RealTy, IRB, OverflowOffset);
3810  OverflowOffset += alignTo(ArgSize, 8);
3811  if (!ShadowBase)
3812  continue;
3813  Value *ShadowPtr, *OriginPtr;
3814  std::tie(ShadowPtr, OriginPtr) =
3815  MSV.getShadowOriginPtr(A, IRB, IRB.getInt8Ty(), kShadowTLSAlignment,
3816  /*isStore*/ false);
3817 
3818  IRB.CreateMemCpy(ShadowBase, kShadowTLSAlignment, ShadowPtr,
3819  kShadowTLSAlignment, ArgSize);
3820  if (MS.TrackOrigins)
3821  IRB.CreateMemCpy(OriginBase, kShadowTLSAlignment, OriginPtr,
3822  kShadowTLSAlignment, ArgSize);
3823  } else {
3824  ArgKind AK = classifyArgument(A);
3825  if (AK == AK_GeneralPurpose && GpOffset >= AMD64GpEndOffset)
3826  AK = AK_Memory;
3827  if (AK == AK_FloatingPoint && FpOffset >= AMD64FpEndOffset)
3828  AK = AK_Memory;
3829  Value *ShadowBase, *OriginBase = nullptr;
3830  switch (AK) {
3831  case AK_GeneralPurpose:
3832  ShadowBase =
3833  getShadowPtrForVAArgument(A->getType(), IRB, GpOffset, 8);
3834  if (MS.TrackOrigins)
3835  OriginBase =
3836  getOriginPtrForVAArgument(A->getType(), IRB, GpOffset);
3837  GpOffset += 8;
3838  break;
3839  case AK_FloatingPoint:
3840  ShadowBase =
3841  getShadowPtrForVAArgument(A->getType(), IRB, FpOffset, 16);
3842  if (MS.TrackOrigins)
3843  OriginBase =
3844  getOriginPtrForVAArgument(A->getType(), IRB, FpOffset);
3845  FpOffset += 16;
3846  break;
3847  case AK_Memory:
3848  if (IsFixed)
3849  continue;
3850  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
3851  ShadowBase =
3852  getShadowPtrForVAArgument(A->getType(), IRB, OverflowOffset, 8);
3853  if (MS.TrackOrigins)
3854  OriginBase =
3855  getOriginPtrForVAArgument(A->getType(), IRB, OverflowOffset);
3856  OverflowOffset += alignTo(ArgSize, 8);
3857  }
3858  // Take fixed arguments into account for GpOffset and FpOffset,
3859  // but don't actually store shadows for them.
3860  // TODO(glider): don't call get*PtrForVAArgument() for them.
3861  if (IsFixed)
3862  continue;
3863  if (!ShadowBase)
3864  continue;
3865  Value *Shadow = MSV.getShadow(A);
3866  IRB.CreateAlignedStore(Shadow, ShadowBase, kShadowTLSAlignment);
3867  if (MS.TrackOrigins) {
3868  Value *Origin = MSV.getOrigin(A);
3869  unsigned StoreSize = DL.getTypeStoreSize(Shadow->getType());
3870  MSV.paintOrigin(IRB, Origin, OriginBase, StoreSize,
3871  std::max(kShadowTLSAlignment, kMinOriginAlignment));
3872  }
3873  }
3874  }
3875  Constant *OverflowSize =
3876  ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AMD64FpEndOffset);
3877  IRB.CreateStore(OverflowSize, MS.VAArgOverflowSizeTLS);
3878  }
3879 
3880  /// Compute the shadow address for a given va_arg.
3881  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
3882  unsigned ArgOffset, unsigned ArgSize) {
3883  // Make sure we don't overflow __msan_va_arg_tls.
3884  if (ArgOffset + ArgSize > kParamTLSSize)
3885  return nullptr;
3886  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
3887  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
3888  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
3889  "_msarg_va_s");
3890  }
3891 
3892  /// Compute the origin address for a given va_arg.
3893  Value *getOriginPtrForVAArgument(Type *Ty, IRBuilder<> &IRB, int ArgOffset) {
3894  Value *Base = IRB.CreatePointerCast(MS.VAArgOriginTLS, MS.IntptrTy);
3895  // getOriginPtrForVAArgument() is always called after
3896  // getShadowPtrForVAArgument(), so __msan_va_arg_origin_tls can never
3897  // overflow.
3898  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
3899  return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
3900  "_msarg_va_o");
3901  }
3902 
3903  void unpoisonVAListTagForInst(IntrinsicInst &I) {
3904  IRBuilder<> IRB(&I);
3905  Value *VAListTag = I.getArgOperand(0);
3906  Value *ShadowPtr, *OriginPtr;
3907  unsigned Alignment = 8;
3908  std::tie(ShadowPtr, OriginPtr) =
3909  MSV.getShadowOriginPtr(VAListTag, IRB, IRB.getInt8Ty(), Alignment,
3910  /*isStore*/ true);
3911 
3912  // Unpoison the whole __va_list_tag.
3913  // FIXME: magic ABI constants.
3914  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
3915  /* size */ 24, Alignment, false);
3916  // We shouldn't need to zero out the origins, as they're only checked for
3917  // nonzero shadow.
3918  }
3919 
3920  void visitVAStartInst(VAStartInst &I) override {
3922  return;
3923  VAStartInstrumentationList.push_back(&I);
3924  unpoisonVAListTagForInst(I);
3925  }
3926 
3927  void visitVACopyInst(VACopyInst &I) override {
3928  if (F.getCallingConv() == CallingConv::Win64) return;
3929  unpoisonVAListTagForInst(I);
3930  }
3931 
3932  void finalizeInstrumentation() override {
3933  assert(!VAArgOverflowSize && !VAArgTLSCopy &&
3934  "finalizeInstrumentation called twice");
3935  if (!VAStartInstrumentationList.empty()) {
3936  // If there is a va_start in this function, make a backup copy of
3937  // va_arg_tls somewhere in the function entry block.
3938  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
3939  VAArgOverflowSize =
3940  IRB.CreateLoad(IRB.getInt64Ty(), MS.VAArgOverflowSizeTLS);
3941  Value *CopySize =
3942  IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, AMD64FpEndOffset),
3943  VAArgOverflowSize);
3944  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
3945  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
3946  if (MS.TrackOrigins) {
3947  VAArgTLSOriginCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
3948  IRB.CreateMemCpy(VAArgTLSOriginCopy, 8, MS.VAArgOriginTLS, 8, CopySize);
3949  }
3950  }
3951 
3952  // Instrument va_start.
3953  // Copy va_list shadow from the backup copy of the TLS contents.
3954  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
3955  CallInst *OrigInst = VAStartInstrumentationList[i];
3956  IRBuilder<> IRB(OrigInst->getNextNode());
3957  Value *VAListTag = OrigInst->getArgOperand(0);
3958 
3959  Type *RegSaveAreaPtrTy = Type::getInt64PtrTy(*MS.C);
3960  Value *RegSaveAreaPtrPtr = IRB.CreateIntToPtr(
3961  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
3962  ConstantInt::get(MS.IntptrTy, 16)),
3963  PointerType::get(RegSaveAreaPtrTy, 0));
3964  Value *RegSaveAreaPtr =
3965  IRB.CreateLoad(RegSaveAreaPtrTy, RegSaveAreaPtrPtr);
3966  Value *RegSaveAreaShadowPtr, *RegSaveAreaOriginPtr;
3967  unsigned Alignment = 16;
3968  std::tie(RegSaveAreaShadowPtr, RegSaveAreaOriginPtr) =
3969  MSV.getShadowOriginPtr(RegSaveAreaPtr, IRB, IRB.getInt8Ty(),
3970  Alignment, /*isStore*/ true);
3971  IRB.CreateMemCpy(RegSaveAreaShadowPtr, Alignment, VAArgTLSCopy, Alignment,
3972  AMD64FpEndOffset);
3973  if (MS.TrackOrigins)
3974  IRB.CreateMemCpy(RegSaveAreaOriginPtr, Alignment, VAArgTLSOriginCopy,
3975  Alignment, AMD64FpEndOffset);
3976  Type *OverflowArgAreaPtrTy = Type::getInt64PtrTy(*MS.C);
3977  Value *OverflowArgAreaPtrPtr = IRB.CreateIntToPtr(
3978  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
3979  ConstantInt::get(MS.IntptrTy, 8)),
3980  PointerType::get(OverflowArgAreaPtrTy, 0));
3981  Value *OverflowArgAreaPtr =
3982  IRB.CreateLoad(OverflowArgAreaPtrTy, OverflowArgAreaPtrPtr);
3983  Value *OverflowArgAreaShadowPtr, *OverflowArgAreaOriginPtr;
3984  std::tie(OverflowArgAreaShadowPtr, OverflowArgAreaOriginPtr) =
3985  MSV.getShadowOriginPtr(OverflowArgAreaPtr, IRB, IRB.getInt8Ty(),
3986  Alignment, /*isStore*/ true);
3987  Value *SrcPtr = IRB.CreateConstGEP1_32(IRB.getInt8Ty(), VAArgTLSCopy,
3988  AMD64FpEndOffset);
3989  IRB.CreateMemCpy(OverflowArgAreaShadowPtr, Alignment, SrcPtr, Alignment,
3990  VAArgOverflowSize);
3991  if (MS.TrackOrigins) {
3992  SrcPtr = IRB.CreateConstGEP1_32(IRB.getInt8Ty(), VAArgTLSOriginCopy,
3993  AMD64FpEndOffset);
3994  IRB.CreateMemCpy(OverflowArgAreaOriginPtr, Alignment, SrcPtr, Alignment,
3995  VAArgOverflowSize);
3996  }
3997  }
3998  }
3999 };
4000 
4001 /// MIPS64-specific implementation of VarArgHelper.
4002 struct VarArgMIPS64Helper : public VarArgHelper {
4003  Function &F;
4004  MemorySanitizer &MS;
4005  MemorySanitizerVisitor &MSV;
4006  Value *VAArgTLSCopy = nullptr;
4007  Value *VAArgSize = nullptr;
4008 
4009  SmallVector<CallInst*, 16> VAStartInstrumentationList;
4010 
4011  VarArgMIPS64Helper(Function &F, MemorySanitizer &MS,
4012  MemorySanitizerVisitor &MSV) : F(F), MS(MS), MSV(MSV) {}
4013 
4014  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
4015  unsigned VAArgOffset = 0;
4016  const DataLayout &DL = F.getParent()->getDataLayout();
4017  for (CallSite::arg_iterator ArgIt = CS.arg_begin() +
4018  CS.getFunctionType()->getNumParams(), End = CS.arg_end();
4019  ArgIt != End; ++ArgIt) {
4020  Triple TargetTriple(F.getParent()->getTargetTriple());
4021  Value *A = *ArgIt;
4022  Value *Base;
4023  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
4024  if (TargetTriple.getArch() == Triple::mips64) {
4025  // Adjusting the shadow for argument with size < 8 to match the placement
4026  // of bits in big endian system
4027  if (ArgSize < 8)
4028  VAArgOffset += (8 - ArgSize);
4029  }
4030  Base = getShadowPtrForVAArgument(A->getType(), IRB, VAArgOffset, ArgSize);
4031  VAArgOffset += ArgSize;
4032  VAArgOffset = alignTo(VAArgOffset, 8);
4033  if (!Base)
4034  continue;
4035  IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
4036  }
4037 
4038  Constant *TotalVAArgSize = ConstantInt::get(IRB.getInt64Ty(), VAArgOffset);
4039  // Here using VAArgOverflowSizeTLS as VAArgSizeTLS to avoid creation of
4040  // a new class member i.e. it is the total size of all VarArgs.
4041  IRB.CreateStore(TotalVAArgSize, MS.VAArgOverflowSizeTLS);
4042  }
4043 
4044  /// Compute the shadow address for a given va_arg.
4045  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
4046  unsigned ArgOffset, unsigned ArgSize) {
4047  // Make sure we don't overflow __msan_va_arg_tls.
4048  if (ArgOffset + ArgSize > kParamTLSSize)
4049  return nullptr;
4050  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
4051  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
4052  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
4053  "_msarg");
4054  }
4055 
4056  void visitVAStartInst(VAStartInst &I) override {
4057  IRBuilder<> IRB(&I);
4058  VAStartInstrumentationList.push_back(&I);
4059  Value *VAListTag = I.getArgOperand(0);
4060  Value *ShadowPtr, *OriginPtr;
4061  unsigned Alignment = 8;
4062  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4063  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4064  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4065  /* size */ 8, Alignment, false);
4066  }
4067 
4068  void visitVACopyInst(VACopyInst &I) override {
4069  IRBuilder<> IRB(&I);
4070  VAStartInstrumentationList.push_back(&I);
4071  Value *VAListTag = I.getArgOperand(0);
4072  Value *ShadowPtr, *OriginPtr;
4073  unsigned Alignment = 8;
4074  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4075  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4076  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4077  /* size */ 8, Alignment, false);
4078  }
4079 
4080  void finalizeInstrumentation() override {
4081  assert(!VAArgSize && !VAArgTLSCopy &&
4082  "finalizeInstrumentation called twice");
4083  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
4084  VAArgSize = IRB.CreateLoad(IRB.getInt64Ty(), MS.VAArgOverflowSizeTLS);
4085  Value *CopySize = IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, 0),
4086  VAArgSize);
4087 
4088  if (!VAStartInstrumentationList.empty()) {
4089  // If there is a va_start in this function, make a backup copy of
4090  // va_arg_tls somewhere in the function entry block.
4091  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
4092  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
4093  }
4094 
4095  // Instrument va_start.
4096  // Copy va_list shadow from the backup copy of the TLS contents.
4097  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
4098  CallInst *OrigInst = VAStartInstrumentationList[i];
4099  IRBuilder<> IRB(OrigInst->getNextNode());
4100  Value *VAListTag = OrigInst->getArgOperand(0);
4101  Type *RegSaveAreaPtrTy = Type::getInt64PtrTy(*MS.C);
4102  Value *RegSaveAreaPtrPtr =
4103  IRB.CreateIntToPtr(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
4104  PointerType::get(RegSaveAreaPtrTy, 0));
4105  Value *RegSaveAreaPtr =
4106  IRB.CreateLoad(RegSaveAreaPtrTy, RegSaveAreaPtrPtr);
4107  Value *RegSaveAreaShadowPtr, *RegSaveAreaOriginPtr;
4108  unsigned Alignment = 8;
4109  std::tie(RegSaveAreaShadowPtr, RegSaveAreaOriginPtr) =
4110  MSV.getShadowOriginPtr(RegSaveAreaPtr, IRB, IRB.getInt8Ty(),
4111  Alignment, /*isStore*/ true);
4112  IRB.CreateMemCpy(RegSaveAreaShadowPtr, Alignment, VAArgTLSCopy, Alignment,
4113  CopySize);
4114  }
4115  }
4116 };
4117 
4118 /// AArch64-specific implementation of VarArgHelper.
4119 struct VarArgAArch64Helper : public VarArgHelper {
4120  static const unsigned kAArch64GrArgSize = 64;
4121  static const unsigned kAArch64VrArgSize = 128;
4122 
4123  static const unsigned AArch64GrBegOffset = 0;
4124  static const unsigned AArch64GrEndOffset = kAArch64GrArgSize;
4125  // Make VR space aligned to 16 bytes.
4126  static const unsigned AArch64VrBegOffset = AArch64GrEndOffset;
4127  static const unsigned AArch64VrEndOffset = AArch64VrBegOffset
4128  + kAArch64VrArgSize;
4129  static const unsigned AArch64VAEndOffset = AArch64VrEndOffset;
4130 
4131  Function &F;
4132  MemorySanitizer &MS;
4133  MemorySanitizerVisitor &MSV;
4134  Value *VAArgTLSCopy = nullptr;
4135  Value *VAArgOverflowSize = nullptr;
4136 
4137  SmallVector<CallInst*, 16> VAStartInstrumentationList;
4138 
4139  enum ArgKind { AK_GeneralPurpose, AK_FloatingPoint, AK_Memory };
4140 
4141  VarArgAArch64Helper(Function &F, MemorySanitizer &MS,
4142  MemorySanitizerVisitor &MSV) : F(F), MS(MS), MSV(MSV) {}
4143 
4144  ArgKind classifyArgument(Value* arg) {
4145  Type *T = arg->getType();
4146  if (T->isFPOrFPVectorTy())
4147  return AK_FloatingPoint;
4148  if ((T->isIntegerTy() && T->getPrimitiveSizeInBits() <= 64)
4149  || (T->isPointerTy()))
4150  return AK_GeneralPurpose;
4151  return AK_Memory;
4152  }
4153 
4154  // The instrumentation stores the argument shadow in a non ABI-specific
4155  // format because it does not know which argument is named (since Clang,
4156  // like x86_64 case, lowers the va_args in the frontend and this pass only
4157  // sees the low level code that deals with va_list internals).
4158  // The first seven GR registers are saved in the first 56 bytes of the
4159  // va_arg tls arra, followers by the first 8 FP/SIMD registers, and then
4160  // the remaining arguments.
4161  // Using constant offset within the va_arg TLS array allows fast copy
4162  // in the finalize instrumentation.
4163  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
4164  unsigned GrOffset = AArch64GrBegOffset;
4165  unsigned VrOffset = AArch64VrBegOffset;
4166  unsigned OverflowOffset = AArch64VAEndOffset;
4167 
4168  const DataLayout &DL = F.getParent()->getDataLayout();
4169  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
4170  ArgIt != End; ++ArgIt) {
4171  Value *A = *ArgIt;
4172  unsigned ArgNo = CS.getArgumentNo(ArgIt);
4173  bool IsFixed = ArgNo < CS.getFunctionType()->getNumParams();
4174  ArgKind AK = classifyArgument(A);
4175  if (AK == AK_GeneralPurpose && GrOffset >= AArch64GrEndOffset)
4176  AK = AK_Memory;
4177  if (AK == AK_FloatingPoint && VrOffset >= AArch64VrEndOffset)
4178  AK = AK_Memory;
4179  Value *Base;
4180  switch (AK) {
4181  case AK_GeneralPurpose:
4182  Base = getShadowPtrForVAArgument(A->getType(), IRB, GrOffset, 8);
4183  GrOffset += 8;
4184  break;
4185  case AK_FloatingPoint:
4186  Base = getShadowPtrForVAArgument(A->getType(), IRB, VrOffset, 8);
4187  VrOffset += 16;
4188  break;
4189  case AK_Memory:
4190  // Don't count fixed arguments in the overflow area - va_start will
4191  // skip right over them.
4192  if (IsFixed)
4193  continue;
4194  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
4195  Base = getShadowPtrForVAArgument(A->getType(), IRB, OverflowOffset,
4196  alignTo(ArgSize, 8));
4197  OverflowOffset += alignTo(ArgSize, 8);
4198  break;
4199  }
4200  // Count Gp/Vr fixed arguments to their respective offsets, but don't
4201  // bother to actually store a shadow.
4202  if (IsFixed)
4203  continue;
4204  if (!Base)
4205  continue;
4206  IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
4207  }
4208  Constant *OverflowSize =
4209  ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AArch64VAEndOffset);
4210  IRB.CreateStore(OverflowSize, MS.VAArgOverflowSizeTLS);
4211  }
4212 
4213  /// Compute the shadow address for a given va_arg.
4214  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
4215  unsigned ArgOffset, unsigned ArgSize) {
4216  // Make sure we don't overflow __msan_va_arg_tls.
4217  if (ArgOffset + ArgSize > kParamTLSSize)
4218  return nullptr;
4219  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
4220  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
4221  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
4222  "_msarg");
4223  }
4224 
4225  void visitVAStartInst(VAStartInst &I) override {
4226  IRBuilder<> IRB(&I);
4227  VAStartInstrumentationList.push_back(&I);
4228  Value *VAListTag = I.getArgOperand(0);
4229  Value *ShadowPtr, *OriginPtr;
4230  unsigned Alignment = 8;
4231  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4232  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4233  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4234  /* size */ 32, Alignment, false);
4235  }
4236 
4237  void visitVACopyInst(VACopyInst &I) override {
4238  IRBuilder<> IRB(&I);
4239  VAStartInstrumentationList.push_back(&I);
4240  Value *VAListTag = I.getArgOperand(0);
4241  Value *ShadowPtr, *OriginPtr;
4242  unsigned Alignment = 8;
4243  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4244  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4245  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4246  /* size */ 32, Alignment, false);
4247  }
4248 
4249  // Retrieve a va_list field of 'void*' size.
4250  Value* getVAField64(IRBuilder<> &IRB, Value *VAListTag, int offset) {
4251  Value *SaveAreaPtrPtr =
4252  IRB.CreateIntToPtr(
4253  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
4254  ConstantInt::get(MS.IntptrTy, offset)),
4255  Type::getInt64PtrTy(*MS.C));
4256  return IRB.CreateLoad(Type::getInt64Ty(*MS.C), SaveAreaPtrPtr);
4257  }
4258 
4259  // Retrieve a va_list field of 'int' size.
4260  Value* getVAField32(IRBuilder<> &IRB, Value *VAListTag, int offset) {
4261  Value *SaveAreaPtr =
4262  IRB.CreateIntToPtr(
4263  IRB.CreateAdd(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
4264  ConstantInt::get(MS.IntptrTy, offset)),
4265  Type::getInt32PtrTy(*MS.C));
4266  Value *SaveArea32 = IRB.CreateLoad(IRB.getInt32Ty(), SaveAreaPtr);
4267  return IRB.CreateSExt(SaveArea32, MS.IntptrTy);
4268  }
4269 
4270  void finalizeInstrumentation() override {
4271  assert(!VAArgOverflowSize && !VAArgTLSCopy &&
4272  "finalizeInstrumentation called twice");
4273  if (!VAStartInstrumentationList.empty()) {
4274  // If there is a va_start in this function, make a backup copy of
4275  // va_arg_tls somewhere in the function entry block.
4276  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
4277  VAArgOverflowSize =
4278  IRB.CreateLoad(IRB.getInt64Ty(), MS.VAArgOverflowSizeTLS);
4279  Value *CopySize =
4280  IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, AArch64VAEndOffset),
4281  VAArgOverflowSize);
4282  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
4283  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
4284  }
4285 
4286  Value *GrArgSize = ConstantInt::get(MS.IntptrTy, kAArch64GrArgSize);
4287  Value *VrArgSize = ConstantInt::get(MS.IntptrTy, kAArch64VrArgSize);
4288 
4289  // Instrument va_start, copy va_list shadow from the backup copy of
4290  // the TLS contents.
4291  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
4292  CallInst *OrigInst = VAStartInstrumentationList[i];
4293  IRBuilder<> IRB(OrigInst->getNextNode());
4294 
4295  Value *VAListTag = OrigInst->getArgOperand(0);
4296 
4297  // The variadic ABI for AArch64 creates two areas to save the incoming
4298  // argument registers (one for 64-bit general register xn-x7 and another
4299  // for 128-bit FP/SIMD vn-v7).
4300  // We need then to propagate the shadow arguments on both regions
4301  // 'va::__gr_top + va::__gr_offs' and 'va::__vr_top + va::__vr_offs'.
4302  // The remaning arguments are saved on shadow for 'va::stack'.
4303  // One caveat is it requires only to propagate the non-named arguments,
4304  // however on the call site instrumentation 'all' the arguments are
4305  // saved. So to copy the shadow values from the va_arg TLS array
4306  // we need to adjust the offset for both GR and VR fields based on
4307  // the __{gr,vr}_offs value (since they are stores based on incoming
4308  // named arguments).
4309 
4310  // Read the stack pointer from the va_list.
4311  Value *StackSaveAreaPtr = getVAField64(IRB, VAListTag, 0);
4312 
4313  // Read both the __gr_top and __gr_off and add them up.
4314  Value *GrTopSaveAreaPtr = getVAField64(IRB, VAListTag, 8);
4315  Value *GrOffSaveArea = getVAField32(IRB, VAListTag, 24);
4316 
4317  Value *GrRegSaveAreaPtr = IRB.CreateAdd(GrTopSaveAreaPtr, GrOffSaveArea);
4318 
4319  // Read both the __vr_top and __vr_off and add them up.
4320  Value *VrTopSaveAreaPtr = getVAField64(IRB, VAListTag, 16);
4321  Value *VrOffSaveArea = getVAField32(IRB, VAListTag, 28);
4322 
4323  Value *VrRegSaveAreaPtr = IRB.CreateAdd(VrTopSaveAreaPtr, VrOffSaveArea);
4324 
4325  // It does not know how many named arguments is being used and, on the
4326  // callsite all the arguments were saved. Since __gr_off is defined as
4327  // '0 - ((8 - named_gr) * 8)', the idea is to just propagate the variadic
4328  // argument by ignoring the bytes of shadow from named arguments.
4329  Value *GrRegSaveAreaShadowPtrOff =
4330  IRB.CreateAdd(GrArgSize, GrOffSaveArea);
4331 
4332  Value *GrRegSaveAreaShadowPtr =
4333  MSV.getShadowOriginPtr(GrRegSaveAreaPtr, IRB, IRB.getInt8Ty(),
4334  /*Alignment*/ 8, /*isStore*/ true)
4335  .first;
4336 
4337  Value *GrSrcPtr = IRB.CreateInBoundsGEP(IRB.getInt8Ty(), VAArgTLSCopy,
4338  GrRegSaveAreaShadowPtrOff);
4339  Value *GrCopySize = IRB.CreateSub(GrArgSize, GrRegSaveAreaShadowPtrOff);
4340 
4341  IRB.CreateMemCpy(GrRegSaveAreaShadowPtr, 8, GrSrcPtr, 8, GrCopySize);
4342 
4343  // Again, but for FP/SIMD values.
4344  Value *VrRegSaveAreaShadowPtrOff =
4345  IRB.CreateAdd(VrArgSize, VrOffSaveArea);
4346 
4347  Value *VrRegSaveAreaShadowPtr =
4348  MSV.getShadowOriginPtr(VrRegSaveAreaPtr, IRB, IRB.getInt8Ty(),
4349  /*Alignment*/ 8, /*isStore*/ true)
4350  .first;
4351 
4352  Value *VrSrcPtr = IRB.CreateInBoundsGEP(
4353  IRB.getInt8Ty(),
4354  IRB.CreateInBoundsGEP(IRB.getInt8Ty(), VAArgTLSCopy,
4355  IRB.getInt32(AArch64VrBegOffset)),
4356  VrRegSaveAreaShadowPtrOff);
4357  Value *VrCopySize = IRB.CreateSub(VrArgSize, VrRegSaveAreaShadowPtrOff);
4358 
4359  IRB.CreateMemCpy(VrRegSaveAreaShadowPtr, 8, VrSrcPtr, 8, VrCopySize);
4360 
4361  // And finally for remaining arguments.
4362  Value *StackSaveAreaShadowPtr =
4363  MSV.getShadowOriginPtr(StackSaveAreaPtr, IRB, IRB.getInt8Ty(),
4364  /*Alignment*/ 16, /*isStore*/ true)
4365  .first;
4366 
4367  Value *StackSrcPtr =
4368  IRB.CreateInBoundsGEP(IRB.getInt8Ty(), VAArgTLSCopy,
4369  IRB.getInt32(AArch64VAEndOffset));
4370 
4371  IRB.CreateMemCpy(StackSaveAreaShadowPtr, 16, StackSrcPtr, 16,
4372  VAArgOverflowSize);
4373  }
4374  }
4375 };
4376 
4377 /// PowerPC64-specific implementation of VarArgHelper.
4378 struct VarArgPowerPC64Helper : public VarArgHelper {
4379  Function &F;
4380  MemorySanitizer &MS;
4381  MemorySanitizerVisitor &MSV;
4382  Value *VAArgTLSCopy = nullptr;
4383  Value *VAArgSize = nullptr;
4384 
4385  SmallVector<CallInst*, 16> VAStartInstrumentationList;
4386 
4387  VarArgPowerPC64Helper(Function &F, MemorySanitizer &MS,
4388  MemorySanitizerVisitor &MSV) : F(F), MS(MS), MSV(MSV) {}
4389 
4390  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {
4391  // For PowerPC, we need to deal with alignment of stack arguments -
4392  // they are mostly aligned to 8 bytes, but vectors and i128 arrays
4393  // are aligned to 16 bytes, byvals can be aligned to 8 or 16 bytes,
4394  // and QPX vectors are aligned to 32 bytes. For that reason, we
4395  // compute current offset from stack pointer (which is always properly
4396  // aligned), and offset for the first vararg, then subtract them.
4397  unsigned VAArgBase;
4398  Triple TargetTriple(F.getParent()->getTargetTriple());
4399  // Parameter save area starts at 48 bytes from frame pointer for ABIv1,
4400  // and 32 bytes for ABIv2. This is usually determined by target
4401  // endianness, but in theory could be overriden by function attribute.
4402  // For simplicity, we ignore it here (it'd only matter for QPX vectors).
4403  if (TargetTriple.getArch() == Triple::ppc64)
4404  VAArgBase = 48;
4405  else
4406  VAArgBase = 32;
4407  unsigned VAArgOffset = VAArgBase;
4408  const DataLayout &DL = F.getParent()->getDataLayout();
4409  for (CallSite::arg_iterator ArgIt = CS.arg_begin(), End = CS.arg_end();
4410  ArgIt != End; ++ArgIt) {
4411  Value *A = *ArgIt;
4412  unsigned ArgNo = CS.getArgumentNo(ArgIt);
4413  bool IsFixed = ArgNo < CS.getFunctionType()->getNumParams();
4414  bool IsByVal = CS.paramHasAttr(ArgNo, Attribute::ByVal);
4415  if (IsByVal) {
4416  assert(A->getType()->isPointerTy());
4417  Type *RealTy = A->getType()->getPointerElementType();
4418  uint64_t ArgSize = DL.getTypeAllocSize(RealTy);
4419  uint64_t ArgAlign = CS.getParamAlignment(ArgNo);
4420  if (ArgAlign < 8)
4421  ArgAlign = 8;
4422  VAArgOffset = alignTo(VAArgOffset, ArgAlign);
4423  if (!IsFixed) {
4424  Value *Base = getShadowPtrForVAArgument(
4425  RealTy, IRB, VAArgOffset - VAArgBase, ArgSize);
4426  if (Base) {
4427  Value *AShadowPtr, *AOriginPtr;
4428  std::tie(AShadowPtr, AOriginPtr) =
4429  MSV.getShadowOriginPtr(A, IRB, IRB.getInt8Ty(),
4430  kShadowTLSAlignment, /*isStore*/ false);
4431 
4432  IRB.CreateMemCpy(Base, kShadowTLSAlignment, AShadowPtr,
4433  kShadowTLSAlignment, ArgSize);
4434  }
4435  }
4436  VAArgOffset += alignTo(ArgSize, 8);
4437  } else {
4438  Value *Base;
4439  uint64_t ArgSize = DL.getTypeAllocSize(A->getType());
4440  uint64_t ArgAlign = 8;
4441  if (A->getType()->isArrayTy()) {
4442  // Arrays are aligned to element size, except for long double
4443  // arrays, which are aligned to 8 bytes.
4444  Type *ElementTy = A->getType()->getArrayElementType();
4445  if (!ElementTy->isPPC_FP128Ty())
4446  ArgAlign = DL.getTypeAllocSize(ElementTy);
4447  } else if (A->getType()->isVectorTy()) {
4448  // Vectors are naturally aligned.
4449  ArgAlign = DL.getTypeAllocSize(A->getType());
4450  }
4451  if (ArgAlign < 8)
4452  ArgAlign = 8;
4453  VAArgOffset = alignTo(VAArgOffset, ArgAlign);
4454  if (DL.isBigEndian()) {
4455  // Adjusting the shadow for argument with size < 8 to match the placement
4456  // of bits in big endian system
4457  if (ArgSize < 8)
4458  VAArgOffset += (8 - ArgSize);
4459  }
4460  if (!IsFixed) {
4461  Base = getShadowPtrForVAArgument(A->getType(), IRB,
4462  VAArgOffset - VAArgBase, ArgSize);
4463  if (Base)
4464  IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
4465  }
4466  VAArgOffset += ArgSize;
4467  VAArgOffset = alignTo(VAArgOffset, 8);
4468  }
4469  if (IsFixed)
4470  VAArgBase = VAArgOffset;
4471  }
4472 
4473  Constant *TotalVAArgSize = ConstantInt::get(IRB.getInt64Ty(),
4474  VAArgOffset - VAArgBase);
4475  // Here using VAArgOverflowSizeTLS as VAArgSizeTLS to avoid creation of
4476  // a new class member i.e. it is the total size of all VarArgs.
4477  IRB.CreateStore(TotalVAArgSize, MS.VAArgOverflowSizeTLS);
4478  }
4479 
4480  /// Compute the shadow address for a given va_arg.
4481  Value *getShadowPtrForVAArgument(Type *Ty, IRBuilder<> &IRB,
4482  unsigned ArgOffset, unsigned ArgSize) {
4483  // Make sure we don't overflow __msan_va_arg_tls.
4484  if (ArgOffset + ArgSize > kParamTLSSize)
4485  return nullptr;
4486  Value *Base = IRB.CreatePointerCast(MS.VAArgTLS, MS.IntptrTy);
4487  Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
4488  return IRB.CreateIntToPtr(Base, PointerType::get(MSV.getShadowTy(Ty), 0),
4489  "_msarg");
4490  }
4491 
4492  void visitVAStartInst(VAStartInst &I) override {
4493  IRBuilder<> IRB(&I);
4494  VAStartInstrumentationList.push_back(&I);
4495  Value *VAListTag = I.getArgOperand(0);
4496  Value *ShadowPtr, *OriginPtr;
4497  unsigned Alignment = 8;
4498  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4499  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4500  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4501  /* size */ 8, Alignment, false);
4502  }
4503 
4504  void visitVACopyInst(VACopyInst &I) override {
4505  IRBuilder<> IRB(&I);
4506  Value *VAListTag = I.getArgOperand(0);
4507  Value *ShadowPtr, *OriginPtr;
4508  unsigned Alignment = 8;
4509  std::tie(ShadowPtr, OriginPtr) = MSV.getShadowOriginPtr(
4510  VAListTag, IRB, IRB.getInt8Ty(), Alignment, /*isStore*/ true);
4511  // Unpoison the whole __va_list_tag.
4512  // FIXME: magic ABI constants.
4513  IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
4514  /* size */ 8, Alignment, false);
4515  }
4516 
4517  void finalizeInstrumentation() override {
4518  assert(!VAArgSize && !VAArgTLSCopy &&
4519  "finalizeInstrumentation called twice");
4520  IRBuilder<> IRB(MSV.ActualFnStart->getFirstNonPHI());
4521  VAArgSize = IRB.CreateLoad(IRB.getInt64Ty(), MS.VAArgOverflowSizeTLS);
4522  Value *CopySize = IRB.CreateAdd(ConstantInt::get(MS.IntptrTy, 0),
4523  VAArgSize);
4524 
4525  if (!VAStartInstrumentationList.empty()) {
4526  // If there is a va_start in this function, make a backup copy of
4527  // va_arg_tls somewhere in the function entry block.
4528  VAArgTLSCopy = IRB.CreateAlloca(Type::getInt8Ty(*MS.C), CopySize);
4529  IRB.CreateMemCpy(VAArgTLSCopy, 8, MS.VAArgTLS, 8, CopySize);
4530  }
4531 
4532  // Instrument va_start.
4533  // Copy va_list shadow from the backup copy of the TLS contents.
4534  for (size_t i = 0, n = VAStartInstrumentationList.size(); i < n; i++) {
4535  CallInst *OrigInst = VAStartInstrumentationList[i];
4536  IRBuilder<> IRB(OrigInst->getNextNode());
4537  Value *VAListTag = OrigInst->getArgOperand(0);
4538  Type *RegSaveAreaPtrTy = Type::getInt64PtrTy(*MS.C);
4539  Value *RegSaveAreaPtrPtr =
4540  IRB.CreateIntToPtr(IRB.CreatePtrToInt(VAListTag, MS.IntptrTy),
4541  PointerType::get(RegSaveAreaPtrTy, 0));
4542  Value *RegSaveAreaPtr =
4543  IRB.CreateLoad(RegSaveAreaPtrTy, RegSaveAreaPtrPtr);
4544  Value *RegSaveAreaShadowPtr, *RegSaveAreaOriginPtr;
4545  unsigned Alignment = 8;
4546  std::tie(RegSaveAreaShadowPtr, RegSaveAreaOriginPtr) =
4547  MSV.getShadowOriginPtr(RegSaveAreaPtr, IRB, IRB.getInt8Ty(),
4548  Alignment, /*isStore*/ true);
4549  IRB.CreateMemCpy(RegSaveAreaShadowPtr, Alignment, VAArgTLSCopy, Alignment,
4550  CopySize);
4551  }
4552  }
4553 };
4554 
4555 /// A no-op implementation of VarArgHelper.
4556 struct VarArgNoOpHelper : public VarArgHelper {
4557  VarArgNoOpHelper(Function &F, MemorySanitizer &MS,
4558  MemorySanitizerVisitor &MSV) {}
4559 
4560  void visitCallSite(CallSite &CS, IRBuilder<> &IRB) override {}
4561 
4562  void visitVAStartInst(VAStartInst &I) override {}
4563 
4564  void visitVACopyInst(VACopyInst &I) override {}
4565 
4566  void finalizeInstrumentation() override {}
4567 };
4568 
4569 } // end anonymous namespace
4570 
4571 static VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
4572  MemorySanitizerVisitor &Visitor) {
4573  // VarArg handling is only implemented on AMD64. False positives are possible
4574  // on other platforms.
4575  Triple TargetTriple(Func.getParent()->getTargetTriple());
4576  if (TargetTriple.getArch() == Triple::x86_64)
4577  return new VarArgAMD64Helper(Func, Msan, Visitor);
4578  else if (TargetTriple.isMIPS64())
4579  return new VarArgMIPS64Helper(Func, Msan, Visitor);
4580  else if (TargetTriple.getArch() == Triple::aarch64)
4581  return new VarArgAArch64Helper(Func, Msan, Visitor);
4582  else if (TargetTriple.getArch() == Triple::ppc64 ||
4583  TargetTriple.getArch() == Triple::ppc64le)
4584  return new VarArgPowerPC64Helper(Func, Msan, Visitor);
4585  else
4586  return new VarArgNoOpHelper(Func, Msan, Visitor);
4587 }
4588 
4589 bool MemorySanitizer::sanitizeFunction(Function &F, TargetLibraryInfo &TLI) {
4590  if (!CompileKernel && F.getName() == kMsanModuleCtorName)
4591  return false;
4592 
4593  MemorySanitizerVisitor Visitor(F, *this, TLI);
4594 
4595  // Clear out readonly/readnone attributes.
4596  AttrBuilder B;
4597  B.addAttribute(Attribute::ReadOnly)
4598  .addAttribute(Attribute::ReadNone);
4600 
4601  return Visitor.runOnFunction();
4602 }
Value * CreateInBoundsGEP(Value *Ptr, ArrayRef< Value *> IdxList, const Twine &Name="")
Definition: IRBuilder.h:1695
Type * getVectorElementType() const
Definition: Type.h:376
IterTy arg_end() const
Definition: CallSite.h:588
uint64_t CallInst * C
Return a value (possibly void), from a function.
SymbolTableList< Instruction >::iterator eraseFromParent()
This method unlinks &#39;this&#39; from the containing basic block and deletes it.
Definition: Instruction.cpp:67
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:598
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:241
static const MemoryMapParams Linux_PowerPC64_MemoryMapParams
Value * CreateICmp(CmpInst::Predicate P, Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:2211
Value * CreateConstGEP1_32(Value *Ptr, unsigned Idx0, const Twine &Name="")
Definition: IRBuilder.h:1734
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:2579
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:1458
This class represents an incoming formal argument to a Function.
Definition: Argument.h:29
static const PlatformMemoryMapParams Linux_PowerPC_MemoryMapParams
AllocaInst * CreateAlloca(Type *Ty, unsigned AddrSpace, Value *ArraySize=nullptr, const Twine &Name="")
Definition: IRBuilder.h:1562
bool doesNotAccessMemory(unsigned OpNo) const
Definition: InstrTypes.h:1551
Base class for instruction visitors.
Definition: InstVisitor.h:80
Value * getAggregateOperand()
Value * CreateICmpNE(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:2105
Atomic ordering constants.
NodeTy * getNextNode()
Get the next node, or nullptr for the list tail.
Definition: ilist_node.h:288
PassT::Result & getResult(IRUnitT &IR, ExtraArgTs... ExtraArgs)
Get the result of an analysis pass for a given IR unit.
Definition: PassManager.h:777
IterTy arg_begin() const
Definition: CallSite.h:584
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
This class represents lattice values for constants.
Definition: AllocatorList.h:23
BinaryOps getOpcode() const
Definition: InstrTypes.h:402
bool isAtomic() const
Return true if this instruction has an AtomicOrdering of unordered or higher.
LoadInst * CreateLoad(Type *Ty, Value *Ptr, const char *Name)
Provided to resolve &#39;CreateLoad(Ty, Ptr, "...")&#39; correctly, instead of converting the string to &#39;bool...
Definition: IRBuilder.h:1575
Value * CreateXor(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1320
A Module instance is used to store all the information related to an LLVM module. ...
Definition: Module.h:66
bool isSized(SmallPtrSetImpl< Type *> *Visited=nullptr) const
Return true if it makes sense to take the size of this type.
Definition: Type.h:265
LoadInst * CreateAlignedLoad(Type *Ty, Value *Ptr, unsigned Align, const char *Name)
Provided to resolve &#39;CreateAlignedLoad(Ptr, Align, "...")&#39; correctly, instead of converting the strin...
Definition: IRBuilder.h:1611
void setOrdering(AtomicOrdering Ordering)
Sets the ordering constraint of this rmw instruction.
Definition: Instructions.h:802
static const MemoryMapParams Linux_I386_MemoryMapParams
Same, but only replaced by something equivalent.
Definition: GlobalValue.h:53
amdgpu Simplify well known AMD library false FunctionCallee Value const Twine & Name
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:536
static const MemoryMapParams NetBSD_X86_64_MemoryMapParams
A handy container for a FunctionType+Callee-pointer pair, which can be passed around as a single enti...
Definition: DerivedTypes.h:170
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
Value * CreateICmpSLT(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:2133
This class represents a function call, abstracting a target machine&#39;s calling convention.
static cl::opt< bool > ClHandleAsmConservative("msan-handle-asm-conservative", cl::desc("conservative handling of inline assembly"), cl::Hidden, cl::init(true))
static PointerType * getInt32PtrTy(LLVMContext &C, unsigned AS=0)
Definition: Type.cpp:232
void setOrdering(AtomicOrdering Ordering)
Sets the ordering constraint of this load instruction.
Definition: Instructions.h:258
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:637
const Value * getTrueValue() const
AtomicOrdering getOrdering() const
Returns the ordering constraint of this load instruction.
Definition: Instructions.h:252
Like Internal, but omit from symbol table.
Definition: GlobalValue.h:56
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:48
bool hasFnAttribute(Attribute::AttrKind Kind) const
Return true if the function has the attribute.
Definition: Function.h:323
A raw_ostream that writes to an SmallVector or SmallString.
Definition: raw_ostream.h:530
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:1881
Metadata node.
Definition: Metadata.h:863
static unsigned TypeSizeToSizeIndex(unsigned TypeSize)
Base class for all callable instructions (InvokeInst and CallInst) Holds everything related to callin...
Definition: InstrTypes.h:1100
F(f)
This class represents a sign extension of integer types.
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:440
An instruction for reading from memory.
Definition: Instructions.h:169
static IntegerType * getInt64Ty(LLVMContext &C)
Definition: Type.cpp:181
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:699
Hexagon Common GEP
TypeSize getTypeSizeInBits(Type *Ty) const
Size examples:
Definition: DataLayout.h:624
bool isVectorTy() const
True if this is an instance of VectorType.
Definition: Type.h:230
#define op(i)
BasicBlock * SplitBlock(BasicBlock *Old, Instruction *SplitPt, DominatorTree *DT=nullptr, LoopInfo *LI=nullptr, MemorySSAUpdater *MSSAU=nullptr, const Twine &BBName="")
Split the specified block at the specified instruction - everything before SplitPt stays in Old and e...
bool isMustTailCall() const
static Type * getX86_MMXTy(LLVMContext &C)
Definition: Type.cpp:175
Use * op_iterator
Definition: User.h:224
bool isPPC_FP128Ty() const
Return true if this is powerpc long double.
Definition: Type.h:159
bool erase(const T &V)
Definition: SmallSet.h:207
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:229
static PointerType * getInt64PtrTy(LLVMContext &C, unsigned AS=0)
Definition: Type.cpp:236
static Constant * get(ArrayType *T, ArrayRef< Constant *> V)
Definition: Constants.cpp:1014
FunctionType * getFunctionType() const
Definition: CallSite.h:328
unsigned getBitWidth() const
Return the number of bits in the APInt.
Definition: APInt.h:1517
static Constant * getNullValue(Type *Ty)
Constructor to create a &#39;0&#39; constant of arbitrary type.
Definition: Constants.cpp:289
Value * CreateNot(Value *V, const Twine &Name="")
Definition: IRBuilder.h:1523
StoreInst * CreateAlignedStore(Value *Val, Value *Ptr, unsigned Align, bool isVolatile=false)
Definition: IRBuilder.h:1648
IntegerType * getInt32Ty()
Fetch the type representing a 32-bit integer.
Definition: IRBuilder.h:383
Value * getArgOperand(unsigned i) const
Definition: InstrTypes.h:1241
unsigned countTrailingZeros() const
Count the number of trailing zero bits.
Definition: APInt.h:1640
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:50
bool isSigned() const
Definition: InstrTypes.h:902
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:381
const DataLayout & getDataLayout() const
Get the data layout for the module&#39;s target platform.
Definition: Module.cpp:369
unsigned getAlignment() const
Return the alignment of the memory that is being allocated by the instruction.
Definition: Instructions.h:112
IntegerType * getInt64Ty()
Fetch the type representing a 64-bit integer.
Definition: IRBuilder.h:388
This is the base class for all instructions that perform data casts.
Definition: InstrTypes.h:439
ArrayRef< T > makeArrayRef(const T &OneElt)
Construct an ArrayRef from a single element.
Definition: ArrayRef.h:450
&#39;undef&#39; values are things that do not have specified contents.
Definition: Constants.h:1285
This class wraps the llvm.memmove intrinsic.
Class to represent struct types.
Definition: DerivedTypes.h:238
LLVMContext & getContext() const
Get the global data context.
Definition: Module.h:245
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:659
bool isUnsigned() const
Definition: InstrTypes.h:908
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:779
IntegerType * getIntPtrTy(const DataLayout &DL, unsigned AddrSpace=0)
Fetch the type representing a pointer to an integer value.
Definition: IRBuilder.h:426
unsigned getArgumentNo(Value::const_user_iterator I) const
Given a value use iterator, returns the argument that corresponds to it.
Definition: CallSite.h:206
This file contains the simple types necessary to represent the attributes associated with functions a...
Value * CreateAdd(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1118
void setName(const Twine &Name)
Change the name of the value.
Definition: Value.cpp:285
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:346
This file implements a class to represent arbitrary precision integral constant values and operations...
Type * getVoidTy()
Fetch the type representing void.
Definition: IRBuilder.h:416
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:1604
InstrTy * getInstruction() const
Definition: CallSite.h:96
Value * CreateIntToPtr(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1958
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:108
Value * CreateBitCast(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1963
Type * getType() const
All values are typed, get the type of this value.
Definition: Value.h:246
This represents the llvm.va_start intrinsic.
static const char *const kMsanInitName
TypeSize getTypeStoreSize(Type *Ty) const
Returns the maximum number of bytes that may be overwritten by storing the specified type...
Definition: DataLayout.h:454
std::string itostr(int64_t X)
Definition: StringExtras.h:238
AtomicOrdering getSuccessOrdering() const
Returns the success ordering constraint of this cmpxchg instruction.
Definition: Instructions.h:590
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:408
std::pair< Function *, FunctionCallee > getOrCreateSanitizerCtorAndInitFunctions(Module &M, StringRef CtorName, StringRef InitName, ArrayRef< Type *> InitArgTypes, ArrayRef< Value *> InitArgs, function_ref< void(Function *, FunctionCallee)> FunctionsCreatedCallback, StringRef VersionCheckName=StringRef())
Creates sanitizer constructor function lazily.
static cl::opt< uint64_t > ClOriginBase("msan-origin-base", cl::desc("Define custom MSan OriginBase"), cl::Hidden, cl::init(0))
This instruction compares its operands according to the predicate given to the constructor.
static bool isStore(int Opcode)
void setComdat(Comdat *C)
Definition: GlobalObject.h:122
bool isVarArg() const
Definition: DerivedTypes.h:128
This class represents a no-op cast from one type to another.
MDNode * getMetadata(unsigned KindID) const
Get the metadata of given kind attached to this Instruction.
Definition: Instruction.h:244
Value * getInsertedValueOperand()
SmallString - A SmallString is just a SmallVector with methods and accessors that make it work better...
Definition: SmallString.h:25
bool paramHasAttr(unsigned ArgNo, Attribute::AttrKind Kind) const
Return true if the call or the callee has the given attribute.
Definition: CallSite.h:385
Value * CreateSub(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1135
AttributeList getAttributes() const
Return the attribute list for this Function.
Definition: Function.h:223
An instruction for storing to memory.
Definition: Instructions.h:325
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:1877
ConstraintPrefix Type
Type - The basic type of the constraint: input/output/clobber.
Definition: InlineAsm.h:120
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:1093
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:132
Value * getOperand(unsigned i) const
Definition: User.h:169
Analysis containing CSE Info
Definition: CSEInfo.cpp:20
static PreservedAnalyses none()
Convenience factory function for the empty preserved set.
Definition: PassManager.h:157
static const PlatformMemoryMapParams Linux_MIPS_MemoryMapParams
bool isRelational() const
Return true if the predicate is relational (not EQ or NE).
Value * CreateOr(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1294
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:359
Type * getScalarType() const
If this is a vector type, return the element type, otherwise return &#39;this&#39;.
Definition: Type.h:307
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:669
an instruction for type-safe pointer arithmetic to access elements of arrays and structs ...
Definition: Instructions.h:881
static bool runOnFunction(Function &F, bool PostInlining)
initializer< Ty > init(const Ty &Val)
Definition: CommandLine.h:432
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:196
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
A set of analyses that are preserved following a run of a transformation pass.
Definition: PassManager.h:154
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:223
const BasicBlock * getSinglePredecessor() const
Return the predecessor of this block if it has a single predecessor block.
Definition: BasicBlock.cpp:240
Value * getCalledValue() const
Definition: InstrTypes.h:1280
static Constant * getOrInsertGlobal(Module &M, StringRef Name, Type *Ty)
LLVM Basic Block Representation.
Definition: BasicBlock.h:57
TypeSize getPrimitiveSizeInBits() const LLVM_READONLY
Return the basic size of this type if it is a primitive type.
Definition: Type.cpp:115
static cl::opt< uint64_t > ClShadowBase("msan-shadow-base", cl::desc("Define custom MSan ShadowBase"), cl::Hidden, cl::init(0))
The instances of the Type class are immutable: once they are created, they are never changed...
Definition: Type.h:46
This is an important class for using LLVM in a threaded context.
Definition: LLVMContext.h:64
const char * getOpcodeName() const
Definition: Instruction.h:127
This is an important base class in LLVM.
Definition: Constant.h:41
Resume the propagation of an exception.
SmallSet - This maintains a set of unique values, optimizing for the case when the set is small (less...
Definition: SmallSet.h:134
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:2287
bool isPointerTy() const
True if this is an instance of PointerType.
Definition: Type.h:224
static cl::opt< uint64_t > ClXorMask("msan-xor-mask", cl::desc("Define custom MSan XorMask"), cl::Hidden, cl::init(0))
TypeSize getTypeAllocSize(Type *Ty) const
Returns the offset in bytes between successive objects of the specified type, including alignment pad...
Definition: DataLayout.h:486
FunctionType * getFunctionType() const
Definition: InstrTypes.h:1144
unsigned getNumParams() const
Return the number of fixed parameters this function type requires.
Definition: DerivedTypes.h:144
int getNumOccurrences() const
Definition: CommandLine.h:393
Represent the analysis usage information of a pass.
op_iterator op_end()
Definition: User.h:231
AllocaInst * findAllocaForValue(Value *V, DenseMap< Value *, AllocaInst *> &AllocaForValue)
Finds alloca where the value comes from.
Value * CreateNeg(Value *V, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1486
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:732
FunctionPass class - This class is used to implement most global optimizations.
Definition: Pass.h:284
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:301
static Constant * get(StructType *T, ArrayRef< Constant *> V)
Definition: Constants.cpp:1075
Value * getPointerOperand()
Definition: Instructions.h:289
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:2101
self_iterator getIterator()
Definition: ilist_node.h:81
Class to represent integer types.
Definition: DerivedTypes.h:40
std::pair< NoneType, bool > insert(const T &V)
insert - Insert an element into the set if it isn&#39;t already there.
Definition: SmallSet.h:180
IntegerType * getIntNTy(unsigned N)
Fetch the type representing an N-bit integer.
Definition: IRBuilder.h:396
Value * CreateExtractElement(Value *Vec, Value *Idx, const Twine &Name="")
Definition: IRBuilder.h:2309
This class represents a cast from an integer to a pointer.
const Value * getCondition() const
static Constant * getAllOnesValue(Type *Ty)
Definition: Constants.cpp:343
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:482
INITIALIZE_PASS_BEGIN(MemorySanitizerLegacyPass, "msan", "MemorySanitizer: detects uninitialized reads.", false, false) INITIALIZE_PASS_END(MemorySanitizerLegacyPass
static PreservedAnalyses all()
Construct a special preserved set that preserves all passes.
Definition: PassManager.h:160
const Value * getArraySize() const
Get the number of elements allocated.
Definition: Instructions.h:92
Value * CreateExtractValue(Value *Agg, ArrayRef< unsigned > Idxs, const Twine &Name="")
Definition: IRBuilder.h:2351
size_t size() const
Definition: SmallVector.h:52
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:421
MDNode * createBranchWeights(uint32_t TrueWeight, uint32_t FalseWeight)
Return metadata containing two branch weights.
Definition: MDBuilder.cpp:37
AtomicOrdering getOrdering() const
Returns the ordering constraint of this rmw instruction.
Definition: Instructions.h:797
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:1152
Value * CreateTrunc(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1873
This struct is a compact representation of a valid (non-zero power of two) alignment.
Definition: Alignment.h:40
Type * getAllocatedType() const
Return the type that is being allocated by the instruction.
Definition: Instructions.h:105
Triple - Helper class for working with autoconf configuration names.
Definition: Triple.h:43
signed greater than
Definition: InstrTypes.h:759
std::vector< ConstraintInfo > ConstraintInfoVector
Definition: InlineAsm.h:115
unsigned first
Intrinsic::ID getIntrinsicID() const
Return the intrinsic ID of this intrinsic.
Definition: IntrinsicInst.h:50
PHINode * CreatePHI(Type *Ty, unsigned NumReservedValues, const Twine &Name="")
Definition: IRBuilder.h:2231
Value * CreateGEP(Value *Ptr, ArrayRef< Value *> IdxList, const Twine &Name="")
Definition: IRBuilder.h:1676
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:244
Type * getSequentialElementType() const
Definition: Type.h:361
Iterator for intrusive lists based on ilist_node.
unsigned getNumOperands() const
Definition: User.h:191
See the file comment.
Definition: ValueMap.h:85
This is the shared class of boolean and integer constants.
Definition: Constants.h:83
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:1146
Align max(MaybeAlign Lhs, Align Rhs)
Definition: Alignment.h:390
iterator end()
Definition: BasicBlock.h:275
unsigned getScalarSizeInBits() const LLVM_READONLY
If this is a vector type, return the getPrimitiveSizeInBits value for the element type...
Definition: Type.cpp:134
CallingConv::ID getCallingConv() const
getCallingConv()/setCallingConv(CC) - These method get and set the calling convention of this functio...
Definition: Function.h:212
Value * CreateIntCast(Value *V, Type *DestTy, bool isSigned, const Twine &Name="")
Definition: IRBuilder.h:2032
This is a &#39;vector&#39; (really, a variable-sized array), optimized for the case when the array is small...
Definition: SmallVector.h:837
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:2322
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:755
bool isAggregateType() const
Return true if the type is an aggregate type.
Definition: Type.h:258
signed less than
Definition: InstrTypes.h:761
CallInst * CreateMaskedStore(Value *Val, Value *Ptr, unsigned Align, Value *Mask)
Create a call to Masked Store intrinsic.
Definition: IRBuilder.cpp:491
CHAIN = SC CHAIN, Imm128 - System call.
ConstantInt * getInt32(uint32_t C)
Get a constant 32-bit value.
Definition: IRBuilder.h:343
static IntegerType * getIntNTy(LLVMContext &C, unsigned N)
Definition: Type.cpp:184
StringRef str()
Return a StringRef for the vector contents.
Definition: raw_ostream.h:555
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:482
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:653
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:63
FunctionCallee getOrInsertFunction(StringRef Name, FunctionType *T, AttributeList AttributeList)
Look up the specified function in the module symbol table.
Definition: Module.cpp:143
CallInst * CreateMaskedLoad(Value *Ptr, unsigned Align, Value *Mask, Value *PassThru=nullptr, const Twine &Name="")
Create a call to Masked Load intrinsic.
Definition: IRBuilder.cpp:470
User::op_iterator arg_iterator
The type of iterator to use when looping over actual arguments at this call site. ...
Definition: CallSite.h:220
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:132
unsigned getVectorNumElements() const
Definition: DerivedTypes.h:566
signed less or equal
Definition: InstrTypes.h:762
Class to represent vector types.
Definition: DerivedTypes.h:432
const Module * getModule() const
Return the module owning the function this instruction belongs to or nullptr it the function does not...
Definition: Instruction.cpp:55
Class for arbitrary precision integers.
Definition: APInt.h:69
IntegerType * getInt8Ty()
Fetch the type representing an 8-bit integer.
Definition: IRBuilder.h:373
Value * CreateShl(Value *LHS, Value *RHS, const Twine &Name="", bool HasNUW=false, bool HasNSW=false)
Definition: IRBuilder.h:1207
const Value * getFalseValue() const
Value * CreatePointerCast(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:2009
static cl::opt< uint64_t > ClAndMask("msan-and-mask", cl::desc("Define custom MSan AndMask"), cl::Hidden, cl::init(0))
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")))
void removeAttributes(unsigned i, const AttrBuilder &Attrs)
removes the attributes from the list of attributes.
Definition: Function.cpp:429
Predicate getPredicate() const
Return the predicate for this instruction.
Definition: InstrTypes.h:807
uint64_t alignTo(uint64_t Size, Align A)
Returns a multiple of A needed to store Size bytes.
Definition: Alignment.h:163
unsigned getNumArgOperands() const
Definition: InstrTypes.h:1239
static cl::opt< bool > ClEnableKmsan("msan-kernel", cl::desc("Enable KernelMemorySanitizer instrumentation"), cl::Hidden, cl::init(false))
The C convention as implemented on Windows/x86-64 and AArch64.
Definition: CallingConv.h:158
Constant * getOrInsertGlobal(StringRef Name, Type *Ty, function_ref< GlobalVariable *()> CreateGlobalCallback)
Look up the specified global in the module symbol table.
Definition: Module.cpp:204
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:242
LLVM_NODISCARD bool empty() const
Definition: SmallVector.h:55
static VectorType * get(Type *ElementType, ElementCount EC)
This static method is the primary way to construct an VectorType.
Definition: Type.cpp:614
StringRef getName() const
Return a constant reference to the value&#39;s name.
Definition: Value.cpp:214
Function * getCalledFunction() const
Returns the function called, or null if this is an indirect function invocation.
Definition: InstrTypes.h:1287
bool onlyReadsMemory(unsigned OpNo) const
Definition: InstrTypes.h:1557
#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:587
static cl::opt< bool > ClHandleLifetimeIntrinsics("msan-handle-lifetime-intrinsics", cl::desc("when possible, poison scoped variables at the beginning of the scope " "(slower, but more precise)"), cl::Hidden, cl::init(true))
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:332
This instruction extracts a single (scalar) element from a VectorType value.
uint32_t Size
Definition: Profile.cpp:46
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:2891
CallInst * CreateCall(FunctionType *FTy, Value *Callee, ArrayRef< Value *> Args=None, const Twine &Name="", MDNode *FPMathTag=nullptr)
Definition: IRBuilder.h:2239
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:42
PreservedAnalyses run(Function &F, FunctionAnalysisManager &FAM)
bool removeUnreachableBlocks(Function &F, DomTreeUpdater *DTU=nullptr, MemorySSAUpdater *MSSAU=nullptr)
Remove all blocks that can not be reached from the function&#39;s entry.
Definition: Local.cpp:2214
static cl::opt< bool > ClCheckConstantShadow("msan-check-constant-shadow", cl::desc("Insert checks for constant shadow values"), cl::Hidden, cl::init(false))
bool isInvoke() const
Return true if a InvokeInst is enclosed.
Definition: CallSite.h:91
static const MemoryMapParams Linux_MIPS64_MemoryMapParams
Value * CreateAnd(Value *LHS, Value *RHS, const Twine &Name="")
Definition: IRBuilder.h:1268
Value * CreatePtrToInt(Value *V, Type *DestTy, const Twine &Name="")
Definition: IRBuilder.h:1953
static const unsigned kOriginSize
const std::string to_string(const T &Value)
Definition: ScopedPrinter.h:61
static const MemoryMapParams FreeBSD_I386_MemoryMapParams
bool isCallBr() const
Return true if a CallBrInst is enclosed.
Definition: CallSite.h:94
Analysis pass providing the TargetLibraryInfo.
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)
bool isCall() const
Return true if a CallInst is enclosed.
Definition: CallSite.h:87
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:152
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
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))
void setSuccessOrdering(AtomicOrdering Ordering)
Sets the success ordering constraint of this cmpxchg instruction.
Definition: Instructions.h:595
ArrayRef< unsigned > getIndices() const
Module * getParent()
Get the module that this global value is contained inside of...
Definition: GlobalValue.h:575
LLVM Value Representation.
Definition: Value.h:74
unsigned getParamAlignment(unsigned ArgNo) const
Extract the alignment for a call or parameter (0=unknown).
Definition: CallSite.h:414
#define LLVM_FALLTHROUGH
LLVM_FALLTHROUGH - Mark fallthrough cases in switch statements.
Definition: Compiler.h:273
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:80
AttributeSet getFnAttributes() const
The function attributes are returned.
BasicBlock::iterator GetInsertPoint() const
Definition: IRBuilder.h:127
Value * CreateLShr(Value *LHS, Value *RHS, const Twine &Name="", bool isExact=false)
Definition: IRBuilder.h:1228
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:333
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:48