1//===-- Lint.cpp - Check for common errors in LLVM IR ---------------------===//
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// This pass statically checks for common and easily-identified constructs
10// which produce undefined or likely unintended behavior in LLVM IR.
11//
12// It is not a guarantee of correctness, in two ways. First, it isn't
13// comprehensive. There are checks which could be done statically which are
14// not yet implemented. Some of these are indicated by TODO comments, but
15// those aren't comprehensive either. Second, many conditions cannot be
16// checked statically. This pass does no dynamic instrumentation, so it
17// can't check for all possible problems.
18//
19// Another limitation is that it assumes all code will be executed. A store
20// through a null pointer in a basic block which is never reached is harmless,
21// but this pass will warn about it anyway. This is the main reason why most
22// of these checks live here instead of in the Verifier pass.
23//
24// Optimization passes may make conditions that this pass checks for more or
25// less obvious. If an optimization pass appears to be introducing a warning,
26// it may be that the optimization pass is merely exposing an existing
27// condition in the code.
28//
29// This code may be run before instcombine. In many cases, instcombine checks
30// for the same kinds of things and turns instructions with undefined behavior
31// into unreachable (or equivalent). Because of this, this pass makes some
32// effort to look through bitcasts and so on.
33//
34//===----------------------------------------------------------------------===//
35
36#include "llvm/Analysis/Lint.h"
37#include "llvm/ADT/APInt.h"
38#include "llvm/ADT/ArrayRef.h"
39#include "llvm/ADT/SmallPtrSet.h"
40#include "llvm/ADT/Twine.h"
41#include "llvm/Analysis/AliasAnalysis.h"
42#include "llvm/Analysis/AssumptionCache.h"
43#include "llvm/Analysis/BasicAliasAnalysis.h"
44#include "llvm/Analysis/ConstantFolding.h"
45#include "llvm/Analysis/InstructionSimplify.h"
46#include "llvm/Analysis/Loads.h"
47#include "llvm/Analysis/MemoryLocation.h"
48#include "llvm/Analysis/ScopedNoAliasAA.h"
49#include "llvm/Analysis/TargetLibraryInfo.h"
50#include "llvm/Analysis/TypeBasedAliasAnalysis.h"
51#include "llvm/Analysis/ValueTracking.h"
52#include "llvm/IR/Argument.h"
53#include "llvm/IR/BasicBlock.h"
54#include "llvm/IR/Constant.h"
55#include "llvm/IR/Constants.h"
56#include "llvm/IR/DataLayout.h"
57#include "llvm/IR/DerivedTypes.h"
58#include "llvm/IR/Dominators.h"
59#include "llvm/IR/Function.h"
60#include "llvm/IR/GlobalVariable.h"
61#include "llvm/IR/InstVisitor.h"
62#include "llvm/IR/InstrTypes.h"
63#include "llvm/IR/Instruction.h"
64#include "llvm/IR/Instructions.h"
65#include "llvm/IR/IntrinsicInst.h"
66#include "llvm/IR/Module.h"
67#include "llvm/IR/PassManager.h"
68#include "llvm/IR/Type.h"
69#include "llvm/IR/Value.h"
70#include "llvm/Support/AMDGPUAddrSpace.h"
71#include "llvm/Support/Casting.h"
72#include "llvm/Support/KnownBits.h"
73#include "llvm/Support/raw_ostream.h"
74#include <cassert>
75#include <cstdint>
76#include <iterator>
77#include <string>
78
79using namespace llvm;
80
81namespace {
82namespace MemRef {
83static const unsigned Read = 1;
84static const unsigned Write = 2;
85static const unsigned Callee = 4;
86static const unsigned Branchee = 8;
87} // end namespace MemRef
88
89class Lint : public InstVisitor<Lint> {
90 friend class InstVisitor<Lint>;
91
92 void visitFunction(Function &F);
93
94 void visitCallBase(CallBase &CB);
95 void visitMemoryReference(Instruction &I, const MemoryLocation &Loc,
96 MaybeAlign Alignment, Type *Ty, unsigned Flags);
97
98 void visitReturnInst(ReturnInst &I);
99 void visitLoadInst(LoadInst &I);
100 void visitStoreInst(StoreInst &I);
101 void visitAtomicCmpXchgInst(AtomicCmpXchgInst &I);
102 void visitAtomicRMWInst(AtomicRMWInst &I);
103 void visitXor(BinaryOperator &I);
104 void visitSub(BinaryOperator &I);
105 void visitLShr(BinaryOperator &I);
106 void visitAShr(BinaryOperator &I);
107 void visitShl(BinaryOperator &I);
108 void visitSDiv(BinaryOperator &I);
109 void visitUDiv(BinaryOperator &I);
110 void visitSRem(BinaryOperator &I);
111 void visitURem(BinaryOperator &I);
112 void visitAllocaInst(AllocaInst &I);
113 void visitVAArgInst(VAArgInst &I);
114 void visitIndirectBrInst(IndirectBrInst &I);
115 void visitExtractElementInst(ExtractElementInst &I);
116 void visitInsertElementInst(InsertElementInst &I);
117 void visitUnreachableInst(UnreachableInst &I);
118
119 Value *findValue(Value *V, bool OffsetOk) const;
120 Value *findValueImpl(Value *V, bool OffsetOk,
121 SmallPtrSetImpl<Value *> &Visited) const;
122
123public:
124 Module *Mod;
125 const Triple &TT;
126 const DataLayout *DL;
127 AliasAnalysis *AA;
128 AssumptionCache *AC;
129 DominatorTree *DT;
130 TargetLibraryInfo *TLI;
131
132 std::string Messages;
133 raw_string_ostream MessagesStr;
134
135 Lint(Module *Mod, const DataLayout *DL, AliasAnalysis *AA,
136 AssumptionCache *AC, DominatorTree *DT, TargetLibraryInfo *TLI)
137 : Mod(Mod), TT(Mod->getTargetTriple()), DL(DL), AA(AA), AC(AC), DT(DT),
138 TLI(TLI), MessagesStr(Messages) {}
139
140 void WriteValues(ArrayRef<const Value *> Vs) {
141 for (const Value *V : Vs) {
142 if (!V)
143 continue;
144 if (isa<Instruction>(Val: V)) {
145 MessagesStr << *V << '\n';
146 } else {
147 V->printAsOperand(O&: MessagesStr, PrintType: true, M: Mod);
148 MessagesStr << '\n';
149 }
150 }
151 }
152
153 /// A check failed, so printout out the condition and the message.
154 ///
155 /// This provides a nice place to put a breakpoint if you want to see why
156 /// something is not correct.
157 void CheckFailed(const Twine &Message) { MessagesStr << Message << '\n'; }
158
159 /// A check failed (with values to print).
160 ///
161 /// This calls the Message-only version so that the above is easier to set
162 /// a breakpoint on.
163 template <typename T1, typename... Ts>
164 void CheckFailed(const Twine &Message, const T1 &V1, const Ts &... Vs) {
165 CheckFailed(Message);
166 WriteValues(Vs: {V1, Vs...});
167 }
168};
169} // end anonymous namespace
170
171// Check - We know that cond should be true, if not print an error message.
172#define Check(C, ...) \
173 do { \
174 if (!(C)) { \
175 CheckFailed(__VA_ARGS__); \
176 return; \
177 } \
178 } while (false)
179
180void Lint::visitFunction(Function &F) {
181 // This isn't undefined behavior, it's just a little unusual, and it's a
182 // fairly common mistake to neglect to name a function.
183 Check(F.hasName() || F.hasLocalLinkage(),
184 "Unusual: Unnamed function with non-local linkage", &F);
185
186 // TODO: Check for irreducible control flow.
187}
188
189void Lint::visitCallBase(CallBase &I) {
190 Value *Callee = I.getCalledOperand();
191
192 visitMemoryReference(I, Loc: MemoryLocation::getAfter(Ptr: Callee), Alignment: std::nullopt,
193 Ty: nullptr, Flags: MemRef::Callee);
194
195 if (Function *F = dyn_cast<Function>(Val: findValue(V: Callee,
196 /*OffsetOk=*/false))) {
197 Check(I.getCallingConv() == F->getCallingConv(),
198 "Undefined behavior: Caller and callee calling convention differ",
199 &I);
200
201 FunctionType *FT = F->getFunctionType();
202 unsigned NumActualArgs = I.arg_size();
203
204 Check(FT->isVarArg() ? FT->getNumParams() <= NumActualArgs
205 : FT->getNumParams() == NumActualArgs,
206 "Undefined behavior: Call argument count mismatches callee "
207 "argument count",
208 &I);
209
210 Check(FT->getReturnType() == I.getType(),
211 "Undefined behavior: Call return type mismatches "
212 "callee return type",
213 &I);
214
215 // Check argument types (in case the callee was casted) and attributes.
216 // TODO: Verify that caller and callee attributes are compatible.
217 Function::arg_iterator PI = F->arg_begin(), PE = F->arg_end();
218 auto AI = I.arg_begin(), AE = I.arg_end();
219 for (; AI != AE; ++AI) {
220 Value *Actual = *AI;
221 if (PI != PE) {
222 Argument *Formal = &*PI++;
223 Check(Formal->getType() == Actual->getType(),
224 "Undefined behavior: Call argument type mismatches "
225 "callee parameter type",
226 &I);
227
228 // Check that noalias arguments don't alias other arguments. This is
229 // not fully precise because we don't know the sizes of the dereferenced
230 // memory regions.
231 if (Formal->hasNoAliasAttr() && Actual->getType()->isPointerTy()) {
232 AttributeList PAL = I.getAttributes();
233 unsigned ArgNo = 0;
234 for (auto *BI = I.arg_begin(); BI != AE; ++BI, ++ArgNo) {
235 // Skip ByVal arguments since they will be memcpy'd to the callee's
236 // stack so we're not really passing the pointer anyway.
237 if (PAL.hasParamAttr(ArgNo, Kind: Attribute::ByVal))
238 continue;
239 // If both arguments are readonly, they have no dependence.
240 if (Formal->onlyReadsMemory() && I.onlyReadsMemory(OpNo: ArgNo))
241 continue;
242 // Skip readnone arguments since those are guaranteed not to be
243 // dereferenced anyway.
244 if (I.doesNotAccessMemory(OpNo: ArgNo))
245 continue;
246 if (AI != BI && (*BI)->getType()->isPointerTy() &&
247 !isa<ConstantPointerNull>(Val: *BI)) {
248 AliasResult Result = AA->alias(V1: *AI, V2: *BI);
249 Check(Result != AliasResult::MustAlias &&
250 Result != AliasResult::PartialAlias,
251 "Unusual: noalias argument aliases another argument", &I);
252 }
253 }
254 }
255
256 // Check that an sret argument points to valid memory.
257 if (Formal->hasStructRetAttr() && Actual->getType()->isPointerTy()) {
258 Type *Ty = Formal->getParamStructRetType();
259 MemoryLocation Loc(
260 Actual, LocationSize::precise(Value: DL->getTypeStoreSize(Ty)));
261 visitMemoryReference(I, Loc, Alignment: DL->getABITypeAlign(Ty), Ty,
262 Flags: MemRef::Read | MemRef::Write);
263 }
264
265 // Check that ABI attributes for the function and call-site match.
266 unsigned ArgNo = AI->getOperandNo();
267 Attribute::AttrKind ABIAttributes[] = {
268 Attribute::ZExt, Attribute::SExt, Attribute::InReg,
269 Attribute::ByVal, Attribute::ByRef, Attribute::InAlloca,
270 Attribute::Preallocated, Attribute::StructRet};
271 AttributeList CallAttrs = I.getAttributes();
272 for (Attribute::AttrKind Attr : ABIAttributes) {
273 Attribute CallAttr = CallAttrs.getParamAttr(ArgNo, Kind: Attr);
274 Attribute FnAttr = F->getParamAttribute(ArgNo, Kind: Attr);
275 Check(CallAttr.isValid() == FnAttr.isValid(),
276 Twine("Undefined behavior: ABI attribute ") +
277 Attribute::getNameFromAttrKind(Attr) +
278 " not present on both function and call-site",
279 &I);
280 if (CallAttr.isValid() && FnAttr.isValid()) {
281 Check(CallAttr == FnAttr,
282 Twine("Undefined behavior: ABI attribute ") +
283 Attribute::getNameFromAttrKind(Attr) +
284 " does not have same argument for function and call-site",
285 &I);
286 }
287 }
288 }
289 }
290 }
291
292 if (const auto *CI = dyn_cast<CallInst>(Val: &I)) {
293 if (CI->isTailCall()) {
294 const AttributeList &PAL = CI->getAttributes();
295 unsigned ArgNo = 0;
296 for (Value *Arg : I.args()) {
297 // Skip ByVal arguments since they will be memcpy'd to the callee's
298 // stack anyway.
299 if (PAL.hasParamAttr(ArgNo: ArgNo++, Kind: Attribute::ByVal))
300 continue;
301 Value *Obj = findValue(V: Arg, /*OffsetOk=*/true);
302 Check(!isa<AllocaInst>(Obj),
303 "Undefined behavior: Call with \"tail\" keyword references "
304 "alloca",
305 &I);
306 }
307 }
308 }
309
310 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Val: &I))
311 switch (II->getIntrinsicID()) {
312 default:
313 break;
314
315 // TODO: Check more intrinsics
316
317 case Intrinsic::memcpy:
318 case Intrinsic::memcpy_inline: {
319 MemCpyInst *MCI = cast<MemCpyInst>(Val: &I);
320 visitMemoryReference(I, Loc: MemoryLocation::getForDest(MI: MCI),
321 Alignment: MCI->getDestAlign(), Ty: nullptr, Flags: MemRef::Write);
322 visitMemoryReference(I, Loc: MemoryLocation::getForSource(MTI: MCI),
323 Alignment: MCI->getSourceAlign(), Ty: nullptr, Flags: MemRef::Read);
324
325 // Check that the memcpy arguments don't overlap. The AliasAnalysis API
326 // isn't expressive enough for what we really want to do. Known partial
327 // overlap is not distinguished from the case where nothing is known.
328 auto Size = LocationSize::afterPointer();
329 if (const ConstantInt *Len =
330 dyn_cast<ConstantInt>(Val: findValue(V: MCI->getLength(),
331 /*OffsetOk=*/false)))
332 if (Len->getValue().isIntN(N: 32))
333 Size = LocationSize::precise(Value: Len->getValue().getZExtValue());
334 Check(AA->alias(MCI->getSource(), Size, MCI->getDest(), Size) !=
335 AliasResult::MustAlias,
336 "Undefined behavior: memcpy source and destination overlap", &I);
337 break;
338 }
339 case Intrinsic::memmove: {
340 MemMoveInst *MMI = cast<MemMoveInst>(Val: &I);
341 visitMemoryReference(I, Loc: MemoryLocation::getForDest(MI: MMI),
342 Alignment: MMI->getDestAlign(), Ty: nullptr, Flags: MemRef::Write);
343 visitMemoryReference(I, Loc: MemoryLocation::getForSource(MTI: MMI),
344 Alignment: MMI->getSourceAlign(), Ty: nullptr, Flags: MemRef::Read);
345 break;
346 }
347 case Intrinsic::memset:
348 case Intrinsic::memset_inline: {
349 MemSetInst *MSI = cast<MemSetInst>(Val: &I);
350 visitMemoryReference(I, Loc: MemoryLocation::getForDest(MI: MSI),
351 Alignment: MSI->getDestAlign(), Ty: nullptr, Flags: MemRef::Write);
352 break;
353 }
354 case Intrinsic::vastart:
355 // vastart in non-varargs function is rejected by the verifier
356 visitMemoryReference(I, Loc: MemoryLocation::getForArgument(Call: &I, ArgIdx: 0, TLI),
357 Alignment: std::nullopt, Ty: nullptr, Flags: MemRef::Read | MemRef::Write);
358 break;
359 case Intrinsic::vacopy:
360 visitMemoryReference(I, Loc: MemoryLocation::getForArgument(Call: &I, ArgIdx: 0, TLI),
361 Alignment: std::nullopt, Ty: nullptr, Flags: MemRef::Write);
362 visitMemoryReference(I, Loc: MemoryLocation::getForArgument(Call: &I, ArgIdx: 1, TLI),
363 Alignment: std::nullopt, Ty: nullptr, Flags: MemRef::Read);
364 break;
365 case Intrinsic::vaend:
366 visitMemoryReference(I, Loc: MemoryLocation::getForArgument(Call: &I, ArgIdx: 0, TLI),
367 Alignment: std::nullopt, Ty: nullptr, Flags: MemRef::Read | MemRef::Write);
368 break;
369
370 case Intrinsic::stackrestore:
371 // Stackrestore doesn't read or write memory, but it sets the
372 // stack pointer, which the compiler may read from or write to
373 // at any time, so check it for both readability and writeability.
374 visitMemoryReference(I, Loc: MemoryLocation::getForArgument(Call: &I, ArgIdx: 0, TLI),
375 Alignment: std::nullopt, Ty: nullptr, Flags: MemRef::Read | MemRef::Write);
376 break;
377 case Intrinsic::get_active_lane_mask:
378 if (auto *TripCount = dyn_cast<ConstantInt>(Val: I.getArgOperand(i: 1)))
379 Check(!TripCount->isZero(),
380 "get_active_lane_mask: operand #2 "
381 "must be greater than 0",
382 &I);
383 break;
384 }
385}
386
387void Lint::visitReturnInst(ReturnInst &I) {
388 Function *F = I.getParent()->getParent();
389 Check(!F->doesNotReturn(),
390 "Unusual: Return statement in function with noreturn attribute", &I);
391
392 if (Value *V = I.getReturnValue()) {
393 Value *Obj = findValue(V, /*OffsetOk=*/true);
394 Check(!isa<AllocaInst>(Obj), "Unusual: Returning alloca value", &I);
395 }
396}
397
398// TODO: Check that the reference is in bounds.
399// TODO: Check readnone/readonly function attributes.
400void Lint::visitMemoryReference(Instruction &I, const MemoryLocation &Loc,
401 MaybeAlign Align, Type *Ty, unsigned Flags) {
402 // If no memory is being referenced, it doesn't matter if the pointer
403 // is valid.
404 if (Loc.Size.isZero())
405 return;
406
407 Value *Ptr = const_cast<Value *>(Loc.Ptr);
408 Value *UnderlyingObject = findValue(V: Ptr, /*OffsetOk=*/true);
409 Check(!isa<ConstantPointerNull>(UnderlyingObject),
410 "Undefined behavior: Null pointer dereference", &I);
411 Check(!isa<UndefValue>(UnderlyingObject),
412 "Undefined behavior: Undef pointer dereference", &I);
413 Check(!isa<ConstantInt>(UnderlyingObject) ||
414 !cast<ConstantInt>(UnderlyingObject)->isMinusOne(),
415 "Unusual: All-ones pointer dereference", &I);
416 Check(!isa<ConstantInt>(UnderlyingObject) ||
417 !cast<ConstantInt>(UnderlyingObject)->isOne(),
418 "Unusual: Address one pointer dereference", &I);
419
420 if (Flags & MemRef::Write) {
421 if (TT.isAMDGPU())
422 Check(!AMDGPU::isConstantAddressSpace(
423 UnderlyingObject->getType()->getPointerAddressSpace()),
424 "Undefined behavior: Write to memory in const addrspace", &I);
425
426 if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(Val: UnderlyingObject))
427 Check(!GV->isConstant(), "Undefined behavior: Write to read-only memory",
428 &I);
429 Check(!isa<Function>(UnderlyingObject) &&
430 !isa<BlockAddress>(UnderlyingObject),
431 "Undefined behavior: Write to text section", &I);
432 }
433 if (Flags & MemRef::Read) {
434 Check(!isa<Function>(UnderlyingObject), "Unusual: Load from function body",
435 &I);
436 Check(!isa<BlockAddress>(UnderlyingObject),
437 "Undefined behavior: Load from block address", &I);
438 }
439 if (Flags & MemRef::Callee) {
440 Check(!isa<BlockAddress>(UnderlyingObject),
441 "Undefined behavior: Call to block address", &I);
442 }
443 if (Flags & MemRef::Branchee) {
444 Check(!isa<Constant>(UnderlyingObject) ||
445 isa<BlockAddress>(UnderlyingObject),
446 "Undefined behavior: Branch to non-blockaddress", &I);
447 }
448
449 // Check for buffer overflows and misalignment.
450 // Only handles memory references that read/write something simple like an
451 // alloca instruction or a global variable.
452 int64_t Offset = 0;
453 if (Value *Base = GetPointerBaseWithConstantOffset(Ptr, Offset, DL: *DL)) {
454 // OK, so the access is to a constant offset from Ptr. Check that Ptr is
455 // something we can handle and if so extract the size of this base object
456 // along with its alignment.
457 uint64_t BaseSize = MemoryLocation::UnknownSize;
458 MaybeAlign BaseAlign;
459
460 if (AllocaInst *AI = dyn_cast<AllocaInst>(Val: Base)) {
461 Type *ATy = AI->getAllocatedType();
462 if (!AI->isArrayAllocation() && ATy->isSized() && !ATy->isScalableTy())
463 BaseSize = DL->getTypeAllocSize(Ty: ATy).getFixedValue();
464 BaseAlign = AI->getAlign();
465 } else if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Val: Base)) {
466 // If the global may be defined differently in another compilation unit
467 // then don't warn about funky memory accesses.
468 if (GV->hasDefinitiveInitializer()) {
469 Type *GTy = GV->getValueType();
470 if (GTy->isSized())
471 BaseSize = DL->getTypeAllocSize(Ty: GTy);
472 BaseAlign = GV->getAlign();
473 if (!BaseAlign && GTy->isSized())
474 BaseAlign = DL->getABITypeAlign(Ty: GTy);
475 }
476 }
477
478 // Accesses from before the start or after the end of the object are not
479 // defined.
480 Check(!Loc.Size.hasValue() || Loc.Size.isScalable() ||
481 BaseSize == MemoryLocation::UnknownSize ||
482 (Offset >= 0 && Offset + Loc.Size.getValue() <= BaseSize),
483 "Undefined behavior: Buffer overflow", &I);
484
485 // Accesses that say that the memory is more aligned than it is are not
486 // defined.
487 if (!Align && Ty && Ty->isSized())
488 Align = DL->getABITypeAlign(Ty);
489 if (BaseAlign && Align)
490 Check(*Align <= commonAlignment(*BaseAlign, Offset),
491 "Undefined behavior: Memory reference address is misaligned", &I);
492 }
493}
494
495void Lint::visitLoadInst(LoadInst &I) {
496 visitMemoryReference(I, Loc: MemoryLocation::get(LI: &I), Align: I.getAlign(), Ty: I.getType(),
497 Flags: MemRef::Read);
498}
499
500void Lint::visitStoreInst(StoreInst &I) {
501 visitMemoryReference(I, Loc: MemoryLocation::get(SI: &I), Align: I.getAlign(),
502 Ty: I.getOperand(i_nocapture: 0)->getType(), Flags: MemRef::Write);
503}
504
505void Lint::visitAtomicCmpXchgInst(AtomicCmpXchgInst &I) {
506 visitMemoryReference(I, Loc: MemoryLocation::get(CXI: &I), Align: I.getAlign(),
507 Ty: I.getOperand(i_nocapture: 0)->getType(), Flags: MemRef::Write);
508}
509
510void Lint::visitAtomicRMWInst(AtomicRMWInst &I) {
511 visitMemoryReference(I, Loc: MemoryLocation::get(RMWI: &I), Align: I.getAlign(),
512 Ty: I.getOperand(i_nocapture: 0)->getType(), Flags: MemRef::Write);
513}
514
515void Lint::visitXor(BinaryOperator &I) {
516 Check(!isa<UndefValue>(I.getOperand(0)) || !isa<UndefValue>(I.getOperand(1)),
517 "Undefined result: xor(undef, undef)", &I);
518}
519
520void Lint::visitSub(BinaryOperator &I) {
521 Check(!isa<UndefValue>(I.getOperand(0)) || !isa<UndefValue>(I.getOperand(1)),
522 "Undefined result: sub(undef, undef)", &I);
523}
524
525void Lint::visitLShr(BinaryOperator &I) {
526 if (ConstantInt *CI = dyn_cast<ConstantInt>(Val: findValue(V: I.getOperand(i_nocapture: 1),
527 /*OffsetOk=*/false)))
528 Check(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
529 "Undefined result: Shift count out of range", &I);
530}
531
532void Lint::visitAShr(BinaryOperator &I) {
533 if (ConstantInt *CI =
534 dyn_cast<ConstantInt>(Val: findValue(V: I.getOperand(i_nocapture: 1), /*OffsetOk=*/false)))
535 Check(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
536 "Undefined result: Shift count out of range", &I);
537}
538
539void Lint::visitShl(BinaryOperator &I) {
540 if (ConstantInt *CI =
541 dyn_cast<ConstantInt>(Val: findValue(V: I.getOperand(i_nocapture: 1), /*OffsetOk=*/false)))
542 Check(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
543 "Undefined result: Shift count out of range", &I);
544}
545
546static bool isZero(Value *V, const DataLayout &DL, DominatorTree *DT,
547 AssumptionCache *AC) {
548 // Assume undef could be zero.
549 if (isa<UndefValue>(Val: V))
550 return true;
551
552 VectorType *VecTy = dyn_cast<VectorType>(Val: V->getType());
553 if (!VecTy) {
554 KnownBits Known = computeKnownBits(V, DL, AC, CxtI: dyn_cast<Instruction>(Val: V), DT);
555 return Known.isZero();
556 }
557
558 // Per-component check doesn't work with zeroinitializer
559 Constant *C = dyn_cast<Constant>(Val: V);
560 if (!C)
561 return false;
562
563 if (C->isZeroValue())
564 return true;
565
566 // For a vector, KnownZero will only be true if all values are zero, so check
567 // this per component
568 for (unsigned I = 0, N = cast<FixedVectorType>(Val: VecTy)->getNumElements();
569 I != N; ++I) {
570 Constant *Elem = C->getAggregateElement(Elt: I);
571 if (isa<UndefValue>(Val: Elem))
572 return true;
573
574 KnownBits Known = computeKnownBits(V: Elem, DL);
575 if (Known.isZero())
576 return true;
577 }
578
579 return false;
580}
581
582void Lint::visitSDiv(BinaryOperator &I) {
583 Check(!isZero(I.getOperand(1), I.getDataLayout(), DT, AC),
584 "Undefined behavior: Division by zero", &I);
585}
586
587void Lint::visitUDiv(BinaryOperator &I) {
588 Check(!isZero(I.getOperand(1), I.getDataLayout(), DT, AC),
589 "Undefined behavior: Division by zero", &I);
590}
591
592void Lint::visitSRem(BinaryOperator &I) {
593 Check(!isZero(I.getOperand(1), I.getDataLayout(), DT, AC),
594 "Undefined behavior: Division by zero", &I);
595}
596
597void Lint::visitURem(BinaryOperator &I) {
598 Check(!isZero(I.getOperand(1), I.getDataLayout(), DT, AC),
599 "Undefined behavior: Division by zero", &I);
600}
601
602void Lint::visitAllocaInst(AllocaInst &I) {
603 if (isa<ConstantInt>(Val: I.getArraySize()))
604 // This isn't undefined behavior, it's just an obvious pessimization.
605 Check(&I.getParent()->getParent()->getEntryBlock() == I.getParent(),
606 "Pessimization: Static alloca outside of entry block", &I);
607
608 // TODO: Check for an unusual size (MSB set?)
609}
610
611void Lint::visitVAArgInst(VAArgInst &I) {
612 visitMemoryReference(I, Loc: MemoryLocation::get(VI: &I), Align: std::nullopt, Ty: nullptr,
613 Flags: MemRef::Read | MemRef::Write);
614}
615
616void Lint::visitIndirectBrInst(IndirectBrInst &I) {
617 visitMemoryReference(I, Loc: MemoryLocation::getAfter(Ptr: I.getAddress()),
618 Align: std::nullopt, Ty: nullptr, Flags: MemRef::Branchee);
619
620 Check(I.getNumDestinations() != 0,
621 "Undefined behavior: indirectbr with no destinations", &I);
622}
623
624void Lint::visitExtractElementInst(ExtractElementInst &I) {
625 if (ConstantInt *CI = dyn_cast<ConstantInt>(Val: findValue(V: I.getIndexOperand(),
626 /*OffsetOk=*/false))) {
627 ElementCount EC = I.getVectorOperandType()->getElementCount();
628 Check(EC.isScalable() || CI->getValue().ult(EC.getFixedValue()),
629 "Undefined result: extractelement index out of range", &I);
630 }
631}
632
633void Lint::visitInsertElementInst(InsertElementInst &I) {
634 if (ConstantInt *CI = dyn_cast<ConstantInt>(Val: findValue(V: I.getOperand(i_nocapture: 2),
635 /*OffsetOk=*/false))) {
636 ElementCount EC = I.getType()->getElementCount();
637 Check(EC.isScalable() || CI->getValue().ult(EC.getFixedValue()),
638 "Undefined result: insertelement index out of range", &I);
639 }
640}
641
642void Lint::visitUnreachableInst(UnreachableInst &I) {
643 // This isn't undefined behavior, it's merely suspicious.
644 Check(&I == &I.getParent()->front() ||
645 std::prev(I.getIterator())->mayHaveSideEffects(),
646 "Unusual: unreachable immediately preceded by instruction without "
647 "side effects",
648 &I);
649}
650
651/// findValue - Look through bitcasts and simple memory reference patterns
652/// to identify an equivalent, but more informative, value. If OffsetOk
653/// is true, look through getelementptrs with non-zero offsets too.
654///
655/// Most analysis passes don't require this logic, because instcombine
656/// will simplify most of these kinds of things away. But it's a goal of
657/// this Lint pass to be useful even on non-optimized IR.
658Value *Lint::findValue(Value *V, bool OffsetOk) const {
659 SmallPtrSet<Value *, 4> Visited;
660 return findValueImpl(V, OffsetOk, Visited);
661}
662
663/// findValueImpl - Implementation helper for findValue.
664Value *Lint::findValueImpl(Value *V, bool OffsetOk,
665 SmallPtrSetImpl<Value *> &Visited) const {
666 // Detect self-referential values.
667 if (!Visited.insert(Ptr: V).second)
668 return PoisonValue::get(T: V->getType());
669
670 // TODO: Look through sext or zext cast, when the result is known to
671 // be interpreted as signed or unsigned, respectively.
672 // TODO: Look through eliminable cast pairs.
673 // TODO: Look through calls with unique return values.
674 // TODO: Look through vector insert/extract/shuffle.
675 V = OffsetOk ? getUnderlyingObject(V) : V->stripPointerCasts();
676 if (LoadInst *L = dyn_cast<LoadInst>(Val: V)) {
677 BasicBlock::iterator BBI = L->getIterator();
678 BasicBlock *BB = L->getParent();
679 SmallPtrSet<BasicBlock *, 4> VisitedBlocks;
680 BatchAAResults BatchAA(*AA);
681 for (;;) {
682 if (!VisitedBlocks.insert(Ptr: BB).second)
683 break;
684 if (Value *U =
685 FindAvailableLoadedValue(Load: L, ScanBB: BB, ScanFrom&: BBI, MaxInstsToScan: DefMaxInstsToScan, AA: &BatchAA))
686 return findValueImpl(V: U, OffsetOk, Visited);
687 if (BBI != BB->begin())
688 break;
689 BB = BB->getUniquePredecessor();
690 if (!BB)
691 break;
692 BBI = BB->end();
693 }
694 } else if (PHINode *PN = dyn_cast<PHINode>(Val: V)) {
695 if (Value *W = PN->hasConstantValue())
696 return findValueImpl(V: W, OffsetOk, Visited);
697 } else if (CastInst *CI = dyn_cast<CastInst>(Val: V)) {
698 if (CI->isNoopCast(DL: *DL))
699 return findValueImpl(V: CI->getOperand(i_nocapture: 0), OffsetOk, Visited);
700 } else if (ExtractValueInst *Ex = dyn_cast<ExtractValueInst>(Val: V)) {
701 if (Value *W =
702 FindInsertedValue(V: Ex->getAggregateOperand(), idx_range: Ex->getIndices()))
703 if (W != V)
704 return findValueImpl(V: W, OffsetOk, Visited);
705 } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Val: V)) {
706 // Same as above, but for ConstantExpr instead of Instruction.
707 if (Instruction::isCast(Opcode: CE->getOpcode())) {
708 if (CastInst::isNoopCast(Opcode: Instruction::CastOps(CE->getOpcode()),
709 SrcTy: CE->getOperand(i_nocapture: 0)->getType(), DstTy: CE->getType(),
710 DL: *DL))
711 return findValueImpl(V: CE->getOperand(i_nocapture: 0), OffsetOk, Visited);
712 }
713 }
714
715 // As a last resort, try SimplifyInstruction or constant folding.
716 if (Instruction *Inst = dyn_cast<Instruction>(Val: V)) {
717 if (Value *W = simplifyInstruction(I: Inst, Q: {*DL, TLI, DT, AC}))
718 return findValueImpl(V: W, OffsetOk, Visited);
719 } else if (auto *C = dyn_cast<Constant>(Val: V)) {
720 Value *W = ConstantFoldConstant(C, DL: *DL, TLI);
721 if (W != V)
722 return findValueImpl(V: W, OffsetOk, Visited);
723 }
724
725 return V;
726}
727
728PreservedAnalyses LintPass::run(Function &F, FunctionAnalysisManager &AM) {
729 auto *Mod = F.getParent();
730 auto *DL = &F.getDataLayout();
731 auto *AA = &AM.getResult<AAManager>(IR&: F);
732 auto *AC = &AM.getResult<AssumptionAnalysis>(IR&: F);
733 auto *DT = &AM.getResult<DominatorTreeAnalysis>(IR&: F);
734 auto *TLI = &AM.getResult<TargetLibraryAnalysis>(IR&: F);
735 Lint L(Mod, DL, AA, AC, DT, TLI);
736 L.visit(F);
737 dbgs() << L.MessagesStr.str();
738 if (AbortOnError && !L.MessagesStr.str().empty())
739 report_fatal_error(
740 reason: "linter found errors, aborting. (enabled by abort-on-error)", gen_crash_diag: false);
741 return PreservedAnalyses::all();
742}
743
744void LintPass::printPipeline(
745 raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) {
746 PassInfoMixin<LintPass>::printPipeline(OS, MapClassName2PassName);
747 if (AbortOnError)
748 OS << "<abort-on-error>";
749}
750
751//===----------------------------------------------------------------------===//
752// Implement the public interfaces to this file...
753//===----------------------------------------------------------------------===//
754
755/// lintFunction - Check a function for errors, printing messages on stderr.
756///
757void llvm::lintFunction(const Function &f, bool AbortOnError) {
758 Function &F = const_cast<Function &>(f);
759 assert(!F.isDeclaration() && "Cannot lint external functions");
760
761 FunctionAnalysisManager FAM;
762 FAM.registerPass(PassBuilder: [&] { return TargetLibraryAnalysis(); });
763 FAM.registerPass(PassBuilder: [&] { return DominatorTreeAnalysis(); });
764 FAM.registerPass(PassBuilder: [&] { return AssumptionAnalysis(); });
765 FAM.registerPass(PassBuilder: [&] {
766 AAManager AA;
767 AA.registerFunctionAnalysis<BasicAA>();
768 AA.registerFunctionAnalysis<ScopedNoAliasAA>();
769 AA.registerFunctionAnalysis<TypeBasedAA>();
770 return AA;
771 });
772 LintPass(AbortOnError).run(F, AM&: FAM);
773}
774
775/// lintModule - Check a module for errors, printing messages on stderr.
776///
777void llvm::lintModule(const Module &M, bool AbortOnError) {
778 for (const Function &F : M) {
779 if (!F.isDeclaration())
780 lintFunction(f: F, AbortOnError);
781 }
782}
783