[llvm-branch-commits] [llvm] [LLVM] Add a ConcurrencySanitizer LLVM pass (PR #225780)
Yaxun Liu via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Fri Oct 30 04:43:23 PDT 2026
================
@@ -0,0 +1,617 @@
+//===- ConcurrencySanitizer.cpp - watchpoint race detector ----------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements instrumentation for ConcurrencySanitizer, a sampling
+// data-race detector. Memory accesses are preceded by runtime probes while the
+// original operations remain unchanged.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Transforms/Instrumentation/ConcurrencySanitizer.h"
+#include "llvm/ADT/BitmaskEnum.h"
+#include "llvm/ADT/DenseSet.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/bit.h"
+#include "llvm/Analysis/CaptureTracking.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Module.h"
+#include "llvm/ProfileData/InstrProf.h"
+#include "llvm/Support/AMDGPUAddrSpace.h"
+#include "llvm/Support/AtomicOrdering.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/NVPTXAddrSpace.h"
+#include "llvm/TargetParser/Triple.h"
+#include "llvm/Transforms/Utils/EscapeEnumerator.h"
+#include "llvm/Transforms/Utils/Instrumentation.h"
+#include "llvm/Transforms/Utils/Local.h"
+#include "llvm/Transforms/Utils/ModuleUtils.h"
+#include <optional>
+
+using namespace llvm;
+
+#define DEBUG_TYPE "csan"
+
+static cl::opt<bool> ClInstrumentMemoryAccesses(
+ "csan-instrument-memory-accesses", cl::init(true),
+ cl::desc("Instrument memory accesses"), cl::Hidden);
+static cl::opt<bool>
+ ClInstrumentFuncEntryExit("csan-instrument-func-entry-exit", cl::init(true),
+ cl::desc("Instrument function entry and exit"),
+ cl::Hidden);
+static cl::opt<bool> ClHandleCxxExceptions(
+ "csan-handle-cxx-exceptions", cl::init(true),
+ cl::desc("Handle C++ exceptions (insert cleanup blocks for unwinding)"),
+ cl::Hidden);
+static cl::opt<bool> ClInstrumentAtomics("csan-instrument-atomics",
+ cl::init(true),
+ cl::desc("Instrument atomics"),
+ cl::Hidden);
+static cl::opt<bool> ClInstrumentMemIntrinsics(
+ "csan-instrument-memintrinsics", cl::init(true),
+ cl::desc("Instrument memintrinsics (memset/memcpy/memmove)"), cl::Hidden);
+static cl::opt<bool> ClDistinguishVolatile(
+ "csan-distinguish-volatile", cl::init(false),
+ cl::desc("Treat aligned volatile accesses up to 8 bytes as atomic"),
+ cl::Hidden);
+STATISTIC(NumInstrumentedReads, "Number of instrumented reads");
+STATISTIC(NumInstrumentedWrites, "Number of instrumented writes");
+STATISTIC(NumOmittedReadsBeforeWrite,
+ "Number of reads ignored due to following writes");
+
+static constexpr char kCsanModuleCtorName[] = "csan.module_ctor";
+static constexpr char kCsanInitName[] = "__csan_init";
+
+// Accesses sizes are powers of two: 1, 2, 4, 8, 16.
+static constexpr size_t kNumAccessSizes = 5;
+static constexpr StringLiteral kCsanReadNames[kNumAccessSizes] = {
+ "__csan_read1", "__csan_read2", "__csan_read4", "__csan_read8",
+ "__csan_read16"};
+static constexpr StringLiteral kCsanWriteNames[kNumAccessSizes] = {
+ "__csan_write1", "__csan_write2", "__csan_write4", "__csan_write8",
+ "__csan_write16"};
+static constexpr StringLiteral kCsanUnalignedReadNames[kNumAccessSizes] = {
+ "__csan_unaligned_read1", "__csan_unaligned_read2",
+ "__csan_unaligned_read4", "__csan_unaligned_read8",
+ "__csan_unaligned_read16"};
+static constexpr StringLiteral kCsanUnalignedWriteNames[kNumAccessSizes] = {
+ "__csan_unaligned_write1", "__csan_unaligned_write2",
+ "__csan_unaligned_write4", "__csan_unaligned_write8",
+ "__csan_unaligned_write16"};
+static constexpr StringLiteral kCsanCompoundRWNames[kNumAccessSizes] = {
+ "__csan_read_write1", "__csan_read_write2", "__csan_read_write4",
+ "__csan_read_write8", "__csan_read_write16"};
+static constexpr StringLiteral kCsanUnalignedCompoundRWNames[kNumAccessSizes] =
+ {"__csan_unaligned_read_write1", "__csan_unaligned_read_write2",
+ "__csan_unaligned_read_write4", "__csan_unaligned_read_write8",
+ "__csan_unaligned_read_write16"};
+
+namespace {
+
+/// Must match CSAN_ACCESS_* in compiler-rt/lib/csan/csan_defs.h.
+enum class AccessFlags : unsigned {
+ None = 0,
+ Atomic = 1u << 0,
+ Compound = 1u << 1,
+ LLVM_MARK_AS_BITMASK_ENUM(Compound),
+};
+
+static bool isAtomicMemoryAccess(const Instruction *I) {
+ std::optional<SyncScope::ID> SSID = getAtomicSyncScopeID(I);
+ if (!SSID)
+ return false;
+ if (isa<LoadInst>(I) || isa<StoreInst>(I))
+ return *SSID != SyncScope::SingleThread;
+ return true;
+}
+
+static Value *getCallbackAddress(IRBuilderBase &IRB, Value *Addr) {
+ return IRB.CreateAddrSpaceCast(Addr, IRB.getPtrTy());
+}
+
+static ConstantInt *createOrdering(IRBuilderBase &IRB, AtomicOrdering Ord) {
+ return IRB.getInt32(static_cast<uint32_t>(toCABI(Ord)));
+}
+
+static int getAccessSizeIndex(Type *Ty, const DataLayout &DL) {
+ assert(Ty->isSized());
+ if (Ty->isScalableTy())
+ return -1;
+ uint32_t TypeSize = DL.getTypeStoreSizeInBits(Ty);
+ if (TypeSize != 8 && TypeSize != 16 && TypeSize != 32 && TypeSize != 64 &&
+ TypeSize != 128)
+ return -1;
+ unsigned Idx = llvm::countr_zero(TypeSize / 8);
+ return static_cast<int>(Idx);
+}
+
+// A later write of the same pointer drops an earlier read in this window.
+static void selectMemoryAccesses(SmallVectorImpl<Instruction *> &Local,
+ SmallVectorImpl<Instruction *> &Out) {
+ DenseSet<Value *> WriteTargets;
+ for (Instruction *Inst : reverse(Local)) {
+ const bool IsWrite = isa<StoreInst>(Inst);
+ Value *Addr = getLoadStorePointerOperand(Inst);
+ if (!IsWrite && WriteTargets.contains(Addr)) {
----------------
yxsamliu wrote:
Nit: TSan keeps the read check here if `-tsan-distinguish-volatile` is on and either access is volatile. Was it left out on purpose? Without it, a plain read before a volatile write only gets an atomic write check, so the race can't be caught.
https://github.com/llvm/llvm-project/pull/225780
More information about the llvm-branch-commits
mailing list