[MSAN] Reintroduce libatomic load/store instrumentation

[guiand]

Summary

Have the front-end use the `nounwind` attribute on atomic libcalls.
This prevents us from seeing `invoke __atomic_load` in MSAN, which
is problematic as it has no successor for instrumentation to be added.

[guiand]

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3518

TODO: Remove

[eugenis]

__libatomic_load might come at the end of the function, with no succeeding BB

Not exactly. It may come at the end of a BB.

We should probably fix it in clang by adding nounwind to libatomic calls, and ignore invokes here.

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3508

Check if the function already exists.

3510

addVisibility(GlobalValue::HiddenVisibility);

[guiand]

Simplified by returning to the old implementation, but having libatomic calls made nounwind (so we never see them as invokes).

[guiand]

Rebased on master

[guiand]

Rebased on master (again)

[eugenis]

clang/lib/CodeGen/CGAtomic.cpp

315 ↗

(On Diff #284080)

This needs a clang test, and better move it to a separate change.

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3509

Better explicitly check that this an invoke and not call. Call always has a next instruction.

And move this check to the parent function to apply regular call instrumentation to the unrecognized call.

3511

It has a successor. Two successors, actually.
I'd rather say something like "invoke of __atomic_load".

[guiand]

Separated out the frontend change. Addressed other comments.

[guiand]

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3513

TODO: Remove this redudant call

[eugenis]

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3513

you've already checked it with assert(isa<CallInst>) - which could be changed to isTerminator(), and this could be simplified to NextIRB(CB.getNextNode()).

[guiand]

Addressed comments

[eugenis]

LGTM with 1 comment

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3578

Probably need the same check for atomic_store.

[guiand]

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3578

Although it would be good for consistency I don't think it's strictly needed since with atomic_store we don't need to use getNextNode()

[eugenis]

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

3578

ah, good point. LGTM

[guiand]

Committed as 97de0188dd5d845ff90c8ac779a2ea09688b17df

compiler-rt/test/msan/libatomic.c

+// RUN: %clang_msan -fsanitize-memory-track-origins=2 -latomic -DTEST_STORE -O0 %s -o %t && %run %t 2>&1
+// RUN: %clang_msan -fsanitize-memory-track-origins=0 -latomic -DTEST_LOAD -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s --check-prefix=CHECK
+// RUN: %clang_msan -fsanitize-memory-track-origins=2 -latomic -DTEST_LOAD -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-SHADOW
+
+// PPC has no libatomic
+// UNSUPPORTED: powerpc64-target-arch
+// UNSUPPORTED: powerpc64le-target-arch
+
+#include <sanitizer/msan_interface.h>
+#include <stdatomic.h>
+
+typedef struct __attribute((packed)) {
+  uint8_t val[3];
+} i24;
+
+void copy(i24 *dst, i24 *src);
+
+int main() {
+  i24 uninit;
+  i24 init = {0};
+
+  __msan_check_mem_is_initialized(&init, 3);
+  copy(&init, &uninit);
+  __msan_check_mem_is_initialized(&init, 3);
+}
+
+void copy(i24 *dst, i24 *src) {
+#ifdef TEST_LOAD
+  __atomic_load(src, dst, __ATOMIC_RELAXED);
+
+  // CHECK: MemorySanitizer: use-of-uninitialized-value
+  // CHECK: #0 {{0x[a-f0-9]+}} in main{{.*}}libatomic.c:[[@LINE-8]]
+
+  // CHECK-SHADOW: Uninitialized value was stored to memory at
+  // CHECK-SHADOW: #0 {{0x[a-f0-9]+}} in copy{{.*}}libatomic.c:[[@LINE-6]]
+#endif
+#ifdef TEST_STORE
+  // Store always writes a clean shadow
+  __atomic_store(src, dst, __ATOMIC_RELAXED);
+#endif
+}

compiler-rt/test/msan/libatomic_load_exceptions.cpp

+// RUN: %clangxx_msan -fexceptions -fsanitize-memory-track-origins=2 -latomic -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-SHADOW
+
+// PPC has no libatomic
+// UNSUPPORTED: powerpc64-target-arch
+// UNSUPPORTED: powerpc64le-target-arch
+
+#include <sanitizer/msan_interface.h>
+#include <stdatomic.h>
+
+typedef struct __attribute((packed)) {
+  uint8_t val[3];
+} i24;
+
+void copy(i24 *dst, i24 *src);
+
+int main() {
+  i24 uninit;
+  i24 init = {0};
+
+  __msan_check_mem_is_initialized(&init, 3);
+  copy(&init, &uninit);
+  __msan_check_mem_is_initialized(&init, 3);
+}
+
+void copy(i24 *dst, i24 *src) {
+  try {
+    __atomic_load(src, dst, __ATOMIC_RELAXED);
+  } catch (...) {
+  }
+}
+
+// CHECK: MemorySanitizer: use-of-uninitialized-value
+// CHECK: #0 {{0x[a-f0-9]+}} in main{{.*}}libatomic_load_exceptions.cpp:[[@LINE-11]]
+
+// CHECK-SHADOW: Uninitialized value was stored to memory at
+// CHECK-SHADOW: #0 {{0x[a-f0-9]+}} in copy{{.*}}libatomic_load_exceptions.cpp:[[@LINE-9]]

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

private:
 
   /// Run-time helper that poisons stack on function entry.
   FunctionCallee MsanPoisonStackFn;
 
   /// Run-time helper that records a store (or any event) of an
   /// uninitialized value and returns an updated origin id encoding this info.
   FunctionCallee MsanChainOriginFn;
 
+  /// Run-time helper that paints an origin over a region.
+  FunctionCallee MsanSetOriginFn;
+
   /// MSan runtime replacements for memmove, memcpy and memset.
   FunctionCallee MemmoveFn, MemcpyFn, MemsetFn;
 
   /// KMSAN callback for task-local function argument shadow.
   StructType *MsanContextStateTy;
   FunctionCallee MsanGetContextStateFn;
 
   /// Functions for poisoning/unpoisoning local variables
void MemorySanitizer::initializeCallbacks(Module &M) {
   if (CallbacksInitialized)
     return;
 
   IRBuilder<> IRB(*C);
   // Initialize callbacks that are common for kernel and userspace
   // instrumentation.
   MsanChainOriginFn = M.getOrInsertFunction(
     "__msan_chain_origin", IRB.getInt32Ty(), IRB.getInt32Ty());
+  MsanSetOriginFn =
+      M.getOrInsertFunction("__msan_set_origin", IRB.getVoidTy(),
+                            IRB.getInt8PtrTy(), IntptrTy, IRB.getInt32Ty());
   MemmoveFn = M.getOrInsertFunction(
     "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
     IRB.getInt8PtrTy(), IntptrTy);
   MemcpyFn = M.getOrInsertFunction(
     "__msan_memcpy", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
     IntptrTy);
   MemsetFn = M.getOrInsertFunction(
     "__msan_memset", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt32Ty(),
switch (a) {
       case AtomicOrdering::AcquireRelease:
         return AtomicOrdering::AcquireRelease;
       case AtomicOrdering::SequentiallyConsistent:
         return AtomicOrdering::SequentiallyConsistent;
     }
     llvm_unreachable("Unknown ordering");
   }
 
+  Value *makeAddReleaseOrderingTable(IRBuilder<> &IRB) {
+    constexpr int NumOrderings = (int)AtomicOrderingCABI::seq_cst + 1;
+    uint32_t OrderingTable[NumOrderings] = {};
+
+    OrderingTable[(int)AtomicOrderingCABI::relaxed] =
+        OrderingTable[(int)AtomicOrderingCABI::release] =
+            (int)AtomicOrderingCABI::release;
+    OrderingTable[(int)AtomicOrderingCABI::consume] =
+        OrderingTable[(int)AtomicOrderingCABI::acquire] =
+            OrderingTable[(int)AtomicOrderingCABI::acq_rel] =
+                (int)AtomicOrderingCABI::acq_rel;
+    OrderingTable[(int)AtomicOrderingCABI::seq_cst] =
+        (int)AtomicOrderingCABI::seq_cst;
+
+    return ConstantDataVector::get(IRB.getContext(),
+                                   makeArrayRef(OrderingTable, NumOrderings));
+  }
+
   AtomicOrdering addAcquireOrdering(AtomicOrdering a) {
     switch (a) {
       case AtomicOrdering::NotAtomic:
         return AtomicOrdering::NotAtomic;
       case AtomicOrdering::Unordered:
       case AtomicOrdering::Monotonic:
       case AtomicOrdering::Acquire:
         return AtomicOrdering::Acquire;
       case AtomicOrdering::Release:
       case AtomicOrdering::AcquireRelease:
         return AtomicOrdering::AcquireRelease;
       case AtomicOrdering::SequentiallyConsistent:
         return AtomicOrdering::SequentiallyConsistent;
     }
     llvm_unreachable("Unknown ordering");
   }
 
+  Value *makeAddAcquireOrderingTable(IRBuilder<> &IRB) {
+    constexpr int NumOrderings = (int)AtomicOrderingCABI::seq_cst + 1;
+    uint32_t OrderingTable[NumOrderings] = {};
+
+    OrderingTable[(int)AtomicOrderingCABI::relaxed] =
+        OrderingTable[(int)AtomicOrderingCABI::acquire] =
+            OrderingTable[(int)AtomicOrderingCABI::consume] =
+                (int)AtomicOrderingCABI::acquire;
+    OrderingTable[(int)AtomicOrderingCABI::release] =
+        OrderingTable[(int)AtomicOrderingCABI::acq_rel] =
+            (int)AtomicOrderingCABI::acq_rel;
+    OrderingTable[(int)AtomicOrderingCABI::seq_cst] =
+        (int)AtomicOrderingCABI::seq_cst;
+
+    return ConstantDataVector::get(IRB.getContext(),
+                                   makeArrayRef(OrderingTable, NumOrderings));
+  }
+
   // ------------------- Visitors.
   using InstVisitor<MemorySanitizerVisitor>::visit;
   void visit(Instruction &I) {
     if (!I.getMetadata("nosanitize"))
       InstVisitor<MemorySanitizerVisitor>::visit(I);
   }
 
   /// Instrument LoadInst
void visitIntrinsicInst(IntrinsicInst &I) {
 
     default:
       if (!handleUnknownIntrinsic(I))
         visitInstruction(I);
       break;
     }
   }
 
+  void visitLibAtomicLoad(CallBase &CB) {
+    // Since we use getNextNode here, we can't have CB terminate the BB.
+    assert(isa<CallInst>(CB));
+
+    IRBuilder<> IRB(&CB);
+    Value *Size = CB.getArgOperand(0);
+    Value *SrcPtr = CB.getArgOperand(1);
+    Value *DstPtr = CB.getArgOperand(2);
+    Value *Ordering = CB.getArgOperand(3);
+    // Convert the call to have at least Acquire ordering to make sure
+    // the shadow operations aren't reordered before it.
+    Value *NewOrdering =
+        IRB.CreateExtractElement(makeAddAcquireOrderingTable(IRB), Ordering);
+    CB.setArgOperand(3, NewOrdering);
+
+    Instruction *InsPoint = CB.getNextNode();
+    assert(InsPoint);
+    if (!InsPoint) {
+      return;
+    }
+    IRBuilder<> NextIRB(InsPoint);
+    NextIRB.SetCurrentDebugLocation(CB.getDebugLoc());
+
+    Value *SrcShadowPtr, *SrcOriginPtr;
+    std::tie(SrcShadowPtr, SrcOriginPtr) =
+        getShadowOriginPtr(SrcPtr, NextIRB, NextIRB.getInt8Ty(), Align(1),
+                           /*isStore*/ false);
+    Value *DstShadowPtr =
+        getShadowOriginPtr(DstPtr, NextIRB, NextIRB.getInt8Ty(), Align(1),
+                           /*isStore*/ true)
+            .first;
+
+    NextIRB.CreateMemCpy(DstShadowPtr, Align(1), SrcShadowPtr, Align(1), Size);
+    if (MS.TrackOrigins) {
+      Value *SrcOrigin = NextIRB.CreateAlignedLoad(MS.OriginTy, SrcOriginPtr,
+                                                   kMinOriginAlignment);
+      Value *NewOrigin = updateOrigin(SrcOrigin, NextIRB);
+      NextIRB.CreateCall(MS.MsanSetOriginFn, {DstPtr, Size, NewOrigin});
+    }
+  }
+
+  void visitLibAtomicStore(CallBase &CB) {
+    IRBuilder<> IRB(&CB);
+    Value *Size = CB.getArgOperand(0);
+    Value *DstPtr = CB.getArgOperand(2);
+    Value *Ordering = CB.getArgOperand(3);
+    // Convert the call to have at least Release ordering to make sure
+    // the shadow operations aren't reordered after it.
+    Value *NewOrdering =
+        IRB.CreateExtractElement(makeAddReleaseOrderingTable(IRB), Ordering);
+    CB.setArgOperand(3, NewOrdering);
+
+    Value *DstShadowPtr =
+        getShadowOriginPtr(DstPtr, IRB, IRB.getInt8Ty(), Align(1),
+                           /*isStore*/ true)
+            .first;
+
+    // Atomic store always paints clean shadow/origin. See file header.
+    IRB.CreateMemSet(DstShadowPtr, getCleanShadow(IRB.getInt8Ty()), Size,
+                     Align(1));
+  }
+
   void visitCallBase(CallBase &CB) {
     assert(!CB.getMetadata("nosanitize"));
     if (CB.isInlineAsm()) {
       // For inline asm (either a call to asm function, or callbr instruction),
       // do the usual thing: check argument shadow and mark all outputs as
       // clean. Note that any side effects of the inline asm that are not
       // immediately visible in its constraints are not handled.
       if (ClHandleAsmConservative && MS.CompileKernel)
         visitAsmInstruction(CB);
       else
         visitInstruction(CB);
       return;
     }
+    LibFunc LF;
+    if (TLI->getLibFunc(CB, LF)) {
+      // libatomic.a functions need to have special handling because there isn't
+      // a good way to intercept them or compile the library with
+      // instrumentation.
+      switch (LF) {
+      case LibFunc_atomic_load:
+        if (!isa<CallInst>(CB)) {
+          llvm::errs() << "MSAN -- cannot instrument invoke of libatomic load."
+                          "Ignoring!\n";
+          break;
+        }
+        visitLibAtomicLoad(CB);
+        return;
+      case LibFunc_atomic_store:
+        visitLibAtomicStore(CB);
+        return;
+      default:
+        break;
+      }
+    }
+
     if (auto *Call = dyn_cast<CallInst>(&CB)) {
       assert(!isa<IntrinsicInst>(Call) && "intrinsics are handled elsewhere");
 
       // We are going to insert code that relies on the fact that the callee
       // will become a non-readonly function after it is instrumented by us. To
       // prevent this code from being optimized out, mark that function
       // non-readonly in advance.
       AttrBuilder B;

llvm/test/Instrumentation/MemorySanitizer/libatomic.ll

+; RUN: opt < %s -msan-check-access-address=0 -S -passes=msan 2>&1 | FileCheck %s
+; RUN: opt < %s -msan-check-access-address=0 -msan-track-origins=2 -S -passes=msan 2>&1 | FileCheck %s -check-prefixes=CHECK,CHECK-ORIGIN
+; RUN: opt < %s -msan -msan-check-access-address=0 -S | FileCheck %s
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+declare void @__atomic_load(i64, i8*, i8*, i32)
+declare void @__atomic_store(i64, i8*, i8*, i32)
+
+define i24 @odd_sized_load(i24* %ptr) sanitize_memory {
+; CHECK: @odd_sized_load(i24* {{.*}}[[PTR:%.+]])
+; CHECK: [[VAL_PTR:%.*]] = alloca i24, align 1
+; CHECK-ORIGIN: @__msan_set_alloca_origin
+; CHECK: [[VAL_PTR_I8:%.*]] = bitcast i24* [[VAL_PTR]] to i8*
+; CHECK: [[PTR_I8:%.*]] = bitcast i24* [[PTR]] to i8*
+; CHECK: call void @__atomic_load(i64 3, i8* [[PTR_I8]], i8* [[VAL_PTR_I8]], i32 2)
+
+; CHECK: ptrtoint i8* [[PTR_I8]]
+; CHECK: xor
+; CHECK: [[SPTR_I8:%.*]] = inttoptr
+; CHECK-ORIGIN: add
+; CHECK-ORIGIN: and
+; CHECK-ORIGIN: [[OPTR:%.*]] = inttoptr
+
+; CHECK: ptrtoint i8* [[VAL_PTR_I8]]
+; CHECK: xor
+; CHECK: [[VAL_SPTR_I8:%.*]] = inttoptr
+; CHECK-ORIGIN: add
+; CHECK-ORIGIN: and
+; CHECK-ORIGIN: [[VAL_OPTR:%.*]] = inttoptr
+
+; CHECK: call void @llvm.memcpy{{.*}}(i8* align 1 [[VAL_SPTR_I8]], i8* align 1 [[SPTR_I8]], i64 3
+
+; CHECK-ORIGIN: [[ARG_ORIGIN:%.*]] = load i32, i32* [[OPTR]]
+; CHECK-ORIGIN: [[VAL_ORIGIN:%.*]] = call i32 @__msan_chain_origin(i32 [[ARG_ORIGIN]])
+; CHECK-ORIGIN: call void @__msan_set_origin(i8* [[VAL_PTR_I8]], i64 3, i32 [[VAL_ORIGIN]])
+
+; CHECK: [[VAL:%.*]] = load i24, i24* [[VAL_PTR]]
+; CHECK: ret i24 [[VAL]]
+  %val_ptr = alloca i24, align 1
+  %val_ptr_i8 = bitcast i24* %val_ptr to i8*
+  %ptr_i8 = bitcast i24* %ptr to i8*
+  call void @__atomic_load(i64 3, i8* %ptr_i8, i8* %val_ptr_i8, i32 0)
+  %val = load i24, i24* %val_ptr
+  ret i24 %val
+}
+
+define void @odd_sized_store(i24* %ptr, i24 %val) sanitize_memory {
+; CHECK: @odd_sized_store(i24* {{.*}}[[PTR:%.+]], i24 {{.*}}[[VAL:%.+]])
+; CHECK: [[VAL_PTR:%.*]] = alloca i24, align 1
+; CHECK: store i24 [[VAL]], i24* [[VAL_PTR]]
+; CHECK: [[VAL_PTR_I8:%.*]] = bitcast i24* [[VAL_PTR]] to i8*
+; CHECK: [[PTR_I8:%.*]] = bitcast i24* [[PTR]] to i8*
+
+; CHECK: ptrtoint i8* [[PTR_I8]]
+; CHECK: xor
+; CHECK: [[SPTR_I8:%.*]] = inttoptr
+; CHECK: call void @llvm.memset{{.*}}(i8* align 1 [[SPTR_I8]], i8 0, i64 3
+
+; CHECK: call void @__atomic_store(i64 3, i8* [[VAL_PTR_I8]], i8* [[PTR_I8]], i32 3)
+; CHECK: ret void
+  %val_ptr = alloca i24, align 1
+  store i24 %val, i24* %val_ptr
+  %val_ptr_i8 = bitcast i24* %val_ptr to i8*
+  %ptr_i8 = bitcast i24* %ptr to i8*
+  call void @__atomic_store(i64 3, i8* %val_ptr_i8, i8* %ptr_i8, i32 0)
+  ret void
+}
+