cppgc: TSAN support for stack scanning

We cannot rely on the clang compiler to generate the trampoline entry
and the right mangling because `__attribute__((naked))` does not
prevent clang from generating TSAN function entry stub
(`__tsan_func_entry`). Even with `__attribute__((no_sanitize_thread)`
annotation clang generates the entry stub.

Upstream bug: https://bugs.llvm.org/show_bug.cgi?id=45400

Bug: chromium:1056170
Change-Id: I7a2063d7d205ee071e6a41ce4d9cb2d8d6423987
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2132797Reviewed-by: Omer Katz <omerkatz@chromium.org>
Reviewed-by: Anton Bikineev <bikineev@chromium.org>
Commit-Queue: Michael Lippautz <mlippautz@chromium.org>
Cr-Commit-Position: refs/heads/master@{#66976}

src/heap/cppgc/asm/x64/push_registers_clang.cc

@@ -5,12 +5,13 @@
 // Push all callee-saved registers to get them on the stack for conservative
 // stack scanning.
 //
-// Do not add any C code to the function. The function is naked to avoid
-// emitting a prologue/epilogue that could violate alginment computations.
-extern "C" __attribute__((naked, noinline)) void
-PushAllRegistersAndIterateStack(void* /* {Stack*} */,
-                                void* /* {StackVisitor*} */,
-                                void* /* {IterateStackCallback} */) {
+// We cannot rely on clang generating the function and right symbol mangling
+// as `__attribite__((naked))` does not prevent clang from generating TSAN
+// function entry stubs (`__tsan_func_entry`). Even with
+// `__attribute__((no_sanitize_thread)` annotation clang generates the entry
+// stub.
+// See https://bugs.llvm.org/show_bug.cgi?id=45400.
+
 // Do not depend on V8_TARGET_OS_* defines as some embedders may override the
 // GN toolchain (e.g. ChromeOS) and not provide them.
 // _WIN64 Defined as 1 when the compilation target is 64-bit ARM or x64.
@@ -21,32 +22,33 @@ PushAllRegistersAndIterateStack(void* /* {Stack*} */,
   // on the stack and we push 72 bytes which maintains 16-byte stack alignment
   // at the call.
   // Source: https://docs.microsoft.com/en-us/cpp/build/x64-calling-convention
-  asm volatile(
-      // rbp is callee-saved. Maintain proper frame pointer for debugging.
-      " push %rbp           \n"
-      " mov %rsp, %rbp      \n"
-      // Dummy for alignment.
-      " push $0xCDCDCD      \n"
-      " push %rsi           \n"
-      " push %rdi           \n"
-      " push %rbx           \n"
-      " push %r12           \n"
-      " push %r13           \n"
-      " push %r14           \n"
-      " push %r15           \n"
-      // Pass 1st parameter (rcx) unchanged (Stack*).
-      // Pass 2nd parameter (rdx) unchanged (StackVisitor*).
-      // Save 3rd parameter (r8; IterateStackCallback)
-      " mov %r8, %r9        \n"
-      // Pass 3rd parameter as rsp (stack pointer).
-      " mov %rsp, %r8       \n"
-      // Call the callback.
-      " call *%r9           \n"
-      // Pop the callee-saved registers.
-      " add $64, %rsp       \n"
-      // Restore rbp as it was used as frame pointer.
-      " pop %rbp            \n"
-      " ret                 \n");
+asm(".globl PushAllRegistersAndIterateStack             \n"
+    "PushAllRegistersAndIterateStack:                   \n"
+    // rbp is callee-saved. Maintain proper frame pointer for debugging.
+    "  push %rbp                                        \n"
+    "  mov %rsp, %rbp                                   \n"
+    // Dummy for alignment.
+    "  push $0xCDCDCD                                   \n"
+    "  push %rsi                                        \n"
+    "  push %rdi                                        \n"
+    "  push %rbx                                        \n"
+    "  push %r12                                        \n"
+    "  push %r13                                        \n"
+    "  push %r14                                        \n"
+    "  push %r15                                        \n"
+    // Pass 1st parameter (rcx) unchanged (Stack*).
+    // Pass 2nd parameter (rdx) unchanged (StackVisitor*).
+    // Save 3rd parameter (r8; IterateStackCallback)
+    "  mov %r8, %r9                                     \n"
+    // Pass 3rd parameter as rsp (stack pointer).
+    "  mov %rsp, %r8                                    \n"
+    // Call the callback.
+    "  call *%r9                                        \n"
+    // Pop the callee-saved registers.
+    "  add $64, %rsp                                    \n"
+    // Restore rbp as it was used as frame pointer.
+    "  pop %rbp                                         \n"
+    "  ret                                              \n");
 
 #else  // !_WIN64
 
@@ -54,30 +56,38 @@ PushAllRegistersAndIterateStack(void* /* {Stack*} */,
   // on the stack and we push 56 bytes which maintains 16-byte stack alignment
   // at the call.
   // Source: https://github.com/hjl-tools/x86-psABI/wiki/x86-64-psABI-1.0.pdf
-  asm volatile(
-      // rbp is callee-saved. Maintain proper frame pointer for debugging.
-      " push %rbp           \n"
-      " mov %rsp, %rbp      \n"
-      // Dummy for alignment.
-      " push $0xCDCDCD      \n"
-      " push %rbx           \n"
-      " push %r12           \n"
-      " push %r13           \n"
-      " push %r14           \n"
-      " push %r15           \n"
-      // Pass 1st parameter (rdi) unchanged (Stack*).
-      // Pass 2nd parameter (rsi) unchanged (StackVisitor*).
-      // Save 3rd parameter (rdx; IterateStackCallback)
-      " mov %rdx, %r8       \n"
-      // Pass 3rd parameter as rsp (stack pointer).
-      " mov %rsp, %rdx      \n"
-      // Call the callback.
-      " call *%r8           \n"
-      // Pop the callee-saved registers.
-      " add $48, %rsp       \n"
-      // Restore rbp as it was used as frame pointer.
-      " pop %rbp            \n"
-      " ret                 \n");
+asm(
+#ifdef __APPLE__
+    ".globl _PushAllRegistersAndIterateStack            \n"
+    ".private_extern _PushAllRegistersAndIterateStack   \n"
+    "_PushAllRegistersAndIterateStack:                  \n"
+#else   // !__APPLE__
+    ".globl PushAllRegistersAndIterateStack             \n"
+    ".hidden PushAllRegistersAndIterateStack            \n"
+    "PushAllRegistersAndIterateStack:                   \n"
+#endif  // !__APPLE__
+    // rbp is callee-saved. Maintain proper frame pointer for debugging.
+    "  push %rbp                                        \n"
+    "  mov %rsp, %rbp                                   \n"
+    // Dummy for alignment.
+    "  push $0xCDCDCD                                   \n"
+    "  push %rbx                                        \n"
+    "  push %r12                                        \n"
+    "  push %r13                                        \n"
+    "  push %r14                                        \n"
+    "  push %r15                                        \n"
+    // Pass 1st parameter (rdi) unchanged (Stack*).
+    // Pass 2nd parameter (rsi) unchanged (StackVisitor*).
+    // Save 3rd parameter (rdx; IterateStackCallback)
+    "  mov %rdx, %r8                                    \n"
+    // Pass 3rd parameter as rsp (stack pointer).
+    "  mov %rsp, %rdx                                   \n"
+    // Call the callback.
+    "  call *%r8                                        \n"
+    // Pop the callee-saved registers.
+    "  add $48, %rsp                                    \n"
+    // Restore rbp as it was used as frame pointer.
+    "  pop %rbp                                         \n"
+    "  ret                                              \n");
 
 #endif  // !_WIN64
-}

test/unittests/heap/cppgc/stack_unittest.cc

@@ -47,7 +47,6 @@ TEST_F(GCStackTest, IsOnStackForHeapValue) {
 }
 
 #ifdef CPPGC_SUPPORTS_CONSERVATIVE_STACK_SCAN
-#ifndef THREAD_SANITIZER
 
 namespace {
 
@@ -357,7 +356,6 @@ TEST_F(GCStackTest, StackAlignment) {
 #endif  // V8_OS_LINUX && (V8_HOST_ARCH_IA32 || V8_HOST_ARCH_X64)
 
 #endif  // CPPGC_SUPPORTS_CONSERVATIVE_STACK_SCAN
-#endif  // !THREAD_SANITIZER
 
 }  // namespace internal
 }  // namespace cppgc