Vendor import of compiler-rt trunk r321017:vendor/compiler-rt/compiler-rt-trunk-r321017

https://llvm.org/svn/llvm-project/compiler-rt/trunk@321017

33 files changed, 1003 insertions, 698 deletions

diff --git a/lib/tsan/CMakeLists.txt b/lib/tsan/CMakeLists.txt
index 193158c54e62..3697ecb21381 100644
--- a/lib/tsan/CMakeLists.txt
+++ b/lib/tsan/CMakeLists.txt
@@ -93,21 +93,15 @@ set(TSAN_HEADERS
   rtl/tsan_symbolize.h
   rtl/tsan_sync.h
   rtl/tsan_trace.h
-  rtl/tsan_update_shadow_word_inl.h
-  rtl/tsan_vector.h)
+  rtl/tsan_update_shadow_word_inl.h)
 
 set(TSAN_RUNTIME_LIBRARIES)
 add_compiler_rt_component(tsan)
 
 if(APPLE)
-  set(TSAN_ASM_SOURCES rtl/tsan_rtl_amd64.S rtl/tsan_rtl_aarch64.S)
-  # Xcode will try to compile this file as C ('clang -x c'), and that will fail.
-  if (${CMAKE_GENERATOR} STREQUAL "Xcode")
-    enable_language(ASM)
-  else()
-    # Pass ASM file directly to the C++ compiler.
-    set_source_files_properties(${TSAN_ASM_SOURCES} PROPERTIES LANGUAGE C)
-  endif()
+  add_asm_sources(TSAN_ASM_SOURCES rtl/tsan_rtl_amd64.S rtl/tsan_rtl_aarch64.S)
+
+  set(TSAN_LINK_LIBS ${SANITIZER_COMMON_LINK_LIBS})
 
   add_weak_symbols("ubsan" WEAK_SYMBOL_LINK_FLAGS)
   add_weak_symbols("sanitizer_common" WEAK_SYMBOL_LINK_FLAGS)
@@ -122,7 +116,8 @@ if(APPLE)
                 RTSanitizerCommonLibc
                 RTUbsan
     CFLAGS ${TSAN_RTL_CFLAGS}
-    LINK_FLAGS ${WEAK_SYMBOL_LINK_FLAGS}
+    LINK_FLAGS ${SANITIZER_COMMON_LINK_FLAGS} ${WEAK_SYMBOL_LINK_FLAGS}
+    LINK_LIBS ${TSAN_LINK_LIBS}
     PARENT_TARGET tsan)
   add_compiler_rt_object_libraries(RTTsan_dynamic
     OS ${TSAN_SUPPORTED_OS}
@@ -142,10 +137,7 @@ if(APPLE)
 else()
   foreach(arch ${TSAN_SUPPORTED_ARCH})
     if(arch STREQUAL "x86_64")
-      set(TSAN_ASM_SOURCES rtl/tsan_rtl_amd64.S)
-      # Pass ASM file directly to the C++ compiler.
-      set_source_files_properties(${TSAN_ASM_SOURCES} PROPERTIES
-        LANGUAGE C)
+      add_asm_sources(TSAN_ASM_SOURCES rtl/tsan_rtl_amd64.S)
       # Sanity check for Go runtime.
       set(BUILDGO_SCRIPT ${CMAKE_CURRENT_SOURCE_DIR}/go/buildgo.sh)
       add_custom_target(GotsanRuntimeCheck
@@ -156,20 +148,11 @@ else()
         COMMENT "Checking TSan Go runtime..."
         VERBATIM)
     elseif(arch STREQUAL "aarch64")
-      set(TSAN_ASM_SOURCES rtl/tsan_rtl_aarch64.S)
-      # Pass ASM file directly to the C++ compiler.
-      set_source_files_properties(${TSAN_ASM_SOURCES} PROPERTIES
-        LANGUAGE C)
-   elseif(arch MATCHES "powerpc64|powerpc64le")
-     set(TSAN_ASM_SOURCES rtl/tsan_rtl_ppc64.S)
-     # Pass ASM file directly to the C++ compiler.
-     set_source_files_properties(${TSAN_ASM_SOURCES} PROPERTIES
-       LANGUAGE C)
+      add_asm_sources(TSAN_ASM_SOURCES rtl/tsan_rtl_aarch64.S)
+    elseif(arch MATCHES "powerpc64|powerpc64le")
+      add_asm_sources(TSAN_ASM_SOURCES rtl/tsan_rtl_ppc64.S)
     elseif(arch MATCHES "mips64|mips64le")
-     set(TSAN_ASM_SOURCES rtl/tsan_rtl_mips64.S)
-     # Pass ASM file directly to the C++ compiler.
-     set_source_files_properties(${TSAN_ASM_SOURCES} PROPERTIES
-       LANGUAGE C)
+      add_asm_sources(TSAN_ASM_SOURCES rtl/tsan_rtl_mips64.S)
     else()
       set(TSAN_ASM_SOURCES)
     endif()
@@ -181,13 +164,15 @@ else()
               $<TARGET_OBJECTS:RTSanitizerCommon.${arch}>
               $<TARGET_OBJECTS:RTSanitizerCommonLibc.${arch}>
               $<TARGET_OBJECTS:RTUbsan.${arch}>
-      CFLAGS ${TSAN_RTL_CFLAGS})
+      CFLAGS ${TSAN_RTL_CFLAGS}
+      PARENT_TARGET tsan)
     add_compiler_rt_runtime(clang_rt.tsan_cxx
       STATIC
       ARCHS ${arch}
       SOURCES ${TSAN_CXX_SOURCES}
               $<TARGET_OBJECTS:RTUbsan_cxx.${arch}>
-      CFLAGS ${TSAN_RTL_CFLAGS})
+      CFLAGS ${TSAN_RTL_CFLAGS}
+      PARENT_TARGET tsan)
     list(APPEND TSAN_RUNTIME_LIBRARIES clang_rt.tsan-${arch}
                                        clang_rt.tsan_cxx-${arch})
     add_sanitizer_rt_symbols(clang_rt.tsan
@@ -218,8 +203,9 @@ if(COMPILER_RT_HAS_SYSROOT_FLAG AND NOT CMAKE_SYSTEM_NAME MATCHES "FreeBSD")
 endif()
 
 # Build libcxx instrumented with TSan.
-if(COMPILER_RT_HAS_LIBCXX_SOURCES AND
-   COMPILER_RT_TEST_COMPILER_ID STREQUAL "Clang")
+if(COMPILER_RT_LIBCXX_PATH AND
+   COMPILER_RT_TEST_COMPILER_ID STREQUAL "Clang" AND
+   NOT ANDROID)
   set(libcxx_tsan_deps)
   foreach(arch ${TSAN_SUPPORTED_ARCH})
     get_target_flags_for_arch(${arch} TARGET_CFLAGS)
diff --git a/lib/tsan/check_analyze.sh b/lib/tsan/check_analyze.sh
index 54dd1b0232dc..9b5abc317fbc 100755
--- a/lib/tsan/check_analyze.sh
+++ b/lib/tsan/check_analyze.sh
@@ -2,6 +2,14 @@
 #
 # Script that checks that critical functions in TSan runtime have correct number
 # of push/pop/rsp instructions to verify that runtime is efficient enough.
+#
+# This test can fail when backend code generation changes the output for various
+# tsan interceptors. When such a change happens, you can ensure that the
+# performance has not regressed by running the following benchmarks before and
+# after the breaking change to verify that the values in this file are safe to
+# update:
+# ./projects/compiler-rt/lib/tsan/tests/rtl/TsanRtlTest
+#   --gtest_also_run_disabled_tests --gtest_filter=DISABLED_BENCH.Mop*
 
 set -u
 
@@ -35,7 +43,7 @@ done
 for f in read1 read2 read4 read8; do
   check $f rsp 1
   check $f push 3
-  check $f pop 3
+  check $f pop 18
 done
 
 for f in func_entry func_exit; do
diff --git a/lib/tsan/dd/CMakeLists.txt b/lib/tsan/dd/CMakeLists.txt
index bcff35f20b36..07fc300535cd 100644
--- a/lib/tsan/dd/CMakeLists.txt
+++ b/lib/tsan/dd/CMakeLists.txt
@@ -10,7 +10,8 @@ set(DD_SOURCES
   dd_interceptors.cc
 )
 
-set(DD_LINKLIBS)
+set(DD_LINKLIBS ${SANITIZER_CXX_ABI_LIBRARY} ${SANITIZER_COMMON_LINK_LIBS})
+
 append_list_if(COMPILER_RT_HAS_LIBDL dl DD_LINKLIBS)
 append_list_if(COMPILER_RT_HAS_LIBRT rt DD_LINKLIBS)
 append_list_if(COMPILER_RT_HAS_LIBPTHREAD pthread DD_LINKLIBS)
@@ -40,6 +41,7 @@ if(CAN_TARGET_x86_64 AND UNIX AND NOT APPLE AND NOT ANDROID)
             $<TARGET_OBJECTS:RTInterception.${arch}>
             $<TARGET_OBJECTS:RTSanitizerCommon.${arch}>
             $<TARGET_OBJECTS:RTSanitizerCommonLibc.${arch}>
+    LINK_FLAGS ${SANITIZER_COMMON_LINK_FLAGS}
     LINK_LIBS ${DD_LINKLIBS}
     PARENT_TARGET dd)
 endif()
diff --git a/lib/tsan/go/buildgo.sh b/lib/tsan/go/buildgo.sh
index 617dd9e11d27..62ff0fc38ba9 100755
--- a/lib/tsan/go/buildgo.sh
+++ b/lib/tsan/go/buildgo.sh
@@ -24,6 +24,7 @@ SRCS="
 	../../sanitizer_common/sanitizer_common.cc
 	../../sanitizer_common/sanitizer_common_libcdep.cc
 	../../sanitizer_common/sanitizer_deadlock_detector2.cc
+	../../sanitizer_common/sanitizer_file.cc
 	../../sanitizer_common/sanitizer_flag_parser.cc
 	../../sanitizer_common/sanitizer_flags.cc
 	../../sanitizer_common/sanitizer_libc.cc
@@ -67,6 +68,21 @@ elif [ "`uname -a | grep FreeBSD`" != "" ]; then
 		../../sanitizer_common/sanitizer_linux_libcdep.cc
 		../../sanitizer_common/sanitizer_stoptheworld_linux_libcdep.cc
 	"
+elif [ "`uname -a | grep NetBSD`" != "" ]; then
+	SUFFIX="netbsd_amd64"
+	OSCFLAGS="-fno-strict-aliasing -fPIC -Werror"
+	OSLDFLAGS="-lpthread -fPIC -fpie"
+	SRCS="
+		$SRCS
+		../rtl/tsan_platform_linux.cc
+		../../sanitizer_common/sanitizer_posix.cc
+		../../sanitizer_common/sanitizer_posix_libcdep.cc
+		../../sanitizer_common/sanitizer_procmaps_common.cc
+		../../sanitizer_common/sanitizer_procmaps_freebsd.cc
+		../../sanitizer_common/sanitizer_linux.cc
+		../../sanitizer_common/sanitizer_linux_libcdep.cc
+		../../sanitizer_common/sanitizer_stoptheworld_linux_libcdep.cc
+	"
 elif [ "`uname -a | grep Darwin`" != "" ]; then
 	SUFFIX="darwin_amd64"
 	OSCFLAGS="-fPIC -Wno-unused-const-variable -Wno-unknown-warning-option -isysroot $(xcodebuild -version -sdk macosx Path) -mmacosx-version-min=10.7"
@@ -126,7 +142,7 @@ if [ "$SILENT" != "1" ]; then
 fi
 $CC $DIR/gotsan.cc -c -o $DIR/race_$SUFFIX.syso $FLAGS $CFLAGS
 
-$CC $OSCFLAGS test.c $DIR/race_$SUFFIX.syso -m64 -g -o $DIR/test $OSLDFLAGS
+$CC $OSCFLAGS test.c $DIR/race_$SUFFIX.syso -m64 -g -o $DIR/test $OSLDFLAGS $LDFLAGS
 
 export GORACE="exitcode=0 atexit_sleep_ms=0"
 if [ "$SILENT" != "1" ]; then
diff --git a/lib/tsan/rtl/tsan_fd.cc b/lib/tsan/rtl/tsan_fd.cc
index d84df4a6413c..f13a7432ec95 100644
--- a/lib/tsan/rtl/tsan_fd.cc
+++ b/lib/tsan/rtl/tsan_fd.cc
@@ -48,8 +48,8 @@ static bool bogusfd(int fd) {
 }
 
 static FdSync *allocsync(ThreadState *thr, uptr pc) {
-  FdSync *s = (FdSync*)user_alloc(thr, pc, sizeof(FdSync), kDefaultAlignment,
-      false);
+  FdSync *s = (FdSync*)user_alloc_internal(thr, pc, sizeof(FdSync),
+      kDefaultAlignment, false);
   atomic_store(&s->rc, 1, memory_order_relaxed);
   return s;
 }
@@ -79,7 +79,7 @@ static FdDesc *fddesc(ThreadState *thr, uptr pc, int fd) {
   if (l1 == 0) {
     uptr size = kTableSizeL2 * sizeof(FdDesc);
     // We need this to reside in user memory to properly catch races on it.
-    void *p = user_alloc(thr, pc, size, kDefaultAlignment, false);
+    void *p = user_alloc_internal(thr, pc, size, kDefaultAlignment, false);
     internal_memset(p, 0, size);
     MemoryResetRange(thr, (uptr)&fddesc, (uptr)p, size);
     if (atomic_compare_exchange_strong(pl1, &l1, (uptr)p, memory_order_acq_rel))
diff --git a/lib/tsan/rtl/tsan_interceptors.cc b/lib/tsan/rtl/tsan_interceptors.cc
index 001123f4941e..d92976ad6e9e 100644
--- a/lib/tsan/rtl/tsan_interceptors.cc
+++ b/lib/tsan/rtl/tsan_interceptors.cc
@@ -17,6 +17,7 @@
 #include "sanitizer_common/sanitizer_errno.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_linux.h"
+#include "sanitizer_common/sanitizer_platform_limits_netbsd.h"
 #include "sanitizer_common/sanitizer_platform_limits_posix.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "sanitizer_common/sanitizer_posix.h"
@@ -39,6 +40,21 @@ using namespace __tsan;  // NOLINT
 #define stderr __stderrp
 #endif
 
+#if SANITIZER_NETBSD
+#define dirfd(dirp) (*(int *)(dirp))
+#define fileno_unlocked fileno
+
+#if _LP64
+#define __sF_size 152
+#else
+#define __sF_size 88
+#endif
+
+#define stdout ((char*)&__sF + (__sF_size * 1))
+#define stderr ((char*)&__sF + (__sF_size * 2))
+
+#endif
+
 #if SANITIZER_ANDROID
 #define mallopt(a, b)
 #endif
@@ -49,11 +65,6 @@ const int kSigCount = 129;
 const int kSigCount = 65;
 #endif
 
-struct my_siginfo_t {
-  // The size is determined by looking at sizeof of real siginfo_t on linux.
-  u64 opaque[128 / sizeof(u64)];
-};
-
 #ifdef __mips__
 struct ucontext_t {
   u64 opaque[768 / sizeof(u64) + 1];
@@ -84,19 +95,25 @@ DECLARE_REAL_AND_INTERCEPTOR(void, free, void *ptr)
 extern "C" void *pthread_self();
 extern "C" void _exit(int status);
 extern "C" int fileno_unlocked(void *stream);
+#if !SANITIZER_NETBSD
 extern "C" int dirfd(void *dirp);
-#if !SANITIZER_FREEBSD && !SANITIZER_ANDROID
+#endif
+#if !SANITIZER_FREEBSD && !SANITIZER_ANDROID && !SANITIZER_NETBSD
 extern "C" int mallopt(int param, int value);
 #endif
+#if SANITIZER_NETBSD
+extern __sanitizer_FILE __sF[];
+#else
 extern __sanitizer_FILE *stdout, *stderr;
-#if !SANITIZER_FREEBSD && !SANITIZER_MAC
+#endif
+#if !SANITIZER_FREEBSD && !SANITIZER_MAC && !SANITIZER_NETBSD
 const int PTHREAD_MUTEX_RECURSIVE = 1;
 const int PTHREAD_MUTEX_RECURSIVE_NP = 1;
 #else
 const int PTHREAD_MUTEX_RECURSIVE = 2;
 const int PTHREAD_MUTEX_RECURSIVE_NP = 2;
 #endif
-#if !SANITIZER_FREEBSD && !SANITIZER_MAC
+#if !SANITIZER_FREEBSD && !SANITIZER_MAC && !SANITIZER_NETBSD
 const int EPOLL_CTL_ADD = 1;
 #endif
 const int SIGILL = 4;
@@ -105,7 +122,7 @@ const int SIGFPE = 8;
 const int SIGSEGV = 11;
 const int SIGPIPE = 13;
 const int SIGTERM = 15;
-#if defined(__mips__) || SANITIZER_FREEBSD || SANITIZER_MAC
+#if defined(__mips__) || SANITIZER_FREEBSD || SANITIZER_MAC || SANITIZER_NETBSD
 const int SIGBUS = 10;
 const int SIGSYS = 12;
 #else
@@ -113,7 +130,9 @@ const int SIGBUS = 7;
 const int SIGSYS = 31;
 #endif
 void *const MAP_FAILED = (void*)-1;
-#if !SANITIZER_MAC
+#if SANITIZER_NETBSD
+const int PTHREAD_BARRIER_SERIAL_THREAD = 1234567;
+#elif !SANITIZER_MAC
 const int PTHREAD_BARRIER_SERIAL_THREAD = -1;
 #endif
 const int MAP_FIXED = 0x10;
@@ -125,48 +144,7 @@ typedef long long_t;  // NOLINT
 # define F_TLOCK 2      /* Test and lock a region for exclusive use.  */
 # define F_TEST  3      /* Test a region for other processes locks.  */
 
-typedef void (*sighandler_t)(int sig);
-typedef void (*sigactionhandler_t)(int sig, my_siginfo_t *siginfo, void *uctx);
-
-#if SANITIZER_ANDROID
-struct sigaction_t {
-  u32 sa_flags;
-  union {
-    sighandler_t sa_handler;
-    sigactionhandler_t sa_sigaction;
-  };
-  __sanitizer_sigset_t sa_mask;
-  void (*sa_restorer)();
-};
-#else
-struct sigaction_t {
-#ifdef __mips__
-  u32 sa_flags;
-#endif
-  union {
-    sighandler_t sa_handler;
-    sigactionhandler_t sa_sigaction;
-  };
-#if SANITIZER_FREEBSD
-  int sa_flags;
-  __sanitizer_sigset_t sa_mask;
-#elif SANITIZER_MAC
-  __sanitizer_sigset_t sa_mask;
-  int sa_flags;
-#else
-  __sanitizer_sigset_t sa_mask;
-#ifndef __mips__
-  int sa_flags;
-#endif
-  void (*sa_restorer)();
-#endif
-};
-#endif
-
-const sighandler_t SIG_DFL = (sighandler_t)0;
-const sighandler_t SIG_IGN = (sighandler_t)1;
-const sighandler_t SIG_ERR = (sighandler_t)-1;
-#if SANITIZER_FREEBSD || SANITIZER_MAC
+#if SANITIZER_FREEBSD || SANITIZER_MAC || SANITIZER_NETBSD
 const int SA_SIGINFO = 0x40;
 const int SIG_SETMASK = 3;
 #elif defined(__mips__)
@@ -180,13 +158,11 @@ const int SIG_SETMASK = 2;
 #define COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED \
   (!cur_thread()->is_inited)
 
-static sigaction_t sigactions[kSigCount];
-
 namespace __tsan {
 struct SignalDesc {
   bool armed;
   bool sigaction;
-  my_siginfo_t siginfo;
+  __sanitizer_siginfo siginfo;
   ucontext_t ctx;
 };
 
@@ -200,11 +176,41 @@ struct ThreadSignalContext {
   __sanitizer_sigset_t oldset;
 };
 
-// The object is 64-byte aligned, because we want hot data to be located in
-// a single cache line if possible (it's accessed in every interceptor).
-static ALIGNED(64) char libignore_placeholder[sizeof(LibIgnore)];
+// The sole reason tsan wraps atexit callbacks is to establish synchronization
+// between callback setup and callback execution.
+struct AtExitCtx {
+  void (*f)();
+  void *arg;
+};
+
+// InterceptorContext holds all global data required for interceptors.
+// It's explicitly constructed in InitializeInterceptors with placement new
+// and is never destroyed. This allows usage of members with non-trivial
+// constructors and destructors.
+struct InterceptorContext {
+  // The object is 64-byte aligned, because we want hot data to be located
+  // in a single cache line if possible (it's accessed in every interceptor).
+  ALIGNED(64) LibIgnore libignore;
+  __sanitizer_sigaction sigactions[kSigCount];
+#if !SANITIZER_MAC && !SANITIZER_NETBSD
+  unsigned finalize_key;
+#endif
+
+  BlockingMutex atexit_mu;
+  Vector<struct AtExitCtx *> AtExitStack;
+
+  InterceptorContext()
+      : libignore(LINKER_INITIALIZED), AtExitStack() {
+  }
+};
+
+static ALIGNED(64) char interceptor_placeholder[sizeof(InterceptorContext)];
+InterceptorContext *interceptor_ctx() {
+  return reinterpret_cast<InterceptorContext*>(&interceptor_placeholder[0]);
+}
+
 LibIgnore *libignore() {
-  return reinterpret_cast<LibIgnore*>(&libignore_placeholder[0]);
+  return &interceptor_ctx()->libignore;
 }
 
 void InitializeLibIgnore() {
@@ -232,10 +238,6 @@ static ThreadSignalContext *SigCtx(ThreadState *thr) {
   return ctx;
 }
 
-#if !SANITIZER_MAC
-static unsigned g_thread_finalize_key;
-#endif
-
 ScopedInterceptor::ScopedInterceptor(ThreadState *thr, const char *fname,
                                      uptr pc)
     : thr_(thr), pc_(pc), in_ignored_lib_(false), ignoring_(false) {
@@ -284,8 +286,18 @@ void ScopedInterceptor::DisableIgnores() {
 #define TSAN_INTERCEPT(func) INTERCEPT_FUNCTION(func)
 #if SANITIZER_FREEBSD
 # define TSAN_INTERCEPT_VER(func, ver) INTERCEPT_FUNCTION(func)
+# define TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(func)
+# define TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS_THR(func)
+#elif SANITIZER_NETBSD
+# define TSAN_INTERCEPT_VER(func, ver) INTERCEPT_FUNCTION(func)
+# define TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(func) \
+         INTERCEPT_FUNCTION(__libc_##func)
+# define TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS_THR(func) \
+         INTERCEPT_FUNCTION(__libc_thr_##func)
 #else
 # define TSAN_INTERCEPT_VER(func, ver) INTERCEPT_FUNCTION_VER(func, ver)
+# define TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(func)
+# define TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS_THR(func)
 #endif
 
 #define READ_STRING_OF_LEN(thr, pc, s, len, n)                 \
@@ -346,17 +358,30 @@ TSAN_INTERCEPTOR(int, nanosleep, void *req, void *rem) {
   return res;
 }
 
-// The sole reason tsan wraps atexit callbacks is to establish synchronization
-// between callback setup and callback execution.
-struct AtExitCtx {
-  void (*f)();
-  void *arg;
-};
+TSAN_INTERCEPTOR(int, pause, int fake) {
+  SCOPED_TSAN_INTERCEPTOR(pause, fake);
+  return BLOCK_REAL(pause)(fake);
+}
 
-static void at_exit_wrapper(void *arg) {
-  ThreadState *thr = cur_thread();
-  uptr pc = 0;
-  Acquire(thr, pc, (uptr)arg);
+static void at_exit_wrapper() {
+  AtExitCtx *ctx;
+  {
+    // Ensure thread-safety.
+    BlockingMutexLock l(&interceptor_ctx()->atexit_mu);
+
+    // Pop AtExitCtx from the top of the stack of callback functions
+    uptr element = interceptor_ctx()->AtExitStack.Size() - 1;
+    ctx = interceptor_ctx()->AtExitStack[element];
+    interceptor_ctx()->AtExitStack.PopBack();
+  }
+
+  Acquire(cur_thread(), (uptr)0, (uptr)ctx);
+  ((void(*)())ctx->f)();
+  InternalFree(ctx);
+}
+
+static void cxa_at_exit_wrapper(void *arg) {
+  Acquire(cur_thread(), 0, (uptr)arg);
   AtExitCtx *ctx = (AtExitCtx*)arg;
   ((void(*)(void *arg))ctx->f)(ctx->arg);
   InternalFree(ctx);
@@ -367,7 +392,7 @@ static int setup_at_exit_wrapper(ThreadState *thr, uptr pc, void(*f)(),
 
 #if !SANITIZER_ANDROID
 TSAN_INTERCEPTOR(int, atexit, void (*f)()) {
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return 0;
   // We want to setup the atexit callback even if we are in ignored lib
   // or after fork.
@@ -377,7 +402,7 @@ TSAN_INTERCEPTOR(int, atexit, void (*f)()) {
 #endif
 
 TSAN_INTERCEPTOR(int, __cxa_atexit, void (*f)(void *a), void *arg, void *dso) {
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return 0;
   SCOPED_TSAN_INTERCEPTOR(__cxa_atexit, f, arg, dso);
   return setup_at_exit_wrapper(thr, pc, (void(*)())f, arg, dso);
@@ -392,12 +417,27 @@ static int setup_at_exit_wrapper(ThreadState *thr, uptr pc, void(*f)(),
   // Memory allocation in __cxa_atexit will race with free during exit,
   // because we do not see synchronization around atexit callback list.
   ThreadIgnoreBegin(thr, pc);
-  int res = REAL(__cxa_atexit)(at_exit_wrapper, ctx, dso);
+  int res;
+  if (!dso) {
+    // NetBSD does not preserve the 2nd argument if dso is equal to 0
+    // Store ctx in a local stack-like structure
+
+    // Ensure thread-safety.
+    BlockingMutexLock l(&interceptor_ctx()->atexit_mu);
+
+    res = REAL(__cxa_atexit)((void (*)(void *a))at_exit_wrapper, 0, 0);
+    // Push AtExitCtx on the top of the stack of callback functions
+    if (!res) {
+      interceptor_ctx()->AtExitStack.PushBack(ctx);
+    }
+  } else {
+    res = REAL(__cxa_atexit)(cxa_at_exit_wrapper, ctx, dso);
+  }
   ThreadIgnoreEnd(thr, pc);
   return res;
 }
 
-#if !SANITIZER_MAC
+#if !SANITIZER_MAC && !SANITIZER_NETBSD
 static void on_exit_wrapper(int status, void *arg) {
   ThreadState *thr = cur_thread();
   uptr pc = 0;
@@ -408,7 +448,7 @@ static void on_exit_wrapper(int status, void *arg) {
 }
 
 TSAN_INTERCEPTOR(int, on_exit, void(*f)(int, void*), void *arg) {
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return 0;
   SCOPED_TSAN_INTERCEPTOR(on_exit, f, arg);
   AtExitCtx *ctx = (AtExitCtx*)InternalAlloc(sizeof(AtExitCtx));
@@ -422,6 +462,9 @@ TSAN_INTERCEPTOR(int, on_exit, void(*f)(int, void*), void *arg) {
   ThreadIgnoreEnd(thr, pc);
   return res;
 }
+#define TSAN_MAYBE_INTERCEPT_ON_EXIT TSAN_INTERCEPT(on_exit)
+#else
+#define TSAN_MAYBE_INTERCEPT_ON_EXIT
 #endif
 
 // Cleanup old bufs.
@@ -461,13 +504,15 @@ static void LongJmp(ThreadState *thr, uptr *env) {
   uptr mangled_sp = env[0];
 #elif SANITIZER_FREEBSD
   uptr mangled_sp = env[2];
+#elif SANITIZER_NETBSD
+  uptr mangled_sp = env[6];
 #elif SANITIZER_MAC
 # ifdef __aarch64__
     uptr mangled_sp = env[13];
 # else
     uptr mangled_sp = env[2];
 # endif
-#elif defined(SANITIZER_LINUX)
+#elif SANITIZER_LINUX
 # ifdef __aarch64__
   uptr mangled_sp = env[13];
 # elif defined(__mips64)
@@ -510,10 +555,29 @@ TSAN_INTERCEPTOR(int, setjmp, void *env);
 TSAN_INTERCEPTOR(int, _setjmp, void *env);
 TSAN_INTERCEPTOR(int, sigsetjmp, void *env);
 #else  // SANITIZER_MAC
+
+#if SANITIZER_NETBSD
+#define setjmp_symname __setjmp14
+#define sigsetjmp_symname __sigsetjmp14
+#else
+#define setjmp_symname setjmp
+#define sigsetjmp_symname sigsetjmp
+#endif
+
+#define TSAN_INTERCEPTOR_SETJMP_(x) __interceptor_ ## x
+#define TSAN_INTERCEPTOR_SETJMP__(x) TSAN_INTERCEPTOR_SETJMP_(x)
+#define TSAN_INTERCEPTOR_SETJMP TSAN_INTERCEPTOR_SETJMP__(setjmp_symname)
+#define TSAN_INTERCEPTOR_SIGSETJMP TSAN_INTERCEPTOR_SETJMP__(sigsetjmp_symname)
+
+#define TSAN_STRING_SETJMP_(x) # x
+#define TSAN_STRING_SETJMP__(x) TSAN_STRING_SETJMP_(x)
+#define TSAN_STRING_SETJMP TSAN_STRING_SETJMP__(setjmp_symname)
+#define TSAN_STRING_SIGSETJMP TSAN_STRING_SETJMP__(sigsetjmp_symname)
+
 // Not called.  Merely to satisfy TSAN_INTERCEPT().
 extern "C" SANITIZER_INTERFACE_ATTRIBUTE
-int __interceptor_setjmp(void *env);
-extern "C" int __interceptor_setjmp(void *env) {
+int TSAN_INTERCEPTOR_SETJMP(void *env);
+extern "C" int TSAN_INTERCEPTOR_SETJMP(void *env) {
   CHECK(0);
   return 0;
 }
@@ -527,51 +591,75 @@ extern "C" int __interceptor__setjmp(void *env) {
 }
 
 extern "C" SANITIZER_INTERFACE_ATTRIBUTE
-int __interceptor_sigsetjmp(void *env);
-extern "C" int __interceptor_sigsetjmp(void *env) {
+int TSAN_INTERCEPTOR_SIGSETJMP(void *env);
+extern "C" int TSAN_INTERCEPTOR_SIGSETJMP(void *env) {
   CHECK(0);
   return 0;
 }
 
+#if !SANITIEZER_NETBSD
 extern "C" SANITIZER_INTERFACE_ATTRIBUTE
 int __interceptor___sigsetjmp(void *env);
 extern "C" int __interceptor___sigsetjmp(void *env) {
   CHECK(0);
   return 0;
 }
+#endif
 
-extern "C" int setjmp(void *env);
+extern "C" int setjmp_symname(void *env);
 extern "C" int _setjmp(void *env);
-extern "C" int sigsetjmp(void *env);
+extern "C" int sigsetjmp_symname(void *env);
+#if !SANITIEZER_NETBSD
 extern "C" int __sigsetjmp(void *env);
-DEFINE_REAL(int, setjmp, void *env)
+#endif
+DEFINE_REAL(int, setjmp_symname, void *env)
 DEFINE_REAL(int, _setjmp, void *env)
-DEFINE_REAL(int, sigsetjmp, void *env)
+DEFINE_REAL(int, sigsetjmp_symname, void *env)
+#if !SANITIEZER_NETBSD
 DEFINE_REAL(int, __sigsetjmp, void *env)
+#endif
 #endif  // SANITIZER_MAC
 
-TSAN_INTERCEPTOR(void, longjmp, uptr *env, int val) {
+#if SANITIZER_NETBSD
+#define longjmp_symname __longjmp14
+#define siglongjmp_symname __siglongjmp14
+#else
+#define longjmp_symname longjmp
+#define siglongjmp_symname siglongjmp
+#endif
+
+TSAN_INTERCEPTOR(void, longjmp_symname, uptr *env, int val) {
   // Note: if we call REAL(longjmp) in the context of ScopedInterceptor,
   // bad things will happen. We will jump over ScopedInterceptor dtor and can
   // leave thr->in_ignored_lib set.
   {
-    SCOPED_INTERCEPTOR_RAW(longjmp, env, val);
+    SCOPED_INTERCEPTOR_RAW(longjmp_symname, env, val);
   }
   LongJmp(cur_thread(), env);
-  REAL(longjmp)(env, val);
+  REAL(longjmp_symname)(env, val);
 }
 
-TSAN_INTERCEPTOR(void, siglongjmp, uptr *env, int val) {
+TSAN_INTERCEPTOR(void, siglongjmp_symname, uptr *env, int val) {
   {
-    SCOPED_INTERCEPTOR_RAW(siglongjmp, env, val);
+    SCOPED_INTERCEPTOR_RAW(siglongjmp_symname, env, val);
   }
   LongJmp(cur_thread(), env);
-  REAL(siglongjmp)(env, val);
+  REAL(siglongjmp_symname)(env, val);
 }
 
+#if SANITIZER_NETBSD
+TSAN_INTERCEPTOR(void, _longjmp, uptr *env, int val) {
+  {
+    SCOPED_INTERCEPTOR_RAW(_longjmp, env, val);
+  }
+  LongJmp(cur_thread(), env);
+  REAL(_longjmp)(env, val);
+}
+#endif
+
 #if !SANITIZER_MAC
 TSAN_INTERCEPTOR(void*, malloc, uptr size) {
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return InternalAlloc(size);
   void *p = 0;
   {
@@ -584,11 +672,11 @@ TSAN_INTERCEPTOR(void*, malloc, uptr size) {
 
 TSAN_INTERCEPTOR(void*, __libc_memalign, uptr align, uptr sz) {
   SCOPED_TSAN_INTERCEPTOR(__libc_memalign, align, sz);
-  return user_alloc(thr, pc, sz, align);
+  return user_memalign(thr, pc, align, sz);
 }
 
 TSAN_INTERCEPTOR(void*, calloc, uptr size, uptr n) {
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return InternalCalloc(size, n);
   void *p = 0;
   {
@@ -600,7 +688,7 @@ TSAN_INTERCEPTOR(void*, calloc, uptr size, uptr n) {
 }
 
 TSAN_INTERCEPTOR(void*, realloc, void *p, uptr size) {
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return InternalRealloc(p, size);
   if (p)
     invoke_free_hook(p);
@@ -615,7 +703,7 @@ TSAN_INTERCEPTOR(void*, realloc, void *p, uptr size) {
 TSAN_INTERCEPTOR(void, free, void *p) {
   if (p == 0)
     return;
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return InternalFree(p);
   invoke_free_hook(p);
   SCOPED_INTERCEPTOR_RAW(free, p);
@@ -625,7 +713,7 @@ TSAN_INTERCEPTOR(void, free, void *p) {
 TSAN_INTERCEPTOR(void, cfree, void *p) {
   if (p == 0)
     return;
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return InternalFree(p);
   invoke_free_hook(p);
   SCOPED_INTERCEPTOR_RAW(cfree, p);
@@ -730,7 +818,7 @@ TSAN_INTERCEPTOR(int, munmap, void *addr, long_t sz) {
 #if SANITIZER_LINUX
 TSAN_INTERCEPTOR(void*, memalign, uptr align, uptr sz) {
   SCOPED_INTERCEPTOR_RAW(memalign, align, sz);
-  return user_alloc(thr, pc, sz, align);
+  return user_memalign(thr, pc, align, sz);
 }
 #define TSAN_MAYBE_INTERCEPT_MEMALIGN TSAN_INTERCEPT(memalign)
 #else
@@ -739,21 +827,29 @@ TSAN_INTERCEPTOR(void*, memalign, uptr align, uptr sz) {
 
 #if !SANITIZER_MAC
 TSAN_INTERCEPTOR(void*, aligned_alloc, uptr align, uptr sz) {
-  SCOPED_INTERCEPTOR_RAW(memalign, align, sz);
-  return user_alloc(thr, pc, sz, align);
+  if (UNLIKELY(cur_thread()->in_symbolizer))
+    return InternalAlloc(sz, nullptr, align);
+  SCOPED_INTERCEPTOR_RAW(aligned_alloc, align, sz);
+  return user_aligned_alloc(thr, pc, align, sz);
 }
 
 TSAN_INTERCEPTOR(void*, valloc, uptr sz) {
+  if (UNLIKELY(cur_thread()->in_symbolizer))
+    return InternalAlloc(sz, nullptr, GetPageSizeCached());
   SCOPED_INTERCEPTOR_RAW(valloc, sz);
-  return user_alloc(thr, pc, sz, GetPageSizeCached());
+  return user_valloc(thr, pc, sz);
 }
 #endif
 
 #if SANITIZER_LINUX
 TSAN_INTERCEPTOR(void*, pvalloc, uptr sz) {
+  if (UNLIKELY(cur_thread()->in_symbolizer)) {
+    uptr PageSize = GetPageSizeCached();
+    sz = sz ? RoundUpTo(sz, PageSize) : PageSize;
+    return InternalAlloc(sz, nullptr, PageSize);
+  }
   SCOPED_INTERCEPTOR_RAW(pvalloc, sz);
-  sz = RoundUp(sz, GetPageSizeCached());
-  return user_alloc(thr, pc, sz, GetPageSizeCached());
+  return user_pvalloc(thr, pc, sz);
 }
 #define TSAN_MAYBE_INTERCEPT_PVALLOC TSAN_INTERCEPT(pvalloc)
 #else
@@ -762,9 +858,15 @@ TSAN_INTERCEPTOR(void*, pvalloc, uptr sz) {
 
 #if !SANITIZER_MAC
 TSAN_INTERCEPTOR(int, posix_memalign, void **memptr, uptr align, uptr sz) {
+  if (UNLIKELY(cur_thread()->in_symbolizer)) {
+    void *p = InternalAlloc(sz, nullptr, align);
+    if (!p)
+      return errno_ENOMEM;
+    *memptr = p;
+    return 0;
+  }
   SCOPED_INTERCEPTOR_RAW(posix_memalign, memptr, align, sz);
-  *memptr = user_alloc(thr, pc, sz, align);
-  return 0;
+  return user_posix_memalign(thr, pc, memptr, align, sz);
 }
 #endif
 
@@ -830,11 +932,12 @@ void DestroyThreadState() {
 }
 }  // namespace __tsan
 
-#if !SANITIZER_MAC
+#if !SANITIZER_MAC && !SANITIZER_NETBSD
 static void thread_finalize(void *v) {
   uptr iter = (uptr)v;
   if (iter > 1) {
-    if (pthread_setspecific(g_thread_finalize_key, (void*)(iter - 1))) {
+    if (pthread_setspecific(interceptor_ctx()->finalize_key,
+        (void*)(iter - 1))) {
       Printf("ThreadSanitizer: failed to set thread key\n");
       Die();
     }
@@ -860,9 +963,9 @@ extern "C" void *__tsan_thread_start_func(void *arg) {
     ThreadState *thr = cur_thread();
     // Thread-local state is not initialized yet.
     ScopedIgnoreInterceptors ignore;
-#if !SANITIZER_MAC
+#if !SANITIZER_MAC && !SANITIZER_NETBSD
     ThreadIgnoreBegin(thr, 0);
-    if (pthread_setspecific(g_thread_finalize_key,
+    if (pthread_setspecific(interceptor_ctx()->finalize_key,
                             (void *)GetPthreadDestructorIterations())) {
       Printf("ThreadSanitizer: failed to set thread key\n");
       Die();
@@ -887,6 +990,9 @@ extern "C" void *__tsan_thread_start_func(void *arg) {
 TSAN_INTERCEPTOR(int, pthread_create,
     void *th, void *attr, void *(*callback)(void*), void * param) {
   SCOPED_INTERCEPTOR_RAW(pthread_create, th, attr, callback, param);
+
+  MaybeSpawnBackgroundThread();
+
   if (ctx->after_multithreaded_fork) {
     if (flags()->die_after_fork) {
       Report("ThreadSanitizer: starting new threads after multi-threaded "
@@ -1312,10 +1418,15 @@ TSAN_INTERCEPTOR(int, pthread_once, void *o, void (*f)()) {
   if (o == 0 || f == 0)
     return errno_EINVAL;
   atomic_uint32_t *a;
-  if (!SANITIZER_MAC)
-    a = static_cast<atomic_uint32_t*>(o);
-  else  // On OS X, pthread_once_t has a header with a long-sized signature.
+
+  if (SANITIZER_MAC)
     a = static_cast<atomic_uint32_t*>((void *)((char *)o + sizeof(long_t)));
+  else if (SANITIZER_NETBSD)
+    a = static_cast<atomic_uint32_t*>
+          ((void *)((char *)o + __sanitizer::pthread_mutex_t_sz));
+  else
+    a = static_cast<atomic_uint32_t*>(o);
+
   u32 v = atomic_load(a, memory_order_acquire);
   if (v == 0 && atomic_compare_exchange_strong(a, &v, 1,
                                                memory_order_relaxed)) {
@@ -1347,7 +1458,7 @@ TSAN_INTERCEPTOR(int, __fxstat, int version, int fd, void *buf) {
 #endif
 
 TSAN_INTERCEPTOR(int, fstat, int fd, void *buf) {
-#if SANITIZER_FREEBSD || SANITIZER_MAC || SANITIZER_ANDROID
+#if SANITIZER_FREEBSD || SANITIZER_MAC || SANITIZER_ANDROID || SANITIZER_NETBSD
   SCOPED_TSAN_INTERCEPTOR(fstat, fd, buf);
   if (fd > 0)
     FdAccess(thr, pc, fd);
@@ -1762,7 +1873,9 @@ TSAN_INTERCEPTOR(int, pthread_sigmask, int how, const __sanitizer_sigset_t *set,
 namespace __tsan {
 
 static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
-    bool sigact, int sig, my_siginfo_t *info, void *uctx) {
+                                  bool sigact, int sig,
+                                  __sanitizer_siginfo *info, void *uctx) {
+  __sanitizer_sigaction *sigactions = interceptor_ctx()->sigactions;
   if (acquire)
     Acquire(thr, 0, (uptr)&sigactions[sig]);
   // Signals are generally asynchronous, so if we receive a signals when
@@ -1784,14 +1897,13 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
   // This code races with sigaction. Be careful to not read sa_sigaction twice.
   // Also need to remember pc for reporting before the call,
   // because the handler can reset it.
-  volatile uptr pc = sigact ?
-     (uptr)sigactions[sig].sa_sigaction :
-     (uptr)sigactions[sig].sa_handler;
-  if (pc != (uptr)SIG_DFL && pc != (uptr)SIG_IGN) {
+  volatile uptr pc =
+      sigact ? (uptr)sigactions[sig].sigaction : (uptr)sigactions[sig].handler;
+  if (pc != sig_dfl && pc != sig_ign) {
     if (sigact)
-      ((sigactionhandler_t)pc)(sig, info, uctx);
+      ((__sanitizer_sigactionhandler_ptr)pc)(sig, info, uctx);
     else
-      ((sighandler_t)pc)(sig);
+      ((__sanitizer_sighandler_ptr)pc)(sig);
   }
   if (!ctx->after_multithreaded_fork) {
     thr->ignore_reads_and_writes = ignore_reads_and_writes;
@@ -1855,7 +1967,8 @@ static bool is_sync_signal(ThreadSignalContext *sctx, int sig) {
 }
 
 void ALWAYS_INLINE rtl_generic_sighandler(bool sigact, int sig,
-    my_siginfo_t *info, void *ctx) {
+                                          __sanitizer_siginfo *info,
+                                          void *ctx) {
   ThreadState *thr = cur_thread();
   ThreadSignalContext *sctx = SigCtx(thr);
   if (sig < 0 || sig >= kSigCount) {
@@ -1905,58 +2018,10 @@ static void rtl_sighandler(int sig) {
   rtl_generic_sighandler(false, sig, 0, 0);
 }
 
-static void rtl_sigaction(int sig, my_siginfo_t *info, void *ctx) {
+static void rtl_sigaction(int sig, __sanitizer_siginfo *info, void *ctx) {
   rtl_generic_sighandler(true, sig, info, ctx);
 }
 
-TSAN_INTERCEPTOR(int, sigaction, int sig, sigaction_t *act, sigaction_t *old) {
-  // Note: if we call REAL(sigaction) directly for any reason without proxying
-  // the signal handler through rtl_sigaction, very bad things will happen.
-  // The handler will run synchronously and corrupt tsan per-thread state.
-  SCOPED_INTERCEPTOR_RAW(sigaction, sig, act, old);
-  if (old)
-    internal_memcpy(old, &sigactions[sig], sizeof(*old));
-  if (act == 0)
-    return 0;
-  // Copy act into sigactions[sig].
-  // Can't use struct copy, because compiler can emit call to memcpy.
-  // Can't use internal_memcpy, because it copies byte-by-byte,
-  // and signal handler reads the sa_handler concurrently. It it can read
-  // some bytes from old value and some bytes from new value.
-  // Use volatile to prevent insertion of memcpy.
-  sigactions[sig].sa_handler = *(volatile sighandler_t*)&act->sa_handler;
-  sigactions[sig].sa_flags = *(volatile int*)&act->sa_flags;
-  internal_memcpy(&sigactions[sig].sa_mask, &act->sa_mask,
-      sizeof(sigactions[sig].sa_mask));
-#if !SANITIZER_FREEBSD && !SANITIZER_MAC
-  sigactions[sig].sa_restorer = act->sa_restorer;
-#endif
-  sigaction_t newact;
-  internal_memcpy(&newact, act, sizeof(newact));
-  internal_sigfillset(&newact.sa_mask);
-  if (act->sa_handler != SIG_IGN && act->sa_handler != SIG_DFL) {
-    if (newact.sa_flags & SA_SIGINFO)
-      newact.sa_sigaction = rtl_sigaction;
-    else
-      newact.sa_handler = rtl_sighandler;
-  }
-  ReleaseStore(thr, pc, (uptr)&sigactions[sig]);
-  int res = REAL(sigaction)(sig, &newact, 0);
-  return res;
-}
-
-TSAN_INTERCEPTOR(sighandler_t, signal, int sig, sighandler_t h) {
-  sigaction_t act;
-  act.sa_handler = h;
-  internal_memset(&act.sa_mask, -1, sizeof(act.sa_mask));
-  act.sa_flags = 0;
-  sigaction_t old;
-  int res = sigaction(sig, &act, &old);
-  if (res)
-    return SIG_ERR;
-  return old.sa_handler;
-}
-
 TSAN_INTERCEPTOR(int, raise, int sig) {
   SCOPED_TSAN_INTERCEPTOR(raise, sig);
   ThreadSignalContext *sctx = SigCtx(thr);
@@ -2020,7 +2085,7 @@ TSAN_INTERCEPTOR(int, getaddrinfo, void *node, void *service,
 }
 
 TSAN_INTERCEPTOR(int, fork, int fake) {
-  if (cur_thread()->in_symbolizer)
+  if (UNLIKELY(cur_thread()->in_symbolizer))
     return REAL(fork)(fake);
   SCOPED_INTERCEPTOR_RAW(fork, fake);
   ForkBefore(thr, pc);
@@ -2270,6 +2335,77 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
 
 #include "sanitizer_common/sanitizer_common_interceptors.inc"
 
+static int sigaction_impl(int sig, const __sanitizer_sigaction *act,
+                          __sanitizer_sigaction *old);
+static __sanitizer_sighandler_ptr signal_impl(int sig,
+                                              __sanitizer_sighandler_ptr h);
+
+#define SIGNAL_INTERCEPTOR_SIGACTION_IMPL(signo, act, oldact) \
+  { return sigaction_impl(signo, act, oldact); }
+
+#define SIGNAL_INTERCEPTOR_SIGNAL_IMPL(func, signo, handler) \
+  { return (uptr)signal_impl(signo, (__sanitizer_sighandler_ptr)handler); }
+
+#include "sanitizer_common/sanitizer_signal_interceptors.inc"
+
+int sigaction_impl(int sig, const __sanitizer_sigaction *act,
+                   __sanitizer_sigaction *old) {
+  // Note: if we call REAL(sigaction) directly for any reason without proxying
+  // the signal handler through rtl_sigaction, very bad things will happen.
+  // The handler will run synchronously and corrupt tsan per-thread state.
+  SCOPED_INTERCEPTOR_RAW(sigaction, sig, act, old);
+  __sanitizer_sigaction *sigactions = interceptor_ctx()->sigactions;
+  __sanitizer_sigaction old_stored;
+  if (old) internal_memcpy(&old_stored, &sigactions[sig], sizeof(old_stored));
+  __sanitizer_sigaction newact;
+  if (act) {
+    // Copy act into sigactions[sig].
+    // Can't use struct copy, because compiler can emit call to memcpy.
+    // Can't use internal_memcpy, because it copies byte-by-byte,
+    // and signal handler reads the handler concurrently. It it can read
+    // some bytes from old value and some bytes from new value.
+    // Use volatile to prevent insertion of memcpy.
+    sigactions[sig].handler =
+        *(volatile __sanitizer_sighandler_ptr const *)&act->handler;
+    sigactions[sig].sa_flags = *(volatile int const *)&act->sa_flags;
+    internal_memcpy(&sigactions[sig].sa_mask, &act->sa_mask,
+                    sizeof(sigactions[sig].sa_mask));
+#if !SANITIZER_FREEBSD && !SANITIZER_MAC && !SANITIZER_NETBSD
+    sigactions[sig].sa_restorer = act->sa_restorer;
+#endif
+    internal_memcpy(&newact, act, sizeof(newact));
+    internal_sigfillset(&newact.sa_mask);
+    if ((uptr)act->handler != sig_ign && (uptr)act->handler != sig_dfl) {
+      if (newact.sa_flags & SA_SIGINFO)
+        newact.sigaction = rtl_sigaction;
+      else
+        newact.handler = rtl_sighandler;
+    }
+    ReleaseStore(thr, pc, (uptr)&sigactions[sig]);
+    act = &newact;
+  }
+  int res = REAL(sigaction)(sig, act, old);
+  if (res == 0 && old) {
+    uptr cb = (uptr)old->sigaction;
+    if (cb == (uptr)rtl_sigaction || cb == (uptr)rtl_sighandler) {
+      internal_memcpy(old, &old_stored, sizeof(*old));
+    }
+  }
+  return res;
+}
+
+static __sanitizer_sighandler_ptr signal_impl(int sig,
+                                              __sanitizer_sighandler_ptr h) {
+  __sanitizer_sigaction act;
+  act.handler = h;
+  internal_memset(&act.sa_mask, -1, sizeof(act.sa_mask));
+  act.sa_flags = 0;
+  __sanitizer_sigaction old;
+  int res = sigaction_symname(sig, &act, &old);
+  if (res) return (__sanitizer_sighandler_ptr)sig_err;
+  return old.handler;
+}
+
 #define TSAN_SYSCALL() \
   ThreadState *thr = cur_thread(); \
   if (thr->ignore_interceptors) \
@@ -2290,7 +2426,7 @@ struct ScopedSyscall {
   }
 };
 
-#if !SANITIZER_FREEBSD && !SANITIZER_MAC
+#if !SANITIZER_FREEBSD && !SANITIZER_MAC && !SANITIZER_NETBSD
 static void syscall_access_range(uptr pc, uptr p, uptr s, bool write) {
   TSAN_SYSCALL();
   MemoryAccessRange(thr, pc, p, s, write);
@@ -2409,6 +2545,34 @@ TSAN_INTERCEPTOR(void *, __tls_get_addr, void *arg) {
 }
 #endif
 
+#if SANITIZER_NETBSD
+TSAN_INTERCEPTOR(void, _lwp_exit) {
+  SCOPED_TSAN_INTERCEPTOR(_lwp_exit);
+  DestroyThreadState();
+  REAL(_lwp_exit)();
+}
+#define TSAN_MAYBE_INTERCEPT__LWP_EXIT TSAN_INTERCEPT(_lwp_exit)
+#else
+#define TSAN_MAYBE_INTERCEPT__LWP_EXIT
+#endif
+
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, cond_init, void *c, void *a);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, cond_signal, void *c);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, cond_broadcast, void *c);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, cond_wait, void *c, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, cond_destroy, void *c);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, mutex_init, void *m, void *a);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, mutex_destroy, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, mutex_trylock, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, rwlock_init, void *m, void *a);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, rwlock_destroy, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, rwlock_rdlock, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, rwlock_tryrdlock, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, rwlock_wrlock, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, rwlock_trywrlock, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS(int, rwlock_unlock, void *m);
+TSAN_INTERCEPTOR_NETBSD_ALIAS_THR(int, once, void *o, void (*f)());
+
 namespace __tsan {
 
 static void finalize(void *arg) {
@@ -2440,20 +2604,30 @@ void InitializeInterceptors() {
   mallopt(-3, 32*1024);  // M_MMAP_THRESHOLD
 #endif
 
+  new(interceptor_ctx()) InterceptorContext();
+
   InitializeCommonInterceptors();
+  InitializeSignalInterceptors();
 
 #if !SANITIZER_MAC
   // We can not use TSAN_INTERCEPT to get setjmp addr,
   // because it does &setjmp and setjmp is not present in some versions of libc.
   using __interception::GetRealFunctionAddress;
-  GetRealFunctionAddress("setjmp", (uptr*)&REAL(setjmp), 0, 0);
+  GetRealFunctionAddress(TSAN_STRING_SETJMP,
+                         (uptr*)&REAL(setjmp_symname), 0, 0);
   GetRealFunctionAddress("_setjmp", (uptr*)&REAL(_setjmp), 0, 0);
-  GetRealFunctionAddress("sigsetjmp", (uptr*)&REAL(sigsetjmp), 0, 0);
+  GetRealFunctionAddress(TSAN_STRING_SIGSETJMP,
+                         (uptr*)&REAL(sigsetjmp_symname), 0, 0);
+#if !SANITIZER_NETBSD
   GetRealFunctionAddress("__sigsetjmp", (uptr*)&REAL(__sigsetjmp), 0, 0);
 #endif
+#endif
 
-  TSAN_INTERCEPT(longjmp);
-  TSAN_INTERCEPT(siglongjmp);
+  TSAN_INTERCEPT(longjmp_symname);
+  TSAN_INTERCEPT(siglongjmp_symname);
+#if SANITIZER_NETBSD
+  TSAN_INTERCEPT(_longjmp);
+#endif
 
   TSAN_INTERCEPT(malloc);
   TSAN_INTERCEPT(__libc_memalign);
@@ -2548,8 +2722,6 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(rmdir);
   TSAN_INTERCEPT(closedir);
 
-  TSAN_INTERCEPT(sigaction);
-  TSAN_INTERCEPT(signal);
   TSAN_INTERCEPT(sigsuspend);
   TSAN_INTERCEPT(sigblock);
   TSAN_INTERCEPT(sigsetmask);
@@ -2560,6 +2732,7 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(sleep);
   TSAN_INTERCEPT(usleep);
   TSAN_INTERCEPT(nanosleep);
+  TSAN_INTERCEPT(pause);
   TSAN_INTERCEPT(gettimeofday);
   TSAN_INTERCEPT(getaddrinfo);
 
@@ -2568,7 +2741,7 @@ void InitializeInterceptors() {
 #if !SANITIZER_ANDROID
   TSAN_INTERCEPT(dl_iterate_phdr);
 #endif
-  TSAN_INTERCEPT(on_exit);
+  TSAN_MAYBE_INTERCEPT_ON_EXIT;
   TSAN_INTERCEPT(__cxa_atexit);
   TSAN_INTERCEPT(_exit);
 
@@ -2576,6 +2749,8 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(__tls_get_addr);
 #endif
 
+  TSAN_MAYBE_INTERCEPT__LWP_EXIT;
+
 #if !SANITIZER_MAC && !SANITIZER_ANDROID
   // Need to setup it, because interceptors check that the function is resolved.
   // But atexit is emitted directly into the module, so can't be resolved.
@@ -2587,13 +2762,30 @@ void InitializeInterceptors() {
     Die();
   }
 
-#if !SANITIZER_MAC
-  if (pthread_key_create(&g_thread_finalize_key, &thread_finalize)) {
+#if !SANITIZER_MAC && !SANITIZER_NETBSD
+  if (pthread_key_create(&interceptor_ctx()->finalize_key, &thread_finalize)) {
     Printf("ThreadSanitizer: failed to create thread key\n");
     Die();
   }
 #endif
 
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(cond_init);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(cond_signal);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(cond_broadcast);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(cond_wait);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(cond_destroy);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(mutex_init);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(mutex_destroy);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(mutex_trylock);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(rwlock_init);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(rwlock_destroy);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(rwlock_rdlock);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(rwlock_tryrdlock);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(rwlock_wrlock);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(rwlock_trywrlock);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS(rwlock_unlock);
+  TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS_THR(once);
+
   FdInit();
 }
 
diff --git a/lib/tsan/rtl/tsan_interceptors.h b/lib/tsan/rtl/tsan_interceptors.h
index de47466501da..959a39465e33 100644
--- a/lib/tsan/rtl/tsan_interceptors.h
+++ b/lib/tsan/rtl/tsan_interceptors.h
@@ -49,4 +49,16 @@ LibIgnore *libignore();
 
 #define TSAN_INTERCEPTOR(ret, func, ...) INTERCEPTOR(ret, func, __VA_ARGS__)
 
+#if SANITIZER_NETBSD
+# define TSAN_INTERCEPTOR_NETBSD_ALIAS(ret, func, ...) \
+  TSAN_INTERCEPTOR(ret, __libc_##func, __VA_ARGS__) \
+  ALIAS(WRAPPER_NAME(pthread_##func));
+# define TSAN_INTERCEPTOR_NETBSD_ALIAS_THR(ret, func, ...) \
+  TSAN_INTERCEPTOR(ret, __libc_thr_##func, __VA_ARGS__) \
+  ALIAS(WRAPPER_NAME(pthread_##func));
+#else
+# define TSAN_INTERCEPTOR_NETBSD_ALIAS(ret, func, ...)
+# define TSAN_INTERCEPTOR_NETBSD_ALIAS_THR(ret, func, ...)
+#endif
+
 #endif  // TSAN_INTERCEPTORS_H
diff --git a/lib/tsan/rtl/tsan_interceptors_mac.cc b/lib/tsan/rtl/tsan_interceptors_mac.cc
index 4f1079467331..3b3a4a2815c3 100644
--- a/lib/tsan/rtl/tsan_interceptors_mac.cc
+++ b/lib/tsan/rtl/tsan_interceptors_mac.cc
@@ -100,7 +100,7 @@ OSATOMIC_INTERCEPTORS_BITWISE(OSAtomicXor, fetch_xor,
   TSAN_INTERCEPTOR(bool, f, t old_value, t new_value, t volatile *ptr) {    \
     SCOPED_TSAN_INTERCEPTOR(f, old_value, new_value, ptr);                  \
     return tsan_atomic_f##_compare_exchange_strong(                         \
-        (tsan_t *)ptr, (tsan_t *)&old_value, (tsan_t)new_value,             \
+        (volatile tsan_t *)ptr, (tsan_t *)&old_value, (tsan_t)new_value,    \
         kMacOrderNonBarrier, kMacOrderNonBarrier);                          \
   }                                                                         \
                                                                             \
@@ -108,7 +108,7 @@ OSATOMIC_INTERCEPTORS_BITWISE(OSAtomicXor, fetch_xor,
                    t volatile *ptr) {                                       \
     SCOPED_TSAN_INTERCEPTOR(f##Barrier, old_value, new_value, ptr);         \
     return tsan_atomic_f##_compare_exchange_strong(                         \
-        (tsan_t *)ptr, (tsan_t *)&old_value, (tsan_t)new_value,             \
+        (volatile tsan_t *)ptr, (tsan_t *)&old_value, (tsan_t)new_value,    \
         kMacOrderBarrier, kMacOrderNonBarrier);                             \
   }
 
@@ -122,14 +122,14 @@ OSATOMIC_INTERCEPTORS_CAS(OSAtomicCompareAndSwap32, __tsan_atomic32, a32,
 OSATOMIC_INTERCEPTORS_CAS(OSAtomicCompareAndSwap64, __tsan_atomic64, a64,
                           int64_t)
 
-#define OSATOMIC_INTERCEPTOR_BITOP(f, op, clear, mo)          \
-  TSAN_INTERCEPTOR(bool, f, uint32_t n, volatile void *ptr) { \
-    SCOPED_TSAN_INTERCEPTOR(f, n, ptr);                       \
-    char *byte_ptr = ((char *)ptr) + (n >> 3);                \
-    char bit = 0x80u >> (n & 7);                              \
-    char mask = clear ? ~bit : bit;                           \
-    char orig_byte = op((a8 *)byte_ptr, mask, mo);            \
-    return orig_byte & bit;                                   \
+#define OSATOMIC_INTERCEPTOR_BITOP(f, op, clear, mo)             \
+  TSAN_INTERCEPTOR(bool, f, uint32_t n, volatile void *ptr) {    \
+    SCOPED_TSAN_INTERCEPTOR(f, n, ptr);                          \
+    volatile char *byte_ptr = ((volatile char *)ptr) + (n >> 3); \
+    char bit = 0x80u >> (n & 7);                                 \
+    char mask = clear ? ~bit : bit;                              \
+    char orig_byte = op((volatile a8 *)byte_ptr, mask, mo);      \
+    return orig_byte & bit;                                      \
   }
 
 #define OSATOMIC_INTERCEPTORS_BITOP(f, op, clear)               \
diff --git a/lib/tsan/rtl/tsan_interface_ann.cc b/lib/tsan/rtl/tsan_interface_ann.cc
index f68a0468de53..a7922a42e42c 100644
--- a/lib/tsan/rtl/tsan_interface_ann.cc
+++ b/lib/tsan/rtl/tsan_interface_ann.cc
@@ -14,6 +14,7 @@
 #include "sanitizer_common/sanitizer_internal_defs.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "sanitizer_common/sanitizer_stacktrace.h"
+#include "sanitizer_common/sanitizer_vector.h"
 #include "tsan_interface_ann.h"
 #include "tsan_mutex.h"
 #include "tsan_report.h"
@@ -21,7 +22,6 @@
 #include "tsan_mman.h"
 #include "tsan_flags.h"
 #include "tsan_platform.h"
-#include "tsan_vector.h"
 
 #define CALLERPC ((uptr)__builtin_return_address(0))
 
@@ -185,10 +185,10 @@ void PrintMatchedBenignRaces() {
   int unique_count = 0;
   int hit_count = 0;
   int add_count = 0;
-  Vector<ExpectRace> hit_matched(MBlockScopedBuf);
+  Vector<ExpectRace> hit_matched;
   CollectMatchedBenignRaces(&hit_matched, &unique_count, &hit_count,
       &ExpectRace::hitcount);
-  Vector<ExpectRace> add_matched(MBlockScopedBuf);
+  Vector<ExpectRace> add_matched;
   CollectMatchedBenignRaces(&add_matched, &unique_count, &add_count,
       &ExpectRace::addcount);
   if (hit_matched.Size()) {
diff --git a/lib/tsan/rtl/tsan_libdispatch_mac.cc b/lib/tsan/rtl/tsan_libdispatch_mac.cc
index 8c759a3be4e1..eb22e4baa710 100644
--- a/lib/tsan/rtl/tsan_libdispatch_mac.cc
+++ b/lib/tsan/rtl/tsan_libdispatch_mac.cc
@@ -86,7 +86,8 @@ static tsan_block_context_t *AllocContext(ThreadState *thr, uptr pc,
                                           void *orig_context,
                                           dispatch_function_t orig_work) {
   tsan_block_context_t *new_context =
-      (tsan_block_context_t *)user_alloc(thr, pc, sizeof(tsan_block_context_t));
+      (tsan_block_context_t *)user_alloc_internal(thr, pc,
+                                                  sizeof(tsan_block_context_t));
   new_context->queue = queue;
   new_context->orig_context = orig_context;
   new_context->orig_work = orig_work;
@@ -176,7 +177,8 @@ static void invoke_and_release_block(void *param) {
   }
 
 #define DISPATCH_INTERCEPT_SYNC_B(name, barrier)                             \
-  TSAN_INTERCEPTOR(void, name, dispatch_queue_t q, dispatch_block_t block) { \
+  TSAN_INTERCEPTOR(void, name, dispatch_queue_t q,                           \
+                   DISPATCH_NOESCAPE dispatch_block_t block) {               \
     SCOPED_TSAN_INTERCEPTOR(name, q, block);                                 \
     SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();                           \
     dispatch_block_t heap_block = Block_copy(block);                         \
@@ -266,7 +268,7 @@ TSAN_INTERCEPTOR(void, dispatch_after_f, dispatch_time_t when,
 // need to undefine the macro.
 #undef dispatch_once
 TSAN_INTERCEPTOR(void, dispatch_once, dispatch_once_t *predicate,
-                 dispatch_block_t block) {
+                 DISPATCH_NOESCAPE dispatch_block_t block) {
   SCOPED_INTERCEPTOR_RAW(dispatch_once, predicate, block);
   atomic_uint32_t *a = reinterpret_cast<atomic_uint32_t *>(predicate);
   u32 v = atomic_load(a, memory_order_acquire);
@@ -476,7 +478,8 @@ TSAN_INTERCEPTOR(void, dispatch_source_set_registration_handler_f,
 }
 
 TSAN_INTERCEPTOR(void, dispatch_apply, size_t iterations,
-                 dispatch_queue_t queue, void (^block)(size_t)) {
+                 dispatch_queue_t queue,
+                 DISPATCH_NOESCAPE void (^block)(size_t)) {
   SCOPED_TSAN_INTERCEPTOR(dispatch_apply, iterations, queue, block);
 
   void *parent_to_child_sync = nullptr;
@@ -519,9 +522,9 @@ TSAN_INTERCEPTOR(dispatch_data_t, dispatch_data_create, const void *buffer,
     return REAL(dispatch_data_create)(buffer, size, q, destructor);
 
   if (destructor == DISPATCH_DATA_DESTRUCTOR_FREE)
-    destructor = ^(void) { WRAP(free)((void *)buffer); };
+    destructor = ^(void) { WRAP(free)((void *)(uintptr_t)buffer); };
   else if (destructor == DISPATCH_DATA_DESTRUCTOR_MUNMAP)
-    destructor = ^(void) { WRAP(munmap)((void *)buffer, size); };
+    destructor = ^(void) { WRAP(munmap)((void *)(uintptr_t)buffer, size); };
 
   SCOPED_TSAN_INTERCEPTOR_USER_CALLBACK_START();
   dispatch_block_t heap_block = Block_copy(destructor);
diff --git a/lib/tsan/rtl/tsan_malloc_mac.cc b/lib/tsan/rtl/tsan_malloc_mac.cc
index 8d31ccbcaaca..455c95df6c50 100644
--- a/lib/tsan/rtl/tsan_malloc_mac.cc
+++ b/lib/tsan/rtl/tsan_malloc_mac.cc
@@ -26,7 +26,7 @@ using namespace __tsan;
 #define COMMON_MALLOC_FORCE_UNLOCK()
 #define COMMON_MALLOC_MEMALIGN(alignment, size) \
   void *p =                                     \
-      user_alloc(cur_thread(), StackTrace::GetCurrentPc(), size, alignment)
+      user_memalign(cur_thread(), StackTrace::GetCurrentPc(), alignment, size)
 #define COMMON_MALLOC_MALLOC(size)                             \
   if (cur_thread()->in_symbolizer) return InternalAlloc(size); \
   SCOPED_INTERCEPTOR_RAW(malloc, size);                        \
@@ -43,7 +43,7 @@ using namespace __tsan;
   if (cur_thread()->in_symbolizer)                            \
     return InternalAlloc(size, nullptr, GetPageSizeCached()); \
   SCOPED_INTERCEPTOR_RAW(valloc, size);                       \
-  void *p = user_alloc(thr, pc, size, GetPageSizeCached())
+  void *p = user_valloc(thr, pc, size)
 #define COMMON_MALLOC_FREE(ptr)                              \
   if (cur_thread()->in_symbolizer) return InternalFree(ptr); \
   SCOPED_INTERCEPTOR_RAW(free, ptr);                         \
diff --git a/lib/tsan/rtl/tsan_mman.cc b/lib/tsan/rtl/tsan_mman.cc
index f79dccddba9f..19680238bf76 100644
--- a/lib/tsan/rtl/tsan_mman.cc
+++ b/lib/tsan/rtl/tsan_mman.cc
@@ -13,6 +13,7 @@
 #include "sanitizer_common/sanitizer_allocator_checks.h"
 #include "sanitizer_common/sanitizer_allocator_interface.h"
 #include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_errno.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "tsan_mman.h"
 #include "tsan_rtl.h"
@@ -149,11 +150,12 @@ static void SignalUnsafeCall(ThreadState *thr, uptr pc) {
   OutputReport(thr, rep);
 }
 
-void *user_alloc(ThreadState *thr, uptr pc, uptr sz, uptr align, bool signal) {
+void *user_alloc_internal(ThreadState *thr, uptr pc, uptr sz, uptr align,
+                          bool signal) {
   if ((sz >= (1ull << 40)) || (align >= (1ull << 40)))
     return Allocator::FailureHandler::OnBadRequest();
   void *p = allocator()->Allocate(&thr->proc()->alloc_cache, sz, align);
-  if (p == 0)
+  if (UNLIKELY(p == 0))
     return 0;
   if (ctx && ctx->initialized)
     OnUserAlloc(thr, pc, (uptr)p, sz, true);
@@ -162,15 +164,6 @@ void *user_alloc(ThreadState *thr, uptr pc, uptr sz, uptr align, bool signal) {
   return p;
 }
 
-void *user_calloc(ThreadState *thr, uptr pc, uptr size, uptr n) {
-  if (CheckForCallocOverflow(size, n))
-    return Allocator::FailureHandler::OnBadRequest();
-  void *p = user_alloc(thr, pc, n * size);
-  if (p)
-    internal_memset(p, 0, n * size);
-  return p;
-}
-
 void user_free(ThreadState *thr, uptr pc, void *p, bool signal) {
   ScopedGlobalProcessor sgp;
   if (ctx && ctx->initialized)
@@ -180,6 +173,19 @@ void user_free(ThreadState *thr, uptr pc, void *p, bool signal) {
     SignalUnsafeCall(thr, pc);
 }
 
+void *user_alloc(ThreadState *thr, uptr pc, uptr sz) {
+  return SetErrnoOnNull(user_alloc_internal(thr, pc, sz, kDefaultAlignment));
+}
+
+void *user_calloc(ThreadState *thr, uptr pc, uptr size, uptr n) {
+  if (UNLIKELY(CheckForCallocOverflow(size, n)))
+    return SetErrnoOnNull(Allocator::FailureHandler::OnBadRequest());
+  void *p = user_alloc_internal(thr, pc, n * size);
+  if (p)
+    internal_memset(p, 0, n * size);
+  return SetErrnoOnNull(p);
+}
+
 void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write) {
   DPrintf("#%d: alloc(%zu) = %p\n", thr->tid, sz, p);
   ctx->metamap.AllocBlock(thr, pc, p, sz);
@@ -200,15 +206,64 @@ void OnUserFree(ThreadState *thr, uptr pc, uptr p, bool write) {
 void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz) {
   // FIXME: Handle "shrinking" more efficiently,
   // it seems that some software actually does this.
-  void *p2 = user_alloc(thr, pc, sz);
-  if (p2 == 0)
-    return 0;
-  if (p) {
-    uptr oldsz = user_alloc_usable_size(p);
-    internal_memcpy(p2, p, min(oldsz, sz));
+  if (!p)
+    return SetErrnoOnNull(user_alloc_internal(thr, pc, sz));
+  if (!sz) {
+    user_free(thr, pc, p);
+    return nullptr;
+  }
+  void *new_p = user_alloc_internal(thr, pc, sz);
+  if (new_p) {
+    uptr old_sz = user_alloc_usable_size(p);
+    internal_memcpy(new_p, p, min(old_sz, sz));
     user_free(thr, pc, p);
   }
-  return p2;
+  return SetErrnoOnNull(new_p);
+}
+
+void *user_memalign(ThreadState *thr, uptr pc, uptr align, uptr sz) {
+  if (UNLIKELY(!IsPowerOfTwo(align))) {
+    errno = errno_EINVAL;
+    return Allocator::FailureHandler::OnBadRequest();
+  }
+  return SetErrnoOnNull(user_alloc_internal(thr, pc, sz, align));
+}
+
+int user_posix_memalign(ThreadState *thr, uptr pc, void **memptr, uptr align,
+                        uptr sz) {
+  if (UNLIKELY(!CheckPosixMemalignAlignment(align))) {
+    Allocator::FailureHandler::OnBadRequest();
+    return errno_EINVAL;
+  }
+  void *ptr = user_alloc_internal(thr, pc, sz, align);
+  if (UNLIKELY(!ptr))
+    return errno_ENOMEM;
+  CHECK(IsAligned((uptr)ptr, align));
+  *memptr = ptr;
+  return 0;
+}
+
+void *user_aligned_alloc(ThreadState *thr, uptr pc, uptr align, uptr sz) {
+  if (UNLIKELY(!CheckAlignedAllocAlignmentAndSize(align, sz))) {
+    errno = errno_EINVAL;
+    return Allocator::FailureHandler::OnBadRequest();
+  }
+  return SetErrnoOnNull(user_alloc_internal(thr, pc, sz, align));
+}
+
+void *user_valloc(ThreadState *thr, uptr pc, uptr sz) {
+  return SetErrnoOnNull(user_alloc_internal(thr, pc, sz, GetPageSizeCached()));
+}
+
+void *user_pvalloc(ThreadState *thr, uptr pc, uptr sz) {
+  uptr PageSize = GetPageSizeCached();
+  if (UNLIKELY(CheckForPvallocOverflow(sz, PageSize))) {
+    errno = errno_ENOMEM;
+    return Allocator::FailureHandler::OnBadRequest();
+  }
+  // pvalloc(0) should allocate one page.
+  sz = sz ? RoundUpTo(sz, PageSize) : PageSize;
+  return SetErrnoOnNull(user_alloc_internal(thr, pc, sz, PageSize));
 }
 
 uptr user_alloc_usable_size(const void *p) {
diff --git a/lib/tsan/rtl/tsan_mman.h b/lib/tsan/rtl/tsan_mman.h
index 8cdeeb35aa6b..6042c5c5d96d 100644
--- a/lib/tsan/rtl/tsan_mman.h
+++ b/lib/tsan/rtl/tsan_mman.h
@@ -27,13 +27,20 @@ void AllocatorProcFinish(Processor *proc);
 void AllocatorPrintStats();
 
 // For user allocations.
-void *user_alloc(ThreadState *thr, uptr pc, uptr sz,
-                 uptr align = kDefaultAlignment, bool signal = true);
-void *user_calloc(ThreadState *thr, uptr pc, uptr sz, uptr n);
+void *user_alloc_internal(ThreadState *thr, uptr pc, uptr sz,
+                          uptr align = kDefaultAlignment, bool signal = true);
 // Does not accept NULL.
 void user_free(ThreadState *thr, uptr pc, void *p, bool signal = true);
+// Interceptor implementations.
+void *user_alloc(ThreadState *thr, uptr pc, uptr sz);
+void *user_calloc(ThreadState *thr, uptr pc, uptr sz, uptr n);
 void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz);
-void *user_alloc_aligned(ThreadState *thr, uptr pc, uptr sz, uptr align);
+void *user_memalign(ThreadState *thr, uptr pc, uptr align, uptr sz);
+int user_posix_memalign(ThreadState *thr, uptr pc, void **memptr, uptr align,
+                        uptr sz);
+void *user_aligned_alloc(ThreadState *thr, uptr pc, uptr align, uptr sz);
+void *user_valloc(ThreadState *thr, uptr pc, uptr sz);
+void *user_pvalloc(ThreadState *thr, uptr pc, uptr sz);
 uptr user_alloc_usable_size(const void *p);
 
 // Invoking malloc/free hooks that may be installed by the user.
diff --git a/lib/tsan/rtl/tsan_platform.h b/lib/tsan/rtl/tsan_platform.h
index bea1daba3952..bfac70df12d9 100644
--- a/lib/tsan/rtl/tsan_platform.h
+++ b/lib/tsan/rtl/tsan_platform.h
@@ -42,6 +42,19 @@ C/C++ on linux/x86_64 and freebsd/x86_64
 7b00 0000 0000 - 7c00 0000 0000: heap
 7c00 0000 0000 - 7e80 0000 0000: -
 7e80 0000 0000 - 8000 0000 0000: modules and main thread stack
+
+C/C++ on netbsd/amd64 can reuse the same mapping:
+ * The address space starts from 0x1000 (option with 0x0) and ends with
+   0x7f7ffffff000.
+ * LoAppMem-kHeapMemEnd can be reused as it is.
+ * No VDSO support.
+ * No MidAppMem region.
+ * No additional HeapMem region.
+ * HiAppMem contains the stack, loader, shared libraries and heap.
+ * Stack on NetBSD/amd64 has prereserved 128MB.
+ * Heap grows downwards (top-down).
+ * ASLR must be disabled per-process or globally.
+
 */
 struct Mapping {
   static const uptr kMetaShadowBeg = 0x300000000000ull;
@@ -111,7 +124,8 @@ C/C++ on Darwin/iOS/ARM64 (36-bit VMA, 64 GB VM)
 0c00 0000 00 - 0d00 0000 00: -                                    (4 GB)
 0d00 0000 00 - 0e00 0000 00: metainfo                             (4 GB)
 0e00 0000 00 - 0f00 0000 00: -                                    (4 GB)
-0f00 0000 00 - 1000 0000 00: traces                               (4 GB)
+0f00 0000 00 - 0fc0 0000 00: traces                               (3 GB)
+0fc0 0000 00 - 1000 0000 00: -
 */
 struct Mapping {
   static const uptr kLoAppMemBeg   = 0x0100000000ull;
@@ -123,9 +137,9 @@ struct Mapping {
   static const uptr kMetaShadowBeg = 0x0d00000000ull;
   static const uptr kMetaShadowEnd = 0x0e00000000ull;
   static const uptr kTraceMemBeg   = 0x0f00000000ull;
-  static const uptr kTraceMemEnd   = 0x1000000000ull;
-  static const uptr kHiAppMemBeg   = 0x1000000000ull;
-  static const uptr kHiAppMemEnd   = 0x1000000000ull;
+  static const uptr kTraceMemEnd   = 0x0fc0000000ull;
+  static const uptr kHiAppMemBeg   = 0x0fc0000000ull;
+  static const uptr kHiAppMemEnd   = 0x0fc0000000ull;
   static const uptr kAppMemMsk     =          0x0ull;
   static const uptr kAppMemXor     =          0x0ull;
   static const uptr kVdsoBeg       = 0x7000000000000000ull;
@@ -303,6 +317,38 @@ struct Mapping46 {
   static const uptr kVdsoBeg       = 0x7800000000000000ull;
 };
 
+/*
+C/C++ on linux/powerpc64 (47-bit VMA)
+0000 0000 1000 - 0100 0000 0000: main binary
+0100 0000 0000 - 0200 0000 0000: -
+0100 0000 0000 - 1000 0000 0000: shadow
+1000 0000 0000 - 1000 0000 0000: -
+1000 0000 0000 - 2000 0000 0000: metainfo (memory blocks and sync objects)
+2000 0000 0000 - 2000 0000 0000: -
+2000 0000 0000 - 2200 0000 0000: traces
+2200 0000 0000 - 7d00 0000 0000: -
+7d00 0000 0000 - 7e00 0000 0000: heap
+7e00 0000 0000 - 7e80 0000 0000: -
+7e80 0000 0000 - 8000 0000 0000: modules and main thread stack
+*/
+struct Mapping47 {
+  static const uptr kMetaShadowBeg = 0x100000000000ull;
+  static const uptr kMetaShadowEnd = 0x200000000000ull;
+  static const uptr kTraceMemBeg   = 0x200000000000ull;
+  static const uptr kTraceMemEnd   = 0x220000000000ull;
+  static const uptr kShadowBeg     = 0x010000000000ull;
+  static const uptr kShadowEnd     = 0x100000000000ull;
+  static const uptr kHeapMemBeg    = 0x7d0000000000ull;
+  static const uptr kHeapMemEnd    = 0x7e0000000000ull;
+  static const uptr kLoAppMemBeg   = 0x000000001000ull;
+  static const uptr kLoAppMemEnd   = 0x010000000000ull;
+  static const uptr kHiAppMemBeg   = 0x7e8000000000ull;
+  static const uptr kHiAppMemEnd   = 0x800000000000ull; // 47 bits
+  static const uptr kAppMemMsk     = 0x7c0000000000ull;
+  static const uptr kAppMemXor     = 0x020000000000ull;
+  static const uptr kVdsoBeg       = 0x7800000000000000ull;
+};
+
 // Indicates the runtime will define the memory regions at runtime.
 #define TSAN_RUNTIME_VMA 1
 #endif
@@ -429,11 +475,13 @@ uptr MappingArchImpl(void) {
   DCHECK(0);
   return 0;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return MappingImpl<Mapping44, Type>();
-  else
-    return MappingImpl<Mapping46, Type>();
+  switch (vmaSize) {
+    case 44: return MappingImpl<Mapping44, Type>();
+    case 46: return MappingImpl<Mapping46, Type>();
+    case 47: return MappingImpl<Mapping47, Type>();
+  }
   DCHECK(0);
+  return 0;
 #else
   return MappingImpl<Mapping, Type>();
 #endif
@@ -582,11 +630,13 @@ bool IsAppMem(uptr mem) {
   DCHECK(0);
   return false;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return IsAppMemImpl<Mapping44>(mem);
-  else
-    return IsAppMemImpl<Mapping46>(mem);
+  switch (vmaSize) {
+    case 44: return IsAppMemImpl<Mapping44>(mem);
+    case 46: return IsAppMemImpl<Mapping46>(mem);
+    case 47: return IsAppMemImpl<Mapping47>(mem);
+  }
   DCHECK(0);
+  return false;
 #else
   return IsAppMemImpl<Mapping>(mem);
 #endif
@@ -609,11 +659,13 @@ bool IsShadowMem(uptr mem) {
   DCHECK(0);
   return false;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return IsShadowMemImpl<Mapping44>(mem);
-  else
-    return IsShadowMemImpl<Mapping46>(mem);
+  switch (vmaSize) {
+    case 44: return IsShadowMemImpl<Mapping44>(mem);
+    case 46: return IsShadowMemImpl<Mapping46>(mem);
+    case 47: return IsShadowMemImpl<Mapping47>(mem);
+  }
   DCHECK(0);
+  return false;
 #else
   return IsShadowMemImpl<Mapping>(mem);
 #endif
@@ -636,11 +688,13 @@ bool IsMetaMem(uptr mem) {
   DCHECK(0);
   return false;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return IsMetaMemImpl<Mapping44>(mem);
-  else
-    return IsMetaMemImpl<Mapping46>(mem);
+  switch (vmaSize) {
+    case 44: return IsMetaMemImpl<Mapping44>(mem);
+    case 46: return IsMetaMemImpl<Mapping46>(mem);
+    case 47: return IsMetaMemImpl<Mapping47>(mem);
+  }
   DCHECK(0);
+  return false;
 #else
   return IsMetaMemImpl<Mapping>(mem);
 #endif
@@ -673,11 +727,13 @@ uptr MemToShadow(uptr x) {
   DCHECK(0);
   return 0;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return MemToShadowImpl<Mapping44>(x);
-  else
-    return MemToShadowImpl<Mapping46>(x);
+  switch (vmaSize) {
+    case 44: return MemToShadowImpl<Mapping44>(x);
+    case 46: return MemToShadowImpl<Mapping46>(x);
+    case 47: return MemToShadowImpl<Mapping47>(x);
+  }
   DCHECK(0);
+  return 0;
 #else
   return MemToShadowImpl<Mapping>(x);
 #endif
@@ -712,11 +768,13 @@ u32 *MemToMeta(uptr x) {
   DCHECK(0);
   return 0;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return MemToMetaImpl<Mapping44>(x);
-  else
-    return MemToMetaImpl<Mapping46>(x);
+  switch (vmaSize) {
+    case 44: return MemToMetaImpl<Mapping44>(x);
+    case 46: return MemToMetaImpl<Mapping46>(x);
+    case 47: return MemToMetaImpl<Mapping47>(x);
+  }
   DCHECK(0);
+  return 0;
 #else
   return MemToMetaImpl<Mapping>(x);
 #endif
@@ -764,11 +822,13 @@ uptr ShadowToMem(uptr s) {
   DCHECK(0);
   return 0;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return ShadowToMemImpl<Mapping44>(s);
-  else
-    return ShadowToMemImpl<Mapping46>(s);
+  switch (vmaSize) {
+    case 44: return ShadowToMemImpl<Mapping44>(s);
+    case 46: return ShadowToMemImpl<Mapping46>(s);
+    case 47: return ShadowToMemImpl<Mapping47>(s);
+  }
   DCHECK(0);
+  return 0;
 #else
   return ShadowToMemImpl<Mapping>(s);
 #endif
@@ -799,11 +859,13 @@ uptr GetThreadTrace(int tid) {
   DCHECK(0);
   return 0;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return GetThreadTraceImpl<Mapping44>(tid);
-  else
-    return GetThreadTraceImpl<Mapping46>(tid);
+  switch (vmaSize) {
+    case 44: return GetThreadTraceImpl<Mapping44>(tid);
+    case 46: return GetThreadTraceImpl<Mapping46>(tid);
+    case 47: return GetThreadTraceImpl<Mapping47>(tid);
+  }
   DCHECK(0);
+  return 0;
 #else
   return GetThreadTraceImpl<Mapping>(tid);
 #endif
@@ -829,11 +891,13 @@ uptr GetThreadTraceHeader(int tid) {
   DCHECK(0);
   return 0;
 #elif defined(__powerpc64__)
-  if (vmaSize == 44)
-    return GetThreadTraceHeaderImpl<Mapping44>(tid);
-  else
-    return GetThreadTraceHeaderImpl<Mapping46>(tid);
+  switch (vmaSize) {
+    case 44: return GetThreadTraceHeaderImpl<Mapping44>(tid);
+    case 46: return GetThreadTraceHeaderImpl<Mapping46>(tid);
+    case 47: return GetThreadTraceHeaderImpl<Mapping47>(tid);
+  }
   DCHECK(0);
+  return 0;
 #else
   return GetThreadTraceHeaderImpl<Mapping>(tid);
 #endif
diff --git a/lib/tsan/rtl/tsan_platform_linux.cc b/lib/tsan/rtl/tsan_platform_linux.cc
index ead1e5704989..e14d5f575a2e 100644
--- a/lib/tsan/rtl/tsan_platform_linux.cc
+++ b/lib/tsan/rtl/tsan_platform_linux.cc
@@ -14,11 +14,12 @@
 
 
 #include "sanitizer_common/sanitizer_platform.h"
-#if SANITIZER_LINUX || SANITIZER_FREEBSD
+#if SANITIZER_LINUX || SANITIZER_FREEBSD || SANITIZER_NETBSD
 
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_linux.h"
+#include "sanitizer_common/sanitizer_platform_limits_netbsd.h"
 #include "sanitizer_common/sanitizer_platform_limits_posix.h"
 #include "sanitizer_common/sanitizer_posix.h"
 #include "sanitizer_common/sanitizer_procmaps.h"
@@ -216,9 +217,9 @@ void InitializePlatformEarly() {
     Die();
   }
 #elif defined(__powerpc64__)
-  if (vmaSize != 44 && vmaSize != 46) {
+  if (vmaSize != 44 && vmaSize != 46 && vmaSize != 47) {
     Printf("FATAL: ThreadSanitizer: unsupported VMA range\n");
-    Printf("FATAL: Found %d - Supported 44 and 46\n", vmaSize);
+    Printf("FATAL: Found %d - Supported 44, 46, and 47\n", vmaSize);
     Die();
   }
 #endif
@@ -401,4 +402,4 @@ void cur_thread_finalize() {
 
 }  // namespace __tsan
 
-#endif  // SANITIZER_LINUX || SANITIZER_FREEBSD
+#endif  // SANITIZER_LINUX || SANITIZER_FREEBSD || SANITIZER_NETBSD
diff --git a/lib/tsan/rtl/tsan_platform_mac.cc b/lib/tsan/rtl/tsan_platform_mac.cc
index 73a656ffca5e..f8d7324b763c 100644
--- a/lib/tsan/rtl/tsan_platform_mac.cc
+++ b/lib/tsan/rtl/tsan_platform_mac.cc
@@ -167,8 +167,8 @@ void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
 #else  // !SANITIZER_GO
     "app      (0x%016zx-0x%016zx): resident %zd kB, dirty %zd kB\n"
 #endif
-    "stacks: %ld unique IDs, %ld kB allocated\n"
-    "threads: %ld total, %ld live\n"
+    "stacks: %zd unique IDs, %zd kB allocated\n"
+    "threads: %zd total, %zd live\n"
     "------------------------------\n",
     ShadowBeg(), ShadowEnd(), shadow_res / 1024, shadow_dirty / 1024,
     MetaShadowBeg(), MetaShadowEnd(), meta_res / 1024, meta_dirty / 1024,
@@ -231,7 +231,7 @@ static void my_pthread_introspection_hook(unsigned int event, pthread_t thread,
 
 void InitializePlatformEarly() {
 #if defined(__aarch64__)
-  uptr max_vm = GetMaxVirtualAddress() + 1;
+  uptr max_vm = GetMaxUserVirtualAddress() + 1;
   if (max_vm != Mapping::kHiAppMemEnd) {
     Printf("ThreadSanitizer: unsupported vm address limit %p, expected %p.\n",
            max_vm, Mapping::kHiAppMemEnd);
diff --git a/lib/tsan/rtl/tsan_report.cc b/lib/tsan/rtl/tsan_report.cc
index 32cc3325fe40..af47076968d3 100644
--- a/lib/tsan/rtl/tsan_report.cc
+++ b/lib/tsan/rtl/tsan_report.cc
@@ -13,6 +13,7 @@
 #include "tsan_report.h"
 #include "tsan_platform.h"
 #include "tsan_rtl.h"
+#include "sanitizer_common/sanitizer_file.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "sanitizer_common/sanitizer_report_decorator.h"
 #include "sanitizer_common/sanitizer_stacktrace_printer.h"
@@ -38,34 +39,27 @@ ReportLocation *ReportLocation::New(ReportLocationType type) {
 class Decorator: public __sanitizer::SanitizerCommonDecorator {
  public:
   Decorator() : SanitizerCommonDecorator() { }
-  const char *Warning()    { return Red(); }
-  const char *EndWarning() { return Default(); }
   const char *Access()     { return Blue(); }
-  const char *EndAccess()  { return Default(); }
   const char *ThreadDescription()    { return Cyan(); }
-  const char *EndThreadDescription() { return Default(); }
   const char *Location()   { return Green(); }
-  const char *EndLocation() { return Default(); }
   const char *Sleep()   { return Yellow(); }
-  const char *EndSleep() { return Default(); }
   const char *Mutex()   { return Magenta(); }
-  const char *EndMutex() { return Default(); }
 };
 
 ReportDesc::ReportDesc()
     : tag(kExternalTagNone)
-    , stacks(MBlockReportStack)
-    , mops(MBlockReportMop)
-    , locs(MBlockReportLoc)
-    , mutexes(MBlockReportMutex)
-    , threads(MBlockReportThread)
-    , unique_tids(MBlockReportThread)
+    , stacks()
+    , mops()
+    , locs()
+    , mutexes()
+    , threads()
+    , unique_tids()
     , sleep()
     , count() {
 }
 
 ReportMop::ReportMop()
-    : mset(MBlockReportMutex) {
+    : mset() {
 }
 
 ReportDesc::~ReportDesc() {
@@ -180,7 +174,7 @@ static void PrintMop(const ReportMop *mop, bool first) {
   }
   PrintMutexSet(mop->mset);
   Printf(":\n");
-  Printf("%s", d.EndAccess());
+  Printf("%s", d.Default());
   PrintStack(mop->stack);
 }
 
@@ -221,20 +215,20 @@ static void PrintLocation(const ReportLocation *loc) {
         loc->fd, thread_name(thrbuf, loc->tid));
     print_stack = true;
   }
-  Printf("%s", d.EndLocation());
+  Printf("%s", d.Default());
   if (print_stack)
     PrintStack(loc->stack);
 }
 
 static void PrintMutexShort(const ReportMutex *rm, const char *after) {
   Decorator d;
-  Printf("%sM%zd%s%s", d.Mutex(), rm->id, d.EndMutex(), after);
+  Printf("%sM%zd%s%s", d.Mutex(), rm->id, d.Default(), after);
 }
 
 static void PrintMutexShortWithAddress(const ReportMutex *rm,
                                        const char *after) {
   Decorator d;
-  Printf("%sM%zd (%p)%s%s", d.Mutex(), rm->id, rm->addr, d.EndMutex(), after);
+  Printf("%sM%zd (%p)%s%s", d.Mutex(), rm->id, rm->addr, d.Default(), after);
 }
 
 static void PrintMutex(const ReportMutex *rm) {
@@ -242,11 +236,11 @@ static void PrintMutex(const ReportMutex *rm) {
   if (rm->destroyed) {
     Printf("%s", d.Mutex());
     Printf("  Mutex M%llu is already destroyed.\n\n", rm->id);
-    Printf("%s", d.EndMutex());
+    Printf("%s", d.Default());
   } else {
     Printf("%s", d.Mutex());
     Printf("  Mutex M%llu (%p) created at:\n", rm->id, rm->addr);
-    Printf("%s", d.EndMutex());
+    Printf("%s", d.Default());
     PrintStack(rm->stack);
   }
 }
@@ -264,7 +258,7 @@ static void PrintThread(const ReportThread *rt) {
   if (rt->workerthread) {
     Printf(" (tid=%zu, %s) is a GCD worker thread\n", rt->os_id, thread_status);
     Printf("\n");
-    Printf("%s", d.EndThreadDescription());
+    Printf("%s", d.Default());
     return;
   }
   Printf(" (tid=%zu, %s) created by %s", rt->os_id, thread_status,
@@ -272,7 +266,7 @@ static void PrintThread(const ReportThread *rt) {
   if (rt->stack)
     Printf(" at:");
   Printf("\n");
-  Printf("%s", d.EndThreadDescription());
+  Printf("%s", d.Default());
   PrintStack(rt->stack);
 }
 
@@ -280,7 +274,7 @@ static void PrintSleep(const ReportStack *s) {
   Decorator d;
   Printf("%s", d.Sleep());
   Printf("  As if synchronized via sleep:\n");
-  Printf("%s", d.EndSleep());
+  Printf("%s", d.Default());
   PrintStack(s);
 }
 
@@ -324,7 +318,7 @@ void PrintReport(const ReportDesc *rep) {
   Printf("%s", d.Warning());
   Printf("WARNING: ThreadSanitizer: %s (pid=%d)\n", rep_typ_str,
          (int)internal_getpid());
-  Printf("%s", d.EndWarning());
+  Printf("%s", d.Default());
 
   if (rep->typ == ReportTypeDeadlock) {
     char thrbuf[kThreadBufSize];
@@ -342,7 +336,7 @@ void PrintReport(const ReportDesc *rep) {
       PrintMutexShort(rep->mutexes[i], " in ");
       Printf("%s", d.ThreadDescription());
       Printf("%s:\n", thread_name(thrbuf, rep->unique_tids[i]));
-      Printf("%s", d.EndThreadDescription());
+      Printf("%s", d.Default());
       if (flags()->second_deadlock_stack) {
         PrintStack(rep->stacks[2*i]);
         Printf("  Mutex ");
diff --git a/lib/tsan/rtl/tsan_report.h b/lib/tsan/rtl/tsan_report.h
index bc1582f90fe4..cdc999c6af1d 100644
--- a/lib/tsan/rtl/tsan_report.h
+++ b/lib/tsan/rtl/tsan_report.h
@@ -14,8 +14,8 @@
 #define TSAN_REPORT_H
 
 #include "sanitizer_common/sanitizer_symbolizer.h"
+#include "sanitizer_common/sanitizer_vector.h"
 #include "tsan_defs.h"
-#include "tsan_vector.h"
 
 namespace __tsan {
 
diff --git a/lib/tsan/rtl/tsan_rtl.cc b/lib/tsan/rtl/tsan_rtl.cc
index a01525302b02..17b820977c26 100644
--- a/lib/tsan/rtl/tsan_rtl.cc
+++ b/lib/tsan/rtl/tsan_rtl.cc
@@ -14,6 +14,7 @@
 
 #include "sanitizer_common/sanitizer_atomic.h"
 #include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_file.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_stackdepot.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
@@ -101,8 +102,8 @@ Context::Context()
   , thread_registry(new(thread_registry_placeholder) ThreadRegistry(
       CreateThreadContext, kMaxTid, kThreadQuarantineSize, kMaxTidReuse))
   , racy_mtx(MutexTypeRacy, StatMtxRacy)
-  , racy_stacks(MBlockRacyStacks)
-  , racy_addresses(MBlockRacyAddresses)
+  , racy_stacks()
+  , racy_addresses()
   , fired_suppressions_mtx(MutexTypeFired, StatMtxFired)
   , fired_suppressions(8)
   , clock_alloc("clock allocator") {
@@ -120,7 +121,7 @@ ThreadState::ThreadState(Context *ctx, int tid, int unique_id, u64 epoch,
   // , ignore_interceptors()
   , clock(tid, reuse_count)
 #if !SANITIZER_GO
-  , jmp_bufs(MBlockJmpBuf)
+  , jmp_bufs()
 #endif
   , tid(tid)
   , unique_id(unique_id)
@@ -213,7 +214,7 @@ static void BackgroundThread(void *arg) {
                              memory_order_relaxed);
       if (last != 0 && last + flags()->flush_symbolizer_ms * kMs2Ns < now) {
         Lock l(&ctx->report_mtx);
-        SpinMutexLock l2(&CommonSanitizerReportMutex);
+        ScopedErrorReportLock l2;
         SymbolizeFlush();
         atomic_store(&ctx->last_symbolize_time_ns, 0, memory_order_relaxed);
       }
@@ -323,6 +324,21 @@ static void CheckShadowMapping() {
   }
 }
 
+#if !SANITIZER_GO
+static void OnStackUnwind(const SignalContext &sig, const void *,
+                          BufferedStackTrace *stack) {
+  uptr top = 0;
+  uptr bottom = 0;
+  bool fast = common_flags()->fast_unwind_on_fatal;
+  if (fast) GetThreadStackTopAndBottom(false, &top, &bottom);
+  stack->Unwind(kStackTraceMax, sig.pc, sig.bp, sig.context, top, bottom, fast);
+}
+
+static void TsanOnDeadlySignal(int signo, void *siginfo, void *context) {
+  HandleDeadlySignal(siginfo, context, GetTid(), &OnStackUnwind, nullptr);
+}
+#endif
+
 void Initialize(ThreadState *thr) {
   // Thread safe because done before all threads exist.
   static bool is_initialized = false;
@@ -360,6 +376,7 @@ void Initialize(ThreadState *thr) {
 #if !SANITIZER_GO
   InitializeShadowMemory();
   InitializeAllocatorLate();
+  InstallDeadlySignalHandlers(TsanOnDeadlySignal);
 #endif
   // Setup correct file descriptor for error reports.
   __sanitizer_set_report_path(common_flags()->log_path);
@@ -367,13 +384,6 @@ void Initialize(ThreadState *thr) {
 #if !SANITIZER_GO
   InitializeLibIgnore();
   Symbolizer::GetOrInit()->AddHooks(EnterSymbolizer, ExitSymbolizer);
-  // On MIPS, TSan initialization is run before
-  // __pthread_initialize_minimal_internal() is finished, so we can not spawn
-  // new threads.
-#ifndef __mips__
-  StartBackgroundThread();
-  SetSandboxingCallback(StopBackgroundThread);
-#endif
 #endif
 
   VPrintf(1, "***** Running under ThreadSanitizer v2 (pid %d) *****\n",
@@ -402,6 +412,21 @@ void Initialize(ThreadState *thr) {
   OnInitialize();
 }
 
+void MaybeSpawnBackgroundThread() {
+  // On MIPS, TSan initialization is run before
+  // __pthread_initialize_minimal_internal() is finished, so we can not spawn
+  // new threads.
+#if !SANITIZER_GO && !defined(__mips__)
+  static atomic_uint32_t bg_thread = {};
+  if (atomic_load(&bg_thread, memory_order_relaxed) == 0 &&
+      atomic_exchange(&bg_thread, 1, memory_order_relaxed) == 0) {
+    StartBackgroundThread();
+    SetSandboxingCallback(StopBackgroundThread);
+  }
+#endif
+}
+
+
 int Finalize(ThreadState *thr) {
   bool failed = false;
 
@@ -412,8 +437,7 @@ int Finalize(ThreadState *thr) {
 
   // Wait for pending reports.
   ctx->report_mtx.Lock();
-  CommonSanitizerReportMutex.Lock();
-  CommonSanitizerReportMutex.Unlock();
+  { ScopedErrorReportLock l; }
   ctx->report_mtx.Unlock();
 
 #if !SANITIZER_GO
diff --git a/lib/tsan/rtl/tsan_rtl.h b/lib/tsan/rtl/tsan_rtl.h
index 2cf2e168454d..7d44ccae9893 100644
--- a/lib/tsan/rtl/tsan_rtl.h
+++ b/lib/tsan/rtl/tsan_rtl.h
@@ -34,12 +34,13 @@
 #include "sanitizer_common/sanitizer_libignore.h"
 #include "sanitizer_common/sanitizer_suppressions.h"
 #include "sanitizer_common/sanitizer_thread_registry.h"
+#include "sanitizer_common/sanitizer_vector.h"
 #include "tsan_clock.h"
 #include "tsan_defs.h"
 #include "tsan_flags.h"
+#include "tsan_mman.h"
 #include "tsan_sync.h"
 #include "tsan_trace.h"
-#include "tsan_vector.h"
 #include "tsan_report.h"
 #include "tsan_platform.h"
 #include "tsan_mutexset.h"
@@ -574,11 +575,8 @@ const char *GetObjectTypeFromTag(uptr tag);
 const char *GetReportHeaderFromTag(uptr tag);
 uptr TagFromShadowStackFrame(uptr pc);
 
-class ScopedReport {
+class ScopedReportBase {
  public:
-  explicit ScopedReport(ReportType typ, uptr tag = kExternalTagNone);
-  ~ScopedReport();
-
   void AddMemoryAccess(uptr addr, uptr external_tag, Shadow s, StackTrace stack,
                        const MutexSet *mset);
   void AddStack(StackTrace stack, bool suppressable = false);
@@ -593,6 +591,10 @@ class ScopedReport {
 
   const ReportDesc *GetReport() const;
 
+ protected:
+  ScopedReportBase(ReportType typ, uptr tag);
+  ~ScopedReportBase();
+
  private:
   ReportDesc *rep_;
   // Symbolizer makes lots of intercepted calls. If we try to process them,
@@ -601,8 +603,17 @@ class ScopedReport {
 
   void AddDeadMutex(u64 id);
 
-  ScopedReport(const ScopedReport&);
-  void operator = (const ScopedReport&);
+  ScopedReportBase(const ScopedReportBase &) = delete;
+  void operator=(const ScopedReportBase &) = delete;
+};
+
+class ScopedReport : public ScopedReportBase {
+ public:
+  explicit ScopedReport(ReportType typ, uptr tag = kExternalTagNone);
+  ~ScopedReport();
+
+ private:
+  ScopedErrorReportLock lock_;
 };
 
 ThreadContext *IsThreadStackOrTls(uptr addr, bool *is_stack);
@@ -690,6 +701,7 @@ void PrintCurrentStack(ThreadState *thr, uptr pc);
 void PrintCurrentStackSlow(uptr pc);  // uses libunwind
 
 void Initialize(ThreadState *thr);
+void MaybeSpawnBackgroundThread();
 int Finalize(ThreadState *thr);
 
 void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write);
@@ -825,7 +837,7 @@ void ALWAYS_INLINE TraceAddEvent(ThreadState *thr, FastState fs,
     return;
   DCHECK_GE((int)typ, 0);
   DCHECK_LE((int)typ, 7);
-  DCHECK_EQ(GetLsb(addr, 61), addr);
+  DCHECK_EQ(GetLsb(addr, kEventPCBits), addr);
   StatInc(thr, StatEvents);
   u64 pos = fs.GetTracePos();
   if (UNLIKELY((pos % kTracePartSize) == 0)) {
@@ -837,7 +849,7 @@ void ALWAYS_INLINE TraceAddEvent(ThreadState *thr, FastState fs,
   }
   Event *trace = (Event*)GetThreadTrace(fs.tid());
   Event *evp = &trace[pos];
-  Event ev = (u64)addr | ((u64)typ << 61);
+  Event ev = (u64)addr | ((u64)typ << kEventPCBits);
   *evp = ev;
 }
 
diff --git a/lib/tsan/rtl/tsan_rtl_aarch64.S b/lib/tsan/rtl/tsan_rtl_aarch64.S
index 61171d635c18..844c2e23db4f 100644
--- a/lib/tsan/rtl/tsan_rtl_aarch64.S
+++ b/lib/tsan/rtl/tsan_rtl_aarch64.S
@@ -6,7 +6,7 @@
 #if !defined(__APPLE__)
 .section .bss
 .type	__tsan_pointer_chk_guard, %object
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(__tsan_pointer_chk_guard))
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(__tsan_pointer_chk_guard))
 __tsan_pointer_chk_guard:
 .zero	8
 #endif
@@ -51,7 +51,7 @@ _sigsetjmp$non_lazy_ptr:
 // original ones.
 ASM_HIDDEN(_Z18InitializeGuardPtrv)
 .global _Z18InitializeGuardPtrv
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(_Z18InitializeGuardPtrv))
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(_Z18InitializeGuardPtrv))
 _Z18InitializeGuardPtrv:
   CFI_STARTPROC
   // Allocates a jmp_buf for the setjmp call.
@@ -88,14 +88,14 @@ _Z18InitializeGuardPtrv:
   CFI_DEF_CFA (31, 0)
   ret
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(_Z18InitializeGuardPtrv))
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(_Z18InitializeGuardPtrv))
 #endif
 
 ASM_HIDDEN(__tsan_setjmp)
 .comm _ZN14__interception11real_setjmpE,8,8
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp):
+.globl ASM_SYMBOL_INTERCEPTOR(setjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(setjmp))
+ASM_SYMBOL_INTERCEPTOR(setjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -125,7 +125,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp):
 #endif
 
   // call tsan interceptor
-  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
+  bl      ASM_SYMBOL(__tsan_setjmp)
 
   // restore env parameter
   mov     x0, x19
@@ -148,12 +148,12 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp):
   br      x1
 
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp))
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(setjmp))
 
 .comm _ZN14__interception12real__setjmpE,8,8
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp):
+.globl ASM_SYMBOL_INTERCEPTOR(_setjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(_setjmp))
+ASM_SYMBOL_INTERCEPTOR(_setjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -183,7 +183,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp):
 #endif
 
   // call tsan interceptor
-  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
+  bl      ASM_SYMBOL(__tsan_setjmp)
 
   // Restore jmp_buf parameter
   mov     x0, x19
@@ -206,12 +206,12 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp):
   br      x1
 
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp))
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(_setjmp))
 
 .comm _ZN14__interception14real_sigsetjmpE,8,8
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
+.globl ASM_SYMBOL_INTERCEPTOR(sigsetjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(sigsetjmp))
+ASM_SYMBOL_INTERCEPTOR(sigsetjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -243,7 +243,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
 #endif
 
   // call tsan interceptor
-  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
+  bl      ASM_SYMBOL(__tsan_setjmp)
 
   // restore env parameter
   mov     w1, w20
@@ -268,13 +268,13 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
 #endif
   br      x2
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp))
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(sigsetjmp))
 
 #if !defined(__APPLE__)
 .comm _ZN14__interception16real___sigsetjmpE,8,8
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp):
+.globl ASM_SYMBOL_INTERCEPTOR(__sigsetjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(__sigsetjmp))
+ASM_SYMBOL_INTERCEPTOR(__sigsetjmp):
   CFI_STARTPROC
 
   // save env parameters for function call
@@ -303,7 +303,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp):
 #endif
 
   // call tsan interceptor
-  bl      ASM_TSAN_SYMBOL(__tsan_setjmp)
+  bl      ASM_SYMBOL(__tsan_setjmp)
 
   mov     w1, w20
   mov     x0, x19
@@ -321,12 +321,12 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp):
   ldr     x2, [x2, #:got_lo12:_ZN14__interception16real___sigsetjmpE]
   ldr     x2, [x2]
 #else
-  adrp    x2, ASM_TSAN_SYMBOL(__sigsetjmp)@page
-  add     x2, x2, ASM_TSAN_SYMBOL(__sigsetjmp)@pageoff
+  adrp    x2, ASM_SYMBOL(__sigsetjmp)@page
+  add     x2, x2, ASM_SYMBOL(__sigsetjmp)@pageoff
 #endif
   br      x2
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp))
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(__sigsetjmp))
 #endif
 
 #if defined(__linux__)
diff --git a/lib/tsan/rtl/tsan_rtl_amd64.S b/lib/tsan/rtl/tsan_rtl_amd64.S
index 98947fd2a1ba..8af61bf0e892 100644
--- a/lib/tsan/rtl/tsan_rtl_amd64.S
+++ b/lib/tsan/rtl/tsan_rtl_amd64.S
@@ -10,8 +10,8 @@
 #endif
 
 ASM_HIDDEN(__tsan_trace_switch)
-.globl ASM_TSAN_SYMBOL(__tsan_trace_switch_thunk)
-ASM_TSAN_SYMBOL(__tsan_trace_switch_thunk):
+.globl ASM_SYMBOL(__tsan_trace_switch_thunk)
+ASM_SYMBOL(__tsan_trace_switch_thunk):
   CFI_STARTPROC
   # Save scratch registers.
   push %rax
@@ -50,7 +50,7 @@ ASM_TSAN_SYMBOL(__tsan_trace_switch_thunk):
   shr $4, %rsp  # clear 4 lsb, align to 16
   shl $4, %rsp
 
-  call ASM_TSAN_SYMBOL(__tsan_trace_switch)
+  call ASM_SYMBOL(__tsan_trace_switch)
 
   # Unalign stack frame back.
   mov %rbx, %rsp  # restore the original rsp
@@ -90,8 +90,8 @@ ASM_TSAN_SYMBOL(__tsan_trace_switch_thunk):
   CFI_ENDPROC
 
 ASM_HIDDEN(__tsan_report_race)
-.globl ASM_TSAN_SYMBOL(__tsan_report_race_thunk)
-ASM_TSAN_SYMBOL(__tsan_report_race_thunk):
+.globl ASM_SYMBOL(__tsan_report_race_thunk)
+ASM_SYMBOL(__tsan_report_race_thunk):
   CFI_STARTPROC
   # Save scratch registers.
   push %rax
@@ -130,7 +130,7 @@ ASM_TSAN_SYMBOL(__tsan_report_race_thunk):
   shr $4, %rsp  # clear 4 lsb, align to 16
   shl $4, %rsp
 
-  call ASM_TSAN_SYMBOL(__tsan_report_race)
+  call ASM_SYMBOL(__tsan_report_race)
 
   # Unalign stack frame back.
   mov %rbx, %rsp  # restore the original rsp
@@ -170,19 +170,27 @@ ASM_TSAN_SYMBOL(__tsan_report_race_thunk):
   CFI_ENDPROC
 
 ASM_HIDDEN(__tsan_setjmp)
-#if !defined(__APPLE__)
+#if defined(__NetBSD__)
+.comm _ZN14__interception15real___setjmp14E,8,8
+#elif !defined(__APPLE__)
 .comm _ZN14__interception11real_setjmpE,8,8
 #endif
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp):
+#if defined(__NetBSD__)
+.globl ASM_SYMBOL_INTERCEPTOR(__setjmp14)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(__setjmp14))
+ASM_SYMBOL_INTERCEPTOR(__setjmp14):
+#else
+.globl ASM_SYMBOL_INTERCEPTOR(setjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(setjmp))
+ASM_SYMBOL_INTERCEPTOR(setjmp):
+#endif
   CFI_STARTPROC
   // save env parameter
   push %rdi
   CFI_ADJUST_CFA_OFFSET(8)
   CFI_REL_OFFSET(%rdi, 0)
   // obtain %rsp
-#if defined(__FreeBSD__)
+#if defined(__FreeBSD__) || defined(__NetBSD__)
   lea 8(%rsp), %rdi
   mov %rdi, %rsi
 #elif defined(__APPLE__)
@@ -197,33 +205,40 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp):
 # error "Unknown platform"
 #endif
   // call tsan interceptor
-  call ASM_TSAN_SYMBOL(__tsan_setjmp)
+  call ASM_SYMBOL(__tsan_setjmp)
   // restore env parameter
   pop %rdi
   CFI_ADJUST_CFA_OFFSET(-8)
   CFI_RESTORE(%rdi)
   // tail jump to libc setjmp
   movl $0, %eax
-#if !defined(__APPLE__)
+#if defined(__NetBSD__)
+  movq _ZN14__interception15real___setjmp14E@GOTPCREL(%rip), %rdx
+  jmp *(%rdx)
+#elif !defined(__APPLE__)
   movq _ZN14__interception11real_setjmpE@GOTPCREL(%rip), %rdx
   jmp *(%rdx)
 #else
-  jmp ASM_TSAN_SYMBOL(setjmp)
+  jmp ASM_SYMBOL(setjmp)
 #endif
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(setjmp))
+#if defined(__NetBSD__)
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(__setjmp14))
+#else
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(setjmp))
+#endif
 
 .comm _ZN14__interception12real__setjmpE,8,8
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp):
+.globl ASM_SYMBOL_INTERCEPTOR(_setjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(_setjmp))
+ASM_SYMBOL_INTERCEPTOR(_setjmp):
   CFI_STARTPROC
   // save env parameter
   push %rdi
   CFI_ADJUST_CFA_OFFSET(8)
   CFI_REL_OFFSET(%rdi, 0)
   // obtain %rsp
-#if defined(__FreeBSD__)
+#if defined(__FreeBSD__) || defined(__NetBSD__)
   lea 8(%rsp), %rdi
   mov %rdi, %rsi
 #elif defined(__APPLE__)
@@ -238,7 +253,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp):
 # error "Unknown platform"
 #endif
   // call tsan interceptor
-  call ASM_TSAN_SYMBOL(__tsan_setjmp)
+  call ASM_SYMBOL(__tsan_setjmp)
   // restore env parameter
   pop %rdi
   CFI_ADJUST_CFA_OFFSET(-8)
@@ -249,15 +264,22 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp):
   movq _ZN14__interception12real__setjmpE@GOTPCREL(%rip), %rdx
   jmp *(%rdx)
 #else
-  jmp ASM_TSAN_SYMBOL(_setjmp)
+  jmp ASM_SYMBOL(_setjmp)
 #endif
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(_setjmp))
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(_setjmp))
 
+#if defined(__NetBSD__)
+.comm _ZN14__interception18real___sigsetjmp14E,8,8
+.globl ASM_SYMBOL_INTERCEPTOR(__sigsetjmp14)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(__sigsetjmp14))
+ASM_SYMBOL_INTERCEPTOR(__sigsetjmp14):
+#else
 .comm _ZN14__interception14real_sigsetjmpE,8,8
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
+.globl ASM_SYMBOL_INTERCEPTOR(sigsetjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(sigsetjmp))
+ASM_SYMBOL_INTERCEPTOR(sigsetjmp):
+#endif
   CFI_STARTPROC
   // save env parameter
   push %rdi
@@ -271,7 +293,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
   sub $8, %rsp
   CFI_ADJUST_CFA_OFFSET(8)
   // obtain %rsp
-#if defined(__FreeBSD__)
+#if defined(__FreeBSD__) || defined(__NetBSD__)
   lea 24(%rsp), %rdi
   mov %rdi, %rsi
 #elif defined(__APPLE__)
@@ -286,7 +308,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
 # error "Unknown platform"
 #endif
   // call tsan interceptor
-  call ASM_TSAN_SYMBOL(__tsan_setjmp)
+  call ASM_SYMBOL(__tsan_setjmp)
   // unalign stack frame
   add $8, %rsp
   CFI_ADJUST_CFA_OFFSET(-8)
@@ -300,20 +322,27 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp):
   CFI_RESTORE(%rdi)
   // tail jump to libc sigsetjmp
   movl $0, %eax
-#if !defined(__APPLE__)
+#if defined(__NetBSD__)
+  movq _ZN14__interception18real___sigsetjmp14E@GOTPCREL(%rip), %rdx
+  jmp *(%rdx)
+#elif !defined(__APPLE__)
   movq _ZN14__interception14real_sigsetjmpE@GOTPCREL(%rip), %rdx
   jmp *(%rdx)
 #else
-  jmp ASM_TSAN_SYMBOL(sigsetjmp)
+  jmp ASM_SYMBOL(sigsetjmp)
 #endif
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(sigsetjmp))
+#if defined(__NetBSD__)
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(__sigsetjmp14))
+#else
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(sigsetjmp))
+#endif
 
-#if !defined(__APPLE__)
+#if !defined(__APPLE__) && !defined(__NetBSD__)
 .comm _ZN14__interception16real___sigsetjmpE,8,8
-.globl ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp)
-ASM_TYPE_FUNCTION(ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp))
-ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp):
+.globl ASM_SYMBOL_INTERCEPTOR(__sigsetjmp)
+ASM_TYPE_FUNCTION(ASM_SYMBOL_INTERCEPTOR(__sigsetjmp))
+ASM_SYMBOL_INTERCEPTOR(__sigsetjmp):
   CFI_STARTPROC
   // save env parameter
   push %rdi
@@ -337,7 +366,7 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp):
   rol $0x11, %rsi
 #endif
   // call tsan interceptor
-  call ASM_TSAN_SYMBOL(__tsan_setjmp)
+  call ASM_SYMBOL(__tsan_setjmp)
   // unalign stack frame
   add $8, %rsp
   CFI_ADJUST_CFA_OFFSET(-8)
@@ -354,11 +383,12 @@ ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp):
   movq _ZN14__interception16real___sigsetjmpE@GOTPCREL(%rip), %rdx
   jmp *(%rdx)
   CFI_ENDPROC
-ASM_SIZE(ASM_TSAN_SYMBOL_INTERCEPTOR(__sigsetjmp))
-#endif  // !defined(__APPLE__)
+ASM_SIZE(ASM_SYMBOL_INTERCEPTOR(__sigsetjmp))
+#endif  // !defined(__APPLE__) && !defined(__NetBSD__)
 
 #if defined(__FreeBSD__) || defined(__linux__)
 /* We do not need executable stack.  */
+/* This note is not needed on NetBSD. */
 .section        .note.GNU-stack,"",@progbits
 #endif
 
diff --git a/lib/tsan/rtl/tsan_rtl_mutex.cc b/lib/tsan/rtl/tsan_rtl_mutex.cc
index 2f85811620f1..152b965ad535 100644
--- a/lib/tsan/rtl/tsan_rtl_mutex.cc
+++ b/lib/tsan/rtl/tsan_rtl_mutex.cc
@@ -84,7 +84,9 @@ void MutexDestroy(ThreadState *thr, uptr pc, uptr addr, u32 flagz) {
   SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr, true);
   if (s == 0)
     return;
-  if ((flagz & MutexFlagLinkerInit) || s->IsFlagSet(MutexFlagLinkerInit)) {
+  if ((flagz & MutexFlagLinkerInit)
+      || s->IsFlagSet(MutexFlagLinkerInit)
+      || ((flagz & MutexFlagNotStatic) && !s->IsFlagSet(MutexFlagNotStatic))) {
     // Destroy is no-op for linker-initialized mutexes.
     s->mtx.Unlock();
     return;
diff --git a/lib/tsan/rtl/tsan_rtl_report.cc b/lib/tsan/rtl/tsan_rtl_report.cc
index 85a982941ed7..cc582ab50190 100644
--- a/lib/tsan/rtl/tsan_rtl_report.cc
+++ b/lib/tsan/rtl/tsan_rtl_report.cc
@@ -143,30 +143,28 @@ static ReportStack *SymbolizeStack(StackTrace trace) {
   return stack;
 }
 
-ScopedReport::ScopedReport(ReportType typ, uptr tag) {
+ScopedReportBase::ScopedReportBase(ReportType typ, uptr tag) {
   ctx->thread_registry->CheckLocked();
   void *mem = internal_alloc(MBlockReport, sizeof(ReportDesc));
   rep_ = new(mem) ReportDesc;
   rep_->typ = typ;
   rep_->tag = tag;
   ctx->report_mtx.Lock();
-  CommonSanitizerReportMutex.Lock();
 }
 
-ScopedReport::~ScopedReport() {
-  CommonSanitizerReportMutex.Unlock();
+ScopedReportBase::~ScopedReportBase() {
   ctx->report_mtx.Unlock();
   DestroyAndFree(rep_);
 }
 
-void ScopedReport::AddStack(StackTrace stack, bool suppressable) {
+void ScopedReportBase::AddStack(StackTrace stack, bool suppressable) {
   ReportStack **rs = rep_->stacks.PushBack();
   *rs = SymbolizeStack(stack);
   (*rs)->suppressable = suppressable;
 }
 
-void ScopedReport::AddMemoryAccess(uptr addr, uptr external_tag, Shadow s,
-                                   StackTrace stack, const MutexSet *mset) {
+void ScopedReportBase::AddMemoryAccess(uptr addr, uptr external_tag, Shadow s,
+                                       StackTrace stack, const MutexSet *mset) {
   void *mem = internal_alloc(MBlockReportMop, sizeof(ReportMop));
   ReportMop *mop = new(mem) ReportMop;
   rep_->mops.PushBack(mop);
@@ -187,11 +185,11 @@ void ScopedReport::AddMemoryAccess(uptr addr, uptr external_tag, Shadow s,
   }
 }
 
-void ScopedReport::AddUniqueTid(int unique_tid) {
+void ScopedReportBase::AddUniqueTid(int unique_tid) {
   rep_->unique_tids.PushBack(unique_tid);
 }
 
-void ScopedReport::AddThread(const ThreadContext *tctx, bool suppressable) {
+void ScopedReportBase::AddThread(const ThreadContext *tctx, bool suppressable) {
   for (uptr i = 0; i < rep_->threads.Size(); i++) {
     if ((u32)rep_->threads[i]->id == tctx->tid)
       return;
@@ -255,14 +253,14 @@ ThreadContext *IsThreadStackOrTls(uptr addr, bool *is_stack) {
 }
 #endif
 
-void ScopedReport::AddThread(int unique_tid, bool suppressable) {
+void ScopedReportBase::AddThread(int unique_tid, bool suppressable) {
 #if !SANITIZER_GO
   if (const ThreadContext *tctx = FindThreadByUidLocked(unique_tid))
     AddThread(tctx, suppressable);
 #endif
 }
 
-void ScopedReport::AddMutex(const SyncVar *s) {
+void ScopedReportBase::AddMutex(const SyncVar *s) {
   for (uptr i = 0; i < rep_->mutexes.Size(); i++) {
     if (rep_->mutexes[i]->id == s->uid)
       return;
@@ -276,7 +274,7 @@ void ScopedReport::AddMutex(const SyncVar *s) {
   rm->stack = SymbolizeStackId(s->creation_stack_id);
 }
 
-u64 ScopedReport::AddMutex(u64 id) {
+u64 ScopedReportBase::AddMutex(u64 id) {
   u64 uid = 0;
   u64 mid = id;
   uptr addr = SyncVar::SplitId(id, &uid);
@@ -295,7 +293,7 @@ u64 ScopedReport::AddMutex(u64 id) {
   return mid;
 }
 
-void ScopedReport::AddDeadMutex(u64 id) {
+void ScopedReportBase::AddDeadMutex(u64 id) {
   for (uptr i = 0; i < rep_->mutexes.Size(); i++) {
     if (rep_->mutexes[i]->id == id)
       return;
@@ -309,7 +307,7 @@ void ScopedReport::AddDeadMutex(u64 id) {
   rm->stack = 0;
 }
 
-void ScopedReport::AddLocation(uptr addr, uptr size) {
+void ScopedReportBase::AddLocation(uptr addr, uptr size) {
   if (addr == 0)
     return;
 #if !SANITIZER_GO
@@ -364,18 +362,19 @@ void ScopedReport::AddLocation(uptr addr, uptr size) {
 }
 
 #if !SANITIZER_GO
-void ScopedReport::AddSleep(u32 stack_id) {
+void ScopedReportBase::AddSleep(u32 stack_id) {
   rep_->sleep = SymbolizeStackId(stack_id);
 }
 #endif
 
-void ScopedReport::SetCount(int count) {
-  rep_->count = count;
-}
+void ScopedReportBase::SetCount(int count) { rep_->count = count; }
 
-const ReportDesc *ScopedReport::GetReport() const {
-  return rep_;
-}
+const ReportDesc *ScopedReportBase::GetReport() const { return rep_; }
+
+ScopedReport::ScopedReport(ReportType typ, uptr tag)
+    : ScopedReportBase(typ, tag) {}
+
+ScopedReport::~ScopedReport() {}
 
 void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
                   MutexSet *mset, uptr *tag) {
@@ -394,7 +393,7 @@ void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
   const u64 ebegin = RoundDown(eend, kTracePartSize);
   DPrintf("#%d: RestoreStack epoch=%zu ebegin=%zu eend=%zu partidx=%d\n",
           tid, (uptr)epoch, (uptr)ebegin, (uptr)eend, partidx);
-  Vector<uptr> stack(MBlockReportStack);
+  Vector<uptr> stack;
   stack.Resize(hdr->stack0.size + 64);
   for (uptr i = 0; i < hdr->stack0.size; i++) {
     stack[i] = hdr->stack0.trace[i];
@@ -406,8 +405,8 @@ void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
   Event *events = (Event*)GetThreadTrace(tid);
   for (uptr i = ebegin; i <= eend; i++) {
     Event ev = events[i];
-    EventType typ = (EventType)(ev >> 61);
-    uptr pc = (uptr)(ev & ((1ull << 61) - 1));
+    EventType typ = (EventType)(ev >> kEventPCBits);
+    uptr pc = (uptr)(ev & ((1ull << kEventPCBits) - 1));
     DPrintf2("  %zu typ=%d pc=%zx\n", i, typ, pc);
     if (typ == EventTypeMop) {
       stack[pos] = pc;
@@ -634,13 +633,33 @@ void ReportRace(ThreadState *thr) {
   const uptr kMop = 2;
   VarSizeStackTrace traces[kMop];
   uptr tags[kMop] = {kExternalTagNone};
-  const uptr toppc = TraceTopPC(thr);
+  uptr toppc = TraceTopPC(thr);
+  if (toppc >> kEventPCBits) {
+    // This is a work-around for a known issue.
+    // The scenario where this happens is rather elaborate and requires
+    // an instrumented __sanitizer_report_error_summary callback and
+    // a __tsan_symbolize_external callback and a race during a range memory
+    // access larger than 8 bytes. MemoryAccessRange adds the current PC to
+    // the trace and starts processing memory accesses. A first memory access
+    // triggers a race, we report it and call the instrumented
+    // __sanitizer_report_error_summary, which adds more stuff to the trace
+    // since it is intrumented. Then a second memory access in MemoryAccessRange
+    // also triggers a race and we get here and call TraceTopPC to get the
+    // current PC, however now it contains some unrelated events from the
+    // callback. Most likely, TraceTopPC will now return a EventTypeFuncExit
+    // event. Later we subtract -1 from it (in GetPreviousInstructionPc)
+    // and the resulting PC has kExternalPCBit set, so we pass it to
+    // __tsan_symbolize_external. __tsan_symbolize_external is within its rights
+    // to crash since the PC is completely bogus.
+    // test/tsan/double_race.cc contains a test case for this.
+    toppc = 0;
+  }
   ObtainCurrentStack(thr, toppc, &traces[0], &tags[0]);
   if (IsFiredSuppression(ctx, typ, traces[0]))
     return;
 
   // MutexSet is too large to live on stack.
-  Vector<u64> mset_buffer(MBlockScopedBuf);
+  Vector<u64> mset_buffer;
   mset_buffer.Resize(sizeof(MutexSet) / sizeof(u64) + 1);
   MutexSet *mset2 = new(&mset_buffer[0]) MutexSet();
 
diff --git a/lib/tsan/rtl/tsan_rtl_thread.cc b/lib/tsan/rtl/tsan_rtl_thread.cc
index 83fab082afe3..e4d65b9a909b 100644
--- a/lib/tsan/rtl/tsan_rtl_thread.cc
+++ b/lib/tsan/rtl/tsan_rtl_thread.cc
@@ -211,7 +211,7 @@ void ThreadFinalize(ThreadState *thr) {
   if (!flags()->report_thread_leaks)
     return;
   ThreadRegistryLock l(ctx->thread_registry);
-  Vector<ThreadLeak> leaks(MBlockScopedBuf);
+  Vector<ThreadLeak> leaks;
   ctx->thread_registry->RunCallbackForEachThreadLocked(
       MaybeReportThreadLeak, &leaks);
   for (uptr i = 0; i < leaks.Size(); i++) {
diff --git a/lib/tsan/rtl/tsan_sync.h b/lib/tsan/rtl/tsan_sync.h
index b83c09ff784e..9039970bcf86 100644
--- a/lib/tsan/rtl/tsan_sync.h
+++ b/lib/tsan/rtl/tsan_sync.h
@@ -34,6 +34,7 @@ enum MutexFlags {
   MutexFlagTryLockFailed       = 1 << 5, // __tsan_mutex_try_lock_failed
   MutexFlagRecursiveLock       = 1 << 6, // __tsan_mutex_recursive_lock
   MutexFlagRecursiveUnlock     = 1 << 7, // __tsan_mutex_recursive_unlock
+  MutexFlagNotStatic           = 1 << 8, // __tsan_mutex_not_static
 
   // The following flags are runtime private.
   // Mutex API misuse was detected, so don't report any more.
@@ -43,7 +44,8 @@ enum MutexFlags {
   // Must list all mutex creation flags.
   MutexCreationFlagMask        = MutexFlagLinkerInit |
                                  MutexFlagWriteReentrant |
-                                 MutexFlagReadReentrant,
+                                 MutexFlagReadReentrant |
+                                 MutexFlagNotStatic,
 };
 
 struct SyncVar {
diff --git a/lib/tsan/rtl/tsan_trace.h b/lib/tsan/rtl/tsan_trace.h
index 96a18ac4101a..9aef375cbfe6 100644
--- a/lib/tsan/rtl/tsan_trace.h
+++ b/lib/tsan/rtl/tsan_trace.h
@@ -41,6 +41,8 @@ enum EventType {
 // u64 addr : 61;  // Associated pc.
 typedef u64 Event;
 
+const uptr kEventPCBits = 61;
+
 struct TraceHeader {
 #if !SANITIZER_GO
   BufferedStackTrace stack0;  // Start stack for the trace.
diff --git a/lib/tsan/rtl/tsan_vector.h b/lib/tsan/rtl/tsan_vector.h
deleted file mode 100644
index a7fb3fa58d54..000000000000
--- a/lib/tsan/rtl/tsan_vector.h
+++ /dev/null
@@ -1,127 +0,0 @@
-//===-- tsan_vector.h -------------------------------------------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file is a part of ThreadSanitizer (TSan), a race detector.
-//
-//===----------------------------------------------------------------------===//
-
-// Low-fat STL-like vector container.
-
-#ifndef TSAN_VECTOR_H
-#define TSAN_VECTOR_H
-
-#include "tsan_defs.h"
-#include "tsan_mman.h"
-
-namespace __tsan {
-
-template<typename T>
-class Vector {
- public:
-  explicit Vector(MBlockType typ)
-      : typ_(typ)
-      , begin_()
-      , end_()
-      , last_() {
-  }
-
-  ~Vector() {
-    if (begin_)
-      internal_free(begin_);
-  }
-
-  void Reset() {
-    if (begin_)
-      internal_free(begin_);
-    begin_ = 0;
-    end_ = 0;
-    last_ = 0;
-  }
-
-  uptr Size() const {
-    return end_ - begin_;
-  }
-
-  T &operator[](uptr i) {
-    DCHECK_LT(i, end_ - begin_);
-    return begin_[i];
-  }
-
-  const T &operator[](uptr i) const {
-    DCHECK_LT(i, end_ - begin_);
-    return begin_[i];
-  }
-
-  T *PushBack() {
-    EnsureSize(Size() + 1);
-    T *p = &end_[-1];
-    internal_memset(p, 0, sizeof(*p));
-    return p;
-  }
-
-  T *PushBack(const T& v) {
-    EnsureSize(Size() + 1);
-    T *p = &end_[-1];
-    internal_memcpy(p, &v, sizeof(*p));
-    return p;
-  }
-
-  void PopBack() {
-    DCHECK_GT(end_, begin_);
-    end_--;
-  }
-
-  void Resize(uptr size) {
-    if (size == 0) {
-      end_ = begin_;
-      return;
-    }
-    uptr old_size = Size();
-    EnsureSize(size);
-    if (old_size < size) {
-      for (uptr i = old_size; i < size; i++)
-        internal_memset(&begin_[i], 0, sizeof(begin_[i]));
-    }
-  }
-
- private:
-  const MBlockType typ_;
-  T *begin_;
-  T *end_;
-  T *last_;
-
-  void EnsureSize(uptr size) {
-    if (size <= Size())
-      return;
-    if (size <= (uptr)(last_ - begin_)) {
-      end_ = begin_ + size;
-      return;
-    }
-    uptr cap0 = last_ - begin_;
-    uptr cap = cap0 * 5 / 4;  // 25% growth
-    if (cap == 0)
-      cap = 16;
-    if (cap < size)
-      cap = size;
-    T *p = (T*)internal_alloc(typ_, cap * sizeof(T));
-    if (cap0) {
-      internal_memcpy(p, begin_, cap0 * sizeof(T));
-      internal_free(begin_);
-    }
-    begin_ = p;
-    end_ = begin_ + size;
-    last_ = begin_ + cap;
-  }
-
-  Vector(const Vector&);
-  void operator=(const Vector&);
-};
-}  // namespace __tsan
-
-#endif  // #ifndef TSAN_VECTOR_H
diff --git a/lib/tsan/tests/CMakeLists.txt b/lib/tsan/tests/CMakeLists.txt
index f8aec6854bb3..ad8d02ed331f 100644
--- a/lib/tsan/tests/CMakeLists.txt
+++ b/lib/tsan/tests/CMakeLists.txt
@@ -2,7 +2,7 @@ include_directories(../rtl)
 
 add_custom_target(TsanUnitTests)
 set_target_properties(TsanUnitTests PROPERTIES
-  FOLDER "TSan unittests")
+  FOLDER "Compiler-RT Tests")
 
 set(TSAN_UNITTEST_CFLAGS
   ${TSAN_CFLAGS}
@@ -13,9 +13,29 @@ set(TSAN_UNITTEST_CFLAGS
   -I${COMPILER_RT_SOURCE_DIR}/lib/tsan/rtl
   -DGTEST_HAS_RTTI=0)
 
+set(TSAN_TEST_ARCH ${TSAN_SUPPORTED_ARCH})
 if(APPLE)
+
+  # Create a static library for test dependencies.
+  set(TSAN_TEST_RUNTIME_OBJECTS
+    $<TARGET_OBJECTS:RTTsan_dynamic.osx>
+    $<TARGET_OBJECTS:RTInterception.osx>
+    $<TARGET_OBJECTS:RTSanitizerCommon.osx>
+    $<TARGET_OBJECTS:RTSanitizerCommonLibc.osx>
+    $<TARGET_OBJECTS:RTUbsan.osx>)
+  set(TSAN_TEST_RUNTIME RTTsanTest)
+  add_library(${TSAN_TEST_RUNTIME} STATIC ${TSAN_TEST_RUNTIME_OBJECTS})
+  set_target_properties(${TSAN_TEST_RUNTIME} PROPERTIES
+    ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
+
+  darwin_filter_host_archs(TSAN_SUPPORTED_ARCH TSAN_TEST_ARCH)
   list(APPEND TSAN_UNITTEST_CFLAGS ${DARWIN_osx_CFLAGS})
-  list(APPEND TSAN_UNITTEST_LINKFLAGS ${DARWIN_osx_LINKFLAGS})
+
+  set(LINK_FLAGS "-lc++")
+  add_weak_symbols("ubsan" LINK_FLAGS)
+  add_weak_symbols("sanitizer_common" LINK_FLAGS)
+else()
+  set(LINK_FLAGS "-fsanitize=thread;-lstdc++;-lm")
 endif()
 
 set(TSAN_RTL_HEADERS)
@@ -23,79 +43,27 @@ foreach (header ${TSAN_HEADERS})
   list(APPEND TSAN_RTL_HEADERS ${CMAKE_CURRENT_SOURCE_DIR}/../${header})
 endforeach()
 
-# tsan_compile(obj_list, source, arch, {headers})
-macro(tsan_compile obj_list source arch)
-  get_filename_component(basename ${source} NAME)
-  set(output_obj "${basename}.${arch}.o")
-  get_target_flags_for_arch(${arch} TARGET_CFLAGS)
-  set(COMPILE_DEPS ${TSAN_RTL_HEADERS} ${ARGN})
-  if(NOT COMPILER_RT_STANDALONE_BUILD)
-    list(APPEND COMPILE_DEPS gtest tsan)
-  endif()
-  clang_compile(${output_obj} ${source}
-          CFLAGS ${TSAN_UNITTEST_CFLAGS} ${TARGET_CFLAGS}
-          DEPS ${COMPILE_DEPS})
-  list(APPEND ${obj_list} ${output_obj})
-endmacro()
-
+# add_tsan_unittest(<name>
+#                   SOURCES <sources list>
+#                   HEADERS <extra headers list>)
 macro(add_tsan_unittest testname)
-  set(TSAN_TEST_ARCH ${TSAN_SUPPORTED_ARCH})
-  if(APPLE)
-    darwin_filter_host_archs(TSAN_SUPPORTED_ARCH TSAN_TEST_ARCH)
-  endif()
+  cmake_parse_arguments(TEST "" "" "SOURCES;HEADERS" ${ARGN})
   if(UNIX)
     foreach(arch ${TSAN_TEST_ARCH})
-      cmake_parse_arguments(TEST "" "" "SOURCES;HEADERS" ${ARGN})
-      set(TEST_OBJECTS)
-      foreach(SOURCE ${TEST_SOURCES} ${COMPILER_RT_GTEST_SOURCE})
-        tsan_compile(TEST_OBJECTS ${SOURCE} ${arch} ${TEST_HEADERS})
-      endforeach()
-      get_target_flags_for_arch(${arch} TARGET_LINK_FLAGS)
-      set(TEST_DEPS ${TEST_OBJECTS})
-      if(NOT COMPILER_RT_STANDALONE_BUILD)
-        list(APPEND TEST_DEPS tsan)
-      endif()
-      if(NOT APPLE)
-        # FIXME: Looks like we should link TSan with just-built runtime,
-        # and not rely on -fsanitize=thread, as these tests are essentially
-        # unit tests.
-        add_compiler_rt_test(TsanUnitTests ${testname}
-                OBJECTS ${TEST_OBJECTS}
-                DEPS ${TEST_DEPS}
-                LINK_FLAGS ${TARGET_LINK_FLAGS}
-                           -fsanitize=thread
-                           -lstdc++ -lm)
-      else()
-        set(TSAN_TEST_RUNTIME_OBJECTS
-          $<TARGET_OBJECTS:RTTsan_dynamic.osx>
-          $<TARGET_OBJECTS:RTInterception.osx>
-          $<TARGET_OBJECTS:RTSanitizerCommon.osx>
-          $<TARGET_OBJECTS:RTSanitizerCommonLibc.osx>
-          $<TARGET_OBJECTS:RTUbsan.osx>)
-        set(TSAN_TEST_RUNTIME RTTsanTest.${testname}.${arch})
-        add_library(${TSAN_TEST_RUNTIME} STATIC ${TSAN_TEST_RUNTIME_OBJECTS})
-        set_target_properties(${TSAN_TEST_RUNTIME} PROPERTIES
-          ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
-        list(APPEND TEST_OBJECTS lib${TSAN_TEST_RUNTIME}.a)
-        list(APPEND TEST_DEPS ${TSAN_TEST_RUNTIME})
-
-        add_weak_symbols("ubsan" WEAK_SYMBOL_LINK_FLAGS)
-        add_weak_symbols("sanitizer_common" WEAK_SYMBOL_LINK_FLAGS)
-
-        # Intentionally do *not* link with `-fsanitize=thread`. We already link
-        # against a static version of the runtime, and we don't want the dynamic
-        # one.
-        add_compiler_rt_test(TsanUnitTests "${testname}-${arch}-Test"
-                OBJECTS ${TEST_OBJECTS}
-                DEPS ${TEST_DEPS}
-                LINK_FLAGS ${TARGET_LINK_FLAGS} ${DARWIN_osx_LINK_FLAGS}
-                           ${WEAK_SYMBOL_LINK_FLAGS} -lc++)
-      endif()
+      set(TsanUnitTestsObjects)
+      generate_compiler_rt_tests(TsanUnitTestsObjects TsanUnitTests
+        "${testname}-${arch}-Test" ${arch}
+        SOURCES ${TEST_SOURCES} ${COMPILER_RT_GTEST_SOURCE}
+        RUNTIME ${TSAN_TEST_RUNTIME}
+        COMPILE_DEPS ${TEST_HEADERS} ${TSAN_RTL_HEADERS}
+        DEPS gtest tsan
+        CFLAGS ${TSAN_UNITTEST_CFLAGS}
+        LINK_FLAGS ${LINK_FLAGS})
     endforeach()
   endif()
 endmacro()
 
-if(COMPILER_RT_CAN_EXECUTE_TESTS)
+if(COMPILER_RT_CAN_EXECUTE_TESTS AND NOT ANDROID)
   add_subdirectory(rtl)
   add_subdirectory(unit)
 endif()
diff --git a/lib/tsan/tests/rtl/tsan_test_util_posix.cc b/lib/tsan/tests/rtl/tsan_test_util_posix.cc
index 834a271aa8c0..d00e26dd5ca6 100644
--- a/lib/tsan/tests/rtl/tsan_test_util_posix.cc
+++ b/lib/tsan/tests/rtl/tsan_test_util_posix.cc
@@ -9,7 +9,7 @@
 //
 // This file is a part of ThreadSanitizer (TSan), a race detector.
 //
-// Test utils, Linux, FreeBSD and Darwin implementation.
+// Test utils, Linux, FreeBSD, NetBSD and Darwin implementation.
 //===----------------------------------------------------------------------===//
 
 #include "sanitizer_common/sanitizer_atomic.h"
@@ -270,7 +270,7 @@ void ScopedThread::Impl::HandleEvent(Event *ev) {
       }
     }
     CHECK_NE(tsan_mop, 0);
-#if defined(__FreeBSD__) || defined(__APPLE__)
+#if defined(__FreeBSD__) || defined(__APPLE__) || defined(__NetBSD__)
     const int ErrCode = ESOCKTNOSUPPORT;
 #else
     const int ErrCode = ECHRNG;
diff --git a/lib/tsan/tests/unit/CMakeLists.txt b/lib/tsan/tests/unit/CMakeLists.txt
index 6898f641d6a0..c08508d50791 100644
--- a/lib/tsan/tests/unit/CMakeLists.txt
+++ b/lib/tsan/tests/unit/CMakeLists.txt
@@ -6,8 +6,7 @@ set(TSAN_UNIT_TEST_SOURCES
   tsan_shadow_test.cc
   tsan_stack_test.cc
   tsan_sync_test.cc
-  tsan_unit_test_main.cc
-  tsan_vector_test.cc)
+  tsan_unit_test_main.cc)
 
 add_tsan_unittest(TsanUnitTest
   SOURCES ${TSAN_UNIT_TEST_SOURCES})
diff --git a/lib/tsan/tests/unit/tsan_mman_test.cc b/lib/tsan/tests/unit/tsan_mman_test.cc
index 60dea3d43240..05ae4286704c 100644
--- a/lib/tsan/tests/unit/tsan_mman_test.cc
+++ b/lib/tsan/tests/unit/tsan_mman_test.cc
@@ -56,6 +56,7 @@ TEST(Mman, UserRealloc) {
     // Realloc(NULL, N) is equivalent to malloc(N), thus must return
     // non-NULL pointer.
     EXPECT_NE(p, (void*)0);
+    user_free(thr, pc, p);
   }
   {
     void *p = user_realloc(thr, pc, 0, 100);
@@ -67,8 +68,9 @@ TEST(Mman, UserRealloc) {
     void *p = user_alloc(thr, pc, 100);
     EXPECT_NE(p, (void*)0);
     memset(p, 0xde, 100);
+    // Realloc(P, 0) is equivalent to free(P) and returns NULL.
     void *p2 = user_realloc(thr, pc, p, 0);
-    EXPECT_NE(p2, (void*)0);
+    EXPECT_EQ(p2, (void*)0);
   }
   {
     void *p = user_realloc(thr, pc, 0, 100);
@@ -135,12 +137,34 @@ TEST(Mman, Stats) {
   EXPECT_EQ(unmapped0, __sanitizer_get_unmapped_bytes());
 }
 
+TEST(Mman, Valloc) {
+  ThreadState *thr = cur_thread();
+  uptr page_size = GetPageSizeCached();
+
+  void *p = user_valloc(thr, 0, 100);
+  EXPECT_NE(p, (void*)0);
+  user_free(thr, 0, p);
+
+  p = user_pvalloc(thr, 0, 100);
+  EXPECT_NE(p, (void*)0);
+  user_free(thr, 0, p);
+
+  p = user_pvalloc(thr, 0, 0);
+  EXPECT_NE(p, (void*)0);
+  EXPECT_EQ(page_size, __sanitizer_get_allocated_size(p));
+  user_free(thr, 0, p);
+
+  EXPECT_DEATH(p = user_pvalloc(thr, 0, (uptr)-(page_size - 1)),
+               "allocator is terminating the process instead of returning 0");
+  EXPECT_DEATH(p = user_pvalloc(thr, 0, (uptr)-1),
+               "allocator is terminating the process instead of returning 0");
+}
+
+#if !SANITIZER_DEBUG
+// EXPECT_DEATH clones a thread with 4K stack,
+// which is overflown by tsan memory accesses functions in debug mode.
+
 TEST(Mman, CallocOverflow) {
-#if SANITIZER_DEBUG
-  // EXPECT_DEATH clones a thread with 4K stack,
-  // which is overflown by tsan memory accesses functions in debug mode.
-  return;
-#endif
   ThreadState *thr = cur_thread();
   uptr pc = 0;
   size_t kArraySize = 4096;
@@ -152,4 +176,57 @@ TEST(Mman, CallocOverflow) {
   EXPECT_EQ(0L, p);
 }
 
+TEST(Mman, Memalign) {
+  ThreadState *thr = cur_thread();
+
+  void *p = user_memalign(thr, 0, 8, 100);
+  EXPECT_NE(p, (void*)0);
+  user_free(thr, 0, p);
+
+  p = NULL;
+  EXPECT_DEATH(p = user_memalign(thr, 0, 7, 100),
+               "allocator is terminating the process instead of returning 0");
+  EXPECT_EQ(0L, p);
+}
+
+TEST(Mman, PosixMemalign) {
+  ThreadState *thr = cur_thread();
+
+  void *p = NULL;
+  int res = user_posix_memalign(thr, 0, &p, 8, 100);
+  EXPECT_NE(p, (void*)0);
+  EXPECT_EQ(res, 0);
+  user_free(thr, 0, p);
+
+  p = NULL;
+  // Alignment is not a power of two, although is a multiple of sizeof(void*).
+  EXPECT_DEATH(res = user_posix_memalign(thr, 0, &p, 3 * sizeof(p), 100),
+               "allocator is terminating the process instead of returning 0");
+  EXPECT_EQ(0L, p);
+  // Alignment is not a multiple of sizeof(void*), although is a power of 2.
+  EXPECT_DEATH(res = user_posix_memalign(thr, 0, &p, 2, 100),
+               "allocator is terminating the process instead of returning 0");
+  EXPECT_EQ(0L, p);
+}
+
+TEST(Mman, AlignedAlloc) {
+  ThreadState *thr = cur_thread();
+
+  void *p = user_aligned_alloc(thr, 0, 8, 64);
+  EXPECT_NE(p, (void*)0);
+  user_free(thr, 0, p);
+
+  p = NULL;
+  // Alignement is not a power of 2.
+  EXPECT_DEATH(p = user_aligned_alloc(thr, 0, 7, 100),
+               "allocator is terminating the process instead of returning 0");
+  EXPECT_EQ(0L, p);
+  // Size is not a multiple of alignment.
+  EXPECT_DEATH(p = user_aligned_alloc(thr, 0, 8, 100),
+               "allocator is terminating the process instead of returning 0");
+  EXPECT_EQ(0L, p);
+}
+
+#endif
+
 }  // namespace __tsan
diff --git a/lib/tsan/tests/unit/tsan_vector_test.cc b/lib/tsan/tests/unit/tsan_vector_test.cc
deleted file mode 100644
index c54ac1ee6de9..000000000000
--- a/lib/tsan/tests/unit/tsan_vector_test.cc
+++ /dev/null
@@ -1,43 +0,0 @@
-//===-- tsan_vector_test.cc -----------------------------------------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file is a part of ThreadSanitizer (TSan), a race detector.
-//
-//===----------------------------------------------------------------------===//
-#include "tsan_vector.h"
-#include "tsan_rtl.h"
-#include "gtest/gtest.h"
-
-namespace __tsan {
-
-TEST(Vector, Basic) {
-  Vector<int> v(MBlockScopedBuf);
-  EXPECT_EQ(v.Size(), (uptr)0);
-  v.PushBack(42);
-  EXPECT_EQ(v.Size(), (uptr)1);
-  EXPECT_EQ(v[0], 42);
-  v.PushBack(43);
-  EXPECT_EQ(v.Size(), (uptr)2);
-  EXPECT_EQ(v[0], 42);
-  EXPECT_EQ(v[1], 43);
-}
-
-TEST(Vector, Stride) {
-  Vector<int> v(MBlockScopedBuf);
-  for (int i = 0; i < 1000; i++) {
-    v.PushBack(i);
-    EXPECT_EQ(v.Size(), (uptr)(i + 1));
-    EXPECT_EQ(v[i], i);
-  }
-  for (int i = 0; i < 1000; i++) {
-    EXPECT_EQ(v[i], i);
-  }
-}
-
-}  // namespace __tsan