libsanitizer merge from upstream r250806.

libsanitizer/ 2015-10-20 Maxim Ostapenko <m.ostapenko@partner.samsung.com> * All source files: Merge from upstream r250806. * configure.ac (link_sanitizer_common): Add -lrt flag. * configure.tgt: Enable TSAN and LSAN for aarch64-linux targets. Set CXX_ABI_NEEDED=true for darwin. * asan/Makefile.am (asan_files): Add new files. (DEFS): Add DCAN_SANITIZE_UB=0 and remove unused and legacy DASAN_FLEXIBLE_MAPPING_AND_OFFSET=0. * asan/Makefile.in: Regenerate. * ubsan/Makefile.am (ubsan_files): Add new files. (DEFS): Add DCAN_SANITIZE_UB=1. (libubsan_la_LIBADD): Add -lc++abi if CXX_ABI_NEEDED is true. * ubsan/Makefile.in: Regenerate. * tsan/Makefile.am (tsan_files): Add new files. (DEFS): Add DCAN_SANITIZE_UB=0. * tsan/Makefile.in: Regenerate. * sanitizer_common/Makefile.am (sanitizer_common_files): Add new files. * sanitizer_common/Makefile.in: Regenerate. * asan/libtool-version: Bump the libasan SONAME. git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@229111 138bc75d-0d04-0410-961f-82ee72b054a4
author: chefmax <chefmax@138bc75d-0d04-0410-961f-82ee72b054a4> 2015-10-21 07:32:45 +0000
committer: chefmax <chefmax@138bc75d-0d04-0410-961f-82ee72b054a4> 2015-10-21 07:32:45 +0000
commit: 5645a48f7ebd0f97a072b7a2eb40b27cea9d4318 (patch)
tree: 2bdaf703dd35e1806b59bd7d74c7eee290a1054f /libsanitizer/tsan
parent: 397881d34f32eddf4a6665789f1a7cdd5ff3695e (diff)
download: gcc-5645a48f7ebd0f97a072b7a2eb40b27cea9d4318.tar.gz
46 files changed, 1519 insertions, 1038 deletions
diff --git a/libsanitizer/tsan/Makefile.am b/libsanitizer/tsan/Makefile.am
index abfafb783fd..5c732cb50c3 100644
--- a/libsanitizer/tsan/Makefile.am
+++ b/libsanitizer/tsan/Makefile.am
@@ -3,7 +3,7 @@ AM_CPPFLAGS = -I $(top_srcdir) -I $(top_srcdir)/include
 # May be used by toolexeclibdir.
 gcc_version := $(shell cat $(top_srcdir)/../gcc/BASE-VER)
 
-DEFS = -D_GNU_SOURCE -D_DEBUG -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS 
+DEFS = -D_GNU_SOURCE -D_DEBUG -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -DCAN_SANITIZE_UB=0
 AM_CXXFLAGS = -Wall -W -Wno-unused-parameter -Wwrite-strings -pedantic -Wno-long-long  -fPIC -fno-builtin -fno-exceptions -fno-rtti -fomit-frame-pointer -funwind-tables -fvisibility=hidden -Wno-variadic-macros
 AM_CXXFLAGS += $(LIBSTDCXX_RAW_CXX_CXXFLAGS)
 AM_CXXFLAGS += -std=gnu++11
@@ -25,6 +25,7 @@ tsan_files = \
 	tsan_mman.cc \
 	tsan_mutex.cc \
 	tsan_mutexset.cc \
+	tsan_new_delete.cc \
 	tsan_platform_linux.cc \
 	tsan_platform_mac.cc \
 	tsan_platform_windows.cc \
diff --git a/libsanitizer/tsan/Makefile.in b/libsanitizer/tsan/Makefile.in
index 8b20d285901..bd196073e03 100644
--- a/libsanitizer/tsan/Makefile.in
+++ b/libsanitizer/tsan/Makefile.in
@@ -108,11 +108,11 @@ am__objects_1 = tsan_clock.lo tsan_fd.lo tsan_flags.lo \
 	tsan_ignoreset.lo tsan_interceptors.lo tsan_interface_ann.lo \
 	tsan_interface_atomic.lo tsan_interface.lo \
 	tsan_interface_java.lo tsan_md5.lo tsan_mman.lo tsan_mutex.lo \
-	tsan_mutexset.lo tsan_platform_linux.lo tsan_platform_mac.lo \
-	tsan_platform_windows.lo tsan_report.lo tsan_rtl.lo \
-	tsan_rtl_mutex.lo tsan_rtl_report.lo tsan_rtl_thread.lo \
-	tsan_stack_trace.lo tsan_stat.lo tsan_suppressions.lo \
-	tsan_symbolize.lo tsan_sync.lo
+	tsan_mutexset.lo tsan_new_delete.lo tsan_platform_linux.lo \
+	tsan_platform_mac.lo tsan_platform_windows.lo tsan_report.lo \
+	tsan_rtl.lo tsan_rtl_mutex.lo tsan_rtl_report.lo \
+	tsan_rtl_thread.lo tsan_stack_trace.lo tsan_stat.lo \
+	tsan_suppressions.lo tsan_symbolize.lo tsan_sync.lo
 am_libtsan_la_OBJECTS = $(am__objects_1)
 libtsan_la_OBJECTS = $(am_libtsan_la_OBJECTS)
 libtsan_la_LINK = $(LIBTOOL) --tag=CXX $(AM_LIBTOOLFLAGS) \
@@ -177,7 +177,7 @@ CXXCPP = @CXXCPP@
 CXXDEPMODE = @CXXDEPMODE@
 CXXFLAGS = @CXXFLAGS@
 CYGPATH_W = @CYGPATH_W@
-DEFS = -D_GNU_SOURCE -D_DEBUG -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS 
+DEFS = -D_GNU_SOURCE -D_DEBUG -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -DCAN_SANITIZE_UB=0
 DEPDIR = @DEPDIR@
 DSYMUTIL = @DSYMUTIL@
 DUMPBIN = @DUMPBIN@
@@ -321,6 +321,7 @@ tsan_files = \
 	tsan_mman.cc \
 	tsan_mutex.cc \
 	tsan_mutexset.cc \
+	tsan_new_delete.cc \
 	tsan_platform_linux.cc \
 	tsan_platform_mac.cc \
 	tsan_platform_windows.cc \
@@ -475,6 +476,7 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_mman.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_mutex.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_mutexset.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_new_delete.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_platform_linux.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_platform_mac.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_platform_windows.Plo@am__quote@
diff --git a/libsanitizer/tsan/tsan_clock.cc b/libsanitizer/tsan/tsan_clock.cc
index a84caa952a6..d925c9cfd83 100644
--- a/libsanitizer/tsan/tsan_clock.cc
+++ b/libsanitizer/tsan/tsan_clock.cc
@@ -80,7 +80,7 @@
 
 // We don't have ThreadState in these methods, so this is an ugly hack that
 // works only in C++.
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 # define CPP_STAT_INC(typ) StatInc(cur_thread(), typ)
 #else
 # define CPP_STAT_INC(typ) (void)0
@@ -102,8 +102,8 @@ ThreadClock::ThreadClock(unsigned tid, unsigned reused)
 }
 
 void ThreadClock::acquire(ClockCache *c, const SyncClock *src) {
-  DCHECK(nclk_ <= kMaxTid);
-  DCHECK(src->size_ <= kMaxTid);
+  DCHECK_LE(nclk_, kMaxTid);
+  DCHECK_LE(src->size_, kMaxTid);
   CPP_STAT_INC(StatClockAcquire);
 
   // Check if it's empty -> no need to do anything.
@@ -213,8 +213,8 @@ void ThreadClock::release(ClockCache *c, SyncClock *dst) const {
 }
 
 void ThreadClock::ReleaseStore(ClockCache *c, SyncClock *dst) const {
-  DCHECK(nclk_ <= kMaxTid);
-  DCHECK(dst->size_ <= kMaxTid);
+  DCHECK_LE(nclk_, kMaxTid);
+  DCHECK_LE(dst->size_, kMaxTid);
   CPP_STAT_INC(StatClockStore);
 
   // Check if we need to resize dst.
diff --git a/libsanitizer/tsan/tsan_defs.h b/libsanitizer/tsan/tsan_defs.h
index 076de73c9be..21b68356e7b 100644
--- a/libsanitizer/tsan/tsan_defs.h
+++ b/libsanitizer/tsan/tsan_defs.h
@@ -15,14 +15,24 @@
 #include "sanitizer_common/sanitizer_internal_defs.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "tsan_stat.h"
+#include "ubsan/ubsan_platform.h"
 
-#ifndef TSAN_DEBUG
-#define TSAN_DEBUG 0
-#endif  // TSAN_DEBUG
+// Setup defaults for compile definitions.
+#ifndef TSAN_NO_HISTORY
+# define TSAN_NO_HISTORY 0
+#endif
+
+#ifndef TSAN_COLLECT_STATS
+# define TSAN_COLLECT_STATS 0
+#endif
+
+#ifndef TSAN_CONTAINS_UBSAN
+# define TSAN_CONTAINS_UBSAN (CAN_SANITIZE_UB && !defined(SANITIZER_GO))
+#endif
 
 namespace __tsan {
 
-#ifdef TSAN_GO
+#ifdef SANITIZER_GO
 const bool kGoMode = true;
 const bool kCppMode = false;
 const char *const kTsanOptionsEnv = "GORACE";
@@ -37,23 +47,17 @@ const char *const kTsanOptionsEnv = "TSAN_OPTIONS";
 
 const int kTidBits = 13;
 const unsigned kMaxTid = 1 << kTidBits;
+#ifndef SANITIZER_GO
 const unsigned kMaxTidInClock = kMaxTid * 2;  // This includes msb 'freed' bit.
+#else
+const unsigned kMaxTidInClock = kMaxTid;  // Go does not track freed memory.
+#endif
 const int kClkBits = 42;
 const unsigned kMaxTidReuse = (1 << (64 - kClkBits)) - 1;
 const uptr kShadowStackSize = 64 * 1024;
 
-#ifdef TSAN_SHADOW_COUNT
-# if TSAN_SHADOW_COUNT == 2 \
-  || TSAN_SHADOW_COUNT == 4 || TSAN_SHADOW_COUNT == 8
-const uptr kShadowCnt = TSAN_SHADOW_COUNT;
-# else
-#   error "TSAN_SHADOW_COUNT must be one of 2,4,8"
-# endif
-#else
 // Count of shadow values in a shadow cell.
-#define TSAN_SHADOW_COUNT 4
 const uptr kShadowCnt = 4;
-#endif
 
 // That many user bytes are mapped onto a single shadow cell.
 const uptr kShadowCell = 8;
@@ -71,22 +75,16 @@ const uptr kMetaShadowCell = 8;
 // Size of a single meta shadow value (u32).
 const uptr kMetaShadowSize = 4;
 
-#if defined(TSAN_NO_HISTORY) && TSAN_NO_HISTORY
+#if TSAN_NO_HISTORY
 const bool kCollectHistory = false;
 #else
 const bool kCollectHistory = true;
 #endif
 
-#if defined(TSAN_COLLECT_STATS) && TSAN_COLLECT_STATS
-const bool kCollectStats = true;
-#else
-const bool kCollectStats = false;
-#endif
-
 // The following "build consistency" machinery ensures that all source files
 // are built in the same configuration. Inconsistent builds lead to
 // hard to debug crashes.
-#if TSAN_DEBUG
+#if SANITIZER_DEBUG
 void build_consistency_debug();
 #else
 void build_consistency_release();
@@ -98,18 +96,8 @@ void build_consistency_stats();
 void build_consistency_nostats();
 #endif
 
-#if TSAN_SHADOW_COUNT == 1
-void build_consistency_shadow1();
-#elif TSAN_SHADOW_COUNT == 2
-void build_consistency_shadow2();
-#elif TSAN_SHADOW_COUNT == 4
-void build_consistency_shadow4();
-#else
-void build_consistency_shadow8();
-#endif
-
 static inline void USED build_consistency() {
-#if TSAN_DEBUG
+#if SANITIZER_DEBUG
   build_consistency_debug();
 #else
   build_consistency_release();
@@ -119,15 +107,6 @@ static inline void USED build_consistency() {
 #else
   build_consistency_nostats();
 #endif
-#if TSAN_SHADOW_COUNT == 1
-  build_consistency_shadow1();
-#elif TSAN_SHADOW_COUNT == 2
-  build_consistency_shadow2();
-#elif TSAN_SHADOW_COUNT == 4
-  build_consistency_shadow4();
-#else
-  build_consistency_shadow8();
-#endif
 }
 
 template<typename T>
diff --git a/libsanitizer/tsan/tsan_fd.cc b/libsanitizer/tsan/tsan_fd.cc
index 10582035f24..8f75a28200d 100644
--- a/libsanitizer/tsan/tsan_fd.cc
+++ b/libsanitizer/tsan/tsan_fd.cc
@@ -89,7 +89,8 @@ static FdDesc *fddesc(ThreadState *thr, uptr pc, int fd) {
 }
 
 // pd must be already ref'ed.
-static void init(ThreadState *thr, uptr pc, int fd, FdSync *s) {
+static void init(ThreadState *thr, uptr pc, int fd, FdSync *s,
+    bool write = true) {
   FdDesc *d = fddesc(thr, pc, fd);
   // As a matter of fact, we don't intercept all close calls.
   // See e.g. libc __res_iclose().
@@ -107,8 +108,13 @@ static void init(ThreadState *thr, uptr pc, int fd, FdSync *s) {
   }
   d->creation_tid = thr->tid;
   d->creation_stack = CurrentStackId(thr, pc);
-  // To catch races between fd usage and open.
-  MemoryRangeImitateWrite(thr, pc, (uptr)d, 8);
+  if (write) {
+    // To catch races between fd usage and open.
+    MemoryRangeImitateWrite(thr, pc, (uptr)d, 8);
+  } else {
+    // See the dup-related comment in FdClose.
+    MemoryRead(thr, pc, (uptr)d, kSizeLog8);
+  }
 }
 
 void FdInit() {
@@ -179,13 +185,25 @@ void FdAccess(ThreadState *thr, uptr pc, int fd) {
   MemoryRead(thr, pc, (uptr)d, kSizeLog8);
 }
 
-void FdClose(ThreadState *thr, uptr pc, int fd) {
+void FdClose(ThreadState *thr, uptr pc, int fd, bool write) {
   DPrintf("#%d: FdClose(%d)\n", thr->tid, fd);
   if (bogusfd(fd))
     return;
   FdDesc *d = fddesc(thr, pc, fd);
-  // To catch races between fd usage and close.
-  MemoryWrite(thr, pc, (uptr)d, kSizeLog8);
+  if (write) {
+    // To catch races between fd usage and close.
+    MemoryWrite(thr, pc, (uptr)d, kSizeLog8);
+  } else {
+    // This path is used only by dup2/dup3 calls.
+    // We do read instead of write because there is a number of legitimate
+    // cases where write would lead to false positives:
+    // 1. Some software dups a closed pipe in place of a socket before closing
+    //    the socket (to prevent races actually).
+    // 2. Some daemons dup /dev/null in place of stdin/stdout.
+    // On the other hand we have not seen cases when write here catches real
+    // bugs.
+    MemoryRead(thr, pc, (uptr)d, kSizeLog8);
+  }
   // We need to clear it, because if we do not intercept any call out there
   // that creates fd, we will hit false postives.
   MemoryResetRange(thr, pc, (uptr)d, 8);
@@ -202,15 +220,15 @@ void FdFileCreate(ThreadState *thr, uptr pc, int fd) {
   init(thr, pc, fd, &fdctx.filesync);
 }
 
-void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd) {
+void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd, bool write) {
   DPrintf("#%d: FdDup(%d, %d)\n", thr->tid, oldfd, newfd);
   if (bogusfd(oldfd) || bogusfd(newfd))
     return;
   // Ignore the case when user dups not yet connected socket.
   FdDesc *od = fddesc(thr, pc, oldfd);
   MemoryRead(thr, pc, (uptr)od, kSizeLog8);
-  FdClose(thr, pc, newfd);
-  init(thr, pc, newfd, ref(od->sync));
+  FdClose(thr, pc, newfd, write);
+  init(thr, pc, newfd, ref(od->sync), write);
 }
 
 void FdPipeCreate(ThreadState *thr, uptr pc, int rfd, int wfd) {
diff --git a/libsanitizer/tsan/tsan_fd.h b/libsanitizer/tsan/tsan_fd.h
index 57ae62cfe8b..4d9236c9903 100644
--- a/libsanitizer/tsan/tsan_fd.h
+++ b/libsanitizer/tsan/tsan_fd.h
@@ -40,9 +40,9 @@ void FdInit();
 void FdAcquire(ThreadState *thr, uptr pc, int fd);
 void FdRelease(ThreadState *thr, uptr pc, int fd);
 void FdAccess(ThreadState *thr, uptr pc, int fd);
-void FdClose(ThreadState *thr, uptr pc, int fd);
+void FdClose(ThreadState *thr, uptr pc, int fd, bool write = true);
 void FdFileCreate(ThreadState *thr, uptr pc, int fd);
-void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd);
+void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd, bool write);
 void FdPipeCreate(ThreadState *thr, uptr pc, int rfd, int wfd);
 void FdEventCreate(ThreadState *thr, uptr pc, int fd);
 void FdSignalCreate(ThreadState *thr, uptr pc, int fd);
diff --git a/libsanitizer/tsan/tsan_flags.cc b/libsanitizer/tsan/tsan_flags.cc
index 6059f2771ba..3e89375187d 100644
--- a/libsanitizer/tsan/tsan_flags.cc
+++ b/libsanitizer/tsan/tsan_flags.cc
@@ -10,10 +10,12 @@
 //===----------------------------------------------------------------------===//
 
 #include "sanitizer_common/sanitizer_flags.h"
+#include "sanitizer_common/sanitizer_flag_parser.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "tsan_flags.h"
 #include "tsan_rtl.h"
 #include "tsan_mman.h"
+#include "ubsan/ubsan_flags.h"
 
 namespace __tsan {
 
@@ -31,80 +33,67 @@ const char *WEAK __tsan_default_options() {
 }
 #endif
 
-static void ParseFlags(Flags *f, const char *env) {
-  ParseFlag(env, &f->enable_annotations, "enable_annotations", "");
-  ParseFlag(env, &f->suppress_equal_stacks, "suppress_equal_stacks", "");
-  ParseFlag(env, &f->suppress_equal_addresses, "suppress_equal_addresses", "");
-  ParseFlag(env, &f->report_bugs, "report_bugs", "");
-  ParseFlag(env, &f->report_thread_leaks, "report_thread_leaks", "");
-  ParseFlag(env, &f->report_destroy_locked, "report_destroy_locked", "");
-  ParseFlag(env, &f->report_mutex_bugs, "report_mutex_bugs", "");
-  ParseFlag(env, &f->report_signal_unsafe, "report_signal_unsafe", "");
-  ParseFlag(env, &f->report_atomic_races, "report_atomic_races", "");
-  ParseFlag(env, &f->force_seq_cst_atomics, "force_seq_cst_atomics", "");
-  ParseFlag(env, &f->print_benign, "print_benign", "");
-  ParseFlag(env, &f->exitcode, "exitcode", "");
-  ParseFlag(env, &f->halt_on_error, "halt_on_error", "");
-  ParseFlag(env, &f->atexit_sleep_ms, "atexit_sleep_ms", "");
-  ParseFlag(env, &f->profile_memory, "profile_memory", "");
-  ParseFlag(env, &f->flush_memory_ms, "flush_memory_ms", "");
-  ParseFlag(env, &f->flush_symbolizer_ms, "flush_symbolizer_ms", "");
-  ParseFlag(env, &f->memory_limit_mb, "memory_limit_mb", "");
-  ParseFlag(env, &f->stop_on_start, "stop_on_start", "");
-  ParseFlag(env, &f->running_on_valgrind, "running_on_valgrind", "");
-  ParseFlag(env, &f->history_size, "history_size", "");
-  ParseFlag(env, &f->io_sync, "io_sync", "");
-  ParseFlag(env, &f->die_after_fork, "die_after_fork", "");
+void Flags::SetDefaults() {
+#define TSAN_FLAG(Type, Name, DefaultValue, Description) Name = DefaultValue;
+#include "tsan_flags.inc"
+#undef TSAN_FLAG
+  // DDFlags
+  second_deadlock_stack = false;
+}
 
+void RegisterTsanFlags(FlagParser *parser, Flags *f) {
+#define TSAN_FLAG(Type, Name, DefaultValue, Description) \
+  RegisterFlag(parser, #Name, Description, &f->Name);
+#include "tsan_flags.inc"
+#undef TSAN_FLAG
   // DDFlags
-  ParseFlag(env, &f->second_deadlock_stack, "second_deadlock_stack", "");
+  RegisterFlag(parser, "second_deadlock_stack",
+      "Report where each mutex is locked in deadlock reports",
+      &f->second_deadlock_stack);
 }
 
 void InitializeFlags(Flags *f, const char *env) {
-  internal_memset(f, 0, sizeof(*f));
-
-  // Default values.
-  f->enable_annotations = true;
-  f->suppress_equal_stacks = true;
-  f->suppress_equal_addresses = true;
-  f->report_bugs = true;
-  f->report_thread_leaks = true;
-  f->report_destroy_locked = true;
-  f->report_mutex_bugs = true;
-  f->report_signal_unsafe = true;
-  f->report_atomic_races = true;
-  f->force_seq_cst_atomics = false;
-  f->print_benign = false;
-  f->exitcode = 66;
-  f->halt_on_error = false;
-  f->atexit_sleep_ms = 1000;
-  f->profile_memory = "";
-  f->flush_memory_ms = 0;
-  f->flush_symbolizer_ms = 5000;
-  f->memory_limit_mb = 0;
-  f->stop_on_start = false;
-  f->running_on_valgrind = false;
-  f->history_size = kGoMode ? 1 : 2;  // There are a lot of goroutines in Go.
-  f->io_sync = 1;
-  f->die_after_fork = true;
+  SetCommonFlagsDefaults();
+  {
+    // Override some common flags defaults.
+    CommonFlags cf;
+    cf.CopyFrom(*common_flags());
+    cf.allow_addr2line = true;
+#ifndef SANITIZER_GO
+    cf.detect_deadlocks = true;
+#endif
+    cf.print_suppressions = false;
+    cf.stack_trace_format = "    #%n %f %S %M";
+    cf.exitcode = 66;
+    OverrideCommonFlags(cf);
+  }
 
-  // DDFlags
-  f->second_deadlock_stack = false;
+  f->SetDefaults();
+
+  FlagParser parser;
+  RegisterTsanFlags(&parser, f);
+  RegisterCommonFlags(&parser);
 
-  CommonFlags *cf = common_flags();
-  SetCommonFlagsDefaults(cf);
-  // Override some common flags defaults.
-  cf->allow_addr2line = true;
-  cf->detect_deadlocks = true;
-  cf->print_suppressions = false;
-  cf->stack_trace_format = "    #%n %f %S %M";
+#if TSAN_CONTAINS_UBSAN
+  __ubsan::Flags *uf = __ubsan::flags();
+  uf->SetDefaults();
+
+  FlagParser ubsan_parser;
+  __ubsan::RegisterUbsanFlags(&ubsan_parser, uf);
+  RegisterCommonFlags(&ubsan_parser);
+#endif
 
   // Let a frontend override.
-  ParseFlags(f, __tsan_default_options());
-  ParseCommonFlagsFromString(cf, __tsan_default_options());
+  parser.ParseString(__tsan_default_options());
+#if TSAN_CONTAINS_UBSAN
+  const char *ubsan_default_options = __ubsan::MaybeCallUbsanDefaultOptions();
+  ubsan_parser.ParseString(ubsan_default_options);
+#endif
   // Override from command line.
-  ParseFlags(f, env);
-  ParseCommonFlagsFromString(cf, env);
+  parser.ParseString(env);
+#if TSAN_CONTAINS_UBSAN
+  ubsan_parser.ParseString(GetEnv("UBSAN_OPTIONS"));
+#endif
 
   // Sanity check.
   if (!f->report_bugs) {
@@ -113,7 +102,11 @@ void InitializeFlags(Flags *f, const char *env) {
     f->report_signal_unsafe = false;
   }
 
-  if (cf->help) PrintFlagDescriptions();
+  SetVerbosity(common_flags()->verbosity);
+
+  if (Verbosity()) ReportUnrecognizedFlags();
+
+  if (common_flags()->help) parser.PrintFlagDescriptions();
 
   if (f->history_size < 0 || f->history_size > 7) {
     Printf("ThreadSanitizer: incorrect value for history_size"
diff --git a/libsanitizer/tsan/tsan_flags.h b/libsanitizer/tsan/tsan_flags.h
index 182d9b298af..3d58ff3793e 100644
--- a/libsanitizer/tsan/tsan_flags.h
+++ b/libsanitizer/tsan/tsan_flags.h
@@ -18,65 +18,12 @@
 namespace __tsan {
 
 struct Flags : DDFlags {
-  // Enable dynamic annotations, otherwise they are no-ops.
-  bool enable_annotations;
-  // Suppress a race report if we've already output another race report
-  // with the same stack.
-  bool suppress_equal_stacks;
-  // Suppress a race report if we've already output another race report
-  // on the same address.
-  bool suppress_equal_addresses;
-  // Turns off bug reporting entirely (useful for benchmarking).
-  bool report_bugs;
-  // Report thread leaks at exit?
-  bool report_thread_leaks;
-  // Report destruction of a locked mutex?
-  bool report_destroy_locked;
-  // Report incorrect usages of mutexes and mutex annotations?
-  bool report_mutex_bugs;
-  // Report violations of async signal-safety
-  // (e.g. malloc() call from a signal handler).
-  bool report_signal_unsafe;
-  // Report races between atomic and plain memory accesses.
-  bool report_atomic_races;
-  // If set, all atomics are effectively sequentially consistent (seq_cst),
-  // regardless of what user actually specified.
-  bool force_seq_cst_atomics;
-  // Print matched "benign" races at exit.
-  bool print_benign;
-  // Override exit status if something was reported.
-  int exitcode;
-  // Exit after first reported error.
-  bool halt_on_error;
-  // Sleep in main thread before exiting for that many ms
-  // (useful to catch "at exit" races).
-  int atexit_sleep_ms;
-  // If set, periodically write memory profile to that file.
-  const char *profile_memory;
-  // Flush shadow memory every X ms.
-  int flush_memory_ms;
-  // Flush symbolizer caches every X ms.
-  int flush_symbolizer_ms;
-  // Resident memory limit in MB to aim at.
-  // If the process consumes more memory, then TSan will flush shadow memory.
-  int memory_limit_mb;
-  // Stops on start until __tsan_resume() is called (for debugging).
-  bool stop_on_start;
-  // Controls whether RunningOnValgrind() returns true or false.
-  bool running_on_valgrind;
-  // Per-thread history size, controls how many previous memory accesses
-  // are remembered per thread.  Possible values are [0..7].
-  // history_size=0 amounts to 32K memory accesses.  Each next value doubles
-  // the amount of memory accesses, up to history_size=7 that amounts to
-  // 4M memory accesses.  The default value is 2 (128K memory accesses).
-  int history_size;
-  // Controls level of synchronization implied by IO operations.
-  // 0 - no synchronization
-  // 1 - reasonable level of synchronization (write->read)
-  // 2 - global synchronization of all IO operations
-  int io_sync;
-  // Die after multi-threaded fork if the child creates new threads.
-  bool die_after_fork;
+#define TSAN_FLAG(Type, Name, DefaultValue, Description) Type Name;
+#include "tsan_flags.inc"
+#undef TSAN_FLAG
+
+  void SetDefaults();
+  void ParseFromString(const char *str);
 };
 
 Flags *flags();
diff --git a/libsanitizer/tsan/tsan_flags.inc b/libsanitizer/tsan/tsan_flags.inc
new file mode 100644
index 00000000000..822e560b622
--- /dev/null
+++ b/libsanitizer/tsan/tsan_flags.inc
@@ -0,0 +1,76 @@
+//===-- tsan_flags.inc ------------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// TSan runtime flags.
+//
+//===----------------------------------------------------------------------===//
+#ifndef TSAN_FLAG
+# error "Define TSAN_FLAG prior to including this file!"
+#endif
+
+// TSAN_FLAG(Type, Name, DefaultValue, Description)
+// See COMMON_FLAG in sanitizer_flags.inc for more details.
+
+TSAN_FLAG(bool, enable_annotations, true,
+          "Enable dynamic annotations, otherwise they are no-ops.")
+// Suppress a race report if we've already output another race report
+// with the same stack.
+TSAN_FLAG(bool, suppress_equal_stacks, true,
+          "Suppress a race report if we've already output another race report "
+          "with the same stack.")
+TSAN_FLAG(bool, suppress_equal_addresses, true,
+          "Suppress a race report if we've already output another race report "
+          "on the same address.")
+
+TSAN_FLAG(bool, report_bugs, true,
+          "Turns off bug reporting entirely (useful for benchmarking).")
+TSAN_FLAG(bool, report_thread_leaks, true, "Report thread leaks at exit?")
+TSAN_FLAG(bool, report_destroy_locked, true,
+          "Report destruction of a locked mutex?")
+TSAN_FLAG(bool, report_mutex_bugs, true,
+          "Report incorrect usages of mutexes and mutex annotations?")
+TSAN_FLAG(bool, report_signal_unsafe, true,
+          "Report violations of async signal-safety "
+          "(e.g. malloc() call from a signal handler).")
+TSAN_FLAG(bool, report_atomic_races, true,
+          "Report races between atomic and plain memory accesses.")
+TSAN_FLAG(
+    bool, force_seq_cst_atomics, false,
+    "If set, all atomics are effectively sequentially consistent (seq_cst), "
+    "regardless of what user actually specified.")
+TSAN_FLAG(bool, print_benign, false, "Print matched \"benign\" races at exit.")
+TSAN_FLAG(bool, halt_on_error, false, "Exit after first reported error.")
+TSAN_FLAG(int, atexit_sleep_ms, 1000,
+          "Sleep in main thread before exiting for that many ms "
+          "(useful to catch \"at exit\" races).")
+TSAN_FLAG(const char *, profile_memory, "",
+          "If set, periodically write memory profile to that file.")
+TSAN_FLAG(int, flush_memory_ms, 0, "Flush shadow memory every X ms.")
+TSAN_FLAG(int, flush_symbolizer_ms, 5000, "Flush symbolizer caches every X ms.")
+TSAN_FLAG(
+    int, memory_limit_mb, 0,
+    "Resident memory limit in MB to aim at."
+    "If the process consumes more memory, then TSan will flush shadow memory.")
+TSAN_FLAG(bool, stop_on_start, false,
+          "Stops on start until __tsan_resume() is called (for debugging).")
+TSAN_FLAG(bool, running_on_valgrind, false,
+          "Controls whether RunningOnValgrind() returns true or false.")
+TSAN_FLAG(
+    int, history_size, kGoMode ? 1 : 3, // There are a lot of goroutines in Go.
+    "Per-thread history size, controls how many previous memory accesses "
+    "are remembered per thread.  Possible values are [0..7]. "
+    "history_size=0 amounts to 32K memory accesses.  Each next value doubles "
+    "the amount of memory accesses, up to history_size=7 that amounts to "
+    "4M memory accesses.  The default value is 2 (128K memory accesses).")
+TSAN_FLAG(int, io_sync, 1,
+          "Controls level of synchronization implied by IO operations. "
+          "0 - no synchronization "
+          "1 - reasonable level of synchronization (write->read)"
+          "2 - global synchronization of all IO operations.")
+TSAN_FLAG(bool, die_after_fork, true,
+          "Die after multi-threaded fork if the child creates new threads.")
+TSAN_FLAG(const char *, suppressions, "", "Suppressions file name.")
diff --git a/libsanitizer/tsan/tsan_interceptors.cc b/libsanitizer/tsan/tsan_interceptors.cc
index 7f1b6e45a8d..16465b96b6c 100644
--- a/libsanitizer/tsan/tsan_interceptors.cc
+++ b/libsanitizer/tsan/tsan_interceptors.cc
@@ -18,6 +18,7 @@
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "sanitizer_common/sanitizer_stacktrace.h"
 #include "interception/interception.h"
+#include "tsan_interceptors.h"
 #include "tsan_interface.h"
 #include "tsan_platform.h"
 #include "tsan_suppressions.h"
@@ -29,25 +30,46 @@ using namespace __tsan;  // NOLINT
 
 #if SANITIZER_FREEBSD
 #define __errno_location __error
-#define __libc_malloc __malloc
 #define __libc_realloc __realloc
 #define __libc_calloc __calloc
-#define __libc_free __free
 #define stdout __stdoutp
 #define stderr __stderrp
 #endif
 
+#if SANITIZER_LINUX || SANITIZER_FREEBSD
+#define PTHREAD_CREATE_DETACHED 1
+#elif SANITIZER_MAC
+#define PTHREAD_CREATE_DETACHED 2
+#endif
+
+
+#ifdef __mips__
+const int kSigCount = 129;
+#else
 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];
+};
+#else
 struct ucontext_t {
   // The size is determined by looking at sizeof of real ucontext_t on linux.
   u64 opaque[936 / sizeof(u64) + 1];
 };
+#endif
+
+#if defined(__x86_64__) || defined(__mips__)
+#define PTHREAD_ABI_BASE  "GLIBC_2.3.2"
+#elif defined(__aarch64__)
+#define PTHREAD_ABI_BASE  "GLIBC_2.17"
+#endif
 
 extern "C" int pthread_attr_init(void *attr);
 extern "C" int pthread_attr_destroy(void *attr);
@@ -66,10 +88,9 @@ extern "C" void *pthread_self();
 extern "C" void _exit(int status);
 extern "C" int *__errno_location();
 extern "C" int fileno_unlocked(void *stream);
-extern "C" void *__libc_malloc(uptr size);
 extern "C" void *__libc_calloc(uptr size, uptr n);
 extern "C" void *__libc_realloc(void *ptr, uptr size);
-extern "C" void __libc_free(void *ptr);
+extern "C" int dirfd(void *dirp);
 #if !SANITIZER_FREEBSD
 extern "C" int mallopt(int param, int value);
 #endif
@@ -86,8 +107,13 @@ const int SIGFPE = 8;
 const int SIGSEGV = 11;
 const int SIGPIPE = 13;
 const int SIGTERM = 15;
+#ifdef __mips__
+const int SIGBUS = 10;
+const int SIGSYS = 12;
+#else
 const int SIGBUS = 7;
 const int SIGSYS = 31;
+#endif
 void *const MAP_FAILED = (void*)-1;
 const int PTHREAD_BARRIER_SERIAL_THREAD = -1;
 const int MAP_FIXED = 0x10;
@@ -99,21 +125,27 @@ 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);
-
 #define errno (*__errno_location())
 
+typedef void (*sighandler_t)(int sig);
+typedef void (*sigactionhandler_t)(int sig, my_siginfo_t *siginfo, void *uctx);
+
 struct sigaction_t {
+#ifdef __mips__
+  u32 sa_flags;
+#endif
   union {
     sighandler_t sa_handler;
-    void (*sa_sigaction)(int sig, my_siginfo_t *siginfo, void *uctx);
+    sigactionhandler_t sa_sigaction;
   };
 #if SANITIZER_FREEBSD
   int sa_flags;
   __sanitizer_sigset_t sa_mask;
 #else
   __sanitizer_sigset_t sa_mask;
+#ifndef __mips__
   int sa_flags;
+#endif
   void (*sa_restorer)();
 #endif
 };
@@ -121,12 +153,19 @@ struct sigaction_t {
 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
+const int SA_SIGINFO = 0x40;
+const int SIG_SETMASK = 3;
+#elif defined(__mips__)
+const int SA_SIGINFO = 8;
+const int SIG_SETMASK = 3;
+#else
 const int SA_SIGINFO = 4;
 const int SIG_SETMASK = 2;
+#endif
 
-namespace std {
-struct nothrow_t {};
-}  // namespace std
+#define COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED \
+  (!cur_thread()->is_inited)
 
 static sigaction_t sigactions[kSigCount];
 
@@ -138,7 +177,7 @@ struct SignalDesc {
   ucontext_t ctx;
 };
 
-struct SignalContext {
+struct ThreadSignalContext {
   int int_signal_send;
   atomic_uintptr_t in_blocking_func;
   atomic_uintptr_t have_pending_signals;
@@ -153,16 +192,22 @@ static LibIgnore *libignore() {
 }
 
 void InitializeLibIgnore() {
-  libignore()->Init(*SuppressionContext::Get());
+  const SuppressionContext &supp = *Suppressions();
+  const uptr n = supp.SuppressionCount();
+  for (uptr i = 0; i < n; i++) {
+    const Suppression *s = supp.SuppressionAt(i);
+    if (0 == internal_strcmp(s->type, kSuppressionLib))
+      libignore()->AddIgnoredLibrary(s->templ);
+  }
   libignore()->OnLibraryLoaded(0);
 }
 
 }  // namespace __tsan
 
-static SignalContext *SigCtx(ThreadState *thr) {
-  SignalContext *ctx = (SignalContext*)thr->signal_ctx;
+static ThreadSignalContext *SigCtx(ThreadState *thr) {
+  ThreadSignalContext *ctx = (ThreadSignalContext*)thr->signal_ctx;
   if (ctx == 0 && !thr->is_dead) {
-    ctx = (SignalContext*)MmapOrDie(sizeof(*ctx), "SignalContext");
+    ctx = (ThreadSignalContext*)MmapOrDie(sizeof(*ctx), "ThreadSignalContext");
     MemoryResetRange(thr, (uptr)&SigCtx, (uptr)ctx, sizeof(*ctx));
     thr->signal_ctx = ctx;
   }
@@ -171,16 +216,6 @@ static SignalContext *SigCtx(ThreadState *thr) {
 
 static unsigned g_thread_finalize_key;
 
-class ScopedInterceptor {
- public:
-  ScopedInterceptor(ThreadState *thr, const char *fname, uptr pc);
-  ~ScopedInterceptor();
- private:
-  ThreadState *const thr_;
-  const uptr pc_;
-  bool in_ignored_lib_;
-};
-
 ScopedInterceptor::ScopedInterceptor(ThreadState *thr, const char *fname,
                                      uptr pc)
     : thr_(thr)
@@ -210,14 +245,6 @@ ScopedInterceptor::~ScopedInterceptor() {
   }
 }
 
-#define SCOPED_INTERCEPTOR_RAW(func, ...) \
-    ThreadState *thr = cur_thread(); \
-    const uptr caller_pc = GET_CALLER_PC(); \
-    ScopedInterceptor si(thr, #func, caller_pc); \
-    const uptr pc = StackTrace::GetCurrentPc(); \
-    (void)pc; \
-/**/
-
 #define SCOPED_TSAN_INTERCEPTOR(func, ...) \
     SCOPED_INTERCEPTOR_RAW(func, __VA_ARGS__); \
     if (REAL(func) == 0) { \
@@ -236,6 +263,13 @@ ScopedInterceptor::~ScopedInterceptor() {
 # define TSAN_INTERCEPT_VER(func, ver) INTERCEPT_FUNCTION_VER(func, ver)
 #endif
 
+#define READ_STRING_OF_LEN(thr, pc, s, len, n)                 \
+  MemoryAccessRange((thr), (pc), (uptr)(s),                         \
+    common_flags()->strict_string_checks ? (len) + 1 : (n), false)
+
+#define READ_STRING(thr, pc, s, n)                             \
+    READ_STRING_OF_LEN((thr), (pc), (s), internal_strlen(s), (n))
+
 #define BLOCK_REAL(name) (BlockingCall(thr), REAL(name))
 
 struct BlockingCall {
@@ -263,7 +297,7 @@ struct BlockingCall {
   }
 
   ThreadState *thr;
-  SignalContext *ctx;
+  ThreadSignalContext *ctx;
 };
 
 TSAN_INTERCEPTOR(unsigned, sleep, unsigned sec) {
@@ -375,7 +409,7 @@ static void JmpBufGarbageCollect(ThreadState *thr, uptr sp) {
 }
 
 static void SetJmp(ThreadState *thr, uptr sp, uptr mangled_sp) {
-  if (thr->shadow_stack_pos == 0)  // called from libc guts during bootstrap
+  if (!thr->is_inited)  // called from libc guts during bootstrap
     return;
   // Cleanup old bufs.
   JmpBufGarbageCollect(thr, sp);
@@ -384,7 +418,7 @@ static void SetJmp(ThreadState *thr, uptr sp, uptr mangled_sp) {
   buf->sp = sp;
   buf->mangled_sp = mangled_sp;
   buf->shadow_stack_pos = thr->shadow_stack_pos;
-  SignalContext *sctx = SigCtx(thr);
+  ThreadSignalContext *sctx = SigCtx(thr);
   buf->int_signal_send = sctx ? sctx->int_signal_send : 0;
   buf->in_blocking_func = sctx ?
       atomic_load(&sctx->in_blocking_func, memory_order_relaxed) :
@@ -407,7 +441,7 @@ static void LongJmp(ThreadState *thr, uptr *env) {
       // Unwind the stack.
       while (thr->shadow_stack_pos > buf->shadow_stack_pos)
         FuncExit(thr);
-      SignalContext *sctx = SigCtx(thr);
+      ThreadSignalContext *sctx = SigCtx(thr);
       if (sctx) {
         sctx->int_signal_send = buf->int_signal_send;
         atomic_store(&sctx->in_blocking_func, buf->in_blocking_func,
@@ -503,14 +537,10 @@ TSAN_INTERCEPTOR(void*, __libc_memalign, uptr align, uptr sz) {
 TSAN_INTERCEPTOR(void*, calloc, uptr size, uptr n) {
   if (cur_thread()->in_symbolizer)
     return __libc_calloc(size, n);
-  if (__sanitizer::CallocShouldReturnNullDueToOverflow(size, n))
-    return AllocatorReturnNull();
   void *p = 0;
   {
     SCOPED_INTERCEPTOR_RAW(calloc, size, n);
-    p = user_alloc(thr, pc, n * size);
-    if (p)
-      internal_memset(p, 0, n * size);
+    p = user_calloc(thr, pc, size, n);
   }
   invoke_malloc_hook(p, n * size);
   return p;
@@ -554,73 +584,6 @@ TSAN_INTERCEPTOR(uptr, malloc_usable_size, void *p) {
   return user_alloc_usable_size(p);
 }
 
-#define OPERATOR_NEW_BODY(mangled_name) \
-  if (cur_thread()->in_symbolizer) \
-    return __libc_malloc(size); \
-  void *p = 0; \
-  {  \
-    SCOPED_INTERCEPTOR_RAW(mangled_name, size); \
-    p = user_alloc(thr, pc, size); \
-  }  \
-  invoke_malloc_hook(p, size);  \
-  return p;
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void *operator new(__sanitizer::uptr size);
-void *operator new(__sanitizer::uptr size) {
-  OPERATOR_NEW_BODY(_Znwm);
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void *operator new[](__sanitizer::uptr size);
-void *operator new[](__sanitizer::uptr size) {
-  OPERATOR_NEW_BODY(_Znam);
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void *operator new(__sanitizer::uptr size, std::nothrow_t const&);
-void *operator new(__sanitizer::uptr size, std::nothrow_t const&) {
-  OPERATOR_NEW_BODY(_ZnwmRKSt9nothrow_t);
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void *operator new[](__sanitizer::uptr size, std::nothrow_t const&);
-void *operator new[](__sanitizer::uptr size, std::nothrow_t const&) {
-  OPERATOR_NEW_BODY(_ZnamRKSt9nothrow_t);
-}
-
-#define OPERATOR_DELETE_BODY(mangled_name) \
-  if (ptr == 0) return;  \
-  if (cur_thread()->in_symbolizer) \
-    return __libc_free(ptr); \
-  invoke_free_hook(ptr);  \
-  SCOPED_INTERCEPTOR_RAW(mangled_name, ptr);  \
-  user_free(thr, pc, ptr);
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void operator delete(void *ptr) throw();
-void operator delete(void *ptr) throw() {
-  OPERATOR_DELETE_BODY(_ZdlPv);
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void operator delete[](void *ptr) throw();
-void operator delete[](void *ptr) throw() {
-  OPERATOR_DELETE_BODY(_ZdaPv);
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void operator delete(void *ptr, std::nothrow_t const&);
-void operator delete(void *ptr, std::nothrow_t const&) {
-  OPERATOR_DELETE_BODY(_ZdlPvRKSt9nothrow_t);
-}
-
-SANITIZER_INTERFACE_ATTRIBUTE
-void operator delete[](void *ptr, std::nothrow_t const&);
-void operator delete[](void *ptr, std::nothrow_t const&) {
-  OPERATOR_DELETE_BODY(_ZdaPvRKSt9nothrow_t);
-}
-
 TSAN_INTERCEPTOR(uptr, strlen, const char *s) {
   SCOPED_TSAN_INTERCEPTOR(strlen, s);
   uptr len = internal_strlen(s);
@@ -629,29 +592,22 @@ TSAN_INTERCEPTOR(uptr, strlen, const char *s) {
 }
 
 TSAN_INTERCEPTOR(void*, memset, void *dst, int v, uptr size) {
-  SCOPED_TSAN_INTERCEPTOR(memset, dst, v, size);
-  MemoryAccessRange(thr, pc, (uptr)dst, size, true);
+  // On FreeBSD we get here from libthr internals on thread initialization.
+  if (!COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED) {
+    SCOPED_TSAN_INTERCEPTOR(memset, dst, v, size);
+    MemoryAccessRange(thr, pc, (uptr)dst, size, true);
+  }
   return internal_memset(dst, v, size);
 }
 
 TSAN_INTERCEPTOR(void*, memcpy, void *dst, const void *src, uptr size) {
-  SCOPED_TSAN_INTERCEPTOR(memcpy, dst, src, size);
-  MemoryAccessRange(thr, pc, (uptr)dst, size, true);
-  MemoryAccessRange(thr, pc, (uptr)src, size, false);
-  return internal_memcpy(dst, src, size);
-}
-
-TSAN_INTERCEPTOR(int, memcmp, const void *s1, const void *s2, uptr n) {
-  SCOPED_TSAN_INTERCEPTOR(memcmp, s1, s2, n);
-  int res = 0;
-  uptr len = 0;
-  for (; len < n; len++) {
-    if ((res = ((unsigned char*)s1)[len] - ((unsigned char*)s2)[len]))
-      break;
+  // On FreeBSD we get here from libthr internals on thread initialization.
+  if (!COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED) {
+    SCOPED_TSAN_INTERCEPTOR(memcpy, dst, src, size);
+    MemoryAccessRange(thr, pc, (uptr)dst, size, true);
+    MemoryAccessRange(thr, pc, (uptr)src, size, false);
   }
-  MemoryAccessRange(thr, pc, (uptr)s1, len < n ? len + 1 : n, false);
-  MemoryAccessRange(thr, pc, (uptr)s2, len < n ? len + 1 : n, false);
-  return res;
+  return internal_memcpy(dst, src, size);
 }
 
 TSAN_INTERCEPTOR(void*, memmove, void *dst, void *src, uptr n) {
@@ -664,8 +620,9 @@ TSAN_INTERCEPTOR(void*, memmove, void *dst, void *src, uptr n) {
 TSAN_INTERCEPTOR(char*, strchr, char *s, int c) {
   SCOPED_TSAN_INTERCEPTOR(strchr, s, c);
   char *res = REAL(strchr)(s, c);
-  uptr len = res ? (char*)res - (char*)s + 1 : internal_strlen(s) + 1;
-  MemoryAccessRange(thr, pc, (uptr)s, len, false);
+  uptr len = internal_strlen(s);
+  uptr n = res ? (char*)res - (char*)s + 1 : len + 1;
+  READ_STRING_OF_LEN(thr, pc, s, len, n);
   return res;
 }
 
@@ -673,7 +630,7 @@ TSAN_INTERCEPTOR(char*, strchrnul, char *s, int c) {
   SCOPED_TSAN_INTERCEPTOR(strchrnul, s, c);
   char *res = REAL(strchrnul)(s, c);
   uptr len = (char*)res - (char*)s + 1;
-  MemoryAccessRange(thr, pc, (uptr)s, len, false);
+  READ_STRING(thr, pc, s, len);
   return res;
 }
 
@@ -699,16 +656,6 @@ TSAN_INTERCEPTOR(char*, strncpy, char *dst, char *src, uptr n) {
   return REAL(strncpy)(dst, src, n);
 }
 
-TSAN_INTERCEPTOR(const char*, strstr, const char *s1, const char *s2) {
-  SCOPED_TSAN_INTERCEPTOR(strstr, s1, s2);
-  const char *res = REAL(strstr)(s1, s2);
-  uptr len1 = internal_strlen(s1);
-  uptr len2 = internal_strlen(s2);
-  MemoryAccessRange(thr, pc, (uptr)s1, len1 + 1, false);
-  MemoryAccessRange(thr, pc, (uptr)s2, len2 + 1, false);
-  return res;
-}
-
 TSAN_INTERCEPTOR(char*, strdup, const char *str) {
   SCOPED_TSAN_INTERCEPTOR(strdup, str);
   // strdup will call malloc, so no instrumentation is required here.
@@ -764,7 +711,11 @@ TSAN_INTERCEPTOR(void*, mmap64, void *addr, long_t sz, int prot,
 
 TSAN_INTERCEPTOR(int, munmap, void *addr, long_t sz) {
   SCOPED_TSAN_INTERCEPTOR(munmap, addr, sz);
-  DontNeedShadowFor((uptr)addr, sz);
+  if (sz != 0) {
+    // If sz == 0, munmap will return EINVAL and don't unmap any memory.
+    DontNeedShadowFor((uptr)addr, sz);
+    ctx->metamap.ResetRange(thr, pc, (uptr)addr, (uptr)sz);
+  }
   int res = REAL(munmap)(addr, sz);
   return res;
 }
@@ -846,7 +797,7 @@ static void thread_finalize(void *v) {
   {
     ThreadState *thr = cur_thread();
     ThreadFinish(thr);
-    SignalContext *sctx = thr->signal_ctx;
+    ThreadSignalContext *sctx = thr->signal_ctx;
     if (sctx) {
       thr->signal_ctx = 0;
       UnmapOrDie(sctx, sizeof(*sctx));
@@ -872,15 +823,15 @@ extern "C" void *__tsan_thread_start_func(void *arg) {
     ScopedIgnoreInterceptors ignore;
     ThreadIgnoreBegin(thr, 0);
     if (pthread_setspecific(g_thread_finalize_key,
-                            (void *)kPthreadDestructorIterations)) {
+                            (void *)GetPthreadDestructorIterations())) {
       Printf("ThreadSanitizer: failed to set thread key\n");
       Die();
     }
     ThreadIgnoreEnd(thr, 0);
     while ((tid = atomic_load(&p->tid, memory_order_acquire)) == 0)
       pthread_yield();
-    atomic_store(&p->tid, 0, memory_order_release);
     ThreadStart(thr, tid, GetTid());
+    atomic_store(&p->tid, 0, memory_order_release);
   }
   void *res = callback(param);
   // Prevent the callback from being tail called,
@@ -926,8 +877,16 @@ TSAN_INTERCEPTOR(int, pthread_create,
     ThreadIgnoreEnd(thr, pc);
   }
   if (res == 0) {
-    int tid = ThreadCreate(thr, pc, *(uptr*)th, detached);
+    int tid = ThreadCreate(thr, pc, *(uptr*)th,
+                           detached == PTHREAD_CREATE_DETACHED);
     CHECK_NE(tid, 0);
+    // Synchronization on p.tid serves two purposes:
+    // 1. ThreadCreate must finish before the new thread starts.
+    //    Otherwise the new thread can call pthread_detach, but the pthread_t
+    //    identifier is not yet registered in ThreadRegistry by ThreadCreate.
+    // 2. ThreadStart must finish before this thread continues.
+    //    Otherwise, this thread can call pthread_detach and reset thr->sync
+    //    before the new thread got a chance to acquire from it in ThreadStart.
     atomic_store(&p.tid, tid, memory_order_release);
     while (atomic_load(&p.tid, memory_order_acquire) != 0)
       pthread_yield();
@@ -949,6 +908,8 @@ TSAN_INTERCEPTOR(int, pthread_join, void *th, void **ret) {
   return res;
 }
 
+DEFINE_REAL_PTHREAD_FUNCTIONS
+
 TSAN_INTERCEPTOR(int, pthread_detach, void *th) {
   SCOPED_TSAN_INTERCEPTOR(pthread_detach, th);
   int tid = ThreadTid(thr, pc, (uptr)th);
@@ -998,13 +959,25 @@ static void *init_cond(void *c, bool force = false) {
 }
 
 struct CondMutexUnlockCtx {
+  ScopedInterceptor *si;
   ThreadState *thr;
   uptr pc;
   void *m;
 };
 
 static void cond_mutex_unlock(CondMutexUnlockCtx *arg) {
+  // pthread_cond_wait interceptor has enabled async signal delivery
+  // (see BlockingCall below). Disable async signals since we are running
+  // tsan code. Also ScopedInterceptor and BlockingCall destructors won't run
+  // since the thread is cancelled, so we have to manually execute them
+  // (the thread still can run some user code due to pthread_cleanup_push).
+  ThreadSignalContext *ctx = SigCtx(arg->thr);
+  CHECK_EQ(atomic_load(&ctx->in_blocking_func, memory_order_relaxed), 1);
+  atomic_store(&ctx->in_blocking_func, 0, memory_order_relaxed);
   MutexLock(arg->thr, arg->pc, (uptr)arg->m);
+  // Undo BlockingCall ctor effects.
+  arg->thr->ignore_interceptors--;
+  arg->si->~ScopedInterceptor();
 }
 
 INTERCEPTOR(int, pthread_cond_init, void *c, void *a) {
@@ -1017,14 +990,19 @@ INTERCEPTOR(int, pthread_cond_init, void *c, void *a) {
 INTERCEPTOR(int, pthread_cond_wait, void *c, void *m) {
   void *cond = init_cond(c);
   SCOPED_TSAN_INTERCEPTOR(pthread_cond_wait, cond, m);
-  MutexUnlock(thr, pc, (uptr)m);
   MemoryAccessRange(thr, pc, (uptr)c, sizeof(uptr), false);
-  CondMutexUnlockCtx arg = {thr, pc, m};
+  MutexUnlock(thr, pc, (uptr)m);
+  CondMutexUnlockCtx arg = {&si, thr, pc, m};
+  int res = 0;
   // This ensures that we handle mutex lock even in case of pthread_cancel.
   // See test/tsan/cond_cancel.cc.
-  int res = call_pthread_cancel_with_cleanup(
-      (int(*)(void *c, void *m, void *abstime))REAL(pthread_cond_wait),
-      cond, m, 0, (void(*)(void *arg))cond_mutex_unlock, &arg);
+  {
+    // Enable signal delivery while the thread is blocked.
+    BlockingCall bc(thr);
+    res = call_pthread_cancel_with_cleanup(
+        (int(*)(void *c, void *m, void *abstime))REAL(pthread_cond_wait),
+        cond, m, 0, (void(*)(void *arg))cond_mutex_unlock, &arg);
+  }
   if (res == errno_EOWNERDEAD)
     MutexRepair(thr, pc, (uptr)m);
   MutexLock(thr, pc, (uptr)m);
@@ -1034,14 +1012,18 @@ INTERCEPTOR(int, pthread_cond_wait, void *c, void *m) {
 INTERCEPTOR(int, pthread_cond_timedwait, void *c, void *m, void *abstime) {
   void *cond = init_cond(c);
   SCOPED_TSAN_INTERCEPTOR(pthread_cond_timedwait, cond, m, abstime);
-  MutexUnlock(thr, pc, (uptr)m);
   MemoryAccessRange(thr, pc, (uptr)c, sizeof(uptr), false);
-  CondMutexUnlockCtx arg = {thr, pc, m};
+  MutexUnlock(thr, pc, (uptr)m);
+  CondMutexUnlockCtx arg = {&si, thr, pc, m};
+  int res = 0;
   // This ensures that we handle mutex lock even in case of pthread_cancel.
   // See test/tsan/cond_cancel.cc.
-  int res = call_pthread_cancel_with_cleanup(
-      REAL(pthread_cond_timedwait), cond, m, abstime,
-      (void(*)(void *arg))cond_mutex_unlock, &arg);
+  {
+    BlockingCall bc(thr);
+    res = call_pthread_cancel_with_cleanup(
+        REAL(pthread_cond_timedwait), cond, m, abstime,
+        (void(*)(void *arg))cond_mutex_unlock, &arg);
+  }
   if (res == errno_EOWNERDEAD)
     MutexRepair(thr, pc, (uptr)m);
   MutexLock(thr, pc, (uptr)m);
@@ -1290,64 +1272,10 @@ TSAN_INTERCEPTOR(int, pthread_once, void *o, void (*f)()) {
   return 0;
 }
 
-TSAN_INTERCEPTOR(int, sem_init, void *s, int pshared, unsigned value) {
-  SCOPED_TSAN_INTERCEPTOR(sem_init, s, pshared, value);
-  int res = REAL(sem_init)(s, pshared, value);
-  return res;
-}
-
-TSAN_INTERCEPTOR(int, sem_destroy, void *s) {
-  SCOPED_TSAN_INTERCEPTOR(sem_destroy, s);
-  int res = REAL(sem_destroy)(s);
-  return res;
-}
-
-TSAN_INTERCEPTOR(int, sem_wait, void *s) {
-  SCOPED_TSAN_INTERCEPTOR(sem_wait, s);
-  int res = BLOCK_REAL(sem_wait)(s);
-  if (res == 0) {
-    Acquire(thr, pc, (uptr)s);
-  }
-  return res;
-}
-
-TSAN_INTERCEPTOR(int, sem_trywait, void *s) {
-  SCOPED_TSAN_INTERCEPTOR(sem_trywait, s);
-  int res = BLOCK_REAL(sem_trywait)(s);
-  if (res == 0) {
-    Acquire(thr, pc, (uptr)s);
-  }
-  return res;
-}
-
-TSAN_INTERCEPTOR(int, sem_timedwait, void *s, void *abstime) {
-  SCOPED_TSAN_INTERCEPTOR(sem_timedwait, s, abstime);
-  int res = BLOCK_REAL(sem_timedwait)(s, abstime);
-  if (res == 0) {
-    Acquire(thr, pc, (uptr)s);
-  }
-  return res;
-}
-
-TSAN_INTERCEPTOR(int, sem_post, void *s) {
-  SCOPED_TSAN_INTERCEPTOR(sem_post, s);
-  Release(thr, pc, (uptr)s);
-  int res = REAL(sem_post)(s);
-  return res;
-}
-
-TSAN_INTERCEPTOR(int, sem_getvalue, void *s, int *sval) {
-  SCOPED_TSAN_INTERCEPTOR(sem_getvalue, s, sval);
-  int res = REAL(sem_getvalue)(s, sval);
-  if (res == 0) {
-    Acquire(thr, pc, (uptr)s);
-  }
-  return res;
-}
-
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, __xstat, int version, const char *path, void *buf) {
   SCOPED_TSAN_INTERCEPTOR(__xstat, version, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__xstat)(version, path, buf);
 }
 #define TSAN_MAYBE_INTERCEPT___XSTAT TSAN_INTERCEPT(__xstat)
@@ -1358,9 +1286,11 @@ TSAN_INTERCEPTOR(int, __xstat, int version, const char *path, void *buf) {
 TSAN_INTERCEPTOR(int, stat, const char *path, void *buf) {
 #if SANITIZER_FREEBSD
   SCOPED_TSAN_INTERCEPTOR(stat, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(stat)(path, buf);
 #else
   SCOPED_TSAN_INTERCEPTOR(__xstat, 0, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__xstat)(0, path, buf);
 #endif
 }
@@ -1368,6 +1298,7 @@ TSAN_INTERCEPTOR(int, stat, const char *path, void *buf) {
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, __xstat64, int version, const char *path, void *buf) {
   SCOPED_TSAN_INTERCEPTOR(__xstat64, version, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__xstat64)(version, path, buf);
 }
 #define TSAN_MAYBE_INTERCEPT___XSTAT64 TSAN_INTERCEPT(__xstat64)
@@ -1378,6 +1309,7 @@ TSAN_INTERCEPTOR(int, __xstat64, int version, const char *path, void *buf) {
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, stat64, const char *path, void *buf) {
   SCOPED_TSAN_INTERCEPTOR(__xstat64, 0, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__xstat64)(0, path, buf);
 }
 #define TSAN_MAYBE_INTERCEPT_STAT64 TSAN_INTERCEPT(stat64)
@@ -1388,6 +1320,7 @@ TSAN_INTERCEPTOR(int, stat64, const char *path, void *buf) {
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, __lxstat, int version, const char *path, void *buf) {
   SCOPED_TSAN_INTERCEPTOR(__lxstat, version, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__lxstat)(version, path, buf);
 }
 #define TSAN_MAYBE_INTERCEPT___LXSTAT TSAN_INTERCEPT(__lxstat)
@@ -1398,9 +1331,11 @@ TSAN_INTERCEPTOR(int, __lxstat, int version, const char *path, void *buf) {
 TSAN_INTERCEPTOR(int, lstat, const char *path, void *buf) {
 #if SANITIZER_FREEBSD
   SCOPED_TSAN_INTERCEPTOR(lstat, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(lstat)(path, buf);
 #else
   SCOPED_TSAN_INTERCEPTOR(__lxstat, 0, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__lxstat)(0, path, buf);
 #endif
 }
@@ -1408,6 +1343,7 @@ TSAN_INTERCEPTOR(int, lstat, const char *path, void *buf) {
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, __lxstat64, int version, const char *path, void *buf) {
   SCOPED_TSAN_INTERCEPTOR(__lxstat64, version, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__lxstat64)(version, path, buf);
 }
 #define TSAN_MAYBE_INTERCEPT___LXSTAT64 TSAN_INTERCEPT(__lxstat64)
@@ -1418,6 +1354,7 @@ TSAN_INTERCEPTOR(int, __lxstat64, int version, const char *path, void *buf) {
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, lstat64, const char *path, void *buf) {
   SCOPED_TSAN_INTERCEPTOR(__lxstat64, 0, path, buf);
+  READ_STRING(thr, pc, path, 0);
   return REAL(__lxstat64)(0, path, buf);
 }
 #define TSAN_MAYBE_INTERCEPT_LSTAT64 TSAN_INTERCEPT(lstat64)
@@ -1477,6 +1414,7 @@ TSAN_INTERCEPTOR(int, fstat64, int fd, void *buf) {
 
 TSAN_INTERCEPTOR(int, open, const char *name, int flags, int mode) {
   SCOPED_TSAN_INTERCEPTOR(open, name, flags, mode);
+  READ_STRING(thr, pc, name, 0);
   int fd = REAL(open)(name, flags, mode);
   if (fd >= 0)
     FdFileCreate(thr, pc, fd);
@@ -1486,6 +1424,7 @@ TSAN_INTERCEPTOR(int, open, const char *name, int flags, int mode) {
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, open64, const char *name, int flags, int mode) {
   SCOPED_TSAN_INTERCEPTOR(open64, name, flags, mode);
+  READ_STRING(thr, pc, name, 0);
   int fd = REAL(open64)(name, flags, mode);
   if (fd >= 0)
     FdFileCreate(thr, pc, fd);
@@ -1498,6 +1437,7 @@ TSAN_INTERCEPTOR(int, open64, const char *name, int flags, int mode) {
 
 TSAN_INTERCEPTOR(int, creat, const char *name, int mode) {
   SCOPED_TSAN_INTERCEPTOR(creat, name, mode);
+  READ_STRING(thr, pc, name, 0);
   int fd = REAL(creat)(name, mode);
   if (fd >= 0)
     FdFileCreate(thr, pc, fd);
@@ -1507,6 +1447,7 @@ TSAN_INTERCEPTOR(int, creat, const char *name, int mode) {
 #if !SANITIZER_FREEBSD
 TSAN_INTERCEPTOR(int, creat64, const char *name, int mode) {
   SCOPED_TSAN_INTERCEPTOR(creat64, name, mode);
+  READ_STRING(thr, pc, name, 0);
   int fd = REAL(creat64)(name, mode);
   if (fd >= 0)
     FdFileCreate(thr, pc, fd);
@@ -1521,7 +1462,7 @@ TSAN_INTERCEPTOR(int, dup, int oldfd) {
   SCOPED_TSAN_INTERCEPTOR(dup, oldfd);
   int newfd = REAL(dup)(oldfd);
   if (oldfd >= 0 && newfd >= 0 && newfd != oldfd)
-    FdDup(thr, pc, oldfd, newfd);
+    FdDup(thr, pc, oldfd, newfd, true);
   return newfd;
 }
 
@@ -1529,7 +1470,7 @@ TSAN_INTERCEPTOR(int, dup2, int oldfd, int newfd) {
   SCOPED_TSAN_INTERCEPTOR(dup2, oldfd, newfd);
   int newfd2 = REAL(dup2)(oldfd, newfd);
   if (oldfd >= 0 && newfd2 >= 0 && newfd2 != oldfd)
-    FdDup(thr, pc, oldfd, newfd2);
+    FdDup(thr, pc, oldfd, newfd2, false);
   return newfd2;
 }
 
@@ -1537,7 +1478,7 @@ TSAN_INTERCEPTOR(int, dup3, int oldfd, int newfd, int flags) {
   SCOPED_TSAN_INTERCEPTOR(dup3, oldfd, newfd, flags);
   int newfd2 = REAL(dup3)(oldfd, newfd, flags);
   if (oldfd >= 0 && newfd2 >= 0 && newfd2 != oldfd)
-    FdDup(thr, pc, oldfd, newfd2);
+    FdDup(thr, pc, oldfd, newfd2, false);
   return newfd2;
 }
 
@@ -1797,9 +1738,16 @@ TSAN_INTERCEPTOR(uptr, fwrite, const void *p, uptr size, uptr nmemb, void *f) {
   return REAL(fwrite)(p, size, nmemb, f);
 }
 
+static void FlushStreams() {
+  // Flushing all the streams here may freeze the process if a child thread is
+  // performing file stream operations at the same time.
+  REAL(fflush)(stdout);
+  REAL(fflush)(stderr);
+}
+
 TSAN_INTERCEPTOR(void, abort, int fake) {
   SCOPED_TSAN_INTERCEPTOR(abort, fake);
-  REAL(fflush)(0);
+  FlushStreams();
   REAL(abort)(fake);
 }
 
@@ -1816,12 +1764,11 @@ TSAN_INTERCEPTOR(int, rmdir, char *path) {
   return res;
 }
 
-TSAN_INTERCEPTOR(void*, opendir, char *path) {
-  SCOPED_TSAN_INTERCEPTOR(opendir, path);
-  void *res = REAL(opendir)(path);
-  if (res != 0)
-    Acquire(thr, pc, Dir2addr(path));
-  return res;
+TSAN_INTERCEPTOR(int, closedir, void *dirp) {
+  SCOPED_TSAN_INTERCEPTOR(closedir, dirp);
+  int fd = dirfd(dirp);
+  FdClose(thr, pc, fd);
+  return REAL(closedir)(dirp);
 }
 
 #if !SANITIZER_FREEBSD
@@ -1865,15 +1812,18 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
   // Ensure that the handler does not spoil errno.
   const int saved_errno = errno;
   errno = 99;
-  // Need to remember pc before the call, because the handler can reset it.
-  uptr pc = sigact ?
+  // 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;
-  pc += 1;  // return address is expected, OutputReport() will undo this
-  if (sigact)
-    sigactions[sig].sa_sigaction(sig, info, uctx);
-  else
-    sigactions[sig].sa_handler(sig);
+  if (pc != (uptr)SIG_DFL && pc != (uptr)SIG_IGN) {
+    if (sigact)
+      ((sigactionhandler_t)pc)(sig, info, uctx);
+    else
+      ((sighandler_t)pc)(sig);
+  }
   // We do not detect errno spoiling for SIGTERM,
   // because some SIGTERM handlers do spoil errno but reraise SIGTERM,
   // tsan reports false positive in such case.
@@ -1883,10 +1833,12 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
   // signal; and it looks too fragile to intercept all ways to reraise a signal.
   if (flags()->report_bugs && !sync && sig != SIGTERM && errno != 99) {
     VarSizeStackTrace stack;
-    ObtainCurrentStack(thr, pc, &stack);
+    // StackTrace::GetNestInstructionPc(pc) is used because return address is
+    // expected, OutputReport() will undo this.
+    ObtainCurrentStack(thr, StackTrace::GetNextInstructionPc(pc), &stack);
     ThreadRegistryLock l(ctx->thread_registry);
     ScopedReport rep(ReportTypeErrnoInSignal);
-    if (!IsFiredSuppression(ctx, rep, stack)) {
+    if (!IsFiredSuppression(ctx, ReportTypeErrnoInSignal, stack)) {
       rep.AddStack(stack, true);
       OutputReport(thr, rep);
     }
@@ -1895,7 +1847,7 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
 }
 
 void ProcessPendingSignals(ThreadState *thr) {
-  SignalContext *sctx = SigCtx(thr);
+  ThreadSignalContext *sctx = SigCtx(thr);
   if (sctx == 0 ||
       atomic_load(&sctx->have_pending_signals, memory_order_relaxed) == 0)
     return;
@@ -1903,26 +1855,23 @@ void ProcessPendingSignals(ThreadState *thr) {
   atomic_fetch_add(&thr->in_signal_handler, 1, memory_order_relaxed);
   // These are too big for stack.
   static THREADLOCAL __sanitizer_sigset_t emptyset, oldset;
-  REAL(sigfillset)(&emptyset);
-  pthread_sigmask(SIG_SETMASK, &emptyset, &oldset);
+  CHECK_EQ(0, REAL(sigfillset)(&emptyset));
+  CHECK_EQ(0, pthread_sigmask(SIG_SETMASK, &emptyset, &oldset));
   for (int sig = 0; sig < kSigCount; sig++) {
     SignalDesc *signal = &sctx->pending_signals[sig];
     if (signal->armed) {
       signal->armed = false;
-      if (sigactions[sig].sa_handler != SIG_DFL
-          && sigactions[sig].sa_handler != SIG_IGN) {
-        CallUserSignalHandler(thr, false, true, signal->sigaction,
-            sig, &signal->siginfo, &signal->ctx);
-      }
+      CallUserSignalHandler(thr, false, true, signal->sigaction, sig,
+          &signal->siginfo, &signal->ctx);
     }
   }
-  pthread_sigmask(SIG_SETMASK, &oldset, 0);
+  CHECK_EQ(0, pthread_sigmask(SIG_SETMASK, &oldset, 0));
   atomic_fetch_add(&thr->in_signal_handler, -1, memory_order_relaxed);
 }
 
 }  // namespace __tsan
 
-static bool is_sync_signal(SignalContext *sctx, int sig) {
+static bool is_sync_signal(ThreadSignalContext *sctx, int sig) {
   return sig == SIGSEGV || sig == SIGBUS || sig == SIGILL ||
       sig == SIGABRT || sig == SIGFPE || sig == SIGPIPE || sig == SIGSYS ||
       // If we are sending signal to ourselves, we must process it now.
@@ -1932,7 +1881,7 @@ static bool is_sync_signal(SignalContext *sctx, int sig) {
 void ALWAYS_INLINE rtl_generic_sighandler(bool sigact, int sig,
     my_siginfo_t *info, void *ctx) {
   ThreadState *thr = cur_thread();
-  SignalContext *sctx = SigCtx(thr);
+  ThreadSignalContext *sctx = SigCtx(thr);
   if (sig < 0 || sig >= kSigCount) {
     VPrintf(1, "ThreadSanitizer: ignoring signal %d\n", sig);
     return;
@@ -1995,7 +1944,19 @@ TSAN_INTERCEPTOR(int, sigaction, int sig, sigaction_t *act, sigaction_t *old) {
     internal_memcpy(old, &sigactions[sig], sizeof(*old));
   if (act == 0)
     return 0;
-  internal_memcpy(&sigactions[sig], act, sizeof(*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 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
+  sigactions[sig].sa_restorer = act->sa_restorer;
+#endif
   sigaction_t newact;
   internal_memcpy(&newact, act, sizeof(newact));
   REAL(sigfillset)(&newact.sa_mask);
@@ -2029,7 +1990,7 @@ TSAN_INTERCEPTOR(int, sigsuspend, const __sanitizer_sigset_t *mask) {
 
 TSAN_INTERCEPTOR(int, raise, int sig) {
   SCOPED_TSAN_INTERCEPTOR(raise, sig);
-  SignalContext *sctx = SigCtx(thr);
+  ThreadSignalContext *sctx = SigCtx(thr);
   CHECK_NE(sctx, 0);
   int prev = sctx->int_signal_send;
   sctx->int_signal_send = sig;
@@ -2041,7 +2002,7 @@ TSAN_INTERCEPTOR(int, raise, int sig) {
 
 TSAN_INTERCEPTOR(int, kill, int pid, int sig) {
   SCOPED_TSAN_INTERCEPTOR(kill, pid, sig);
-  SignalContext *sctx = SigCtx(thr);
+  ThreadSignalContext *sctx = SigCtx(thr);
   CHECK_NE(sctx, 0);
   int prev = sctx->int_signal_send;
   if (pid == (int)internal_getpid()) {
@@ -2057,7 +2018,7 @@ TSAN_INTERCEPTOR(int, kill, int pid, int sig) {
 
 TSAN_INTERCEPTOR(int, pthread_kill, void *tid, int sig) {
   SCOPED_TSAN_INTERCEPTOR(pthread_kill, tid, sig);
-  SignalContext *sctx = SigCtx(thr);
+  ThreadSignalContext *sctx = SigCtx(thr);
   CHECK_NE(sctx, 0);
   int prev = sctx->int_signal_send;
   if (tid == pthread_self()) {
@@ -2126,9 +2087,53 @@ TSAN_INTERCEPTOR(int, vfork, int fake) {
   return WRAP(fork)(fake);
 }
 
+typedef int (*dl_iterate_phdr_cb_t)(__sanitizer_dl_phdr_info *info, SIZE_T size,
+                                    void *data);
+struct dl_iterate_phdr_data {
+  ThreadState *thr;
+  uptr pc;
+  dl_iterate_phdr_cb_t cb;
+  void *data;
+};
+
+static bool IsAppNotRodata(uptr addr) {
+  return IsAppMem(addr) && *(u64*)MemToShadow(addr) != kShadowRodata;
+}
+
+static int dl_iterate_phdr_cb(__sanitizer_dl_phdr_info *info, SIZE_T size,
+                              void *data) {
+  dl_iterate_phdr_data *cbdata = (dl_iterate_phdr_data *)data;
+  // dlopen/dlclose allocate/free dynamic-linker-internal memory, which is later
+  // accessible in dl_iterate_phdr callback. But we don't see synchronization
+  // inside of dynamic linker, so we "unpoison" it here in order to not
+  // produce false reports. Ignoring malloc/free in dlopen/dlclose is not enough
+  // because some libc functions call __libc_dlopen.
+  if (info && IsAppNotRodata((uptr)info->dlpi_name))
+    MemoryResetRange(cbdata->thr, cbdata->pc, (uptr)info->dlpi_name,
+                     internal_strlen(info->dlpi_name));
+  int res = cbdata->cb(info, size, cbdata->data);
+  // Perform the check one more time in case info->dlpi_name was overwritten
+  // by user callback.
+  if (info && IsAppNotRodata((uptr)info->dlpi_name))
+    MemoryResetRange(cbdata->thr, cbdata->pc, (uptr)info->dlpi_name,
+                     internal_strlen(info->dlpi_name));
+  return res;
+}
+
+TSAN_INTERCEPTOR(int, dl_iterate_phdr, dl_iterate_phdr_cb_t cb, void *data) {
+  SCOPED_TSAN_INTERCEPTOR(dl_iterate_phdr, cb, data);
+  dl_iterate_phdr_data cbdata;
+  cbdata.thr = thr;
+  cbdata.pc = pc;
+  cbdata.cb = cb;
+  cbdata.data = data;
+  int res = REAL(dl_iterate_phdr)(dl_iterate_phdr_cb, &cbdata);
+  return res;
+}
+
 static int OnExit(ThreadState *thr) {
   int status = Finalize(thr);
-  REAL(fflush)(0);
+  FlushStreams();
   return status;
 }
 
@@ -2161,6 +2166,16 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
 #undef SANITIZER_INTERCEPT_FGETPWENT
 #undef SANITIZER_INTERCEPT_GETPWNAM_AND_FRIENDS
 #undef SANITIZER_INTERCEPT_GETPWNAM_R_AND_FRIENDS
+// __tls_get_addr can be called with mis-aligned stack due to:
+// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58066
+// There are two potential issues:
+// 1. Sanitizer code contains a MOVDQA spill (it does not seem to be the case
+// right now). or 2. ProcessPendingSignal calls user handler which contains
+// MOVDQA spill (this happens right now).
+// Since the interceptor only initializes memory for msan, the simplest solution
+// is to disable the interceptor in tsan (other sanitizers do not call
+// signal handlers from COMMON_INTERCEPTOR_ENTER).
+#undef SANITIZER_INTERCEPT_TLS_GET_ADDR
 
 #define COMMON_INTERCEPT_FUNCTION(name) INTERCEPT_FUNCTION(name)
 
@@ -2199,12 +2214,21 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
     if (fd >= 0) FdClose(thr, pc, fd);           \
   }
 
-#define COMMON_INTERCEPTOR_LIBRARY_LOADED(filename, res)  \
+#define COMMON_INTERCEPTOR_LIBRARY_LOADED(filename, handle) \
   libignore()->OnLibraryLoaded(filename)
 
 #define COMMON_INTERCEPTOR_LIBRARY_UNLOADED() \
   libignore()->OnLibraryUnloaded()
 
+#define COMMON_INTERCEPTOR_ACQUIRE(ctx, u) \
+  Acquire(((TsanInterceptorContext *) ctx)->thr, pc, u)
+
+#define COMMON_INTERCEPTOR_RELEASE(ctx, u) \
+  Release(((TsanInterceptorContext *) ctx)->thr, pc, u)
+
+#define COMMON_INTERCEPTOR_DIR_ACQUIRE(ctx, path) \
+  Acquire(((TsanInterceptorContext *) ctx)->thr, pc, Dir2addr(path))
+
 #define COMMON_INTERCEPTOR_FD_ACQUIRE(ctx, fd) \
   FdAcquire(((TsanInterceptorContext *) ctx)->thr, pc, fd)
 
@@ -2244,6 +2268,14 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
   HandleRecvmsg(((TsanInterceptorContext *)ctx)->thr, \
       ((TsanInterceptorContext *)ctx)->pc, msg)
 
+#define COMMON_INTERCEPTOR_GET_TLS_RANGE(begin, end)                           \
+  if (TsanThread *t = GetCurrentThread()) {                                    \
+    *begin = t->tls_begin();                                                   \
+    *end = t->tls_end();                                                       \
+  } else {                                                                     \
+    *begin = *end = 0;                                                         \
+  }
+
 #include "sanitizer_common/sanitizer_common_interceptors.inc"
 
 #define TSAN_SYSCALL() \
@@ -2364,12 +2396,9 @@ static void finalize(void *arg) {
   ThreadState *thr = cur_thread();
   int status = Finalize(thr);
   // Make sure the output is not lost.
-  // Flushing all the streams here may freeze the process if a child thread is
-  // performing file stream operations at the same time.
-  REAL(fflush)(stdout);
-  REAL(fflush)(stderr);
+  FlushStreams();
   if (status)
-    REAL(_exit)(status);
+    Die();
 }
 
 static void unreachable() {
@@ -2381,7 +2410,6 @@ void InitializeInterceptors() {
   // We need to setup it early, because functions like dlsym() can call it.
   REAL(memset) = internal_memset;
   REAL(memcpy) = internal_memcpy;
-  REAL(memcmp) = internal_memcmp;
 
   // Instruct libc malloc to consume less memory.
 #if !SANITIZER_FREEBSD
@@ -2420,25 +2448,23 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(memset);
   TSAN_INTERCEPT(memcpy);
   TSAN_INTERCEPT(memmove);
-  TSAN_INTERCEPT(memcmp);
   TSAN_INTERCEPT(strchr);
   TSAN_INTERCEPT(strchrnul);
   TSAN_INTERCEPT(strrchr);
   TSAN_INTERCEPT(strcpy);  // NOLINT
   TSAN_INTERCEPT(strncpy);
-  TSAN_INTERCEPT(strstr);
   TSAN_INTERCEPT(strdup);
 
   TSAN_INTERCEPT(pthread_create);
   TSAN_INTERCEPT(pthread_join);
   TSAN_INTERCEPT(pthread_detach);
 
-  TSAN_INTERCEPT_VER(pthread_cond_init, "GLIBC_2.3.2");
-  TSAN_INTERCEPT_VER(pthread_cond_signal, "GLIBC_2.3.2");
-  TSAN_INTERCEPT_VER(pthread_cond_broadcast, "GLIBC_2.3.2");
-  TSAN_INTERCEPT_VER(pthread_cond_wait, "GLIBC_2.3.2");
-  TSAN_INTERCEPT_VER(pthread_cond_timedwait, "GLIBC_2.3.2");
-  TSAN_INTERCEPT_VER(pthread_cond_destroy, "GLIBC_2.3.2");
+  TSAN_INTERCEPT_VER(pthread_cond_init, PTHREAD_ABI_BASE);
+  TSAN_INTERCEPT_VER(pthread_cond_signal, PTHREAD_ABI_BASE);
+  TSAN_INTERCEPT_VER(pthread_cond_broadcast, PTHREAD_ABI_BASE);
+  TSAN_INTERCEPT_VER(pthread_cond_wait, PTHREAD_ABI_BASE);
+  TSAN_INTERCEPT_VER(pthread_cond_timedwait, PTHREAD_ABI_BASE);
+  TSAN_INTERCEPT_VER(pthread_cond_destroy, PTHREAD_ABI_BASE);
 
   TSAN_INTERCEPT(pthread_mutex_init);
   TSAN_INTERCEPT(pthread_mutex_destroy);
@@ -2467,14 +2493,6 @@ void InitializeInterceptors() {
 
   TSAN_INTERCEPT(pthread_once);
 
-  TSAN_INTERCEPT(sem_init);
-  TSAN_INTERCEPT(sem_destroy);
-  TSAN_INTERCEPT(sem_wait);
-  TSAN_INTERCEPT(sem_trywait);
-  TSAN_INTERCEPT(sem_timedwait);
-  TSAN_INTERCEPT(sem_post);
-  TSAN_INTERCEPT(sem_getvalue);
-
   TSAN_INTERCEPT(stat);
   TSAN_MAYBE_INTERCEPT___XSTAT;
   TSAN_MAYBE_INTERCEPT_STAT64;
@@ -2523,7 +2541,7 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(abort);
   TSAN_INTERCEPT(puts);
   TSAN_INTERCEPT(rmdir);
-  TSAN_INTERCEPT(opendir);
+  TSAN_INTERCEPT(closedir);
 
   TSAN_MAYBE_INTERCEPT_EPOLL_CTL;
   TSAN_MAYBE_INTERCEPT_EPOLL_WAIT;
@@ -2542,6 +2560,7 @@ void InitializeInterceptors() {
 
   TSAN_INTERCEPT(fork);
   TSAN_INTERCEPT(vfork);
+  TSAN_INTERCEPT(dl_iterate_phdr);
   TSAN_INTERCEPT(on_exit);
   TSAN_INTERCEPT(__cxa_atexit);
   TSAN_INTERCEPT(_exit);
@@ -2562,19 +2581,4 @@ void InitializeInterceptors() {
   FdInit();
 }
 
-void *internal_start_thread(void(*func)(void *arg), void *arg) {
-  // Start the thread with signals blocked, otherwise it can steal user signals.
-  __sanitizer_sigset_t set, old;
-  internal_sigfillset(&set);
-  internal_sigprocmask(SIG_SETMASK, &set, &old);
-  void *th;
-  REAL(pthread_create)(&th, 0, (void*(*)(void *arg))func, arg);
-  internal_sigprocmask(SIG_SETMASK, &old, 0);
-  return th;
-}
-
-void internal_join_thread(void *th) {
-  REAL(pthread_join)(th, 0);
-}
-
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_interceptors.h b/libsanitizer/tsan/tsan_interceptors.h
new file mode 100644
index 00000000000..4a246c647a0
--- /dev/null
+++ b/libsanitizer/tsan/tsan_interceptors.h
@@ -0,0 +1,35 @@
+#ifndef TSAN_INTERCEPTORS_H
+#include "sanitizer_common/sanitizer_stacktrace.h"
+#include "tsan_rtl.h"
+
+namespace __tsan {
+
+class ScopedInterceptor {
+ public:
+  ScopedInterceptor(ThreadState *thr, const char *fname, uptr pc);
+  ~ScopedInterceptor();
+ private:
+  ThreadState *const thr_;
+  const uptr pc_;
+  bool in_ignored_lib_;
+};
+
+}  // namespace __tsan
+
+#define SCOPED_INTERCEPTOR_RAW(func, ...) \
+    ThreadState *thr = cur_thread(); \
+    const uptr caller_pc = GET_CALLER_PC(); \
+    ScopedInterceptor si(thr, #func, caller_pc); \
+    const uptr pc = StackTrace::GetCurrentPc(); \
+    (void)pc; \
+/**/
+
+#if SANITIZER_FREEBSD
+#define __libc_free __free
+#define __libc_malloc __malloc
+#endif
+
+extern "C" void __libc_free(void *ptr);
+extern "C" void *__libc_malloc(uptr size);
+
+#endif  // TSAN_INTERCEPTORS_H
diff --git a/libsanitizer/tsan/tsan_interface.cc b/libsanitizer/tsan/tsan_interface.cc
index e056bd465bf..ee9a627076b 100644
--- a/libsanitizer/tsan/tsan_interface.cc
+++ b/libsanitizer/tsan/tsan_interface.cc
@@ -36,56 +36,89 @@ void __tsan_write16(void *addr) {
   MemoryWrite(cur_thread(), CALLERPC, (uptr)addr + 8, kSizeLog8);
 }
 
-u16 __tsan_unaligned_read2(const uu16 *addr) {
+void __tsan_read16_pc(void *addr, void *pc) {
+  MemoryRead(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog8);
+  MemoryRead(cur_thread(), (uptr)pc, (uptr)addr + 8, kSizeLog8);
+}
+
+void __tsan_write16_pc(void *addr, void *pc) {
+  MemoryWrite(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog8);
+  MemoryWrite(cur_thread(), (uptr)pc, (uptr)addr + 8, kSizeLog8);
+}
+
+// __tsan_unaligned_read/write calls are emitted by compiler.
+
+void __tsan_unaligned_read2(const void *addr) {
   UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, false, false);
-  return *addr;
 }
 
-u32 __tsan_unaligned_read4(const uu32 *addr) {
+void __tsan_unaligned_read4(const void *addr) {
   UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 4, false, false);
-  return *addr;
 }
 
-u64 __tsan_unaligned_read8(const uu64 *addr) {
+void __tsan_unaligned_read8(const void *addr) {
   UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 8, false, false);
-  return *addr;
 }
 
-void __tsan_unaligned_write2(uu16 *addr, u16 v) {
+void __tsan_unaligned_read16(const void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 16, false, false);
+}
+
+void __tsan_unaligned_write2(void *addr) {
   UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 2, true, false);
-  *addr = v;
 }
 
-void __tsan_unaligned_write4(uu32 *addr, u32 v) {
+void __tsan_unaligned_write4(void *addr) {
   UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 4, true, false);
-  *addr = v;
 }
 
-void __tsan_unaligned_write8(uu64 *addr, u64 v) {
+void __tsan_unaligned_write8(void *addr) {
   UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 8, true, false);
-  *addr = v;
 }
 
+void __tsan_unaligned_write16(void *addr) {
+  UnalignedMemoryAccess(cur_thread(), CALLERPC, (uptr)addr, 16, true, false);
+}
+
+// __sanitizer_unaligned_load/store are for user instrumentation.
+
 extern "C" {
 SANITIZER_INTERFACE_ATTRIBUTE
-uint16_t __sanitizer_unaligned_load16(void *addr)
-    ALIAS("__tsan_unaligned_read2");
+u16 __sanitizer_unaligned_load16(const uu16 *addr) {
+  __tsan_unaligned_read2(addr);
+  return *addr;
+}
+
 SANITIZER_INTERFACE_ATTRIBUTE
-uint32_t __sanitizer_unaligned_load32(void *addr)
-    ALIAS("__tsan_unaligned_read4");
+u32 __sanitizer_unaligned_load32(const uu32 *addr) {
+  __tsan_unaligned_read4(addr);
+  return *addr;
+}
+
 SANITIZER_INTERFACE_ATTRIBUTE
-uint64_t __sanitizer_unaligned_load64(void *addr)
-    ALIAS("__tsan_unaligned_read8");
+u64 __sanitizer_unaligned_load64(const uu64 *addr) {
+  __tsan_unaligned_read8(addr);
+  return *addr;
+}
+
 SANITIZER_INTERFACE_ATTRIBUTE
-void __sanitizer_unaligned_store16(void *addr, uint16_t v)
-    ALIAS("__tsan_unaligned_write2");
+void __sanitizer_unaligned_store16(uu16 *addr, u16 v) {
+  __tsan_unaligned_write2(addr);
+  *addr = v;
+}
+
 SANITIZER_INTERFACE_ATTRIBUTE
-void __sanitizer_unaligned_store32(void *addr, uint32_t v)
-    ALIAS("__tsan_unaligned_write4");
+void __sanitizer_unaligned_store32(uu32 *addr, u32 v) {
+  __tsan_unaligned_write4(addr);
+  *addr = v;
+}
+
 SANITIZER_INTERFACE_ATTRIBUTE
-void __sanitizer_unaligned_store64(void *addr, uint64_t v)
-    ALIAS("__tsan_unaligned_write8");
+void __sanitizer_unaligned_store64(uu64 *addr, u64 v) {
+  __tsan_unaligned_write8(addr);
+  *addr = v;
 }
+}  // extern "C"
 
 void __tsan_acquire(void *addr) {
   Acquire(cur_thread(), CALLERPC, (uptr)addr);
diff --git a/libsanitizer/tsan/tsan_interface.h b/libsanitizer/tsan/tsan_interface.h
index 63de7b2ab5c..4ff6fa1831e 100644
--- a/libsanitizer/tsan/tsan_interface.h
+++ b/libsanitizer/tsan/tsan_interface.h
@@ -39,12 +39,27 @@ SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write4(void *addr);
 SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write8(void *addr);
 SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write16(void *addr);
 
-SANITIZER_INTERFACE_ATTRIBUTE u16 __tsan_unaligned_read2(const uu16 *addr);
-SANITIZER_INTERFACE_ATTRIBUTE u32 __tsan_unaligned_read4(const uu32 *addr);
-SANITIZER_INTERFACE_ATTRIBUTE u64 __tsan_unaligned_read8(const uu64 *addr);
-SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_write2(uu16 *addr, u16 v);
-SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_write4(uu32 *addr, u32 v);
-SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_write8(uu64 *addr, u64 v);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_read2(const void *addr);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_read4(const void *addr);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_read8(const void *addr);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_read16(const void *addr);
+
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_write2(void *addr);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_write4(void *addr);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_write8(void *addr);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_unaligned_write16(void *addr);
+
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_read1_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_read2_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_read4_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_read8_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_read16_pc(void *addr, void *pc);
+
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write1_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write2_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write4_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write8_pc(void *addr, void *pc);
+SANITIZER_INTERFACE_ATTRIBUTE void __tsan_write16_pc(void *addr, void *pc);
 
 SANITIZER_INTERFACE_ATTRIBUTE void __tsan_vptr_read(void **vptr_p);
 SANITIZER_INTERFACE_ATTRIBUTE
diff --git a/libsanitizer/tsan/tsan_interface_ann.cc b/libsanitizer/tsan/tsan_interface_ann.cc
index b44e56898d6..19ff4050abc 100644
--- a/libsanitizer/tsan/tsan_interface_ann.cc
+++ b/libsanitizer/tsan/tsan_interface_ann.cc
@@ -61,8 +61,8 @@ static const int kMaxDescLen = 128;
 struct ExpectRace {
   ExpectRace *next;
   ExpectRace *prev;
-  int hitcount;
-  int addcount;
+  atomic_uintptr_t hitcount;
+  atomic_uintptr_t addcount;
   uptr addr;
   uptr size;
   char *file;
@@ -88,7 +88,8 @@ static void AddExpectRace(ExpectRace *list,
   ExpectRace *race = list->next;
   for (; race != list; race = race->next) {
     if (race->addr == addr && race->size == size) {
-      race->addcount++;
+      atomic_store_relaxed(&race->addcount,
+          atomic_load_relaxed(&race->addcount) + 1);
       return;
     }
   }
@@ -98,8 +99,8 @@ static void AddExpectRace(ExpectRace *list,
   race->file = f;
   race->line = l;
   race->desc[0] = 0;
-  race->hitcount = 0;
-  race->addcount = 1;
+  atomic_store_relaxed(&race->hitcount, 0);
+  atomic_store_relaxed(&race->addcount, 1);
   if (desc) {
     int i = 0;
     for (; i < kMaxDescLen - 1 && desc[i]; i++)
@@ -128,7 +129,7 @@ static bool CheckContains(ExpectRace *list, uptr addr, uptr size) {
     return false;
   DPrintf("Hit expected/benign race: %s addr=%zx:%d %s:%d\n",
       race->desc, race->addr, (int)race->size, race->file, race->line);
-  race->hitcount++;
+  atomic_fetch_add(&race->hitcount, 1, memory_order_relaxed);
   return true;
 }
 
@@ -144,7 +145,7 @@ void InitializeDynamicAnnotations() {
 }
 
 bool IsExpectedReport(uptr addr, uptr size) {
-  Lock lock(&dyn_ann_ctx->mtx);
+  ReadLock lock(&dyn_ann_ctx->mtx);
   if (CheckContains(&dyn_ann_ctx->expect, addr, size))
     return true;
   if (CheckContains(&dyn_ann_ctx->benign, addr, size))
@@ -153,20 +154,21 @@ bool IsExpectedReport(uptr addr, uptr size) {
 }
 
 static void CollectMatchedBenignRaces(Vector<ExpectRace> *matched,
-    int *unique_count, int *hit_count, int ExpectRace::*counter) {
+    int *unique_count, int *hit_count, atomic_uintptr_t ExpectRace::*counter) {
   ExpectRace *list = &dyn_ann_ctx->benign;
   for (ExpectRace *race = list->next; race != list; race = race->next) {
     (*unique_count)++;
-    if (race->*counter == 0)
+    const uptr cnt = atomic_load_relaxed(&(race->*counter));
+    if (cnt == 0)
       continue;
-    (*hit_count) += race->*counter;
+    *hit_count += cnt;
     uptr i = 0;
     for (; i < matched->Size(); i++) {
       ExpectRace *race0 = &(*matched)[i];
       if (race->line == race0->line
           && internal_strcmp(race->file, race0->file) == 0
           && internal_strcmp(race->desc, race0->desc) == 0) {
-        race0->*counter += race->*counter;
+        atomic_fetch_add(&(race0->*counter), cnt, memory_order_relaxed);
         break;
       }
     }
@@ -191,8 +193,8 @@ void PrintMatchedBenignRaces() {
         hit_count, (int)internal_getpid());
     for (uptr i = 0; i < hit_matched.Size(); i++) {
       Printf("%d %s:%d %s\n",
-          hit_matched[i].hitcount, hit_matched[i].file,
-          hit_matched[i].line, hit_matched[i].desc);
+          atomic_load_relaxed(&hit_matched[i].hitcount),
+          hit_matched[i].file, hit_matched[i].line, hit_matched[i].desc);
     }
   }
   if (hit_matched.Size()) {
@@ -201,8 +203,8 @@ void PrintMatchedBenignRaces() {
         add_count, unique_count, (int)internal_getpid());
     for (uptr i = 0; i < add_matched.Size(); i++) {
       Printf("%d %s:%d %s\n",
-          add_matched[i].addcount, add_matched[i].file,
-          add_matched[i].line, add_matched[i].desc);
+          atomic_load_relaxed(&add_matched[i].addcount),
+          add_matched[i].file, add_matched[i].line, add_matched[i].desc);
     }
   }
 }
@@ -301,7 +303,7 @@ void INTERFACE_ATTRIBUTE AnnotateFlushExpectedRaces(char *f, int l) {
   Lock lock(&dyn_ann_ctx->mtx);
   while (dyn_ann_ctx->expect.next != &dyn_ann_ctx->expect) {
     ExpectRace *race = dyn_ann_ctx->expect.next;
-    if (race->hitcount == 0) {
+    if (atomic_load_relaxed(&race->hitcount) == 0) {
       ctx->nmissed_expected++;
       ReportMissedExpectedRace(race);
     }
diff --git a/libsanitizer/tsan/tsan_interface_atomic.cc b/libsanitizer/tsan/tsan_interface_atomic.cc
index 15fc21c4627..0ded1aa1099 100644
--- a/libsanitizer/tsan/tsan_interface_atomic.cc
+++ b/libsanitizer/tsan/tsan_interface_atomic.cc
@@ -30,15 +30,15 @@ typedef unsigned char      a8;
 typedef unsigned short     a16;  // NOLINT
 typedef unsigned int       a32;
 typedef unsigned long long a64;  // NOLINT
-#if !defined(TSAN_GO) && (defined(__SIZEOF_INT128__) \
-    || (__clang_major__ * 100 + __clang_minor__ >= 302))
+#if !defined(SANITIZER_GO) && (defined(__SIZEOF_INT128__) \
+    || (__clang_major__ * 100 + __clang_minor__ >= 302)) && !defined(__mips64)
 __extension__ typedef __int128 a128;
 # define __TSAN_HAS_INT128 1
 #else
 # define __TSAN_HAS_INT128 0
 #endif
 
-#ifndef TSAN_GO
+#if !defined(SANITIZER_GO) && __TSAN_HAS_INT128
 // Protects emulation of 128-bit atomic operations.
 static StaticSpinMutex mutex128;
 #endif
@@ -123,7 +123,8 @@ template<typename T> T func_cas(volatile T *v, T cmp, T xch) {
 // Atomic ops are executed under tsan internal mutex,
 // here we assume that the atomic variables are not accessed
 // from non-instrumented code.
-#if !defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_16) && !defined(TSAN_GO)
+#if !defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_16) && !defined(SANITIZER_GO) \
+    && __TSAN_HAS_INT128
 a128 func_xchg(volatile a128 *v, a128 op) {
   SpinMutexLock lock(&mutex128);
   a128 cmp = *v;
@@ -196,22 +197,22 @@ static int SizeLog() {
   // this leads to false negatives only in very obscure cases.
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 static atomic_uint8_t *to_atomic(const volatile a8 *a) {
-  return (atomic_uint8_t*)a;
+  return reinterpret_cast<atomic_uint8_t *>(const_cast<a8 *>(a));
 }
 
 static atomic_uint16_t *to_atomic(const volatile a16 *a) {
-  return (atomic_uint16_t*)a;
+  return reinterpret_cast<atomic_uint16_t *>(const_cast<a16 *>(a));
 }
 #endif
 
 static atomic_uint32_t *to_atomic(const volatile a32 *a) {
-  return (atomic_uint32_t*)a;
+  return reinterpret_cast<atomic_uint32_t *>(const_cast<a32 *>(a));
 }
 
 static atomic_uint64_t *to_atomic(const volatile a64 *a) {
-  return (atomic_uint64_t*)a;
+  return reinterpret_cast<atomic_uint64_t *>(const_cast<a64 *>(a));
 }
 
 static memory_order to_mo(morder mo) {
@@ -232,7 +233,7 @@ static T NoTsanAtomicLoad(const volatile T *a, morder mo) {
   return atomic_load(to_atomic(a), to_mo(mo));
 }
 
-#if __TSAN_HAS_INT128 && !defined(TSAN_GO)
+#if __TSAN_HAS_INT128 && !defined(SANITIZER_GO)
 static a128 NoTsanAtomicLoad(const volatile a128 *a, morder mo) {
   SpinMutexLock lock(&mutex128);
   return *a;
@@ -262,7 +263,7 @@ static void NoTsanAtomicStore(volatile T *a, T v, morder mo) {
   atomic_store(to_atomic(a), v, to_mo(mo));
 }
 
-#if __TSAN_HAS_INT128 && !defined(TSAN_GO)
+#if __TSAN_HAS_INT128 && !defined(SANITIZER_GO)
 static void NoTsanAtomicStore(volatile a128 *a, a128 v, morder mo) {
   SpinMutexLock lock(&mutex128);
   *a = v;
@@ -454,7 +455,7 @@ static T AtomicCAS(ThreadState *thr, uptr pc,
   return c;
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 static void NoTsanAtomicFence(morder mo) {
   __sync_synchronize();
 }
@@ -466,7 +467,7 @@ static void AtomicFence(ThreadState *thr, uptr pc, morder mo) {
 #endif
 
 // Interface functions follow.
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 
 // C/C++
 
@@ -865,7 +866,7 @@ void __tsan_atomic_signal_fence(morder mo) {
 }
 }  // extern "C"
 
-#else  // #ifndef TSAN_GO
+#else  // #ifndef SANITIZER_GO
 
 // Go
 
@@ -948,4 +949,4 @@ void __tsan_go_atomic64_compare_exchange(
   *(bool*)(a+24) = (cur == cmp);
 }
 }  // extern "C"
-#endif  // #ifndef TSAN_GO
+#endif  // #ifndef SANITIZER_GO
diff --git a/libsanitizer/tsan/tsan_interface_inl.h b/libsanitizer/tsan/tsan_interface_inl.h
index 4d6e10bcf7b..67a91aa341a 100644
--- a/libsanitizer/tsan/tsan_interface_inl.h
+++ b/libsanitizer/tsan/tsan_interface_inl.h
@@ -48,6 +48,38 @@ void __tsan_write8(void *addr) {
   MemoryWrite(cur_thread(), CALLERPC, (uptr)addr, kSizeLog8);
 }
 
+void __tsan_read1_pc(void *addr, void *pc) {
+  MemoryRead(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog1);
+}
+
+void __tsan_read2_pc(void *addr, void *pc) {
+  MemoryRead(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog2);
+}
+
+void __tsan_read4_pc(void *addr, void *pc) {
+  MemoryRead(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog4);
+}
+
+void __tsan_read8_pc(void *addr, void *pc) {
+  MemoryRead(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog8);
+}
+
+void __tsan_write1_pc(void *addr, void *pc) {
+  MemoryWrite(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog1);
+}
+
+void __tsan_write2_pc(void *addr, void *pc) {
+  MemoryWrite(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog2);
+}
+
+void __tsan_write4_pc(void *addr, void *pc) {
+  MemoryWrite(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog4);
+}
+
+void __tsan_write8_pc(void *addr, void *pc) {
+  MemoryWrite(cur_thread(), (uptr)pc, (uptr)addr, kSizeLog8);
+}
+
 void __tsan_vptr_update(void **vptr_p, void *new_val) {
   CHECK_EQ(sizeof(vptr_p), 8);
   if (*vptr_p != new_val) {
diff --git a/libsanitizer/tsan/tsan_interface_java.cc b/libsanitizer/tsan/tsan_interface_java.cc
index 6e1ec6de359..934ccc3ef9c 100644
--- a/libsanitizer/tsan/tsan_interface_java.cc
+++ b/libsanitizer/tsan/tsan_interface_java.cc
@@ -217,3 +217,33 @@ int __tsan_java_mutex_unlock_rec(jptr addr) {
 
   return MutexUnlock(thr, pc, addr, true);
 }
+
+void __tsan_java_acquire(jptr addr) {
+  SCOPED_JAVA_FUNC(__tsan_java_acquire);
+  DPrintf("#%d: java_acquire(%p)\n", thr->tid, addr);
+  CHECK_NE(jctx, 0);
+  CHECK_GE(addr, jctx->heap_begin);
+  CHECK_LT(addr, jctx->heap_begin + jctx->heap_size);
+
+  Acquire(thr, caller_pc, addr);
+}
+
+void __tsan_java_release(jptr addr) {
+  SCOPED_JAVA_FUNC(__tsan_java_release);
+  DPrintf("#%d: java_release(%p)\n", thr->tid, addr);
+  CHECK_NE(jctx, 0);
+  CHECK_GE(addr, jctx->heap_begin);
+  CHECK_LT(addr, jctx->heap_begin + jctx->heap_size);
+
+  Release(thr, caller_pc, addr);
+}
+
+void __tsan_java_release_store(jptr addr) {
+  SCOPED_JAVA_FUNC(__tsan_java_release);
+  DPrintf("#%d: java_release_store(%p)\n", thr->tid, addr);
+  CHECK_NE(jctx, 0);
+  CHECK_GE(addr, jctx->heap_begin);
+  CHECK_LT(addr, jctx->heap_begin + jctx->heap_size);
+
+  ReleaseStore(thr, caller_pc, addr);
+}
diff --git a/libsanitizer/tsan/tsan_interface_java.h b/libsanitizer/tsan/tsan_interface_java.h
index 9af1f3fe21a..04e52031dd3 100644
--- a/libsanitizer/tsan/tsan_interface_java.h
+++ b/libsanitizer/tsan/tsan_interface_java.h
@@ -77,6 +77,14 @@ void __tsan_java_mutex_lock_rec(jptr addr, int rec) INTERFACE_ATTRIBUTE;
 // the same recursion level.
 int __tsan_java_mutex_unlock_rec(jptr addr) INTERFACE_ATTRIBUTE;
 
+// Raw acquire/release primitives.
+// Can be used to establish happens-before edges on volatile/final fields,
+// in atomic operations, etc. release_store is the same as release, but it
+// breaks release sequence on addr (see C++ standard 1.10/7 for details).
+void __tsan_java_acquire(jptr addr) INTERFACE_ATTRIBUTE;
+void __tsan_java_release(jptr addr) INTERFACE_ATTRIBUTE;
+void __tsan_java_release_store(jptr addr) INTERFACE_ATTRIBUTE;
+
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/libsanitizer/tsan/tsan_md5.cc b/libsanitizer/tsan/tsan_md5.cc
index 883239c2e71..f299dfc59b6 100644
--- a/libsanitizer/tsan/tsan_md5.cc
+++ b/libsanitizer/tsan/tsan_md5.cc
@@ -23,7 +23,7 @@ namespace __tsan {
   (a) += (b);
 
 #define SET(n) \
-  (*(MD5_u32plus *)&ptr[(n) * 4])
+  (*(const MD5_u32plus *)&ptr[(n) * 4])
 #define GET(n) \
   SET(n)
 
@@ -37,13 +37,11 @@ typedef struct {
   MD5_u32plus block[16];
 } MD5_CTX;
 
-static void *body(MD5_CTX *ctx, void *data, ulong_t size) {
-  unsigned char *ptr;
+static const void *body(MD5_CTX *ctx, const void *data, ulong_t size) {
+  const unsigned char *ptr = (const unsigned char *)data;
   MD5_u32plus a, b, c, d;
   MD5_u32plus saved_a, saved_b, saved_c, saved_d;
 
-  ptr = (unsigned char*)data;
-
   a = ctx->a;
   b = ctx->b;
   c = ctx->c;
@@ -149,7 +147,7 @@ void MD5_Init(MD5_CTX *ctx) {
   ctx->hi = 0;
 }
 
-void MD5_Update(MD5_CTX *ctx, void *data, ulong_t size) {
+void MD5_Update(MD5_CTX *ctx, const void *data, ulong_t size) {
   MD5_u32plus saved_lo;
   ulong_t used, free;
 
@@ -169,7 +167,7 @@ void MD5_Update(MD5_CTX *ctx, void *data, ulong_t size) {
     }
 
     internal_memcpy(&ctx->buffer[used], data, free);
-    data = (unsigned char *)data + free;
+    data = (const unsigned char *)data + free;
     size -= free;
     body(ctx, ctx->buffer, 64);
   }
@@ -236,7 +234,7 @@ MD5Hash md5_hash(const void *data, uptr size) {
   MD5Hash res;
   MD5_CTX ctx;
   MD5_Init(&ctx);
-  MD5_Update(&ctx, (void*)data, size);
+  MD5_Update(&ctx, data, size);
   MD5_Final((unsigned char*)&res.hash[0], &ctx);
   return res;
 }
diff --git a/libsanitizer/tsan/tsan_mman.cc b/libsanitizer/tsan/tsan_mman.cc
index d89610a2cc3..c8a5a6a264a 100644
--- a/libsanitizer/tsan/tsan_mman.cc
+++ b/libsanitizer/tsan/tsan_mman.cc
@@ -34,6 +34,23 @@ struct MapUnmapCallback {
     // We are about to unmap a chunk of user memory.
     // Mark the corresponding shadow memory as not needed.
     DontNeedShadowFor(p, size);
+    // Mark the corresponding meta shadow memory as not needed.
+    // Note the block does not contain any meta info at this point
+    // (this happens after free).
+    const uptr kMetaRatio = kMetaShadowCell / kMetaShadowSize;
+    const uptr kPageSize = GetPageSizeCached() * kMetaRatio;
+    // Block came from LargeMmapAllocator, so must be large.
+    // We rely on this in the calculations below.
+    CHECK_GE(size, 2 * kPageSize);
+    uptr diff = RoundUp(p, kPageSize) - p;
+    if (diff != 0) {
+      p += diff;
+      size -= diff;
+    }
+    diff = p + size - RoundDown(p + size, kPageSize);
+    if (diff != 0)
+      size -= diff;
+    FlushUnneededShadowMemory((uptr)MemToMeta(p), size / kMetaRatio);
   }
 };
 
@@ -43,7 +60,7 @@ Allocator *allocator() {
 }
 
 void InitializeAllocator() {
-  allocator()->Init();
+  allocator()->Init(common_flags()->allocator_may_return_null);
 }
 
 void AllocatorThreadStart(ThreadState *thr) {
@@ -61,22 +78,22 @@ void AllocatorPrintStats() {
 }
 
 static void SignalUnsafeCall(ThreadState *thr, uptr pc) {
-  if (atomic_load(&thr->in_signal_handler, memory_order_relaxed) == 0 ||
+  if (atomic_load_relaxed(&thr->in_signal_handler) == 0 ||
       !flags()->report_signal_unsafe)
     return;
   VarSizeStackTrace stack;
   ObtainCurrentStack(thr, pc, &stack);
+  if (IsFiredSuppression(ctx, ReportTypeSignalUnsafe, stack))
+    return;
   ThreadRegistryLock l(ctx->thread_registry);
   ScopedReport rep(ReportTypeSignalUnsafe);
-  if (!IsFiredSuppression(ctx, rep, stack)) {
-    rep.AddStack(stack, true);
-    OutputReport(thr, rep);
-  }
+  rep.AddStack(stack, true);
+  OutputReport(thr, rep);
 }
 
 void *user_alloc(ThreadState *thr, uptr pc, uptr sz, uptr align, bool signal) {
   if ((sz >= (1ull << 40)) || (align >= (1ull << 40)))
-    return AllocatorReturnNull();
+    return allocator()->ReturnNullOrDie();
   void *p = allocator()->Allocate(&thr->alloc_cache, sz, align);
   if (p == 0)
     return 0;
@@ -87,6 +104,15 @@ 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 (CallocShouldReturnNullDueToOverflow(size, n))
+    return allocator()->ReturnNullOrDie();
+  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) {
   if (ctx && ctx->initialized)
     OnUserFree(thr, pc, (uptr)p, true);
diff --git a/libsanitizer/tsan/tsan_mman.h b/libsanitizer/tsan/tsan_mman.h
index fe020af9d2d..da7cf211940 100644
--- a/libsanitizer/tsan/tsan_mman.h
+++ b/libsanitizer/tsan/tsan_mman.h
@@ -25,6 +25,7 @@ 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);
 // Does not accept NULL.
 void user_free(ThreadState *thr, uptr pc, void *p, bool signal = true);
 void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz);
diff --git a/libsanitizer/tsan/tsan_mutex.cc b/libsanitizer/tsan/tsan_mutex.cc
index 2e49b9d2de9..3d3f6cc82b9 100644
--- a/libsanitizer/tsan/tsan_mutex.cc
+++ b/libsanitizer/tsan/tsan_mutex.cc
@@ -23,7 +23,7 @@ namespace __tsan {
 // then Report mutex can be locked while under Threads mutex.
 // The leaf mutexes can be locked under any other mutexes.
 // Recursive locking is not supported.
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
 const MutexType MutexTypeLeaf = (MutexType)-1;
 static MutexType CanLockTab[MutexTypeCount][MutexTypeCount] = {
   /*0  MutexTypeInvalid*/     {},
@@ -39,13 +39,15 @@ static MutexType CanLockTab[MutexTypeCount][MutexTypeCount] = {
   /*9  MutexTypeMBlock*/      {MutexTypeSyncVar},
   /*10 MutexTypeJavaMBlock*/  {MutexTypeSyncVar},
   /*11 MutexTypeDDetector*/   {},
+  /*12 MutexTypeFired*/       {MutexTypeLeaf},
+  /*13 MutexTypeRacy*/        {MutexTypeLeaf},
 };
 
 static bool CanLockAdj[MutexTypeCount][MutexTypeCount];
 #endif
 
 void InitializeMutex() {
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
   // Build the "can lock" adjacency matrix.
   // If [i][j]==true, then one can lock mutex j while under mutex i.
   const int N = MutexTypeCount;
@@ -126,7 +128,7 @@ InternalDeadlockDetector::InternalDeadlockDetector() {
   // Rely on zero initialization because some mutexes can be locked before ctor.
 }
 
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
 void InternalDeadlockDetector::Lock(MutexType t) {
   // Printf("LOCK %d @%zu\n", t, seq_ + 1);
   CHECK_GT(t, MutexTypeInvalid);
@@ -168,7 +170,7 @@ void InternalDeadlockDetector::CheckNoLocks() {
 #endif
 
 void CheckNoLocks(ThreadState *thr) {
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
   thr->internal_deadlock_detector.CheckNoLocks();
 #endif
 }
@@ -206,7 +208,7 @@ class Backoff {
 Mutex::Mutex(MutexType type, StatType stat_type) {
   CHECK_GT(type, MutexTypeInvalid);
   CHECK_LT(type, MutexTypeCount);
-#if TSAN_DEBUG
+#if SANITIZER_DEBUG
   type_ = type;
 #endif
 #if TSAN_COLLECT_STATS
@@ -220,7 +222,7 @@ Mutex::~Mutex() {
 }
 
 void Mutex::Lock() {
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
   cur_thread()->internal_deadlock_detector.Lock(type_);
 #endif
   uptr cmp = kUnlocked;
@@ -232,7 +234,7 @@ void Mutex::Lock() {
       cmp = kUnlocked;
       if (atomic_compare_exchange_weak(&state_, &cmp, kWriteLock,
                                        memory_order_acquire)) {
-#if TSAN_COLLECT_STATS && !TSAN_GO
+#if TSAN_COLLECT_STATS && !SANITIZER_GO
         StatInc(cur_thread(), stat_type_, backoff.Contention());
 #endif
         return;
@@ -245,13 +247,13 @@ void Mutex::Unlock() {
   uptr prev = atomic_fetch_sub(&state_, kWriteLock, memory_order_release);
   (void)prev;
   DCHECK_NE(prev & kWriteLock, 0);
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
   cur_thread()->internal_deadlock_detector.Unlock(type_);
 #endif
 }
 
 void Mutex::ReadLock() {
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
   cur_thread()->internal_deadlock_detector.Lock(type_);
 #endif
   uptr prev = atomic_fetch_add(&state_, kReadLock, memory_order_acquire);
@@ -260,7 +262,7 @@ void Mutex::ReadLock() {
   for (Backoff backoff; backoff.Do();) {
     prev = atomic_load(&state_, memory_order_acquire);
     if ((prev & kWriteLock) == 0) {
-#if TSAN_COLLECT_STATS && !TSAN_GO
+#if TSAN_COLLECT_STATS && !SANITIZER_GO
       StatInc(cur_thread(), stat_type_, backoff.Contention());
 #endif
       return;
@@ -273,7 +275,7 @@ void Mutex::ReadUnlock() {
   (void)prev;
   DCHECK_EQ(prev & kWriteLock, 0);
   DCHECK_GT(prev & ~kWriteLock, 0);
-#if TSAN_DEBUG && !TSAN_GO
+#if SANITIZER_DEBUG && !SANITIZER_GO
   cur_thread()->internal_deadlock_detector.Unlock(type_);
 #endif
 }
diff --git a/libsanitizer/tsan/tsan_mutex.h b/libsanitizer/tsan/tsan_mutex.h
index c27bc2b4459..9960eee9bc8 100644
--- a/libsanitizer/tsan/tsan_mutex.h
+++ b/libsanitizer/tsan/tsan_mutex.h
@@ -30,6 +30,8 @@ enum MutexType {
   MutexTypeMBlock,
   MutexTypeJavaMBlock,
   MutexTypeDDetector,
+  MutexTypeFired,
+  MutexTypeRacy,
 
   // This must be the last.
   MutexTypeCount
@@ -50,7 +52,7 @@ class Mutex {
 
  private:
   atomic_uintptr_t state_;
-#if TSAN_DEBUG
+#if SANITIZER_DEBUG
   MutexType type_;
 #endif
 #if TSAN_COLLECT_STATS
diff --git a/libsanitizer/tsan/tsan_mutexset.h b/libsanitizer/tsan/tsan_mutexset.h
index 7fdc194109f..d5406ae9f12 100644
--- a/libsanitizer/tsan/tsan_mutexset.h
+++ b/libsanitizer/tsan/tsan_mutexset.h
@@ -41,7 +41,7 @@ class MutexSet {
   }
 
  private:
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   uptr size_;
   Desc descs_[kMaxSize];
 #endif
@@ -53,7 +53,7 @@ class MutexSet {
 // Go does not have mutexes, so do not spend memory and time.
 // (Go sync.Mutex is actually a semaphore -- can be unlocked
 // in different goroutine).
-#ifdef TSAN_GO
+#ifdef SANITIZER_GO
 MutexSet::MutexSet() {}
 void MutexSet::Add(u64 id, bool write, u64 epoch) {}
 void MutexSet::Del(u64 id, bool write) {}
diff --git a/libsanitizer/tsan/tsan_new_delete.cc b/libsanitizer/tsan/tsan_new_delete.cc
new file mode 100644
index 00000000000..ad4ed449751
--- /dev/null
+++ b/libsanitizer/tsan/tsan_new_delete.cc
@@ -0,0 +1,86 @@
+//===-- tsan_new_delete.cc ----------------------------------------------===//
+//
+// 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.
+//
+// Interceptors for operators new and delete.
+//===----------------------------------------------------------------------===//
+#include "sanitizer_common/sanitizer_internal_defs.h"
+#include "tsan_interceptors.h"
+
+using namespace __tsan;  // NOLINT
+
+namespace std {
+struct nothrow_t {};
+}  // namespace std
+
+#define OPERATOR_NEW_BODY(mangled_name) \
+  if (cur_thread()->in_symbolizer) \
+    return __libc_malloc(size); \
+  void *p = 0; \
+  {  \
+    SCOPED_INTERCEPTOR_RAW(mangled_name, size); \
+    p = user_alloc(thr, pc, size); \
+  }  \
+  invoke_malloc_hook(p, size);  \
+  return p;
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void *operator new(__sanitizer::uptr size);
+void *operator new(__sanitizer::uptr size) {
+  OPERATOR_NEW_BODY(_Znwm);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void *operator new[](__sanitizer::uptr size);
+void *operator new[](__sanitizer::uptr size) {
+  OPERATOR_NEW_BODY(_Znam);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void *operator new(__sanitizer::uptr size, std::nothrow_t const&);
+void *operator new(__sanitizer::uptr size, std::nothrow_t const&) {
+  OPERATOR_NEW_BODY(_ZnwmRKSt9nothrow_t);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void *operator new[](__sanitizer::uptr size, std::nothrow_t const&);
+void *operator new[](__sanitizer::uptr size, std::nothrow_t const&) {
+  OPERATOR_NEW_BODY(_ZnamRKSt9nothrow_t);
+}
+
+#define OPERATOR_DELETE_BODY(mangled_name) \
+  if (ptr == 0) return;  \
+  if (cur_thread()->in_symbolizer) \
+    return __libc_free(ptr); \
+  invoke_free_hook(ptr);  \
+  SCOPED_INTERCEPTOR_RAW(mangled_name, ptr);  \
+  user_free(thr, pc, ptr);
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void operator delete(void *ptr) NOEXCEPT;
+void operator delete(void *ptr) NOEXCEPT {
+  OPERATOR_DELETE_BODY(_ZdlPv);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void operator delete[](void *ptr) NOEXCEPT;
+void operator delete[](void *ptr) NOEXCEPT {
+  OPERATOR_DELETE_BODY(_ZdaPv);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void operator delete(void *ptr, std::nothrow_t const&);
+void operator delete(void *ptr, std::nothrow_t const&) {
+  OPERATOR_DELETE_BODY(_ZdlPvRKSt9nothrow_t);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void operator delete[](void *ptr, std::nothrow_t const&);
+void operator delete[](void *ptr, std::nothrow_t const&) {
+  OPERATOR_DELETE_BODY(_ZdaPvRKSt9nothrow_t);
+}
diff --git a/libsanitizer/tsan/tsan_platform.h b/libsanitizer/tsan/tsan_platform.h
index 35837d2132a..07d11e572ff 100644
--- a/libsanitizer/tsan/tsan_platform.h
+++ b/libsanitizer/tsan/tsan_platform.h
@@ -22,10 +22,11 @@
 
 namespace __tsan {
 
-#if !defined(TSAN_GO)
+#if !defined(SANITIZER_GO)
 
+#if defined(__x86_64__)
 /*
-C/C++ on linux and freebsd
+C/C++ on linux/x86_64 and freebsd/x86_64
 0000 0000 1000 - 0100 0000 0000: main binary and/or MAP_32BIT mappings
 0100 0000 0000 - 0200 0000 0000: -
 0200 0000 0000 - 1000 0000 0000: shadow
@@ -38,7 +39,6 @@ C/C++ on linux and freebsd
 7e00 0000 0000 - 7e80 0000 0000: -
 7e80 0000 0000 - 8000 0000 0000: modules and main thread stack
 */
-
 const uptr kMetaShadowBeg = 0x300000000000ull;
 const uptr kMetaShadowEnd = 0x400000000000ull;
 const uptr kTraceMemBeg   = 0x600000000000ull;
@@ -53,6 +53,97 @@ const uptr kHiAppMemBeg   = 0x7e8000000000ull;
 const uptr kHiAppMemEnd   = 0x800000000000ull;
 const uptr kAppMemMsk     = 0x7c0000000000ull;
 const uptr kAppMemXor     = 0x020000000000ull;
+const uptr kVdsoBeg       = 0xf000000000000000ull;
+#elif defined(__mips64)
+/*
+C/C++ on linux/mips64
+0100 0000 00 - 0200 0000 00: main binary
+0200 0000 00 - 1400 0000 00: -
+1400 0000 00 - 2400 0000 00: shadow
+2400 0000 00 - 3000 0000 00: -
+3000 0000 00 - 4000 0000 00: metainfo (memory blocks and sync objects)
+4000 0000 00 - 6000 0000 00: -
+6000 0000 00 - 6200 0000 00: traces
+6200 0000 00 - fe00 0000 00: -
+fe00 0000 00 - ff00 0000 00: heap
+ff00 0000 00 - ff80 0000 00: -
+ff80 0000 00 - ffff ffff ff: modules and main thread stack
+*/
+const uptr kMetaShadowBeg = 0x3000000000ull;
+const uptr kMetaShadowEnd = 0x4000000000ull;
+const uptr kTraceMemBeg   = 0x6000000000ull;
+const uptr kTraceMemEnd   = 0x6200000000ull;
+const uptr kShadowBeg     = 0x1400000000ull;
+const uptr kShadowEnd     = 0x2400000000ull;
+const uptr kHeapMemBeg    = 0xfe00000000ull;
+const uptr kHeapMemEnd    = 0xff00000000ull;
+const uptr kLoAppMemBeg   = 0x0100000000ull;
+const uptr kLoAppMemEnd   = 0x0200000000ull;
+const uptr kHiAppMemBeg   = 0xff80000000ull;
+const uptr kHiAppMemEnd   = 0xffffffffffull;
+const uptr kAppMemMsk     = 0xfc00000000ull;
+const uptr kAppMemXor     = 0x0400000000ull;
+const uptr kVdsoBeg       = 0xfffff00000ull;
+#elif defined(__aarch64__)
+# if SANITIZER_AARCH64_VMA == 39
+/*
+C/C++ on linux/aarch64 (39-bit VMA)
+0000 4000 00 - 0200 0000 00: main binary
+2000 0000 00 - 4000 0000 00: shadow memory
+4000 0000 00 - 5000 0000 00: metainfo
+5000 0000 00 - 6000 0000 00: -
+6000 0000 00 - 6200 0000 00: traces
+6200 0000 00 - 7d00 0000 00: -
+7d00 0000 00 - 7e00 0000 00: heap
+7e00 0000 00 - 7fff ffff ff: modules and main thread stack
+*/
+const uptr kLoAppMemBeg   = 0x0000400000ull;
+const uptr kLoAppMemEnd   = 0x0200000000ull;
+const uptr kShadowBeg     = 0x2000000000ull;
+const uptr kShadowEnd     = 0x4000000000ull;
+const uptr kMetaShadowBeg = 0x4000000000ull;
+const uptr kMetaShadowEnd = 0x5000000000ull;
+const uptr kTraceMemBeg   = 0x6000000000ull;
+const uptr kTraceMemEnd   = 0x6200000000ull;
+const uptr kHeapMemBeg    = 0x7d00000000ull;
+const uptr kHeapMemEnd    = 0x7e00000000ull;
+const uptr kHiAppMemBeg   = 0x7e00000000ull;
+const uptr kHiAppMemEnd   = 0x7fffffffffull;
+const uptr kAppMemMsk     = 0x7800000000ull;
+const uptr kAppMemXor     = 0x0800000000ull;
+const uptr kVdsoBeg       = 0x7f00000000ull;
+# elif SANITIZER_AARCH64_VMA == 42
+/*
+C/C++ on linux/aarch64 (42-bit VMA)
+00000 4000 00 - 01000 0000 00: main binary
+01000 0000 00 - 10000 0000 00: -
+10000 0000 00 - 20000 0000 00: shadow memory
+20000 0000 00 - 26000 0000 00: -
+26000 0000 00 - 28000 0000 00: metainfo
+28000 0000 00 - 36200 0000 00: -
+36200 0000 00 - 36240 0000 00: traces
+36240 0000 00 - 3e000 0000 00: -
+3e000 0000 00 - 3f000 0000 00: heap
+3c000 0000 00 - 3ff00 0000 00: -
+3ff00 0000 00 - 3ffff f000 00: modules and main thread stack
+*/
+const uptr kLoAppMemBeg   = 0x00000400000ull;
+const uptr kLoAppMemEnd   = 0x01000000000ull;
+const uptr kShadowBeg     = 0x10000000000ull;
+const uptr kShadowEnd     = 0x20000000000ull;
+const uptr kMetaShadowBeg = 0x26000000000ull;
+const uptr kMetaShadowEnd = 0x28000000000ull;
+const uptr kTraceMemBeg   = 0x36200000000ull;
+const uptr kTraceMemEnd   = 0x36400000000ull;
+const uptr kHeapMemBeg    = 0x3e000000000ull;
+const uptr kHeapMemEnd    = 0x3f000000000ull;
+const uptr kHiAppMemBeg   = 0x3ff00000000ull;
+const uptr kHiAppMemEnd   = 0x3fffff00000ull;
+const uptr kAppMemMsk     = 0x3c000000000ull;
+const uptr kAppMemXor     = 0x04000000000ull;
+const uptr kVdsoBeg       = 0x37f00000000ull;
+# endif
+#endif
 
 ALWAYS_INLINE
 bool IsAppMem(uptr mem) {
@@ -100,7 +191,7 @@ static USED uptr UserRegions[] = {
   kHeapMemBeg,  kHeapMemEnd,
 };
 
-#elif defined(TSAN_GO) && !SANITIZER_WINDOWS
+#elif defined(SANITIZER_GO) && !SANITIZER_WINDOWS
 
 /* Go on linux, darwin and freebsd
 0000 0000 1000 - 0000 1000 0000: executable
@@ -162,15 +253,15 @@ static USED uptr UserRegions[] = {
   kAppMemBeg, kAppMemEnd,
 };
 
-#elif defined(TSAN_GO) && SANITIZER_WINDOWS
+#elif defined(SANITIZER_GO) && SANITIZER_WINDOWS
 
 /* Go on windows
 0000 0000 1000 - 0000 1000 0000: executable
 0000 1000 0000 - 00f8 0000 0000: -
 00c0 0000 0000 - 00e0 0000 0000: heap
 00e0 0000 0000 - 0100 0000 0000: -
-0100 0000 0000 - 0380 0000 0000: shadow
-0380 0000 0000 - 0560 0000 0000: -
+0100 0000 0000 - 0500 0000 0000: shadow
+0500 0000 0000 - 0560 0000 0000: -
 0560 0000 0000 - 0760 0000 0000: traces
 0760 0000 0000 - 07d0 0000 0000: metainfo (memory blocks and sync objects)
 07d0 0000 0000 - 8000 0000 0000: -
@@ -181,7 +272,7 @@ const uptr kMetaShadowEnd = 0x07d000000000ull;
 const uptr kTraceMemBeg   = 0x056000000000ull;
 const uptr kTraceMemEnd   = 0x076000000000ull;
 const uptr kShadowBeg     = 0x010000000000ull;
-const uptr kShadowEnd     = 0x038000000000ull;
+const uptr kShadowEnd     = 0x050000000000ull;
 const uptr kAppMemBeg     = 0x000000001000ull;
 const uptr kAppMemEnd     = 0x00e000000000ull;
 
@@ -203,21 +294,21 @@ bool IsMetaMem(uptr mem) {
 ALWAYS_INLINE
 uptr MemToShadow(uptr x) {
   DCHECK(IsAppMem(x));
-  return ((x & ~(kShadowCell - 1)) * kShadowCnt) | kShadowBeg;
+  return ((x & ~(kShadowCell - 1)) * kShadowCnt) + kShadowBeg;
 }
 
 ALWAYS_INLINE
 u32 *MemToMeta(uptr x) {
   DCHECK(IsAppMem(x));
   return (u32*)(((x & ~(kMetaShadowCell - 1)) / \
-      kMetaShadowCell * kMetaShadowSize) | kMetaShadowEnd);
+      kMetaShadowCell * kMetaShadowSize) | kMetaShadowBeg);
 }
 
 ALWAYS_INLINE
 uptr ShadowToMem(uptr s) {
   CHECK(IsShadowMem(s));
   // FIXME(dvyukov): this is most likely wrong as the mapping is not bijection.
-  return (x & ~kShadowBeg) / kShadowCnt;
+  return (s - kShadowBeg) / kShadowCnt;
 }
 
 static USED uptr UserRegions[] = {
@@ -249,10 +340,6 @@ uptr ALWAYS_INLINE GetThreadTraceHeader(int tid) {
 void InitializePlatform();
 void FlushShadowMemory();
 void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive);
-uptr GetRSS();
-
-void *internal_start_thread(void(*func)(void*), void *arg);
-void internal_join_thread(void *th);
 
 // Says whether the addr relates to a global var.
 // Guesses with high probability, may yield both false positives and negatives.
diff --git a/libsanitizer/tsan/tsan_platform_linux.cc b/libsanitizer/tsan/tsan_platform_linux.cc
index 32591316ea2..f409a8b221c 100644
--- a/libsanitizer/tsan/tsan_platform_linux.cc
+++ b/libsanitizer/tsan/tsan_platform_linux.cc
@@ -16,6 +16,7 @@
 
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_libc.h"
+#include "sanitizer_common/sanitizer_posix.h"
 #include "sanitizer_common/sanitizer_procmaps.h"
 #include "sanitizer_common/sanitizer_stoptheworld.h"
 #include "sanitizer_common/sanitizer_stackdepot.h"
@@ -64,8 +65,6 @@ namespace __tsan {
 static uptr g_data_start;
 static uptr g_data_end;
 
-const uptr kPageSize = 4096;
-
 enum {
   MemTotal  = 0,
   MemShadow = 1,
@@ -85,7 +84,7 @@ void FillProfileCallback(uptr p, uptr rss, bool file,
     mem[MemShadow] += rss;
   else if (p >= kMetaShadowBeg && p < kMetaShadowEnd)
     mem[MemMeta] += rss;
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   else if (p >= kHeapMemBeg && p < kHeapMemEnd)
     mem[MemHeap] += rss;
   else if (p >= kLoAppMemBeg && p < kLoAppMemEnd)
@@ -116,33 +115,6 @@ void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
       nlive, nthread);
 }
 
-uptr GetRSS() {
-  uptr fd = OpenFile("/proc/self/statm", false);
-  if ((sptr)fd < 0)
-    return 0;
-  char buf[64];
-  uptr len = internal_read(fd, buf, sizeof(buf) - 1);
-  internal_close(fd);
-  if ((sptr)len <= 0)
-    return 0;
-  buf[len] = 0;
-  // The format of the file is:
-  // 1084 89 69 11 0 79 0
-  // We need the second number which is RSS in 4K units.
-  char *pos = buf;
-  // Skip the first number.
-  while (*pos >= '0' && *pos <= '9')
-    pos++;
-  // Skip whitespaces.
-  while (!(*pos >= '0' && *pos <= '9') && *pos != 0)
-    pos++;
-  // Read the number.
-  uptr rss = 0;
-  while (*pos >= '0' && *pos <= '9')
-    rss = rss * 10 + *pos++ - '0';
-  return rss * 4096;
-}
-
 #if SANITIZER_LINUX
 void FlushShadowMemoryCallback(
     const SuspendedThreadsList &suspended_threads_list,
@@ -157,12 +129,12 @@ void FlushShadowMemory() {
 #endif
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 static void ProtectRange(uptr beg, uptr end) {
   CHECK_LE(beg, end);
   if (beg == end)
     return;
-  if (beg != (uptr)Mprotect(beg, end - beg)) {
+  if (beg != (uptr)MmapNoAccess(beg, end - beg)) {
     Printf("FATAL: ThreadSanitizer can not protect [%zx,%zx]\n", beg, end);
     Printf("FATAL: Make sure you are not using unlimited stack\n");
     Die();
@@ -198,7 +170,7 @@ static void MapRodata() {
     *p = kShadowRodata;
   internal_write(fd, marker.data(), marker.size());
   // Map the file into memory.
-  uptr page = internal_mmap(0, kPageSize, PROT_READ | PROT_WRITE,
+  uptr page = internal_mmap(0, GetPageSizeCached(), PROT_READ | PROT_WRITE,
                             MAP_PRIVATE | MAP_ANONYMOUS, fd, 0);
   if (internal_iserror(page)) {
     internal_close(fd);
@@ -228,8 +200,8 @@ static void MapRodata() {
 
 void InitializeShadowMemory() {
   // Map memory shadow.
-  uptr shadow = (uptr)MmapFixedNoReserve(kShadowBeg,
-    kShadowEnd - kShadowBeg);
+  uptr shadow =
+      (uptr)MmapFixedNoReserve(kShadowBeg, kShadowEnd - kShadowBeg, "shadow");
   if (shadow != kShadowBeg) {
     Printf("FATAL: ThreadSanitizer can not mmap the shadow memory\n");
     Printf("FATAL: Make sure to compile with -fPIE and "
@@ -240,23 +212,41 @@ void InitializeShadowMemory() {
   // Frequently a thread uses only a small part of stack and similarly
   // a program uses a small part of large mmap. On some programs
   // we see 20% memory usage reduction without huge pages for this range.
-#ifdef MADV_NOHUGEPAGE
-  madvise((void*)MemToShadow(0x7f0000000000ULL),
-      0x10000000000ULL * kShadowMultiplier, MADV_NOHUGEPAGE);
+  // FIXME: don't use constants here.
+#if defined(__x86_64__)
+  const uptr kMadviseRangeBeg  = 0x7f0000000000ull;
+  const uptr kMadviseRangeSize = 0x010000000000ull;
+#elif defined(__mips64)
+  const uptr kMadviseRangeBeg  = 0xff00000000ull;
+  const uptr kMadviseRangeSize = 0x0100000000ull;
+#elif defined(__aarch64__)
+  const uptr kMadviseRangeBeg  = 0x7e00000000ull;
+  const uptr kMadviseRangeSize = 0x0100000000ull;
 #endif
+  NoHugePagesInRegion(MemToShadow(kMadviseRangeBeg),
+                      kMadviseRangeSize * kShadowMultiplier);
+  // Meta shadow is compressing and we don't flush it,
+  // so it makes sense to mark it as NOHUGEPAGE to not over-allocate memory.
+  // On one program it reduces memory consumption from 5GB to 2.5GB.
+  NoHugePagesInRegion(kMetaShadowBeg, kMetaShadowEnd - kMetaShadowBeg);
+  if (common_flags()->use_madv_dontdump)
+    DontDumpShadowMemory(kShadowBeg, kShadowEnd - kShadowBeg);
   DPrintf("memory shadow: %zx-%zx (%zuGB)\n",
       kShadowBeg, kShadowEnd,
       (kShadowEnd - kShadowBeg) >> 30);
 
   // Map meta shadow.
   uptr meta_size = kMetaShadowEnd - kMetaShadowBeg;
-  uptr meta = (uptr)MmapFixedNoReserve(kMetaShadowBeg, meta_size);
+  uptr meta =
+      (uptr)MmapFixedNoReserve(kMetaShadowBeg, meta_size, "meta shadow");
   if (meta != kMetaShadowBeg) {
     Printf("FATAL: ThreadSanitizer can not mmap the shadow memory\n");
     Printf("FATAL: Make sure to compile with -fPIE and "
                "to link with -pie (%p, %p).\n", meta, kMetaShadowBeg);
     Die();
   }
+  if (common_flags()->use_madv_dontdump)
+    DontDumpShadowMemory(meta, meta_size);
   DPrintf("meta shadow: %zx-%zx (%zuGB)\n",
       meta, meta + meta_size, meta_size >> 30);
 
@@ -311,9 +301,9 @@ static void CheckAndProtect() {
     if (IsAppMem(p))
       continue;
     if (p >= kHeapMemEnd &&
-        p < kHeapMemEnd + PrimaryAllocator::AdditionalSize())
+        p < HeapEnd())
       continue;
-    if (p >= 0xf000000000000000ull)  // vdso
+    if (p >= kVdsoBeg)  // vdso
       break;
     Printf("FATAL: ThreadSanitizer: unexpected memory mapping %p-%p\n", p, end);
     Die();
@@ -322,10 +312,13 @@ static void CheckAndProtect() {
   ProtectRange(kLoAppMemEnd, kShadowBeg);
   ProtectRange(kShadowEnd, kMetaShadowBeg);
   ProtectRange(kMetaShadowEnd, kTraceMemBeg);
+  // Memory for traces is mapped lazily in MapThreadTrace.
+  // Protect the whole range for now, so that user does not map something here.
+  ProtectRange(kTraceMemBeg, kTraceMemEnd);
   ProtectRange(kTraceMemEnd, kHeapMemBeg);
-  ProtectRange(kHeapMemEnd + PrimaryAllocator::AdditionalSize(), kHiAppMemBeg);
+  ProtectRange(HeapEnd(), kHiAppMemBeg);
 }
-#endif  // #ifndef TSAN_GO
+#endif  // #ifndef SANITIZER_GO
 
 void InitializePlatform() {
   DisableCoreDumperIfNecessary();
@@ -359,7 +352,7 @@ void InitializePlatform() {
       ReExec();
   }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   CheckAndProtect();
   InitTlsSize();
   InitDataSeg();
@@ -370,7 +363,7 @@ bool IsGlobalVar(uptr addr) {
   return g_data_start && addr >= g_data_start && addr < g_data_end;
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 // Extract file descriptors passed to glibc internal __res_iclose function.
 // This is required to properly "close" the fds, because we do not see internal
 // closes within glibc. The code is a pure hack.
@@ -408,6 +401,8 @@ int ExtractRecvmsgFDs(void *msgp, int *fds, int nfd) {
   return res;
 }
 
+// Note: this function runs with async signals enabled,
+// so it must not touch any tsan state.
 int call_pthread_cancel_with_cleanup(int(*fn)(void *c, void *m,
     void *abstime), void *c, void *m, void *abstime,
     void(*cleanup)(void *arg), void *arg) {
diff --git a/libsanitizer/tsan/tsan_platform_mac.cc b/libsanitizer/tsan/tsan_platform_mac.cc
index 2091318a674..a2ae26f5341 100644
--- a/libsanitizer/tsan/tsan_platform_mac.cc
+++ b/libsanitizer/tsan/tsan_platform_mac.cc
@@ -48,11 +48,7 @@ void FlushShadowMemory() {
 void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
 }
 
-uptr GetRSS() {
-  return 0;
-}
-
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 void InitializeShadowMemory() {
   uptr shadow = (uptr)MmapFixedNoReserve(kShadowBeg,
     kShadowEnd - kShadowBeg);
@@ -62,6 +58,8 @@ void InitializeShadowMemory() {
            "to link with -pie.\n");
     Die();
   }
+  if (common_flags()->use_madv_dontdump)
+    DontDumpShadowMemory(kShadowBeg, kShadowEnd - kShadowBeg);
   DPrintf("kShadow %zx-%zx (%zuGB)\n",
       kShadowBeg, kShadowEnd,
       (kShadowEnd - kShadowBeg) >> 30);
@@ -75,7 +73,9 @@ void InitializePlatform() {
   DisableCoreDumperIfNecessary();
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
+// Note: this function runs with async signals enabled,
+// so it must not touch any tsan state.
 int call_pthread_cancel_with_cleanup(int(*fn)(void *c, void *m,
     void *abstime), void *c, void *m, void *abstime,
     void(*cleanup)(void *arg), void *arg) {
diff --git a/libsanitizer/tsan/tsan_platform_windows.cc b/libsanitizer/tsan/tsan_platform_windows.cc
index 3f81c54e3eb..ccea22ed66e 100644
--- a/libsanitizer/tsan/tsan_platform_windows.cc
+++ b/libsanitizer/tsan/tsan_platform_windows.cc
@@ -29,10 +29,6 @@ void FlushShadowMemory() {
 void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
 }
 
-uptr GetRSS() {
-  return 0;
-}
-
 void InitializePlatform() {
 }
 
diff --git a/libsanitizer/tsan/tsan_report.cc b/libsanitizer/tsan/tsan_report.cc
index 6feab81383f..a84738dff6e 100644
--- a/libsanitizer/tsan/tsan_report.cc
+++ b/libsanitizer/tsan/tsan_report.cc
@@ -17,13 +17,11 @@
 
 namespace __tsan {
 
-ReportStack::ReportStack() : next(nullptr), info(), suppressable(false) {}
+ReportStack::ReportStack() : frames(nullptr), suppressable(false) {}
 
-ReportStack *ReportStack::New(uptr addr) {
+ReportStack *ReportStack::New() {
   void *mem = internal_alloc(MBlockReportStack, sizeof(ReportStack));
-  ReportStack *res = new(mem) ReportStack();
-  res->info.address = addr;
-  return res;
+  return new(mem) ReportStack();
 }
 
 ReportLocation::ReportLocation(ReportLocationType type)
@@ -71,7 +69,7 @@ ReportDesc::~ReportDesc() {
   // FIXME(dvyukov): it must be leaking a lot of memory.
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 
 const int kThreadBufSize = 32;
 const char *thread_name(char *buf, int tid) {
@@ -112,13 +110,15 @@ static const char *ReportTypeString(ReportType typ) {
 }
 
 void PrintStack(const ReportStack *ent) {
-  if (ent == 0) {
+  if (ent == 0 || ent->frames == 0) {
     Printf("    [failed to restore the stack]\n\n");
     return;
   }
-  for (int i = 0; ent && ent->info.address; ent = ent->next, i++) {
+  SymbolizedStack *frame = ent->frames;
+  for (int i = 0; frame && frame->info.address; frame = frame->next, i++) {
     InternalScopedString res(2 * GetPageSizeCached());
-    RenderFrame(&res, common_flags()->stack_trace_format, i, ent->info,
+    RenderFrame(&res, common_flags()->stack_trace_format, i, frame->info,
+                common_flags()->symbolize_vs_style,
                 common_flags()->strip_path_prefix, "__interceptor_");
     Printf("%s\n", res.data());
   }
@@ -250,10 +250,20 @@ static ReportStack *ChooseSummaryStack(const ReportDesc *rep) {
   return 0;
 }
 
-ReportStack *SkipTsanInternalFrames(ReportStack *ent) {
-  while (FrameIsInternal(ent) && ent->next)
-    ent = ent->next;
-  return ent;
+static bool FrameIsInternal(const SymbolizedStack *frame) {
+  if (frame == 0)
+    return false;
+  const char *file = frame->info.file;
+  return file != 0 &&
+         (internal_strstr(file, "tsan_interceptors.cc") ||
+          internal_strstr(file, "sanitizer_common_interceptors.inc") ||
+          internal_strstr(file, "tsan_interface_"));
+}
+
+static SymbolizedStack *SkipTsanInternalFrames(SymbolizedStack *frames) {
+  while (FrameIsInternal(frames) && frames->next)
+    frames = frames->next;
+  return frames;
 }
 
 void PrintReport(const ReportDesc *rep) {
@@ -323,27 +333,29 @@ void PrintReport(const ReportDesc *rep) {
   if (rep->typ == ReportTypeThreadLeak && rep->count > 1)
     Printf("  And %d more similar thread leaks.\n\n", rep->count - 1);
 
-  if (ReportStack *ent = SkipTsanInternalFrames(ChooseSummaryStack(rep))) {
-    const AddressInfo &info = ent->info;
-    ReportErrorSummary(rep_typ_str, info.file, info.line, info.function);
+  if (ReportStack *stack = ChooseSummaryStack(rep)) {
+    if (SymbolizedStack *frame = SkipTsanInternalFrames(stack->frames))
+      ReportErrorSummary(rep_typ_str, frame->info);
   }
 
   Printf("==================\n");
 }
 
-#else  // #ifndef TSAN_GO
+#else  // #ifndef SANITIZER_GO
 
 const int kMainThreadId = 1;
 
 void PrintStack(const ReportStack *ent) {
-  if (ent == 0) {
+  if (ent == 0 || ent->frames == 0) {
     Printf("  [failed to restore the stack]\n");
     return;
   }
-  for (int i = 0; ent; ent = ent->next, i++) {
-    const AddressInfo &info = ent->info;
-    Printf("  %s()\n      %s:%d +0x%zx\n", info.function, info.file, info.line,
-           (void *)info.module_offset);
+  SymbolizedStack *frame = ent->frames;
+  for (int i = 0; frame; frame = frame->next, i++) {
+    const AddressInfo &info = frame->info;
+    Printf("  %s()\n      %s:%d +0x%zx\n", info.function,
+        StripPathPrefix(info.file, common_flags()->strip_path_prefix),
+        info.line, (void *)info.module_offset);
   }
 }
 
diff --git a/libsanitizer/tsan/tsan_report.h b/libsanitizer/tsan/tsan_report.h
index 3d99c7a7c24..68924edb547 100644
--- a/libsanitizer/tsan/tsan_report.h
+++ b/libsanitizer/tsan/tsan_report.h
@@ -34,10 +34,9 @@ enum ReportType {
 };
 
 struct ReportStack {
-  ReportStack *next;
-  AddressInfo info;
+  SymbolizedStack *frames;
   bool suppressable;
-  static ReportStack *New(uptr addr);
+  static ReportStack *New();
 
  private:
   ReportStack();
diff --git a/libsanitizer/tsan/tsan_rtl.cc b/libsanitizer/tsan/tsan_rtl.cc
index bf971b62da4..e7bdfb51884 100644
--- a/libsanitizer/tsan/tsan_rtl.cc
+++ b/libsanitizer/tsan/tsan_rtl.cc
@@ -22,6 +22,7 @@
 #include "tsan_mman.h"
 #include "tsan_suppressions.h"
 #include "tsan_symbolize.h"
+#include "ubsan/ubsan_init.h"
 
 #ifdef __SSE3__
 // <emmintrin.h> transitively includes <stdlib.h>,
@@ -41,7 +42,7 @@ extern "C" void __tsan_resume() {
 
 namespace __tsan {
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 THREADLOCAL char cur_thread_placeholder[sizeof(ThreadState)] ALIGNED(64);
 #endif
 static char ctx_placeholder[sizeof(Context)] ALIGNED(64);
@@ -64,14 +65,26 @@ static char thread_registry_placeholder[sizeof(ThreadRegistry)];
 
 static ThreadContextBase *CreateThreadContext(u32 tid) {
   // Map thread trace when context is created.
-  MapThreadTrace(GetThreadTrace(tid), TraceSize() * sizeof(Event));
-  MapThreadTrace(GetThreadTraceHeader(tid), sizeof(Trace));
-  new(ThreadTrace(tid)) Trace();
+  char name[50];
+  internal_snprintf(name, sizeof(name), "trace %u", tid);
+  MapThreadTrace(GetThreadTrace(tid), TraceSize() * sizeof(Event), name);
+  const uptr hdr = GetThreadTraceHeader(tid);
+  internal_snprintf(name, sizeof(name), "trace header %u", tid);
+  MapThreadTrace(hdr, sizeof(Trace), name);
+  new((void*)hdr) Trace();
+  // We are going to use only a small part of the trace with the default
+  // value of history_size. However, the constructor writes to the whole trace.
+  // Unmap the unused part.
+  uptr hdr_end = hdr + sizeof(Trace);
+  hdr_end -= sizeof(TraceHeader) * (kTraceParts - TraceParts());
+  hdr_end = RoundUp(hdr_end, GetPageSizeCached());
+  if (hdr_end < hdr + sizeof(Trace))
+    UnmapOrDie((void*)hdr_end, hdr + sizeof(Trace) - hdr_end);
   void *mem = internal_alloc(MBlockThreadContex, sizeof(ThreadContext));
   return new(mem) ThreadContext(tid);
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 static const u32 kThreadQuarantineSize = 16;
 #else
 static const u32 kThreadQuarantineSize = 64;
@@ -84,8 +97,10 @@ Context::Context()
   , nmissed_expected()
   , thread_registry(new(thread_registry_placeholder) ThreadRegistry(
       CreateThreadContext, kMaxTid, kThreadQuarantineSize, kMaxTidReuse))
+  , racy_mtx(MutexTypeRacy, StatMtxRacy)
   , racy_stacks(MBlockRacyStacks)
   , racy_addresses(MBlockRacyAddresses)
+  , fired_suppressions_mtx(MutexTypeFired, StatMtxFired)
   , fired_suppressions(8) {
 }
 
@@ -100,7 +115,7 @@ ThreadState::ThreadState(Context *ctx, int tid, int unique_id, u64 epoch,
   // , ignore_reads_and_writes()
   // , ignore_interceptors()
   , clock(tid, reuse_count)
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   , jmp_bufs(MBlockJmpBuf)
 #endif
   , tid(tid)
@@ -109,29 +124,28 @@ ThreadState::ThreadState(Context *ctx, int tid, int unique_id, u64 epoch,
   , stk_size(stk_size)
   , tls_addr(tls_addr)
   , tls_size(tls_size)
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   , last_sleep_clock(tid)
 #endif
 {
 }
 
+#ifndef SANITIZER_GO
 static void MemoryProfiler(Context *ctx, fd_t fd, int i) {
   uptr n_threads;
   uptr n_running_threads;
   ctx->thread_registry->GetNumberOfThreads(&n_threads, &n_running_threads);
   InternalScopedBuffer<char> buf(4096);
   WriteMemoryProfile(buf.data(), buf.size(), n_threads, n_running_threads);
-  internal_write(fd, buf.data(), internal_strlen(buf.data()));
+  WriteToFile(fd, buf.data(), internal_strlen(buf.data()));
 }
 
 static void BackgroundThread(void *arg) {
-#ifndef TSAN_GO
   // This is a non-initialized non-user thread, nothing to see here.
   // We don't use ScopedIgnoreInterceptors, because we want ignores to be
   // enabled even when the thread function exits (e.g. during pthread thread
   // shutdown code).
   cur_thread()->ignore_interceptors++;
-#endif
   const u64 kMs2Ns = 1000 * 1000;
 
   fd_t mprof_fd = kInvalidFd;
@@ -141,15 +155,14 @@ static void BackgroundThread(void *arg) {
     } else if (internal_strcmp(flags()->profile_memory, "stderr") == 0) {
       mprof_fd = 2;
     } else {
-      InternalScopedBuffer<char> filename(4096);
-      internal_snprintf(filename.data(), filename.size(), "%s.%d",
-          flags()->profile_memory, (int)internal_getpid());
-      uptr openrv = OpenFile(filename.data(), true);
-      if (internal_iserror(openrv)) {
+      InternalScopedString filename(kMaxPathLength);
+      filename.append("%s.%d", flags()->profile_memory, (int)internal_getpid());
+      fd_t fd = OpenFile(filename.data(), WrOnly);
+      if (fd == kInvalidFd) {
         Printf("ThreadSanitizer: failed to open memory profile file '%s'\n",
             &filename[0]);
       } else {
-        mprof_fd = openrv;
+        mprof_fd = fd;
       }
     }
   }
@@ -190,7 +203,6 @@ static void BackgroundThread(void *arg) {
     if (mprof_fd != kInvalidFd)
       MemoryProfiler(ctx, mprof_fd, i);
 
-#ifndef TSAN_GO
     // Flush symbolizer cache if requested.
     if (flags()->flush_symbolizer_ms > 0) {
       u64 last = atomic_load(&ctx->last_symbolize_time_ns,
@@ -202,7 +214,6 @@ static void BackgroundThread(void *arg) {
         atomic_store(&ctx->last_symbolize_time_ns, 0, memory_order_relaxed);
       }
     }
-#endif
   }
 }
 
@@ -210,13 +221,14 @@ static void StartBackgroundThread() {
   ctx->background_thread = internal_start_thread(&BackgroundThread, 0);
 }
 
-#ifndef TSAN_GO
+#ifndef __mips__
 static void StopBackgroundThread() {
   atomic_store(&ctx->stop_background_thread, 1, memory_order_relaxed);
   internal_join_thread(ctx->background_thread);
   ctx->background_thread = 0;
 }
 #endif
+#endif
 
 void DontNeedShadowFor(uptr addr, uptr size) {
   uptr shadow_beg = MemToShadow(addr);
@@ -228,7 +240,7 @@ void MapShadow(uptr addr, uptr size) {
   // Global data is not 64K aligned, but there are no adjacent mappings,
   // so we can get away with unaligned mapping.
   // CHECK_EQ(addr, addr & ~((64 << 10) - 1));  // windows wants 64K alignment
-  MmapFixedNoReserve(MemToShadow(addr), size * kShadowMultiplier);
+  MmapFixedNoReserve(MemToShadow(addr), size * kShadowMultiplier, "shadow");
 
   // Meta shadow is 2:1, so tread carefully.
   static bool data_mapped = false;
@@ -240,7 +252,7 @@ void MapShadow(uptr addr, uptr size) {
   if (!data_mapped) {
     // First call maps data+bss.
     data_mapped = true;
-    MmapFixedNoReserve(meta_begin, meta_end - meta_begin);
+    MmapFixedNoReserve(meta_begin, meta_end - meta_begin, "meta shadow");
   } else {
     // Mapping continous heap.
     // Windows wants 64K alignment.
@@ -250,19 +262,19 @@ void MapShadow(uptr addr, uptr size) {
       return;
     if (meta_begin < mapped_meta_end)
       meta_begin = mapped_meta_end;
-    MmapFixedNoReserve(meta_begin, meta_end - meta_begin);
+    MmapFixedNoReserve(meta_begin, meta_end - meta_begin, "meta shadow");
     mapped_meta_end = meta_end;
   }
   VPrintf(2, "mapped meta shadow for (%p-%p) at (%p-%p)\n",
       addr, addr+size, meta_begin, meta_end);
 }
 
-void MapThreadTrace(uptr addr, uptr size) {
+void MapThreadTrace(uptr addr, uptr size, const char *name) {
   DPrintf("#0: Mapping trace at %p-%p(0x%zx)\n", addr, addr + size, size);
   CHECK_GE(addr, kTraceMemBeg);
   CHECK_LE(addr + size, kTraceMemEnd);
   CHECK_EQ(addr, addr & ~((64 << 10) - 1));  // windows wants 64K alignment
-  uptr addr1 = (uptr)MmapFixedNoReserve(addr, size);
+  uptr addr1 = (uptr)MmapFixedNoReserve(addr, size, name);
   if (addr1 != addr) {
     Printf("FATAL: ThreadSanitizer can not mmap thread trace (%p/%p->%p)\n",
         addr, size, addr1);
@@ -281,11 +293,11 @@ static void CheckShadowMapping() {
         if (p < beg || p >= end)
           continue;
         const uptr s = MemToShadow(p);
-        VPrintf(3, "  checking pointer %p -> %p\n", p, s);
+        const uptr m = (uptr)MemToMeta(p);
+        VPrintf(3, "  checking pointer %p: shadow=%p meta=%p\n", p, s, m);
         CHECK(IsAppMem(p));
         CHECK(IsShadowMem(s));
         CHECK_EQ(p & ~(kShadowCell - 1), ShadowToMem(s));
-        const uptr m = (uptr)MemToMeta(p);
         CHECK(IsMetaMem(m));
       }
     }
@@ -306,8 +318,10 @@ void Initialize(ThreadState *thr) {
 
   ctx = new(ctx_placeholder) Context;
   const char *options = GetEnv(kTsanOptionsEnv);
+  CacheBinaryName();
   InitializeFlags(&ctx->flags, options);
-#ifndef TSAN_GO
+  CheckVMASize();
+#ifndef SANITIZER_GO
   InitializeAllocator();
 #endif
   InitializeInterceptors();
@@ -315,20 +329,23 @@ void Initialize(ThreadState *thr) {
   InitializePlatform();
   InitializeMutex();
   InitializeDynamicAnnotations();
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   InitializeShadowMemory();
 #endif
   // Setup correct file descriptor for error reports.
   __sanitizer_set_report_path(common_flags()->log_path);
   InitializeSuppressions();
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   InitializeLibIgnore();
   Symbolizer::GetOrInit()->AddHooks(EnterSymbolizer, ExitSymbolizer);
-#endif
+  // On MIPS, TSan initialization is run before
+  // __pthread_initialize_minimal_internal() is finished, so we can not spawn
+  // new threads.
+#ifndef __mips__
   StartBackgroundThread();
-#ifndef TSAN_GO
   SetSandboxingCallback(StopBackgroundThread);
 #endif
+#endif
   if (common_flags()->detect_deadlocks)
     ctx->dd = DDetector::Create(flags());
 
@@ -339,6 +356,9 @@ void Initialize(ThreadState *thr) {
   int tid = ThreadCreate(thr, 0, 0, true);
   CHECK_EQ(tid, 0);
   ThreadStart(thr, tid, internal_getpid());
+#if TSAN_CONTAINS_UBSAN
+  __ubsan::InitAsPlugin();
+#endif
   ctx->initialized = true;
 
   if (flags()->stop_on_start) {
@@ -363,16 +383,15 @@ int Finalize(ThreadState *thr) {
   CommonSanitizerReportMutex.Unlock();
   ctx->report_mtx.Unlock();
 
-#ifndef TSAN_GO
-  if (common_flags()->verbosity)
-    AllocatorPrintStats();
+#ifndef SANITIZER_GO
+  if (Verbosity()) AllocatorPrintStats();
 #endif
 
   ThreadFinalize(thr);
 
   if (ctx->nreported) {
     failed = true;
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
     Printf("ThreadSanitizer: reported %d warnings\n", ctx->nreported);
 #else
     Printf("Found %d data race(s)\n", ctx->nreported);
@@ -387,19 +406,22 @@ int Finalize(ThreadState *thr) {
 
   if (common_flags()->print_suppressions)
     PrintMatchedSuppressions();
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   if (flags()->print_benign)
     PrintMatchedBenignRaces();
 #endif
 
   failed = OnFinalize(failed);
 
+#if TSAN_COLLECT_STATS
   StatAggregate(ctx->stat, thr->stat);
   StatOutput(ctx->stat);
-  return failed ? flags()->exitcode : 0;
+#endif
+
+  return failed ? common_flags()->exitcode : 0;
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 void ForkBefore(ThreadState *thr, uptr pc) {
   ctx->thread_registry->Lock();
   ctx->report_mtx.Lock();
@@ -419,7 +441,7 @@ void ForkChildAfter(ThreadState *thr, uptr pc) {
   VPrintf(1, "ThreadSanitizer: forked new process with pid %d,"
       " parent had %d threads\n", (int)internal_getpid(), (int)nthread);
   if (nthread == 1) {
-    internal_start_thread(&BackgroundThread, 0);
+    StartBackgroundThread();
   } else {
     // We've just forked a multi-threaded process. We cannot reasonably function
     // after that (some mutexes may be locked before fork). So just enable
@@ -432,7 +454,7 @@ void ForkChildAfter(ThreadState *thr, uptr pc) {
 }
 #endif
 
-#ifdef TSAN_GO
+#ifdef SANITIZER_GO
 NOINLINE
 void GrowShadowStack(ThreadState *thr) {
   const int sz = thr->shadow_stack_end - thr->shadow_stack;
@@ -448,10 +470,10 @@ void GrowShadowStack(ThreadState *thr) {
 #endif
 
 u32 CurrentStackId(ThreadState *thr, uptr pc) {
-  if (thr->shadow_stack_pos == 0)  // May happen during bootstrap.
+  if (!thr->is_inited)  // May happen during bootstrap.
     return 0;
   if (pc != 0) {
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
     DCHECK_LT(thr->shadow_stack_pos, thr->shadow_stack_end);
 #else
     if (thr->shadow_stack_pos == thr->shadow_stack_end)
@@ -497,7 +519,7 @@ uptr TraceParts() {
   return TraceSize() / kTracePartSize;
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 extern "C" void __tsan_trace_switch() {
   TraceSwitch(cur_thread());
 }
@@ -530,7 +552,7 @@ void HandleRace(ThreadState *thr, u64 *shadow_mem,
   thr->racy_state[0] = cur.raw();
   thr->racy_state[1] = old.raw();
   thr->racy_shadow_addr = shadow_mem;
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   HACKY_CALL(__tsan_report_race);
 #else
   ReportRace(thr);
@@ -564,43 +586,26 @@ void MemoryAccessImpl1(ThreadState *thr, uptr addr,
   // it's just not worth it (performance- and complexity-wise).
 
   Shadow old(0);
-  if (kShadowCnt == 1) {
-    int idx = 0;
-#include "tsan_update_shadow_word_inl.h"
-  } else if (kShadowCnt == 2) {
-    int idx = 0;
-#include "tsan_update_shadow_word_inl.h"
-    idx = 1;
-#include "tsan_update_shadow_word_inl.h"
-  } else if (kShadowCnt == 4) {
-    int idx = 0;
-#include "tsan_update_shadow_word_inl.h"
-    idx = 1;
-#include "tsan_update_shadow_word_inl.h"
-    idx = 2;
-#include "tsan_update_shadow_word_inl.h"
-    idx = 3;
-#include "tsan_update_shadow_word_inl.h"
-  } else if (kShadowCnt == 8) {
-    int idx = 0;
-#include "tsan_update_shadow_word_inl.h"
-    idx = 1;
-#include "tsan_update_shadow_word_inl.h"
-    idx = 2;
-#include "tsan_update_shadow_word_inl.h"
-    idx = 3;
+
+  // It release mode we manually unroll the loop,
+  // because empirically gcc generates better code this way.
+  // However, we can't afford unrolling in debug mode, because the function
+  // consumes almost 4K of stack. Gtest gives only 4K of stack to death test
+  // threads, which is not enough for the unrolled loop.
+#if SANITIZER_DEBUG
+  for (int idx = 0; idx < 4; idx++) {
 #include "tsan_update_shadow_word_inl.h"
-    idx = 4;
+  }
+#else
+  int idx = 0;
 #include "tsan_update_shadow_word_inl.h"
-    idx = 5;
+  idx = 1;
 #include "tsan_update_shadow_word_inl.h"
-    idx = 6;
+  idx = 2;
 #include "tsan_update_shadow_word_inl.h"
-    idx = 7;
+  idx = 3;
 #include "tsan_update_shadow_word_inl.h"
-  } else {
-    CHECK(false);
-  }
+#endif
 
   // we did not find any races and had already stored
   // the current access info, so we are done
@@ -651,7 +656,7 @@ bool ContainsSameAccessSlow(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
   return false;
 }
 
-#if defined(__SSE3__) && TSAN_SHADOW_COUNT == 4
+#if defined(__SSE3__)
 #define SHUF(v0, v1, i0, i1, i2, i3) _mm_castps_si128(_mm_shuffle_ps( \
     _mm_castsi128_ps(v0), _mm_castsi128_ps(v1), \
     (i0)*1 + (i1)*4 + (i2)*16 + (i3)*64))
@@ -711,11 +716,12 @@ bool ContainsSameAccessFast(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
 
 ALWAYS_INLINE
 bool ContainsSameAccess(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
-#if defined(__SSE3__) && TSAN_SHADOW_COUNT == 4
+#if defined(__SSE3__)
   bool res = ContainsSameAccessFast(s, a, sync_epoch, is_write);
   // NOTE: this check can fail if the shadow is concurrently mutated
-  // by other threads.
-  DCHECK_EQ(res, ContainsSameAccessSlow(s, a, sync_epoch, is_write));
+  // by other threads. But it still can be useful if you modify
+  // ContainsSameAccessFast and want to ensure that it's not completely broken.
+  // DCHECK_EQ(res, ContainsSameAccessSlow(s, a, sync_epoch, is_write));
   return res;
 #else
   return ContainsSameAccessSlow(s, a, sync_epoch, is_write);
@@ -732,7 +738,7 @@ void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
       (int)(1 << kAccessSizeLog), kAccessIsWrite, shadow_mem,
       (uptr)shadow_mem[0], (uptr)shadow_mem[1],
       (uptr)shadow_mem[2], (uptr)shadow_mem[3]);
-#if TSAN_DEBUG
+#if SANITIZER_DEBUG
   if (!IsAppMem(addr)) {
     Printf("Access to non app mem %zx\n", addr);
     DCHECK(IsAppMem(addr));
@@ -842,7 +848,7 @@ static void MemoryRangeSet(ThreadState *thr, uptr pc, uptr addr, uptr size,
     }
   } else {
     // The region is big, reset only beginning and end.
-    const uptr kPageSize = 4096;
+    const uptr kPageSize = GetPageSizeCached();
     u64 *begin = (u64*)MemToShadow(addr);
     u64 *end = begin + size / kShadowCell * kShadowCnt;
     u64 *p = begin;
@@ -916,7 +922,7 @@ void FuncEntry(ThreadState *thr, uptr pc) {
   // Shadow stack maintenance can be replaced with
   // stack unwinding during trace switch (which presumably must be faster).
   DCHECK_GE(thr->shadow_stack_pos, thr->shadow_stack);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   DCHECK_LT(thr->shadow_stack_pos, thr->shadow_stack_end);
 #else
   if (thr->shadow_stack_pos == thr->shadow_stack_end)
@@ -936,7 +942,7 @@ void FuncExit(ThreadState *thr) {
   }
 
   DCHECK_GT(thr->shadow_stack_pos, thr->shadow_stack);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   DCHECK_LT(thr->shadow_stack_pos, thr->shadow_stack_end);
 #endif
   thr->shadow_stack_pos--;
@@ -947,7 +953,7 @@ void ThreadIgnoreBegin(ThreadState *thr, uptr pc) {
   thr->ignore_reads_and_writes++;
   CHECK_GT(thr->ignore_reads_and_writes, 0);
   thr->fast_state.SetIgnoreBit();
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   if (!ctx->after_multithreaded_fork)
     thr->mop_ignore_set.Add(CurrentStackId(thr, pc));
 #endif
@@ -959,7 +965,7 @@ void ThreadIgnoreEnd(ThreadState *thr, uptr pc) {
   CHECK_GE(thr->ignore_reads_and_writes, 0);
   if (thr->ignore_reads_and_writes == 0) {
     thr->fast_state.ClearIgnoreBit();
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
     thr->mop_ignore_set.Reset();
 #endif
   }
@@ -969,7 +975,7 @@ void ThreadIgnoreSyncBegin(ThreadState *thr, uptr pc) {
   DPrintf("#%d: ThreadIgnoreSyncBegin\n", thr->tid);
   thr->ignore_sync++;
   CHECK_GT(thr->ignore_sync, 0);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   if (!ctx->after_multithreaded_fork)
     thr->sync_ignore_set.Add(CurrentStackId(thr, pc));
 #endif
@@ -979,7 +985,7 @@ void ThreadIgnoreSyncEnd(ThreadState *thr, uptr pc) {
   DPrintf("#%d: ThreadIgnoreSyncEnd\n", thr->tid);
   thr->ignore_sync--;
   CHECK_GE(thr->ignore_sync, 0);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   if (thr->ignore_sync == 0)
     thr->sync_ignore_set.Reset();
 #endif
@@ -989,7 +995,7 @@ bool MD5Hash::operator==(const MD5Hash &other) const {
   return hash[0] == other.hash[0] && hash[1] == other.hash[1];
 }
 
-#if TSAN_DEBUG
+#if SANITIZER_DEBUG
 void build_consistency_debug() {}
 #else
 void build_consistency_release() {}
@@ -1001,19 +1007,9 @@ void build_consistency_stats() {}
 void build_consistency_nostats() {}
 #endif
 
-#if TSAN_SHADOW_COUNT == 1
-void build_consistency_shadow1() {}
-#elif TSAN_SHADOW_COUNT == 2
-void build_consistency_shadow2() {}
-#elif TSAN_SHADOW_COUNT == 4
-void build_consistency_shadow4() {}
-#else
-void build_consistency_shadow8() {}
-#endif
-
 }  // namespace __tsan
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 // Must be included in this file to make sure everything is inlined.
 #include "tsan_interface_inl.h"
 #endif
diff --git a/libsanitizer/tsan/tsan_rtl.h b/libsanitizer/tsan/tsan_rtl.h
index a5b88e0c358..34b009b15d7 100644
--- a/libsanitizer/tsan/tsan_rtl.h
+++ b/libsanitizer/tsan/tsan_rtl.h
@@ -50,10 +50,23 @@
 
 namespace __tsan {
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 struct MapUnmapCallback;
+#if defined(__mips64) || defined(__aarch64__)
+static const uptr kAllocatorSpace = 0;
+static const uptr kAllocatorSize = SANITIZER_MMAP_RANGE_SIZE;
+static const uptr kAllocatorRegionSizeLog = 20;
+static const uptr kAllocatorNumRegions =
+    kAllocatorSize >> kAllocatorRegionSizeLog;
+typedef TwoLevelByteMap<(kAllocatorNumRegions >> 12), 1 << 12,
+    MapUnmapCallback> ByteMap;
+typedef SizeClassAllocator32<kAllocatorSpace, kAllocatorSize, 0,
+    CompactSizeClassMap, kAllocatorRegionSizeLog, ByteMap,
+    MapUnmapCallback> PrimaryAllocator;
+#else
 typedef SizeClassAllocator64<kHeapMemBeg, kHeapMemEnd - kHeapMemBeg, 0,
     DefaultSizeClassMap, MapUnmapCallback> PrimaryAllocator;
+#endif
 typedef SizeClassAllocatorLocalCache<PrimaryAllocator> AllocatorCache;
 typedef LargeMmapAllocator<MapUnmapCallback> SecondaryAllocator;
 typedef CombinedAllocator<PrimaryAllocator, AllocatorCache,
@@ -298,7 +311,7 @@ class Shadow : public FastState {
   }
 };
 
-struct SignalContext;
+struct ThreadSignalContext;
 
 struct JmpBuf {
   uptr sp;
@@ -330,7 +343,7 @@ struct ThreadState {
   int ignore_reads_and_writes;
   int ignore_sync;
   // Go does not support ignores.
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   IgnoreSet mop_ignore_set;
   IgnoreSet sync_ignore_set;
 #endif
@@ -343,17 +356,20 @@ struct ThreadState {
   u64 racy_state[2];
   MutexSet mset;
   ThreadClock clock;
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   AllocatorCache alloc_cache;
   InternalAllocatorCache internal_alloc_cache;
   Vector<JmpBuf> jmp_bufs;
   int ignore_interceptors;
 #endif
+#if TSAN_COLLECT_STATS
   u64 stat[StatCnt];
+#endif
   const int tid;
   const int unique_id;
   bool in_symbolizer;
   bool in_ignored_lib;
+  bool is_inited;
   bool is_dead;
   bool is_freeing;
   bool is_vptr_access;
@@ -363,18 +379,20 @@ struct ThreadState {
   const uptr tls_size;
   ThreadContext *tctx;
 
+#if SANITIZER_DEBUG && !SANITIZER_GO
   InternalDeadlockDetector internal_deadlock_detector;
+#endif
   DDPhysicalThread *dd_pt;
   DDLogicalThread *dd_lt;
 
   atomic_uintptr_t in_signal_handler;
-  SignalContext *signal_ctx;
+  ThreadSignalContext *signal_ctx;
 
   DenseSlabAllocCache block_cache;
   DenseSlabAllocCache sync_cache;
   DenseSlabAllocCache clock_cache;
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   u32 last_sleep_stack_id;
   ThreadClock last_sleep_clock;
 #endif
@@ -389,7 +407,7 @@ struct ThreadState {
                        uptr tls_addr, uptr tls_size);
 };
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 __attribute__((tls_model("initial-exec")))
 extern THREADLOCAL char cur_thread_placeholder[];
 INLINE ThreadState *cur_thread() {
@@ -411,13 +429,13 @@ class ThreadContext : public ThreadContextBase {
   u64 epoch1;
 
   // Override superclass callbacks.
-  void OnDead();
-  void OnJoined(void *arg);
-  void OnFinished();
-  void OnStarted(void *arg);
-  void OnCreated(void *arg);
-  void OnReset();
-  void OnDetached(void *arg);
+  void OnDead() override;
+  void OnJoined(void *arg) override;
+  void OnFinished() override;
+  void OnStarted(void *arg) override;
+  void OnCreated(void *arg) override;
+  void OnReset() override;
+  void OnDetached(void *arg) override;
 };
 
 struct RacyStacks {
@@ -438,7 +456,7 @@ struct RacyAddress {
 
 struct FiredSuppression {
   ReportType type;
-  uptr pc;
+  uptr pc_or_addr;
   Suppression *supp;
 };
 
@@ -460,9 +478,11 @@ struct Context {
 
   ThreadRegistry *thread_registry;
 
+  Mutex racy_mtx;
   Vector<RacyStacks> racy_stacks;
   Vector<RacyAddress> racy_addresses;
   // Number of fired suppressions may be large enough.
+  Mutex fired_suppressions_mtx;
   InternalMmapVector<FiredSuppression> fired_suppressions;
   DDetector *dd;
 
@@ -479,13 +499,13 @@ extern Context *ctx;  // The one and the only global runtime context.
 
 struct ScopedIgnoreInterceptors {
   ScopedIgnoreInterceptors() {
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
     cur_thread()->ignore_interceptors++;
 #endif
   }
 
   ~ScopedIgnoreInterceptors() {
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
     cur_thread()->ignore_interceptors--;
 #endif
   }
@@ -537,19 +557,24 @@ void ObtainCurrentStack(ThreadState *thr, uptr toppc, StackTraceTy *stack) {
 }
 
 
+#if TSAN_COLLECT_STATS
 void StatAggregate(u64 *dst, u64 *src);
 void StatOutput(u64 *stat);
+#endif
+
 void ALWAYS_INLINE StatInc(ThreadState *thr, StatType typ, u64 n = 1) {
-  if (kCollectStats)
-    thr->stat[typ] += n;
+#if TSAN_COLLECT_STATS
+  thr->stat[typ] += n;
+#endif
 }
 void ALWAYS_INLINE StatSet(ThreadState *thr, StatType typ, u64 n) {
-  if (kCollectStats)
-    thr->stat[typ] = n;
+#if TSAN_COLLECT_STATS
+  thr->stat[typ] = n;
+#endif
 }
 
 void MapShadow(uptr addr, uptr size);
-void MapThreadTrace(uptr addr, uptr size);
+void MapThreadTrace(uptr addr, uptr size, const char *name);
 void DontNeedShadowFor(uptr addr, uptr size);
 void InitializeShadowMemory();
 void InitializeInterceptors();
@@ -562,12 +587,9 @@ void ForkChildAfter(ThreadState *thr, uptr pc);
 
 void ReportRace(ThreadState *thr);
 bool OutputReport(ThreadState *thr, const ScopedReport &srep);
-bool IsFiredSuppression(Context *ctx, const ScopedReport &srep,
-                        StackTrace trace);
+bool IsFiredSuppression(Context *ctx, ReportType type, StackTrace trace);
 bool IsExpectedReport(uptr addr, uptr size);
 void PrintMatchedBenignRaces();
-bool FrameIsInternal(const ReportStack *frame);
-ReportStack *SkipTsanInternalFrames(ReportStack *ent);
 
 #if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 1
 # define DPrintf Printf
@@ -664,6 +686,12 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr);
 void MutexRepair(ThreadState *thr, uptr pc, uptr addr);  // call on EOWNERDEAD
 
 void Acquire(ThreadState *thr, uptr pc, uptr addr);
+// AcquireGlobal synchronizes the current thread with all other threads.
+// In terms of happens-before relation, it draws a HB edge from all threads
+// (where they happen to execute right now) to the current thread. We use it to
+// handle Go finalizers. Namely, finalizer goroutine executes AcquireGlobal
+// right before executing finalizers. This provides a coarse, but simple
+// approximation of the actual required synchronization.
 void AcquireGlobal(ThreadState *thr, uptr pc);
 void Release(ThreadState *thr, uptr pc, uptr addr);
 void ReleaseStore(ThreadState *thr, uptr pc, uptr addr);
@@ -679,7 +707,7 @@ void AcquireReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c);
 // The trick is that the call preserves all registers and the compiler
 // does not treat it as a call.
 // If it does not work for you, use normal call.
-#if TSAN_DEBUG == 0 && defined(__x86_64__)
+#if !SANITIZER_DEBUG && defined(__x86_64__)
 // The caller may not create the stack frame for itself at all,
 // so we create a reserve stack frame for it (1024b must be enough).
 #define HACKY_CALL(f) \
@@ -711,7 +739,7 @@ void ALWAYS_INLINE TraceAddEvent(ThreadState *thr, FastState fs,
   StatInc(thr, StatEvents);
   u64 pos = fs.GetTracePos();
   if (UNLIKELY((pos % kTracePartSize) == 0)) {
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
     HACKY_CALL(__tsan_trace_switch);
 #else
     TraceSwitch(thr);
@@ -723,6 +751,12 @@ void ALWAYS_INLINE TraceAddEvent(ThreadState *thr, FastState fs,
   *evp = ev;
 }
 
+#ifndef SANITIZER_GO
+uptr ALWAYS_INLINE HeapEnd() {
+  return kHeapMemEnd + PrimaryAllocator::AdditionalSize();
+}
+#endif
+
 }  // namespace __tsan
 
 #endif  // TSAN_RTL_H
diff --git a/libsanitizer/tsan/tsan_rtl_mutex.cc b/libsanitizer/tsan/tsan_rtl_mutex.cc
index d731b4bfb59..deb7722f521 100644
--- a/libsanitizer/tsan/tsan_rtl_mutex.cc
+++ b/libsanitizer/tsan/tsan_rtl_mutex.cc
@@ -34,12 +34,8 @@ struct Callback : DDCallback {
     DDCallback::lt = thr->dd_lt;
   }
 
-  virtual u32 Unwind() {
-    return CurrentStackId(thr, pc);
-  }
-  virtual int UniqueTid() {
-    return thr->unique_id;
-  }
+  u32 Unwind() override { return CurrentStackId(thr, pc); }
+  int UniqueTid() override { return thr->unique_id; }
 };
 
 void DDMutexInit(ThreadState *thr, uptr pc, SyncVar *s) {
@@ -86,7 +82,7 @@ void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
 void MutexDestroy(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexDestroy %zx\n", thr->tid, addr);
   StatInc(thr, StatMutexDestroy);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   // Global mutexes not marked as LINKER_INITIALIZED
   // cause tons of not interesting reports, so just ignore it.
   if (IsGlobalVar(addr))
@@ -403,7 +399,7 @@ void ReleaseStore(ThreadState *thr, uptr pc, uptr addr) {
   s->mtx.Unlock();
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 static void UpdateSleepClockCallback(ThreadContextBase *tctx_base, void *arg) {
   ThreadState *thr = reinterpret_cast<ThreadState*>(arg);
   ThreadContext *tctx = static_cast<ThreadContext*>(tctx_base);
@@ -474,7 +470,7 @@ void ReportDeadlock(ThreadState *thr, uptr pc, DDReport *r) {
   for (int i = 0; i < r->n; i++) {
     for (int j = 0; j < (flags()->second_deadlock_stack ? 2 : 1); j++) {
       u32 stk = r->loop[i].stk[j];
-      if (stk) {
+      if (stk && stk != 0xffffffff) {
         rep.AddStack(StackDepotGet(stk), true);
       } else {
         // Sometimes we fail to extract the stack trace (FIXME: investigate),
diff --git a/libsanitizer/tsan/tsan_rtl_report.cc b/libsanitizer/tsan/tsan_rtl_report.cc
index f86cfd4681d..d0d1fbaf45a 100644
--- a/libsanitizer/tsan/tsan_rtl_report.cc
+++ b/libsanitizer/tsan/tsan_rtl_report.cc
@@ -54,40 +54,43 @@ bool WEAK OnReport(const ReportDesc *rep, bool suppressed) {
 }
 #endif
 
-static void StackStripMain(ReportStack *stack) {
-  ReportStack *last_frame = 0;
-  ReportStack *last_frame2 = 0;
-  for (ReportStack *ent = stack; ent; ent = ent->next) {
+static void StackStripMain(SymbolizedStack *frames) {
+  SymbolizedStack *last_frame = nullptr;
+  SymbolizedStack *last_frame2 = nullptr;
+  for (SymbolizedStack *cur = frames; cur; cur = cur->next) {
     last_frame2 = last_frame;
-    last_frame = ent;
+    last_frame = cur;
   }
 
   if (last_frame2 == 0)
     return;
+#ifndef SANITIZER_GO
   const char *last = last_frame->info.function;
-#ifndef TSAN_GO
   const char *last2 = last_frame2->info.function;
   // Strip frame above 'main'
   if (last2 && 0 == internal_strcmp(last2, "main")) {
-    last_frame2->next = 0;
+    last_frame->ClearAll();
+    last_frame2->next = nullptr;
   // Strip our internal thread start routine.
   } else if (last && 0 == internal_strcmp(last, "__tsan_thread_start_func")) {
-    last_frame2->next = 0;
+    last_frame->ClearAll();
+    last_frame2->next = nullptr;
   // Strip global ctors init.
   } else if (last && 0 == internal_strcmp(last, "__do_global_ctors_aux")) {
-    last_frame2->next = 0;
+    last_frame->ClearAll();
+    last_frame2->next = nullptr;
   // If both are 0, then we probably just failed to symbolize.
   } else if (last || last2) {
     // Ensure that we recovered stack completely. Trimmed stack
     // can actually happen if we do not instrument some code,
     // so it's only a debug print. However we must try hard to not miss it
     // due to our fault.
-    DPrintf("Bottom stack frame of stack %zx is missed\n", stack->pc);
+    DPrintf("Bottom stack frame is missed\n");
   }
 #else
   // The last frame always point into runtime (gosched0, goexit0, runtime.main).
-  last_frame2->next = 0;
-  (void)last;
+  last_frame->ClearAll();
+  last_frame2->next = nullptr;
 #endif
 }
 
@@ -103,31 +106,29 @@ ReportStack *SymbolizeStackId(u32 stack_id) {
 static ReportStack *SymbolizeStack(StackTrace trace) {
   if (trace.size == 0)
     return 0;
-  ReportStack *stack = 0;
+  SymbolizedStack *top = nullptr;
   for (uptr si = 0; si < trace.size; si++) {
     const uptr pc = trace.trace[si];
-#ifndef TSAN_GO
-    // We obtain the return address, that is, address of the next instruction,
-    // so offset it by 1 byte.
-    const uptr pc1 = StackTrace::GetPreviousInstructionPc(pc);
-#else
-    // FIXME(dvyukov): Go sometimes uses address of a function as top pc.
     uptr pc1 = pc;
-    if (si != trace.size - 1)
-      pc1 -= 1;
-#endif
-    ReportStack *ent = SymbolizeCode(pc1);
+    // We obtain the return address, but we're interested in the previous
+    // instruction.
+    if ((pc & kExternalPCBit) == 0)
+      pc1 = StackTrace::GetPreviousInstructionPc(pc);
+    SymbolizedStack *ent = SymbolizeCode(pc1);
     CHECK_NE(ent, 0);
-    ReportStack *last = ent;
+    SymbolizedStack *last = ent;
     while (last->next) {
       last->info.address = pc;  // restore original pc for report
       last = last->next;
     }
     last->info.address = pc;  // restore original pc for report
-    last->next = stack;
-    stack = ent;
+    last->next = top;
+    top = ent;
   }
-  StackStripMain(stack);
+  StackStripMain(top);
+
+  ReportStack *stack = ReportStack::New();
+  stack->frames = top;
   return stack;
 }
 
@@ -196,7 +197,7 @@ void ScopedReport::AddThread(const ThreadContext *tctx, bool suppressable) {
     rt->stack->suppressable = suppressable;
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 static ThreadContext *FindThreadByUidLocked(int unique_id) {
   ctx->thread_registry->CheckLocked();
   for (unsigned i = 0; i < kMaxTid; i++) {
@@ -241,7 +242,7 @@ ThreadContext *IsThreadStackOrTls(uptr addr, bool *is_stack) {
 #endif
 
 void ScopedReport::AddThread(int unique_tid, bool suppressable) {
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   if (const ThreadContext *tctx = FindThreadByUidLocked(unique_tid))
     AddThread(tctx, suppressable);
 #endif
@@ -297,7 +298,7 @@ void ScopedReport::AddDeadMutex(u64 id) {
 void ScopedReport::AddLocation(uptr addr, uptr size) {
   if (addr == 0)
     return;
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   int fd = -1;
   int creat_tid = -1;
   u32 creat_stack = 0;
@@ -347,7 +348,7 @@ void ScopedReport::AddLocation(uptr addr, uptr size) {
 #endif
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 void ScopedReport::AddSleep(u32 stack_id) {
   rep_->sleep = SymbolizeStackId(stack_id);
 }
@@ -366,30 +367,23 @@ void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
   // This function restores stack trace and mutex set for the thread/epoch.
   // It does so by getting stack trace and mutex set at the beginning of
   // trace part, and then replaying the trace till the given epoch.
-  ctx->thread_registry->CheckLocked();
-  ThreadContext *tctx = static_cast<ThreadContext*>(
-      ctx->thread_registry->GetThreadLocked(tid));
-  if (tctx == 0)
-    return;
-  if (tctx->status != ThreadStatusRunning
-      && tctx->status != ThreadStatusFinished
-      && tctx->status != ThreadStatusDead)
-    return;
-  Trace* trace = ThreadTrace(tctx->tid);
-  Lock l(&trace->mtx);
+  Trace* trace = ThreadTrace(tid);
+  ReadLock l(&trace->mtx);
   const int partidx = (epoch / kTracePartSize) % TraceParts();
   TraceHeader* hdr = &trace->headers[partidx];
-  if (epoch < hdr->epoch0)
+  if (epoch < hdr->epoch0 || epoch >= hdr->epoch0 + kTracePartSize)
     return;
+  CHECK_EQ(RoundDown(epoch, kTracePartSize), hdr->epoch0);
   const u64 epoch0 = RoundDown(epoch, TraceSize());
   const u64 eend = epoch % TraceSize();
   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);
-  InternalScopedBuffer<uptr> stack(kShadowStackSize);
+  Vector<uptr> stack(MBlockReportStack);
+  stack.Resize(hdr->stack0.size + 64);
   for (uptr i = 0; i < hdr->stack0.size; i++) {
     stack[i] = hdr->stack0.trace[i];
-    DPrintf2("  #%02lu: pc=%zx\n", i, stack[i]);
+    DPrintf2("  #%02zu: pc=%zx\n", i, stack[i]);
   }
   if (mset)
     *mset = hdr->mset0;
@@ -403,6 +397,8 @@ void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
     if (typ == EventTypeMop) {
       stack[pos] = pc;
     } else if (typ == EventTypeFuncEnter) {
+      if (stack.Size() < pos + 2)
+        stack.Resize(pos + 2);
       stack[pos++] = pc;
     } else if (typ == EventTypeFuncExit) {
       if (pos > 0)
@@ -425,50 +421,58 @@ void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
   if (pos == 0 && stack[0] == 0)
     return;
   pos++;
-  stk->Init(stack.data(), pos);
+  stk->Init(&stack[0], pos);
 }
 
 static bool HandleRacyStacks(ThreadState *thr, VarSizeStackTrace traces[2],
                              uptr addr_min, uptr addr_max) {
   bool equal_stack = false;
   RacyStacks hash;
-  if (flags()->suppress_equal_stacks) {
-    hash.hash[0] = md5_hash(traces[0].trace, traces[0].size * sizeof(uptr));
-    hash.hash[1] = md5_hash(traces[1].trace, traces[1].size * sizeof(uptr));
-    for (uptr i = 0; i < ctx->racy_stacks.Size(); i++) {
-      if (hash == ctx->racy_stacks[i]) {
-        DPrintf("ThreadSanitizer: suppressing report as doubled (stack)\n");
-        equal_stack = true;
-        break;
-      }
-    }
-  }
   bool equal_address = false;
   RacyAddress ra0 = {addr_min, addr_max};
-  if (flags()->suppress_equal_addresses) {
-    for (uptr i = 0; i < ctx->racy_addresses.Size(); i++) {
-      RacyAddress ra2 = ctx->racy_addresses[i];
-      uptr maxbeg = max(ra0.addr_min, ra2.addr_min);
-      uptr minend = min(ra0.addr_max, ra2.addr_max);
-      if (maxbeg < minend) {
-        DPrintf("ThreadSanitizer: suppressing report as doubled (addr)\n");
-        equal_address = true;
-        break;
+  {
+    ReadLock lock(&ctx->racy_mtx);
+    if (flags()->suppress_equal_stacks) {
+      hash.hash[0] = md5_hash(traces[0].trace, traces[0].size * sizeof(uptr));
+      hash.hash[1] = md5_hash(traces[1].trace, traces[1].size * sizeof(uptr));
+      for (uptr i = 0; i < ctx->racy_stacks.Size(); i++) {
+        if (hash == ctx->racy_stacks[i]) {
+          VPrintf(2,
+              "ThreadSanitizer: suppressing report as doubled (stack)\n");
+          equal_stack = true;
+          break;
+        }
+      }
+    }
+    if (flags()->suppress_equal_addresses) {
+      for (uptr i = 0; i < ctx->racy_addresses.Size(); i++) {
+        RacyAddress ra2 = ctx->racy_addresses[i];
+        uptr maxbeg = max(ra0.addr_min, ra2.addr_min);
+        uptr minend = min(ra0.addr_max, ra2.addr_max);
+        if (maxbeg < minend) {
+          VPrintf(2, "ThreadSanitizer: suppressing report as doubled (addr)\n");
+          equal_address = true;
+          break;
+        }
       }
     }
   }
-  if (equal_stack || equal_address) {
-    if (!equal_stack)
-      ctx->racy_stacks.PushBack(hash);
-    if (!equal_address)
-      ctx->racy_addresses.PushBack(ra0);
-    return true;
+  if (!equal_stack && !equal_address)
+    return false;
+  if (!equal_stack) {
+    Lock lock(&ctx->racy_mtx);
+    ctx->racy_stacks.PushBack(hash);
   }
-  return false;
+  if (!equal_address) {
+    Lock lock(&ctx->racy_mtx);
+    ctx->racy_addresses.PushBack(ra0);
+  }
+  return true;
 }
 
 static void AddRacyStacks(ThreadState *thr, VarSizeStackTrace traces[2],
                           uptr addr_min, uptr addr_max) {
+  Lock lock(&ctx->racy_mtx);
   if (flags()->suppress_equal_stacks) {
     RacyStacks hash;
     hash.hash[0] = md5_hash(traces[0].trace, traces[0].size * sizeof(uptr));
@@ -482,26 +486,29 @@ static void AddRacyStacks(ThreadState *thr, VarSizeStackTrace traces[2],
 }
 
 bool OutputReport(ThreadState *thr, const ScopedReport &srep) {
-  atomic_store(&ctx->last_symbolize_time_ns, NanoTime(), memory_order_relaxed);
+  if (!flags()->report_bugs)
+    return false;
+  atomic_store_relaxed(&ctx->last_symbolize_time_ns, NanoTime());
   const ReportDesc *rep = srep.GetReport();
   Suppression *supp = 0;
-  uptr suppress_pc = 0;
-  for (uptr i = 0; suppress_pc == 0 && i < rep->mops.Size(); i++)
-    suppress_pc = IsSuppressed(rep->typ, rep->mops[i]->stack, &supp);
-  for (uptr i = 0; suppress_pc == 0 && i < rep->stacks.Size(); i++)
-    suppress_pc = IsSuppressed(rep->typ, rep->stacks[i], &supp);
-  for (uptr i = 0; suppress_pc == 0 && i < rep->threads.Size(); i++)
-    suppress_pc = IsSuppressed(rep->typ, rep->threads[i]->stack, &supp);
-  for (uptr i = 0; suppress_pc == 0 && i < rep->locs.Size(); i++)
-    suppress_pc = IsSuppressed(rep->typ, rep->locs[i], &supp);
-  if (suppress_pc != 0) {
-    FiredSuppression s = {srep.GetReport()->typ, suppress_pc, supp};
+  uptr pc_or_addr = 0;
+  for (uptr i = 0; pc_or_addr == 0 && i < rep->mops.Size(); i++)
+    pc_or_addr = IsSuppressed(rep->typ, rep->mops[i]->stack, &supp);
+  for (uptr i = 0; pc_or_addr == 0 && i < rep->stacks.Size(); i++)
+    pc_or_addr = IsSuppressed(rep->typ, rep->stacks[i], &supp);
+  for (uptr i = 0; pc_or_addr == 0 && i < rep->threads.Size(); i++)
+    pc_or_addr = IsSuppressed(rep->typ, rep->threads[i]->stack, &supp);
+  for (uptr i = 0; pc_or_addr == 0 && i < rep->locs.Size(); i++)
+    pc_or_addr = IsSuppressed(rep->typ, rep->locs[i], &supp);
+  if (pc_or_addr != 0) {
+    Lock lock(&ctx->fired_suppressions_mtx);
+    FiredSuppression s = {srep.GetReport()->typ, pc_or_addr, supp};
     ctx->fired_suppressions.push_back(s);
   }
   {
     bool old_is_freeing = thr->is_freeing;
     thr->is_freeing = false;
-    bool suppressed = OnReport(rep, suppress_pc != 0);
+    bool suppressed = OnReport(rep, pc_or_addr != 0);
     thr->is_freeing = old_is_freeing;
     if (suppressed)
       return false;
@@ -509,20 +516,20 @@ bool OutputReport(ThreadState *thr, const ScopedReport &srep) {
   PrintReport(rep);
   ctx->nreported++;
   if (flags()->halt_on_error)
-    internal__exit(flags()->exitcode);
+    Die();
   return true;
 }
 
-bool IsFiredSuppression(Context *ctx, const ScopedReport &srep,
-                        StackTrace trace) {
+bool IsFiredSuppression(Context *ctx, ReportType type, StackTrace trace) {
+  ReadLock lock(&ctx->fired_suppressions_mtx);
   for (uptr k = 0; k < ctx->fired_suppressions.size(); k++) {
-    if (ctx->fired_suppressions[k].type != srep.GetReport()->typ)
+    if (ctx->fired_suppressions[k].type != type)
       continue;
     for (uptr j = 0; j < trace.size; j++) {
       FiredSuppression *s = &ctx->fired_suppressions[k];
-      if (trace.trace[j] == s->pc) {
+      if (trace.trace[j] == s->pc_or_addr) {
         if (s->supp)
-          s->supp->hit_count++;
+          atomic_fetch_add(&s->supp->hit_count, 1, memory_order_relaxed);
         return true;
       }
     }
@@ -530,32 +537,21 @@ bool IsFiredSuppression(Context *ctx, const ScopedReport &srep,
   return false;
 }
 
-static bool IsFiredSuppression(Context *ctx,
-                               const ScopedReport &srep,
-                               uptr addr) {
+static bool IsFiredSuppression(Context *ctx, ReportType type, uptr addr) {
+  ReadLock lock(&ctx->fired_suppressions_mtx);
   for (uptr k = 0; k < ctx->fired_suppressions.size(); k++) {
-    if (ctx->fired_suppressions[k].type != srep.GetReport()->typ)
+    if (ctx->fired_suppressions[k].type != type)
       continue;
     FiredSuppression *s = &ctx->fired_suppressions[k];
-    if (addr == s->pc) {
+    if (addr == s->pc_or_addr) {
       if (s->supp)
-        s->supp->hit_count++;
+        atomic_fetch_add(&s->supp->hit_count, 1, memory_order_relaxed);
       return true;
     }
   }
   return false;
 }
 
-bool FrameIsInternal(const ReportStack *frame) {
-  if (frame == 0)
-    return false;
-  const char *file = frame->info.file;
-  return file != 0 &&
-         (internal_strstr(file, "tsan_interceptors.cc") ||
-          internal_strstr(file, "sanitizer_common_interceptors.inc") ||
-          internal_strstr(file, "tsan_interface_"));
-}
-
 static bool RaceBetweenAtomicAndFree(ThreadState *thr) {
   Shadow s0(thr->racy_state[0]);
   Shadow s1(thr->racy_state[1]);
@@ -602,8 +598,6 @@ void ReportRace(ThreadState *thr) {
       return;
   }
 
-  ThreadRegistryLock l0(ctx->thread_registry);
-
   ReportType typ = ReportTypeRace;
   if (thr->is_vptr_access && freed)
     typ = ReportTypeVptrUseAfterFree;
@@ -611,29 +605,35 @@ void ReportRace(ThreadState *thr) {
     typ = ReportTypeVptrRace;
   else if (freed)
     typ = ReportTypeUseAfterFree;
-  ScopedReport rep(typ);
-  if (IsFiredSuppression(ctx, rep, addr))
+
+  if (IsFiredSuppression(ctx, typ, addr))
     return;
+
   const uptr kMop = 2;
   VarSizeStackTrace traces[kMop];
   const uptr toppc = TraceTopPC(thr);
   ObtainCurrentStack(thr, toppc, &traces[0]);
-  if (IsFiredSuppression(ctx, rep, traces[0]))
+  if (IsFiredSuppression(ctx, typ, traces[0]))
     return;
-  InternalScopedBuffer<MutexSet> mset2(1);
-  new(mset2.data()) MutexSet();
+
+  // MutexSet is too large to live on stack.
+  Vector<u64> mset_buffer(MBlockScopedBuf);
+  mset_buffer.Resize(sizeof(MutexSet) / sizeof(u64) + 1);
+  MutexSet *mset2 = new(&mset_buffer[0]) MutexSet();
+
   Shadow s2(thr->racy_state[1]);
-  RestoreStack(s2.tid(), s2.epoch(), &traces[1], mset2.data());
-  if (IsFiredSuppression(ctx, rep, traces[1]))
+  RestoreStack(s2.tid(), s2.epoch(), &traces[1], mset2);
+  if (IsFiredSuppression(ctx, typ, traces[1]))
     return;
 
   if (HandleRacyStacks(thr, traces, addr_min, addr_max))
     return;
 
+  ThreadRegistryLock l0(ctx->thread_registry);
+  ScopedReport rep(typ);
   for (uptr i = 0; i < kMop; i++) {
     Shadow s(thr->racy_state[i]);
-    rep.AddMemoryAccess(addr, s, traces[i],
-                        i == 0 ? &thr->mset : mset2.data());
+    rep.AddMemoryAccess(addr, s, traces[i], i == 0 ? &thr->mset : mset2);
   }
 
   for (uptr i = 0; i < kMop; i++) {
@@ -647,7 +647,7 @@ void ReportRace(ThreadState *thr) {
 
   rep.AddLocation(addr_min, addr_max - addr_min);
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   {  // NOLINT
     Shadow s(thr->racy_state[1]);
     if (s.epoch() <= thr->last_sleep_clock.get(s.tid()))
@@ -668,7 +668,7 @@ void PrintCurrentStack(ThreadState *thr, uptr pc) {
 }
 
 void PrintCurrentStackSlow(uptr pc) {
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   BufferedStackTrace *ptrace =
       new(internal_alloc(MBlockStackTrace, sizeof(BufferedStackTrace)))
           BufferedStackTrace();
diff --git a/libsanitizer/tsan/tsan_rtl_thread.cc b/libsanitizer/tsan/tsan_rtl_thread.cc
index d75445aa8bb..9d7e2d3028e 100644
--- a/libsanitizer/tsan/tsan_rtl_thread.cc
+++ b/libsanitizer/tsan/tsan_rtl_thread.cc
@@ -28,7 +28,7 @@ ThreadContext::ThreadContext(int tid)
   , epoch1() {
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 ThreadContext::~ThreadContext() {
 }
 #endif
@@ -90,7 +90,7 @@ void ThreadContext::OnStarted(void *arg) {
   epoch1 = (u64)-1;
   new(thr) ThreadState(ctx, tid, unique_id, epoch0, reuse_count,
       args->stk_addr, args->stk_size, args->tls_addr, args->tls_size);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   thr->shadow_stack = &ThreadTrace(thr->tid)->shadow_stack[0];
   thr->shadow_stack_pos = thr->shadow_stack;
   thr->shadow_stack_end = thr->shadow_stack + kShadowStackSize;
@@ -102,7 +102,7 @@ void ThreadContext::OnStarted(void *arg) {
   thr->shadow_stack_pos = thr->shadow_stack;
   thr->shadow_stack_end = thr->shadow_stack + kInitStackSize;
 #endif
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   AllocatorThreadStart(thr);
 #endif
   if (common_flags()->detect_deadlocks) {
@@ -118,6 +118,7 @@ void ThreadContext::OnStarted(void *arg) {
   AcquireImpl(thr, 0, &sync);
   StatInc(thr, StatSyncAcquire);
   sync.Reset(&thr->clock_cache);
+  thr->is_inited = true;
   DPrintf("#%d: ThreadStart epoch=%zu stk_addr=%zx stk_size=%zx "
           "tls_addr=%zx tls_size=%zx\n",
           tid, (uptr)epoch0, args->stk_addr, args->stk_size,
@@ -139,15 +140,17 @@ void ThreadContext::OnFinished() {
   }
   ctx->clock_alloc.FlushCache(&thr->clock_cache);
   ctx->metamap.OnThreadIdle(thr);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   AllocatorThreadFinish(thr);
 #endif
   thr->~ThreadState();
+#if TSAN_COLLECT_STATS
   StatAggregate(ctx->stat, thr->stat);
+#endif
   thr = 0;
 }
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 struct ThreadLeak {
   ThreadContext *tctx;
   int count;
@@ -169,7 +172,7 @@ static void MaybeReportThreadLeak(ThreadContextBase *tctx_base, void *arg) {
 }
 #endif
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
 static void ReportIgnoresEnabled(ThreadContext *tctx, IgnoreSet *set) {
   if (tctx->tid == 0) {
     Printf("ThreadSanitizer: main thread finished with ignores enabled\n");
@@ -201,7 +204,7 @@ static void ThreadCheckIgnore(ThreadState *thr) {}
 
 void ThreadFinalize(ThreadState *thr) {
   ThreadCheckIgnore(thr);
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   if (!flags()->report_thread_leaks)
     return;
   ThreadRegistryLock l(ctx->thread_registry);
@@ -237,6 +240,7 @@ void ThreadStart(ThreadState *thr, int tid, uptr os_id) {
   uptr stk_size = 0;
   uptr tls_addr = 0;
   uptr tls_size = 0;
+#ifndef SANITIZER_GO
   GetThreadStackAndTls(tid == 0, &stk_addr, &stk_size, &tls_addr, &tls_size);
 
   if (tid) {
@@ -257,6 +261,7 @@ void ThreadStart(ThreadState *thr, int tid, uptr os_id) {
           thr_end, tls_addr + tls_size - thr_end);
     }
   }
+#endif
 
   ThreadRegistry *tr = ctx->thread_registry;
   OnStartedArgs args = { thr, stk_addr, stk_size, tls_addr, tls_size };
@@ -266,7 +271,7 @@ void ThreadStart(ThreadState *thr, int tid, uptr os_id) {
   thr->tctx = (ThreadContext*)tr->GetThreadLocked(tid);
   tr->Unlock();
 
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   if (ctx->after_multithreaded_fork) {
     thr->ignore_interceptors++;
     ThreadIgnoreBegin(thr, 0);
@@ -328,7 +333,7 @@ void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
       thr->tid, (void*)pc, (void*)addr,
       (int)size, is_write);
 
-#if TSAN_DEBUG
+#if SANITIZER_DEBUG
   if (!IsAppMem(addr)) {
     Printf("Access to non app mem %zx\n", addr);
     DCHECK(IsAppMem(addr));
diff --git a/libsanitizer/tsan/tsan_stat.cc b/libsanitizer/tsan/tsan_stat.cc
index cdb48358551..5eccca69312 100644
--- a/libsanitizer/tsan/tsan_stat.cc
+++ b/libsanitizer/tsan/tsan_stat.cc
@@ -13,17 +13,14 @@
 
 namespace __tsan {
 
+#if TSAN_COLLECT_STATS
+
 void StatAggregate(u64 *dst, u64 *src) {
-  if (!kCollectStats)
-    return;
   for (int i = 0; i < StatCnt; i++)
     dst[i] += src[i];
 }
 
 void StatOutput(u64 *stat) {
-  if (!kCollectStats)
-    return;
-
   stat[StatShadowNonZero] = stat[StatShadowProcessed] - stat[StatShadowZero];
 
   static const char *name[StatCnt] = {};
@@ -165,8 +162,9 @@ void StatOutput(u64 *stat) {
   name[StatMtxAtExit]                    = "  Atexit                          ";
   name[StatMtxAnnotations]               = "  Annotations                     ";
   name[StatMtxMBlock]                    = "  MBlock                          ";
-  name[StatMtxJavaMBlock]                = "  JavaMBlock                      ";
   name[StatMtxDeadlockDetector]          = "  DeadlockDetector                ";
+  name[StatMtxFired]                     = "  FiredSuppressions               ";
+  name[StatMtxRacy]                      = "  RacyStacks                      ";
   name[StatMtxFD]                        = "  FD                              ";
 
   Printf("Statistics:\n");
@@ -174,4 +172,6 @@ void StatOutput(u64 *stat) {
     Printf("%s: %16zu\n", name[i], (uptr)stat[i]);
 }
 
+#endif
+
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_stat.h b/libsanitizer/tsan/tsan_stat.h
index 132656f035f..002570f42f1 100644
--- a/libsanitizer/tsan/tsan_stat.h
+++ b/libsanitizer/tsan/tsan_stat.h
@@ -167,8 +167,9 @@ enum StatType {
   StatMtxAnnotations,
   StatMtxAtExit,
   StatMtxMBlock,
-  StatMtxJavaMBlock,
   StatMtxDeadlockDetector,
+  StatMtxFired,
+  StatMtxRacy,
   StatMtxFD,
 
   // This must be the last.
diff --git a/libsanitizer/tsan/tsan_suppressions.cc b/libsanitizer/tsan/tsan_suppressions.cc
index 76460d90d92..5c8d03dddd2 100644
--- a/libsanitizer/tsan/tsan_suppressions.cc
+++ b/libsanitizer/tsan/tsan_suppressions.cc
@@ -19,6 +19,7 @@
 #include "tsan_mman.h"
 #include "tsan_platform.h"
 
+#ifndef SANITIZER_GO
 // Suppressions for true/false positives in standard libraries.
 static const char *const std_suppressions =
 // Libstdc++ 4.4 has data races in std::string.
@@ -31,7 +32,6 @@ static const char *const std_suppressions =
 "race:std::_Sp_counted_ptr_inplace<std::thread::_Impl\n";
 
 // Can be overriden in frontend.
-#ifndef TSAN_GO
 extern "C" const char *WEAK __tsan_default_suppressions() {
   return 0;
 }
@@ -39,84 +39,105 @@ extern "C" const char *WEAK __tsan_default_suppressions() {
 
 namespace __tsan {
 
-static bool suppressions_inited = false;
+ALIGNED(64) static char suppression_placeholder[sizeof(SuppressionContext)];
+static SuppressionContext *suppression_ctx = nullptr;
+static const char *kSuppressionTypes[] = {
+    kSuppressionRace,   kSuppressionRaceTop, kSuppressionMutex,
+    kSuppressionThread, kSuppressionSignal, kSuppressionLib,
+    kSuppressionDeadlock};
 
 void InitializeSuppressions() {
-  CHECK(!suppressions_inited);
-  SuppressionContext::InitIfNecessary();
-#ifndef TSAN_GO
-  SuppressionContext::Get()->Parse(__tsan_default_suppressions());
-  SuppressionContext::Get()->Parse(std_suppressions);
+  CHECK_EQ(nullptr, suppression_ctx);
+  suppression_ctx = new (suppression_placeholder) // NOLINT
+      SuppressionContext(kSuppressionTypes, ARRAY_SIZE(kSuppressionTypes));
+  suppression_ctx->ParseFromFile(flags()->suppressions);
+#ifndef SANITIZER_GO
+  suppression_ctx->Parse(__tsan_default_suppressions());
+  suppression_ctx->Parse(std_suppressions);
 #endif
-  suppressions_inited = true;
 }
 
-SuppressionType conv(ReportType typ) {
+SuppressionContext *Suppressions() {
+  CHECK(suppression_ctx);
+  return suppression_ctx;
+}
+
+static const char *conv(ReportType typ) {
   if (typ == ReportTypeRace)
-    return SuppressionRace;
+    return kSuppressionRace;
   else if (typ == ReportTypeVptrRace)
-    return SuppressionRace;
+    return kSuppressionRace;
   else if (typ == ReportTypeUseAfterFree)
-    return SuppressionRace;
+    return kSuppressionRace;
   else if (typ == ReportTypeVptrUseAfterFree)
-    return SuppressionRace;
+    return kSuppressionRace;
   else if (typ == ReportTypeThreadLeak)
-    return SuppressionThread;
+    return kSuppressionThread;
   else if (typ == ReportTypeMutexDestroyLocked)
-    return SuppressionMutex;
+    return kSuppressionMutex;
   else if (typ == ReportTypeMutexDoubleLock)
-    return SuppressionMutex;
+    return kSuppressionMutex;
   else if (typ == ReportTypeMutexBadUnlock)
-    return SuppressionMutex;
+    return kSuppressionMutex;
   else if (typ == ReportTypeMutexBadReadLock)
-    return SuppressionMutex;
+    return kSuppressionMutex;
   else if (typ == ReportTypeMutexBadReadUnlock)
-    return SuppressionMutex;
+    return kSuppressionMutex;
   else if (typ == ReportTypeSignalUnsafe)
-    return SuppressionSignal;
+    return kSuppressionSignal;
   else if (typ == ReportTypeErrnoInSignal)
-    return SuppressionNone;
+    return kSuppressionNone;
   else if (typ == ReportTypeDeadlock)
-    return SuppressionDeadlock;
+    return kSuppressionDeadlock;
   Printf("ThreadSanitizer: unknown report type %d\n", typ),
   Die();
 }
 
+static uptr IsSuppressed(const char *stype, const AddressInfo &info,
+    Suppression **sp) {
+  if (suppression_ctx->Match(info.function, stype, sp) ||
+      suppression_ctx->Match(info.file, stype, sp) ||
+      suppression_ctx->Match(info.module, stype, sp)) {
+    VPrintf(2, "ThreadSanitizer: matched suppression '%s'\n", (*sp)->templ);
+    atomic_fetch_add(&(*sp)->hit_count, 1, memory_order_relaxed);
+    return info.address;
+  }
+  return 0;
+}
+
 uptr IsSuppressed(ReportType typ, const ReportStack *stack, Suppression **sp) {
-  if (!SuppressionContext::Get()->SuppressionCount() || stack == 0 ||
+  CHECK(suppression_ctx);
+  if (!suppression_ctx->SuppressionCount() || stack == 0 ||
       !stack->suppressable)
     return 0;
-  SuppressionType stype = conv(typ);
-  if (stype == SuppressionNone)
+  const char *stype = conv(typ);
+  if (0 == internal_strcmp(stype, kSuppressionNone))
     return 0;
-  Suppression *s;
-  for (const ReportStack *frame = stack; frame; frame = frame->next) {
-    const AddressInfo &info = frame->info;
-    if (SuppressionContext::Get()->Match(info.function, stype, &s) ||
-        SuppressionContext::Get()->Match(info.file, stype, &s) ||
-        SuppressionContext::Get()->Match(info.module, stype, &s)) {
-      DPrintf("ThreadSanitizer: matched suppression '%s'\n", s->templ);
-      s->hit_count++;
-      *sp = s;
-      return info.address;
-    }
+  for (const SymbolizedStack *frame = stack->frames; frame;
+      frame = frame->next) {
+    uptr pc = IsSuppressed(stype, frame->info, sp);
+    if (pc != 0)
+      return pc;
   }
+  if (0 == internal_strcmp(stype, kSuppressionRace) && stack->frames != nullptr)
+    return IsSuppressed(kSuppressionRaceTop, stack->frames->info, sp);
   return 0;
 }
 
 uptr IsSuppressed(ReportType typ, const ReportLocation *loc, Suppression **sp) {
-  if (!SuppressionContext::Get()->SuppressionCount() || loc == 0 ||
+  CHECK(suppression_ctx);
+  if (!suppression_ctx->SuppressionCount() || loc == 0 ||
       loc->type != ReportLocationGlobal || !loc->suppressable)
     return 0;
-  SuppressionType stype = conv(typ);
-  if (stype == SuppressionNone)
+  const char *stype = conv(typ);
+  if (0 == internal_strcmp(stype, kSuppressionNone))
     return 0;
   Suppression *s;
   const DataInfo &global = loc->global;
-  if (SuppressionContext::Get()->Match(global.name, stype, &s) ||
-      SuppressionContext::Get()->Match(global.module, stype, &s)) {
-      DPrintf("ThreadSanitizer: matched suppression '%s'\n", s->templ);
-      s->hit_count++;
+  if (suppression_ctx->Match(global.name, stype, &s) ||
+      suppression_ctx->Match(global.module, stype, &s)) {
+      VPrintf(2, "ThreadSanitizer: matched suppression '%s'\n", s->templ);
+      atomic_fetch_add(&s->hit_count, 1, memory_order_relaxed);
       *sp = s;
       return global.start;
   }
@@ -125,17 +146,18 @@ uptr IsSuppressed(ReportType typ, const ReportLocation *loc, Suppression **sp) {
 
 void PrintMatchedSuppressions() {
   InternalMmapVector<Suppression *> matched(1);
-  SuppressionContext::Get()->GetMatched(&matched);
+  CHECK(suppression_ctx);
+  suppression_ctx->GetMatched(&matched);
   if (!matched.size())
     return;
   int hit_count = 0;
   for (uptr i = 0; i < matched.size(); i++)
-    hit_count += matched[i]->hit_count;
+    hit_count += atomic_load_relaxed(&matched[i]->hit_count);
   Printf("ThreadSanitizer: Matched %d suppressions (pid=%d):\n", hit_count,
          (int)internal_getpid());
   for (uptr i = 0; i < matched.size(); i++) {
-    Printf("%d %s:%s\n", matched[i]->hit_count,
-           SuppressionTypeString(matched[i]->type), matched[i]->templ);
+    Printf("%d %s:%s\n", matched[i]->hit_count, matched[i]->type,
+           matched[i]->templ);
   }
 }
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_suppressions.h b/libsanitizer/tsan/tsan_suppressions.h
index e38d81ece85..58951975075 100644
--- a/libsanitizer/tsan/tsan_suppressions.h
+++ b/libsanitizer/tsan/tsan_suppressions.h
@@ -16,7 +16,17 @@
 
 namespace __tsan {
 
+const char kSuppressionNone[] = "none";
+const char kSuppressionRace[] = "race";
+const char kSuppressionRaceTop[] = "race_top";
+const char kSuppressionMutex[] = "mutex";
+const char kSuppressionThread[] = "thread";
+const char kSuppressionSignal[] = "signal";
+const char kSuppressionLib[] = "called_from_lib";
+const char kSuppressionDeadlock[] = "deadlock";
+
 void InitializeSuppressions();
+SuppressionContext *Suppressions();
 void PrintMatchedSuppressions();
 uptr IsSuppressed(ReportType typ, const ReportStack *stack, Suppression **sp);
 uptr IsSuppressed(ReportType typ, const ReportLocation *loc, Suppression **sp);
diff --git a/libsanitizer/tsan/tsan_symbolize.cc b/libsanitizer/tsan/tsan_symbolize.cc
index 795f838991b..0e54562e385 100644
--- a/libsanitizer/tsan/tsan_symbolize.cc
+++ b/libsanitizer/tsan/tsan_symbolize.cc
@@ -34,26 +34,16 @@ void ExitSymbolizer() {
   thr->ignore_interceptors--;
 }
 
-// Denotes fake PC values that come from JIT/JAVA/etc.
-// For such PC values __tsan_symbolize_external() will be called.
-const uptr kExternalPCBit = 1ULL << 60;
-
 // May be overriden by JIT/JAVA/etc,
 // whatever produces PCs marked with kExternalPCBit.
-extern "C" bool __tsan_symbolize_external(uptr pc,
-                               char *func_buf, uptr func_siz,
-                               char *file_buf, uptr file_siz,
-                               int *line, int *col)
-                               SANITIZER_WEAK_ATTRIBUTE;
-
-bool __tsan_symbolize_external(uptr pc,
-                               char *func_buf, uptr func_siz,
-                               char *file_buf, uptr file_siz,
-                               int *line, int *col) {
+extern "C" bool WEAK __tsan_symbolize_external(uptr pc,
+                                               char *func_buf, uptr func_siz,
+                                               char *file_buf, uptr file_siz,
+                                               int *line, int *col) {
   return false;
 }
 
-ReportStack *SymbolizeCode(uptr addr) {
+SymbolizedStack *SymbolizeCode(uptr addr) {
   // Check if PC comes from non-native land.
   if (addr & kExternalPCBit) {
     // Declare static to not consume too much stack space.
@@ -61,36 +51,17 @@ ReportStack *SymbolizeCode(uptr addr) {
     static char func_buf[1024];
     static char file_buf[1024];
     int line, col;
-    ReportStack *ent = ReportStack::New(addr);
-    if (!__tsan_symbolize_external(addr, func_buf, sizeof(func_buf),
-                                  file_buf, sizeof(file_buf), &line, &col))
-      return ent;
-    ent->info.function = internal_strdup(func_buf);
-    ent->info.file = internal_strdup(file_buf);
-    ent->info.line = line;
-    ent->info.column = col;
-    return ent;
-  }
-  static const uptr kMaxAddrFrames = 16;
-  InternalScopedBuffer<AddressInfo> addr_frames(kMaxAddrFrames);
-  for (uptr i = 0; i < kMaxAddrFrames; i++)
-    new(&addr_frames[i]) AddressInfo();
-  uptr addr_frames_num = Symbolizer::GetOrInit()->SymbolizePC(
-      addr, addr_frames.data(), kMaxAddrFrames);
-  if (addr_frames_num == 0)
-    return ReportStack::New(addr);
-  ReportStack *top = 0;
-  ReportStack *bottom = 0;
-  for (uptr i = 0; i < addr_frames_num; i++) {
-    ReportStack *cur_entry = ReportStack::New(addr);
-    cur_entry->info = addr_frames[i];
-    if (i == 0)
-      top = cur_entry;
-    else
-      bottom->next = cur_entry;
-    bottom = cur_entry;
+    SymbolizedStack *frame = SymbolizedStack::New(addr);
+    if (__tsan_symbolize_external(addr, func_buf, sizeof(func_buf), file_buf,
+                                  sizeof(file_buf), &line, &col)) {
+      frame->info.function = internal_strdup(func_buf);
+      frame->info.file = internal_strdup(file_buf);
+      frame->info.line = line;
+      frame->info.column = col;
+    }
+    return frame;
   }
-  return top;
+  return Symbolizer::GetOrInit()->SymbolizePC(addr);
 }
 
 ReportLocation *SymbolizeData(uptr addr) {
diff --git a/libsanitizer/tsan/tsan_symbolize.h b/libsanitizer/tsan/tsan_symbolize.h
index 0d9077ed379..a859f6318b0 100644
--- a/libsanitizer/tsan/tsan_symbolize.h
+++ b/libsanitizer/tsan/tsan_symbolize.h
@@ -18,7 +18,7 @@ namespace __tsan {
 
 void EnterSymbolizer();
 void ExitSymbolizer();
-ReportStack *SymbolizeCode(uptr addr);
+SymbolizedStack *SymbolizeCode(uptr addr);
 ReportLocation *SymbolizeData(uptr addr);
 void SymbolizeFlush();
 
diff --git a/libsanitizer/tsan/tsan_sync.cc b/libsanitizer/tsan/tsan_sync.cc
index 2209199ac48..91ad8c8b228 100644
--- a/libsanitizer/tsan/tsan_sync.cc
+++ b/libsanitizer/tsan/tsan_sync.cc
@@ -78,17 +78,21 @@ uptr MetaMap::FreeBlock(ThreadState *thr, uptr pc, uptr p) {
   return sz;
 }
 
-void MetaMap::FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
+bool MetaMap::FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
+  bool has_something = false;
   u32 *meta = MemToMeta(p);
   u32 *end = MemToMeta(p + sz);
   if (end == meta)
     end++;
   for (; meta < end; meta++) {
     u32 idx = *meta;
+    if (idx == 0) {
+      // Note: don't write to meta in this case -- the block can be huge.
+      continue;
+    }
     *meta = 0;
-    for (;;) {
-      if (idx == 0)
-        break;
+    has_something = true;
+    while (idx != 0) {
       if (idx & kFlagBlock) {
         block_alloc_.Free(&thr->block_cache, idx & ~kFlagMask);
         break;
@@ -104,6 +108,64 @@ void MetaMap::FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
       }
     }
   }
+  return has_something;
+}
+
+// ResetRange removes all meta objects from the range.
+// It is called for large mmap-ed regions. The function is best-effort wrt
+// freeing of meta objects, because we don't want to page in the whole range
+// which can be huge. The function probes pages one-by-one until it finds a page
+// without meta objects, at this point it stops freeing meta objects. Because
+// thread stacks grow top-down, we do the same starting from end as well.
+void MetaMap::ResetRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
+  const uptr kMetaRatio = kMetaShadowCell / kMetaShadowSize;
+  const uptr kPageSize = GetPageSizeCached() * kMetaRatio;
+  if (sz <= 4 * kPageSize) {
+    // If the range is small, just do the normal free procedure.
+    FreeRange(thr, pc, p, sz);
+    return;
+  }
+  // First, round both ends of the range to page size.
+  uptr diff = RoundUp(p, kPageSize) - p;
+  if (diff != 0) {
+    FreeRange(thr, pc, p, diff);
+    p += diff;
+    sz -= diff;
+  }
+  diff = p + sz - RoundDown(p + sz, kPageSize);
+  if (diff != 0) {
+    FreeRange(thr, pc, p + sz - diff, diff);
+    sz -= diff;
+  }
+  // Now we must have a non-empty page-aligned range.
+  CHECK_GT(sz, 0);
+  CHECK_EQ(p, RoundUp(p, kPageSize));
+  CHECK_EQ(sz, RoundUp(sz, kPageSize));
+  const uptr p0 = p;
+  const uptr sz0 = sz;
+  // Probe start of the range.
+  while (sz > 0) {
+    bool has_something = FreeRange(thr, pc, p, kPageSize);
+    p += kPageSize;
+    sz -= kPageSize;
+    if (!has_something)
+      break;
+  }
+  // Probe end of the range.
+  while (sz > 0) {
+    bool has_something = FreeRange(thr, pc, p - kPageSize, kPageSize);
+    sz -= kPageSize;
+    if (!has_something)
+      break;
+  }
+  // Finally, page out the whole range (including the parts that we've just
+  // freed). Note: we can't simply madvise, because we need to leave a zeroed
+  // range (otherwise __tsan_java_move can crash if it encounters a left-over
+  // meta objects in java heap).
+  uptr metap = (uptr)MemToMeta(p0);
+  uptr metasz = sz0 / kMetaRatio;
+  UnmapOrDie((void*)metap, metasz);
+  MmapFixedNoReserve(metap, metasz);
 }
 
 MBlock* MetaMap::GetBlock(uptr p) {
diff --git a/libsanitizer/tsan/tsan_sync.h b/libsanitizer/tsan/tsan_sync.h
index 6ed3715ee0a..50bc872275d 100644
--- a/libsanitizer/tsan/tsan_sync.h
+++ b/libsanitizer/tsan/tsan_sync.h
@@ -71,7 +71,8 @@ class MetaMap {
 
   void AllocBlock(ThreadState *thr, uptr pc, uptr p, uptr sz);
   uptr FreeBlock(ThreadState *thr, uptr pc, uptr p);
-  void FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz);
+  bool FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz);
+  void ResetRange(ThreadState *thr, uptr pc, uptr p, uptr sz);
   MBlock* GetBlock(uptr p);
 
   SyncVar* GetOrCreateAndLock(ThreadState *thr, uptr pc,
@@ -83,9 +84,9 @@ class MetaMap {
   void OnThreadIdle(ThreadState *thr);
 
  private:
-  static const u32 kFlagMask  = 3 << 30;
-  static const u32 kFlagBlock = 1 << 30;
-  static const u32 kFlagSync  = 2 << 30;
+  static const u32 kFlagMask  = 3u << 30;
+  static const u32 kFlagBlock = 1u << 30;
+  static const u32 kFlagSync  = 2u << 30;
   typedef DenseSlabAlloc<MBlock, 1<<16, 1<<12> BlockAlloc;
   typedef DenseSlabAlloc<SyncVar, 1<<16, 1<<10> SyncAlloc;
   BlockAlloc block_alloc_;
diff --git a/libsanitizer/tsan/tsan_trace.h b/libsanitizer/tsan/tsan_trace.h
index 8eceb634a53..a27efa9fd7a 100644
--- a/libsanitizer/tsan/tsan_trace.h
+++ b/libsanitizer/tsan/tsan_trace.h
@@ -18,9 +18,9 @@
 
 namespace __tsan {
 
-const int kTracePartSizeBits = 14;
+const int kTracePartSizeBits = 13;
 const int kTracePartSize = 1 << kTracePartSizeBits;
-const int kTraceParts = 4 * 1024 * 1024 / kTracePartSize;
+const int kTraceParts = 2 * 1024 * 1024 / kTracePartSize;
 const int kTraceSize = kTracePartSize * kTraceParts;
 
 // Must fit into 3 bits.
@@ -40,7 +40,7 @@ enum EventType {
 typedef u64 Event;
 
 struct TraceHeader {
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   BufferedStackTrace stack0;  // Start stack for the trace.
 #else
   VarSizeStackTrace stack0;
@@ -52,13 +52,15 @@ struct TraceHeader {
 };
 
 struct Trace {
-  TraceHeader headers[kTraceParts];
   Mutex mtx;
-#ifndef TSAN_GO
+#ifndef SANITIZER_GO
   // Must be last to catch overflow as paging fault.
   // Go shadow stack is dynamically allocated.
   uptr shadow_stack[kShadowStackSize];
 #endif
+  // Must be the last field, because we unmap the unused part in
+  // CreateThreadContext.
+  TraceHeader headers[kTraceParts];
 
   Trace()
     : mtx(MutexTypeTrace, StatMtxTrace) {
diff --git a/libsanitizer/tsan/tsan_update_shadow_word_inl.h b/libsanitizer/tsan/tsan_update_shadow_word_inl.h
index 91def7bb1e8..2ea74283818 100644
--- a/libsanitizer/tsan/tsan_update_shadow_word_inl.h
+++ b/libsanitizer/tsan/tsan_update_shadow_word_inl.h
@@ -36,7 +36,8 @@ do {
     }
     StatInc(thr, StatShadowAnotherThread);
     if (HappensBefore(old, thr)) {
-      StoreIfNotYetStored(sp, &store_word);
+      if (old.IsRWWeakerOrEqual(kAccessIsWrite, kIsAtomic))
+        StoreIfNotYetStored(sp, &store_word);
       break;
     }
     if (old.IsBothReadsOrAtomic(kAccessIsWrite, kIsAtomic))
author	chefmax <chefmax@138bc75d-0d04-0410-961f-82ee72b054a4>	2015-10-21 07:32:45 +0000
committer	chefmax <chefmax@138bc75d-0d04-0410-961f-82ee72b054a4>	2015-10-21 07:32:45 +0000
commit	5645a48f7ebd0f97a072b7a2eb40b27cea9d4318 (patch)
tree	2bdaf703dd35e1806b59bd7d74c7eee290a1054f /libsanitizer/tsan
parent	397881d34f32eddf4a6665789f1a7cdd5ff3695e (diff)
download	gcc-5645a48f7ebd0f97a072b7a2eb40b27cea9d4318.tar.gz