configure: add --enable-tsan flag + fiber annotations for coroutine-ucontext

We tried running QEMU under tsan in 2016, but tsan's lack of support for
longjmp-based fibers was a blocker:
  https://groups.google.com/forum/#!topic/thread-sanitizer/se0YuzfWazw

Fortunately, thread sanitizer gained fiber support in early 2019:
  https://reviews.llvm.org/D54889

This patch brings tsan support upstream by importing the patch that annotated
QEMU's coroutines as tsan fibers in Android's QEMU fork:
  https://android-review.googlesource.com/c/platform/external/qemu/+/844675

Tested with '--enable-tsan --cc=clang-9 --cxx=clang++-9 --disable-werror'
configure flags.

Signed-off-by: Lingfeng Yang <lfy@google.com>
Signed-off-by: Emilio G. Cota <cota@braap.org>
[cota: minor modifications + configure changes]
Signed-off-by: Robert Foley <robert.foley@linaro.org>
[RF: configure changes, coroutine fix + minor modifications]
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200609200738.445-2-robert.foley@linaro.org>
Message-Id: <20200612190237.30436-5-alex.bennee@linaro.org>

configure

@@ -395,6 +395,7 @@ gprof="no"
 debug_tcg="no"
 debug="no"
 sanitizers="no"
+tsan="no"
 fortify_source=""
 strip_opt="yes"
 tcg_interpreter="no"
@@ -1152,6 +1153,10 @@ for opt do
   ;;
   --disable-sanitizers) sanitizers="no"
   ;;
+  --enable-tsan) tsan="yes"
+  ;;
+  --disable-tsan) tsan="no"
+  ;;
   --enable-sparse) sparse="yes"
   ;;
   --disable-sparse) sparse="no"
@@ -1764,6 +1769,7 @@ Advanced options (experts only):
   --with-pkgversion=VERS   use specified string as sub-version of the package
   --enable-debug           enable common debug build options
   --enable-sanitizers      enable default sanitizers
+  --enable-tsan            enable thread sanitizer
   --disable-strip          disable stripping binaries
   --disable-werror         disable compilation abort on warning
   --disable-stack-protector disable compiler-provided stack protection
@@ -6220,6 +6226,30 @@ if test "$fuzzing" = "yes" ; then
   fi
 fi
 
+# Thread sanitizer is, for now, much noisier than the other sanitizers;
+# keep it separate until that is not the case.
+if test "$tsan" = "yes" && test "$sanitizers" = "yes"; then
+  error_exit "TSAN is not supported with other sanitiziers."
+fi
+have_tsan=no
+have_tsan_iface_fiber=no
+if test "$tsan" = "yes" ; then
+  write_c_skeleton
+  if compile_prog "$CPU_CFLAGS -Werror -fsanitize=thread" "" ; then
+      have_tsan=yes
+  fi
+  cat > $TMPC << EOF
+#include <sanitizer/tsan_interface.h>
+int main(void) {
+  __tsan_create_fiber(0);
+  return 0;
+}
+EOF
+  if compile_prog "$CPU_CFLAGS -Werror -fsanitize=thread" "" ; then
+      have_tsan_iface_fiber=yes
+  fi
+fi
+
 ##########################################
 # check for libpmem
 
@@ -6377,6 +6407,16 @@ if test "$have_asan" = "yes"; then
            "Without code annotation, the report may be inferior."
   fi
 fi
+if test "$have_tsan" = "yes" ; then
+  if test "$have_tsan_iface_fiber" = "yes" ; then
+    QEMU_CFLAGS="-fsanitize=thread $QEMU_CFLAGS"
+    QEMU_LDFLAGS="-fsanitize=thread $QEMU_LDFLAGS"
+  else
+    error_exit "Cannot enable TSAN due to missing fiber annotation interface."
+  fi
+elif test "$tsan" = "yes" ; then
+  error_exit "Cannot enable TSAN due to missing sanitize thread interface."
+fi
 if test "$have_ubsan" = "yes"; then
   QEMU_CFLAGS="-fsanitize=undefined $QEMU_CFLAGS"
   QEMU_LDFLAGS="-fsanitize=undefined $QEMU_LDFLAGS"
@@ -6412,7 +6452,8 @@ if test "$werror" = "yes"; then
     QEMU_CFLAGS="-Werror $QEMU_CFLAGS"
 fi
 
-if test "$solaris" = "no" ; then
+# Exclude --warn-common with TSan to suppress warnings from the TSan libraries.
+if test "$solaris" = "no" && test "$tsan" = "no"; then
     if $ld --version 2>/dev/null | grep "GNU ld" >/dev/null 2>/dev/null ; then
         QEMU_LDFLAGS="-Wl,--warn-common $QEMU_LDFLAGS"
     fi
@@ -7476,6 +7517,10 @@ if test "$have_asan_iface_fiber" = "yes" ; then
     echo "CONFIG_ASAN_IFACE_FIBER=y" >> $config_host_mak
 fi
 
+if test "$have_tsan" = "yes" && test "$have_tsan_iface_fiber" = "yes" ; then
+    echo "CONFIG_TSAN=y" >> $config_host_mak
+fi
+
 if test "$has_environ" = "yes" ; then
   echo "CONFIG_HAS_ENVIRON=y" >> $config_host_mak
 fi

util/coroutine-ucontext.c

@@ -37,12 +37,19 @@
 #endif
 #endif
 
+#ifdef CONFIG_TSAN
+#include <sanitizer/tsan_interface.h>
+#endif
+
 typedef struct {
     Coroutine base;
     void *stack;
     size_t stack_size;
     sigjmp_buf env;
 
+    void *tsan_co_fiber;
+    void *tsan_caller_fiber;
+
 #ifdef CONFIG_VALGRIND_H
     unsigned int valgrind_stack_id;
 #endif
@@ -65,7 +72,18 @@ union cc_arg {
     int i[2];
 };
 
-static void finish_switch_fiber(void *fake_stack_save)
+/* QEMU_ALWAYS_INLINE only does so if __OPTIMIZE__, so we cannot use it. */
+static inline __attribute__((always_inline))
+void on_new_fiber(CoroutineUContext *co)
+{
+#ifdef CONFIG_TSAN
+    co->tsan_co_fiber = __tsan_create_fiber(0); /* flags: sync on switch */
+    co->tsan_caller_fiber = __tsan_get_current_fiber();
+#endif
+}
+
+static inline __attribute__((always_inline))
+void finish_switch_fiber(void *fake_stack_save)
 {
 #ifdef CONFIG_ASAN
     const void *bottom_old;
@@ -78,13 +96,30 @@ static void finish_switch_fiber(void *fake_stack_save)
         leader.stack_size = size_old;
     }
 #endif
+#ifdef CONFIG_TSAN
+    if (fake_stack_save) {
+        __tsan_release(fake_stack_save);
+        __tsan_switch_to_fiber(fake_stack_save, 0);  /* 0=synchronize */
+    }
+#endif
 }
 
-static void start_switch_fiber(void **fake_stack_save,
+static inline __attribute__((always_inline)) void start_switch_fiber(
-                               const void *bottom, size_t size)
+    CoroutineAction action, void **fake_stack_save,
+    const void *bottom, size_t size, void *new_fiber)
 {
 #ifdef CONFIG_ASAN
-    __sanitizer_start_switch_fiber(fake_stack_save, bottom, size);
+    __sanitizer_start_switch_fiber(
+            action == COROUTINE_TERMINATE ? NULL : fake_stack_save,
+            bottom, size);
+#endif
+#ifdef CONFIG_TSAN
+    void *curr_fiber =
+        __tsan_get_current_fiber();
+    __tsan_acquire(curr_fiber);
+
+    *fake_stack_save = curr_fiber;
+    __tsan_switch_to_fiber(new_fiber, 0);  /* 0=synchronize */
 #endif
 }
 
@@ -104,8 +139,12 @@ static void coroutine_trampoline(int i0, int i1)
 
     /* Initialize longjmp environment and switch back the caller */
     if (!sigsetjmp(self->env, 0)) {
-        start_switch_fiber(&fake_stack_save,
+        start_switch_fiber(
-                           leader.stack, leader.stack_size);
+            COROUTINE_YIELD,
+            &fake_stack_save,
+            leader.stack,
+            leader.stack_size,
+            self->tsan_caller_fiber);
         siglongjmp(*(sigjmp_buf *)co->entry_arg, 1);
     }
 
@@ -154,12 +193,16 @@ Coroutine *qemu_coroutine_new(void)
 
     arg.p = co;
 
+    on_new_fiber(co);
     makecontext(&uc, (void (*)(void))coroutine_trampoline,
                 2, arg.i[0], arg.i[1]);
 
     /* swapcontext() in, siglongjmp() back out */
     if (!sigsetjmp(old_env, 0)) {
-        start_switch_fiber(&fake_stack_save, co->stack, co->stack_size);
+        start_switch_fiber(
+            COROUTINE_YIELD,
+            &fake_stack_save,
+            co->stack, co->stack_size, co->tsan_co_fiber);
         swapcontext(&old_uc, &uc);
     }
 
@@ -216,8 +259,8 @@ qemu_coroutine_switch(Coroutine *from_, Coroutine *to_,
 
     ret = sigsetjmp(from->env, 0);
     if (ret == 0) {
-        start_switch_fiber(action == COROUTINE_TERMINATE ?
+        start_switch_fiber(action, &fake_stack_save,
-                           NULL : &fake_stack_save, to->stack, to->stack_size);
+                           to->stack, to->stack_size, to->tsan_co_fiber);
         siglongjmp(to->env, action);
     }
 
@@ -231,6 +274,11 @@ Coroutine *qemu_coroutine_self(void)
     if (!current) {
         current = &leader.base;
     }
+#ifdef CONFIG_TSAN
+    if (!leader.tsan_co_fiber) {
+        leader.tsan_co_fiber = __tsan_get_current_fiber();
+    }
+#endif
     return current;
 }