Test toleration of transient errors in "rr replay -a"

CMakeLists.txt

@@ -1704,6 +1704,7 @@ set(TESTS_WITHOUT_PROGRAM
   trace_version
   term_trace_cpu
   trace_events
+  transient_fault_replay_all
   tty
   unmap_vdso
   unwind_on_signal

src/ReplaySession.cc

@@ -2160,6 +2160,11 @@ void ReplaySession::reattach_tasks(ScopedFd new_tracee_socket, ScopedFd new_trac
   }
 }
 
+bool ReplaySession::mark_stdio() const {
+  return Session::mark_stdio() &&
+    current_frame_time() >= suppress_stdio_before_event_;
+}
+
 bool ReplaySession::echo_stdio() const {
   return flags().redirect_stdio &&
     current_frame_time() >= suppress_stdio_before_event_;

src/ReplaySession.h

@@ -361,6 +361,7 @@ class ReplaySession final : public Session {
   void notify_detected_transient_error() { detected_transient_error_ = true; }
 
   void set_suppress_stdio_before_event(FrameTime event) { suppress_stdio_before_event_ = event; }
+  bool mark_stdio() const override;
   bool echo_stdio() const;
 
 private:

src/Session.h

@@ -425,7 +425,7 @@ class Session {
      or replay. */
   bool intel_pt_enabled() const { return intel_pt_; }
 
-  bool mark_stdio() const;
+  virtual bool mark_stdio() const;
 
 protected:
   Session();

src/Task.cc

@@ -2139,6 +2139,26 @@ void Task::set_aarch64_tls_register(uintptr_t val) {
      we tried to set. */
 }
 
+static FrameTime simulate_error_at_event() {
+  const char* s = getenv("RR_SIMULATE_ERROR_AT_EVENT");
+  if (s) {
+    return atoi(s);
+  }
+  return INT64_MAX;
+}
+
+static bool simulate_transient_error(Task* t) {
+  static bool simulated_error = false;
+  static FrameTime simulate_error_at_event_ = simulate_error_at_event();
+
+  if (simulated_error || !t->session().is_replaying() ||
+      static_cast<ReplayTask*>(t)->session().trace_stream()->time() < simulate_error_at_event_) {
+    return false;
+  }
+  simulated_error = true;
+  return true;
+}
+
 bool Task::did_waitpid(WaitStatus status) {
   if (is_detached_proxy() &&
       (status.stop_sig() == SIGSTOP || status.stop_sig() == SIGCONT)) {
@@ -2275,6 +2295,9 @@ bool Task::did_waitpid(WaitStatus status) {
         in_unexpected_exit = true;
         return false;
       }
+      if (simulate_transient_error(this)) {
+        error_state = PerfCounters::Error::Transient;
+      }
       if (detect_transient_error &&
         *detect_transient_error == PerfCounters::Error::Transient) {
         session().as_replay()->notify_detected_transient_error();

src/test/transient_fault_replay_all.run

@@ -0,0 +1,8 @@
+source `dirname $0`/util.sh
+
+RECORD_ARGS=-M
+just_record seq "1 100"
+RR_SIMULATE_ERROR_AT_EVENT=300 replay -M
+check 100 || exit 1
+RR_SIMULATE_ERROR_AT_EVENT=301 replay -M
+check 100