Maybe the root process exiting is not sufficient for content_browser to clean up NaCl.

Ken, do you think the root process should send some other kind of signal to content_browser to tell it to terminate? Or maybe something needs to be wired into content_browser's OnStop() (???) to trigger a clean shutdown?

This complicates analysis of other test failures, so I'll take a look now...

No repro with ToT chrome r458750.

I wonder if I see this in tests because other processes are crashing and that causes hangs.

I cannot reproduce this any more.

Note that it is "normal" to see signal 9 from session_manager to chrome in autotests, because session_manager wants to kill chrome quickly and not wait for shutdown. This is not a hang:

2017-03-23T11:00:21.955803-07:00 INFO session_manager[4372]: [INFO:browser_job.cc(149)] Terminating process group: Restarting browser on-demand.
2017-03-23T11:00:21.955866-07:00 INFO session_manager[4372]: [INFO:system_utils_impl.cc(110)] Sending 9 to -4434 as 1000
2017-03-23T11:00:21.958390-07:00 INFO session_manager[4372]: [INFO:child_exit_handler.cc(77)] Handling 4434 exit.
2017-03-23T11:00:21.959511-07:00 ERR session_manager[4372]: [ERROR:child_exit_handler.cc(85)]   Exited with signal 9

See login_manager SessionManagerImpl::EnableChromeTesting and SessionManagerService::RestartBrowserWithArgs

void SessionManagerService::RestartBrowserWithArgs(
    const std::vector<std::string>& args,
    bool args_are_extra) {
  // Waiting for Chrome to shutdown takes too much time.
  // We're killing it immediately hoping that data Chrome uses before
  // logging in is not corrupted.
  // TODO(avayvod): Remove RestartJob when crosbug.com/6924 is fixed.
  browser_->KillEverything(SIGKILL, "Restarting browser on-demand.");
  if (args_are_extra)
    browser_->SetExtraArguments(args);
  else
    browser_->SetArguments(args);
  // The browser will be restarted in HandleExit().
}