While working on an autotest for --mash I added an intentional CHECK(false) into services/ui/ws/window_server.cc in WindowServer::OnGpuServiceInitialized().

When running chrome --mash this causes the window server to go into a crash loop. On desktop this isn't so bad, but under tests on device this causes core files to be generated continuously for 2 minutes until the test times out. This makes /var/spool/crash huge and slows down test completion tremendously (as it rsyncs all the core files somewhere).

Is there some way we could stop trying to restart the window server if it crashes multiple times on startup? Is there explicit restart code somewhere or is this a side effect of other processes trying to connect to the window server service?