In the veyron_minnie-release build at http://stainless/browse/chromeos-autotest-results/221311356-chromeos-test/, the ui.ChromeLogin Tast test failed with an error that I haven't seen before: "Navigating to http://127.0.0.1:35570 failed: cdp.Page: DOMContentEventFired Recv: rpcc: the connection is closing".

It looks like Chrome's GPU process segfaulted, resulting in a SIGABRT in the browser process:

...
[6374:6374:0727/145144.629122:VERBOSE1:user_session_manager.cc(1960)] Launching browser...
[6374:6374:0727/145144.837050:ERROR:login_display_host_mojo.cc(233)] Not implemented reached in virtual void chromeos::LoginDisplayHostMojo::OnBrowserCreated()
[6374:6374:0727/145144.896838:VERBOSE1:input_events_blocker.cc(29)] InputEventsBlocker 0xff6f520 destroyed.
[6374:6374:0727/145144.913508:ERROR:CONSOLE(217)] "Unexpected condition on chrome://oobe/gaia-signin: Could not find value for languageList", source: chrome://resources/js/load_time_data.js (217)
[6374:6374:0727/145145.357993:VERBOSE1:connection_holder.h(384)] Instance arc.mojom.LockScreenInstance not available.
[6374:6374:0727/145146.017609:VERBOSE1:arc_session_impl.cc(530)] Notified that ARC instance is stopped 1
[6374:6374:0727/145146.017702:INFO:arc_session_runner.cc(264)] ARC stopped: SHUTDOWN
[6374:6479:0727/145154.378593:WARNING:crash_handler_host_linux.cc(334)] Could not translate tid, attempt = 1 retry ...
[6374:6479:0727/145158.487911:FATAL:gpu_data_manager_impl_private.cc(855)] GPU process isn't usable. Goodbye.
[7306:7306:0727/145159.706435:ERROR:broker_posix.cc(43)] Invalid node channel message
<EOF>

Unfortunately, the symbolized stacks that I've looked at appear to be garbage -- every thread has a frame like this in it, for instance:

  chrome!__sort4<bool (*&)(const blink::GridTrack *, const blink::GridTrack *), blink::GridTrack **> + 0x7f27c1

I think that the broken stack may be due to issue 851851, which also references the "GPU process isn't usable. Goodbye." error: "All crashes caused by LOG(FATAL) or any variant thereof will end up with garbage stack traces due to this issue, and may be attributed to BreakDebugger or to other random things found via stack scanning."

Now that that's fixed, hopefully we'll get better stacks soon for the underlying GPU issue the next time this happens.