Currently, the test job must simply timeout and fail, so the test will end. LAVA spends a lot of time parsing the output of the DUT - if that output stops, LAVA cannot assume that the kernel has crashed and not all kernel crashes produce recognisable messages, so the only option is a timeout.