Repository navigation
[FEATURE] Support reboot cycles/stress-tests #1038
Description
While debugging on MTL_RVP_HDA, I encountered multiple cases of stuck reboots, which can only be solved with a power-cycle.
It's my understanding that we do not test reboot on our CI and daily test devices, so we don't know what we don't know.
I am not sure if this can be implemented with sof-test 'run-all-test', this may require an external device to send the reboot commands and check what the state is. Or this may be a script in sof-test that executes on a host device, with commands sent over ssh to the DUT?
Comments and ideas welcome @keqiaozhang @marc-hb @fredoh9 @ranj063
cc:
Activity
marc-hb commentedon May 19, 2023
Testing a device that does up and down is not easy, it requires relatively smart outside control and sof-test is not that because sof-test runs on the device.
In fact we already had some discussions about this in internal issue 272. They were focused on suspend/resume which is easier but not 100% reliable either. So I think the next step is to extract from 272 information and pointers relevant to this.
EDIT, not sure what this
check-reboot.shdoes:36dac39
plbossart commentedon May 19, 2023
It's not easy, that's a given. But I am really worried about the cost of doing nothing. probably time to think about the scripts and whether the premise that they all run locally is correct. we could have parts of sof-test that are dedicated to remote control, instead of having this in some random CI hidden repo.