Checked both logs. This is a Class 4 failure — DOE mailbox stuck with PCIe link collapsed to x0 on 000f:01:00.0. Zero bandwidth available, BERT error record from the previous session confirms the EC left the mailbox in a failed state.
Cold power cycle is the only recovery — you already know this works. If it keeps recurring, run sudo rasdaemon --errors after the next occurrence to identify what’s triggering the BERT reset.