Just wanted to report issue I encountered - messages look like:

nvme nvme0: controller is down; will reset: CSTS=0xffffffff, PCI_STATUS=0x10
nvme0n1: detected capacity change from 7814037168 to 0

btrfs goes then to read only. If you have started early-shell
(systemd.debug-shell=1) you can check dmesg what happened switching to
ctrl+alt+F9.

I added "nvme_core.default_ps_max_latency_us=0" and runs now for 3 weeks
without any issues.

Found here:
https://lore.kernel.org/lkml/YnNeTsSzFJqEK%2Fs+@kbusch-mbp.dhcp.thefacebook.com/T/