Comment 17 for bug 1814874

I don't recall seeing a soft lockup at all, not a related one anyway. Sadly I don't have a specific command but I did find it triggered when:

1. Memory is overcommited. It doesn't have to be heavily overcommitted.
2. I/O is very high.
3. I “leave“ an SSH session, releasing a small amount of memory.

The patch seems to suggest that it is this release that causes the problem. My guess would be that part of the session is already in the compressed cache, and it is released while another process attempts to enter the compressed cache. I don't even know if that makes sense.

To add to the confusion: it crashes fairly consistently across VMs and bare metal, but occasionally a server under the same sort of stress, and otherwise identical setup, can run weeks or months without issue.

At this point, though, I know of no specific command to trigger it. If I get time I can have a play around but that may never happen.

I usually have zswap.max_pool_percent=50, but everything else can vary and the result is the same.