Comment 7 for bug 1846237

Manoj,

This is a soft lockup in schedule() - a task is waiting I/O completion - inside a KVM guest.

The guest kernel has a completion that will be awaken, re-scheduling the task back to a CPU run queue, as soon as the I/O is finished (the I/O usually contains a handle that, as soon as I/O is confirmed to be sent by transport layer, calls a callback function to "release" the completion and let the application to be re-scheduled. Depending on how the cache is configured, this logic MIGHT also check for I/O being committed in I/O server, only allowing the task to continue its logic after that: That is also considered as a soft lockup (tasks keeps re-scheduling itself until the I/o is done).

The guest has just panic'ed because it had "panic on hung" configured.

It is highly probable that the "issue" here is I/O contention causing the lockup inside the Guest, nothing else. There isn't any I/O timeouts - because bad transport and/or block device layer - or any other hard lockup due to a dead lock, for example.

So, unless something else, undocumented in this bug, is happening, there is not much to be done without more information. To help kernel team, it would be good for IBM to provide more information on what was being done on the HOST, how the I/O devices are configured in KVM guest, etc.

Thats my 5 cents.