There is another thread, started by Richard Hansen, about a somewhat separate issue. There seem to be multiple bugs in handling running out of memory (KVA more than pages??). You might try Richard's patch to the pagedaemon. I am guessing that there's a bug in swap-on-cgd (that KASSSERTed) and a separate bug in swap-on-raidframe. I would suggests reading the code around the KASSERT that fired, and see if you can understand what went wrong, or if you can add an earlier KASSERT. I suspect your add-regular-add-raidframe-remove-regular case will fail eventually; it would be odd if that were somehow different. I wonder if the code to handle paging out is running out of kva. Do you have an elevated maxvnodes?
Attachment:
pgpvbkiV5j_Lh.pgp
Description: PGP signature