pgcl-2.5.65-1
(1) PCI DMA fix (zwane probably did it and I accidentally dropped it later)
(2) brute-force merge to 2.5.65
Given the brutality of the merge I wouldn't be surprised by a few bugs.
The stronger antifragmentation heuristics are still being debugged. I wouldn't expect very good performance until they're sorted out.
Also, badari's working on a fix for q->max_sectors*512 < PAGE_SIZE for fs/mpage.c and possibly others. If anyone besides me and badari are interested in that, either he or I will send the patch out for review.
Tested on 48GB NUMA-Q only (successfully so). No testing elsewhere yet.
Non-incremental vs. virgin 2.5.65.
As usual, available from ftp://ftp.kernel.org/pub/linux/kernel/people/wli/vm/pgcl/
-- wli