On Sun, Oct 31, 2010 at 02:27:33PM +0900, Masao Uebayashi wrote: (snip) > > > > - as you guessed in your mail, I don't like the new PQ_FIXED flag > > > > either. > > > > it looks like it's only used for preventing the XIP pages from > > > > being put on the paging queues, but that would be better done by > > > > just initializing pg->wire_count to 1 so that UVM treats them > > > > as permanently non-pageable (which they actually are). > > > > > > Are you OK to prevent XIP pages from being loaned? ;) > > > > > > The actual problems here are > > > > > > a) the generic fault handler doesn't know how to handle pages that > > > never be paged out (device pages), and (snip)
I'll revert the PQ_FIXED part. Full vm_page is allocated for the time being.