Re: CVS commit: src/sys/arch

[Taylor R Campbell <campbell+netbsd-source-changes-d%mumble.net@localhost>]

   Date: Fri, 25 Jul 2014 10:23:31 +1000
   from: matthew green <mrg%eterna.com.au@localhost>

   "Taylor R Campbell" writes:
   > Add a FIRST1G page freelist to x86, for old graphics devices.

   other devices that do this didn't have to change the
   platform code to do it.  see bce(4), which is limited
   to 1GB of ram on laptops that have 2GB.

Yes...but those can just use bus_dmamem_alloc, which in turn can use
uvm_pglistalloc.  The graphics devices need to constrain the pages
allocated for uvm objects, which use uvm_pagealloc.  I proposed a
patch to allow configuring uao to allocate pages with uvm_pglistalloc
instead:

https://mail-index.netbsd.org/tech-kern/2014/05/20/msg017092.html

matt@ objected and suggested using page freelists instead, so I did
that.  Nobody on port-x86 objected to the page freelist code I added:

https://mail-index.netbsd.org/port-amd64/2014/05/21/msg002062.html
https://mail-index.netbsd.org/port-i386/2014/05/21/msg003277.html

So this week I asked about making the mechanism for choosing a
suitable page freelist a MI interface:

https://mail-index.netbsd.org/tech-kern/2014/07/22/msg017391.html

I would like to nail this down pronto so that we have a chance to pull
up graphics fixes to netbsd-7 without having to bring in whole new
bus_dma gizmos.  I don't care how it works -- I just want some MI way
to demand that the pages allocated for a uao be fit for bus_dma,
whether it be by page freelists or uvm_pglistalloc or something else
altogether.