Merging 64 bit changes to -HEAD - part 2

[Juli Mallett]

[ Picking the most recent message to reply to ]

On Thu, Jun 17, 2010 at 10:05, M. Warner Losh <imp at bsdimp.com> wrote:
> It was also a name-space collision, so we were using PG_x instead of
> PG_y in the PTE code due to the overlap.  Maybe it all works now, but
> that was the motivation for the change.

The header and source I checked in to my branch date from my original
work on FreeBSD for MIPS and I decided to use the existing spellings
in those files rather than changing them to use the ones in the tree
because using PG_ is fairly traditional and normal.  I don't have an
objection to PTE_ as such.  JC is just trying to reduce diffs with my
branch, but I don't care what the prefix on the name is as long as
they're consistent and correct.

I'm torn on the matter of cache attributes.  I'd like for us to use
consistent spellings for the cache coherency bits and to lump them all
together in naming, which is why I like PG_C_UC and PG_C_CNC.  The
latter bit is defined in the various user's guides I've read as always
meaning cacheable non-coherent.  Is that same bit used on CPUs we care
about to mean cacheable coherent?  I don't really care, I suppose, as
long as the bits are grouped (e.g. PG_C_*) and consistent between
XKPHYS and PTE names.  I'd prefer for us to not have cases where the
same PTE with the same bits set are coherent on some CPUs and not
coherent on others, but it sounds like that may be beyond our control.

Thanks,
Juli.