mtrr setup?

[kwall@kurtwerks.com]

Feigning erudition, David A. Bandel wrote:
% -----BEGIN PGP SIGNED MESSAGE-----
% Hash: SHA1
% 
% On Mon, 6 Jan 2003 10:56:37 -0500
% begin  Douglas J Hunley <doug at hunley.homeip.net> spewed forth:
% 
% > -----BEGIN PGP SIGNED MESSAGE-----
% > Hash: SHA1
% > 
% > David A. Bandel spewed electrons into the ether that resembled:
% > > What's in your /proc/mtrr file anyway and are you sure you can do
% > > better?
% > 
% > reg00: base=0x00000000 ( 0MB), size=2048MB; write-back, count=1
% 
% you have 2Gb of memory?
% 
% > 
% > I'm pretty sure that is not optimal
% 
% reg00: base=0x00000000 (   0MB), size= 256MB: write-back, count=1
% reg01: base=0xe4000000 (3648MB), size=  64MB: write-combining, count=1
% reg02: base=0xdc000000 (3520MB), size=   4MB: write-combining, count=1

Mine is comparable:

reg00: base=0x00000000 (   0MB), size= 256MB: write-back, count=1
reg01: base=0xf2000000 (3872MB), size=  32MB: write-combining, count=3
reg02: base=0xf8000000 (3968MB), size=  64MB: write-combining, count=1

% IIRC, you can only have one write-back entry, the rest will be
% write-combining.  Do you have other entries?  Those others are the ones
% you should worry about.

>From the the Documentation/mtrr.txt, it appears that write-combining
is preferable because writes are bunched together before being shoved
across the bus (PCI or AGP) -- very nice for the video adapter.

My framebuffer lives at at 0xf2000000, as /var/log/XFree86.0.log says:

(--) MGA(0): Linear framebuffer at 0xF2000000

This matches up nicely with the reg01 entry, so, clearly, the X servers
is in control of that MTRR. As far as I can tell, though, there's very
little optimization that can be done manually. 

Kurt
-- 
Comparing information and knowledge is like asking whether the fatness
of a pig is more or less green than the designated hitter rule."
		-- David Guaspari