Xorg 6.9.0, FreeBSD-6.1 and a Radeon or two

[Matt Dawson]

Hi all,
	I suspect this has been done to death. RingTFM and Google, however, seem not 
to be my friends at the moment. I have the following R300 cards:

Radeon 9800 non pro, 128 bit bus, eight pipelines, 256MB memory;
Radeon 9700 non pro, used to be a 9500 with four pipes disabled but the 
hardware mod (I have SMT rework kit) and a BIOS flash soon put a stop to that 
nonsense, now eight pipes, 256bit bus, 128MB memory.

The latter was tried on xorg-6.9.0 (xorg-server from ports with stock dri) and 
locked hard before the modification. The same result when modified, although 
it did pick up the change of ID from the new BIOS and reported the card as a 
9700 with an R300 core in dmesg. The 9800, unsurprisingly, locks hard as 
well, but I expected this from others' experiences with the 9800, like the 
note on Eric Anholt's homepage.

Incidentally, the modded 9500 has been tested on an evil box (I have to keep 
one in the workshop for some things, such as flashing firmware to things and 
for my EPROM burner), and works flawlessly with the official catalyst 6.4 
drivers, giving me a 3DMark 2001 score of 12000+ and a DroneZmark score of 
~190, just in case anyone thinks the card itself is buggy. All eight pipes 
work and there are no chequerboard artifacts and ATITool passes my BIOS set 
clocks (core underclocked slightly to sync with memory clock) with no 
artefacts.

My request is really this: Does anyone have a definitive list of what works 
and what doesn't with Radeons i.e. xorg ports, patches etc.? My everyday 
boxen use 9200s, so theres no problem there. DRM in the kernel seems to 
attach and detect the hardware correctly with the later cards, so I doubt 
that is a problem either. With the 9700 on my main AMD64 box, it managed to 
play briquolo for a minute and a half before it locked solid (three figure 
framerate, too). 2D is never a problem, it just seems like Radeons >9200 
still do not support 3D at all, contrary to what I have seen lately on 
mailing lists and Google, which seems to point at the 9800 being the card 
that hardlocks, with true R300 cores working correctly to a point. Unless, of 
course, I'm doing something stupid and the 6.9.0 ports are not the ones to 
use. The very fact that it worked at all makes me hopeful it is just a silly 
mistake on my part.

Advice? Flames? Anything at all to get the correct information, even if it 
means not wasting my time with the R300s until a later date, would be very 
much appreciated. If logs are required, I can hook the 9700 back up in this 
box and get them, although there was nothing glaringly obvious in dmesg (agp 
and drm devices detected and attached correctly, info: [drm] Loading R300 
Microcode displayed as X starts) /var/log/Xorg.0.log (no EE lines, the usual 
WW R300 code is very beta etc.) or glxinfo | grep -i render (direct 
rendering: Yes, OpenGL renderer string: Mesa DRI R300 20041207 AGP 8x TCL and 
the usual R300 warn once message).
-- 
Matt Dawson.

matt at mattsnetwork.co.uk
MTD15-RIPE OpenNIC M_D9
MD51-6BONE