4x card in 8x AGP KT600 = locked solid (AGP bug?)

4x card in 8x AGP KT600 = locked solid (AGP bug?)

[Ian McConnell]

When I try to use DRI with a 4x AGP Radeon 9100 card in my 8x AGP KT600
motherboard, the screen goes black and the machine locks up solid - only a
hard reset reboots the machine. I can't find any error messages or Oops
logged.

With the latest BIOS and drivers, the card plays 3D games under Win2k ok.
However with debian testing, stock kernel 2.6.5 and XFree86 4.3.0.1, the
machine hangs unless I disable DRI in /etc/X11/XF86Config-4 with 
   Option          "ForcePCIMode"          "on"
then X works fine (using software GL)

I wasn't sure if this was a DRI or AGP bug, so I downloaded and installed
ATI's own 3d drivers (fglrx-4.3.0-3.7.6.i386.rpm). This time the screen goes
black and the X server sits burning 100% CPU, but I can log in remotely. I
cannot kill the X server (with kill -9 or ctrl-alt-backspace) and 
strace -p<xserver pid> just hangs with no output.

There is a thread of my experiences and more detail at
      http://www.rage3d.com/board/showthread.php?s=&threadid=33756033


So two difference implementations of DRI hanging makes me suspect that there
is a bug with AGP (Also the same video card, kernel and X worked well with
an older 4xAGP KT133 motherboard)

I'm using a stock kernel-2.6.5 with an AthlonXP 2700 and modules:
  via_agp                 5824  1 
  agpgart                28072  2 via_agp
  radeon                115184  2 

which show
  Linux agpgart interface v0.100 (c) Dave Jones
  agpgart: Detected VIA KT400/KT400A/KT600 chipset
  agpgart: Maximum main memory to use for agp memory: 439M
  agpgart: AGP aperture is 64M @ 0xe8000000
  [drm] Initialized radeon 1.9.0 20020828 on minor 0

The card is a Sapphire Radeon 9100 which supports up 4x AGP and the bios
R1.04 (dated Jan 20 2004) settings are:
  AGP Aperture 64M
  AGP Mode 4x
  AGP Driving control Auto
  AGP Fastwrite Enable
  AGP Master 1 WS Write Disable
  AGP Master 1 WS Read Disable
Changing the BIOS settings doesn't make any noticeable difference.

I've tried kernel-2.4.25, but that fails loading agpgart.o with
  Linux agpgart interface v0.99 (c) Jeff Hartmann
  agpgart: Maximum main memory to use for agp memory: 439M
  agpgart: Detected Via Apollo Pro KT400 chipset
  agpgart: unable to determine aperture size.



Is there any way to test AGP without using X? Any suggestions as to how I
can track down where/what is freezing the machine? How compatible is the
KT600 running in an AGP v2 compatibility mode?

Re: 4x card in 8x AGP KT600 = locked solid (AGP bug?)

[Ian McConnell]

Ian McConnell <kernel@emit.demon.co.uk> writes:

> So two difference implementations of DRI hanging makes me suspect that there
> is a bug with AGP (Also the same video card, kernel and X worked well with
> an older 4xAGP KT133 motherboard)

Looks like I was wrong about it being an AGP bug. I found testgart and 
everything works fine:

# testgart 
agpgart: Found an AGP 3.5 compliant device at 0000:00:00.0.
agpgart: Device is in legacy mode, falling back to 2.x
agpgart: Putting AGP V2 device at 0000:00:00.0 into 4x mode
agpgart: Putting AGP V2 device at 0000:01:00.0 into 4x mode
version: 0.100
bridge id: 0x31891106
agp_mode: 0x1f000a17
aper_base: 0xe8000000
aper_size: 64
pg_total: 112384
pg_system: 112384
pg_used: 0
entry.key : 0
entry.key : 1
Allocated 8 megs of GART memory
MemoryBenchmark: 1173 mb/s
MemoryBenchmark: 1766 mb/s
MemoryBenchmark: 1772 mb/s
Average speed: 1570 mb/s
Testing data integrity (1st pass): passed on first pass.
Testing data integrity (2nd pass): passed on second pass.