GL apps lock up system w/4349

[mmastrac]

Just installed the latest 4349 drivers. 2D works great, X starts perfectly. Running any OpenGL app locks the system hard - no /var/log/messages entry, no kernel panic, etc. Just a hard lock.

I have done the following:
- checked that the GL libs are correctly installed
- checked the perms (666) on the /dev/nv* devices
- dropped AGP to 2x

I'm running a custom 2.4.20 kernel. Is there a kernel config option I've missed?

lspci -v:

01:00.0 VGA compatible controller: nVidia Corporation NV25 [GeForce4 Ti4400] (rev a3) (prog-if 00 [VGA])
Subsystem: Micro-star International Co Ltd: Unknown device 8710
Flags: bus master, 66Mhz, medium devsel, latency 248, IRQ 12
Memory at e8000000 (32-bit, non-prefetchable) [size=16M]
Memory at d8000000 (32-bit, prefetchable) [size=128M]
Memory at e0000000 (32-bit, prefetchable) [size=512K]
Expansion ROM at <unassigned> [disabled] [size=128K]
Capabilities: [60] Power Management version 2
Capabilities: [44] AGP version 2.0

/proc entries:

[matthew@localhost driver]$ cat nvidia/agp/card
Fast Writes: Supported
SBA: Supported
AGP Rates: 4x 2x 1x
Registers: 0x1f000217:0x1f000102

[matthew@localhost nvidia]$ cat agp/status
Status: Enabled
Driver: AGPGART
AGP Rate: 2x
Fast Writes: Disabled
SBA: Disabled

[matthew@localhost nvidia]$ cat agp/host-bridge
Host Bridge: VIA Technologies, Inc. VT8363/8365 [KT133/KM133]
Fast Writes: Not Supported
SBA: Supported
AGP Rates: 2x 1x
Registers: 0x1f000203:0x00000102

[matthew@localhost nvidia]$ cat version
NVRM version: NVIDIA Linux x86 nvidia.o Kernel Module 1.0-4349 Thu Mar 27 19:00:02 PST 2003
GCC version: gcc version 3.2 20020903 (Red Hat Linux 8.0 3.2-7)

[matthew@localhost nvidia]$ cat cards/0
Model: GeForce4 Ti 4400
IRQ: 12
Video BIOS: 04.25.00.22.37
Card Type: AGP

[mmastrac]

Here's the XF86 log file - nothing too excited in there.

[Andy Mecham]

Do the lockups happen with AGP disabled?

--andy

[mmastrac]

Works fine with NvAgp = 0.
I think that the NVidia AGP driver is causing the instability - it only started happening when I added 512MB to my existing 512MB RAM. When only 512MB were in the system (either my old or the new ram), everything worked solid.

I'm running a HIGHMEM enabled kernel:

[matthew@localhost linux-2.4.20]$ cat .config | grep "HIGHMEM"
# CONFIG_NOHIGHMEM is not set
CONFIG_HIGHMEM4G=y
# CONFIG_HIGHMEM64G is not set
CONFIG_HIGHMEM=y
# CONFIG_DEBUG_HIGHMEM is not set


Are there any known bugs with the NVidia AGP driver and >= 1GB memory?

[Andy Mecham]

No, but you should be using HIMEM.

There have been several reports of the KT133 being unstable with AGP; that might be the issue. Have you tried AGPGART?

--andy

[mmastrac]

Tried AGPGART this morning - still locks up. It usually manages to draw the backgrounds of the gdm login windows, but doesn't actually draw any pixmaps.

After looking at dmesg, it appears that agpgart is capable of grabbing a huge chunk of memory:

agpgart: Maximum main memory to use for agp memory: 941M
agpgart: Detected Via Apollo Pro KT133 chipset
agpgart: AGP aperture is 128M @ 0xd0000000

Unfortunately, there are no ways to limit this by module parameters. I might try to patch the source to limit it to a certain amount.

I'm going to try setting the AGP drive strength to EA in the BIOS tonight. Fast writes are already disabled. Any other recommended AGP bios settings?

I'm willing to work with the nVidia developers to debug this (even manually patching kernel source/running test code), if it turns out to be a driver bug, or something that can be worked around. Is this something that they will do?

[bwkaz]

You can try using a smaller AGP aperture setting in your BIOS. That'll limit the 128M number -- mine's set at 64MB (IIRC, that is), and working fine.

[mmastrac]

I was wondering whether it would be helpful to limit the maximum addressable main memory for AGP, rather than the aperture size. Since my system becomes unstable at > 768MB, I was wondering if limiting the amount of main memory potentially available would help with stability.

I've also discovered the following page that has some tips on BIOS settings:

http://www.tech-report.com/reviews/2000q3/abit-ka7/index5.x

Just need to figure out if I have a BIOS setting to control I/O voltage. :)

[mmastrac]

No such luck with any BIOS settings/agpgart tweaks/etc. AGPGART locks up using OpenGL and NVAGP reboots the system.

I'm going to try compiling 2.5.66 and seeing if the new AGPGART stuff makes things more stable.

[mmastrac]

Replaced RAM - no luck. Instant lockup on anything OpenGL-related.
Dropped to 768MB - works perfectly. AGP 4x even runs fine.

It seems to be an interaction between something and 1GB of RAM. I don't know which piece of the system is causing this issue, however.

[Andy Mecham]

Do you have a card that needs the bttv driver in your system?

--andy

[mmastrac]

I have a bt848-based TV tuner card. Is there an issue with this chipset and my setup?

[mmastrac]

Here is my complete lspci listing

[mmastrac]

And here is the mtrr listing, just in case:

[mmastrac]

Is there an interaction between the bt848 chip and the nvidia drivers?

[Andy Mecham]

No, not really. In the interest of problem isolation, does disabling the bttv driver return stability to your system? Does AGP work at that point?

--andy

[mmastrac]

I'll test and get back to you.