Frequent lockups after 2.6.19 upgrade (or so I believe)
 

Hi,
I had the "screen frozen, pointer moves" bug in the past, which appeared and disappeared as the driver was updated. I generally found some stability after setting AGP4X on BIOS. The RenderAccel option didn't seem to affect.

Now, after I upgraded to kernel 2.6.19 (at least, I think that is what changed the situation) I had really frequent lockups one every 1 or two days.

I have a P4S800-MX motherboard and an NVIDIA GeForce4 MX440 Agp8X. It works perfect on windows.
I just suspect of my motherboard from long time ago since the 3.3 line is always at 2.99 or 2.98. I don't know if this means I have a weak PSU or a bad mobo (or neither).

But, this time, I think this is clearly an NVIDIA bug (or at least, a kernel bug), since it hits NULL dereference, causing it to oops. I've got this information because some of the lockups only lock X but the machine seems to be in control, since I can CTRL+ALT+DEL and reboot normally. Other times it is a hard lockup which causes a general freeze (I've recently compiled SysRq key support into the kernel, to save myself from this situation also, since I imagine that I can just kill X and get control back).

I post the bug report and the kernel.log that informs about the kernel oops.

BTW, I'm a little tired about this card (both in terms of performance and stability) so I think I'm going to buy a newer AGP card (if I still trust the mobo) or a new mobo with PCI-E and a PCI-E card. Probably a new PSU also.

Thanks

Re: Frequent lockups after 2.6.19 upgrade (or so I believe)
 

Please, someone from NVIDIA, could you have a look at this? I got the exact same kernel oops again, this time with Xid 13's.
It really looks like a software error (not necesarilly on nvidia's module, but I think a NULL dereference must be a software error somewhere).

Re: Frequent lockups after 2.6.19 upgrade (or so I believe)
 

I have a few questions:
0) How can this problem be reproduced?
1) If you set NvAGP to 0 in xorg.conf, does that have any impact?
2) Have you verified that you're using the latest motherboard BIOS?
3) If you set RenderAccel to false in xorg.conf, does that have any impact?

Thanks,
Lonni

Re: Frequent lockups after 2.6.19 upgrade (or so I believe)
 

It isn't very deterministic. It triggers while I'm using my X server normally. I think both times it got triggered while using my X terminal mrxvt. In any case, it seems to be a bug that appears after a day and a half of uptime (during which the X server needs to be running, If I shutdown every day it doesn't seem to trigger).

I haven't tried yet. I just tried switching from AGPgart to NvAGP and it happened also. I'm currently running with NvAGP 0. I will post an update in a few days if all goes well.

I am, but the last update to the BIOS is quite old and ASUS just added support for newer processors.
I was getting a few ACPI and APIC errors, so I tried playing with that. I tried booting with noapic and setting interrupt mode from APIC to PIC (and that is visible now from /proc/interrupts). This changes were made before the last freeze, so they didn't affect anything.

As far as I can tell, no. If it freezes again I'll disable that and try.


BTW, does the kernel oops report that I've attached help to locate where the NULL dereference is being made?

Thank you for the response,
Matt.

Re: Frequent lockups after 2.6.19 upgrade (or so I believe)
 

Well, just disabling AGP (NvAGP 0 and blacklisting agpgart, just in case) solved the stability problems.
Does this mean that the board will use system memory? I never understood what "disabling AGP" meant (I would imagine this to disable the card completely, but that doesn't seem to be the case).

Does this mean this card has problems with AGP? Or my mother?

Re: Frequent lockups after 2.6.19 upgrade (or so I believe)
 

Disabling AGP means your 3D performance is gone.

Re: Frequent lockups after 2.6.19 upgrade (or so I believe)
 

No, disabling AGP does not mean that 3D performance is gone. It means that the driver is falling back to PCI bus speed instead of an AGP rate.