Lockups / lag on F8 / AMD64

[Doc Ok]

Hi guys,

I'm running F8 with x86_64 kernel version 2.6.25.14-69.fc8 on an AMD Athlon 64X2 4800+.

I'm wondering if anybody else is experiencing similar trouble: when I boot the kernel with the default flags, the systems locks up hard after a while of using it intensively. Especially on 3D graphics and video playback. Only recourse is power cycling; doesn't even seem to reply to ping.

I dug around some, and if I boot with kernel options nohz=off highres=off it does not lock up (or at least so rarely that is hasn't happened yet), but after a while of intensive use I get strange "lag," for lack of a better word. For example, when I open a terminal and press a key, it takes about 1/2 second for the character to show up in the terminal. If I press two keys in succession, both show up relatively quickly, but not as fast as usual. When I run "top", it takes about 30 seconds for the initial process list to show up, and then the list only updates every 30 seconds or so. Normally, it updates every 5 seconds.

I somehow get the feeling I'm losing timer interrupts (but there is no indication in dmesg or /var/log/messages), or that the timer chip is misprogrammed. I tried googling for these symptoms, but had no luck.

Anybody have any idea what this could be? I'm running F8 on a couple of other machines as well, from P4 to Core2Quad, and none of them have the same problem. Known (not to me) AMD issue? This is really annoying. :)

[Doc Ok]

... sorry, forgot one symptom. I normally use the "ondemand" CPU frequency governor, and when the lag thingy starts happening, the governor does not increase the CPU frequency beyond the 1000MHz baseline anymore, even under CPU utilization. Since the CPU usage sampler is timer-based, that might give another hint.

[Hlingler]

You have nvidia graphics chipsets and/or running a compositing window manager (e.g., Compiz-Fusion) on those machines?

V

[Doc Ok]

You have nvidia graphics chipsets and/or running a compositing window manager (e.g., Compiz-Fusion) on those machines?

D'oh, should have mentioned that. My bad.

Yes, I have an Nvidia GeForce 7800 GS, with Nvidia driver version 173.14.05, but no compositing window manager. I disabled the "Composite" extension in the xorg configuration file; don't remember for what reason right now. All my other computers have Nvidia cards as well, with the same/similar driver versions.

[Hlingler]

Well, that was just a hunch, and a long-shot at that. If no compositing in use, it's not likely to be graphics-related (but keep that in the back of your mind just in case...).

Please search the forums, as I believe I have seen other issues recently in RE: multi-core CPU load balancing and throttling.

V

P.S. If you find nothing relevant, then post back here.

[savage]

Maybe boot off a LiveCD/DVD and run memtest, check you don't have a dodgy memory address in the high end regions that causes the machine to crash when it's used.

I used to have 4GB RAM, the first 2GB sticks were fine, load anything requiring a lot of memory, and down goes the system.

Not sure if that could cause your "lags", I guess if the software is trying to recover, daemons crashing and reloading? Guess work here, and maybe some straws and clutching involved too, but another thing to rule out.

Do you have another network card you could try (based on the no-ping)? Also, do you run any network dependant services? Mounted NFS shares and the like?