My notice was that it especially happend when the CPU usage grew, so I suppose it was related to the "powernow" feature of the CPU, because at least on Opteron systems SuSE's "power management" changes the CPU clock based on the load.
After I upgraded to the most recent vanilla kernel the problem was gone.
The powernow-k8 driver in SL9.3 caused clock instability on SMP machines that did not support HPET timers; SL10.0 and the mainstream kernel support PMTimer for those machines. That's most likely what you were seeing.
The single processor, dual-core problem is believed to be something different and is under investigation at AMD. We will make a statement as soon as we have definite information.
So can you confirm that the kernel options clock=pmtmr and notsc will not have any effect on this problem, and we should wait for a resolution from AMD?
I cannot confirm that. They may help. We're working on finding the answer.
I suspect that clock=pmtmr will help, for what it's worth.
Sorry for the request for clarification, but I'm stumbling around in the dark on this matter, and don't want to waste any more time on the problem - especially if it's all fruitless.
All I can officially say at this point is that AMD is aware of the problem and will be releasing a statement when we have more information.
Unofficially, clock=pmtmr should at least reduce the amount of lost ticks.
-Mark Langsdorf AMD, Inc.
On Thursday 27 October 2005 19:42, Langsdorf, Mark wrote:
Unofficially, clock=pmtmr should at least reduce the amount of lost ticks.
I doubt it will help on 64bit kernels, which don't have this option (ok placebos are sometimes known to help on computers too ;-) clock= only exists on 32bit.
notsc might do the trick on 64bit though. However it will only work on multi processor systems. On uniprocessor there is currently no way to force another clock.
-Andi
Hi Andi,
And on uniprocessors, with 2 cores? ;)
Thanks for the info, I will remove the clock=pmtmr from my boot line.
Cheers, Jon.
Andi Kleen wrote:
On Thursday 27 October 2005 19:42, Langsdorf, Mark wrote:
Unofficially, clock=pmtmr should at least reduce the amount of lost ticks.
I doubt it will help on 64bit kernels, which don't have this option (ok placebos are sometimes known to help on computers too ;-) clock= only exists on 32bit.
notsc might do the trick on 64bit though. However it will only work on multi processor systems. On uniprocessor there is currently no way to force another clock.
-Andi
-
Andi Kleen -
Jonathan Brooks -
Langsdorf, Mark