[XEyedBear]

Having recently marked a thread here, on 'Linux System Freeze', as closed, I find that I need to do more investigation - hence the question in this thread title.

To recap: my Ubuntu 10.10 was freezing far too frequently. With help from this forum I was guided into changing the source for timing from its kernel default to use hpet timers. This appeared to fix the problem. Since then there has been a kernel upgrade (from ...27 to ...28) and the freeze problem is back. The Linux boot option 'hpet=force clocksource=hpet' is still in place.

The messages part of my system log show, at the start of boot:

Mar 20 15:17:33 Advent kernel: [    0.000000] Fast TSC calibration using PIT

then

Mar 20 15:17:33 Advent kernel: [    0.184114] HPET: 3 timers in total, 0 timers will be used for per-cpu timer
Mar 20 15:17:33 Advent kernel: [    0.184121] hpet0: at MMIO 0xfed00000, IRQs 2, 8, 0
Mar 20 15:17:33 Advent kernel: [    0.184126] hpet0: 3 comparators, 32-bit 14.318180 MHz counter
Mar 20 15:17:33 Advent kernel: [    0.192041] Switching to clocksource hpet

and then

Mar 21 09:22:51 Advent kernel: [   85.144090] Clocksource tsc unstable (delta = -250014055 ns)

and at that stage the system freezes.

Other posts on Ubuntu Forum show people continuing to be able to use their system after the 'clock source unstable' message. Indeed it is pointed out that with AMD processors the motherboard is frequently changing the CPU frequency - implying an unstable clock - but those systems continue to run.

Why does my system freeze?

(And, yes, I have got the BIOS set to 'Auto' for system frequency, which I think means that frequency can be changed as much as 3% depending on CPU load).

 

[hal8000]

In your other thread you were overclocking by 5%. Have you disabled overclocking?

Instability caused by clocking errors is usually down to hardware. The combination of motherboard, CPU and memory all must be inherently stable and compatible. Overclocking can cause instability.

Sometimes hardware can work with one system e.g. windows, but be unstable in another system e.g. linux (or vice versa). In these cases reducing clock speeds may help.

 

[XEyedBear]

Yes, I had changed overclocking from a fixed value of 5% to the standard setting which, if my reading of the inscrutable English of the ASUS motherboard guide is correct, means anything between 0% and 3% overclocking, depending on CPU load.

I have since set this back to a fixed system frequency of 200 Mhz (0% o'clocking) and haven't suffered from a freeze (but without an exhaustive test). So I conclude that it is a pathological combination of my motherboard, CPU, memory (and possibly graphics card) that the Linux Kernel doesn't like.

Bearing in mind that other reports (many) claim the AMD 3800+ X2 CPU is readily overclockable (by more than 10%) and that I am running my memory only at 75% of its rated speed (DDR333 instead of DDR400), The only workable solution that does not require me to replace both CPU and memory is to find a better socket 939 motherboard.

 

[hal8000]

Use google and search:
AMD 3800 socket 939 linux problems

This lists about 251,000 results, (not all problems).
This may help in your quest to find a better socket.

I spent 3 months researching my hardware before I built my new
computer. It had to work and be stable under linux. The search
of various motherboards was the hardest; some had sound chips
or other hardware that users had found problems with.

Luckily my current hardware has had no major issues, my descision
to use an ATI card was perhaps not so good. Video on ATI is not
quite as smooth as Nvidia, but then ATI cards were much cheaper.

The linux hardware compatibility list may also be of help to you:
Home Page - Linux Hardware Compatibility List