On Tue, Feb 26, 2008 at 10:47:34AM +0100, Carl-Daniel Hailfinger wrote:
Temporarily disable the fan control patch from this morning; it turns out to stop the CPU fan on the m57sli v1.1 (PLCC) entirely, which is less than desirable. I did not notice before because my board ran fine for about 15 minutes before the CPU overheated.
The disaster mentioned above shall never repeat. We need a testing procedure which ensures that a board does not overheat.
Quite.
Mprime (Linux variant of Prime95) should be a nice CPU head generator. Together with reading out the K8 internal temperature every 20 seconds, running Mprime for half an hour should definitely detect CPU overheating problems.
Actually I was using burnK7 (from the cpuburn package) in combination with sensors (from the lm-sensors package), and that worked great to produce load on the CPU.
In this particular case, I ran burnK7 for 5 minutes or so, saw with sensors that the fans *were* behaving correctly (and did a visual inspection too). It was not until the next reboot, with a new coreboot image (different payload), that things went south and the CPU fan stopped shortly after boot.
I'm not sure yet why it only failed on the second boot; the new fan code seems to write only absolute values to the registers.
I'll debug further when I'm next at the office, on Friday.
Thanks, Ward.