SMI issue on Compaq Deskpro EN

After reviewing the hardware list in the wiki a while back I noticed that the compaq deskpro EN was listed as a good machine. By sheer coincendence I just happened to have one laying around. It is a nice machine particularly since it is very compact.

I have installed the latest EMC2 on it but am having some issues with SMI jitter. The machine posts excellent latency times (under 10uS normally, max under 17uS or so during heavy graphics). The issue is that when the CPU is heavily loaded (with burnmmx or similar) I get jitter spikes whenever the fan changes speed. I'm talking jitter on the order of 14mS (yes milleseconds). I have already disabled power management in the BIOS as suggested by the wiki entry for the machine. If I load the rtai_smi.ko module the problem goes away (but the fan doesn't change speed either...).

Unfortunately there is no option in the bios for setting the fan to max... altough I can load the machine up and then load the SMI module when the fans aer already at max speed. Reading online it appears that the machine has a "hidden" SMBUS which is where the SMI does its fan control. I may be able to get access to that bus and then run a userspace fan control script. Still looking into this.

Anyone have any tips?


James

Hi

Are you able to measure the voltage at the fan when at full speed?
I imagine the fan is variable voltage and it may well be near 12v at full speed, but you will have to investigate.

I would be very tempted to hard wire the CPU fan to a feed from the main PSU, dropped to the required voltage if necessary, instead of its dedicated plug on the MB.

That should keep the processor cool and enable you to eliminate SMI spikes via the rtai_smi module in a very simple manner.

I would be interested in a lspci -vv print from your machine to match the exact chipset against the list I have, for which the smi fix is effective.

regards

Hot-wiring the fans could be a solution. This machine has three variable speed fans but I could do them all.

I saw some conversations on RTAI groups to the effect that folks who turned on ACPI kernels got fan control and low jitter without disabling SMI... but information was a bit slim.

Here is an lspci dump from the machine.

Hi

Thanks for the print, yes the 82801BA is definitely on the list of those that work.

I saw some conversations on RTAI groups to the effect that folks who turned on ACPI kernels got fan control and low jitter without disabling SMI...

If you fancy taking that further, it would benefit a lot of people if it works.

I found that only one of the ICH8 family of chips which appear in later Intel boards, the ICH8_4, is accepted by the rtai_smi module.
Hacking it to get it to accept the other 7 sub-sets, worked so far as loading the module went, but made matters worse not better, regards latency.

Presumably the bits set by the module were changed in location and/or purpose in these later chips, so any other method of achieving the same would be interesting.

regards

I forgot to follow up on this thread a year ago when I was fooling with this.

I ended up figuring out how to unhide the SMI bus on this machine ( lists.lm-sensors.org/pipermail/lm-sensors/2008-June/023367.html ) then disabled the SMI ( wiki.linuxcnc.org/cgi-bin/wiki.pl?FixingSMIIssues ). Once SMI is disabled and the SMBUS is unhidden you can then control the fan with standard Linux userland applications through the machine's I2c fan control bus. Once this is done the latencies stay nice and low.

Unfortunately for the little Deskpro I managed to find an older (but still much faster than the Deskpro) Athlon machine that I also had laying around which seemed to work perfectly under EMC so I will likely just use this machine instead.


James