Random hangups, ACPI warnings, cannot read voltage sensors

[Kyeno]

Hello,
After few days of trying to nail the problem myself (with a bit of Google's support) I believe I've ultimately hit the wall and thus I'm condemned to ask for help.

I've recently slightly upgraded hardware in my home server (router, firewall, masq, samba, webdev+experiments) from some "garage parts". It currently runs:
- Intel DP35DP Motherboard with Intel P35 chipset (ICH9R)
- Intel Core2 Quad Q8200S 4x2.33GHz low TDP
- 3GB DDR2 667MHz Dual-Channel Asymmetric (2x1GB + 1GB)
- 3Com 3c905B 100mbit/s network card for Internet (eth0)
- Broadcom NetXtreme BCM5782 1gbit/s network card for LAN (eth1)
- S3 ViRGE PCI for console display and virtual framebuffer for headless browser experiments
- 2x500GB HDD3.5" SATA drives with 3 mdadm-based RAID1 mirror arrays
- Pretty much random 400WATT PSU with passive PFC

I've updated the BIOS to the last-recent one:

[Kyeno]

UPDATE:

I have compiled in most of the hardware drivers for this machine. Even when I don't need them - just added them in so the IRQ mapper "knows what it's doing" (at least those were my thoughts).

[cboldt]

When I've had random failures with no trace in logs, I was able to cure them with memory timing (BIOS), CPU clock timing, and voltages. Something is being pushed "just a little too hard." Not a thermal issue either, had good cooling, nothing "hot" inside.

Doesn't help any with your sensors issue, but might get you past the lockups. I have no more specific suggestions on CPU clock, RAM timing, etc., been so long since I had to play there. I do recall that any CPU clock or voltage moves I made, were modest. A few percent. I had an A7NX8 and was pushing (overclocking) CPU, with pretty good success. I think I ended up with 25% improvement in speed - but I'd started and ran for years with the thing underclocked.

[Kyeno]

Hey - thanks for the tips, @cboldt! Actually everything here runs completely stock, even with slightly under-timed mem (5-5-5-15, while those chips are capable of better timings).

Considering your suggestions I have just ordered Kingston Value Ram 2x2GB DualChannel kit so I can just skip using DualChannel Asymmetric - maybe that config is a bit too harsh for the northbridge indeed; despite the clock.

For the sensors I'm still pretty much lost. I think I'll try Lindows (Ubuntu LiveCD) on this machine and see if it detects anything; then if yes - I'll try to debug it from there. Otherwise I'm not sure anymore where to look or who to contact for support here. Intel support consultant/forums? Kernel support forums?

I have a feeling I have missed something very simple and obvious in the kernel config though. ISA Bus support? (still required? I remember times when I2C actually wanted this; now the kernel says to enable it *only* if you have real ISA slots)

[cboldt]

Systemrescuecd boots pretty fast, and is amazingly good at finding hardware.

This forum is as good as most for feedback, but if you want to poke around, a method I've used with some success is google or bing (or whatever) the MoBo plus "sensors" and maybe plus "kernel" or "linux." Let the search engine do the heavy lifting. You'll find out which forums are best in that very narrow line of inquiry.

You say the parts are "garage parts," I don't think you meant that literally, but you may have a "loose connection" somewhere. Take the MoBo apart, RAM and CPU, and put it back together. It doesn't take much of "the right thing" for a system to lockup. You may have done that already (reseat separable parts).

Could be too, power quality issues from MoBo part aging (caps), or (less likely) P/S. All fixable, but not without melting some solder

[C5ace]

My 8 year old desktop occasionally crashed X11. No cause shown in logs. Final cure was to pull video card, clean the connector. Now all OK.

[Kyeno]

@cboldt, @C5ace - thanks again for your tips. :)

And well.. I *built* it from garage parts, not just installed them.
Meaning I have inspected all MoBo and PCI card capacitors for being swollen (at least "using the eye and fingers", I didn't unsolder them for precise measurements), I have cleaned, polished and replaced all heatsinks while adding new, pretty much good quality thermal grease (silver-based), I have cleaned all connector pins of RAM modules and extension cards using acetylene based liquids, I have sprayed compressed air on all MoBo extension and RAM sockets, I have cleaned all fans (including PSU) and applied a drop of new rotor grease to each, etc.

For Googling around - I literally spend about 2 days on doing that before even writing 1st post here and bothering you guys. Surprisingly I find very little info about Intel DP35DP or even P35/82P35 sensors on Linux. Try googling for "DP35DP linux sensors" and you'll likely find this very thread on 1st page of results. :)

*edit*
Actually some Japanese site looks quite promising http://www.bondoffamily-net.com/~kinta-chan/ LOL!

*edit2*
http://cateee.net/lkddb/web-lkddb/INTEL_MEI.html getting closer... ;)

[cboldt]

Good job on the build, just tossing out possible sources of random failure.

For finding sensors, I'm of a mind that MoBo and chipset searches don't necessarily correspond with sensors.

Found this old article http://lm-sensors.lm-sensors.narkive.com/DntGFWsS/intel-dp35dp-p35-sensor-support that suggests module "coretemp," and some remarks about getting info from the isa bus.

More at http://marc.info/?t=119241816300001&r=1&w=2

[Jaglover]

I'd use a real voltmeter to check all voltages on ATX connector.
_________________
My Gentoo installation notes.
Please learn how to denote units correctly!

[Kyeno]

@cboldt Thanks a lot and I really appreciate it! I've seen the thread you posted before, no luck with it though.

@Jaglover Did this and they're fine, but that's random ad-hoc testing. My idea was scripting something that would output the voltages (among with most other sensor reads) every n minutes to some log file, so I'd have a chance of getting a bit more idea of what's going on with the hardware on system hangup.

[Kyeno]

Victory! (at least for one thing) :)

[cboldt]

Just that simple -- Heheheh. I have a couple of thinkpads, both have modules "mei" and "mei_me" loaded, and I try to pare down my kernels and modules to the bare minimum. I watch very little sensors-wise, CPU temp and frequency, and that's it.

[Kyeno]

I hear you - most of the time I do the same; yet in case of very strange hardware behavior it's probably wise to monitor a bit more. :)
And again I hear you about the kernels; I went on a path against my beliefs trying to get those sensors done here.
I'm kind of wondering though (just for the sake of pure experimentation) if that heci-qst solution would read your thinkpad's sensors since you got the modules already there?

Btw, so far the machine seems stable. Maybe I actually added something useful/previously missing to the kernel. I'll do a bit overnight stress test.

[cboldt]

I usually build or try to build so I can see everything, voltages, timings, etc. Once the challenge is met, I'm done with the adventure So done, that I forget how to view the stuff. New adventure with each new MoBo/platform. About 5 adventures in the last 10 years.

[Kyeno]

Haha :)

Btw, I gave a brief try and I somehow cannot get those reads on the Intel DP45SG computer with 3.7 kernel series. I wonder if it's the kernel alone, or the chipset.

[Kyeno]

...and I can pretty much confirm the DP35DP machine hangs on extensive harddrive usage. Just hanged on `perl-cleaner all`

[Kyeno]

Hello again.
Just posting here to let you know that I think I managed to regain the system stability and thus this issue might be marked as resolved (even though I still have some warnings in dmesg).

[steveL]