Hello, I'm using lm_sensors 2.7.0 on a mini-cluster, which consists of 8 Dual-Athlon nodes running on Tyan Tiger motherboards (two of them have older AMD760MP chipset, six have 760MPX). Kernel version is 2.4.19. During last 4 months I've noticed two failures related to lm_sensors i this cluster. In both cases machines stopped responding, logged strange values of temperatures and voltages, and finally, mondo daemon succeeded to shutdown them (mondo was set to protect machines from overheating due to fan failures and does `halt -fp' if something seems to be going wrong). Here comes part of the log from /var/log/messages Aug 1 05:30:03 onyx kernel: i2c-amd756.o: SMBus collision! Aug 1 05:30:08 onyx kernel: i2c-amd756.o: Busy wait timeout! (0800) Aug 1 05:30:08 onyx kernel: i2c-amd756.o: Sending abort. Mondo wants to read sensors every 5 sec, so it was repeated many times, until the machine went down. It's worth to note, that it took ~40 minutes to halt the machine, normally `halt -fp' should do a poweroff within few seconds. Another software, that reads the sensors is gkrellmd - is it possible, that problem comes from simultanous attempts to read the sensors chip? I don't know how to reproduce this behavior, it occurs so rarely (two cases on 8 machines in 4 months gives a MTBF > 1 year), and, unfortunately, I had no chance to get the console just as the error occurred as yet, because I was away from server room. -- Greetings Artur Gawryszczak
