Hi Romain, On Wed, 22 Oct 2014 11:25:03 +0200, Romain Dolbeau wrote: > 2014-10-22 11:13 GMT+02:00 Jean Delvare <jdelvare@xxxxxxx>: > > On many motherboards, for an unknown reason, the thermal sensor seems > > to be disabled and will return a constant temperature value of 36.5 > > degrees Celsius. Don't bind to the device in that case, so that we > > don't report this bogus value to userspace. > > 36.5°C might be a legitimate value - for instance, if the driver is loaded > after a cold boot in a temperature-controlled machine room, the properly > cooled chipset might not have reached 37°C yet, so the chip will legitimately > report 0x7F in TSFSC... I wonder if this can really happen. In my experience these chipsets generate a lot of heat. On my board, the original heatsink was not suited for the case form factor and it would idle at 80°C. Under load it could reach 100°C [1]. [1] http://jdelvare.nerim.net/articles/i5500-heatsink-tuning/ I could instrument the driver to log the temperature value when the driver is loaded. > I _think_ a safer way would be to check variability - if it's not 0x7F > then fine, it's likely working. If is is, then drop TSTHRTHI by ~30°C ; > if TSFSC still reports 0x7F then it's definitely wrong since that would > be 6.5°C (machine room in the [ant]arctic where that would still be legit > shouldn't worry about chipset temperature anyway :-) I am reluctant to change the settings at driver load time, even if we attempt to revert shortly after that. It could easily cause a glitch. For example we have no idea how the chip reacts if the high temperature limit is lower than the "low" temperature limit. -- Jean Delvare SUSE L3 Support _______________________________________________ lm-sensors mailing list lm-sensors@xxxxxxxxxxxxxx http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
