If you do want to go all the way, IPMI supports 8 limits per sensor. Critical, non-critical, non-recoverable, and hysteresis, each for both high and low. I suppose IPMI is a good guideline for the ultimate in possibilities, should you wish to refer to it in the sysfs design. Although I think a sophisticated BMC manager would access the BMC directly rather than through /proc or /sys. So don't be overly concerned with /sys limiting the IPMI information. What I do in bmcsensors is pick the best two of those 8 and present them to the user. IPMI defines hysteresis as a delta from the limit. I subtract one from the other and present that through /proc as absolute values. I think we should continue to do that (i.e. I agree with Mark Hoffman). Jean Delvare wrote: > > > current is in there because the IPMI spec includes current sensors > > so I added support to bmcsensors. > > Never had a report of a BMC that actually measures current, though... > > Took a look at that document, it specifies hysteresis values for all > sensor types. It even specifies two values (one for negative-going > thresholds, one for positive-going thresholds). > > BMC isn't in 2.6 yet anyway, so we can safely remove references to > current hysteresis for now. > > That quick look at the IPMI specification once again raised the question > of what file naming convention to choose for storing hysteresis values > in sysfs. The fact that IPMI provides space for two hysteresis values > per sensor proves that names like "temp_hyst1" (as I did in the patch I > just sent to Greg) won't be sufficent to handle all cases. I'm still in > favor of storing hysteresis as a values linked to another limit, and not > as a limit per se. That is, temp_max1 and temp_max1_hyst (or > temp_max_hyst1, or even temp_hyst_max1, whatever fits better with the > current logic). Following Mark Hoffman's views, value in these files > would be an absolute temperature, regardless of how the chipset actually > handles it. > > I would like everyone to comment on that if needed. If we can agree on a > clean, logic naming convention, I volunteer to update the documentation > accordingly and modify existing drivers. > > -- > Jean Delvare > http://www.ensicaen.ismra.fr/~delvare/