Hello Jean, I have tested the "new" version and the result is better, but we can live with the "old" version too. I have attached sensors.conf and a display output for the "new" version. Thanks and best regards Horst Kretschmer -----Urspr?ngliche Nachricht----- Von: Jean Delvare [mailto:khali at linux-fr.org] Gesendet: Mittwoch, 16. Juni 2004 21:29 An: Kretschmer, Horst Cc: Bervas, Elodie; Oelkers, Hans-Joachim; Duellings, Peter Betreff: Re: AW: LM-Sensors > I have tested -13V with lower values in sensors.conf with the > following results: > > sensors.conf displayed /proc/sys/....../in1 > -13 > -12.99 -13.22 2.42 2.55 2.52 > -12.98 -13.22 2.42 2.55 2.52 > .. > -12.88 -13.22 2.42 2.55 2.52 > > -12.87 -12.88 2.44 2.55 2.52 Hmmm, OK. I get one part of the mystery. The native resolution of the A-D converter is 0.0116V. It's slightly more than 0.01V, which means that, with a magnitude of 2, some values cannot be represented. Among others, 2.43V cannot be represented. This explains why the converted value jumps from -12.88 to -13.22 directly (while you would at least expect -13.05 inbetween). Also, I just realized that there are four successive roundings when setting a limit. libsensors has to round to the driver's magnitude, then the driver rounds to an integer (register value), then you read the value (with sensors), so conversions are done the other way around. For most drivers, this isn't a problem because the ratio between register values and exported values is "nice" (typical A-D converters have a 16mV/bit resolution) so the noise is limited. But the PC87366 has a weird resolution and this probably adds noise to the conversions. I think that using a magnitude of 3 for voltages should reduce that noise and the driver should obey "sensors -s" better. Patch attached. Could you please apply and test? If it improves the situation for you, I'll apply to our CVS repository. Then we'll see if more is needed. Thanks. -- Jean Delvare http://khali.linux-fr.org/ -------------- next part -------------- A non-text attachment was scrubbed... Name: sensors_new.zip Type: application/octet-stream Size: 1167 bytes Desc: not available Url : http://lists.lm-sensors.org/pipermail/lm-sensors/attachments/20040617/4ba2833a/attachment.obj
