sysctl magnitude and sysfs ...

Philip Pokorny ppokorny at penguincomputing.com
Mon Mar 31 09:10:23 CEST 2003


Hmmm...

I now see that I am hopelessly out of touch the current state of lm_sensors 
development.

To fix that, please add me to the list.  Since I'm updating lm85 with the SMSC 
chip and writing the adm1026, plus I want to have some input on the adm1021 
(for xeons...)

So I guess my magnitude suggestion is kinda silly given the long thread about 
sysfs support...

My two cents on the sysfs proposal would be that it takes something that was 
easy to understand (a list of structures defining "files") and a few basic 
functions one for each type of sensor (generally) and replaces them with a 
bizarre set of macros with concatenation operators creating a blizzard of 
functions and entries in the chip directory.

For reference, both the lm85 and adm1026 will have about 100 or more entries 
in the new sysfs scheme.

Some temperature sensors have min/max limits.  Others have over/hyst limits.

One other comment that didn't seem to get acknowledged was that if you're 
going to switch to fixed point and leave off the decimal, then it would be 
helpful to make *everything* milli-volts and milli-degC.  That means * 1000 
not *100.  Most devices have at least 8-bit A/D converters and using only 
1/100 volts will loose precision in some cases.  Some chips have 10-bit or 
more A/D converters.  For more than 10-bits, even milli-volts may not be 
accurate enough.

As for sensor numbering, that's what libsensors and sensors.conf is for.  To 
map the driver names (fan0, fan1, fan2) to the board labels (SYS_FAN1, 
CPU0_FAN, CPU2_FAN, FAN3).  So if we want to change to base 0 for all the 
inputs, that might make sense.

I also notice that the new code being commited have lines more than 80 
characters.  Is there no longer a kernel coding standard of 80 characters?  I 
agree that some of the line breaks in the exisitng code are strange...




More information about the lm-sensors mailing list