Jean Delvare wrote:
> Hi Hans,
>
> On Sat, 07 Jul 2007 21:23:40 +0200, Hans de Goede wrote:
>> Jean Delvare wrote:
>>> The sensors3 patch should be pretty easy, maybe you could take care of
>>> it? It's needed anyway, whether my kernel patch gets applied or not.
>> Ok, here is a proposal:
>>
>> Index: prog/sensors/chips_generic.c
>> ===================================================================
>> --- prog/sensors/chips_generic.c (revision 4566)
>> +++ prog/sensors/chips_generic.c (working copy)
>> @@ -209,6 +209,11 @@
>> /* print out temperature sensor info */
>> if (TEMP_FEATURE(SENSORS_FEATURE_TEMP_SENS)) {
>> int sens = (int)TEMP_FEATURE_VAL(SENSORS_FEATURE_TEMP_SENS);
>> +
>> + /* older kernels / drivers sometimes report a beta value for thermistors */
>> + if (sens > 1000)
>> + sens = 4;
>> +
>> printf("sensor = %s", sens == 0 ? "disabled" :
>> sens == 1 ? "diode" :
>> sens == 2 ? "transistor" :
>>
>> The idea here is that a beta value should (reasonably) always be > 1000, and we
>> don't want to assume thermistor for any unknown value as we might add new types
>> to the list later. Does this look ok?
>
> Yes, that's fine with me.
>
committed
>>> I'm still not sure which minimum version we want to support. I don't
>>> want to support anything older than 2.6.5 for sure, things were just
>>> too badly broken before that. And I don't want to clutter the
>>> libsensors code too much to deal with the non-standard things that some
>>> old kernels had. Unless someone has a better plan, I think we'll make
>>> things work for the systems we developers are using, and release that,
>>> and then if users complain that it doesn't work with some older kernel
>>> they're using, we'll consider adding support on a case-by-case basis,
>>> depending on how intrusive that would be.
>> That sounds like a plan, but not like something one would put in a README,
>> which then brings me to the question what do we put in the readme, just that it
>> requires a kernel > 2.6.5, or maybe we should require one which has the hwmon
>> class?
>
> Well, requiring the hwmon class would let us cleanup a number of things
> for sure, so it may be a good idea. That would mean we "only" support
> kernels >= 2.6.14. But even then, the README will have to mention that
> some features are not available before much more recent kernels. Most
> notably, the alarms for most devices won't be available before 2.6.24.
>
2.6.14 sounds ok as a base, most people shoild be using something newer then
that by now.
Regards,
Hans
