Jean Delvare wrote: > Hi Hans, > > On Sun, 25 Jan 2009 14:47:30 +0100, Hans de Goede wrote: >> Jean Delvare wrote: >>> Really? I agree for bus drivers, but this seems difficult for hwmon >>> chip drivers. Do you have some ideas that you didn't share with us yet? >> Well I hope that for graphiccards, the drm / modesetting driver will also >> (probe for?) and instantiate any known to be (sometimes) present i2c sensors. > > If they are going to do the same probings we have to do for mainboards, > then I don't quite get the point. Well yes and no, most cards based on a certain GPU all use the same hwmon IC (the one from the reference design). But some may deviate, so there only needs to be a probe for one, maybe 2 chips, and a fairly simple probe. There is no certainty that some manufacturer won't deviate, hence the probe is needed, but the situation is nowhere as bad as with mainboards. > If it is possible to instantiate the sensor devices without probing > because you can get enough information from the hardware to know what > and where they are, then that's a different story, and I agree we > should do that. Any probing we can avoid is good. > Well there is no 100% certainty (AFAIK) so we still need a probe, but basicly only for one chip, if the user has one of the rare deviating board he / she is out of luck. >>> My initial goal was to make the configuration file more simple, not >>> more complex ;) The main reason why I wanted to split bus drivers and >>> hardware monitoring chip drivers is that lm_sensors should be >>> considered not configured as long as no hardware monitoring drivers are >>> listed. The fact that this will make things a little clearer to the >>> users is only a side benefit. Oh, and in a few cases, it is more >>> efficient to load the bus drivers first. >>> >>> I don't see much value in splitting the variables further. >>> >>> Bus drivers will almost always auto-load these days, so these variables >>> will almost always be empty. And the trend for graphics adapter is to >>> expose I2C buses from the graphics driver itself, so you aren't loading >>> the bus drivers just for lm-sensors. >>> >>> If we ever have a kernel driver for hard disk temperatures (that would >>> be nice!) then I suspect it will be the same driver for all disks, and >>> hopefully it will auto-load. >>> >>> This leaves us with the hardware monitoring chips themselves. Given >>> that it is possible that the same driver is used for the motherboard >>> and for the graphics adapter, how would you split them? Sure, you could >>> list the module twice, but that's not exactly efficient. >>> >>> What benefits do you see in splitting the variables that way? >> Oneday I hope we have a tool which will generate and /etc/sensors.conf and >> /etc/sysconfig/lm_sensors based on dmi info and a database, then it would be >> good to have separate line in /etc/sysconfig/lm_sensors, for sensors located on >> the motherboard and sensors located on for example a graphic card, so that that >> tool then can only mangle the motherboard lines, without touching any >> (potentially manually added) other line. > > I don't expect /etc/sensors.conf to be ever generated. Instead my plan > is to change libsensors to read all configuration files from > directory /etc/sensors.d. Then our to-be-written configuration tool > would simply drop board-specific configuration files to that directory. > This is IMHO much easier than merging several files > into /etc/sensors.conf. Ack, this sounds like a good plan. > I wanted to work on this before the end of > January but have been delayed by other issues. > > With regards to separate variables in /etc/sysconfig/lm_sensors, it > seems to only make sense if our to-be-written configuration tool allows > the user to reconfigure the sensors separately on the mainboard and on > graphics adapters. Do we have a need for this? At the moment, > sensors-detect doesn't make any difference between the mainboard and > graphics adapters, it probes the graphics adapters' I2C buses as it > does for the SMBus. True, I guess this is something best revisited when it actually becomes an issue. <snip> > > But I don't think there is anything that needs to be done now. The new > configuration file format is easily extended in the future if we need > variables for specific purposes. > Ack. Regards, Hans