Hi Yani, > I have a stablish working ipmisensors driver (the bmcsensors re-write > for kernel 2.6/hwmon - see > http://bmcsensors-26.sourceforge.net/ipmisensors/ for patches), and > I'm tyring to add support to libsensors and sensors for testing. As > far as I can tell the following patch should work, We can obviously apply this, it's not very intrusive ;) However, may I ask why you didn't simply keep the "bmc" prefix? You would have won compatibility with a few previous releases of lm_sensors, and the name change might also confuse the users. You missed one bmc -> impi duplication in etc/sensors.conf.eg. > however after > applying it I still can't see the sensors (yet the entries do appear > in /sys/class/hwmon/hwmon0/*). Any help would be appreciated (as you > can tell I'm new to the userspace side ;) If you are usins lm_sensors 2.10.0, it has a known bug that prevents hardware monitoring chips from being properly detected if there isn't at least one i2c chip with a driver loaded. This is fixed in SVN, so you should use that. Secondly, make sure you created a "name" attribute with the chip type (that would be "ipmi" for you) for your device. Lastly, libsensors has some bus-specific code to generate chip "names" (like "adm1032-i2c-1-4c"). Depending on where in the sysfs tree your device is, the code might fail to generate a proper name. The code doing that is in lib/sysfs.c:sensors_read_sysfs_bus() if you want to take a look. strace might come in handy to diagnose the problem. > I do have some other concerns/ideas on support ipmisensors would be > looking for in libsensors, and considering a possible re-write I feel > I should list these now. > > - fixed number of sensors > ipmisensors has a dynamic ("hotplug" sensors not supported at runtime > at the moment) and unknown number of sensors. As far as I can tell > libsensors needs to know a fixed number. For now, yes. However there are plans floating around to have libsensors probe for chip features (Linux 2.6 only) instead of using an internal hard-coded list for every chip. That's what Mark M. Hoffman and I had been discussing at the restaurant in Ottawa, if you remember. I know Hans de Goede has interest in this too, unfortunately I'm a bit too busy right now to go on. > - labels given > ipmisensors gives the BMC's labels for each sensors in a "label" > entry. This should be easy to add support for (perhaps defaulting to > these labels without user configured labels). Yeah, that would be very convenient for the users. Indeed it shouldn't be too hard to implement. Once again I think I read that Hans de Goede has a similar need for his new uGuru2 driver. > - binary sensors > IPMI supports "non-threshold sensors" for things like power status, > chassis intrusion, etc, which are basically binary valued. You mean _booleans_? > I plan to > add support for these to the driver, but do these belong in lm-sensors > and how hard would it be to add support for them to libsensors? I see no problem here, booleans are simple integers after all, so this is already support. Attributes like pwm1_enable were booleans too until recently, and it works just fine. The extra attributes (e.g. chassis intrusion) might lack proper standardization at the moment, though. -- Jean Delvare
