* Yani Ioannou <yani.ioannou at gmail.com> [2005-05-22 23:37:08 -0400]: > Hi Mark, > > > I thought I would finish the userspace/libsensors modifications first, > > then I could shake out any interface problems. As I wrote earlier: > ... > > I was thinking of passing the class device ID string as a parameter to > > hwmon_device_register() instead of pulling it out of dev->bus_id. If I > > were to re-submit the patch for inclusion (w/ the changes suggested by > > Greg)... what would you use for the class device ID string for bmcsensors? > > err... I don't know...not bus id :-). bmcsensors works using the > kernel IPMI messaging interface, and that really isn't a bus at > all...its more like a messaging protocol, hence the need in my mind to > remove all the i2c references. Sure, I realize that non I2C/SMBus sensors will need to somehow be uniquely identified in the hwmon class. That's why I was looking for feedback on just how to do that. For now, we can just let the hwmon registrants use an ID of their own choosing. > Jean was explaining to me today that this patch just adds the sysfs > clsss and drivers would still be i2c based, I was under the mistaken > impression it was standalone, but as your e-mail stated that is yet to > be implemented. I'm not sure what you mean here. Of course, sensor chips that live on SMBus will still be i2c based. As for the hwmon class, they will all register to that *in addition to* the rest of the existing sysfs interface. > I fear that bmcsensors really needs the full separation from i2c, how > realistic would it be to go ahead and try something like that at this > stage (I can help out)? It would seem to me that basically everything > under drivers/i2c/chips should be moved into drivers/hwmon and perhaps > drivers/i2c/bus to drivers/i2c (although the latter isn't necessary, > it just makes sense to me). Assuming that I modify my previous patch to allow hwmon class registrants to pass their own class ID, your bmcsensors driver would just add a bunch of sysfs device attributes just like existing drivers. You would not need to register with the I2C subsystem for anything. Point the hwmon class at your device by registering, and then userspace (will) know about it. Of course, libsensors is not yet ready for that. > > If we can have such a patch included on the condition that it may still > > change (due to interface / naming conventions) then I will rework it > > and submit it on Wed or Thu. Maybe we can keep it in -mm for an extended > > period until I finish the libsensors stuff. Would that work for you? > > Yes, I need the dynamic sysfs stuff that will be in -mm anyway, I so > that's perfectly fine, I expect all of this to be in testing for a > while. OK, I'll try to submit a patch tomorrow night. Regards, -- Mark M. Hoffman mhoffman at lightlink.com
