> Nope, the I2C core calls asb100_detach_client() three times... once > for the primary client and twice more for the subclients. Ohhhh OK. Now I get it. It's a bit confusing at first. I'd have expected two different detach functions (i.e. two different driver structures). But I agree it works. > The changes in 2.6 are also on purpose. They're necessary because the > I2C client list was implemented differently than in 2.4, which caused > it to try to release the subclients twice. I didn't see much change, except the sysctl_id being gone. > I.e. technically, it's the CVS w83781d driver that's wrong... although > it happens to be benign vs. 2.4 kernel. Yes, I see that now. I'm surprised that it doesn't cause more trouble than it does. Care to correct the 2.4/CVS driver so that it does what your asb100 driver does (i.e. the right thing)? Thanks. -- Jean Delvare http://www.ensicaen.ismra.fr/~delvare/
