On Tue, 30 Dec 2003, Jean Delvare wrote: > I think I remember that Ky?sti was quite interested in classes because > it would let us ignore non-SMBus adapters in sensors. Not sure how > great it is however, since actually we should look for specific chips, > not specific adapters. I was referring to a class_device in the driver-model. Registering a sensor as a class device would at minimum create symbolic links to sensor chips in a common directory under sysfs, no matter which hw bus implementation they have. The class test with adapter and driver is something different. I don't know where this part of the code came from but the purpose seems clear, to prevent, eg. tuner driver unintentionally probing on an SMBus clock generator. A related issue is i2c_clients_command. For every client on an adapter, it calls client->driver->command(client, cmd, data) if specified. This blindly assumes a global enumeration of cmd, which is not true. -- Ky?sti M?lkki <kyosti.malkki at welho.com>
