Hi Greg, On Thu, Feb 03, 2022 at 03:55:19PM +0100, Greg Kroah-Hartman wrote: > On Thu, Feb 03, 2022 at 05:46:55PM +0300, Heikki Krogerus wrote: > > +/* These additional details are only available with vSafe5V supplies */ > > +static struct kobj_attribute dual_role_power_attr = __ATTR_RO(dual_role_power); > > +static struct kobj_attribute usb_suspend_supported_attr = __ATTR_RO(usb_suspend_supported); > > +static struct kobj_attribute unconstrained_power_attr = __ATTR_RO(unconstrained_power); > > +static struct kobj_attribute usb_communication_capable_attr = __ATTR_RO(usb_communication_capable); > > +static struct kobj_attribute dual_role_data_attr = __ATTR_RO(dual_role_data); > > +static struct kobj_attribute > > +unchunked_extended_messages_supported_attr = __ATTR_RO(unchunked_extended_messages_supported); > > Note, no 'struct device' should ever have a "raw" kobject hanging off of > it. If so, something went wrong. > > If you do this, userspace will never be notified of the attributes and > any userspace representation of the tree will be messed up. > > Please, use an attribute directory with a name, or if you really need to > go another level deep, use a real 'struct device'. As-is here, I can't > take it. OK, got it. I don't think we can avoid the deeper levels, not without making this really cryptic, and not really usable in all cases. These objects are trying to represent (parts) of the protocol - the messages, the objects in those messages, and later the responses to those messages. But I'm also trying to avoid having to claim that these objects are "devices", because honestly, claiming that the packages used in communication are devices is confusing, and just wrong. If we take that road, then we really should redefine what struct device is supposed to represent, and rename it also. So would it be OK that, instead of registering these objects as devices, we just introduce a kset where we can group them (/sys/kernel/usb_power_delivery)? thanks, -- heikki
