> On Mar 26, 2024, at 10:18 AM, Jason Thorpe <thor...@me.com> wrote: > > Like I said in a previous message on this thread, I think the better model is > for ugen to always exist for every USB device / interface, and for the kernel > driver to attach there. Then it’s easy to just say “detach whatever kernel > driver is here [assuming it’s not busy providing some other service] so that > user-space can take over”, and “re-attach any kernel driver that might match > this device now that I’m done”.
I guess I should explicitly state the other part of this, which is “and anything other than totally benign control pipe transfers without having CLAIM’d the device/interface must return EBUSY”. It’s the act of CLAIM’ing the device/interface that detaches the kernel driver and the act of RELEASE’ing it that causes the kernel to attempt to re-attach one. -- thorpej