On Sun, Jul 10, 2016 at 06:28:15PM +0000, Michael van Elst wrote: > How 'intelligent' can autoconf respond? Is there a proposal for a > protocol? How do the current methods to control autoconfiguration > fit into the "new scheme"? And of course, where is the benefit?
Slightly off topic, but while rethinking protocols there: we need to think about partial parallel tree visits in autocofig. We have that hacked partially, with ad-hoc synchronization, e.g. in usb or scsi bus discovery. This needs to be done for i2c buses too, e.g. a very significant portion of the boot time of my main machines that I reeboot ~daily are: spdmem and scrolling from drmkmsfb. We could do e.g. differnt types of visits for quiet boots, and parts of this could be controlled by return codes from attach functions too. So: yes, we need to draw the big picture now, before we start rotototilling. Martin
