> >>> On 8/16/2012 at 05:11 PM, David Boyes <dboyes@sine...>
> > Again, missed that SLES now has the code to stop in the right place
> > for this to work properly. Nice to know. Does it still need a EW
> > segment, or is it now well behaved for shared R/O?
> I wasn't aware that it was any other way. Are you perhaps thinking of xip2fs
> in a DCSS?
No, not at all. As you're aware, well behaved NSSes separate their R/O and R/W code such that as much as possible can be shared. There was some work in the mainline kernel tree a while back to do that (mostly for the embedded folks to make Linux more EPROM friendly), and I was wondering if that work had been leveraged here. If it had, then you wouldn't need an EW segment. You'd need two segments, but the SR would be the majority of the system.
There's also steps we need to be inserted to stop the IPL in an appropriate place after all the data structures were initialized and let CP save it at that point.