On 21.07.2011, at 18:46, Scott Wood wrote: > On Thu, 21 Jul 2011 03:27:14 +0200 > Alexander Graf <ag...@suse.de> wrote: > >> All guest CPUs need to be specified in the device trees. Since removing nodes >> is easy with FDT, but adding nodes is not, we just put 32 CPU nodes into the >> device tree and remove them later on init when not used. > > Adding nodes isn't that hard...
Hm. I'll check again and see if I can add a node. If so, we can just completely replace the CPU definitions by self-generated ones. > I'm also hesitant to stick so many CPUs on something modeled after an > e500v2-era mpc85xx. E.g. in the MPIC the upper bits of xIDR are used > for EP/CIx. The newer MPIC used in p4080 and such move crit/mcheck routing > into a separate register. There are similar constraints in guts (power > management), and possibly elsewhere. Good point. See my other mail. I'll fix the CI thing and limit the # of CPUs. Alex