On Mon, 2006-07-31 at 10:59 +0200, Andi Kleen wrote: > On Monday 31 July 2006 10:27, Christian Limpach wrote: > > I really > > don't think either of those are very appealing, especially considering > > that there's a much simpler solution using hypervisor specific entry > > points. > > I think i would prefer Zach's later probe method over the multiple entry > points. It will be likely cleaner. Surely there could be some way to > probe for Xen? We could certainly have probe fns for each paravirt ops and call them in order first up (after some test about ring 0 and/or paging enabled?). These probe fns might be as simple as "is register xxx some magic value?" or whatever. This was my original thought, but seemed like overkill so I dropped it. I've been assuming that it's important that whatever we choose support Xen 3.0 as it currently stands. Jeremy's idea that they'll change to using ELF notes suggests that this compatibility is not as important as I thought? As long as all the per-hypervisor magic is contained in the ->ops structure, I don't really care *what* it is... Rusty. -- Help! Save Australia from the worst of the DMCA: http://linux.org.au/law
