Xen devel - Dec 2005 - RE: Guest-visible phys2mach part of Xen arch-neutral API? was: Uses of &frame_table[xfn]

Thought I should sleep on this before replying... 
> p==m for any domain (even domain0) may have a siginificant effect on 
> the amount of otherwise arch-indep xenlinux code you can share.
Not to be snide, but its hard to call code arch-indep if it only
runs on one machine and has been designed only to meet the needs
of one architecture.  The discussion is about whether this code
should be made more flexible to accommodate other approaches,
or whether another approach should be disallowed.  This has
already been done for the block driver and most "core" code,
but not yet for the balloon driver or network driver.
> > So then is p==m in dom0 (and driver domains) an unacceptable design
> I think *that* is the critical question here.
I think there are two critical questions that are very highly
related.  To me, the critical question is how many changes
to a guest are required to run on Xen.  I''ve argued for a long
time that paravirtualization changes should be minimized/optimized
to only those that are absolutely necessary for functionality
and performance.   DMA-capable domains require either p==m
or non-trivial changes to the guest.  On x86, non-trivial changes
to the guest are necessary anyway due to the x86 memory architecture
so p!=m comes "for free".  This is not necessarily the case
for non-x86 Xen machines.
> something to discuss and think about at the summit?
Sounds good.

Dan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Friday 30 December 2005 13:50, Magenheimer, Dan (HP Labs Fort Collins) 
wrote:
>
> I think there are two critical questions that are very highly
> related.  To me, the critical question is how many changes
> to a guest are required to run on Xen.  I''ve argued for a long
> time that paravirtualization changes should be minimized/optimized
> to only those that are absolutely necessary for functionality
> and performance.   DMA-capable domains require either p==m
> or non-trivial changes to the guest.  On x86, non-trivial changes
> to the guest are necessary anyway due to the x86 memory architecture
> so p!=m comes "for free".  This is not necessarily the case
> for non-x86 Xen machines.
For the record, the PowerPC port is not currently using a p2m table in dom0 
(or any domain for that matter). We are still using the PAPR interface
(IBM''s
enterprise hypervisor ABI) for memory management.

Using Xen terminology, that interface says that domains pass only physical 
addresses to the hypervisor, and the hypervisor performs the physical to 
machine translation.

I think we''re not looking to diverge from this interface unless we
absolutely
have to, and so far (given the maturity of the PPC port) we haven''t had
to.

-- 
Hollis Blanchard
IBM Linux Technology Center

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel