Re: [Xen-ia64-devel] [RFC][PATCH] [RESEND] support special guest optimisations in the hypervisor

[Dietmar Hahn <dietmar.hahn@xxxxxxxxxxxxxxxxxxx>]

Am Freitag, 29. Juni 2007 schrieb Tristan Gingold:
> > attached to this mail are 2 patches:
> > - opt_feature_xen.patch
> >   implements the new hypercall  __HYPERVISOR_opt_feature in the
> > hypervisor. - opt_feature_linux.patch
> >   implements the usage of the hypercall.
>
> I think the hypercall interface is a little bit overkill but who knows
> about the future?
Which part do you mean with overkill?
For the function call I used the same scheme the other hypercalls use.
Or do you mean the structure?
What I wanted to do was a common solution. All the hypervisor needs to handle 
the identity mapping insertion are the access bit of the pte and (maybe in 
the future) the protection key.
Ok, I can remove the key, because there is currently no guest using this. And 
we can add it, when it will be necessary - should I resend the patch without 
the key?

>
> > The xen - patch has real effects in vcpu_translate() on inserting
> > identity mappings in the vhpt/tlb. I'am not sure about some performance
> > impacts. But without such a patch correct future protection key handling
> > is impossible.
I meant here needed is the possibility to disable the auto id mapping.

>
> I don't understand why.  Isn't PK 0 supposed to be always enabled ?
I thought of the possibility one guest may use PK!=0 for identity mapped 
pages. But you are right, currently there is no ia64-OS available using PK - 
so this is a bit overkill currently.

> (But I think it should be possible to disable the auto id mapping
> insertion).
Yes, FreeBSD uses the same scheme with identity mapping on region 7. Maybe a 
FreeBSD PV-port would lead to problems.
Many thanks for your comments!

Dietmar.

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel