[prev in list] [next in list] [prev in thread] [next in thread] 

List:       xen-ia64-devel
Subject:    [Xen-ia64-devel] RE: [PATCH] use shared version of iva/dcr
From:       "Magenheimer, Dan (HP Labs Fort Collins)" <dan.magenheimer () hp ! com>
Date:       2005-08-23 20:56:17
Message-ID: 516F50407E01324991DD6D07B0531AD569516B () cacexc12 ! americas ! cpqcorp ! net
[Download RAW message or body]

The question that I was asking is:

Assume a hostile paravirtualized guest writes, say, -1
into the virtual iva (or dcr).  Is there any possibility that
this could crash Xen if Xen blindly uses the value
to branch to?  You are probably correct that the answer
is "no", but we should make sure of this before moving
it out of the private data structure.

> -----Original Message-----
> From: Matt Chapman [mailto:matthewc@cse.unsw.edu.au] 
> Sent: Tuesday, August 23, 2005 3:53 AM
> To: Magenheimer, Dan (HP Labs Fort Collins)
> Cc: xen-ia64-devel@lists.xensource.com
> Subject: Re: [PATCH] use shared version of iva/dcr
> 
> On Mon, Aug 22, 2005 at 09:56:57AM -0700, Magenheimer, Dan 
> (HP Labs Fort Collins) wrote:
> > For dcr, this may not matter, but for iva, there
> > are some semantics that need to be checked every
> > virtual interrupt delivery.  At a minimum, if virtual
> > iva is not private, bits 0-14 should be zeroed
> > before every use by Xen, correct?
> 
> We could do that, but I don't think it's necessary.  The 
> shared page is
> only written directly by paravirtualised guests, and for 
> those guests we
> can just define the guest interface such that that field must 
> be zero -
> it seems unlikely that any OS would intentionally write that field.
> If an OS were to write that field, it just shoots itself in 
> the foot as
> exceptions are vectored to the wrong address - no loss of isolation
> occurs.
> 
> Matt
> 
> 

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

[prev in list] [next in list] [prev in thread] [next in thread]