[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] RE: XEN monitor/mwait patch for AMD SVM


  • To: "Keir Fraser" <keir.fraser@xxxxxxxxxxxxx>, <xen-devel@xxxxxxxxxxxxxxxxxxx>
  • From: "Boppuri, Navin" <Navin.Boppuri@xxxxxxx>
  • Date: Mon, 25 May 2009 08:26:05 -0500
  • Cc:
  • Delivery-date: Tue, 26 May 2009 07:25:05 -0700
  • List-id: Xen developer discussion <xen-devel.lists.xensource.com>
  • Thread-index: Acna8T+H6taQdEAnQUKV8e71Hl/oFwACq7+nAACgPVAAALuOmwAsRDsQAABzSbcAYe780A==
  • Thread-topic: XEN monitor/mwait patch for AMD SVM

Keir,

Sounds good. As long as we are consistent about the fact that all configuration 
policies are managed using libxc, I'm more than satisfied with this approach to 
prevent use of mwait/monitor instructions in a guest.

Thanks again.
-Navin

> -----Original Message-----
> From: Keir Fraser [mailto:keir.fraser@xxxxxxxxxxxxx]
> Sent: Saturday, May 23, 2009 9:39 AM
> To: Boppuri, Navin; xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: Re: XEN monitor/mwait patch for AMD SVM
> 
> No problem. In general, we aren't too afraid of having 'power user'
> options
> in the domain config file that can cause real problems if misapplied.
> Especially this CPUID config, which most end users have not a prayer of
> understanding and probably should not touch at all unless told to. The
> expectation is that most vendors will wrap up that layer with a GUI wizard
> of some sort to hide most of the complexity and power.
> 
>  -- Keir
> 
> On 23/05/2009 15:26, "Boppuri, Navin" <Navin.Boppuri@xxxxxxx> wrote:
> 
> > :) Point taken! Thank you for considering the request though.
> >
> > -Navin
> >
> >> -----Original Message-----
> >> From: Keir Fraser [mailto:keir.fraser@xxxxxxxxxxxxx]
> >> Sent: Friday, May 22, 2009 12:19 PM
> >> To: Boppuri, Navin; xen-devel@xxxxxxxxxxxxxxxxxxx
> >> Subject: Re: XEN monitor/mwait patch for AMD SVM
> >>
> >> That's firmly in the 'Doctor, it hurts when I...' category I'm afraid.
> No
> >> way will I apply this type of patch.
> >>
> >>  -- Keir
> >>
> >> On 22/05/2009 18:10, "Boppuri, Navin" <Navin.Boppuri@xxxxxxx> wrote:
> >>
> >>> Keir,
> >>>
> >>> Sorry but the patch is against the latest tip of the xen unstable tree
> >>> (changeset: 18646).
> >>>
> >>> I understand that MWAIT is not mentioned in the whitelist for HVM
> >> policy.
> >>> However, one could modify the guest config file and set the cpuid bit
> >> for
> >>> monitor/mwait and enable this feature by force. I was able to do just
> >> this in
> >>> my test configuration. This patch should cover such mods for cpuid in
> >> addition
> >>> to the cpuid policy in libxc.
> >>>
> >>> Thank you.
> >>> -Navin
> >>>
> >>>> -----Original Message-----
> >>>> From: Keir Fraser [mailto:keir.fraser@xxxxxxxxxxxxx]
> >>>> Sent: Friday, May 22, 2009 11:40 AM
> >>>> To: Boppuri, Navin; xen-devel@xxxxxxxxxxxxxxxxxxx
> >>>> Subject: Re: XEN monitor/mwait patch for AMD SVM
> >>>>
> >>>> On 22/05/2009 16:23, "Boppuri, Navin" <Navin.Boppuri@xxxxxxx> wrote:
> >>>>
> >>>>> This patch clears the monitor/mwait bit in the cpuid intercept code
> >> for
> >>>> AMD¹s
> >>>>> SVM. This is to disallow VM¹s that may try and use the monitor/mwait
> >>>> feature
> >>>>> on AMD processors which is currently not supported in XEN.
> >>>>>
> >>>>> Signed-off-by: Navin Boppuri <navin.boppuri@xxxxxxx>
> >>>>
> >>>> What Xen version is this patch against? We already hide MWAIT from
> HVM
> >>>> guests as far as I am aware. See
> >>>> tools/libxc/xc_cpuid_x86.c:xc_cpuid_hvm_policy(), and note that
> >> features
> >>>> are
> >>>> *whitelisted* and also that MWAIT is *not* mentioned.
> >>>>
> >>>> So I think your patch is not needed, and also inappropriate since we
> >>>> implement CPUID filtering policy in libxc these days.
> >>>>
> >>>>  -- Keir
> >>>>
> >>>>
> >>>
> >>>
> >>
> >>
> >
> >
> 
> 



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


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.