Re: [Xen-devel] [PATCH 0/2] MTRR/PAT virtualization

[Keir Fraser <Keir.Fraser@xxxxxxxxxxxx>]

On 8/10/07 09:59, "Su, Disheng" <disheng.su@xxxxxxxxx> wrote:

>> Initialisation of guest MTRR state will need to be moved to hvmloader
>> -- we do not call setvcpucontext for HVM guests any more, so the
> 
> The problem here is to initialize MTRR MSRs of both BP and AP, in order
> to make sure MTRR MSRs are the same. Can do it in hvmloader?
> Another place to do it is in ROMBIOS, but there is lack of
> AP initialization logic. So what's your opinion about it? Do it in
> hvmloader
> , ROMBIOS or other place?

The best way, although a bit of a pain to implement, would be to bring up
all APs in hvmloader, initialise them, and HLT them. Like a proper BIOS.

I wonder whether full MTRR/PAT emulation is necessary or even particularly
desirable, however. I'd like to know what scenarios you are testing where
this improves correctness or performance.

 -- Keir



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