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

Re: [Xen-devel] [V10 PATCH 0/4] pvh dom0 patches...

On Thu, 8 May 2014 14:15:36 +0100
David Vrabel <david.vrabel@xxxxxxxxxx> wrote:

> On 08/05/14 01:12, Mukesh Rathor wrote:
> > On Wed, 7 May 2014 15:38:26 +0200
> > Roger Pau Monnà <roger.pau@xxxxxxxxxx> wrote:
> > 
> >> On 07/05/14 15:20, Konrad Rzeszutek Wilk wrote:
> >>> For example, right now the Linux PV code can run with an E820 that
> >>> looks like swiss cheese - aka like the hosts' one. That is easily
> >>> seen if you boot an PV guest with PCI passthrough devices - as the
> >>> libxl 'e820_host' option gets turned on which creates an E820
> >>> that looks like the hosts. Granted it does not populate the P2M
> >>> as such (it is all linear). But the point is that the Linux
> >>> code is capable of dealing with this and bring the P2M to sync.
> >>>
> >>> Hoisting this up in the hypervisor would be a plus - as the
> >>> Linux code wouldn't have to do this anymore.
> >>
> >> IMHO doing it in the hypervisor is clearly the right solution,
> >> forcing the guest OS to do all this on it's own just promotes code
> >> duplication across the several OSes with PV(H) support.
> > 
> > I am in agreement with you, but my point was code exists for PV
> > already, so if you address pvh only, then that code still has to
> > exist in the guest until PV reaches EOL. 
> > 
> > FWIW, my agreement with previous maintainers was to keep PVH as
> > close to PV as possible.
> PVH support in Linux should be in order of preference:
> 1. Like native.
> 2. Like HVM guests.
> 3. Like PV guests.
> 4. PVH specific.

I've been thinking:

  PVH == PV guest in HVM container with HAP. (Actually, my initial 
         UP cut didn't require hap, but smp got very complicated. Besides
         eliminating pv mmu became seondary goal).

Now, for prefrence (where choice exist):

   For (each path)
       if (HVM performs better)
           pvh == hvm
           pvh == pv

IOW, HVM good, PV better, PVH best!  Altho, now i believe 
  hvm && pv == better

Historical background jfyi, when I talked about pvh to Keir, Ian P,
and can't remember others on the sidelines of xen summit years ago, 
pv ops didn't exist, and my main motivation was performance. I didn't 
think it would take this long, but I am euphoric to see it finally be
done, well, almost :)...


Xen-devel mailing list



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