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

Re: [Xen-devel] [PATCH v3 5/6] ioreq-server: add support for multiple servers



> -----Original Message-----
> From: dunlapg@xxxxxxxxx [mailto:dunlapg@xxxxxxxxx] On Behalf Of
> George Dunlap
> Sent: 18 March 2014 13:24
> To: Paul Durrant
> Cc: xen-devel@xxxxxxxxxxxxx; Ian Campbell
> Subject: Re: [Xen-devel] [PATCH v3 5/6] ioreq-server: add support for
> multiple servers
> 
> On Tue, Mar 18, 2014 at 11:33 AM, Paul Durrant <Paul.Durrant@xxxxxxxxxx>
> wrote:
> >> -----Original Message-----
> >> From: Ian Campbell
> >> Sent: 17 March 2014 14:56
> >> To: Paul Durrant
> >> Cc: xen-devel@xxxxxxxxxxxxx
> >> Subject: Re: [Xen-devel] [PATCH v3 5/6] ioreq-server: add support for
> >> multiple servers
> >>
> >> On Mon, 2014-03-17 at 14:52 +0000, Paul Durrant wrote:
> >> > > -----Original Message-----
> >> > > From: Ian Campbell
> >>
> >> > > It's always struck me as odd to have a Xen<->DM communication
> channel
> >> > > sitting there in guest pfn space (regardless of who the nominal owner
> >> > > is).
> >> > >
> >> > > I don't suppose there is any way to pull these pages out of the guest
> >> > > pfn space while still accounting them to the guest. Or if there is it
> >> > > would probably be a whole other kettle of fish than this series.
> >> > >
> >> >
> >> > The closest analogy I can think of accounting-wise would be shadow
> >> > pages. I'll have a look at how they are handled.
> >>
> >> I think the big difference is that noone outside Xen needs to be able to
> >> refer to a shadow page, whereas the device models need some sort of
> >> handle onto the ring to be able to map them etc. Not insurmountable I
> >> suppose.
> >>
> >
> > Probably not, but it's looking like it will be a bit of a can of worms. Are 
> > you
> ok with sticking to base+range HVM params for secondary emulators that can
> potentially be moved on migration for now? I.e. the save image just contains
> a count. There's still some growth room in the existing area (all pages from
> FE800000 to FF000000 AFACT) so as long as - as George said - we don't bake
> the PFN layout in, I don't think we preclude moving the emulator PFNs
> around in future.
> 
> xentrace has to share pages between Xen and dom0; it just exposes an
> interface for dom0 to get the mfn and then maps those mfns.  Couldn't
> you do something similar?  When you create an ioreq server, Xen could
> allocate the pages internally; and then you could use
> hvm_get_ioreq_server_info to get the MFNs, and xc_map_foreign_range()
> to map them.  Am I missing something?
> 

If you use xc_map_foreign_range() then presumably the page is still in the p2m. 
AFAIK the value supplied to xc_map_foreign_range() is still a guest frame 
number isn't it?

  Paul

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

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