Re: [PATCH] [RFC]: xenpaging: add command to to flush qemu buffer cache via VMCALL

[Tim Deegan <tim@xen.org>]

At 15:42 +0100 on 08 Mar (1331221336), Olaf Hering wrote:
> On Thu, Mar 08, Tim Deegan wrote:
> 
> > At 17:15 +0100 on 26 Feb (1330276512), Olaf Hering wrote:
> > > # HG changeset patch
> > > # User Olaf Hering <olaf@aepfle.de>
> > > # Date 1330272880 -3600
> > > # Node ID e17afed13a35989bdf422b75d44843bfd2fda0f4
> > > # Parent  3ace34b89ebe3bdd6afb0bca85b0c1270f99c1bb
> > > [RFC]: xenpaging: add command to to flush qemu buffer cache via VMCALL
> > > 
> > > Is the approach taken in this patch ok?
> > 
> > Can you explain a bit more about what you're trying to do?
> 
> It looked easy to "inject" a request to send a flush to ioemu in this
> way. But it turned out that this approach will not work.

I think it could be made to work, but it will need more changes than
this.  Since the current mechanism is really only intended for guest
hypercalls it would need to be updated so that the flag is checked in
more places (say, on every VMENTER, or maybe somewhere a little less hot
like on interrupt injection).

How synchronous to you need the flush to be, btw?  If you need to be
sure it's completed then you probably can't go through the hypervisor.

> So both old and new qemu need a change to accept the flush command in
> some way. For old qemu the xenstore approach looks best, for the new
> qemu I need to dig into QMP.

Righto.  Can you please make sure that whatever you do for new qemu
doesn't require xenpaging and qemu to be in the same VM?

Cheers,

Tim.