Re: [Qemu-block] [Xen-devel] [PATCH 2/4] block/xen_disk: remove use of grant map/unmap

[Paul Durrant]

> -----Original Message-----
> From: Roger Pau Monne
> Sent: 30 April 2018 16:28
> To: Paul Durrant <address@hidden>
> Cc: address@hidden; address@hidden; qemu-
> address@hidden; Anthony Perard <address@hidden>; Kevin
> Wolf <address@hidden>; Stefano Stabellini <address@hidden>; Max
> Reitz <address@hidden>
> Subject: Re: [Xen-devel] [PATCH 2/4] block/xen_disk: remove use of grant
> map/unmap
> 
> On Mon, Apr 30, 2018 at 04:16:52PM +0100, Paul Durrant wrote:
> > > -----Original Message-----
> > > From: Roger Pau Monne
> > > Sent: 30 April 2018 16:12
> > > To: Paul Durrant <address@hidden>
> > > Cc: address@hidden; address@hidden; qemu-
> > > address@hidden; Anthony Perard <address@hidden>;
> Kevin
> > > Wolf <address@hidden>; Stefano Stabellini <address@hidden>;
> Max
> > > Reitz <address@hidden>
> > > Subject: Re: [Xen-devel] [PATCH 2/4] block/xen_disk: remove use of
> grant
> > > map/unmap
> > >
> > > On Mon, Apr 30, 2018 at 01:01:37PM +0100, Paul Durrant wrote:
> > > > The grant copy operation was added to libxengnttab in Xen 4.8.0. If
> grant
> > > > copy is available then data from the guest will be copied rather than
> > > > mapped.
> > > > The xen_disk source can be significantly simplified by removing this
> now
> > > > redundant code.
> > >
> > > Hm, I know this is a PITA, but FreeBSD gntdev hasn't implemented the
> > > grant-copy operation yet.
> > >
> > > I could try to implement it, but I can't make any promises on the time
> > > ATM, since I'm quite busy.
> > >
> >
> > I guess we could carry a compat patch in QEMU that implements grant copy
> by doing a map/memcpy/unmap , but QEMU feels like the wrong place for
> that. I could try putting together a similar patch for the freebsd.c component
> of libxengnttab in the xen source rather than it simply failing with ENOSYS as
> it does now. Would either of those help?
> 
> Maybe this could be implemented in gnttab_core.c, so it can also be
> used by MiniOS and Linux versions not supporting the copy ioctl as a
> fallback?

That sounds like a reasonable idea. I'll put something together so that it can 
go in early in 4.12.

  Paul

> 
> Roger.