Re: [Xen-devel] [PATCH RFC] xen/grant_table: deprecate GNTTABOP_unmap_and_replace

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Mon, 5 Aug 2013, Jan Beulich wrote:
> >>> On 22.07.13 at 12:22, Stefano Stabellini 
> >>> <stefano.stabellini@xxxxxxxxxxxxx> wrote:
> > On Mon, 22 Jul 2013, Stefano Stabellini wrote:
> >> On Sun, 21 Jul 2013, Ian Campbell wrote:
> >> > On Sun, 2013-07-21 at 18:34 +0100, Stefano Stabellini wrote:
> >> > > --- a/xen/include/public/grant_table.h
> >> > > +++ b/xen/include/public/grant_table.h
> >> > > @@ -303,12 +303,13 @@ typedef uint16_t grant_status_t;
> >> > >  #define GNTTABOP_transfer             4
> >> > >  #define GNTTABOP_copy                 5
> >> > >  #define GNTTABOP_query_size           6
> >> > > -#define GNTTABOP_unmap_and_replace    7
> >> > > +#define GNTTABOP_unmap_and_replace_legacy    7
> >> > >  #if __XEN_INTERFACE_VERSION__ >= 0x0003020a
> >> > >  #define GNTTABOP_set_version          8
> >> > >  #define GNTTABOP_get_status_frames    9
> >> > >  #define GNTTABOP_get_version          10
> >> > >  #define GNTTABOP_swap_grant_ref             11
> >> > > +#define GNTTABOP_unmap_and_replace    12
> >> > >  #endif /* __XEN_INTERFACE_VERSION__ */
> >> > 
> >> > You need an ifdef here so that users specifying an old
> >> > __XEN_INTERFACE_VERSION__ get the old hypercall under the non-legacy
> >> > name.
> >> 
> >> I guess I need to bump the version too?
> > 
> > Thinking twice about this, even though bumping the interface version
> > would be easy enough, that would prevent any backports of the "fixed"
> > hypercalls to older hypervisors.
> 
> Why?

Jan, I dropped this patch and any Xen changes from my new versions of
the series. I am doing all the changes only on the Linux side.
That means that we don't need a new hypervisor version to have safe
O_DIRECT with files on NFS.

However Roger is now introducing a new hypercall, similar to my new
version of GNTTABOP_unmap_and_replace, to allow batch unmaps without
multicalls. That should be a good performance improvement.

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