Re: [Xen-devel] [PATCH] xen: arm: clarify cacheability requirements of hypercall arguments.

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Tue, 17 Dec 2013, Ian Campbell wrote:
> On Tue, 2013-12-17 at 12:29 +0000, Stefano Stabellini wrote:
> > On Tue, 17 Dec 2013, Ian Campbell wrote:
> > > Accepting hypercall arguments which are either consistently in cached or
> > > uncached is tricky and/or potentially slow, requiring a guest mapping 
> > > lookup
> > > to determine whether/when to do a cache clean or invalidate.
> > > 
> > > There are very few reasons, and no current use cases in practice, for a 
> > > guest
> > > to use uncached memory for their hypercall arguments. Therefore mandate 
> > > that
> > > all hypercall arguments must be mapped inner-cacheable.
> > > 
> > > If use cases arise then we can consider specific exemptions to this rule.
> > > 
> > > Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
> > >
> > >  xen/include/public/arch-arm.h |    3 +++
> > >  1 file changed, 3 insertions(+)
> > > 
> > > diff --git a/xen/include/public/arch-arm.h b/xen/include/public/arch-arm.h
> > > index 475cb4a..7b95237 100644
> > > --- a/xen/include/public/arch-arm.h
> > > +++ b/xen/include/public/arch-arm.h
> > > @@ -58,6 +58,9 @@
> > >   * (AAPCS64). Where there is a conflict the 64-bit standard should be
> > >   * used regardless of guest type. Structures which are passed as
> > >   * hypercall arguments are always little endian.
> > > + *
> > > + * All hypercall arguments passed via a pointer to guest memory must
> > > + * reside in memory which is mapped as Normal Inner-cacheable.
> > 
> > Maybe it is good to be pedantic: should we write "Inner or Outer and
> > Inner" cacheable?
> 
> I think it only matters that it is inner cacheable, since that is where
> Xen's accesses will hit. Whether or not it is outercacheable doesn't
> matter to us, does it?

I agree, I don't think that outer cacheable matters to us, I was just
pointing out that we should probably write that down to avoid any
confusion.


> > Should we mention explicitly that "Write-Back", "Write-Through" or
> > "Write-Back Write-Allocate" are all supported?
> 
> I think we don't care strategy is used, do we?

Same as above.
I think we should write down that the cache strategy doesn't matter (or
a list of cache strategies that we intend to support).

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