Re: [Xen-devel] [PATCH v3 4/6] xen: introduce XEN_GUEST_HANDLE_PARAM

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

On Fri, 17 Aug 2012, Ian Campbell wrote:
> On Fri, 2012-08-17 at 14:58 +0100, Jan Beulich wrote:
> > >>> On 17.08.12 at 15:47, Stefano Stabellini 
> > >>> <stefano.stabellini@xxxxxxxxxxxxx>
> > wrote:
> > > On Fri, 17 Aug 2012, Jan Beulich wrote:
> > >> >>> On 17.08.12 at 10:02, Ian Campbell <Ian.Campbell@xxxxxxxxxx> wrote:
> > >> > On Thu, 2012-08-16 at 18:10 +0100, Stefano Stabellini wrote:
> > >> >> On Thu, 16 Aug 2012, Jan Beulich wrote:
> > >> >> > >>> On 16.08.12 at 17:54, "Jan Beulich" <JBeulich@xxxxxxxx> wrote:
> > >> >> > > Seeing the patch I btw realized that there's no easy way to
> > >> >> > > avoid having the type as a second argument in the conversion
> > >> >> > > macros. Nevertheless I still don't like the explicitly specified 
> > >> >> > > type
> > >> >> > > there.
> > >> >> > 
> > >> >> > Btw - on the architecture(s) where the two handles are identical
> > >> >> > I would prefer you to make the conversion functions trivial (and
> > >> >> > thus avoid making use of the "type" parameter), thus allowing
> > >> >> > the type checking to occur that you currently circumvent.
> > >> >> 
> > >> >> OK, I can do that.
> > >> > 
> > >> > Will this result in the type parameter potentially becoming stale?
> > >> > 
> > >> > Adding a redundant pointer compare is a good way to get the compiler to
> > >> > catch this. Smth like;
> > >> > 
> > >> >         /* Cast a XEN_GUEST_HANDLE_PARAM to XEN_GUEST_HANDLE */
> > >> >         #define guest_handle_from_param(hnd, type) ({
> > >> >             typeof((hnd).p) _x = (hnd).p;
> > >> >             XEN_GUEST_HANDLE(type) _y;
> > >> >             &_y == &_x;
> > >> >             hnd;
> > >> >          })
> > >> 
> > >> Ah yes, that's a good suggestion.
> > >> 
> > >> > I'm not sure which two pointers of members of the various structs need
> > >> > to be compared, maybe it's actually &_y.p and &hnd.p, but you get the
> > >> > idea...
> > >> 
> > >> Right, comparing (hnd).p with _y.p would be the right thing; no
> > >> need for _x, but some other (mechanical) adjustments would be
> > >> necessary.
> > > 
> > > The _x variable is still useful to avoid multiple evaluations of hnd,
> > > even though I know that this is not a public header.
> > 
> > But we had settled on returning hnd unmodified when both
> > handle types are the same.
> > 
> > > What about the following:
> > > 
> > > /* Cast a XEN_GUEST_HANDLE to XEN_GUEST_HANDLE_PARAM */
> > > #define guest_handle_to_param(hnd, type) ({                \
> > >     typeof((hnd).p) _x = (hnd).p;                          \
> > >     XEN_GUEST_HANDLE_PARAM(type) _y = { _x };              \
> > >     if (&_x != &_y.p) BUG();                               \
> > >     _y;                                                    \
> > > })
> > 
> > Since this is not a public header, something like this (untested,
> > so may not compile as is)
> > 
> > #define guest_handle_to_param(hnd, type) ({                \
> >     (void)(typeof((hnd).p)0 == (XEN_GUEST_HANDLE_PARAM(type){}).p); \
> >     (hnd);                                                    \
> > })
> > 
> > is what I was thinking of.
> 
> This evaluates hnd twice, or do we only care about that in public
> headers for some reason? (personally I think principal of least surprise
> suggests avoiding it wherever possible)

the ARM version evaluates hnd only once and would work for x86 too

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