Re: [XenPPC] [rfc] [patch] 64-bitize guest handles

[Hollis Blanchard <hollisb@xxxxxxxxxx>]

On Wed, 2006-06-28 at 15:02 -0400, Jimi Xenidis wrote:
> On Jun 28, 2006, at 1:10 PM, Hollis Blanchard wrote:
> 
> > On Wed, 2006-06-28 at 06:34 -0400, Jimi Xenidis wrote:
> >>
> >>    #define set_xen_guest_handle(hnd, val) \
> >>        do { \
> >>            if (sizeof ((hnd).__pad)) \
> >>                (hnd).__pad = 0; \
> >>            (hnd).p = val; \
> >>        } while (0)
> >
> > I like the union better than a struct with carefully-constructed
> > non-overlapping members.
> 
> hmm, really! Ok, then you will have to also fix:
>    #define get_xen_guest_handle(val, hnd)  do { val = (hnd).p; }  
> while (0)
> 
> or the 32bit caller will only get the high-order 32bit into val.

Oh, yeah. I remember getting that wrong last time this came up, too.

> > Also, the separate assignment to 0 would be avoided.
> 
> In 32 bits the store in of a long long is still 2 stw instructions,  
> and the conditional will get optimized away so the instruction  
> sequence is equivalent.
> 
> I'll leave it to you.

OK, I guess I'll just go with the struct hackery (though I think it's
ugly as hell :) .

Of course, before committing I was hoping to find out from you if it
helps with the memory_reservation problem you were having. Does it?

-- 
Hollis Blanchard
IBM Linux Technology Center


_______________________________________________
Xen-ppc-devel mailing list
Xen-ppc-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ppc-devel