On 04/11/15 14:29, Ian Campbell wrote: >>>>> This looses out on the arm32 hypervisor sanity checking that the >>>>> padding >>>>> bytes are 0 (as required by the ABI) but TBH I haven't checked that >>>>> the >>>>> current version has that property either. >>>> >>>> It's done during the assignation by the compiler: >>>> >>>> (hnd).q = (uint64_t)(uintptr_t)(val); >>> >>> I meant on the reading side. >> >> It's the responsibility of the caller to zero the padding. There is >> nothing to do on the reading side, the hypervisor will use "p" which >> will be the size of the natural pointer. > > For a 32-bit Xen the check would be that a guest was not inadvertently > violating this rule, such a guest would crash if it was run on a 64-bit > hypervisor (which would see the non-zero padding as part of the pointer), > by rejecting such cases on 32-bit Xen we avoid such guests becoming > established and therefore presenting a case for us to relax this rule in > one way or another.
It would add overhead each time we want to copy to/from the guest memory. TBH, I don't think it's our business to check if the guest properly filling out the structure. It could happen only when the guest decides to implement it's own set_guest_handle_macro. As long as we don't crash the hypervisor it's fine. Regards, -- Julien Grall _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
