Il 29/05/2013 14:16, Michael S. Tsirkin ha scritto: >>>> > >> If you really want to use offsetof like this you're >>>> > >> going to need to decorate the structs with QEMU_PACKED. >> > >>> > > Nope. >>> > > These structs are carefully designed not to have any padding. >> > >> > ...on every compiler and OS combination that QEMU builds for? > Yes. All the way back to EGCS and before. > GCC has been like this for many many years. I would still prefer to have QEMU_PACKED (or __attribute((__packed__)) in the kernel). >>> > > And if there was a bug and there was some padding, we still >>> > > can't fix it with PACKED because this structure >>> > > is used to interact with the guest code which does not >>> > > have the packed attribute. >> > >> > The guest code has to use a set of structure offsets and >> > sizes which is fixed by the virtio ABI -- how it implements >> > this is up to the guest (and if it misimplements it that is >> > a guest bug and not our problem). On the other hand, encouraging both userspace and guest to reuse a single set of headers means that the bad offset becomes a spec bug more than a guest bug. > Deviating from driver in random ways is an endless source > of hard to debug issues, and it's a very practical > consideration because virtio spec is constantly > being extended (unlike hardware which is mostly fixed). Agreed---but the driver should use __attribute__((__packed__)) too. Paolo _______________________________________________ Virtualization mailing list Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linuxfoundation.org/mailman/listinfo/virtualization
