On Wed, Dec 4, 2013 at 1:43 PM, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: > On Wed, Dec 04, 2013 at 12:46:42PM -0800, Colin Cross wrote: >> On Wed, Dec 4, 2013 at 10:35 AM, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: >> <snip> >> >> > And finally, is this all really needed? Why not just fix the structures >> > to be "correct", and then fix userspace to use the correct structures as >> > well, thereby not needing a compat layer at all? >> >> Some of the binder ioctls take userspace pointers. Are you suggesting >> storing those pointers in a __u64 to avoid having to have a >> compat_ioctl? > > Yes, that's the best way to solve the issue, right? It's the least code, but in exchange you lose all the type safety and warnings when copying in and out of the pointers, as well as sparse checking on the __user attribute. That doesn't seem like a good tradeoff to me. In addition it requires modifying the existing heavily used 32 bit api, which means a mostly-equivalent compat layer added in libbinder to support old kernels. I would suggest fixing the 32-bit structures to use fixed sizes where appropriate (__u32 instead of unsigned long) while maintaining compatibility, and then using compat_ioctl where necessary to handle pointers. _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
