Up to now PA-RISC could live with a trivial version of access_ok().
Our fault handlers can correctly handle fault cases.

But testcases showed that we need a better access check else we won't
always return correct errno failure codes to userspace.

Problem showed up during 32bit userspace tests in which writev() used a
32bit memory area and length which would then wrap around on 64bit
kernel.
Signed-off-by: NHelge Deller <deller@gmx.de>