Patch from Lennert Buytenhek

The routine that nwfpe uses for converting floats/doubles to
extended precision fails to zero two bytes of kernel stack.  This
is not immediately obvious, as the floatx80 structure has 16 bits
of implicit padding (by design.)  These two bytes are copied to
userspace when an stfe is emulated, causing a possible info leak.

Make the padding explicit and zero it out in the relevant places.
Signed-off-by: NLennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: NRussell King <rmk+kernel@arm.linux.org.uk>