The inline version is used on three NOMMU architectures and is
particularly inefficient when it scans the string one byte at a time
twice. It also lacks a check for user_addr_max(), but this is
probably ok on NOMMU targets.

Consolidate the asm-generic implementation with the library version
that is used everywhere else.  This version is generalized enough to
work efficiently on both MMU and NOMMU targets, and using the
same code everywhere reduces the potential for subtle bugs.

Mark the prototypes as __must_check in the process.
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>