From: David Laight
> Sent: 16 December 2022 13:30
>
> From: Leizhen (ThunderTown)
> > Sent: 16 December 2022 12:02
> >
> ..
> > > Moving the m68k version inside lib/string.c makes the test pass, too.
> > > So it must be related to the function being inline, and gcc making
> > > (incorrect) assumptions...
> >
> > Yes, it's the compiler's fault. I just replied David Laight:
> >
> > I added 'volatile' to prevent compiler optimizations, and it's OK now.
> >
> > diff --git a/arch/m68k/include/asm/string.h b/arch/m68k/include/asm/string.h
> > index f759d944c449940..3db81e5a783c72a 100644
> > --- a/arch/m68k/include/asm/string.h
> > +++ b/arch/m68k/include/asm/string.h
> > @@ -42,9 +42,9 @@ static inline char *strncpy(char *dest, const char *src, size_t n)
> > #define __HAVE_ARCH_STRCMP
> > static inline int strcmp(const char *cs, const char *ct)
> > {
> > - char res;
> > + signed char res;
> >
> > - asm ("\n"
> > + asm volatile ("\n"
> > "1: move.b (%0)+,%2\n" /* get *cs */
> > " cmp.b (%1)+,%2\n" /* compare a byte */
> > " jne 2f\n" /* not equal, break out */
>
> Adding 'volatile' there shouldn't make any real difference.
>
> I'd double-check the asm constraints for the two pointers.
> They are modified by the asm, but the caller's variables
> must not be changed.
>
> I think that means they need to be normal 'input' parameters
> and the result must be in different register (early clobber?).
> Currently the pointers are "+r" - which I think means they
> are input-output and any caller-supplied variable is
> likely to get changed.
Definitely badly broken.
'cs' and 'ct' should be input parameters.
'res' needs to be an early-clobber output parameter "=&r".
Since it is actually a 'static inline' (not just a #define)
then letting cs/ct be changed probably doesn't matter.
But the lack of 'early clobber' will cause grief.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
