On Thu, Jan 05, 2023 at 10:17:24PM +0900, Tetsuo Handa wrote:
> On 2023/01/05 20:54, Daniel Vetter wrote:
> >>> . Plain memset() in arch/x86/include/asm/string_64.h is redirected to __msan_memset()
> >>> but memsetXX() are not redirected to __msan_memsetXX(). That is, memory initialization
> >>> via memsetXX() results in KMSAN's shadow memory being not updated.
> >>>
> >>> KMSAN folks, how should we fix this problem?
> >>> Redirect assembly-implemented memset16(size) to memset(size*2) if KMSAN is enabled?
> >>>
> >>
> >> I think the easiest way to fix it would be disable memsetXX asm
> >> implementations by something like:
> >>
> >> -------------------------------------------------------------------------------------------------
> >> diff --git a/arch/x86/include/asm/string_64.h b/arch/x86/include/asm/string_64.h
> >> index 888731ccf1f67..5fb330150a7d1 100644
> >> --- a/arch/x86/include/asm/string_64.h
> >> +++ b/arch/x86/include/asm/string_64.h
> >> @@ -33,6 +33,7 @@ void *memset(void *s, int c, size_t n);
> >> #endif
> >> void *__memset(void *s, int c, size_t n);
> >>
> >> +#if !defined(__SANITIZE_MEMORY__)
> >> #define __HAVE_ARCH_MEMSET16
> >> static inline void *memset16(uint16_t *s, uint16_t v, size_t n)
> >> {
> >> @@ -68,6 +69,7 @@ static inline void *memset64(uint64_t *s, uint64_t
> >> v, size_t n)
> >> : "memory");
> >> return s;
> >> }
> >> +#endif
> >
> > So ... what should I do here? Can someone please send me a revert or patch
> > to apply. I don't think I should do this, since I already tossed my credit
> > for not looking at stuff carefully enough into the wind :-)
> > -Daniel
> >
> >>
> >> #define __HAVE_ARCH_MEMMOVE
> >> #if defined(__SANITIZE_MEMORY__) && defined(__NO_FORTIFY)
> >> -------------------------------------------------------------------------------------------------
> >>
> >> This way we'll just pick the existing C implementations instead of
> >> reinventing them.
> >>
>
> I'd like to avoid touching per-arch asm/string.h files if possible.
>
> Can't we do like below (i.e. keep asm implementations as-is, but
> automatically redirect to __msan_memset()) ? If yes, we could move all
> __msan_*() redirection from per-arch asm/string.h files to the common
> linux/string.h file?
Oh I was more asking about the fbdev patch. This here sounds a lot more
something that needs to be discussed with kmsan people, that's definitely
not my area.
-Daniel
>
> diff --git a/include/linux/string.h b/include/linux/string.h
> index c062c581a98b..403813b04e00 100644
> --- a/include/linux/string.h
> +++ b/include/linux/string.h
> @@ -360,4 +360,15 @@ static __always_inline size_t str_has_prefix(const char *str, const char *prefix
> return strncmp(str, prefix, len) == 0 ? len : 0;
> }
>
> +#if defined(__SANITIZE_MEMORY__) && defined(__NO_FORTIFY)
> +#undef memset
> +#define memset(dest, src, count) __msan_memset((dest), (src), (count))
> +#undef memset16
> +#define memset16(dest, src, count) __msan_memset((dest), (src), (count) << 1)
> +#undef memset32
> +#define memset32(dest, src, count) __msan_memset((dest), (src), (count) << 2)
> +#undef memset64
> +#define memset64(dest, src, count) __msan_memset((dest), (src), (count) << 3)
> +#endif
> +
> #endif /* _LINUX_STRING_H_ */
>
>
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
