On 2022/12/15 18:36, Geert Uytterhoeven wrote:

The next line is:

        scr_memsetw(save, erase, array3_size(logo_lines, new_cols, 2));

So how can this turn out to be uninitialized later below?

        scr_memcpyw(q, save, array3_size(logo_lines, new_cols, 2));

What am I missing?

Good catch. It turned out that this was a KMSAN problem (i.e. a false positive report).

On x86_64, scr_memsetw() is implemented as

        static inline void scr_memsetw(u16 *s, u16 c, unsigned int count)
        {
                memset16(s, c, count / 2);
        }

and memset16() is implemented as

        static inline void *memset16(uint16_t *s, uint16_t v, size_t n)
        {
        	long d0, d1;
        	asm volatile("rep\n\t"
        		     "stosw"
        		     : "=&c" (d0), "=&D" (d1)
        		     : "a" (v), "1" (s), "0" (n)
        		     : "memory");
        	return s;
        }

. 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?