[PATCH v3] riscv: lib: Fix ZBB strnlen wrap-around regression on huge counts

[Qingfang Deng]

Hi,

On 2026/9/14 16:21, shao.mingyin at zte.com.cn wrote:
> diff --git a/arch/riscv/lib/strnlen.S b/arch/riscv/lib/strnlen.S
> index a8911605c248..528ee7995969 100644
> --- a/arch/riscv/lib/strnlen.S
> +++ b/arch/riscv/lib/strnlen.S
> @@ -67,7 +67,7 @@ strnlen_zbb:
>   	 *   a1 - Max length of string
>   	 *
>   	 * Clobbers
> -	 *   t0, t1, t2, t3, t4
> +	 *   t0, t1, t2, t3, t4, t5
>   	 */
>
>   	/* If maxlen is 0, return 0. */
> @@ -87,9 +87,22 @@ strnlen_zbb:
>   	 * Aligned boundary.  Use the address of the last valid byte
>   	 * (s + count - 1) to avoid loading a word past the count
>   	 * boundary in the loop below.  count == 0 is handled above.
> +	 *
> +	 * Saturate the boundary when s + count wraps around (very large
> +	 * counts, e.g. SIZE_MAX passed by FORTIFY strcat/strlcat with a
> +	 * destination whose size is unknown at compile time).  Without
> +	 * this, the wrapped boundary lands before s and the pre-loop
> +	 * guard below always exits, returning a truncated length.
> +	 * Saturating makes the scan equivalent to strlen().
> +	 *
> +	 * Keep the saturation branchless so that the normal path does
> +	 * not carry a taken branch that could be mispredicted.
>   	 */
>   	add	t4, a0, a1
>   	addi	t4, t4, -1
> +	sltu	t5, t4, a0	/* Did s + count wrap around? */
> +	sub	t5, zero, t5	/* -1 if wrapped, 0 otherwise */
> +	or	t4, t4, t5	/* Saturate to the top of the address space */
You can save one instruction by using Zbb minu instruction.

Equation: saturating_add(s, count - 1) == s + min(count - 1, ~s), given 
that count >= 1.

     addi    t4, a1, -1
     not    t1, a0
     minu    t4, t4, t1
     add    t4, a0, t4
>   	andi	t4, t4, -SZREG
>
>   	/* Get the first word.  */
Kind regards,
Qingfang