Re: [Qemu-devel] [PATCH 3/5] target/hppa: fix log conditions

[Richard Henderson]

On 2/11/19 10:19 AM, Sven Schnelle wrote:
>      switch (cf >> 1) {
> -    case 4: case 5: case 6:
> -        cf &= 1;
> +    case 0: /* never */
> +        cond = cond_make_f();
> +        break;
> +    case 1: /* = all bits are zero */
> +        cond = cond_make_0(TCG_COND_EQ, res);
> +        break;
> +    case 2: /* < leftmost bit is 1 */
> +        cond = cond_make_0(TCG_COND_LT, res);
> +        break;
> +    case 3: /* <= leftmost bit is 1 or all bits 0 */
> +        cond = cond_make_0(TCG_COND_LE, res);
> +        break;
> +    case 7: /* OD rightmost bit is 1 */
> +        tmp = tcg_temp_new();
> +        tcg_gen_andi_reg(tmp, res, 1);
> +        cond = cond_make_0(TCG_COND_NE, tmp);
> +        tcg_temp_free(tmp);
> +        break;
> +    default:
>          break;
>      }

You can't do nothing for cases 4,5,6.  That lets a bad guest crash qemu, since
cond will be uninitialized.  Also, this patch has to be sorted before the
previous, as otherwise you introduce a regression during bisection.

I've fixed this up locally.


r~