Re: [PATCH 11/18] target/i386: move 00-5F opcodes to new decoder

[Richard Henderson]

On 10/14/23 03:01, Paolo Bonzini wrote:
> +    [0x28] = X86_OP_ENTRY2(SUB, E,b, G,b),
> +    [0x29] = X86_OP_ENTRY2(SUB, E,v, G,v),
> +    [0x2A] = X86_OP_ENTRY2(SUB, G,b, E,b),
> +    [0x2B] = X86_OP_ENTRY2(SUB, G,v, E,v),
> +    [0x2C] = X86_OP_ENTRY2(SUB, 0,b, I,b),   /* AL, Ib */
> +    [0x2D] = X86_OP_ENTRY2(SUB, 0,v, I,z),   /* rAX, Iz */
> +    [0x2E] = {},
> +    [0x2F] = X86_OP_ENTRY0(DAS, chk(i64)),
> +
> +    [0x38] = X86_OP_ENTRY2(SUB, E,b, G,b, nowb),
> +    [0x39] = X86_OP_ENTRY2(SUB, E,v, G,v, nowb),
> +    [0x3A] = X86_OP_ENTRY2(SUB, G,b, E,b, nowb),
> +    [0x3B] = X86_OP_ENTRY2(SUB, G,v, E,v, nowb),
> +    [0x3C] = X86_OP_ENTRY2(SUB, 0,b, I,b, nowb),   /* AL, Ib */
> +    [0x3D] = X86_OP_ENTRY2(SUB, 0,v, I,z, nowb),   /* rAX, Iz */
> +    [0x3E] = {},
> +    [0x3F] = X86_OP_ENTRY0(AAS, chk(i64)),
...
> +    case X86_SPECIAL_NoWriteback:
> +        decode.op[0].unit = X86_OP_SKIP;
> +        break;
...
> +static void gen_SUB(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
> +{
> +    MemOp ot = decode->op[0].ot;
> +
> +    if (s->prefix & PREFIX_LOCK) {
> +        tcg_gen_neg_tl(s->T0, s->T1);
> +        tcg_gen_atomic_fetch_add_tl(s->cc_srcT, s->A0, s->T0,
> +                                    s->mem_index, ot | MO_LE);
> +        tcg_gen_sub_tl(s->T0, s->cc_srcT, s->T1);
> +    } else {
> +        tcg_gen_mov_tl(s->cc_srcT, s->T0);
> +        tcg_gen_sub_tl(s->T0, s->T0, s->T1);
> +    }
> +    prepare_update2_cc(decode, s, CC_OP_SUBB + ot);
> +}

Missing a check for CMP vs PREFIX_LOCK.
I'm not sure where the best place to put that.
Whether you use a chk bit or simply hard-code a check for (!lock || !skip).


r~