Re: [Qemu-devel] [PATCH v6 07/17] target/m68k: add chk and chk2

[Laurent Vivier]

Le 03/01/2018 à 22:52, Richard Henderson a écrit :
> On 01/02/2018 03:40 PM, Laurent Vivier wrote:
>> +void HELPER(chk)(CPUM68KState *env, int32_t val, int32_t ub)
>> +{
>> +    if (val < 0 || val > ub) {
>> +        CPUState *cs = CPU(m68k_env_get_cpu(env));
>> +
>> +        /* Recover PC and CC_OP for the beginning of the insn.  */
>> +        cpu_restore_state(cs, GETPC());
>> +
>> +        /* Adjust PC and FLAGS to end of the insn.  */
>> +        env->pc += 2;
>> +        helper_flush_flags(env, env->cc_op);
>> +        env->cc_n = val;
>> +
>> +        cs->exception_index = EXCP_CHK;
>> +        cpu_loop_exit(cs);
>> +    }
>> +}
>> +
> 
> I thought you said for 68040, N is always unset for val >= 0.
> That would suggest
> 
>   helper_flush_flags(env, env->cc_op);
>   env->cc_n = val;
>   if (val < 0 || val > ub) {
>     ...
>   }

ok, my though was it is better to not update the flag if it is not
needed (it should be undefined), but what you suggest is closer to the
real hardware so I will update it.

> Did you examine the real hw change to the other flags?

yes, C is modified, and the logic is:
  C = 0 <= ub ? val < 0 || ub < val : val < 0 && ub < val;
All other flags are not modified.

I'm going to update the patch to reflect the change of N and C by the
real hardware.

> Because they're officially undefined, which suggests
> 
>   env->cc_n = val;
>   env->cc_op = CC_OP_LOGIC;
> 
>> +void HELPER(chk2)(CPUM68KState *env, int32_t val, int32_t lb, int32_t ub)
>> +{
>> +    helper_flush_flags(env, env->cc_op);
>> +
>> +    env->cc_z = val != lb && val != ub;
>> +    env->cc_c = lb <= ub ? val < lb || val > ub : val > ub && val < lb;
>> +
>> +    if (env->cc_c) {
>> +        CPUState *cs = CPU(m68k_env_get_cpu(env));
>> +
>> +        cpu_restore_state(cs, GETPC());
>> +        env->cc_op = CC_OP_FLAGS;
> 
> A comment that we're reverting a change made during unwind would be helpful 
> here.

Ok

Thanks,
Laurent