On 16/09/15 19:14, Richard Henderson wrote:
> On 09/14/2015 05:45 AM, Leon Alrae wrote:
>> -static inline void gen_mfc0_load64 (TCGv arg, target_ulong off)
>> -{
>> - tcg_gen_ld_tl(arg, cpu_env, off);
>> - tcg_gen_ext32s_tl(arg, arg);
>> -}
>> -
>> static inline void gen_mtc0_store32 (TCGv arg, target_ulong off)
>> {
>> TCGv_i32 t0 = tcg_temp_new_i32();
>> @@ -4972,17 +4966,19 @@ static void gen_mfc0(DisasContext *ctx, TCGv arg,
>> int reg, int sel)
>> break;
>> case 4:
>> CP0_CHECK(ctx->insn_flags & ASE_MT);
>> - gen_mfc0_load64(arg, offsetof(CPUMIPSState, CP0_YQMask));
>> + tcg_gen_ld32s_tl(arg, cpu_env, offsetof(CPUMIPSState,
>> CP0_YQMask));
>> rn = "YQMask";
>> break;
>
> This change is broken for 64-bit guest and big-endian host -- one has to
> adjust
> the offset in that case. I suspect that's why the extension was separate to
> begin with...
Uh, I think I wasn't expecting that ld32s_i64 actually does 32-bit load.
It means that CP0_UserLocal needs fixing in current code.
Thanks,
Leon
