Re: [Qemu-ppc] [Qemu-devel] icbt on PPC440 after commit 01662f3e

[Philippe Mathieu-Daudé]

Cc'ing Richard

On 06/09/2018 09:07 AM, BALATON Zoltan wrote:
> On Sat, 2 Jun 2018, BALATON Zoltan wrote:
>> I get an invalid instruction exception for the icbt instruction on the
>> sam460ex machine (PPC440 core 460EX CPU) but I think this should not
>> raise exception as it's valid on BookE that this CPU should support.
>> Commit 01662f3e PPC: Implement e500 (FSL) MMU has introduced an
>> extended instruction type (type2) in PPC opcodes and changed
>> taget/ppc/translate_init.c as:
>>
>> @@ -9437,7 +9492,8 @@ static int create_ppc_opcodes (CPUPPCState *env,
>> const ppc_def_t *def)
>>
>>     fill_new_table(env->opcodes, 0x40);
>>     for (opc = opcodes; opc < &opcodes[ARRAY_SIZE(opcodes)]; opc++) {
>> -        if ((opc->handler.type & def->insns_flags) != 0) {
>> +        if (((opc->handler.type & def->insns_flags) != 0) ||
>> +            ((opc->handler.type2 & def->insns_flags2) != 0)) {
>>             if (register_insn(env->opcodes, opc) < 0) {
>>                 printf("*** ERROR initializing PowerPC instruction "
>>                        "0x%02x 0x%02x 0x%02x\n", opc->opc1, opc->opc2,
>>
>>
>> it also made this change in target/ppc/translate.c:
>>
>> -GEN_HANDLER(mbar, 0x1F, 0x16, 0x1a, 0x001FF801, PPC_BOOKE),
>> -GEN_HANDLER(msync, 0x1F, 0x16, 0x12, 0x03FFF801, PPC_BOOKE),
>> -GEN_HANDLER2(icbt_440, "icbt", 0x1F, 0x16, 0x00, 0x03E00001, PPC_BOOKE),
>> +GEN_HANDLER_E(mbar, 0x1F, 0x16, 0x1a, 0x001FF801,
>> +              PPC_BOOKE, PPC2_BOOKE206),
>> +GEN_HANDLER_E(msync, 0x1F, 0x16, 0x12, 0x03FFF801,
>> +              PPC_BOOKE, PPC2_BOOKE206),
>> +GEN_HANDLER2_E(icbt_440, "icbt", 0x1F, 0x16, 0x00, 0x03E00001,
>> +               PPC_BOOKE, PPC2_BOOKE206),
>>
>> If I revert this change to icbt_440 it works without exception but
>> according to the above it should also work for either type == BOOKE
>> _or_ type2 == BOOKE206 but I still get exception in that case. Any
>> idea why is this not working as expected?
> 
> This has gone for a while but then happened again and this time I've
> found that icbt seems to have two opcodes on PPC440 and got invalid
> instruction exception for 7c063a0c (1f-06-08-06) now. Not sure why I got
> it for other opcode or I was decoding it wrong that time but the problem
> I saw now could be fixed by
> 
> diff --git a/target/ppc/translate.c b/target/ppc/translate.c
> index b28e8b9..40bc02d 100644
> --- a/target/ppc/translate.c
> +++ b/target/ppc/translate.c
> @@ -6699,6 +6699,8 @@ GEN_HANDLER_E(mbar, 0x1F, 0x16, 0x1a, 0x001FF801,
>  GEN_HANDLER(msync_4xx, 0x1F, 0x16, 0x12, 0x03FFF801, PPC_BOOKE),
>  GEN_HANDLER2_E(icbt_440, "icbt", 0x1F, 0x16, 0x00, 0x03E00001,
>                 PPC_BOOKE, PPC2_BOOKE206),
> +GEN_HANDLER2(icbt_440, "icbt", 0x1F, 0x06, 0x08, 0x03E00001,
> +               PPC_440_SPEC),
>  GEN_HANDLER(lvsl, 0x1f, 0x06, 0x00, 0x00000001, PPC_ALTIVEC),
>  GEN_HANDLER(lvsr, 0x1f, 0x06, 0x01, 0x00000001, PPC_ALTIVEC),
>  GEN_HANDLER(mfvscr, 0x04, 0x2, 0x18, 0x001ff800, PPC_ALTIVEC),
> 
> Can anyone confirm that's correct? If so I'll include in my next series.
> 
> Regards,
> BALATON Zoltan
>