Cédric Le Goater <c...@kaod.org> writes:
> Richard,
>
> On 12/9/21 16:05, Fabiano Rosas wrote:
>> Cédric Le Goater <c...@kaod.org> writes:
>>
>>> On 12/9/21 00:06, Fabiano Rosas wrote:
>>>> We cannot have TCG code in powerpc_excp because the function is called
>>>> from kvm-only code via ppc_cpu_do_interrupt:
>>>>
>>>> ../target/ppc/excp_helper.c:463:29: error: implicit declaration of
>>>> function ‘cpu_ldl_code’ [-Werror=implicit-function-declaration]
>>>>
>>>> Fortunately, the Alignment interrupt is not among the ones dispatched
>>>> from kvm-only code, so we can keep it out of the disable-tcg build for
>>>> now.
>>>>
>>>> Fixes: 336e91f853 ("target/ppc: Move SPR_DSISR setting to powerpc_excp")
>>>> Signed-off-by: Fabiano Rosas <faro...@linux.ibm.com>
>>>>
>>>> ---
>>>>
>>>> Perhaps we could make powerpc_excp TCG only and have a separate
>>>> function that only knows the two interrupts that we use with KVM
>>>> (Program, Machine check). But for now this fix will do, I think.
>>>> ---
>>>> target/ppc/excp_helper.c | 2 ++
>>>> 1 file changed, 2 insertions(+)
>>>>
>>>> diff --git a/target/ppc/excp_helper.c b/target/ppc/excp_helper.c
>>>> index 17607adbe4..dcf22440cc 100644
>>>> --- a/target/ppc/excp_helper.c
>>>> +++ b/target/ppc/excp_helper.c
>>>> @@ -453,6 +453,7 @@ static inline void powerpc_excp(PowerPCCPU *cpu, int
>>>> excp_model, int excp)
>>>> }
>>>> break;
>>>> }
>>>> +#ifdef CONFIG_TCG
>>>> case POWERPC_EXCP_ALIGN: /* Alignment exception
>>>> */
>>>> /*
>>>> * Get rS/rD and rA from faulting opcode.
>>>> @@ -464,6 +465,7 @@ static inline void powerpc_excp(PowerPCCPU *cpu, int
>>>> excp_model, int excp)
>>>> env->spr[SPR_DSISR] |= (insn & 0x03FF0000) >> 16;
>>>> }
>>>> break;
>>>> +#endif
>>>> case POWERPC_EXCP_PROGRAM: /* Program exception
>>>> */
>>>> switch (env->error_code & ~0xF) {
>>>> case POWERPC_EXCP_FP:
>>>>
>>>
>>> Shouldn't we move that code under ppc_cpu_do_unaligned_access ?
>>
>> Well, it came from there initially. We could revert 336e91f853 and that
>> would fix the issue as well.
>
> What would you prefer ?
Well none of this interfere with the work I'm doing, so it really makes
no difference. I guess reverting the patch is cleaner than having an
ifdef loose in the middle of the code. I'll send a v2 with the revert.
>
> Thanks,
>
> C.
