On 22.02.2013, at 17:31, Andreas Färber wrote:
> Am 22.02.2013 15:23, schrieb Alexander Graf:
>>
>> On 18.02.2013, at 10:16, Andreas Färber wrote:
>>
>>> diff --git a/target-ppc/kvm.c b/target-ppc/kvm.c
>>> index 2c64c63..e601059 100644
>>> --- a/target-ppc/kvm.c
>>> +++ b/target-ppc/kvm.c
>>> @@ -1263,7 +1263,7 @@ static void kvmppc_host_cpu_initfn(Object *obj)
>>>
>>> assert(kvm_enabled());
>>>
>>> - if (pcc->info->pvr != mfpvr()) {
>>> + if (pcc->pvr != mfpvr()) {
>>> fprintf(stderr, "Your host CPU is unsupported.\n"
>>> "Please choose a supported model instead, see -cpu ?.\n");
>>> exit(1);
>>> @@ -1275,30 +1275,38 @@ static void kvmppc_host_cpu_class_init(ObjectClass *oc, void *data)
>>> PowerPCCPUClass *pcc = POWERPC_CPU_CLASS(oc);
>>> uint32_t host_pvr = mfpvr();
>>> PowerPCCPUClass *pvr_pcc;
>>> - ppc_def_t *spec;
>>> uint32_t vmx = kvmppc_get_vmx();
>>> uint32_t dfp = kvmppc_get_dfp();
>>>
>>> - spec = g_malloc0(sizeof(*spec));
>>> -
>>> pvr_pcc = ppc_cpu_class_by_pvr(host_pvr);
>>> if (pvr_pcc != NULL) {
>>> - memcpy(spec, pvr_pcc->info, sizeof(*spec));
>>> + pcc->pvr = pvr_pcc->pvr;
>>> + pcc->svr = pvr_pcc->svr;
>>> + pcc->insns_flags = pvr_pcc->insns_flags;
>>> + pcc->insns_flags2 = pvr_pcc->insns_flags2;
>>> + pcc->msr_mask = pvr_pcc->msr_mask;
>>> + pcc->mmu_model = pvr_pcc->mmu_model;
>>> + pcc->excp_model = pvr_pcc->excp_model;
>>> + pcc->bus_model = pvr_pcc->bus_model;
>>> + pcc->flags = pvr_pcc->flags;
>>> + pcc->bfd_mach = pvr_pcc->bfd_mach;
>>> +#ifdef TARGET_PPC64
>>> + pcc->sps = pvr_pcc->sps;
>>> +#endif
>>> + pcc->init_proc = pvr_pcc->init_proc;
>>> + pcc->check_pow = pvr_pcc->check_pow;
>>
>> It would be nice to have field copying more streamlined. This way, whoever adds a new field to the class needs to know that he also has to change this piece of code, which is non-obvious.
>>
>> Speaking of which, why aren't you copying parent_reset for example?
>
> parent_reset is already assigned by the .parent's class_init before this
> class_init is executed.
>
>> Or asked differently: Why can't we do a memcpy? We're really trying to do a subclass of the parent class here, no?
>
> I did suggest making it a subclass in the cover letter, as follow-up. :)
>
> The issue is we need to know which parent class. And we do not have any
> guarantee that in ..._register_types() the types corresponding to our
> PVR have already been registered.
>
> Therefore we would need to move host CPU type registration to
> kvm_arch_init(), as suggested by Eduardo for x86. A side effect would be
> that the type is not yet registered at -cpu ? time. If that is
> acceptable to you (we might hard-code its output within CONFIG_KVM), I
> can send you a patch.
Yes, I think that's the most reasonable way forward. We can always print it explicitly in -cpu ?.
Alex
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
