>>>> void icp_set_cppr(ICPState *ss, uint8_t cppr)
>>>> @@ -262,7 +261,7 @@ void icp_set_cppr(ICPState *ss, uint8_t cppr)
>>>> }
>>>> } else {
>>>> if (!XISR(ss)) {
>>>> - icp_resend(ss);
>>>> + icp_resend(XICS_INTERFACE(qdev_get_machine()), ss);
>>>
>>> Here you're assuming that the machine is the implementor of the xics
>>> interface, which is kinda ugly. The ICP should have a pointer to the
>>> xics interface, which will eventually replace the pointer to the
>>> overall xics object it has now.
>>
>> yes. I will try improve that. I don't like those calls to
>> qdev_get_machine()either.
>>
>> There are done in a couple of places though, under spapr_cpu_core
>> to get XICS for instance.
>
> Right, but I'm happier with it there, in code that's definitely
> associated with a particular machine, rather than in the xics code
> which is at least somewhat reusable.
I fixed that with the backlink on the XICSFabric.
C.
