Re: [Qemu-ppc] [PATCH 6/6] xics-kvm: Support for in-kernel XICS interrupt controller

[Alexey Kardashevskiy]

On 08/06/2013 08:12 PM, Andreas Färber wrote:
>> +    /* Call emulated XICS implementation for consistency */
>> +    assert(xics_info && xics_info->cpu_setup);
>> +    xics_info->cpu_setup(icp, cpu);
>> +}
>> +
>> +void xics_kvm_set_nr_irqs(XICSState *icp, uint32_t nr_irqs)
>> +{
>> +    icp->ics = ICS(object_new(TYPE_ICS_KVM));
>> +    object_property_add_child(OBJECT(icp), "ics", OBJECT(icp->ics), NULL);
>> +    icp->ics->icp = icp;
> 
> instance_init + object_initialize()?

Create ICS_KVM in instance_init of XICS_KVM and initialize it here? Why split?

> 
>> +    icp->nr_irqs = icp->ics->nr_irqs = nr_irqs;
>> +}
>> +
>> +void xics_kvm_set_nr_servers(XICSState *icp, uint32_t nr_servers)
>> +{
>> +    int i;
>> +
>> +    icp->nr_servers = nr_servers;
>> +
>> +    icp->ss = g_malloc0(icp->nr_servers*sizeof(ICPState));
>> +    for (i = 0; i < icp->nr_servers; i++) {
>> +        char buffer[32];
>> +        object_initialize(&icp->ss[i], TYPE_ICP_KVM);
>> +        snprintf(buffer, sizeof(buffer), "icp[%d]", i);
>> +        object_property_add_child(OBJECT(icp), buffer, OBJECT(&icp->ss[i]), 
>> NULL);
>> +    }
>> +}
>> +
>> +static void rtas_dummy(PowerPCCPU *cpu, sPAPREnvironment *spapr,
>> +                       uint32_t token,
>> +                       uint32_t nargs, target_ulong args,
>> +                       uint32_t nret, target_ulong rets)
>> +{
>> +    error_report("pseries: %s must never be called for in-kernel XICS\n",
>> +            __func__);
>> +}
>> +
>> +static void xics_kvm_realize(DeviceState *dev, Error **errp)
>> +{
>> +    KVMXICSState *icpkvm = KVM_XICS(dev);
>> +    XICSState *icp = XICS_COMMON(dev);
>> +    int i, rc;
>> +    Error *error = NULL;
>> +    struct kvm_create_device xics_create_device = {
>> +        .type = KVM_DEV_TYPE_XICS,
>> +        .flags = 0,
>> +    };
>> +
>> +    assert(kvm_enabled());
>> +    assert(kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS));
> 
> error_setg() - if device can be created without accel=kvm (which it
> looks as if it can) then you should just error out the nice way here.


I check kvm_enabled() (I thought I check both but was wrong, will fix it)
where I try to create XICS_KVM so we should not be here if that check failed.


-- 
Alexey