Re: [Qemu-devel] [PATCH 1/2] cpu/apic: drop icc bus/bridge/

[Igor Mammedov <imammedo@redhat.com>]

On Thu, 19 Mar 2015 18:08:32 +0800
Chen Fan <chen.fan.fnst@cn.fujitsu.com> wrote:

> 
> On 03/18/2015 12:29 AM, Igor Mammedov wrote:
> > On Tue, 17 Mar 2015 17:08:30 +0800
> > Chen Fan <chen.fan.fnst@cn.fujitsu.com> wrote:
> >
> >> ICC bus was invented only to provide hotplug capability to
> >> CPU and APIC because at the time being hotplug was available only for
> >> BUS attached devices.
> >>
> >> Now this patch is to drop ICC bus impl, and switch to bus-less
> >> CPU+APIC hotplug, handling them in the same manner as pc-dimm.
> >>
> >> Signed-off-by: Chen Fan <chen.fan.fnst@cn.fujitsu.com>
> >> ---
> > [...]
> >
> >> diff --git a/hw/intc/apic_common.c b/hw/intc/apic_common.c
> >> index 0858b45..020219e 100644
> >> --- a/hw/intc/apic_common.c
> >> +++ b/hw/intc/apic_common.c
> >> @@ -27,6 +27,8 @@
> >>   static int apic_irq_delivered;
> >>   bool apic_report_tpr_access;
> >>   
> >> +static MemoryRegion *apic_container;
> >> +
> >>   void cpu_set_apic_base(DeviceState *dev, uint64_t val)
> >>   {
> >>       trace_cpu_set_apic_base(val);
> >> @@ -314,8 +316,11 @@ static void apic_common_realize(DeviceState *dev, Error **errp)
> >>       info = APIC_COMMON_GET_CLASS(s);
> >>       info->realize(dev, errp);
> >>       if (!mmio_registered) {
> >> -        ICCBus *b = ICC_BUS(qdev_get_parent_bus(dev));
> >> -        memory_region_add_subregion(b->apic_address_space, 0, &s->io_memory);
> >> +        memory_region_add_subregion(apic_container, 0, &s->io_memory);
> >> +        memory_region_add_subregion_overlap(CPU(s->cpu)->as->root,
> >> +                                            APIC_DEFAULT_ADDRESS,
> >> +                                            apic_container,
> >> +                                            0x1000);
> >>           mmio_registered = true;
> >>       }
> > Provided per CPU address spaces are used,
> > why do you need static apic_container and mmio_registered thing?
> > Shouldn't it be possible to register &s->io_memory of each APIC
> > in CPU's own AS and avoid using globals?
> it seems only need to do
> memory_region_add_subregion_overlap(CPU(s->cpu)->as->root,
>                                              APIC_DEFAULT_ADDRESS,
>                                              &io_memory,
>                                              0x1000);
> 
> at here, right?
yep, something along theses lines.

I also do not remember if we really need it for every APIC device
that inherits from APIC_COMMON.It might be that kvm_iqrchip don't need it at all.
You  could experiment with -machine kernel_irqchip=on/off options
to switch between QEMU emulated APIC and kernel_irqchip.

Also this change is independent of dropping icc_bridge,
so please split it out into a separate patch. 


> 
> Thanks,
> Chen
> >
> >>   
> >> @@ -431,15 +436,23 @@ static Property apic_properties_common[] = {
> >>       DEFINE_PROP_END_OF_LIST(),
> >>   };
> >>   
> >> +static void apic_common_initfn(Object *obj)
> >> +{
> >> +    if (!apic_container) {
> >> +        apic_container = g_malloc(sizeof(*apic_container));
> >> +        memory_region_init(apic_container, obj, "apic-container",
> >> +                           APIC_SPACE_SIZE);
> >> +    }
> >> +}
> >> +
> >
> >
> > .
> >
>