Re: [Qemu-devel] [PATCH 5/5] pseries: Move XICS initialization before cpu initialization

[Alexander Graf]

Am 16.03.2013 um 06:41 schrieb Benjamin Herrenschmidt <address@hidden>:

> On Sat, 2013-03-16 at 06:33 +0100, Alexander Graf wrote:
>>>> We're changing that notion in the in-kernel XICS discussions.  The
>> flow will look like this:
>>>> 
>>>> * create vcpus
>>>> * create XICS
>>>> * foreach (vcpu)
>>>>     * enable_cap(vcpu, CAP_XICS_SERVER, xics_handle)
>>> 
>>> This is stupid. Why have the VCPU initialize itself for non-kernel
>>> interrupts and *then* switch it over ?
>> 
>> Because non-kernel initialization is a nop.
> 
> And ?

Compared to the previous aproach of a vm wide irq type, we win:

  - less vm global state
  - possibility of defining the 'cpu nr' from the pic's point of view
  - the same mechanism and code paths for hotplug and non-hotplug
  - possibility of supporting multiple pics
  - less implicit assumptions

While the code changes required to make this work should be minimal over the 
"set irq architecture" approach. It really makes life easier for everyone.


Alex