Re: [Qemu-devel] Why is TYPE_CPU no-user?

[Markus Armbruster]

Andreas Färber <address@hidden> writes:

> Hi Markus,
>
> Am 15.10.2013 14:24, schrieb Markus Armbruster:
>> To go beyond RFC with this series, I need to explain why TYPE_CPU
>> cannot_instantiate_with_device_add_yet.  Would you be so kind and help
>> me out with a suitable comment?
>
>>From what I remember this was done when I started the whole process and
> most CPU subtypes did not yet use the QOM instance_init for
> initialization. Most importantly x86 still is not yet self-contained,
> nor is sparc. Such targets need to use cpu_init() et al. rather than
> -device. (This became visible in the first s390x vCPU hotplug series.)
>
> Most boards rely on being able to do postprocessing after they have
> instantiated the CPU: wiring up IRQs, adding reset handlers, halting
> non-first CPUs, ...
> -device would skip that.

This is the prime reason for cannot_instantiate_with_device_add_yet.

> Another aspect is that no CPU subtype has been proven hot-pluggable with
> device_add yet. For s390x we're the closest to date.

Hot-plug support is orthogonal.  We got plenty of devices that support
device-add, but only cold plug.

> We could move the flag from the base type to the targets' base types if
> you prefer. Then we can knock the bad values out one by one rather than
> overriding the inherited value with an explicit positive one.

I'd prefer to make that move when the first CPU is ready.

Thanks!

[...]