Re: [Qemu-devel] [PATCH 00/35] hw/ reorganization, part 2

[Paolo Bonzini <pbonzini@redhat.com>]

Il 20/03/2013 00:44, Peter Maydell ha scritto:
> On 19 March 2013 23:35, Paolo Bonzini <pbonzini@redhat.com> wrote:
>> > Il 19/03/2013 23:34, Peter Maydell ha scritto:
>>> >> On 19 March 2013 22:23, Paolo Bonzini <pbonzini@redhat.com> wrote:
>>>> >>> So, okay to put these in hw/arm and then I'll work on patches moving
>>>> >>> cpu_arm_init to a*mpcore.c?
>>> >>
>>> >> Wrong way round. If you can't put the cpu_arm_init into the a*mpcore
>>> >> in a way that doesn't make you want to put them in hw/arm/ then
>>> >> it should wait until we've QOMified the CPU cores sufficiently
>>> >> that we can do it properly.
>> >
>> > Does that include calling the CPU constructor something else than
>> > cpu_arm_init (which is defined in target-arm/)?  For me that would be
>> > enough to put it in hw/arm.
> The CPU should be created (and thus initialised and realised) the same
> way as any other QOM object or device; the containers shouldn't need
> to call cpu_arm_init directly.

ARM is already quite good in that respect.  However, until all
architectures are converted cpu_*_init needs to remain because of
user-mode targets (where the CPUs are created by common code, there is
no board to encapsulate the target-specific differences).  IMHO waiting
for the demise of cpu_*_init is putting the cart before the horse, or
another similar proverb.

(Also, I don't see much difference between using a function in
target-ARCH and using a TYPE_FOO define from target-ARCH.  They are the
same thing masked through RTTI.  hw/ARCH should really be the bridge
between target-ARCH and hw/everything-else, and a*mpcore.c fits in that
description).

Paolo