Re: [Qemu-devel] [RFC v4 00/12] ISA reconfigurability v4

[Blue Swirl <blauwirbel@gmail.com>]

On Tue, Jun 14, 2011 at 12:09 AM, Andreas Färber <andreas.faerber@web.de> wrote:
> Am 13.06.2011 um 22:08 schrieb Blue Swirl:
>
>> On Wed, Jun 8, 2011 at 9:55 PM, Andreas Färber <andreas.faerber@web.de>
>> wrote:
>>>
>>> I've refined the series to track the state in ISADevice and to expose it
>>> as VMState.
>>> Error handling has been improved, and setting the state multiple times is
>>> no-op now.
>>>
>>> To read the state, I'm introducing support for bool qdev properties.
>>> Some more qdev_prop_get_*() helpers are introduced, too.
>>>
>>> Still need to do some runtime testing, but I'd like to hear if this is
>>> getting
>>> mergeable now, especially wrt VMState.
>>>
>>> Andreas
>>>
>>>
>>> Andreas Färber (11):
>>>  qdev: Add support for property type bool
>>>  qdev: Add helpers for reading properties
>>>  isa: Provide set_state callback
>>>  isa: Allow to un-assign I/O ports
>>>  isa: Allow to un-associate an IRQ
>>
>> I like the patches above.
>>
>> But I think the set_state() interface could be improved. For example,
>> cpu_register_io_memory() gives an index which is passed to
>> sysbus_register_mmio(). Then the board can instantiate the device at
>> desired location without caring about the device internals. With
>> set_state(), the device does everything.
>
> Thanks. On IRC, Juan proposed to replace set_state with enable and disable,
> that's what I hope to post together with VMStateSubsections tonight, as part
> of the large PReP series.
>
> Your proposal I don't understand yet. The ioport handlers are
> device-specific, so must be registered from within the device. We discussed
> consolidating that into helpers at ISA level to avoid one call, Gleb further
> suggested converting to IORange. We can't do that declaratively since some
> are conditional.

Yes, I would have wanted something declarative. But maybe ISA is special then.