On 01/26/2012 01:12 PM, Peter Maydell wrote:
On 26 January 2012 19:00, Anthony Liguori<aligu...@us.ibm.com> wrote:
We need to modeled MemoryRegions and qemu_irq in QOM too.
+1 : this ought to let us get rid of SysBus...
MemoryRegions
shouldn't be that difficult. Our habit of passing qemu_irq's as arrays without
an explicit size will probably require some refactoring but in principle,
supporting irqs should be easy too.
I think that there are probably a lot of cases where we're using an array
of qemu_irqs now but should be using separately named signals of some sort
instead (particularly where we're using them for things which aren't actually
IRQs...)
I started hacking up a Pin object that used a Notifier. It's pretty easy to
plumb that to an existing qemu_irq so I think that's the way to go.
That way we could incrementally remove qemu_irq usage.
I started with this path but the pc initialization was so fubar that I ran into
too many problems. Now I think I can go back and do it again and it will be
more reasonable given this refactoring.
At a high level, a Pin object looks and feels like a qemu_irq. There's a
pin_raise, pin_set_level, etc. But there is also a pin_get_level() (it's
stateful) and there's a pin_add_level_change_notifier() which allows you to
register.
Pins are objects so they can be added to the composition tree which means they
can be addressed. If you have a truly unidirectional path, then you can just
use a child and link and connect them that way.
For a bidirectional path (where software controls the direction at run time),
you can have an intermediate Wire object which consists of two Pin links.
Regards,
Anthony Liguori
-- PMM