Re: [Qemu-devel] [PULL 07/15] q35: Check propery to determine if iommu is set

[Marcel Apfelbaum]

On 11/29/2015 08:22 PM, Bandan Das wrote:
> Hi Marcel,
>
> Marcel Apfelbaum <address@hidden> writes:
> ...
> > Maybe is too late, but this contradicts QEMU usage, as I understand
> Why late ? We can always revert it :)

Well, this is why I am so disappointment in myself that I didn't catch
this earlier. I *really* don't like doing this. :(

On the bright side this is a 're-factoring only' patch, so we can
take our time with it.

> > object_property_get_* should be used when we don't know object's type.
> My understanding is that it's not mandatory to use it only when type
> is unknown. Ofcourse, it makes it redundant when you do know the type.
>
> I tend to follow convention, I noticed another call to qdev_get_machine,
> and so opted for this. I am actually ok either way and don't prefer one way
> over the other.

This part actually makes sense. qdev_get_machine should be preferred over
current_machine global variable that should disappear (IMHO).

But once we have the machine as Object, we can simply cast it to machine
and get the field with MACHINE(qdev_get_machine())->iommu instead of
calling the property 'by name'.

And the wrapper machine_iommu was a "commodity method" requested by (some)
QOM guys who don't like calling the "object internals" in other files.

However since "iommmu" is a simple flag, I suppose we gain nothing from the 
wrapper.


Thanks,
Marcel

> > Why use "iommu" when you can simply call current_machine->iommu ?
> > (if you don't like the wrapper, which is pretty harmless in my opinion)
>
> > Thanks,
> > Marcel
> >
> > >            mch_init_dmar(mch);
> > >        }
> > >    }