Re: [Qemu-devel] [PATCH v3 6/7] pcspk: Convert to qdev

[Anthony Liguori <aliguori@us.ibm.com>]

On 01/31/2012 04:48 PM, Jan Kiszka wrote:
> On 2012-01-31 23:40, Anthony Liguori wrote:
>> Why is what's in the tree not usable?
>>
>> Just don't do pcspk_init as a static inline (which is not that nice to
>> do anyway) and you don't need to worry about the availability of an
>> accessor.
>
> The current pattern requested by some reviewers used to be the one I
> applied here. I dislike it as well when the device can't be seriously
> configured out. But I can switch over, no problem.
>
>>
>> And BTW, there is strict type checking now, which makes it already an
>> improvement over property pointers.
>
> OK, for my slow understanding: I use qdev_property_add_link in the
> device init function to create the property, letting it point to a state
> field. But what should I call from the outside to actually set its
> value?

You have a few options:

1) you can add a generic set_child function to qdev... but don't do this, I'll 
fix it right in my future series

2) You can take advantage of the fact that the only thing that ever sets this is 
pcspk_init, move pcspk_init to pcspk.c, and then since you have access to the 
state structure, directly assign it there.

3) Introduce a pcspk_set_pit() function that just does the same thing as (2).

4) Introduce (1) and then have a pcspk_set_pit() that just calls (1).  This is 
what I think we should do moving forward and I plan to do this in my future 
refactorings.

That all said, I think this isn't going to work for you until my next series is 
merged.  Right now, the link properties have strict type checking.  You really 
need a link<PITCommon> in order to be able to accept either a KVMPIT or a normal 
PIT.  My series on the list relaxes the type checking to use implements instead 
of a strict equality check.

So maybe the best thing to do is drop the ptr type entirely, make pcspk_init() 
be in pcspk.c and touch the private state, and then I'll refactor it later to 
use a link<> property.

> And what C type does this value have?

PITState *.

Regards,

Anthony Liguroi

  A DeviceState * or a char *
> (path)?
>
> Jan