[Qemu-devel] Re: [PATCH 0/3] New VMState table based load/save infrastructure

[Juan Quintela]

Gerd Hoffmann <address@hidden> wrote:
>   Hi,
>
>> - For structures, Kraxel suggested to use a VMSTATE_INCLUDE() instead of
>>    having to declare a new VMStateInfo struct.  As I already have the new
>>    struct code working, I sent it with the struct.  Thinking about how to
>>    implement the VMSTATE_INCLUDE() and which one is easier to use.
>
> Looked at your code and figured the two approaches are not that
> different.  I'd just put the pointer to the chained/included
> VMStateDescription directly into the VMStateField to avoid the (IMHO
> pointless) indirection via VMStateInfo.

Agreed that the appreaches are very similar, basically question of style.

> Also:  It would probably useful to have two include modes here:  One
> for embedded structs, i.e. this:
>
> struct dev_state {
>   PCIDevice dev;
> };
>
> ... and one for referenced structs, i.e. this:
>
> struct dev2_state {
>   PCIDevice *dev;
> };

Trying to do pcibus_save() at the moment, and arrived at the conclusion
that we need the VMSTATE_INCLUDE() like thing.  Basically arrays need to
be handled the same:

struct dev_state {
    int32_t a[5];
}

struct dev_state {
    int32_t *a;
}

I am working on getting something like:

STRUCT(...)
STRUCT_P()
ARRAY()
ARRAY_P()

You get the idea :)

Later, Juan.