Re: [Qemu-ppc] [PATCH 08/25] spapr: introduce a skeleton for the XIVE interrupt controller

[Cédric Le Goater]

>>>> +static const VMStateDescription vmstate_spapr_xive = {
>>>> +    .name = TYPE_SPAPR_XIVE,
>>>> +    .version_id = 1,
>>>> +    .minimum_version_id = 1,
>>>> +    .needed = vmstate_spapr_xive_needed,
>>>> +    .fields = (VMStateField[]) {
>>>> +        VMSTATE_UINT32_EQUAL(nr_irqs, sPAPRXive, NULL),
>>>> +        VMSTATE_STRUCT_VARRAY_UINT32_ALLOC(ivt, sPAPRXive, nr_irqs, 1,
>>>> +                                           vmstate_spapr_xive_ive, 
>>>> XiveIVE),  
>>>
>>> Hmm... this array is allocated at realize and this will cause
>>> the migration code to re-allocate it again with the same size,
>>> and leak memory IIUC.  
>>
>> I thought so but something was going wrong on the receive side (memory 
>> corruption detected by valgrind). I did not find why yet.
>>
> 
> Have you tried VMSTATE_STRUCT_VARRAY_POINTER_UINT32() ?

yes. tcg/intel only though.

C.