Re: [Qemu-devel] [PATCH] lsi: implement basic SBCL functionality

[Paolo Bonzini]

On 04/03/19 11:19, Peter Maydell wrote:
> On Fri, 15 Feb 2019 at 09:38, Paolo Bonzini <address@hidden> wrote:
>>
>> On 14/02/19 23:04, Sven Schnelle wrote:
>>> @@ -2202,6 +2224,7 @@ static const VMStateDescription vmstate_lsi_scsi = {
>>>          VMSTATE_UINT8(stime0, LSIState),
>>>          VMSTATE_UINT8(respid0, LSIState),
>>>          VMSTATE_UINT8(respid1, LSIState),
>>> +        VMSTATE_UINT8(sbcl, LSIState),
>>
>> This breaks live migration.  You need to bump the version number higher
>> in vmstate_lsi_scsi and use "VMSTATE_UINT8_V(sbcl, LSIState, 1)" so that
>> the field is only marshalled/unmarshalled for version 1 of the migration
>> state.
> 
> I thought we preferred to do this with migration subsections
> rather than versioning fields these days ? In this case I
> think a subsection that says "needed if sbcl is non-zero" would
> do the right thing.

Yes, we prefer migration subsections but the only advantage is support
for backwards migration.  LSI is not a device for which I would
guarantee that backwards migration works at all, and it is not the
default controller for any board whose machine types are versioned.

Paolo