Re: [Qemu-devel] Re: [PATCH 11/11] test-vmstate: add test case to verify we don't change VMState

[Jan Kiszka]

On 2011-03-23 17:27, Anthony Liguori wrote:
> On 03/23/2011 10:00 AM, Peter Maydell wrote:
>> On 23 March 2011 14:52, Anthony Liguori<address@hidden>  wrote:
>>> I think we ought to merge VMStateDescription into DeviceInfo.  For
>>> compatibility, we probably need a vmstate_alias name since the device names
>>> don't always map 1-1 with the qdev names.  But this should eliminate the
>>> problem of reusing VMStateDescriptions for multiple devices.
>> That's a feature, not a bug. Consider eg hw/pl110.c -- there
>> are two different DeviceInfo devices but since the underlying
>> implementation is the same you definitely don't want to have
>> two separate VMStateDescription structures to get out of sync.
> 
> No, it's a bug.
> 
> Migration uses the VMStateDescription name as a section identifier.  The 
> section identifiers MUST be unique for a given device.  Otherwise, if 
> both devices are present, migration fails miserably.  It also means that 
> if the wrong devices are created on the destination, instead of 
> predictable failure, you get unpredictable guest corruption.

This is true for incompatible devices, but not for those that are
perfectly exchangeable (like in-kernel vs. user-space irqchip).

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux