Re: [Qemu-devel] [PATCH v2 3/8] fdc: Introduce fdctrl->phase

[Markus Armbruster]

"Dr. David Alan Gilbert" <address@hidden> writes:

> * Peter Maydell (address@hidden) wrote:
>> On 29 May 2015 at 11:34, Dr. David Alan Gilbert <address@hidden> wrote:
>> > It's the destination I'm worried about here, not the source; lets say
>> > you have two devices, a & b.  'a' gets serialised, but then 'b' finds
>> > it has to wait, so we return to running the source and sending pages
>> > across.   However the destination has already loaded the 'a' device state;
>> > so that when we serialise again we send a new 'a' device state'; I'm
>> > worrying here that the destination 'a' state loader would get upset
>> > trying to load the same state twice without somehow resetting it.
>> 
>> You would need to reset the system before resending device state,
>> I think. Device implementations rely on knowing that migration happens
>> into a freshly reset device, and we don't have any way of resetting
>> a single device, only the whole system at once.
>
> I don't think you can do a device (or certainly not a system) reset
> without resending the contents of RAM which was the whole point of
> not restarting the migration from the beginning.

What about resetting all the devices, but keeping contents of RAM, then
send only newly dirtied RAM plus complete device state?