On Mon, Oct 06, 2014 at 03:42:01PM +0200, Paolo Bonzini wrote: > Il 02/10/2014 15:49, Michael S. Tsirkin ha scritto: > >>> The issue is that incoming migration might have a different > >>> fw_cfg size from what we have. > >> > >> Understood now. > >> > >>> I think migrating this value will solve the issue in a cleaner way. > >> > >> Perhaps. The question is whether it would complicate the > >> forwards-migration code beyond what is sane. I think we are practically > >> speaking stuck with RAM. > > > > Migrating RAM size is actually useful too, I think someone asked for it. > > Migrating RAM size was discussed for BIOS and option ROMs, in order to > support migration from old versions of QEMU. It was floated around for > some time, but ultimately we ended up shipping two copies of affected > firmware (128k/256k BIOS, and non-EFI/EFI option ROMs). > > For BIOS it wouldn't be enough, because the BIOS size affects the memory > map. Of course ACPI tables aren't mapped anywhere, but I'd be wary of > adding code to migration that is half-broken and almost never used. > > Also, RAM blocks that have different size would be yet another thing > that makes machine types "almost compatible" with the QEMU version > they're supposed to represent. In a scenario similar to John's, with > mutable RAM sizes, would have likely broken all machine types, because > we would not have bothered doing full backwards-compatibility. > > I'm not an advocate of backwards bug-compatibility, but I think RAM > block sizes are way beyond the line of what we should be allowed to > modify between machine types. > > Paolo
Maybe we should just modify ACPI and rom files in general to use something else, not RAM? It looked like a good fit initially so we went ahead with it, but these things are fairly small, so it's not a problem to migrate them as part of the device state. -- MST
