Oh, and the implication is that devices should fall into a certain schema, and as long as the configuration loading the checkpoint has objects in the same places using the same schemas it should just work. Checking for that could even be an automatic part of loading a checkpoint.
Gabe On Mon, Jun 1, 2020 at 10:18 PM Gabe Black <gabebl...@google.com> wrote: > Hi folks. Another idea I had which I don't have time to look into is to > have more rigorously defined contents and names for things in checkpoints. > Right now, we just dump in whatever we have lying around in whatever way is > convenient, occasionally causing checkpoint incompatibilities and making it > hard for some models to interoperate with others which happen to have > slightly different state which is more convenient to dump in some other way. > > What I think we should do is have a well defined set of state which is > what is required at an architectural level for any given object, for > instance a CPU implementing a particular ISA, a system device like a disk, > a memory, or an interrupt controller. Conforming checkpoints should express > the objects within them in only that way, and conforming objects should > only create conforming checkpoints. > > There are a lot of other ideas I have for how to improve checkpoints, some > of which I've sent to the list in the past, but this is one I'm running > into a lot recently. > > Gabe > _______________________________________________ gem5-dev mailing list -- gem5-dev@gem5.org To unsubscribe send an email to gem5-dev-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
