[Hypervisor Live Update] Notes from February 10, 2025

[Jason Gunthorpe]

On Tue, Feb 18, 2025 at 08:04:47PM -0800, David Rientjes wrote:

>  - the future of guestmemfs and what it becomes, including alignment so
>    prototyping can be done

IMHO we need a generic FDBOX sort of filesystem and the ability to put
guestmemfd, memfd and hugetlbfs (fd) into it. This would completely
replace the need for a special memory-only filesystem.

Most of it should be doable with the folio preservation bitmap scheme
I showed in the v5 thread and is not particularly special.

Ie for a memfd you take all the folios under it, preserve them in the
bitmap, store the folio phys_addr_t's in a linked list of 4k chunks
and put the top of the linked list in the FDBOX for that memfd.

>  - establishing an FSM for all of the various states that are agreed upon
>    with common language (when memory mappings can happen, what is
>    disallowed at certain stages)

I think this is very important, along with explanations where people
expect work to be performed.

Jason