Re: [PATCH 00/13] migration: Threadify loadvm process

[Peter Xu <peterx@redhat.com>]

On Wed, Oct 22, 2025 at 03:25:59PM -0400, Peter Xu wrote:
> This is v1, however not 10.2 material.  The earliest I see fit would still
> be 11.0+ even if everything goes extremely smooth.
> 
> Removal of RFC is only about that I'm more confident this should be able to
> land without breaking something too easily, as I smoked it slightly more
> cross-archs this time.  AFAIU the best (and possibly only..) way to prove
> it solid is to merge it.. likely in the early phase of a dev cycle.
> 
> The plan is we'll try to get to more device setups too soon, before it
> could land.

I forgot to attach a changelog, sorry, here it is:

rfc->v1:
- Collected tags
- Try to split the major patch 5 to smaller one [Dave]
  - One with WITH_BQL_*() macros, rewritten to allow internal "return" or
    nesting [Vladimir]
  - One patch "migration: Pass in bql_held information from
    qemu_loadvm_state()", the patch itself contain no functional change.
    However it should contain all changes relevant to BQL, hence maybe it
    would help review.
  - One patch contains the rest that need to be in one patch.
- Refine commit message for patch "migration/rdma: Change io_create_watch()
  to return immediately" [Fabiano]
- Pre-requisite patch to rework migration_channel_read_peek(), removing the
  sleep functions [Fabiano]
  - Added patch "io: Add qio_channel_wait_cond() helper"
  - Added patch "migration: Properly wait on G_IO_IN when peeking messages"
- Squashed patch "migration/rdma: Remove rdma_cm_poll_handler" into patch
  "migration: Thread-ify precopy vmstate load process", otherwise the patch
  when split can be racy.
- Changed the vfio patch to be at the end, instead of adding dead code
  directly remove the bql lock/unlock, making sure that API is invoked
  without BQL instead.
- Added one patch "migration: Allow blocking mode for incoming live
  migration", so that after the threadify work switching incoming main
  channel to be in blocking mode.  Reason in commit message.

It may not be complete, so there can be small touch ups here and there that
are not mentioned.

-- 
Peter Xu