[0/6] migration: Fix issues during qmp_migrate_cancel

Message

Fabiano Rosas Dec. 2, 2024, 10:01 p.m. UTC

Hi,

While working on downstream issues with postcopy, I ended up writing a
set of tests for issuing qmp_migrate_cancel() at various points during
the migration. That exposed some bugs, which this series attempts to
fix.

There is also a fix for the issue Daniel found:

https://gitlab.com/qemu-project/qemu/-/issues/2633

I'm also sending the test code. It creates one test per
MIGRATION_STATUS_ state. Each test starts a migration, waits for that
specific state to be reached, issues qmp_migrate_cancel() and checks
that the migration state changes to cancelled (for now only cancelling
migration from the source side).

I was initially worried that this would be too racy, but so far each
test has survived 1000 iterations. I'm thinking it's worth merging,
specially because even after working on this I haven't been able to
clear the questions we have in our todo list [1], so we'll probably
need more work around this area in the future.

1- https://wiki.qemu.org/ToDo/LiveMigration#Migration_cancel_concurrency

CI run: https://gitlab.com/farosas/qemu/-/pipelines/1569870481

Thanks

Fabiano Rosas (6):
  tests/qtest/migration: Introduce migration_test_add_suffix
  migration: Kick postcopy threads on cancel
  migration: Fix postcopy listen thread exit
  migration: Make sure postcopy recovery doesn't hang when cancelling
  migration: Fix hang after error in destination setup phase
  tests/qtest/migration: Add a cancel test

 migration/channel.c             |  11 +-
 migration/migration.c           |  58 +++++---
 migration/migration.h           |   2 +-
 migration/postcopy-ram.c        |  14 +-
 migration/savevm.c              |  60 ++++----
 tests/qtest/migration-helpers.c |  24 ++++
 tests/qtest/migration-helpers.h |   2 +
 tests/qtest/migration-test.c    | 243 ++++++++++++++++++++++++++++++++
 8 files changed, 365 insertions(+), 49 deletions(-)