[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH RFC 00/11] migration/block: disk activation rewrite
From: |
Peter Xu |
Subject: |
[PATCH RFC 00/11] migration/block: disk activation rewrite |
Date: |
Tue, 3 Dec 2024 19:51:27 -0500 |
I started looking at this problem as a whole when reviewing Fabiano's
series, especially the patch (for a QEMU crash [1]):
https://lore.kernel.org/r/20241125144612.16194-5-farosas@suse.de
The proposed patch could work, but it's unwanted to add such side effect to
migration. So I start to think about whether we can provide a cleaner
approach, at least remove that "we must active the disk for migration"
dependency, because migration really don't need the disks to be active..
It's also a pure wish that, if bdrv_inactivate_all() could be benign to be
called even if all disks are already inactive. Then problem also gone.
After all, similar call on bdrv_activate_all() upon all-active disks is all
fine. I hope that wish could still be fair.
And when I was looking at that, I found more things spread all over the
place on disk activation. I decided to clean all of them up, while
hopefully fixing the QEMU crash [1] too.
So this is what I came up with as of today. Marking RFC as of now, just to
collect some feedbacks first. At least I'd like to go with one more patch
to deprecate late-block-active - not deprecating its function, but make it
always happen (which is the default as of now for Libvirt), which should
hopefully be migration-ABI-safe.
With the help of Fabiano's test cases, I at least am sure this series works
for the ping pong migrations, and all existing qtests.
Let me know, thanks.
[1] https://gitlab.com/qemu-project/qemu/-/issues/2395
Fabiano Rosas (4):
tests/qtest/migration: Move more code under only_target
tests/qtest/migration: Don't use hardcoded strings for -serial
tests/qtest/migration: Support cleaning up only one side of migration
tests/qtest/migration: Test successive migrations
Peter Xu (7):
migration: Add helper to get target runstate
migration/block: Make late-block-active the default
migration/block: Apply late-block-active behavior to postcopy
migration/block: Fix possible race with block_inactive
migration/block: Merge block reactivations for fail/cancel
migration/block: Extend the migration_block_* API to dest side
migration/block: Apply the migration_block_* API to postcopy
migration/migration.h | 33 ++++-
tests/qtest/migration-helpers.h | 2 +
migration/migration.c | 177 +++++++++++-----------
migration/savevm.c | 32 ++--
tests/qtest/migration-helpers.c | 8 +
tests/qtest/migration-test.c | 252 +++++++++++++++++++++++++-------
6 files changed, 344 insertions(+), 160 deletions(-)
--
2.47.0
- [PATCH RFC 00/11] migration/block: disk activation rewrite,
Peter Xu <=
- [PATCH RFC 01/11] migration: Add helper to get target runstate, Peter Xu, 2024/12/03
- [PATCH RFC 03/11] migration/block: Apply late-block-active behavior to postcopy, Peter Xu, 2024/12/03
- [PATCH RFC 02/11] migration/block: Make late-block-active the default, Peter Xu, 2024/12/03
- [PATCH RFC 06/11] migration/block: Extend the migration_block_* API to dest side, Peter Xu, 2024/12/03
- [PATCH RFC 05/11] migration/block: Merge block reactivations for fail/cancel, Peter Xu, 2024/12/03
- [PATCH RFC 04/11] migration/block: Fix possible race with block_inactive, Peter Xu, 2024/12/03
- [PATCH RFC 08/11] tests/qtest/migration: Move more code under only_target, Peter Xu, 2024/12/03
- [PATCH RFC 10/11] tests/qtest/migration: Support cleaning up only one side of migration, Peter Xu, 2024/12/03
- [PATCH RFC 09/11] tests/qtest/migration: Don't use hardcoded strings for -serial, Peter Xu, 2024/12/03