On Wed, Oct 23, 2024 at 02:02:12PM -0400, Peter Xu wrote:
> This is a follow up of below patch from Avihai as a replacement:
> 
> https://lore.kernel.org/qemu-devel/20241020130108.27148-3-avih...@nvidia.com/
> 
> This is v2 of the series, and it became a more generic rework on how we do
> migration object refcounts, so I skipped a changelog because most of this
> is new things.
> 
> To put it simple, now I introduced another pointer to migration object, and
> here's a simple explanation for both after all change applied (copy-paste
> from one of the patch):
> 
> /*
>  * We have two pointers for the global migration objects.  Both of them are
>  * initialized early during QEMU starts, but they have different lifecycles.
>  *
>  * - current_migration
>  *
>  *   This variable reflects the whole lifecycle of the migration object
>  *   (which each QEMU can only have one).  It is valid until the migration
>  *   object is destroyed.
>  *
>  *   This is the object that internal migration so far use.  For example,
>  *   internal helper migrate_get_current() references it.
>  *
>  *   When all migration code can always pass over a MigrationState* around,
>  *   this variable can logically be dropped.  But we're not yet there.
>  *
>  * - global_migration
>  *
>  *   This is valid only until the migration object is still valid to the
>  *   outside-migration world (until migration_shutdown()).
>  *
>  *   This should normally be always set, cleared or accessed by the main
>  *   thread only, rather than the migration thread.
>  *
>  *   All the exported functions (in include/migration) should reference the
>  *   exported migration object only to avoid race conditions, as
>  *   current_migration can be freed concurrently by migration thread when
>  *   the migration thread holds the last refcount.
>  */
> 
> It allows all misc.h exported helpers to be used for the whole VM
> lifecycle, so as to never crash QEMU with freed migration objects.
> 
> Thanks,
> 
> Peter Xu (4):
>   migration: Unexport dirty_bitmap_mig_init() in misc.h
>   migration: Reset current_migration properly
>   migration: Add global_migration
>   migration: Make all helpers in misc.h safe to use without migration
> 
>  include/migration/misc.h | 29 ++++++++----
>  migration/migration.h    |  4 ++
>  migration/migration.c    | 99 +++++++++++++++++++++++++++++++++++-----
>  3 files changed, 113 insertions(+), 19 deletions(-)

Sent too soon. This breaks device-introspect-test.. Sorry.  I'll look at
that and repost.

Meanwhile please still comment on the idea, especially when one disagrees.

-- 
Peter Xu


Reply via email to