On 11/12/2015 04:33 PM, Wen Congyang wrote:
> Imagine that migration_dirty_pages is slightly too small and we enter ram_save_iterate;
> ram_save_iterate now sends *all* it's pages, it decrements migration_dirty_pages for
> every page sent. At the end of ram_save_iterate, migration_dirty_pages would be negative.
> But migration_dirty_pages is *u*int64_t; so we exit ram_save_iterate,
> go around the main migration_thread loop again and call qemu_savevm_state_pending, and
> it returns a very large number (because it's actually a negative number), so we keep
> going around the loop, because it never gets smaller.
I don't know how to trigger the problem. I think store migration_dirty_pages in BitmapRcu
can fix this problem.
hi, David
It seem that it's not easy to reproduce this problem in my environment.
and the following 2 patches are to fix this issue, can you help to review and test.
thx
Li
|
0002-migration-use-rcu-to-protect-migration_dirty_pages.patch