* Paolo Bonzini (pbonz...@redhat.com) wrote:
> This is a race that can happen when migrating TCG guests under load.
> It was introduced by the change to run vCPUs outside the big QEMU lock.

That does seem to fix the case I was seeing on x86;

rmmod kvm-intel
taskset -c 0 bash
export QTEST_QEMU_BINARY=$PWD/x86_64-softmmu/qemu-system-x86_64
for S in $(seq 1 10); do ./tests/migration-test & done

that would throw a :
'Memory content inconsistency'

most times (may have to increase the 10 to get it to do it
every time?)

Dave


> Paolo Bonzini (4):
>   memory: remove memory_region_test_and_clear_dirty
>   memory: hide memory_region_sync_dirty_bitmap behind
>     DirtyBitmapSnapshot
>   memory: introduce memory_global_after_dirty_log_sync
>   memory: unify loops to sync dirty log bitmap
> 
>  exec.c                       | 31 ++++++++++++++++++
>  hw/display/cg3.c             |  1 -
>  hw/display/exynos4210_fimd.c |  1 -
>  hw/display/framebuffer.c     |  1 -
>  hw/display/g364fb.c          |  1 -
>  hw/display/sm501.c           |  1 -
>  hw/display/tcx.c             |  2 --
>  hw/display/vga.c             |  6 ----
>  include/exec/memory.h        | 47 +++++++++------------------
>  memory.c                     | 76 
> +++++++++++++++++---------------------------
>  migration/ram.c              |  1 +
>  11 files changed, 76 insertions(+), 92 deletions(-)
> 
> -- 
> 2.14.3
> 
--
Dr. David Alan Gilbert / dgilb...@redhat.com / Manchester, UK

Reply via email to