On Sat, Sep 06, 2025 at 07:18:14AM +0200, Richard Henderson wrote:
> We need not call tb_flush once per cpu, only once per vmload.
> Move the call from cpu_common_post_load to a tcg-specific
> vm_change_state_handler.

The change looks correct.  Though the commit message here implies a
conversion from per-cpu flush to per-system flush, which IMHO is only part
of the change.

IIUC, the change is better than that, because previously when the flush is
in post_load(), it means the per-cpu flush will also happen in a live
migration, but here AFAIU the flush is only needed through a HMP
loadvm-styled migration, where there must be some stale data in the first
place. In case of live migrations there isn't any prior data to flush,
iiuc.

Maybe it would be nice to mention the other side of the changes in the
commit message too?  No strong feelings, though.

Thanks,

-- 
Peter Xu


Reply via email to