On Mon, Nov 04, 2024 at 05:26:45PM +0530, Prasad Pandit wrote:
> On Fri, 1 Nov 2024 at 20:09, Peter Xu <pet...@redhat.com> wrote:
> > > + if (migrate_multifd()) {
> > > + RAMBlock *block = pss->block;
> > > + /*
> > > + * While using multifd live migration, we still need to handle
> > > zero
> > > + * page checking on the migration main thread.
> > > + */
> > > + if (migrate_zero_page_detection() == ZERO_PAGE_DETECTION_LEGACY)
> > > {
> > > + if (save_zero_page(rs, pss, offset)) {
> > > + return 1;
> > > + }
> > }
> > There's one more save_zero_page() below. Please consider properly merging
> > them.
>
> if (save_zero_page(rs, pss, offset)) {
> return 1;
> }
>
> * First is called in migrate_multifd() mode, the second (above) is
> called in non-multifd mode. Will check how/if we can conflate them.
Yes, IMHO it's better when merged.
One more note here, that even with ZERO_PAGE_DETECTION_MULTIFD, qemu will
fallback to use LEGACY in reality when !multifd before. We need to keep
that behavior.
>
> > > migration_ops = g_malloc0(sizeof(MigrationOps));
> > > + migration_ops->ram_save_target_page = ram_save_target_page_common;
> >
> > If we want to merge the hooks, we should drop the hook in one shot, then
> > call the new function directly.
> >
>
> * Ie. drop the 'migration_ops' object altogether? And call
> ram_save_target_page() as it used to be before multifd mode?
Yes.
Thanks,
--
Peter Xu
