Re: live-migration performance regression when using pmem

[Peter Xu <peterx@redhat.com>]

On Thu, Jun 12, 2025 at 03:34:35PM +0000, Chaney, Ben wrote:
> On 5/14/25, 9:59 AM, "David Hildenbrand" <david@redhat.com <mailto:david@redhat.com>> wrote:
> 
> 
> 
> >Because we need to make sure the pmem kind memory data is synced
> >after migration, we choose to call pmem_persist() when the migration
> >finish. This will make sure the data of pmem is safe and will not
> >lose if power is off.
> 
> Thank you for clarifying. I think I initially misunderstood the purpose of
> this code path.
> 
> In that case, how about something like this to restrict the sync to only
> run when it would be effective? If the memory region is volatile then
> there is no benefit to syncing.
> 
> Thanks,
>         Ben
> 
> ---
> migration/ram.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/migration/ram.c b/migration/ram.c
> index f25ebd9620..24fb29f0a5 100644
> --- a/migration/ram.c
> +++ b/migration/ram.c
> @@ -3930,7 +3930,9 @@ static int ram_load_cleanup(void *opaque)
>      RAMBlock *rb;
> 
>      RAMBLOCK_FOREACH_NOT_IGNORED(rb) {
> -        qemu_ram_block_writeback(rb);
> +        if (rb->mr->nonvolatile) {
> +            qemu_ram_block_writeback(rb);
> +        }
>      }
> 
>      xbzrle_load_cleanup();
> --
> 2.40.1

Looks good here, I think that's what I mentioned:

https://lore.kernel.org/all/aCTHwhrXROReEPEh@x1.local/

But I guess I got the use case wrong; looks like it worked for you.

In that case please switch to memory_region_is_nonvolatile(), and add
proper Fixes, and copy stable.

Thanks,

-- 
Peter Xu