Re: [RFC PATCH 0/4] mm/damon: add DAMOS_MIGRATE_{HOT,COLD} destination nodes and weights

[Bijan Tabatabai <bijan311@gmail.com>]

On Sat, 21 Jun 2025 10:31:27 -0700 SeongJae Park <sj@kernel.org> wrote:

> DAMOS_MIGRATE_{HOT,COLD} action can migrate pages to only single node.
> It could be useful to allow it migrates pages to multiple nodes with
> different weights.  A work for dynamic interleaving[1] is in progress.
> We also discussed this could be useful for memory tiering, e.g., when we
> want to move pages to multiple noes of same tier.
> 
> Extend the API and ABI interface for specifying the multiple destination
> nodes and their weights.
> 
> Note that this RFC is a prototype of the interface change for Bijan's
> interleaving work[1].  Hence this patch series is only implementing the
> interface part, not the real implementation of the migration behavior.
> 
> [1] https://lore.kernel.org/20250620180458.5041-1-bijan311@gmail.com
> [2] https://lore.kernel.org/20240408175228.91414-1-sj@kernel.org
> 
> SeongJae Park (4):
>   mm/damon: add struct damos_migrate_dest
>   mm/damon/core: add damos->migrate_dest field
>   mm/damon/sysfs-schemes: implement DAMOS action destinations directory
>   mm/damon/sysfs-schemes: set damos->migrate_dest
> 
>  include/linux/damon.h    |  29 ++++-
>  mm/damon/core.c          |   4 +
>  mm/damon/sysfs-schemes.c | 253 ++++++++++++++++++++++++++++++++++++++-
>  3 files changed, 282 insertions(+), 4 deletions(-)
> 
> 
> base-commit: 78745efafbb93197e476717385616ed57ea2df22
> -- 
> 2.39.5

Hi SeongJae,

Thanks for putting this patch together. Sorry for taking a while to
respond to it.

I am finishing up V3 of the interleave patchset using these patches. It
has mostly worked great, but I noticed that damos->migrate_dest was not
being updated in damos_commit(), so new weights would not be applied. This
meant that you could not update the interleave weights by committing the
damon state. I also saw that damos->target_nid was also not being updated
here. I will will have a patch fixing this with the V3 of the interleave
patchset that I will hopefully send out tomorrow.

Thanks,
Bijan