Re: [RFC PATCH 2/4] mm/damon/paddr: Add DAMOS_INTERLEAVE action

From: David Hildenbrand
Date: Fri Jun 13 2025 - 09:46:36 EST


On 12.06.25 20:13, Bijan Tabatabai wrote:
From: Bijan Tabatabai <bijantabatab@xxxxxxxxxx>

This patch adds the DAMOS_INTERLEAVE action.
It interleaves pages inside of a given region according to the weights
in the iw_table. To reuse existing interleaving code, the target nid for
a folio is determined by calling policy_nodemask, therefore only folios
belonging to processes using the MPOL_WEIGHTED_INTERLEAVE policy will
have their pages migrated.

Below is an example of its usage where pages are initially interleaved at
a 1:1 ratio and then changed to be interleaved at a 2:1 ratio. The
alloc_data program simply allocates 1GB of data then sleeps.
$ cd /sys/kernel/mm/damon/admin/kdamonds/0
$ sudo cat ./contexts/0/schemes/0/action
interleave
$ echo 1 | sudo tee /sys/kernel/mm/mempolicy/weighted_interleave/node0
$ echo 1 | sudo tee /sys/kernel/mm/mempolicy/weighted_interleave/node1
$ numactl -w 0,1 ~/alloc_data 1G &
$ numastat -c -p alloc_data

Per-node process memory usage (in MBs) for PID 18473 (alloc_data)
Node 0 Node 1 Total
------ ------ -----
Huge 0 0 0
Heap 0 0 0
Stack 0 0 0
Private 514 514 1027
------- ------ ------ -----
Total 514 514 1028
$ echo 2 | sudo tee /sys/kernel/mm/mempolicy/weighted_interleave/node0
$ numastat -c -p alloc_data

Per-node process memory usage (in MBs) for PID 18473 (alloc_data)
Node 0 Node 1 Total
------ ------ -----
Huge 0 0 0
Heap 0 0 0
Stack 0 0 0
Private 684 343 1027
------- ------ ------ -----
Total 684 343 1027

Signed-off-by: Bijan Tabatabai <bijantabatab@xxxxxxxxxx>
---
Documentation/mm/damon/design.rst | 2 +
include/linux/damon.h | 2 +
mm/damon/paddr.c | 112 ++++++++++++++++++++++++++++++
mm/damon/sysfs-schemes.c | 1 +
4 files changed, 117 insertions(+)

diff --git a/Documentation/mm/damon/design.rst b/Documentation/mm/damon/design.rst
index ddc50db3afa4..c50d2105cea0 100644
--- a/Documentation/mm/damon/design.rst
+++ b/Documentation/mm/damon/design.rst
@@ -455,6 +455,8 @@ that supports each action are as below.
Supported by ``paddr`` operations set.
- ``migrate_cold``: Migrate the regions prioritizing colder regions.
Supported by ``paddr`` operations set.
+ - ``interleave``: Interleave the regions according to the weighted interleave weights.
+ Supported by ``paddr`` operations set.
- ``stat``: Do nothing but count the statistics.
Supported by all operations sets.
diff --git a/include/linux/damon.h b/include/linux/damon.h
index a4011726cb3b..81d26a203337 100644
--- a/include/linux/damon.h
+++ b/include/linux/damon.h
@@ -117,6 +117,7 @@ struct damon_target {
* @DAMOS_LRU_DEPRIO: Deprioritize the region on its LRU lists.
* @DAMOS_MIGRATE_HOT: Migrate the regions prioritizing warmer regions.
* @DAMOS_MIGRATE_COLD: Migrate the regions prioritizing colder regions.
+ * @DAMOS_INTERLEAVE: Interleave the regions by the weighted interleave ratio
* @DAMOS_STAT: Do nothing but count the stat.
* @NR_DAMOS_ACTIONS: Total number of DAMOS actions
*
@@ -136,6 +137,7 @@ enum damos_action {
DAMOS_LRU_DEPRIO,
DAMOS_MIGRATE_HOT,
DAMOS_MIGRATE_COLD,
+ DAMOS_INTERLEAVE,
DAMOS_STAT, /* Do nothing but only record the stat */
NR_DAMOS_ACTIONS,
};
diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
index 4102a8c5f992..e989464635cd 100644
--- a/mm/damon/paddr.c
+++ b/mm/damon/paddr.c
@@ -535,6 +535,114 @@ static unsigned long damon_pa_migrate(struct damon_region *r, struct damos *s,
return applied * PAGE_SIZE;
}
+#if defined(CONFIG_MEMCG) && defined(CONFIG_NUMA)
+struct damos_interleave_private {
+ struct list_head *folio_migration_list;
+ bool putback_lru;
+};
+
+static bool damon_pa_interleave_rmap(struct folio *folio, struct vm_area_struct *vma,
+ unsigned long addr, void *arg)
+{
+ struct mempolicy *pol;
+ struct task_struct *task;
+ pgoff_t ilx;
+ int target_nid;
+ struct damos_interleave_private *priv = arg;
+
+ task = rcu_dereference(vma->vm_mm->owner);
+ if (!task)
+ return true;
+
+ pol = get_task_policy(task);
+ if (!pol)
+ return true;

Why is this not using get_vma_policy(), which will fallback to the task policy in case there is no per-vma policy>

--
Cheers,

David / dhildenb