Skip to contentSkip to conversation

mm: fix node reclaim swappiness handling

PATCH In review Ridong 6 participants 17 messages Jul 16 – Jul 23

Subsystems (7)
lore
Share & tools
lore

Ridong

author
#0 lore raw

v2 cover mm: fix node reclaim swappiness handling

From: Ridong <[email protected]>

The per-node proactive reclaim interface (/sys/devices/system/node/nodeX/reclaim) accepts a swappiness parameter, but it is silently ignored when CONFIG_MEMCG is disabled. The root cause is that sc_swappiness() has separate implementations for CONFIG_MEMCG and !CONFIG_MEMCG, and the latter never checks proactive_swappiness.

Patch 1 moves mem_cgroup_swappiness() and vm_swappiness out of the public include/linux/swap.h into the mm-private mm/swap.h, and makes the helper handle both CONFIG_MEMCG and !CONFIG_MEMCG in a single inline function. This is a prerequisite for unifying sc_swappiness().

Patch 2 consolidates sc_swappiness() into a single definition that works regardless of CONFIG_MEMCG, fixing the node reclaim swappiness bug.

---

v1 -> v2:
- Move mem_cgroup_swappiness() and vm_swappiness to mm/swap.h instead of include/linux/memcontrol.h. Suggested by Barry Song.
- Correct fix tag for patch 2.

v1: https://lore.kernel.org/all/[email protected]/

Ridong Chen (2):
memcg: move mem_cgroup_swappiness and vm_swappiness to mm/swap.h
mm: vmscan: fix node reclaim ignoring swappiness parameter

 include/linux/memcontrol.h |  4 ++--
 include/linux/swap.h       | 19 -------------------
 mm/memcontrol.c            |  4 ++--
 mm/swap.h                  | 20 ++++++++++++++++++++
 mm/vmscan.c                | 19 +++++++------------
 5 files changed, 31 insertions(+), 35 deletions(-)
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH v2 1/2 memcg: move mem_cgroup_swappiness and vm_swappiness to mm/swap.h

From: Ridong Chen <[email protected]>

The per-memcg swappiness knob is v1-only; v2 always uses global vm_swappiness and ignores the per-cgroup field.

Both mem_cgroup_swappiness() and vm_swappiness are only used within mm/ (memcontrol.c, memcontrol-v1.c, vmscan.c), so move them out of the public include/linux/swap.h into the mm-private mm/swap.h. This keeps unrelated declarations out of include/linux/swap.h.

Guard memcg->swappiness with CONFIG_MEMCG_V1 as well, so v2-only kernels drop the unused field.

No functional change for v1; v2-only kernels drop the unused field.

---
 include/linux/memcontrol.h |  4 ++--
 include/linux/swap.h       | 19 -------------------
 mm/memcontrol.c            |  4 ++--
 mm/swap.h                  | 20 ++++++++++++++++++++
 4 files changed, 24 insertions(+), 23 deletions(-)
include/linux/memcontrol.h +2 -2
index 957260677678..e7a794dfb40e 100644
@@ -239,8 +239,6 @@ struct mem_cgroup {
 	 */
 	bool oom_group;
 
-	int swappiness;
-
 	/* memory.events and memory.events.local */
 	struct cgroup_file events_file;
 	struct cgroup_file events_local_file;
@@ -318,6 +316,8 @@ struct mem_cgroup {
 	/* List of events which userspace want to receive */
 	struct list_head event_list;
 	spinlock_t event_list_lock;
+
+	int swappiness;
 #endif /* CONFIG_MEMCG_V1 */
 
 	struct mem_cgroup_per_node *nodeinfo[];
include/linux/swap.h +0 -19
index 696ed01709c2..330a420fd6de 100644
@@ -309,7 +309,6 @@ static inline bool lru_cache_disabled(void)
 }
 
 extern unsigned long shrink_all_memory(unsigned long nr_pages);
-extern int vm_swappiness;
 long remove_mapping(struct address_space *mapping, struct folio *folio);
 
 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
@@ -468,25 +467,7 @@ static inline int add_swap_extent(struct swap_info_struct *sis,
 }
 #endif /* CONFIG_SWAP */
 #ifdef CONFIG_MEMCG
-static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
-{
-	/* Cgroup2 doesn't have per-cgroup swappiness */
-	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
-		return READ_ONCE(vm_swappiness);
-
-	/* root ? */
-	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
-		return READ_ONCE(vm_swappiness);
-
-	return READ_ONCE(memcg->swappiness);
-}
-
 void lru_reparent_memcg(struct mem_cgroup *memcg, struct mem_cgroup *parent, int nid);
-#else
-static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
-{
-	return READ_ONCE(vm_swappiness);
-}
 #endif
 
 #if defined(CONFIG_SWAP) && defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
mm/memcontrol.c +2 -2
index 22f55aeb94f3..d6ba276740ea 100644
@@ -63,6 +63,7 @@
 #include <linux/sched/isolation.h>
 #include <linux/kmemleak.h>
 #include "internal.h"
+#include "swap.h"
 #include "swap_table.h"
 #include <net/sock.h>
 #include <net/ip.h>
@@ -4174,11 +4175,10 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
 #endif
 	page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
 	if (parent) {
-		WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent));
-
 		page_counter_init(&memcg->memory, &parent->memory, memcg_on_dfl);
 		page_counter_init(&memcg->swap, &parent->swap, false);
 #ifdef CONFIG_MEMCG_V1
+		WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent));
 		memcg->memory.track_failcnt = !memcg_on_dfl;
 		WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable));
 		page_counter_init(&memcg->kmem, &parent->kmem, false);
mm/swap.h +20 -0
index b51ad3071a73..a0e04b97a466 100644
@@ -4,6 +4,7 @@
 
 #include <linux/atomic.h> /* for atomic_long_t */
 #include <linux/mm.h> /* for PAGE_SHIFT */
+#include <linux/memcontrol.h> /* for mem_cgroup_swappiness() */
 struct mempolicy;
 struct swap_iocb;
 struct swap_memcg_table;
@@ -76,6 +77,25 @@ enum swap_cluster_flags {
 	CLUSTER_FLAG_MAX,
 };
 
+extern int vm_swappiness;
+
+static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
+{
+#ifdef CONFIG_MEMCG_V1
+	/* Cgroup2 doesn't have per-cgroup swappiness */
+	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
+		return READ_ONCE(vm_swappiness);
+
+	/* root ? */
+	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
+		return READ_ONCE(vm_swappiness);
+
+	return READ_ONCE(memcg->swappiness);
+#else
+	return READ_ONCE(vm_swappiness);
+#endif
+}
+
 #ifdef CONFIG_SWAP
 #include <linux/swapops.h> /* for swp_offset */
 #include <linux/blk_types.h> /* for bio_end_io_t */
Signature · 1 line
2.34.1
author
#6 lore raw

PATCH v2 2/2 mm: vmscan: fix node reclaim ignoring swappiness parameter

In reply to Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

    # cat /proc/sys/vm/swappiness
    60
    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 1840
    pgsteal_anon 25
    pgsteal_file 1815
    # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 18013
    pgsteal_anon 337
    pgsteal_file 17676

After fix (swappiness=max honored, anon pages reclaimed as expected):

    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 0
    pgsteal_anon 0
    pgsteal_file 0
    # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 16283
    pgsteal_anon 16283
    pgsteal_file 0
Fixes b980077899ea ("mm: introduce per-node proactive reclaim interface")
---
 mm/vmscan.c | 19 +++++++------------
 1 file changed, 7 insertions(+), 12 deletions(-)
mm/vmscan.c +7 -12
index 986dde8e7429..bd71595c8b2b 100644
@@ -199,6 +199,13 @@ struct scan_control {
  */
 int vm_swappiness = 60;
 
+static int sc_swappiness(struct scan_control *sc, struct mem_cgroup *memcg)
+{
+	if (sc->proactive && sc->proactive_swappiness)
+		return *sc->proactive_swappiness;
+	return mem_cgroup_swappiness(memcg);
+}
+
 #ifdef CONFIG_MEMCG
 
 /* Returns true for reclaim through cgroup limits or cgroup interfaces. */
@@ -239,13 +246,6 @@ static bool writeback_throttling_sane(struct scan_control *sc)
 #endif
 	return false;
 }
-
-static int sc_swappiness(struct scan_control *sc, struct mem_cgroup *memcg)
-{
-	if (sc->proactive && sc->proactive_swappiness)
-		return *sc->proactive_swappiness;
-	return mem_cgroup_swappiness(memcg);
-}
 #else
 static bool cgroup_reclaim(struct scan_control *sc)
 {
@@ -261,11 +261,6 @@ static bool writeback_throttling_sane(struct scan_control *sc)
 {
 	return true;
 }
-
-static int sc_swappiness(struct scan_control *sc, struct mem_cgroup *memcg)
-{
-	return READ_ONCE(vm_swappiness);
-}
 #endif
 
 static void set_task_reclaim_state(struct task_struct *task,
Signature · 1 line
2.34.1
reviewer Acked-by on v2 2/2
#7 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

Quoted 52 lines · Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

     # cat /proc/sys/vm/swappiness
     60
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 1840
     pgsteal_anon 25
     pgsteal_file 1815
     # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 18013
     pgsteal_anon 337
     pgsteal_file 17676

After fix (swappiness=max honored, anon pages reclaimed as expected):

     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 0
     pgsteal_anon 0
     pgsteal_file 0
     # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 16283
     pgsteal_anon 16283
     pgsteal_file 0
Quoted 8 lines
Fixes b980077899ea ("mm: introduce per-node proactive reclaim interface")
---
  mm/vmscan.c | 19 +++++++------------
  1 file changed, 7 insertions(+), 12 deletions(-)

Thanks for fixing this!

reviewer
#8 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

In reply to Ridong

Hi Ridong,

Quoted 34 lines · Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

     # cat /proc/sys/vm/swappiness
     60
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 1840
     pgsteal_anon 25
     pgsteal_file 1815
     # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 18013
     pgsteal_anon 337
     pgsteal_file 17676
Quoted 2 lines

After fix (swappiness=max honored, anon pages reclaimed as expected):

By the way, in get_scan_count(), it seems we still reclaim file pages without swap space, even if swappiness=max (SWAPPINESS_ANON_ONLY) is set.

This behavior appears to contradict the semantics of SWAPPINESS_ANON_ONLY, which probably needs a fix as well.

Thanks,
Qi

Tested-by Song Hu Reviewed-by Song Hu on v2 2/2
#13 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

In reply to Ridong

Hi Ridong,

Reproduced the bug and verified the fix in a QEMU x86_64 guest with CONFIG_MEMCG disabled. Global vm.swappiness pinned to 1, then:

echo "64M swappiness=<x>" > /sys/devices/system/node/node0/reclaim

  swappiness= | buggy (anon/file)   | patched (anon/file)
  ----------- | -------------------- | --------------------
  max         |      0  /  16385     |  16292  /  0
  200         |      2  /  16384     |  16837  /  0
  1           |      0  /  16386     |      0  /  16467

Unpatched ignores the swappiness arg (always file); patched honors it. Matches the commit message.

Quoted 105 lines · Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

    # cat /proc/sys/vm/swappiness
    60
    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 1840
    pgsteal_anon 25
    pgsteal_file 1815
    # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 18013
    pgsteal_anon 337
    pgsteal_file 17676

After fix (swappiness=max honored, anon pages reclaimed as expected):

    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 0
    pgsteal_anon 0
    pgsteal_file 0
    # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
    # cat /proc/vmstat | grep pgsteal
    pgsteal_kswapd 0
    pgsteal_direct 0
    pgsteal_khugepaged 0
    pgsteal_proactive 16283
    pgsteal_anon 16283
    pgsteal_file 0
Fixes b980077899ea ("mm: introduce per-node proactive reclaim interface")
---
 mm/vmscan.c | 19 +++++++------------
 1 file changed, 7 insertions(+), 12 deletions(-)
mm/vmscan.c +7 -12
index 986dde8e7429..bd71595c8b2b 100644
@@ -199,6 +199,13 @@ struct scan_control {
  */
 int vm_swappiness = 60;
 
+static int sc_swappiness(struct scan_control *sc, struct mem_cgroup *memcg)
+{
+	if (sc->proactive && sc->proactive_swappiness)
+		return *sc->proactive_swappiness;
+	return mem_cgroup_swappiness(memcg);
+}
+
 #ifdef CONFIG_MEMCG
 
 /* Returns true for reclaim through cgroup limits or cgroup interfaces. */
@@ -239,13 +246,6 @@ static bool writeback_throttling_sane(struct scan_control *sc)
 #endif
 	return false;
 }
-
-static int sc_swappiness(struct scan_control *sc, struct mem_cgroup *memcg)
-{
-	if (sc->proactive && sc->proactive_swappiness)
-		return *sc->proactive_swappiness;
-	return mem_cgroup_swappiness(memcg);
-}
 #else
 static bool cgroup_reclaim(struct scan_control *sc)
 {
@@ -261,11 +261,6 @@ static bool writeback_throttling_sane(struct scan_control *sc)
 {
 	return true;
 }
-
-static int sc_swappiness(struct scan_control *sc, struct mem_cgroup *memcg)
-{
-	return READ_ONCE(vm_swappiness);
-}
 #endif
 
 static void set_task_reclaim_state(struct task_struct *task,
Reviewed-by Song Hu on v2 1/2
#2 lore raw

Re: [PATCH v2 1/2] memcg: move mem_cgroup_swappiness and vm_swappiness to mm/swap.h

In reply to Ridong
Quoted 146 lines · Ridong

From: Ridong Chen <[email protected]>

The per-memcg swappiness knob is v1-only; v2 always uses global vm_swappiness and ignores the per-cgroup field.

Both mem_cgroup_swappiness() and vm_swappiness are only used within mm/ (memcontrol.c, memcontrol-v1.c, vmscan.c), so move them out of the public include/linux/swap.h into the mm-private mm/swap.h. This keeps unrelated declarations out of include/linux/swap.h.

Guard memcg->swappiness with CONFIG_MEMCG_V1 as well, so v2-only kernels drop the unused field.

No functional change for v1; v2-only kernels drop the unused field.

---
 include/linux/memcontrol.h |  4 ++--
 include/linux/swap.h       | 19 -------------------
 mm/memcontrol.c            |  4 ++--
 mm/swap.h                  | 20 ++++++++++++++++++++
 4 files changed, 24 insertions(+), 23 deletions(-)
include/linux/memcontrol.h +2 -2
index 957260677678..e7a794dfb40e 100644
@@ -239,8 +239,6 @@ struct mem_cgroup {
 	 */
 	bool oom_group;
 
-	int swappiness;
-
 	/* memory.events and memory.events.local */
 	struct cgroup_file events_file;
 	struct cgroup_file events_local_file;
@@ -318,6 +316,8 @@ struct mem_cgroup {
 	/* List of events which userspace want to receive */
 	struct list_head event_list;
 	spinlock_t event_list_lock;
+
+	int swappiness;
 #endif /* CONFIG_MEMCG_V1 */
 
 	struct mem_cgroup_per_node *nodeinfo[];
include/linux/swap.h +0 -19
index 696ed01709c2..330a420fd6de 100644
@@ -309,7 +309,6 @@ static inline bool lru_cache_disabled(void)
 }
 
 extern unsigned long shrink_all_memory(unsigned long nr_pages);
-extern int vm_swappiness;
 long remove_mapping(struct address_space *mapping, struct folio *folio);
 
 #if defined(CONFIG_SYSFS) && defined(CONFIG_NUMA)
@@ -468,25 +467,7 @@ static inline int add_swap_extent(struct swap_info_struct *sis,
 }
 #endif /* CONFIG_SWAP */
 #ifdef CONFIG_MEMCG
-static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
-{
-	/* Cgroup2 doesn't have per-cgroup swappiness */
-	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
-		return READ_ONCE(vm_swappiness);
-
-	/* root ? */
-	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
-		return READ_ONCE(vm_swappiness);
-
-	return READ_ONCE(memcg->swappiness);
-}
-
 void lru_reparent_memcg(struct mem_cgroup *memcg, struct mem_cgroup *parent, int nid);
-#else
-static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
-{
-	return READ_ONCE(vm_swappiness);
-}
 #endif
 
 #if defined(CONFIG_SWAP) && defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
mm/memcontrol.c +2 -2
index 22f55aeb94f3..d6ba276740ea 100644
@@ -63,6 +63,7 @@
 #include <linux/sched/isolation.h>
 #include <linux/kmemleak.h>
 #include "internal.h"
+#include "swap.h"
 #include "swap_table.h"
 #include <net/sock.h>
 #include <net/ip.h>
@@ -4174,11 +4175,10 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
 #endif
 	page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX);
 	if (parent) {
-		WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent));
-
 		page_counter_init(&memcg->memory, &parent->memory, memcg_on_dfl);
 		page_counter_init(&memcg->swap, &parent->swap, false);
 #ifdef CONFIG_MEMCG_V1
+		WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent));
 		memcg->memory.track_failcnt = !memcg_on_dfl;
 		WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable));
 		page_counter_init(&memcg->kmem, &parent->kmem, false);
mm/swap.h +20 -0
index b51ad3071a73..a0e04b97a466 100644
@@ -4,6 +4,7 @@
 
 #include <linux/atomic.h> /* for atomic_long_t */
 #include <linux/mm.h> /* for PAGE_SHIFT */
+#include <linux/memcontrol.h> /* for mem_cgroup_swappiness() */
 struct mempolicy;
 struct swap_iocb;
 struct swap_memcg_table;
@@ -76,6 +77,25 @@ enum swap_cluster_flags {
 	CLUSTER_FLAG_MAX,
 };
 
+extern int vm_swappiness;
+
+static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
+{
+#ifdef CONFIG_MEMCG_V1
+	/* Cgroup2 doesn't have per-cgroup swappiness */
+	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
+		return READ_ONCE(vm_swappiness);
+
+	/* root ? */
+	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
+		return READ_ONCE(vm_swappiness);
+
+	return READ_ONCE(memcg->swappiness);
+#else
+	return READ_ONCE(vm_swappiness);
+#endif
+}
+
 #ifdef CONFIG_SWAP
 #include <linux/swapops.h> /* for swp_offset */
 #include <linux/blk_types.h> /* for bio_end_io_t */
reviewer Reviewed-by on v2 1/2
#3 lore raw

Re: [PATCH v2 1/2] memcg: move mem_cgroup_swappiness and vm_swappiness to mm/swap.h

In reply to Ridong
Quoted 11 lines · Ridong

From: Ridong Chen <[email protected]>

The per-memcg swappiness knob is v1-only; v2 always uses global vm_swappiness and ignores the per-cgroup field.

Both mem_cgroup_swappiness() and vm_swappiness are only used within mm/ (memcontrol.c, memcontrol-v1.c, vmscan.c), so move them out of the public include/linux/swap.h into the mm-private mm/swap.h. This keeps unrelated declarations out of include/linux/swap.h.

Quoted 7 lines

Guard memcg->swappiness with CONFIG_MEMCG_V1 as well, so v2-only kernels drop the unused field.

No functional change for v1; v2-only kernels drop the unused field.

---

#9 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

In reply to Qi Zheng
Quoted 40 lines · Qi Zheng

Hi Ridong,

Quoted 34 lines · Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

     # cat /proc/sys/vm/swappiness
     60
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 1840
     pgsteal_anon 25
     pgsteal_file 1815
     # echo "64M swappiness=max" > /sys/devices/system/node/node0/reclaim
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 18013
     pgsteal_anon 337
     pgsteal_file 17676
Quoted 2 lines

After fix (swappiness=max honored, anon pages reclaimed as expected):

Quoted 7 lines

By the way, in get_scan_count(), it seems we still reclaim file pages without swap space, even if swappiness=max (SWAPPINESS_ANON_ONLY) is set.

This behavior appears to contradict the semantics of SWAPPINESS_ANON_ONLY, which probably needs a fix as well.

Thanks, Qi.

You're right. Currently, when can_reclaim_anon_pages() returns false, the reclaim logic falls back to scanning file folios even if swappiness == SWAPPINESS_ANON_ONLY is set:

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
			   unsigned long *nr)
{
	....
	/* If we have no swap space, do not bother scanning anon folios. */
	if (!sc->may_swap || !can_reclaim_anon_pages(memcg, pgdat->node_id, sc)) {
		scan_balance = SCAN_FILE;
		goto out;
	}

	...
	/* Proactive reclaim initiated by userspace for anonymous memory only */
	if (swappiness == SWAPPINESS_ANON_ONLY) {
		WARN_ON_ONCE(!sc->proactive);
		scan_balance = SCAN_ANON;
		goto out;
	}
	...
}
```

The !can_reclaim_anon_pages() check above takes precedence over the SWAPPINESS_ANON_ONLY branch, making the latter effectively unreachable when anonymous pages are deemed non-reclaimable.

To fix this, could we move the SWAPPINESS_ANON_ONLY check up before the can_reclaim_anon_pages() check?

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
			   unsigned long *nr)
{
	...
	/* Proactive reclaim initiated by userspace for anonymous memory only */
	if (swappiness == SWAPPINESS_ANON_ONLY) {
		WARN_ON_ONCE(!sc->proactive);
		scan_balance = SCAN_ANON;
		goto out;
	}
	....
	/* If we have no swap space, do not bother scanning anon folios. */
	if (!sc->may_swap || !can_reclaim_anon_pages(memcg, pgdat->node_id, sc)) {
		scan_balance = SCAN_FILE;
		goto out;
	}

	...
}
```
Signature · 2 lines
Best regards
Ridong
#14 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

In reply to Tao Cui
Quoted 13 lines · Tao Cui

Hi Ridong,

Reproduced the bug and verified the fix in a QEMU x86_64 guest with CONFIG_MEMCG disabled. Global vm.swappiness pinned to 1, then:

echo "64M swappiness=<x>" > /sys/devices/system/node/node0/reclaim

   swappiness= | buggy (anon/file)   | patched (anon/file)
   ----------- | -------------------- | --------------------
   max         |      0  /  16385     |  16292  /  0
   200         |      2  /  16384     |  16837  /  0
   1           |      0  /  16386     |      0  /  16467
Quoted 6 lines

Unpatched ignores the swappiness arg (always file); patched honors it. Matches the commit message.

Thanks.

Signature · 2 lines
Best regards
Ridong
reviewer
#10 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

In reply to Ridong Chen
Quoted 82 lines · Ridong Chen
Quoted 40 lines · Qi Zheng

Hi Ridong,

Quoted 34 lines · Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

     # cat /proc/sys/vm/swappiness
     60
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 1840
     pgsteal_anon 25
     pgsteal_file 1815
     # echo "64M swappiness=max" > /sys/devices/system/node/node0/ reclaim
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 18013
     pgsteal_anon 337
     pgsteal_file 17676
Quoted 2 lines

After fix (swappiness=max honored, anon pages reclaimed as expected):

Quoted 7 lines

By the way, in get_scan_count(), it seems we still reclaim file pages without swap space, even if swappiness=max (SWAPPINESS_ANON_ONLY) is set.

This behavior appears to contradict the semantics of SWAPPINESS_ANON_ONLY, which probably needs a fix as well.

Thanks, Qi.

You're right. Currently, when can_reclaim_anon_pages() returns false, the reclaim logic falls back to scanning file folios even if swappiness == SWAPPINESS_ANON_ONLY is set:

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
                unsigned long *nr)
{
     ....
     /* If we have no swap space, do not bother scanning anon folios. */
     if (!sc->may_swap || !can_reclaim_anon_pages(memcg, pgdat->node_id, 
sc)) {
         scan_balance = SCAN_FILE;
         goto out;
     }

     ...
     /* Proactive reclaim initiated by userspace for anonymous memory 
only */
     if (swappiness == SWAPPINESS_ANON_ONLY) {
         WARN_ON_ONCE(!sc->proactive);
         scan_balance = SCAN_ANON;
         goto out;
     }
     ...
}
```

The !can_reclaim_anon_pages() check above takes precedence over the SWAPPINESS_ANON_ONLY branch, making the latter effectively unreachable when anonymous pages are deemed non-reclaimable.

To fix this, could we move the SWAPPINESS_ANON_ONLY check up before the can_reclaim_anon_pages() check?

Quoted 11 lines

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
               unsigned long *nr)
{
    ...
    /* Proactive reclaim initiated by userspace for anonymous memory
only */
    if (swappiness == SWAPPINESS_ANON_ONLY) { WARN_ON_ONCE(!sc->proactive);
        scan_balance = SCAN_ANON;
        goto out;

For SWAPPINESS_ANON_ONLY + !can_reclaim_anon_pages, perhaps we should just return directly instead of goto out, since no pages can be reclaimed in this situation.

Thanks,
Qi

Quoted 13 lines

    }
    ....
    /* If we have no swap space, do not bother scanning anon folios. */ if (!sc->may_swap || !can_reclaim_anon_pages(memcg, pgdat->node_id,
sc)) {
        scan_balance = SCAN_FILE;
        goto out;
    }

    ...
}
```

#11 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

Quoted 95 lines · Qi Zheng
Quoted 81 lines · Ridong Chen
Quoted 40 lines · Qi Zheng

Hi Ridong,

Quoted 34 lines · Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

     # cat /proc/sys/vm/swappiness
     60
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 1840
     pgsteal_anon 25
     pgsteal_file 1815
     # echo "64M swappiness=max" > /sys/devices/system/node/node0/ reclaim
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 18013
     pgsteal_anon 337
     pgsteal_file 17676
Quoted 2 lines

After fix (swappiness=max honored, anon pages reclaimed as expected):

Quoted 7 lines

By the way, in get_scan_count(), it seems we still reclaim file pages without swap space, even if swappiness=max (SWAPPINESS_ANON_ONLY) is set.

This behavior appears to contradict the semantics of SWAPPINESS_ANON_ONLY, which probably needs a fix as well.

Thanks, Qi.

You're right. Currently, when can_reclaim_anon_pages() returns false, the reclaim logic falls back to scanning file folios even if swappiness == SWAPPINESS_ANON_ONLY is set:

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
                unsigned long *nr)
{
     ....
     /* If we have no swap space, do not bother scanning anon folios. */
     if (!sc->may_swap || !can_reclaim_anon_pages(memcg, pgdat- 
>node_id, sc)) {
         scan_balance = SCAN_FILE;
         goto out;
     }

     ...
     /* Proactive reclaim initiated by userspace for anonymous memory only */
     if (swappiness == SWAPPINESS_ANON_ONLY) {
         WARN_ON_ONCE(!sc->proactive);
         scan_balance = SCAN_ANON;
         goto out;
     }
     ...
}
```

The !can_reclaim_anon_pages() check above takes precedence over the SWAPPINESS_ANON_ONLY branch, making the latter effectively unreachable when anonymous pages are deemed non-reclaimable.

To fix this, could we move the SWAPPINESS_ANON_ONLY check up before the can_reclaim_anon_pages() check?

Quoted 10 lines

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, unsigned long *nr)
{
     ...
     /* Proactive reclaim initiated by userspace for anonymous memory only */ if (swappiness == SWAPPINESS_ANON_ONLY) { WARN_ON_ONCE(!sc->proactive);
         scan_balance = SCAN_ANON;
         goto out;

Quoted 4 lines

For SWAPPINESS_ANON_ONLY + !can_reclaim_anon_pages, perhaps we should just return directly instead of goto out, since no pages can be reclaimed in this situation.

Much better.
Would you mind if I send a separate patch to fix this issue?

Quoted 15 lines
Quoted 14 lines

     }
     ....
     /* If we have no swap space, do not bother scanning anon folios. */ if (!sc->may_swap || !can_reclaim_anon_pages(memcg, pgdat-

Quoted 1 line

node_id, sc)) {

         scan_balance = SCAN_FILE;
         goto out;
     }

     ...
}
```

Signature · 2 lines
Best regards
Ridong
reviewer
#12 lore raw

Re: [PATCH v2 2/2] mm: vmscan: fix node reclaim ignoring swappiness parameter

Quoted 98 lines · Ridong Chen
Quoted 95 lines · Qi Zheng
Quoted 91 lines · Ridong Chen
Quoted 40 lines · Qi Zheng

Hi Ridong,

Quoted 34 lines · Ridong

From: Ridong Chen <[email protected]>

sc_swappiness() had two separate definitions depending on CONFIG_MEMCG. The !CONFIG_MEMCG variant simply returned vm_swappiness, ignoring the proactive_swappiness value passed through scan_control. This caused the swappiness parameter written to /sys/devices/system/node/nodeX/reclaim to have no effect when CONFIG_MEMCG is disabled.

Fix this by consolidating sc_swappiness() into a single definition that checks sc->proactive_swappiness first, then falls back to mem_cgroup_swappiness() which already handles both CONFIG_MEMCG and !CONFIG_MEMCG.

Before fix (swappiness=max ignored, mostly file pages reclaimed):

     # cat /proc/sys/vm/swappiness
     60
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 1840
     pgsteal_anon 25
     pgsteal_file 1815
     # echo "64M swappiness=max" > /sys/devices/system/node/node0/ reclaim
     # cat /proc/vmstat | grep pgsteal
     pgsteal_kswapd 0
     pgsteal_direct 0
     pgsteal_khugepaged 0
     pgsteal_proactive 18013
     pgsteal_anon 337
     pgsteal_file 17676
Quoted 2 lines

After fix (swappiness=max honored, anon pages reclaimed as expected):

Quoted 7 lines

By the way, in get_scan_count(), it seems we still reclaim file pages without swap space, even if swappiness=max (SWAPPINESS_ANON_ONLY) is set.

This behavior appears to contradict the semantics of SWAPPINESS_ANON_ONLY, which probably needs a fix as well.

Thanks, Qi.

You're right. Currently, when can_reclaim_anon_pages() returns false, the reclaim logic falls back to scanning file folios even if swappiness == SWAPPINESS_ANON_ONLY is set:

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc,
                unsigned long *nr)
{
     ....
     /* If we have no swap space, do not bother scanning anon folios. */
     if (!sc->may_swap || !can_reclaim_anon_pages(memcg, pgdat- 
>node_id, sc)) {
         scan_balance = SCAN_FILE;
         goto out;
     }

     ...
     /* Proactive reclaim initiated by userspace for anonymous memory only */
     if (swappiness == SWAPPINESS_ANON_ONLY) {
         WARN_ON_ONCE(!sc->proactive);
         scan_balance = SCAN_ANON;
         goto out;
     }
     ...
}
```

The !can_reclaim_anon_pages() check above takes precedence over the SWAPPINESS_ANON_ONLY branch, making the latter effectively unreachable when anonymous pages are deemed non-reclaimable.

To fix this, could we move the SWAPPINESS_ANON_ONLY check up before the can_reclaim_anon_pages() check?

```
static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, unsigned long *nr)
{
     ...
     /* Proactive reclaim initiated by userspace for anonymous memory only */ if (swappiness == SWAPPINESS_ANON_ONLY) { WARN_ON_ONCE(!sc->proactive);
         scan_balance = SCAN_ANON;
         goto out;

Quoted 8 lines
Quoted 4 lines

For SWAPPINESS_ANON_ONLY + !can_reclaim_anon_pages, perhaps we should just return directly instead of goto out, since no pages can be reclaimed in this situation.

Much better.
Would you mind if I send a separate patch to fix this issue?

Sure, thanks for your time on this, and I look forward to your fix!

#15 lore raw
In reply to Ridong
Quoted 29 lines · Ridong

From: Ridong <[email protected]>

The per-node proactive reclaim interface (/sys/devices/system/node/nodeX/reclaim) accepts a swappiness parameter, but it is silently ignored when CONFIG_MEMCG is disabled. The root cause is that sc_swappiness() has separate implementations for CONFIG_MEMCG and !CONFIG_MEMCG, and the latter never checks proactive_swappiness.

Patch 1 moves mem_cgroup_swappiness() and vm_swappiness out of the public include/linux/swap.h into the mm-private mm/swap.h, and makes the helper handle both CONFIG_MEMCG and !CONFIG_MEMCG in a single inline function. This is a prerequisite for unifying sc_swappiness().

Patch 2 consolidates sc_swappiness() into a single definition that works regardless of CONFIG_MEMCG, fixing the node reclaim swappiness bug.

---

v1 -> v2:
- Move mem_cgroup_swappiness() and vm_swappiness to mm/swap.h instead of include/linux/memcontrol.h. Suggested by Barry Song.
- Correct fix tag for patch 2.

v1: https://lore.kernel.org/all/[email protected]/

Ridong Chen (2):
memcg: move mem_cgroup_swappiness and vm_swappiness to mm/swap.h
mm: vmscan: fix node reclaim ignoring swappiness parameter

Quoted 7 lines
  include/linux/memcontrol.h |  4 ++--
  include/linux/swap.h       | 19 -------------------
  mm/memcontrol.c            |  4 ++--
  mm/swap.h                  | 20 ++++++++++++++++++++
  mm/vmscan.c                | 19 +++++++------------
  5 files changed, 31 insertions(+), 35 deletions(-)

Hello all,

Can this series be applied?

Just a reminder in case this gets buried.

Signature · 2 lines
Best regards
Ridong
maintainer
#16 lore raw
Quoted 8 lines · On Thu, 23 Jul 2026 09:13:27 +0800 Ridong Chen <[email protected]> wrote:
Quoted 7 lines
  include/linux/memcontrol.h |  4 ++--
  include/linux/swap.h       | 19 -------------------
  mm/memcontrol.c            |  4 ++--
  mm/swap.h                  | 20 ++++++++++++++++++++
  mm/vmscan.c                | 19 +++++++------------
  5 files changed, 31 insertions(+), 35 deletions(-)
Quoted 5 lines

Hello all,

Can this series be applied?

Just a reminder in case this gets buried.

Got it, and thanks for the poke.

It seems that kvack.org sent me two of the emails from this 13 email thread. This is becoming a problem.

maintainer Acked-by on v2 1/2
#4 lore raw

Re: [PATCH v2 1/2] memcg: move mem_cgroup_swappiness and vm_swappiness to mm/swap.h

In reply to Ridong

One small suggestion:

Quoted 11 lines · Ridong
mm/swap.h +7 -0
@@ -76,6 +77,25 @@ enum swap_cluster_flags {
 	CLUSTER_FLAG_MAX,
 };
 
+extern int vm_swappiness;
+
+static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
+{
+#ifdef CONFIG_MEMCG_V1
+	/* Cgroup2 doesn't have per-cgroup swappiness */
+	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))

mm/swap.h enum swap_cluster_flags { · line 96

Quoted 10 lines
mm/swap.h +10 -0
+		return READ_ONCE(vm_swappiness);
+
+	/* root ? */
+	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
+		return READ_ONCE(vm_swappiness);
+
+	return READ_ONCE(memcg->swappiness);
+#else
+	return READ_ONCE(vm_swappiness);
+#endif

How about:

#ifdef CONFIG_MEMCG_V1
	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
	    !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg))
		return memcg->swappiness;
#endif
	return READ_ONCE(vm_swappiness);
#5 lore raw

Re: [PATCH v2 1/2] memcg: move mem_cgroup_swappiness and vm_swappiness to mm/swap.h

Quoted 29 lines · Johannes Weiner

One small suggestion:

Quoted 11 lines · Ridong
mm/swap.h +7 -0
@@ -76,6 +77,25 @@ enum swap_cluster_flags {
  	CLUSTER_FLAG_MAX,
  };
 
+extern int vm_swappiness;
+
+static inline int mem_cgroup_swappiness(struct mem_cgroup *memcg)
+{
+#ifdef CONFIG_MEMCG_V1
+	/* Cgroup2 doesn't have per-cgroup swappiness */
+	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))

mm/swap.h enum swap_cluster_flags { · line 96

Quoted 10 lines
mm/swap.h +10 -0
+		return READ_ONCE(vm_swappiness);
+
+	/* root ? */
+	if (mem_cgroup_disabled() || mem_cgroup_is_root(memcg))
+		return READ_ONCE(vm_swappiness);
+
+	return READ_ONCE(memcg->swappiness);
+#else
+	return READ_ONCE(vm_swappiness);
+#endif

How about:

Quoted 6 lines
#ifdef CONFIG_MEMCG_V1
	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
	    !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg))
		return memcg->swappiness;
#endif
	return READ_ONCE(vm_swappiness);

This is simpler.

Hi, Andrew, I notice you just applied this series to mm-new branch. Would you mind I update it?

Signature · 2 lines
Best regards
Ridong