diff mbox series

[v4] mm,memcg: provide per-cgroup counters for NUMA balancing operations

Message ID 20240814235122.252309-1-kaiyang2@cs.cmu.edu (mailing list archive)
State New
Headers show
Series [v4] mm,memcg: provide per-cgroup counters for NUMA balancing operations | expand

Commit Message

Kaiyang Zhao Aug. 14, 2024, 11:51 p.m. UTC 
From: Kaiyang Zhao <kaiyang2@cs.cmu.edu>

The ability to observe the demotion and promotion decisions made by the
kernel on a per-cgroup basis is important for monitoring and tuning
containerized workloads on machines equipped with tiered memory.

Different containers in the system may experience drastically different
memory tiering actions that cannot be distinguished from the global
counters alone.

For example, a container running a workload that has a much hotter
memory accesses will likely see more promotions and fewer demotions,
potentially depriving a colocated container of top tier memory to such
an extent that its performance degrades unacceptably.

For another example, some containers may exhibit longer periods between
data reuse, causing much more numa_hint_faults than numa_pages_migrated.
In this case, tuning hot_threshold_ms may be appropriate, but the signal
can easily be lost if only global counters are available.

In the long term, we hope to introduce per-cgroup control of promotion
and demotion actions to implement memory placement policies in tiering.

This patch set adds seven counters to memory.stat in a cgroup:
numa_pages_migrated, numa_pte_updates, numa_hint_faults, pgdemote_kswapd,
pgdemote_khugepaged, pgdemote_direct and pgpromote_success. pgdemote_*
and pgpromote_success are also available in memory.numa_stat.

count_memcg_events_mm() is added to count multiple event occurrences at
once, and get_mem_cgroup_from_folio() is added because we need to get a
reference to the memcg of a folio before it's migrated to track
numa_pages_migrated. The accounting of PGDEMOTE_* is moved to
shrink_inactive_list() before being changed to per-cgroup.

Signed-off-by: Kaiyang Zhao <kaiyang2@cs.cmu.edu>
---
v4:
- added documentation of the memcg counters in cgroup-v2.rst
- added a hint of future directions in the changelog

v3:
- added pgpromote_success as suggested by Wei Xu

v2:
- fixed compilation error when CONFIG_NUMA_BALANCING is off
- fixed doc warning due to missing parameter description in
get_mem_cgroup_from_folio

 Documentation/admin-guide/cgroup-v2.rst | 19 +++++++++++
 include/linux/memcontrol.h              | 24 +++++++++++--
 include/linux/vmstat.h                  |  1 +
 mm/memcontrol.c                         | 45 +++++++++++++++++++++++++
 mm/memory.c                             |  3 ++
 mm/mempolicy.c                          |  4 ++-
 mm/migrate.c                            |  7 ++--
 mm/vmscan.c                             |  8 ++---
 8 files changed, 101 insertions(+), 10 deletions(-)

Comments

Andrew Morton Sept. 2, 2024, 2:12 a.m. UTC | #1 
On Wed, 14 Aug 2024 23:51:22 +0000 kaiyang2@cs.cmu.edu wrote:

> From: Kaiyang Zhao <kaiyang2@cs.cmu.edu>
> 
> The ability to observe the demotion and promotion decisions made by the
> kernel on a per-cgroup basis is important for monitoring and tuning
> containerized workloads on machines equipped with tiered memory.
> 
> Different containers in the system may experience drastically different
> memory tiering actions that cannot be distinguished from the global
> counters alone.
> 
> For example, a container running a workload that has a much hotter
> memory accesses will likely see more promotions and fewer demotions,
> potentially depriving a colocated container of top tier memory to such
> an extent that its performance degrades unacceptably.
> 
> For another example, some containers may exhibit longer periods between
> data reuse, causing much more numa_hint_faults than numa_pages_migrated.
> In this case, tuning hot_threshold_ms may be appropriate, but the signal
> can easily be lost if only global counters are available.
> 
> In the long term, we hope to introduce per-cgroup control of promotion
> and demotion actions to implement memory placement policies in tiering.
> 
> This patch set adds seven counters to memory.stat in a cgroup:
> numa_pages_migrated, numa_pte_updates, numa_hint_faults, pgdemote_kswapd,
> pgdemote_khugepaged, pgdemote_direct and pgpromote_success. pgdemote_*
> and pgpromote_success are also available in memory.numa_stat.
> 
> count_memcg_events_mm() is added to count multiple event occurrences at
> once, and get_mem_cgroup_from_folio() is added because we need to get a
> reference to the memcg of a folio before it's migrated to track
> numa_pages_migrated. The accounting of PGDEMOTE_* is moved to
> shrink_inactive_list() before being changed to per-cgroup.

There appears to have been little reviewer interest in this one?
diff mbox series

Patch

diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
index 49b32cde3801..f11ede7b18d7 100644
--- a/Documentation/admin-guide/cgroup-v2.rst
+++ b/Documentation/admin-guide/cgroup-v2.rst
@@ -1617,6 +1617,25 @@  The following nested keys are defined.
 		Usually because failed to allocate some continuous swap space
 		for the huge page.
 
+	  numa_pages_migrated (npn)
+		Number of pages migrated by NUMA balancing.
+
+	  numa_pte_updates (npn)
+		Number of pages whose page table entries are modified by
+		NUMA balancing to produce NUMA hinting faults on access.
+
+	  numa_hint_faults (npn)
+		Number of NUMA hinting faults.
+
+	  pgdemote_kswapd
+		Number of pages demoted by kswapd.
+
+	  pgdemote_direct
+		Number of pages demoted directly.
+
+	  pgdemote_khugepaged
+		Number of pages demoted by khugepaged.
+
   memory.numa_stat
 	A read-only nested-keyed file which exists on non-root cgroups.
 
diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index 44f7fb7dc0c8..90ecd2dbca06 100644
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -768,6 +768,8 @@  struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm);
 
 struct mem_cgroup *get_mem_cgroup_from_current(void);
 
+struct mem_cgroup *get_mem_cgroup_from_folio(struct folio *folio);
+
 struct lruvec *folio_lruvec_lock(struct folio *folio);
 struct lruvec *folio_lruvec_lock_irq(struct folio *folio);
 struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio,
@@ -1012,8 +1014,8 @@  static inline void count_memcg_folio_events(struct folio *folio,
 		count_memcg_events(memcg, idx, nr);
 }
 
-static inline void count_memcg_event_mm(struct mm_struct *mm,
-					enum vm_event_item idx)
+static inline void count_memcg_events_mm(struct mm_struct *mm,
+					enum vm_event_item idx, unsigned long count)
 {
 	struct mem_cgroup *memcg;
 
@@ -1023,10 +1025,16 @@  static inline void count_memcg_event_mm(struct mm_struct *mm,
 	rcu_read_lock();
 	memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 	if (likely(memcg))
-		count_memcg_events(memcg, idx, 1);
+		count_memcg_events(memcg, idx, count);
 	rcu_read_unlock();
 }
 
+static inline void count_memcg_event_mm(struct mm_struct *mm,
+					enum vm_event_item idx)
+{
+	count_memcg_events_mm(mm, idx, 1);
+}
+
 static inline void memcg_memory_event(struct mem_cgroup *memcg,
 				      enum memcg_memory_event event)
 {
@@ -1246,6 +1254,11 @@  static inline struct mem_cgroup *get_mem_cgroup_from_current(void)
 	return NULL;
 }
 
+static inline struct mem_cgroup *get_mem_cgroup_from_folio(struct folio *folio)
+{
+	return NULL;
+}
+
 static inline
 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *css)
 {
@@ -1468,6 +1481,11 @@  static inline void count_memcg_folio_events(struct folio *folio,
 {
 }
 
+static inline void count_memcg_events_mm(struct mm_struct *mm,
+					enum vm_event_item idx, unsigned long count)
+{
+}
+
 static inline
 void count_memcg_event_mm(struct mm_struct *mm, enum vm_event_item idx)
 {
diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
index 9eb77c9007e6..d2761bf8ff32 100644
--- a/include/linux/vmstat.h
+++ b/include/linux/vmstat.h
@@ -32,6 +32,7 @@  struct reclaim_stat {
 	unsigned nr_ref_keep;
 	unsigned nr_unmap_fail;
 	unsigned nr_lazyfree_fail;
+	unsigned nr_demoted;
 };
 
 /* Stat data for system wide items */
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 4884629f0ce5..9a338978eeae 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -307,6 +307,12 @@  static const unsigned int memcg_node_stat_items[] = {
 #ifdef CONFIG_SWAP
 	NR_SWAPCACHE,
 #endif
+#ifdef CONFIG_NUMA_BALANCING
+	PGPROMOTE_SUCCESS,
+#endif
+	PGDEMOTE_KSWAPD,
+	PGDEMOTE_DIRECT,
+	PGDEMOTE_KHUGEPAGED,
 };
 
 static const unsigned int memcg_stat_items[] = {
@@ -437,6 +443,11 @@  static const unsigned int memcg_vm_event_stat[] = {
 	THP_SWPOUT,
 	THP_SWPOUT_FALLBACK,
 #endif
+#ifdef CONFIG_NUMA_BALANCING
+	NUMA_PAGE_MIGRATE,
+	NUMA_PTE_UPDATES,
+	NUMA_HINT_FAULTS,
+#endif
 };
 
 #define NR_MEMCG_EVENTS ARRAY_SIZE(memcg_vm_event_stat)
@@ -978,6 +989,24 @@  struct mem_cgroup *get_mem_cgroup_from_current(void)
 	return memcg;
 }
 
+/**
+ * get_mem_cgroup_from_folio - Obtain a reference on a given folio's memcg.
+ * @folio: folio from which memcg should be extracted.
+ */
+struct mem_cgroup *get_mem_cgroup_from_folio(struct folio *folio)
+{
+	struct mem_cgroup *memcg = folio_memcg(folio);
+
+	if (mem_cgroup_disabled())
+		return NULL;
+
+	rcu_read_lock();
+	if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css)))
+		memcg = root_mem_cgroup;
+	rcu_read_unlock();
+	return memcg;
+}
+
 /**
  * mem_cgroup_iter - iterate over memory cgroup hierarchy
  * @root: hierarchy root
@@ -1383,6 +1412,13 @@  static const struct memory_stat memory_stats[] = {
 	{ "workingset_restore_anon",	WORKINGSET_RESTORE_ANON		},
 	{ "workingset_restore_file",	WORKINGSET_RESTORE_FILE		},
 	{ "workingset_nodereclaim",	WORKINGSET_NODERECLAIM		},
+
+	{ "pgdemote_kswapd",		PGDEMOTE_KSWAPD		},
+	{ "pgdemote_direct",		PGDEMOTE_DIRECT		},
+	{ "pgdemote_khugepaged",	PGDEMOTE_KHUGEPAGED	},
+#ifdef CONFIG_NUMA_BALANCING
+	{ "pgpromote_success",		PGPROMOTE_SUCCESS	},
+#endif
 };
 
 /* The actual unit of the state item, not the same as the output unit */
@@ -1407,6 +1443,9 @@  static int memcg_page_state_output_unit(int item)
 	/*
 	 * Workingset state is actually in pages, but we export it to userspace
 	 * as a scalar count of events, so special case it here.
+	 *
+	 * Demotion and promotion activities are exported in pages, consistent
+	 * with their global counterparts.
 	 */
 	switch (item) {
 	case WORKINGSET_REFAULT_ANON:
@@ -1416,6 +1455,12 @@  static int memcg_page_state_output_unit(int item)
 	case WORKINGSET_RESTORE_ANON:
 	case WORKINGSET_RESTORE_FILE:
 	case WORKINGSET_NODERECLAIM:
+	case PGDEMOTE_KSWAPD:
+	case PGDEMOTE_DIRECT:
+	case PGDEMOTE_KHUGEPAGED:
+#ifdef CONFIG_NUMA_BALANCING
+	case PGPROMOTE_SUCCESS:
+#endif
 		return 1;
 	default:
 		return memcg_page_state_unit(item);
diff --git a/mm/memory.c b/mm/memory.c
index 0ed3603aaf31..13b679ad182c 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -5400,6 +5400,9 @@  int numa_migrate_check(struct folio *folio, struct vm_fault *vmf,
 	vma_set_access_pid_bit(vma);
 
 	count_vm_numa_event(NUMA_HINT_FAULTS);
+#ifdef CONFIG_NUMA_BALANCING
+	count_memcg_folio_events(folio, NUMA_HINT_FAULTS, 1);
+#endif
 	if (folio_nid(folio) == numa_node_id()) {
 		count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL);
 		*flags |= TNF_FAULT_LOCAL;
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index b3b5f376471f..b646fab3e45e 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -676,8 +676,10 @@  unsigned long change_prot_numa(struct vm_area_struct *vma,
 	tlb_gather_mmu(&tlb, vma->vm_mm);
 
 	nr_updated = change_protection(&tlb, vma, addr, end, MM_CP_PROT_NUMA);
-	if (nr_updated > 0)
+	if (nr_updated > 0) {
 		count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
+		count_memcg_events_mm(vma->vm_mm, NUMA_PTE_UPDATES, nr_updated);
+	}
 
 	tlb_finish_mmu(&tlb);
 
diff --git a/mm/migrate.c b/mm/migrate.c
index 6e32098ac2dc..dbfa910ec24b 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -2668,6 +2668,8 @@  int migrate_misplaced_folio(struct folio *folio, struct vm_area_struct *vma,
 	int nr_remaining;
 	unsigned int nr_succeeded;
 	LIST_HEAD(migratepages);
+	struct mem_cgroup *memcg = get_mem_cgroup_from_folio(folio);
+	struct lruvec *lruvec = mem_cgroup_lruvec(memcg, pgdat);
 
 	list_add(&folio->lru, &migratepages);
 	nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_folio,
@@ -2677,12 +2679,13 @@  int migrate_misplaced_folio(struct folio *folio, struct vm_area_struct *vma,
 		putback_movable_pages(&migratepages);
 	if (nr_succeeded) {
 		count_vm_numa_events(NUMA_PAGE_MIGRATE, nr_succeeded);
+		count_memcg_events(memcg, NUMA_PAGE_MIGRATE, nr_succeeded);
 		if ((sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING)
 		    && !node_is_toptier(folio_nid(folio))
 		    && node_is_toptier(node))
-			mod_node_page_state(pgdat, PGPROMOTE_SUCCESS,
-					    nr_succeeded);
+			mod_lruvec_state(lruvec, PGPROMOTE_SUCCESS, nr_succeeded);
 	}
+	mem_cgroup_put(memcg);
 	BUG_ON(!list_empty(&migratepages));
 	return nr_remaining ? -EAGAIN : 0;
 }
diff --git a/mm/vmscan.c b/mm/vmscan.c
index da6ba3206827..a118a55bbed5 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -1018,9 +1018,6 @@  static unsigned int demote_folio_list(struct list_head *demote_folios,
 		      (unsigned long)&mtc, MIGRATE_ASYNC, MR_DEMOTION,
 		      &nr_succeeded);
 
-	mod_node_page_state(pgdat, PGDEMOTE_KSWAPD + reclaimer_offset(),
-			    nr_succeeded);
-
 	return nr_succeeded;
 }
 
@@ -1519,7 +1516,8 @@  static unsigned int shrink_folio_list(struct list_head *folio_list,
 	/* 'folio_list' is always empty here */
 
 	/* Migrate folios selected for demotion */
-	nr_reclaimed += demote_folio_list(&demote_folios, pgdat);
+	stat->nr_demoted = demote_folio_list(&demote_folios, pgdat);
+	nr_reclaimed += stat->nr_demoted;
 	/* Folios that could not be demoted are still in @demote_folios */
 	if (!list_empty(&demote_folios)) {
 		/* Folios which weren't demoted go back on @folio_list */
@@ -1985,6 +1983,8 @@  static unsigned long shrink_inactive_list(unsigned long nr_to_scan,
 	spin_lock_irq(&lruvec->lru_lock);
 	move_folios_to_lru(lruvec, &folio_list);
 
+	__mod_lruvec_state(lruvec, PGDEMOTE_KSWAPD + reclaimer_offset(),
+					stat.nr_demoted);
 	__mod_node_page_state(pgdat, NR_ISOLATED_ANON + file, -nr_taken);
 	item = PGSTEAL_KSWAPD + reclaimer_offset();
 	if (!cgroup_reclaim(sc))