Commit 42a30035 authored by Johannes Weiner's avatar Johannes Weiner Committed by Linus Torvalds
Browse files

mm: memcontrol: fix recursive statistics correctness & scalabilty 

Right now, when somebody needs to know the recursive memory statistics
and events of a cgroup subtree, they need to walk the entire subtree and
sum up the counters manually.

There are two issues with this:

1. When a cgroup gets deleted, its stats are lost. The state counters
   should all be 0 at that point, of course, but the events are not.
   When this happens, the event counters, which are supposed to be
   monotonic, can go backwards in the parent cgroups.

2. During regular operation, we always have a certain number of lazily
   freed cgroups sitting around that have been deleted, have no tasks,
   but have a few cache pages remaining. These groups' statistics do not
   change until we eventually hit memory pressure, but somebody
   watching, say, memory.stat on an ancestor has to iterate those every
   time.

This patch addresses both issues by introducing recursive counters at
each level that are propagated from the write side when stats change.

Upward propagation happens when the per-cpu caches spill over into the
local atomic counter.  This is the same thing we do during charge and
uncharge, except that the latter uses atomic RMWs, which are more
expensive; stat changes happen at around the same rate.  In a sparse
file test (page faults and reclaim at maximum CPU speed) with 5 cgroup
nesting levels, perf shows __mod_memcg_page state at ~1%.

Link: http://lkml.kernel.org/r/20190412151507.2769-4-hannes@cmpxchg.org


Signed-off-by: default avatarJohannes Weiner <hannes@cmpxchg.org>
Reviewed-by: default avatarShakeel Butt <shakeelb@google.com>
Reviewed-by: default avatarRoman Gushchin <guro@fb.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
parent db9adbcb

include/linux/memcontrol.h

+51 −3
Original line number Diff line number Diff line
@@ -128,6 +128,7 @@ struct mem_cgroup_per_node { 


	struct lruvec_stat __percpu *lruvec_stat_cpu;

	atomic_long_t		lruvec_stat[NR_VM_NODE_STAT_ITEMS];

	atomic_long_t		lruvec_stat_local[NR_VM_NODE_STAT_ITEMS];



	unsigned long		lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
@@ -279,7 +280,11 @@ struct mem_cgroup { 
	MEMCG_PADDING(_pad2_);



	atomic_long_t		vmstats[MEMCG_NR_STAT];

	atomic_long_t		vmstats_local[MEMCG_NR_STAT];



	atomic_long_t		vmevents[NR_VM_EVENT_ITEMS];

	atomic_long_t		vmevents_local[NR_VM_EVENT_ITEMS];



	atomic_long_t		memory_events[MEMCG_NR_MEMORY_EVENTS];



	unsigned long		socket_pressure;
@@ -550,6 +555,20 @@ struct mem_cgroup *lock_page_memcg(struct page *page); 
void __unlock_page_memcg(struct mem_cgroup *memcg);

void unlock_page_memcg(struct page *page);



/*

 * idx can be of type enum memcg_stat_item or node_stat_item.

 * Keep in sync with memcg_exact_page_state().

 */

static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)

{

	long x = atomic_long_read(&memcg->vmstats[idx]);

#ifdef CONFIG_SMP

	if (x < 0)

		x = 0;

#endif

	return x;

}



/*

 * idx can be of type enum memcg_stat_item or node_stat_item.

 * Keep in sync with memcg_exact_page_state().
@@ -557,7 +576,7 @@ void unlock_page_memcg(struct page *page); 
static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg,

						   int idx)

{

	long x = atomic_long_read(&memcg->vmstats[idx]);

	long x = atomic_long_read(&memcg->vmstats_local[idx]);

#ifdef CONFIG_SMP

	if (x < 0)

		x = 0;
@@ -609,7 +628,7 @@ static inline void mod_memcg_page_state(struct page *page, 
		mod_memcg_state(page->mem_cgroup, idx, val);

}



static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,

static inline unsigned long lruvec_page_state(struct lruvec *lruvec,

					      enum node_stat_item idx)

{

	struct mem_cgroup_per_node *pn;
@@ -627,6 +646,24 @@ static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec, 
	return x;

}



static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,

						    enum node_stat_item idx)

{

	struct mem_cgroup_per_node *pn;

	long x;



	if (mem_cgroup_disabled())

		return node_page_state(lruvec_pgdat(lruvec), idx);



	pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);

	x = atomic_long_read(&pn->lruvec_stat_local[idx]);

#ifdef CONFIG_SMP

	if (x < 0)

		x = 0;

#endif

	return x;

}



void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,

			int val);
@@ -959,6 +996,11 @@ static inline void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) 
{

}



static inline unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx)

{

	return 0;

}



static inline unsigned long memcg_page_state_local(struct mem_cgroup *memcg,

						   int idx)

{
@@ -989,6 +1031,12 @@ static inline void mod_memcg_page_state(struct page *page, 
{

}



static inline unsigned long lruvec_page_state(struct lruvec *lruvec,

					      enum node_stat_item idx)

{

	return node_page_state(lruvec_pgdat(lruvec), idx);

}



static inline unsigned long lruvec_page_state_local(struct lruvec *lruvec,

						    enum node_stat_item idx)

{

mm/memcontrol.c

+99 −106
Original line number Diff line number Diff line
@@ -702,12 +702,27 @@ void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) 


	x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]);

	if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {

		atomic_long_add(x, &memcg->vmstats[idx]);

		struct mem_cgroup *mi;



		atomic_long_add(x, &memcg->vmstats_local[idx]);

		for (mi = memcg; mi; mi = parent_mem_cgroup(mi))

			atomic_long_add(x, &mi->vmstats[idx]);

		x = 0;

	}

	__this_cpu_write(memcg->vmstats_percpu->stat[idx], x);

}



static struct mem_cgroup_per_node *

parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid)

{

	struct mem_cgroup *parent;



	parent = parent_mem_cgroup(pn->memcg);

	if (!parent)

		return NULL;

	return mem_cgroup_nodeinfo(parent, nid);

}



/**

 * __mod_lruvec_state - update lruvec memory statistics

 * @lruvec: the lruvec
@@ -721,24 +736,31 @@ void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) 
void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx,

			int val)

{

	pg_data_t *pgdat = lruvec_pgdat(lruvec);

	struct mem_cgroup_per_node *pn;

	struct mem_cgroup *memcg;

	long x;



	/* Update node */

	__mod_node_page_state(lruvec_pgdat(lruvec), idx, val);

	__mod_node_page_state(pgdat, idx, val);



	if (mem_cgroup_disabled())

		return;



	pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec);

	memcg = pn->memcg;



	/* Update memcg */

	__mod_memcg_state(pn->memcg, idx, val);

	__mod_memcg_state(memcg, idx, val);



	/* Update lruvec */

	x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]);

	if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) {

		atomic_long_add(x, &pn->lruvec_stat[idx]);

		struct mem_cgroup_per_node *pi;



		atomic_long_add(x, &pn->lruvec_stat_local[idx]);

		for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id))

			atomic_long_add(x, &pi->lruvec_stat[idx]);

		x = 0;

	}

	__this_cpu_write(pn->lruvec_stat_cpu->count[idx], x);
@@ -760,18 +782,26 @@ void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, 


	x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]);

	if (unlikely(x > MEMCG_CHARGE_BATCH)) {

		atomic_long_add(x, &memcg->vmevents[idx]);

		struct mem_cgroup *mi;



		atomic_long_add(x, &memcg->vmevents_local[idx]);

		for (mi = memcg; mi; mi = parent_mem_cgroup(mi))

			atomic_long_add(x, &mi->vmevents[idx]);

		x = 0;

	}

	__this_cpu_write(memcg->vmstats_percpu->events[idx], x);

}



static unsigned long memcg_events_local(struct mem_cgroup *memcg,

					int event)

static unsigned long memcg_events(struct mem_cgroup *memcg, int event)

{

	return atomic_long_read(&memcg->vmevents[event]);

}



static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event)

{

	return atomic_long_read(&memcg->vmevents_local[event]);

}



static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,

					 struct page *page,

					 bool compound, int nr_pages)
@@ -2157,7 +2187,7 @@ static void drain_all_stock(struct mem_cgroup *root_memcg) 
static int memcg_hotplug_cpu_dead(unsigned int cpu)

{

	struct memcg_stock_pcp *stock;

	struct mem_cgroup *memcg;

	struct mem_cgroup *memcg, *mi;



	stock = &per_cpu(memcg_stock, cpu);

	drain_stock(stock);
@@ -2170,8 +2200,11 @@ static int memcg_hotplug_cpu_dead(unsigned int cpu) 
			long x;



			x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0);

			if (x)

			if (x) {

				atomic_long_add(x, &memcg->vmstats_local[i]);

				for (mi = memcg; mi; mi = parent_mem_cgroup(mi))

					atomic_long_add(x, &memcg->vmstats[i]);

			}



			if (i >= NR_VM_NODE_STAT_ITEMS)

				continue;
@@ -2181,8 +2214,12 @@ static int memcg_hotplug_cpu_dead(unsigned int cpu) 


				pn = mem_cgroup_nodeinfo(memcg, nid);

				x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0);

				if (x)

				if (x) {

					atomic_long_add(x, &pn->lruvec_stat_local[i]);

					do {

						atomic_long_add(x, &pn->lruvec_stat[i]);

					} while ((pn = parent_nodeinfo(pn, nid)));

				}

			}

		}
@@ -2190,10 +2227,13 @@ static int memcg_hotplug_cpu_dead(unsigned int cpu) 
			long x;



			x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0);

			if (x)

			if (x) {

				atomic_long_add(x, &memcg->vmevents_local[i]);

				for (mi = memcg; mi; mi = parent_mem_cgroup(mi))

					atomic_long_add(x, &memcg->vmevents[i]);

			}

		}

	}



	return 0;

}
@@ -3021,54 +3061,15 @@ static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, 
	return retval;

}



struct accumulated_vmstats {

	unsigned long vmstats[MEMCG_NR_STAT];

	unsigned long vmevents[NR_VM_EVENT_ITEMS];

	unsigned long lru_pages[NR_LRU_LISTS];



	/* overrides for v1 */

	const unsigned int *vmstats_array;

	const unsigned int *vmevents_array;



	int vmstats_size;

	int vmevents_size;

};



static void accumulate_vmstats(struct mem_cgroup *memcg,

			       struct accumulated_vmstats *acc)

{

	struct mem_cgroup *mi;

	int i;



	for_each_mem_cgroup_tree(mi, memcg) {

		for (i = 0; i < acc->vmstats_size; i++)

			acc->vmstats[i] += memcg_page_state_local(mi,

				acc->vmstats_array ? acc->vmstats_array[i] : i);



		for (i = 0; i < acc->vmevents_size; i++)

			acc->vmevents[i] += memcg_events_local(mi,

				acc->vmevents_array

				? acc->vmevents_array[i] : i);



		for (i = 0; i < NR_LRU_LISTS; i++)

			acc->lru_pages[i] += memcg_page_state_local(mi,

							      NR_LRU_BASE + i);

	}

}



static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)

{

	unsigned long val = 0;

	unsigned long val;



	if (mem_cgroup_is_root(memcg)) {

		struct mem_cgroup *iter;



		for_each_mem_cgroup_tree(iter, memcg) {

			val += memcg_page_state_local(iter, MEMCG_CACHE);

			val += memcg_page_state_local(iter, MEMCG_RSS);

		val = memcg_page_state(memcg, MEMCG_CACHE) +

			memcg_page_state(memcg, MEMCG_RSS);

		if (swap)

				val += memcg_page_state_local(iter, MEMCG_SWAP);

		}

			val += memcg_page_state(memcg, MEMCG_SWAP);

	} else {

		if (!swap)

			val = page_counter_read(&memcg->memory);
@@ -3499,7 +3500,6 @@ static int memcg_stat_show(struct seq_file *m, void *v) 
	unsigned long memory, memsw;

	struct mem_cgroup *mi;

	unsigned int i;

	struct accumulated_vmstats acc;



	BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));

	BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
@@ -3533,27 +3533,21 @@ static int memcg_stat_show(struct seq_file *m, void *v) 
		seq_printf(m, "hierarchical_memsw_limit %llu\n",

			   (u64)memsw * PAGE_SIZE);



	memset(&acc, 0, sizeof(acc));

	acc.vmstats_size = ARRAY_SIZE(memcg1_stats);

	acc.vmstats_array = memcg1_stats;

	acc.vmevents_size = ARRAY_SIZE(memcg1_events);

	acc.vmevents_array = memcg1_events;

	accumulate_vmstats(memcg, &acc);



	for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {

		if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())

			continue;

		seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i],

			   (u64)acc.vmstats[i] * PAGE_SIZE);

			   (u64)memcg_page_state(memcg, i) * PAGE_SIZE);

	}



	for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)

		seq_printf(m, "total_%s %llu\n", memcg1_event_names[i],

			   (u64)acc.vmevents[i]);

			   (u64)memcg_events(memcg, i));



	for (i = 0; i < NR_LRU_LISTS; i++)

		seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i],

			   (u64)acc.lru_pages[i] * PAGE_SIZE);

			   (u64)memcg_page_state(memcg, NR_LRU_BASE + i) *

			   PAGE_SIZE);



#ifdef CONFIG_DEBUG_VM

	{
@@ -5646,7 +5640,6 @@ static int memory_events_show(struct seq_file *m, void *v) 
static int memory_stat_show(struct seq_file *m, void *v)

{

	struct mem_cgroup *memcg = mem_cgroup_from_seq(m);

	struct accumulated_vmstats acc;

	int i;



	/*
@@ -5660,31 +5653,27 @@ static int memory_stat_show(struct seq_file *m, void *v) 
	 * Current memory state:

	 */



	memset(&acc, 0, sizeof(acc));

	acc.vmstats_size = MEMCG_NR_STAT;

	acc.vmevents_size = NR_VM_EVENT_ITEMS;

	accumulate_vmstats(memcg, &acc);



	seq_printf(m, "anon %llu\n",

		   (u64)acc.vmstats[MEMCG_RSS] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, MEMCG_RSS) * PAGE_SIZE);

	seq_printf(m, "file %llu\n",

		   (u64)acc.vmstats[MEMCG_CACHE] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, MEMCG_CACHE) * PAGE_SIZE);

	seq_printf(m, "kernel_stack %llu\n",

		   (u64)acc.vmstats[MEMCG_KERNEL_STACK_KB] * 1024);

		   (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * 1024);

	seq_printf(m, "slab %llu\n",

		   (u64)(acc.vmstats[NR_SLAB_RECLAIMABLE] +

			 acc.vmstats[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);

		   (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) +

			 memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) *

		   PAGE_SIZE);

	seq_printf(m, "sock %llu\n",

		   (u64)acc.vmstats[MEMCG_SOCK] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, MEMCG_SOCK) * PAGE_SIZE);



	seq_printf(m, "shmem %llu\n",

		   (u64)acc.vmstats[NR_SHMEM] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, NR_SHMEM) * PAGE_SIZE);

	seq_printf(m, "file_mapped %llu\n",

		   (u64)acc.vmstats[NR_FILE_MAPPED] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * PAGE_SIZE);

	seq_printf(m, "file_dirty %llu\n",

		   (u64)acc.vmstats[NR_FILE_DIRTY] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * PAGE_SIZE);

	seq_printf(m, "file_writeback %llu\n",

		   (u64)acc.vmstats[NR_WRITEBACK] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, NR_WRITEBACK) * PAGE_SIZE);



	/*

	 * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter
@@ -5693,43 +5682,47 @@ static int memory_stat_show(struct seq_file *m, void *v) 
	 * where the page->mem_cgroup is set up and stable.

	 */

	seq_printf(m, "anon_thp %llu\n",

		   (u64)acc.vmstats[MEMCG_RSS_HUGE] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * PAGE_SIZE);



	for (i = 0; i < NR_LRU_LISTS; i++)

		seq_printf(m, "%s %llu\n", mem_cgroup_lru_names[i],

			   (u64)acc.lru_pages[i] * PAGE_SIZE);

			   (u64)memcg_page_state(memcg, NR_LRU_BASE + i) *

			   PAGE_SIZE);



	seq_printf(m, "slab_reclaimable %llu\n",

		   (u64)acc.vmstats[NR_SLAB_RECLAIMABLE] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) *

		   PAGE_SIZE);

	seq_printf(m, "slab_unreclaimable %llu\n",

		   (u64)acc.vmstats[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE);

		   (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) *

		   PAGE_SIZE);



	/* Accumulated memory events */



	seq_printf(m, "pgfault %lu\n", acc.vmevents[PGFAULT]);

	seq_printf(m, "pgmajfault %lu\n", acc.vmevents[PGMAJFAULT]);

	seq_printf(m, "pgfault %lu\n", memcg_events(memcg, PGFAULT));

	seq_printf(m, "pgmajfault %lu\n", memcg_events(memcg, PGMAJFAULT));



	seq_printf(m, "workingset_refault %lu\n",

		   acc.vmstats[WORKINGSET_REFAULT]);

		   memcg_page_state(memcg, WORKINGSET_REFAULT));

	seq_printf(m, "workingset_activate %lu\n",

		   acc.vmstats[WORKINGSET_ACTIVATE]);

		   memcg_page_state(memcg, WORKINGSET_ACTIVATE));

	seq_printf(m, "workingset_nodereclaim %lu\n",

		   acc.vmstats[WORKINGSET_NODERECLAIM]);



	seq_printf(m, "pgrefill %lu\n", acc.vmevents[PGREFILL]);

	seq_printf(m, "pgscan %lu\n", acc.vmevents[PGSCAN_KSWAPD] +

		   acc.vmevents[PGSCAN_DIRECT]);

	seq_printf(m, "pgsteal %lu\n", acc.vmevents[PGSTEAL_KSWAPD] +

		   acc.vmevents[PGSTEAL_DIRECT]);

	seq_printf(m, "pgactivate %lu\n", acc.vmevents[PGACTIVATE]);

	seq_printf(m, "pgdeactivate %lu\n", acc.vmevents[PGDEACTIVATE]);

	seq_printf(m, "pglazyfree %lu\n", acc.vmevents[PGLAZYFREE]);

	seq_printf(m, "pglazyfreed %lu\n", acc.vmevents[PGLAZYFREED]);

		   memcg_page_state(memcg, WORKINGSET_NODERECLAIM));



	seq_printf(m, "pgrefill %lu\n", memcg_events(memcg, PGREFILL));

	seq_printf(m, "pgscan %lu\n", memcg_events(memcg, PGSCAN_KSWAPD) +

		   memcg_events(memcg, PGSCAN_DIRECT));

	seq_printf(m, "pgsteal %lu\n", memcg_events(memcg, PGSTEAL_KSWAPD) +

		   memcg_events(memcg, PGSTEAL_DIRECT));

	seq_printf(m, "pgactivate %lu\n", memcg_events(memcg, PGACTIVATE));

	seq_printf(m, "pgdeactivate %lu\n", memcg_events(memcg, PGDEACTIVATE));

	seq_printf(m, "pglazyfree %lu\n", memcg_events(memcg, PGLAZYFREE));

	seq_printf(m, "pglazyfreed %lu\n", memcg_events(memcg, PGLAZYFREED));



#ifdef CONFIG_TRANSPARENT_HUGEPAGE

	seq_printf(m, "thp_fault_alloc %lu\n", acc.vmevents[THP_FAULT_ALLOC]);

	seq_printf(m, "thp_fault_alloc %lu\n",

		   memcg_events(memcg, THP_FAULT_ALLOC));

	seq_printf(m, "thp_collapse_alloc %lu\n",

		   acc.vmevents[THP_COLLAPSE_ALLOC]);

		   memcg_events(memcg, THP_COLLAPSE_ALLOC));

#endif /* CONFIG_TRANSPARENT_HUGEPAGE */



	return 0;

CRA Git | Maintained and supported by SUSTech CRA and CCSE