[Devel] [PATCH RHEL7 COMMIT] mm/memcg: Use per-cpu stock charges for ->kmem and ->cache counters

Vasily Averin vvs at virtuozzo.com
Tue Jul 21 17:59:50 MSK 2020


The commit is pushed to "branch-rh7-3.10.0-1127.10.1.vz7.162.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh7-3.10.0-1127.10.1.vz7.162.13
------>
commit ef2a900bbfcb696536f14020f71112f966fd67c4
Author: Andrey Ryabinin <aryabinin at virtuozzo.com>
Date:   Tue Jul 21 17:59:50 2020 +0300

    mm/memcg: Use per-cpu stock charges for ->kmem and ->cache counters
    
    Currently we use per-cpu stocks to do precharges of the ->memory and ->memsw
    counters. Do this for the ->kmem and ->cache as well to decrease contention
    on these counters as well.
    
    https://jira.sw.ru/browse/PSBM-101300
    Signed-off-by: Andrey Ryabinin <aryabinin at virtuozzo.com>
---
 mm/memcontrol.c | 71 +++++++++++++++++++++++++++++++++++++--------------------
 1 file changed, 46 insertions(+), 25 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 15cd07144d5a0..fa7b2cad93add 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2811,6 +2811,8 @@ void mem_cgroup_update_page_stat(struct page *page,
 struct memcg_stock_pcp {
 	struct mem_cgroup *cached; /* this never be root cgroup */
 	unsigned int nr_pages;
+	unsigned int cache_nr_pages;
+	unsigned int kmem_nr_pages;
 	struct work_struct work;
 	unsigned long flags;
 #define FLUSHING_CACHED_CHARGE	0
@@ -2829,7 +2831,8 @@ static DEFINE_MUTEX(percpu_charge_mutex);
  *
  * returns true if successful, false otherwise.
  */
-static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
+static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages,
+			bool cache, bool kmem)
 {
 	struct memcg_stock_pcp *stock;
 	bool ret = false;
@@ -2838,9 +2841,19 @@ static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
 		return ret;
 
 	stock = &get_cpu_var(memcg_stock);
-	if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
-		stock->nr_pages -= nr_pages;
-		ret = true;
+	if (memcg == stock->cached) {
+		if (cache && stock->cache_nr_pages >= nr_pages) {
+			stock->cache_nr_pages -= nr_pages;
+			ret = true;
+		}
+		if (kmem && stock->kmem_nr_pages >= nr_pages) {
+			stock->kmem_nr_pages -= nr_pages;
+			ret = true;
+		}
+		if (!cache && !kmem && stock->nr_pages >= nr_pages) {
+			stock->nr_pages -= nr_pages;
+			ret = true;
+		}
 	}
 	put_cpu_var(memcg_stock);
 	return ret;
@@ -2852,12 +2865,20 @@ static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
 static void drain_stock(struct memcg_stock_pcp *stock)
 {
 	struct mem_cgroup *old = stock->cached;
+	unsigned long nr_pages = stock->nr_pages + stock->cache_nr_pages + stock->kmem_nr_pages;
+
+	if (stock->cache_nr_pages)
+		page_counter_uncharge(&old->cache, stock->cache_nr_pages);
+	if (stock->kmem_nr_pages)
+		page_counter_uncharge(&old->kmem, stock->kmem_nr_pages);
 
-	if (stock->nr_pages) {
-		page_counter_uncharge(&old->memory, stock->nr_pages);
+	if (nr_pages) {
+		page_counter_uncharge(&old->memory, nr_pages);
 		if (do_swap_account)
-			page_counter_uncharge(&old->memsw, stock->nr_pages);
+			page_counter_uncharge(&old->memsw, nr_pages);
 		stock->nr_pages = 0;
+		stock->kmem_nr_pages = 0;
+		stock->cache_nr_pages = 0;
 	}
 	stock->cached = NULL;
 }
@@ -2888,7 +2909,8 @@ static void __init memcg_stock_init(void)
  * Cache charges(val) to local per_cpu area.
  * This will be consumed by consume_stock() function, later.
  */
-static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
+static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages,
+			bool cache, bool kmem)
 {
 	struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
 
@@ -2896,7 +2918,13 @@ static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
 		drain_stock(stock);
 		stock->cached = memcg;
 	}
-	stock->nr_pages += nr_pages;
+
+	if (cache)
+		stock->cache_nr_pages += nr_pages;
+	else if (kmem)
+		stock->kmem_nr_pages += nr_pages;
+	else
+		stock->nr_pages += nr_pages;
 	put_cpu_var(memcg_stock);
 }
 
@@ -2917,7 +2945,7 @@ static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync)
 		struct mem_cgroup *memcg;
 
 		memcg = stock->cached;
-		if (!memcg || !stock->nr_pages)
+		if (!memcg || !(stock->nr_pages + stock->kmem_nr_pages + stock->cache_nr_pages))
 			continue;
 		if (!mem_cgroup_same_or_subtree(root_memcg, memcg))
 			continue;
@@ -3108,21 +3136,13 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, bool kmem_charge
 	bool drained = false;
 
 	if (mem_cgroup_is_root(memcg))
-		goto done;
+		return 0;
 retry:
 	flags = 0;
 
-	if (consume_stock(memcg, nr_pages)) {
-		if (kmem_charge && !page_counter_try_charge(
-				&memcg->kmem, nr_pages, &counter)) {
-			refill_stock(memcg, nr_pages);
-			goto charge;
-		}
-
+	if (consume_stock(memcg, nr_pages, cache_charge, kmem_charge))
 		goto done;
-	}
 
-charge:
 	mem_over_limit = NULL;
 	if (page_counter_try_charge(&memcg->memory, batch, &counter)) {
 		if (do_swap_account && !page_counter_try_charge(
@@ -3135,7 +3155,7 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, bool kmem_charge
 		mem_over_limit = mem_cgroup_from_counter(counter, memory);
 
 	if (!mem_over_limit && kmem_charge) {
-		if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
+		if (!page_counter_try_charge(&memcg->kmem, batch, &counter)) {
 			flags |= MEM_CGROUP_RECLAIM_KMEM;
 			mem_over_limit = mem_cgroup_from_counter(counter, kmem);
 			page_counter_uncharge(&memcg->memory, batch);
@@ -3264,12 +3284,13 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, bool kmem_charge
 		goto bypass;
 	}
 
-	if (batch > nr_pages)
-		refill_stock(memcg, batch - nr_pages);
-done:
 	if (cache_charge)
-		page_counter_charge(&memcg->cache, nr_pages);
+		page_counter_charge(&memcg->cache, batch);
+
+	if (batch > nr_pages)
+		refill_stock(memcg, batch - nr_pages, cache_charge, kmem_charge);
 
+done:
 	/*
 	 * If the hierarchy is above the normal consumption range, schedule
 	 * reclaim on returning to userland.  We can perform reclaim here


More information about the Devel mailing list