Skip to contentSkip to conversation

mm, memcg: unify reclaim retry limits with page allocator

PATCH In review Chris Down 4 participants 6 messages May 20, 2020 – May 22, 2020

Subsystems (3)
lore
Share & tools
lore

Chris Down

author
#0 lore raw

PATCH 1/1 mm, memcg: unify reclaim retry limits with page allocator

Reclaim retries have been set to 5 since the beginning of time in 66e1707bc346 ("Memory controller: add per cgroup LRU and reclaim"). However, we now have a generally agreed-upon standard for page reclaim:
MAX_RECLAIM_RETRIES (currently 16), added many years later in 0a0337e0d1d1 ("mm, oom: rework oom detection").

In the absence of a compelling reason to declare an OOM earlier in memcg context than page allocator context, it seems reasonable to supplant MEM_CGROUP_RECLAIM_RETRIES with MAX_RECLAIM_RETRIES, making the page allocator and memcg internals more similar in semantics when reclaim fails to produce results, avoiding premature OOMs or throttling.

Cc Andrew Morton <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Michal Hocko <[email protected]>
---
 mm/memcontrol.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)
mm/memcontrol.c +6 -9
index b040951ccd6b..d3b23c57bed4 100644
@@ -73,9 +73,6 @@ EXPORT_SYMBOL(memory_cgrp_subsys);
 
 struct mem_cgroup *root_mem_cgroup __read_mostly;
 
-/* The number of times we should retry reclaim failures before giving up. */
-#define MEM_CGROUP_RECLAIM_RETRIES	5
-
 /* Socket memory accounting disabled? */
 static bool cgroup_memory_nosocket;
 
@@ -2386,7 +2383,7 @@ void mem_cgroup_handle_over_high(void)
 	unsigned long pflags;
 	unsigned long nr_reclaimed;
 	unsigned int nr_pages = current->memcg_nr_pages_over_high;
-	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	int nr_retries = MAX_RECLAIM_RETRIES;
 	struct mem_cgroup *memcg;
 
 	if (likely(!nr_pages))
@@ -2438,7 +2435,7 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
 		      unsigned int nr_pages)
 {
 	unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages);
-	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	int nr_retries = MAX_RECLAIM_RETRIES;
 	struct mem_cgroup *mem_over_limit;
 	struct page_counter *counter;
 	unsigned long nr_reclaimed;
@@ -2557,7 +2554,7 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
 		       get_order(nr_pages * PAGE_SIZE));
 	switch (oom_status) {
 	case OOM_SUCCESS:
-		nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+		nr_retries = MAX_RECLAIM_RETRIES;
 		goto retry;
 	case OOM_FAILED:
 		goto force;
@@ -3168,7 +3165,7 @@ static inline bool memcg_has_children(struct mem_cgroup *memcg)
  */
 static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
 {
-	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	int nr_retries = MAX_RECLAIM_RETRIES;
 
 	/* we call try-to-free pages for make this cgroup empty */
 	lru_add_drain_all();
@@ -6001,7 +5998,7 @@ static ssize_t memory_high_write(struct kernfs_open_file *of,
 				 char *buf, size_t nbytes, loff_t off)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
-	unsigned int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	unsigned int nr_retries = MAX_RECLAIM_RETRIES;
 	bool drained = false;
 	unsigned long high;
 	int err;
@@ -6049,7 +6046,7 @@ static ssize_t memory_max_write(struct kernfs_open_file *of,
 				char *buf, size_t nbytes, loff_t off)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
-	unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
+	unsigned int nr_reclaims = MAX_RECLAIM_RETRIES;
 	bool drained = false;
 	unsigned long max;
 	int err;
Signature · 1 line
2.26.2
maintainer Acked-by on v1 1/1
#1 lore raw
Quoted 12 lines · Chris Down

Reclaim retries have been set to 5 since the beginning of time in 66e1707bc346 ("Memory controller: add per cgroup LRU and reclaim"). However, we now have a generally agreed-upon standard for page reclaim:
MAX_RECLAIM_RETRIES (currently 16), added many years later in 0a0337e0d1d1 ("mm, oom: rework oom detection").

In the absence of a compelling reason to declare an OOM earlier in memcg context than page allocator context, it seems reasonable to supplant MEM_CGROUP_RECLAIM_RETRIES with MAX_RECLAIM_RETRIES, making the page allocator and memcg internals more similar in semantics when reclaim fails to produce results, avoiding premature OOMs or throttling.

Quoted 4 lines
Cc Andrew Morton <[email protected]>
Cc Johannes Weiner <[email protected]>
Cc Michal Hocko <[email protected]>

I have already expressed my dislike to some of the MEM_CGROUP_RECLAIM_RETRIES usage but using MAX_RECLAIM_RETRIES is a reasonable thing to do. There is simply no reason to have two retry limits.

Quoted 74 lines
---
 mm/memcontrol.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)
mm/memcontrol.c +6 -9
index b040951ccd6b..d3b23c57bed4 100644
@@ -73,9 +73,6 @@ EXPORT_SYMBOL(memory_cgrp_subsys);
 
 struct mem_cgroup *root_mem_cgroup __read_mostly;
 
-/* The number of times we should retry reclaim failures before giving up. */
-#define MEM_CGROUP_RECLAIM_RETRIES	5
-
 /* Socket memory accounting disabled? */
 static bool cgroup_memory_nosocket;
 
@@ -2386,7 +2383,7 @@ void mem_cgroup_handle_over_high(void)
 	unsigned long pflags;
 	unsigned long nr_reclaimed;
 	unsigned int nr_pages = current->memcg_nr_pages_over_high;
-	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	int nr_retries = MAX_RECLAIM_RETRIES;
 	struct mem_cgroup *memcg;
 
 	if (likely(!nr_pages))
@@ -2438,7 +2435,7 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
 		      unsigned int nr_pages)
 {
 	unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages);
-	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	int nr_retries = MAX_RECLAIM_RETRIES;
 	struct mem_cgroup *mem_over_limit;
 	struct page_counter *counter;
 	unsigned long nr_reclaimed;
@@ -2557,7 +2554,7 @@ static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
 		       get_order(nr_pages * PAGE_SIZE));
 	switch (oom_status) {
 	case OOM_SUCCESS:
-		nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+		nr_retries = MAX_RECLAIM_RETRIES;
 		goto retry;
 	case OOM_FAILED:
 		goto force;
@@ -3168,7 +3165,7 @@ static inline bool memcg_has_children(struct mem_cgroup *memcg)
  */
 static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
 {
-	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	int nr_retries = MAX_RECLAIM_RETRIES;
 
 	/* we call try-to-free pages for make this cgroup empty */
 	lru_add_drain_all();
@@ -6001,7 +5998,7 @@ static ssize_t memory_high_write(struct kernfs_open_file *of,
 				 char *buf, size_t nbytes, loff_t off)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
-	unsigned int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	unsigned int nr_retries = MAX_RECLAIM_RETRIES;
 	bool drained = false;
 	unsigned long high;
 	int err;
@@ -6049,7 +6046,7 @@ static ssize_t memory_max_write(struct kernfs_open_file *of,
 				char *buf, size_t nbytes, loff_t off)
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
-	unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
+	unsigned int nr_reclaims = MAX_RECLAIM_RETRIES;
 	bool drained = false;
 	unsigned long max;
 	int err;
Signature · 1 line
2.26.2
Signature · 2 lines
Michal Hocko
SUSE Labs
maintainer
#2 lore raw
In reply to Chris Down
Quoted 17 lines · On Wed, 20 May 2020 17:31:42 +0100 Chris Down…

Reclaim retries have been set to 5 since the beginning of time in 66e1707bc346 ("Memory controller: add per cgroup LRU and reclaim"). However, we now have a generally agreed-upon standard for page reclaim:
MAX_RECLAIM_RETRIES (currently 16), added many years later in 0a0337e0d1d1 ("mm, oom: rework oom detection").

In the absence of a compelling reason to declare an OOM earlier in memcg context than page allocator context, it seems reasonable to supplant MEM_CGROUP_RECLAIM_RETRIES with MAX_RECLAIM_RETRIES, making the page allocator and memcg internals more similar in semantics when reclaim fails to produce results, avoiding premature OOMs or throttling.

...

mm/memcontrol.c EXPORT_SYMBOL(memory_cgrp_subsys); · line 76

Quoted 4 lines
mm/memcontrol.c +0 -1
@@ -73,9 +73,6 @@ EXPORT_SYMBOL(memory_cgrp_subsys);
 
 struct mem_cgroup *root_mem_cgroup __read_mostly;
 
-/* The number of times we should retry reclaim failures before giving up. */

hm, what tree is this against?

mm/memcontrol.c void mem_cgroup_handle_over_high(void) · line 2386

Quoted 11 lines
mm/memcontrol.c +1 -3
-#define MEM_CGROUP_RECLAIM_RETRIES	5
-
 /* Socket memory accounting disabled? */
 static bool cgroup_memory_nosocket;
 
@@ -2386,7 +2383,7 @@ void mem_cgroup_handle_over_high(void)
 	unsigned long pflags;
 	unsigned long nr_reclaimed;
 	unsigned int nr_pages = current->memcg_nr_pages_over_high;
-	int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
+	int nr_retries = MAX_RECLAIM_RETRIES;

I can't seem to find a tree in which mem_cgroup_handle_over_high() has a local `nr_retries'.

maintainer
#3 lore raw
Quoted 3 lines · On Wed, 20 May 2020 16:40:37 -0700 Andrew Morton <[email protected]>…
Quoted 1 line

-/* The number of times we should retry reclaim failures before giving up. */

hm, what tree is this against?

Ah, my habit of working in reverse time order sometimes does this ;)

I suggest that "mm, memcg: reclaim more aggressively before high allocator throttling" and this patch become a two-patch series?

author
#4 lore raw

Hey Andrew,

Quoted 4 lines · Andrew Morton

Ah, my habit of working in reverse time order sometimes does this ;)

I suggest that "mm, memcg: reclaim more aggressively before high allocator throttling" and this patch become a two-patch series?

Sure, they can do (sorry, I meant to add a comment mentioning the dependency, but forgot). I just didn't want to conflate discussion for both of them, since they are separate in nature :-)

I'll hold off on sending v2 until the discussion with Michal is finished.

Reviewed-by on v1 1/1
#5 lore raw
In reply to Chris Down
Quoted 7 lines · Chris Down

Reclaim retries have been set to 5 since the beginning of time in 66e1707bc346 ("Memory controller: add per cgroup LRU and reclaim"). However, we now have a generally agreed-upon standard for page reclaim:
MAX_RECLAIM_RETRIES (currently 16), added many years later in 0a0337e0d1d1 ("mm, oom: rework oom detection").

Quoted 7 lines

In the absence of a compelling reason to declare an OOM earlier in memcg context than page allocator context, it seems reasonable to supplant MEM_CGROUP_RECLAIM_RETRIES with MAX_RECLAIM_RETRIES, making the page allocator and memcg internals more similar in semantics when reclaim fails to produce results, avoiding premature OOMs or throttling.