2008-02-07 08:13:50 +00:00
|
|
|
/* memcontrol.c - Memory Controller
|
|
|
|
*
|
|
|
|
* Copyright IBM Corporation, 2007
|
|
|
|
* Author Balbir Singh <balbir@linux.vnet.ibm.com>
|
|
|
|
*
|
2008-02-07 08:13:51 +00:00
|
|
|
* Copyright 2007 OpenVZ SWsoft Inc
|
|
|
|
* Author: Pavel Emelianov <xemul@openvz.org>
|
|
|
|
*
|
2010-03-10 23:22:24 +00:00
|
|
|
* Memory thresholds
|
|
|
|
* Copyright (C) 2009 Nokia Corporation
|
|
|
|
* Author: Kirill A. Shutemov
|
|
|
|
*
|
2008-02-07 08:13:50 +00:00
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of the GNU General Public License as published by
|
|
|
|
* the Free Software Foundation; either version 2 of the License, or
|
|
|
|
* (at your option) any later version.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/res_counter.h>
|
|
|
|
#include <linux/memcontrol.h>
|
|
|
|
#include <linux/cgroup.h>
|
2008-02-07 08:13:51 +00:00
|
|
|
#include <linux/mm.h>
|
2010-03-10 23:22:14 +00:00
|
|
|
#include <linux/hugetlb.h>
|
2009-01-08 02:07:56 +00:00
|
|
|
#include <linux/pagemap.h>
|
2008-02-07 08:14:24 +00:00
|
|
|
#include <linux/smp.h>
|
2008-02-07 08:13:53 +00:00
|
|
|
#include <linux/page-flags.h>
|
2008-02-07 08:13:56 +00:00
|
|
|
#include <linux/backing-dev.h>
|
2008-02-07 08:13:53 +00:00
|
|
|
#include <linux/bit_spinlock.h>
|
|
|
|
#include <linux/rcupdate.h>
|
2009-04-02 23:57:39 +00:00
|
|
|
#include <linux/limits.h>
|
2009-01-08 02:08:00 +00:00
|
|
|
#include <linux/mutex.h>
|
2009-09-23 22:56:37 +00:00
|
|
|
#include <linux/rbtree.h>
|
2008-04-29 08:00:19 +00:00
|
|
|
#include <linux/slab.h>
|
2008-02-07 08:13:56 +00:00
|
|
|
#include <linux/swap.h>
|
2010-03-10 23:22:17 +00:00
|
|
|
#include <linux/swapops.h>
|
2008-02-07 08:13:56 +00:00
|
|
|
#include <linux/spinlock.h>
|
2010-03-10 23:22:24 +00:00
|
|
|
#include <linux/eventfd.h>
|
|
|
|
#include <linux/sort.h>
|
2008-02-07 08:13:56 +00:00
|
|
|
#include <linux/fs.h>
|
2008-02-07 08:14:25 +00:00
|
|
|
#include <linux/seq_file.h>
|
2008-04-29 08:00:24 +00:00
|
|
|
#include <linux/vmalloc.h>
|
2008-10-19 03:26:14 +00:00
|
|
|
#include <linux/mm_inline.h>
|
2008-10-19 03:28:16 +00:00
|
|
|
#include <linux/page_cgroup.h>
|
2009-12-16 00:47:08 +00:00
|
|
|
#include <linux/cpu.h>
|
2010-08-11 01:03:00 +00:00
|
|
|
#include <linux/oom.h>
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
#include "internal.h"
|
2008-02-07 08:13:50 +00:00
|
|
|
|
2008-02-07 08:13:59 +00:00
|
|
|
#include <asm/uaccess.h>
|
|
|
|
|
2010-08-10 00:19:57 +00:00
|
|
|
#include <trace/events/vmscan.h>
|
|
|
|
|
2008-07-25 08:47:08 +00:00
|
|
|
struct cgroup_subsys mem_cgroup_subsys __read_mostly;
|
|
|
|
#define MEM_CGROUP_RECLAIM_RETRIES 5
|
2009-09-23 22:56:32 +00:00
|
|
|
struct mem_cgroup *root_mem_cgroup __read_mostly;
|
2008-02-07 08:13:50 +00:00
|
|
|
|
2009-01-08 02:07:57 +00:00
|
|
|
#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
|
2009-06-17 23:27:15 +00:00
|
|
|
/* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
|
2009-01-08 02:07:57 +00:00
|
|
|
int do_swap_account __read_mostly;
|
|
|
|
static int really_do_swap_account __initdata = 1; /* for remember boot option*/
|
|
|
|
#else
|
|
|
|
#define do_swap_account (0)
|
|
|
|
#endif
|
|
|
|
|
2010-03-10 23:22:31 +00:00
|
|
|
/*
|
|
|
|
* Per memcg event counter is incremented at every pagein/pageout. This counter
|
|
|
|
* is used for trigger some periodic events. This is straightforward and better
|
|
|
|
* than using jiffies etc. to handle periodic memcg event.
|
|
|
|
*
|
|
|
|
* These values will be used as !((event) & ((1 <<(thresh)) - 1))
|
|
|
|
*/
|
|
|
|
#define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
|
|
|
|
#define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
|
2009-01-08 02:07:57 +00:00
|
|
|
|
2008-02-07 08:14:24 +00:00
|
|
|
/*
|
|
|
|
* Statistics for memory cgroup.
|
|
|
|
*/
|
|
|
|
enum mem_cgroup_stat_index {
|
|
|
|
/*
|
|
|
|
* For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
|
|
|
|
*/
|
|
|
|
MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
|
2009-06-17 23:26:34 +00:00
|
|
|
MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */
|
2009-12-16 00:47:09 +00:00
|
|
|
MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */
|
2008-05-01 11:35:12 +00:00
|
|
|
MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
|
|
|
|
MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
|
2009-09-23 22:56:42 +00:00
|
|
|
MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */
|
2010-03-10 23:22:31 +00:00
|
|
|
MEM_CGROUP_EVENTS, /* incremented at every pagein/pageout */
|
2008-02-07 08:14:24 +00:00
|
|
|
|
|
|
|
MEM_CGROUP_STAT_NSTATS,
|
|
|
|
};
|
|
|
|
|
|
|
|
struct mem_cgroup_stat_cpu {
|
|
|
|
s64 count[MEM_CGROUP_STAT_NSTATS];
|
|
|
|
};
|
|
|
|
|
2008-02-07 08:14:31 +00:00
|
|
|
/*
|
|
|
|
* per-zone information in memory controller.
|
|
|
|
*/
|
|
|
|
struct mem_cgroup_per_zone {
|
2008-02-07 08:14:39 +00:00
|
|
|
/*
|
|
|
|
* spin_lock to protect the per cgroup LRU
|
|
|
|
*/
|
2008-10-19 03:26:14 +00:00
|
|
|
struct list_head lists[NR_LRU_LISTS];
|
|
|
|
unsigned long count[NR_LRU_LISTS];
|
2009-01-08 02:08:20 +00:00
|
|
|
|
|
|
|
struct zone_reclaim_stat reclaim_stat;
|
2009-09-23 22:56:37 +00:00
|
|
|
struct rb_node tree_node; /* RB tree node */
|
|
|
|
unsigned long long usage_in_excess;/* Set to the value by which */
|
|
|
|
/* the soft limit is exceeded*/
|
|
|
|
bool on_tree;
|
2009-09-23 22:56:39 +00:00
|
|
|
struct mem_cgroup *mem; /* Back pointer, we cannot */
|
|
|
|
/* use container_of */
|
2008-02-07 08:14:31 +00:00
|
|
|
};
|
|
|
|
/* Macro for accessing counter */
|
|
|
|
#define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
|
|
|
|
|
|
|
|
struct mem_cgroup_per_node {
|
|
|
|
struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
|
|
|
|
};
|
|
|
|
|
|
|
|
struct mem_cgroup_lru_info {
|
|
|
|
struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
|
|
|
|
};
|
|
|
|
|
2009-09-23 22:56:37 +00:00
|
|
|
/*
|
|
|
|
* Cgroups above their limits are maintained in a RB-Tree, independent of
|
|
|
|
* their hierarchy representation
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct mem_cgroup_tree_per_zone {
|
|
|
|
struct rb_root rb_root;
|
|
|
|
spinlock_t lock;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct mem_cgroup_tree_per_node {
|
|
|
|
struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
|
|
|
|
};
|
|
|
|
|
|
|
|
struct mem_cgroup_tree {
|
|
|
|
struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct mem_cgroup_tree soft_limit_tree __read_mostly;
|
|
|
|
|
2010-03-10 23:22:24 +00:00
|
|
|
struct mem_cgroup_threshold {
|
|
|
|
struct eventfd_ctx *eventfd;
|
|
|
|
u64 threshold;
|
|
|
|
};
|
|
|
|
|
2010-05-26 21:42:36 +00:00
|
|
|
/* For threshold */
|
2010-03-10 23:22:24 +00:00
|
|
|
struct mem_cgroup_threshold_ary {
|
|
|
|
/* An array index points to threshold just below usage. */
|
2010-05-26 21:42:42 +00:00
|
|
|
int current_threshold;
|
2010-03-10 23:22:24 +00:00
|
|
|
/* Size of entries[] */
|
|
|
|
unsigned int size;
|
|
|
|
/* Array of thresholds */
|
|
|
|
struct mem_cgroup_threshold entries[0];
|
|
|
|
};
|
2010-05-26 21:42:47 +00:00
|
|
|
|
|
|
|
struct mem_cgroup_thresholds {
|
|
|
|
/* Primary thresholds array */
|
|
|
|
struct mem_cgroup_threshold_ary *primary;
|
|
|
|
/*
|
|
|
|
* Spare threshold array.
|
|
|
|
* This is needed to make mem_cgroup_unregister_event() "never fail".
|
|
|
|
* It must be able to store at least primary->size - 1 entries.
|
|
|
|
*/
|
|
|
|
struct mem_cgroup_threshold_ary *spare;
|
|
|
|
};
|
|
|
|
|
2010-05-26 21:42:36 +00:00
|
|
|
/* for OOM */
|
|
|
|
struct mem_cgroup_eventfd_list {
|
|
|
|
struct list_head list;
|
|
|
|
struct eventfd_ctx *eventfd;
|
|
|
|
};
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
static void mem_cgroup_threshold(struct mem_cgroup *mem);
|
2010-05-26 21:42:36 +00:00
|
|
|
static void mem_cgroup_oom_notify(struct mem_cgroup *mem);
|
2010-03-10 23:22:24 +00:00
|
|
|
|
2008-02-07 08:13:50 +00:00
|
|
|
/*
|
|
|
|
* The memory controller data structure. The memory controller controls both
|
|
|
|
* page cache and RSS per cgroup. We would eventually like to provide
|
|
|
|
* statistics based on the statistics developed by Rik Van Riel for clock-pro,
|
|
|
|
* to help the administrator determine what knobs to tune.
|
|
|
|
*
|
|
|
|
* TODO: Add a water mark for the memory controller. Reclaim will begin when
|
2008-02-07 08:13:53 +00:00
|
|
|
* we hit the water mark. May be even add a low water mark, such that
|
|
|
|
* no reclaim occurs from a cgroup at it's low water mark, this is
|
|
|
|
* a feature that will be implemented much later in the future.
|
2008-02-07 08:13:50 +00:00
|
|
|
*/
|
|
|
|
struct mem_cgroup {
|
|
|
|
struct cgroup_subsys_state css;
|
|
|
|
/*
|
|
|
|
* the counter to account for memory usage
|
|
|
|
*/
|
|
|
|
struct res_counter res;
|
2009-01-08 02:08:00 +00:00
|
|
|
/*
|
|
|
|
* the counter to account for mem+swap usage.
|
|
|
|
*/
|
|
|
|
struct res_counter memsw;
|
2008-02-07 08:13:51 +00:00
|
|
|
/*
|
|
|
|
* Per cgroup active and inactive list, similar to the
|
|
|
|
* per zone LRU lists.
|
|
|
|
*/
|
2008-02-07 08:14:31 +00:00
|
|
|
struct mem_cgroup_lru_info info;
|
2008-02-07 08:14:39 +00:00
|
|
|
|
2009-01-08 02:08:23 +00:00
|
|
|
/*
|
|
|
|
protect against reclaim related member.
|
|
|
|
*/
|
|
|
|
spinlock_t reclaim_param_lock;
|
|
|
|
|
2009-01-08 02:08:06 +00:00
|
|
|
/*
|
tree-wide: fix assorted typos all over the place
That is "success", "unknown", "through", "performance", "[re|un]mapping"
, "access", "default", "reasonable", "[con]currently", "temperature"
, "channel", "[un]used", "application", "example","hierarchy", "therefore"
, "[over|under]flow", "contiguous", "threshold", "enough" and others.
Signed-off-by: André Goddard Rosa <andre.goddard@gmail.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2009-11-14 15:09:05 +00:00
|
|
|
* While reclaiming in a hierarchy, we cache the last child we
|
2009-04-02 23:57:33 +00:00
|
|
|
* reclaimed from.
|
2009-01-08 02:08:06 +00:00
|
|
|
*/
|
2009-04-02 23:57:33 +00:00
|
|
|
int last_scanned_child;
|
2009-01-08 02:08:07 +00:00
|
|
|
/*
|
|
|
|
* Should the accounting and control be hierarchical, per subtree?
|
|
|
|
*/
|
|
|
|
bool use_hierarchy;
|
2010-03-10 23:22:39 +00:00
|
|
|
atomic_t oom_lock;
|
2009-01-08 02:08:00 +00:00
|
|
|
atomic_t refcnt;
|
2009-01-08 02:08:18 +00:00
|
|
|
|
2009-01-08 02:08:24 +00:00
|
|
|
unsigned int swappiness;
|
2010-05-26 21:42:37 +00:00
|
|
|
/* OOM-Killer disable */
|
|
|
|
int oom_kill_disable;
|
2009-01-08 02:08:24 +00:00
|
|
|
|
2009-06-17 23:27:19 +00:00
|
|
|
/* set when res.limit == memsw.limit */
|
|
|
|
bool memsw_is_minimum;
|
|
|
|
|
2010-03-10 23:22:24 +00:00
|
|
|
/* protect arrays of thresholds */
|
|
|
|
struct mutex thresholds_lock;
|
|
|
|
|
|
|
|
/* thresholds for memory usage. RCU-protected */
|
2010-05-26 21:42:47 +00:00
|
|
|
struct mem_cgroup_thresholds thresholds;
|
2010-05-26 21:42:46 +00:00
|
|
|
|
2010-03-10 23:22:24 +00:00
|
|
|
/* thresholds for mem+swap usage. RCU-protected */
|
2010-05-26 21:42:47 +00:00
|
|
|
struct mem_cgroup_thresholds memsw_thresholds;
|
2010-05-26 21:42:46 +00:00
|
|
|
|
2010-05-26 21:42:36 +00:00
|
|
|
/* For oom notifier event fd */
|
|
|
|
struct list_head oom_notify;
|
|
|
|
|
2010-03-10 23:22:13 +00:00
|
|
|
/*
|
|
|
|
* Should we move charges of a task when a task is moved into this
|
|
|
|
* mem_cgroup ? And what type of charges should we move ?
|
|
|
|
*/
|
|
|
|
unsigned long move_charge_at_immigrate;
|
2008-02-07 08:14:24 +00:00
|
|
|
/*
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
* percpu counter.
|
2008-02-07 08:14:24 +00:00
|
|
|
*/
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
struct mem_cgroup_stat_cpu *stat;
|
2008-02-07 08:13:50 +00:00
|
|
|
};
|
|
|
|
|
2010-03-10 23:22:13 +00:00
|
|
|
/* Stuffs for move charges at task migration. */
|
|
|
|
/*
|
|
|
|
* Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
|
|
|
|
* left-shifted bitmap of these types.
|
|
|
|
*/
|
|
|
|
enum move_type {
|
2010-03-10 23:22:14 +00:00
|
|
|
MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */
|
2010-05-26 21:42:39 +00:00
|
|
|
MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */
|
2010-03-10 23:22:13 +00:00
|
|
|
NR_MOVE_TYPE,
|
|
|
|
};
|
|
|
|
|
2010-03-10 23:22:14 +00:00
|
|
|
/* "mc" and its members are protected by cgroup_mutex */
|
|
|
|
static struct move_charge_struct {
|
2010-08-11 01:02:58 +00:00
|
|
|
spinlock_t lock; /* for from, to, moving_task */
|
2010-03-10 23:22:14 +00:00
|
|
|
struct mem_cgroup *from;
|
|
|
|
struct mem_cgroup *to;
|
|
|
|
unsigned long precharge;
|
2010-03-10 23:22:15 +00:00
|
|
|
unsigned long moved_charge;
|
2010-03-10 23:22:18 +00:00
|
|
|
unsigned long moved_swap;
|
2010-03-10 23:22:16 +00:00
|
|
|
struct task_struct *moving_task; /* a task moving charges */
|
|
|
|
wait_queue_head_t waitq; /* a waitq for other context */
|
|
|
|
} mc = {
|
2010-08-11 01:02:58 +00:00
|
|
|
.lock = __SPIN_LOCK_UNLOCKED(mc.lock),
|
2010-03-10 23:22:16 +00:00
|
|
|
.waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
|
|
|
|
};
|
2010-03-10 23:22:14 +00:00
|
|
|
|
2010-05-26 21:42:38 +00:00
|
|
|
static bool move_anon(void)
|
|
|
|
{
|
|
|
|
return test_bit(MOVE_CHARGE_TYPE_ANON,
|
|
|
|
&mc.to->move_charge_at_immigrate);
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:39 +00:00
|
|
|
static bool move_file(void)
|
|
|
|
{
|
|
|
|
return test_bit(MOVE_CHARGE_TYPE_FILE,
|
|
|
|
&mc.to->move_charge_at_immigrate);
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:39 +00:00
|
|
|
/*
|
|
|
|
* Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
|
|
|
|
* limit reclaim to prevent infinite loops, if they ever occur.
|
|
|
|
*/
|
|
|
|
#define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
|
|
|
|
#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
|
|
|
|
|
2008-02-07 08:14:17 +00:00
|
|
|
enum charge_type {
|
|
|
|
MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
|
|
|
|
MEM_CGROUP_CHARGE_TYPE_MAPPED,
|
2008-10-19 03:26:32 +00:00
|
|
|
MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */
|
2008-10-19 03:28:11 +00:00
|
|
|
MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */
|
2009-01-08 02:07:56 +00:00
|
|
|
MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
|
2009-06-17 23:27:17 +00:00
|
|
|
MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */
|
2008-10-19 03:28:11 +00:00
|
|
|
NR_CHARGE_TYPE,
|
|
|
|
};
|
|
|
|
|
2008-10-19 03:28:16 +00:00
|
|
|
/* only for here (for easy reading.) */
|
|
|
|
#define PCGF_CACHE (1UL << PCG_CACHE)
|
|
|
|
#define PCGF_USED (1UL << PCG_USED)
|
|
|
|
#define PCGF_LOCK (1UL << PCG_LOCK)
|
2009-09-23 22:56:32 +00:00
|
|
|
/* Not used, but added here for completeness */
|
|
|
|
#define PCGF_ACCT (1UL << PCG_ACCT)
|
2008-02-07 08:14:17 +00:00
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
/* for encoding cft->private value on file */
|
|
|
|
#define _MEM (0)
|
|
|
|
#define _MEMSWAP (1)
|
2010-05-26 21:42:36 +00:00
|
|
|
#define _OOM_TYPE (2)
|
2009-01-08 02:08:00 +00:00
|
|
|
#define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
|
|
|
|
#define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
|
|
|
|
#define MEMFILE_ATTR(val) ((val) & 0xffff)
|
2010-05-26 21:42:36 +00:00
|
|
|
/* Used for OOM nofiier */
|
|
|
|
#define OOM_CONTROL (0)
|
2009-01-08 02:08:00 +00:00
|
|
|
|
2009-09-23 22:56:38 +00:00
|
|
|
/*
|
|
|
|
* Reclaim flags for mem_cgroup_hierarchical_reclaim
|
|
|
|
*/
|
|
|
|
#define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
|
|
|
|
#define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
|
|
|
|
#define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
|
|
|
|
#define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
|
2009-09-23 22:56:39 +00:00
|
|
|
#define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
|
|
|
|
#define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
|
2009-09-23 22:56:38 +00:00
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
static void mem_cgroup_get(struct mem_cgroup *mem);
|
|
|
|
static void mem_cgroup_put(struct mem_cgroup *mem);
|
2009-01-29 22:25:11 +00:00
|
|
|
static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
|
2009-12-16 00:47:08 +00:00
|
|
|
static void drain_all_stock_async(void);
|
2009-01-08 02:08:00 +00:00
|
|
|
|
2009-09-23 22:56:37 +00:00
|
|
|
static struct mem_cgroup_per_zone *
|
|
|
|
mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
|
|
|
|
{
|
|
|
|
return &mem->info.nodeinfo[nid]->zoneinfo[zid];
|
|
|
|
}
|
|
|
|
|
2009-12-16 11:19:59 +00:00
|
|
|
struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
return &mem->css;
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:37 +00:00
|
|
|
static struct mem_cgroup_per_zone *
|
|
|
|
page_cgroup_zoneinfo(struct page_cgroup *pc)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem = pc->mem_cgroup;
|
|
|
|
int nid = page_cgroup_nid(pc);
|
|
|
|
int zid = page_cgroup_zid(pc);
|
|
|
|
|
|
|
|
if (!mem)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return mem_cgroup_zoneinfo(mem, nid, zid);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct mem_cgroup_tree_per_zone *
|
|
|
|
soft_limit_tree_node_zone(int nid, int zid)
|
|
|
|
{
|
|
|
|
return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct mem_cgroup_tree_per_zone *
|
|
|
|
soft_limit_tree_from_page(struct page *page)
|
|
|
|
{
|
|
|
|
int nid = page_to_nid(page);
|
|
|
|
int zid = page_zonenum(page);
|
|
|
|
|
|
|
|
return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2009-09-23 22:56:39 +00:00
|
|
|
__mem_cgroup_insert_exceeded(struct mem_cgroup *mem,
|
2009-09-23 22:56:37 +00:00
|
|
|
struct mem_cgroup_per_zone *mz,
|
2009-10-01 22:44:12 +00:00
|
|
|
struct mem_cgroup_tree_per_zone *mctz,
|
|
|
|
unsigned long long new_usage_in_excess)
|
2009-09-23 22:56:37 +00:00
|
|
|
{
|
|
|
|
struct rb_node **p = &mctz->rb_root.rb_node;
|
|
|
|
struct rb_node *parent = NULL;
|
|
|
|
struct mem_cgroup_per_zone *mz_node;
|
|
|
|
|
|
|
|
if (mz->on_tree)
|
|
|
|
return;
|
|
|
|
|
2009-10-01 22:44:12 +00:00
|
|
|
mz->usage_in_excess = new_usage_in_excess;
|
|
|
|
if (!mz->usage_in_excess)
|
|
|
|
return;
|
2009-09-23 22:56:37 +00:00
|
|
|
while (*p) {
|
|
|
|
parent = *p;
|
|
|
|
mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
|
|
|
|
tree_node);
|
|
|
|
if (mz->usage_in_excess < mz_node->usage_in_excess)
|
|
|
|
p = &(*p)->rb_left;
|
|
|
|
/*
|
|
|
|
* We can't avoid mem cgroups that are over their soft
|
|
|
|
* limit by the same amount
|
|
|
|
*/
|
|
|
|
else if (mz->usage_in_excess >= mz_node->usage_in_excess)
|
|
|
|
p = &(*p)->rb_right;
|
|
|
|
}
|
|
|
|
rb_link_node(&mz->tree_node, parent, p);
|
|
|
|
rb_insert_color(&mz->tree_node, &mctz->rb_root);
|
|
|
|
mz->on_tree = true;
|
2009-09-23 22:56:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
__mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
|
|
|
|
struct mem_cgroup_per_zone *mz,
|
|
|
|
struct mem_cgroup_tree_per_zone *mctz)
|
|
|
|
{
|
|
|
|
if (!mz->on_tree)
|
|
|
|
return;
|
|
|
|
rb_erase(&mz->tree_node, &mctz->rb_root);
|
|
|
|
mz->on_tree = false;
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:37 +00:00
|
|
|
static void
|
|
|
|
mem_cgroup_remove_exceeded(struct mem_cgroup *mem,
|
|
|
|
struct mem_cgroup_per_zone *mz,
|
|
|
|
struct mem_cgroup_tree_per_zone *mctz)
|
|
|
|
{
|
|
|
|
spin_lock(&mctz->lock);
|
2009-09-23 22:56:39 +00:00
|
|
|
__mem_cgroup_remove_exceeded(mem, mz, mctz);
|
2009-09-23 22:56:37 +00:00
|
|
|
spin_unlock(&mctz->lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page)
|
|
|
|
{
|
2009-10-01 22:44:12 +00:00
|
|
|
unsigned long long excess;
|
2009-09-23 22:56:37 +00:00
|
|
|
struct mem_cgroup_per_zone *mz;
|
|
|
|
struct mem_cgroup_tree_per_zone *mctz;
|
2009-10-01 22:44:11 +00:00
|
|
|
int nid = page_to_nid(page);
|
|
|
|
int zid = page_zonenum(page);
|
2009-09-23 22:56:37 +00:00
|
|
|
mctz = soft_limit_tree_from_page(page);
|
|
|
|
|
|
|
|
/*
|
2009-10-01 22:44:11 +00:00
|
|
|
* Necessary to update all ancestors when hierarchy is used.
|
|
|
|
* because their event counter is not touched.
|
2009-09-23 22:56:37 +00:00
|
|
|
*/
|
2009-10-01 22:44:11 +00:00
|
|
|
for (; mem; mem = parent_mem_cgroup(mem)) {
|
|
|
|
mz = mem_cgroup_zoneinfo(mem, nid, zid);
|
2009-10-01 22:44:12 +00:00
|
|
|
excess = res_counter_soft_limit_excess(&mem->res);
|
2009-10-01 22:44:11 +00:00
|
|
|
/*
|
|
|
|
* We have to update the tree if mz is on RB-tree or
|
|
|
|
* mem is over its softlimit.
|
|
|
|
*/
|
2009-10-01 22:44:12 +00:00
|
|
|
if (excess || mz->on_tree) {
|
2009-10-01 22:44:11 +00:00
|
|
|
spin_lock(&mctz->lock);
|
|
|
|
/* if on-tree, remove it */
|
|
|
|
if (mz->on_tree)
|
|
|
|
__mem_cgroup_remove_exceeded(mem, mz, mctz);
|
|
|
|
/*
|
2009-10-01 22:44:12 +00:00
|
|
|
* Insert again. mz->usage_in_excess will be updated.
|
|
|
|
* If excess is 0, no tree ops.
|
2009-10-01 22:44:11 +00:00
|
|
|
*/
|
2009-10-01 22:44:12 +00:00
|
|
|
__mem_cgroup_insert_exceeded(mem, mz, mctz, excess);
|
2009-10-01 22:44:11 +00:00
|
|
|
spin_unlock(&mctz->lock);
|
|
|
|
}
|
2009-09-23 22:56:37 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
int node, zone;
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
|
|
|
struct mem_cgroup_tree_per_zone *mctz;
|
|
|
|
|
|
|
|
for_each_node_state(node, N_POSSIBLE) {
|
|
|
|
for (zone = 0; zone < MAX_NR_ZONES; zone++) {
|
|
|
|
mz = mem_cgroup_zoneinfo(mem, node, zone);
|
|
|
|
mctz = soft_limit_tree_node_zone(node, zone);
|
|
|
|
mem_cgroup_remove_exceeded(mem, mz, mctz);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:39 +00:00
|
|
|
static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct mem_cgroup_per_zone *
|
|
|
|
__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
|
|
|
|
{
|
|
|
|
struct rb_node *rightmost = NULL;
|
2009-10-01 22:44:08 +00:00
|
|
|
struct mem_cgroup_per_zone *mz;
|
2009-09-23 22:56:39 +00:00
|
|
|
|
|
|
|
retry:
|
2009-10-01 22:44:08 +00:00
|
|
|
mz = NULL;
|
2009-09-23 22:56:39 +00:00
|
|
|
rightmost = rb_last(&mctz->rb_root);
|
|
|
|
if (!rightmost)
|
|
|
|
goto done; /* Nothing to reclaim from */
|
|
|
|
|
|
|
|
mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
|
|
|
|
/*
|
|
|
|
* Remove the node now but someone else can add it back,
|
|
|
|
* we will to add it back at the end of reclaim to its correct
|
|
|
|
* position in the tree.
|
|
|
|
*/
|
|
|
|
__mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
|
|
|
|
if (!res_counter_soft_limit_excess(&mz->mem->res) ||
|
|
|
|
!css_tryget(&mz->mem->css))
|
|
|
|
goto retry;
|
|
|
|
done:
|
|
|
|
return mz;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct mem_cgroup_per_zone *
|
|
|
|
mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
|
|
|
|
|
|
|
spin_lock(&mctz->lock);
|
|
|
|
mz = __mem_cgroup_largest_soft_limit_node(mctz);
|
|
|
|
spin_unlock(&mctz->lock);
|
|
|
|
return mz;
|
|
|
|
}
|
|
|
|
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
|
|
|
|
enum mem_cgroup_stat_index idx)
|
|
|
|
{
|
|
|
|
int cpu;
|
|
|
|
s64 val = 0;
|
|
|
|
|
|
|
|
for_each_possible_cpu(cpu)
|
|
|
|
val += per_cpu(mem->stat->count[idx], cpu);
|
|
|
|
return val;
|
|
|
|
}
|
|
|
|
|
|
|
|
static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
s64 ret;
|
|
|
|
|
|
|
|
ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
|
|
|
|
ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:42 +00:00
|
|
|
static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
|
|
|
|
bool charge)
|
|
|
|
{
|
|
|
|
int val = (charge) ? 1 : -1;
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val);
|
2009-09-23 22:56:42 +00:00
|
|
|
}
|
|
|
|
|
2008-10-19 03:28:11 +00:00
|
|
|
static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
|
|
|
|
struct page_cgroup *pc,
|
|
|
|
bool charge)
|
2008-02-07 08:14:24 +00:00
|
|
|
{
|
2009-09-23 22:56:42 +00:00
|
|
|
int val = (charge) ? 1 : -1;
|
2008-02-07 08:14:24 +00:00
|
|
|
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
preempt_disable();
|
|
|
|
|
2008-10-19 03:28:11 +00:00
|
|
|
if (PageCgroupCache(pc))
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
__this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val);
|
2008-02-07 08:14:24 +00:00
|
|
|
else
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
__this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val);
|
2008-05-01 11:35:12 +00:00
|
|
|
|
|
|
|
if (charge)
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
__this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
|
2008-05-01 11:35:12 +00:00
|
|
|
else
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
__this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
|
2010-03-10 23:22:31 +00:00
|
|
|
__this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]);
|
2010-03-10 23:22:24 +00:00
|
|
|
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
preempt_enable();
|
2008-02-07 08:14:31 +00:00
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
|
2008-10-19 03:26:14 +00:00
|
|
|
enum lru_list idx)
|
2008-02-07 08:14:31 +00:00
|
|
|
{
|
|
|
|
int nid, zid;
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
|
|
|
u64 total = 0;
|
|
|
|
|
|
|
|
for_each_online_node(nid)
|
|
|
|
for (zid = 0; zid < MAX_NR_ZONES; zid++) {
|
|
|
|
mz = mem_cgroup_zoneinfo(mem, nid, zid);
|
|
|
|
total += MEM_CGROUP_ZSTAT(mz, idx);
|
|
|
|
}
|
|
|
|
return total;
|
2008-02-07 08:14:24 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:31 +00:00
|
|
|
static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
|
|
|
|
{
|
|
|
|
s64 val;
|
|
|
|
|
|
|
|
val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]);
|
|
|
|
|
|
|
|
return !(val & ((1 << event_mask_shift) - 1));
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check events in order.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
|
|
|
|
{
|
|
|
|
/* threshold event is triggered in finer grain than soft limit */
|
|
|
|
if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) {
|
|
|
|
mem_cgroup_threshold(mem);
|
|
|
|
if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH)))
|
|
|
|
mem_cgroup_update_tree(mem, page);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-03-04 22:29:10 +00:00
|
|
|
static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
|
2008-02-07 08:13:50 +00:00
|
|
|
{
|
|
|
|
return container_of(cgroup_subsys_state(cont,
|
|
|
|
mem_cgroup_subsys_id), struct mem_cgroup,
|
|
|
|
css);
|
|
|
|
}
|
|
|
|
|
cgroups: add an owner to the mm_struct
Remove the mem_cgroup member from mm_struct and instead adds an owner.
This approach was suggested by Paul Menage. The advantage of this approach
is that, once the mm->owner is known, using the subsystem id, the cgroup
can be determined. It also allows several control groups that are
virtually grouped by mm_struct, to exist independent of the memory
controller i.e., without adding mem_cgroup's for each controller, to
mm_struct.
A new config option CONFIG_MM_OWNER is added and the memory resource
controller selects this config option.
This patch also adds cgroup callbacks to notify subsystems when mm->owner
changes. The mm_cgroup_changed callback is called with the task_lock() of
the new task held and is called just prior to changing the mm->owner.
I am indebted to Paul Menage for the several reviews of this patchset and
helping me make it lighter and simpler.
This patch was tested on a powerpc box, it was compiled with both the
MM_OWNER config turned on and off.
After the thread group leader exits, it's moved to init_css_state by
cgroup_exit(), thus all future charges from runnings threads would be
redirected to the init_css_set's subsystem.
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Sudhir Kumar <skumar@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Hirokazu Takahashi <taka@valinux.co.jp>
Cc: David Rientjes <rientjes@google.com>,
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Reviewed-by: Paul Menage <menage@google.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-29 08:00:16 +00:00
|
|
|
struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
|
2008-02-07 08:13:51 +00:00
|
|
|
{
|
mm owner: fix race between swapoff and exit
There's a race between mm->owner assignment and swapoff, more easily
seen when task slab poisoning is turned on. The condition occurs when
try_to_unuse() runs in parallel with an exiting task. A similar race
can occur with callers of get_task_mm(), such as /proc/<pid>/<mmstats>
or ptrace or page migration.
CPU0 CPU1
try_to_unuse
looks at mm = task0->mm
increments mm->mm_users
task 0 exits
mm->owner needs to be updated, but no
new owner is found (mm_users > 1, but
no other task has task->mm = task0->mm)
mm_update_next_owner() leaves
mmput(mm) decrements mm->mm_users
task0 freed
dereferencing mm->owner fails
The fix is to notify the subsystem via mm_owner_changed callback(),
if no new owner is found, by specifying the new task as NULL.
Jiri Slaby:
mm->owner was set to NULL prior to calling cgroup_mm_owner_callbacks(), but
must be set after that, so as not to pass NULL as old owner causing oops.
Daisuke Nishimura:
mm_update_next_owner() may set mm->owner to NULL, but mem_cgroup_from_task()
and its callers need to take account of this situation to avoid oops.
Hugh Dickins:
Lockdep warning and hang below exec_mmap() when testing these patches.
exit_mm() up_reads mmap_sem before calling mm_update_next_owner(),
so exec_mmap() now needs to do the same. And with that repositioning,
there's now no point in mm_need_new_owner() allowing for NULL mm.
Reported-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-28 22:09:31 +00:00
|
|
|
/*
|
|
|
|
* mm_update_next_owner() may clear mm->owner to NULL
|
|
|
|
* if it races with swapoff, page migration, etc.
|
|
|
|
* So this can be called with p == NULL.
|
|
|
|
*/
|
|
|
|
if (unlikely(!p))
|
|
|
|
return NULL;
|
|
|
|
|
2008-02-07 08:13:51 +00:00
|
|
|
return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
|
|
|
|
struct mem_cgroup, css);
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:33 +00:00
|
|
|
static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem = NULL;
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
|
|
|
|
if (!mm)
|
|
|
|
return NULL;
|
2009-01-08 02:08:33 +00:00
|
|
|
/*
|
|
|
|
* Because we have no locks, mm->owner's may be being moved to other
|
|
|
|
* cgroup. We use css_tryget() here even if this looks
|
|
|
|
* pessimistic (rather than adding locks here).
|
|
|
|
*/
|
|
|
|
rcu_read_lock();
|
|
|
|
do {
|
|
|
|
mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
|
|
|
|
if (unlikely(!mem))
|
|
|
|
break;
|
|
|
|
} while (!css_tryget(&mem->css));
|
|
|
|
rcu_read_unlock();
|
|
|
|
return mem;
|
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
/*
|
|
|
|
* Call callback function against all cgroup under hierarchy tree.
|
|
|
|
*/
|
|
|
|
static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data,
|
|
|
|
int (*func)(struct mem_cgroup *, void *))
|
|
|
|
{
|
|
|
|
int found, ret, nextid;
|
|
|
|
struct cgroup_subsys_state *css;
|
|
|
|
struct mem_cgroup *mem;
|
|
|
|
|
|
|
|
if (!root->use_hierarchy)
|
|
|
|
return (*func)(root, data);
|
|
|
|
|
|
|
|
nextid = 1;
|
|
|
|
do {
|
|
|
|
ret = 0;
|
|
|
|
mem = NULL;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
css = css_get_next(&mem_cgroup_subsys, nextid, &root->css,
|
|
|
|
&found);
|
|
|
|
if (css && css_tryget(css))
|
|
|
|
mem = container_of(css, struct mem_cgroup, css);
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
if (mem) {
|
|
|
|
ret = (*func)(mem, data);
|
|
|
|
css_put(&mem->css);
|
|
|
|
}
|
|
|
|
nextid = found + 1;
|
|
|
|
} while (!ret && css);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:32 +00:00
|
|
|
static inline bool mem_cgroup_is_root(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
return (mem == root_mem_cgroup);
|
|
|
|
}
|
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
/*
|
|
|
|
* Following LRU functions are allowed to be used without PCG_LOCK.
|
|
|
|
* Operations are called by routine of global LRU independently from memcg.
|
|
|
|
* What we have to take care of here is validness of pc->mem_cgroup.
|
|
|
|
*
|
|
|
|
* Changes to pc->mem_cgroup happens when
|
|
|
|
* 1. charge
|
|
|
|
* 2. moving account
|
|
|
|
* In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
|
|
|
|
* It is added to LRU before charge.
|
|
|
|
* If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
|
|
|
|
* When moving account, the page is not on LRU. It's isolated.
|
|
|
|
*/
|
2008-10-19 03:26:32 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
|
|
|
|
{
|
|
|
|
struct page_cgroup *pc;
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
2008-02-07 08:14:31 +00:00
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
return;
|
|
|
|
pc = lookup_page_cgroup(page);
|
|
|
|
/* can happen while we handle swapcache. */
|
2009-09-23 22:56:32 +00:00
|
|
|
if (!TestClearPageCgroupAcctLRU(pc))
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
return;
|
2009-09-23 22:56:32 +00:00
|
|
|
VM_BUG_ON(!pc->mem_cgroup);
|
2009-01-08 02:08:34 +00:00
|
|
|
/*
|
|
|
|
* We don't check PCG_USED bit. It's cleared when the "page" is finally
|
|
|
|
* removed from global LRU.
|
|
|
|
*/
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
mz = page_cgroup_zoneinfo(pc);
|
2008-10-19 03:26:14 +00:00
|
|
|
MEM_CGROUP_ZSTAT(mz, lru) -= 1;
|
2009-09-23 22:56:32 +00:00
|
|
|
if (mem_cgroup_is_root(pc->mem_cgroup))
|
|
|
|
return;
|
|
|
|
VM_BUG_ON(list_empty(&pc->lru));
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
list_del_init(&pc->lru);
|
|
|
|
return;
|
2008-02-07 08:14:31 +00:00
|
|
|
}
|
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
void mem_cgroup_del_lru(struct page *page)
|
2008-02-07 08:14:31 +00:00
|
|
|
{
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
mem_cgroup_del_lru_list(page, page_lru(page));
|
|
|
|
}
|
2008-10-19 03:26:14 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
|
|
|
struct page_cgroup *pc;
|
2008-10-19 03:26:14 +00:00
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
return;
|
2008-02-07 08:14:31 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
pc = lookup_page_cgroup(page);
|
2009-01-15 21:51:11 +00:00
|
|
|
/*
|
|
|
|
* Used bit is set without atomic ops but after smp_wmb().
|
|
|
|
* For making pc->mem_cgroup visible, insert smp_rmb() here.
|
|
|
|
*/
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
smp_rmb();
|
2009-09-23 22:56:32 +00:00
|
|
|
/* unused or root page is not rotated. */
|
|
|
|
if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
return;
|
|
|
|
mz = page_cgroup_zoneinfo(pc);
|
|
|
|
list_move(&pc->lru, &mz->lists[lru]);
|
2008-02-07 08:14:31 +00:00
|
|
|
}
|
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
|
2008-02-07 08:13:56 +00:00
|
|
|
{
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
struct page_cgroup *pc;
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
2008-02-07 08:14:31 +00:00
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
return;
|
|
|
|
pc = lookup_page_cgroup(page);
|
2009-09-23 22:56:32 +00:00
|
|
|
VM_BUG_ON(PageCgroupAcctLRU(pc));
|
2009-01-15 21:51:11 +00:00
|
|
|
/*
|
|
|
|
* Used bit is set without atomic ops but after smp_wmb().
|
|
|
|
* For making pc->mem_cgroup visible, insert smp_rmb() here.
|
|
|
|
*/
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
smp_rmb();
|
|
|
|
if (!PageCgroupUsed(pc))
|
Unevictable LRU Infrastructure
When the system contains lots of mlocked or otherwise unevictable pages,
the pageout code (kswapd) can spend lots of time scanning over these
pages. Worse still, the presence of lots of unevictable pages can confuse
kswapd into thinking that more aggressive pageout modes are required,
resulting in all kinds of bad behaviour.
Infrastructure to manage pages excluded from reclaim--i.e., hidden from
vmscan. Based on a patch by Larry Woodman of Red Hat. Reworked to
maintain "unevictable" pages on a separate per-zone LRU list, to "hide"
them from vmscan.
Kosaki Motohiro added the support for the memory controller unevictable
lru list.
Pages on the unevictable list have both PG_unevictable and PG_lru set.
Thus, PG_unevictable is analogous to and mutually exclusive with
PG_active--it specifies which LRU list the page is on.
The unevictable infrastructure is enabled by a new mm Kconfig option
[CONFIG_]UNEVICTABLE_LRU.
A new function 'page_evictable(page, vma)' in vmscan.c tests whether or
not a page may be evictable. Subsequent patches will add the various
!evictable tests. We'll want to keep these tests light-weight for use in
shrink_active_list() and, possibly, the fault path.
To avoid races between tasks putting pages [back] onto an LRU list and
tasks that might be moving the page from non-evictable to evictable state,
the new function 'putback_lru_page()' -- inverse to 'isolate_lru_page()'
-- tests the "evictability" of a page after placing it on the LRU, before
dropping the reference. If the page has become unevictable,
putback_lru_page() will redo the 'putback', thus moving the page to the
unevictable list. This way, we avoid "stranding" evictable pages on the
unevictable list.
[akpm@linux-foundation.org: fix fallout from out-of-order merge]
[riel@redhat.com: fix UNEVICTABLE_LRU and !PROC_PAGE_MONITOR build]
[nishimura@mxp.nes.nec.co.jp: remove redundant mapping check]
[kosaki.motohiro@jp.fujitsu.com: unevictable-lru-infrastructure: putback_lru_page()/unevictable page handling rework]
[kosaki.motohiro@jp.fujitsu.com: kill unnecessary lock_page() in vmscan.c]
[kosaki.motohiro@jp.fujitsu.com: revert migration change of unevictable lru infrastructure]
[kosaki.motohiro@jp.fujitsu.com: revert to unevictable-lru-infrastructure-kconfig-fix.patch]
[kosaki.motohiro@jp.fujitsu.com: restore patch failure of vmstat-unevictable-and-mlocked-pages-vm-events.patch]
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Debugged-by: Benjamin Kidwell <benjkidwell@yahoo.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 03:26:39 +00:00
|
|
|
return;
|
2008-10-19 03:26:14 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
mz = page_cgroup_zoneinfo(pc);
|
2008-10-19 03:26:14 +00:00
|
|
|
MEM_CGROUP_ZSTAT(mz, lru) += 1;
|
2009-09-23 22:56:32 +00:00
|
|
|
SetPageCgroupAcctLRU(pc);
|
|
|
|
if (mem_cgroup_is_root(pc->mem_cgroup))
|
|
|
|
return;
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
list_add(&pc->lru, &mz->lists[lru]);
|
|
|
|
}
|
2009-01-08 02:08:34 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
/*
|
2009-01-08 02:08:34 +00:00
|
|
|
* At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
|
|
|
|
* lru because the page may.be reused after it's fully uncharged (because of
|
|
|
|
* SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
|
|
|
|
* it again. This function is only used to charge SwapCache. It's done under
|
|
|
|
* lock_page and expected that zone->lru_lock is never held.
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
*/
|
2009-01-08 02:08:34 +00:00
|
|
|
static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page)
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
{
|
2009-01-08 02:08:34 +00:00
|
|
|
unsigned long flags;
|
|
|
|
struct zone *zone = page_zone(page);
|
|
|
|
struct page_cgroup *pc = lookup_page_cgroup(page);
|
|
|
|
|
|
|
|
spin_lock_irqsave(&zone->lru_lock, flags);
|
|
|
|
/*
|
|
|
|
* Forget old LRU when this page_cgroup is *not* used. This Used bit
|
|
|
|
* is guarded by lock_page() because the page is SwapCache.
|
|
|
|
*/
|
|
|
|
if (!PageCgroupUsed(pc))
|
|
|
|
mem_cgroup_del_lru_list(page, page_lru(page));
|
|
|
|
spin_unlock_irqrestore(&zone->lru_lock, flags);
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:34 +00:00
|
|
|
static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
struct zone *zone = page_zone(page);
|
|
|
|
struct page_cgroup *pc = lookup_page_cgroup(page);
|
|
|
|
|
|
|
|
spin_lock_irqsave(&zone->lru_lock, flags);
|
|
|
|
/* link when the page is linked to LRU but page_cgroup isn't */
|
2009-09-23 22:56:32 +00:00
|
|
|
if (PageLRU(page) && !PageCgroupAcctLRU(pc))
|
2009-01-08 02:08:34 +00:00
|
|
|
mem_cgroup_add_lru_list(page, page_lru(page));
|
|
|
|
spin_unlock_irqrestore(&zone->lru_lock, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
void mem_cgroup_move_lists(struct page *page,
|
|
|
|
enum lru_list from, enum lru_list to)
|
|
|
|
{
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
return;
|
|
|
|
mem_cgroup_del_lru_list(page, from);
|
|
|
|
mem_cgroup_add_lru_list(page, to);
|
2008-02-07 08:13:56 +00:00
|
|
|
}
|
|
|
|
|
2008-02-07 08:14:06 +00:00
|
|
|
int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
int ret;
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
struct mem_cgroup *curr = NULL;
|
2010-08-11 01:03:00 +00:00
|
|
|
struct task_struct *p;
|
2008-02-07 08:14:06 +00:00
|
|
|
|
2010-08-11 01:03:00 +00:00
|
|
|
p = find_lock_task_mm(task);
|
|
|
|
if (!p)
|
|
|
|
return 0;
|
|
|
|
curr = try_get_mem_cgroup_from_mm(p->mm);
|
|
|
|
task_unlock(p);
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
if (!curr)
|
|
|
|
return 0;
|
2009-12-16 00:47:12 +00:00
|
|
|
/*
|
|
|
|
* We should check use_hierarchy of "mem" not "curr". Because checking
|
|
|
|
* use_hierarchy of "curr" here make this function true if hierarchy is
|
|
|
|
* enabled in "curr" and "curr" is a child of "mem" in *cgroup*
|
|
|
|
* hierarchy(even if use_hierarchy is disabled in "mem").
|
|
|
|
*/
|
|
|
|
if (mem->use_hierarchy)
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
ret = css_is_ancestor(&curr->css, &mem->css);
|
|
|
|
else
|
|
|
|
ret = (curr == mem);
|
|
|
|
css_put(&curr->css);
|
2008-02-07 08:14:06 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:25 +00:00
|
|
|
static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages)
|
2009-01-08 02:08:18 +00:00
|
|
|
{
|
|
|
|
unsigned long active;
|
|
|
|
unsigned long inactive;
|
2009-01-08 02:08:25 +00:00
|
|
|
unsigned long gb;
|
|
|
|
unsigned long inactive_ratio;
|
2009-01-08 02:08:18 +00:00
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON);
|
|
|
|
active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON);
|
2009-01-08 02:08:18 +00:00
|
|
|
|
2009-01-08 02:08:25 +00:00
|
|
|
gb = (inactive + active) >> (30 - PAGE_SHIFT);
|
|
|
|
if (gb)
|
|
|
|
inactive_ratio = int_sqrt(10 * gb);
|
|
|
|
else
|
|
|
|
inactive_ratio = 1;
|
|
|
|
|
|
|
|
if (present_pages) {
|
|
|
|
present_pages[0] = inactive;
|
|
|
|
present_pages[1] = active;
|
|
|
|
}
|
|
|
|
|
|
|
|
return inactive_ratio;
|
|
|
|
}
|
|
|
|
|
|
|
|
int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg)
|
|
|
|
{
|
|
|
|
unsigned long active;
|
|
|
|
unsigned long inactive;
|
|
|
|
unsigned long present_pages[2];
|
|
|
|
unsigned long inactive_ratio;
|
|
|
|
|
|
|
|
inactive_ratio = calc_inactive_ratio(memcg, present_pages);
|
|
|
|
|
|
|
|
inactive = present_pages[0];
|
|
|
|
active = present_pages[1];
|
|
|
|
|
|
|
|
if (inactive * inactive_ratio < active)
|
2009-01-08 02:08:18 +00:00
|
|
|
return 1;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-06-16 22:32:28 +00:00
|
|
|
int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg)
|
|
|
|
{
|
|
|
|
unsigned long active;
|
|
|
|
unsigned long inactive;
|
|
|
|
|
|
|
|
inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE);
|
|
|
|
active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE);
|
|
|
|
|
|
|
|
return (active > inactive);
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:19 +00:00
|
|
|
unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg,
|
|
|
|
struct zone *zone,
|
|
|
|
enum lru_list lru)
|
|
|
|
{
|
2010-08-11 01:03:06 +00:00
|
|
|
int nid = zone_to_nid(zone);
|
2009-01-08 02:08:19 +00:00
|
|
|
int zid = zone_idx(zone);
|
|
|
|
struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
|
|
|
|
|
|
|
|
return MEM_CGROUP_ZSTAT(mz, lru);
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:20 +00:00
|
|
|
struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg,
|
|
|
|
struct zone *zone)
|
|
|
|
{
|
2010-08-11 01:03:06 +00:00
|
|
|
int nid = zone_to_nid(zone);
|
2009-01-08 02:08:20 +00:00
|
|
|
int zid = zone_idx(zone);
|
|
|
|
struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid);
|
|
|
|
|
|
|
|
return &mz->reclaim_stat;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct zone_reclaim_stat *
|
|
|
|
mem_cgroup_get_reclaim_stat_from_page(struct page *page)
|
|
|
|
{
|
|
|
|
struct page_cgroup *pc;
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
|
|
|
|
|
|
|
if (mem_cgroup_disabled())
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
pc = lookup_page_cgroup(page);
|
2009-01-15 21:51:11 +00:00
|
|
|
/*
|
|
|
|
* Used bit is set without atomic ops but after smp_wmb().
|
|
|
|
* For making pc->mem_cgroup visible, insert smp_rmb() here.
|
|
|
|
*/
|
|
|
|
smp_rmb();
|
|
|
|
if (!PageCgroupUsed(pc))
|
|
|
|
return NULL;
|
|
|
|
|
2009-01-08 02:08:20 +00:00
|
|
|
mz = page_cgroup_zoneinfo(pc);
|
|
|
|
if (!mz)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return &mz->reclaim_stat;
|
|
|
|
}
|
|
|
|
|
2008-02-07 08:13:56 +00:00
|
|
|
unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
|
|
|
|
struct list_head *dst,
|
|
|
|
unsigned long *scanned, int order,
|
|
|
|
int mode, struct zone *z,
|
|
|
|
struct mem_cgroup *mem_cont,
|
2008-10-19 03:26:32 +00:00
|
|
|
int active, int file)
|
2008-02-07 08:13:56 +00:00
|
|
|
{
|
|
|
|
unsigned long nr_taken = 0;
|
|
|
|
struct page *page;
|
|
|
|
unsigned long scan;
|
|
|
|
LIST_HEAD(pc_list);
|
|
|
|
struct list_head *src;
|
2008-02-07 08:14:11 +00:00
|
|
|
struct page_cgroup *pc, *tmp;
|
2010-08-11 01:03:06 +00:00
|
|
|
int nid = zone_to_nid(z);
|
2008-02-07 08:14:38 +00:00
|
|
|
int zid = zone_idx(z);
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
2009-09-22 00:02:56 +00:00
|
|
|
int lru = LRU_FILE * file + active;
|
2009-06-17 23:27:21 +00:00
|
|
|
int ret;
|
2008-02-07 08:13:56 +00:00
|
|
|
|
cgroups: add an owner to the mm_struct
Remove the mem_cgroup member from mm_struct and instead adds an owner.
This approach was suggested by Paul Menage. The advantage of this approach
is that, once the mm->owner is known, using the subsystem id, the cgroup
can be determined. It also allows several control groups that are
virtually grouped by mm_struct, to exist independent of the memory
controller i.e., without adding mem_cgroup's for each controller, to
mm_struct.
A new config option CONFIG_MM_OWNER is added and the memory resource
controller selects this config option.
This patch also adds cgroup callbacks to notify subsystems when mm->owner
changes. The mm_cgroup_changed callback is called with the task_lock() of
the new task held and is called just prior to changing the mm->owner.
I am indebted to Paul Menage for the several reviews of this patchset and
helping me make it lighter and simpler.
This patch was tested on a powerpc box, it was compiled with both the
MM_OWNER config turned on and off.
After the thread group leader exits, it's moved to init_css_state by
cgroup_exit(), thus all future charges from runnings threads would be
redirected to the init_css_set's subsystem.
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Sudhir Kumar <skumar@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Hirokazu Takahashi <taka@valinux.co.jp>
Cc: David Rientjes <rientjes@google.com>,
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Reviewed-by: Paul Menage <menage@google.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-29 08:00:16 +00:00
|
|
|
BUG_ON(!mem_cont);
|
2008-02-07 08:14:38 +00:00
|
|
|
mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
|
2008-10-19 03:26:14 +00:00
|
|
|
src = &mz->lists[lru];
|
2008-02-07 08:13:56 +00:00
|
|
|
|
2008-02-07 08:14:11 +00:00
|
|
|
scan = 0;
|
|
|
|
list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
|
2008-02-07 08:14:12 +00:00
|
|
|
if (scan >= nr_to_scan)
|
2008-02-07 08:14:11 +00:00
|
|
|
break;
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
|
|
|
|
page = pc->page;
|
2008-10-19 03:28:16 +00:00
|
|
|
if (unlikely(!PageCgroupUsed(pc)))
|
|
|
|
continue;
|
2008-02-07 08:14:12 +00:00
|
|
|
if (unlikely(!PageLRU(page)))
|
2008-02-07 08:14:11 +00:00
|
|
|
continue;
|
|
|
|
|
2008-02-07 08:14:12 +00:00
|
|
|
scan++;
|
2009-06-17 23:27:21 +00:00
|
|
|
ret = __isolate_lru_page(page, mode, file);
|
|
|
|
switch (ret) {
|
|
|
|
case 0:
|
2008-02-07 08:13:56 +00:00
|
|
|
list_move(&page->lru, dst);
|
2009-06-17 23:27:21 +00:00
|
|
|
mem_cgroup_del_lru(page);
|
2008-02-07 08:13:56 +00:00
|
|
|
nr_taken++;
|
2009-06-17 23:27:21 +00:00
|
|
|
break;
|
|
|
|
case -EBUSY:
|
|
|
|
/* we don't affect global LRU but rotate in our LRU */
|
|
|
|
mem_cgroup_rotate_lru_list(page, page_lru(page));
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
2008-02-07 08:13:56 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
*scanned = scan;
|
2010-08-10 00:19:57 +00:00
|
|
|
|
|
|
|
trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken,
|
|
|
|
0, 0, 0, mode);
|
|
|
|
|
2008-02-07 08:13:56 +00:00
|
|
|
return nr_taken;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:06 +00:00
|
|
|
#define mem_cgroup_from_res_counter(counter, member) \
|
|
|
|
container_of(counter, struct mem_cgroup, member)
|
|
|
|
|
2009-01-08 02:08:12 +00:00
|
|
|
static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
if (do_swap_account) {
|
|
|
|
if (res_counter_check_under_limit(&mem->res) &&
|
|
|
|
res_counter_check_under_limit(&mem->memsw))
|
|
|
|
return true;
|
|
|
|
} else
|
|
|
|
if (res_counter_check_under_limit(&mem->res))
|
|
|
|
return true;
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:24 +00:00
|
|
|
static unsigned int get_swappiness(struct mem_cgroup *memcg)
|
|
|
|
{
|
|
|
|
struct cgroup *cgrp = memcg->css.cgroup;
|
|
|
|
unsigned int swappiness;
|
|
|
|
|
|
|
|
/* root ? */
|
|
|
|
if (cgrp->parent == NULL)
|
|
|
|
return vm_swappiness;
|
|
|
|
|
|
|
|
spin_lock(&memcg->reclaim_param_lock);
|
|
|
|
swappiness = memcg->swappiness;
|
|
|
|
spin_unlock(&memcg->reclaim_param_lock);
|
|
|
|
|
|
|
|
return swappiness;
|
|
|
|
}
|
|
|
|
|
2010-08-11 01:02:57 +00:00
|
|
|
/* A routine for testing mem is not under move_account */
|
|
|
|
|
|
|
|
static bool mem_cgroup_under_move(struct mem_cgroup *mem)
|
|
|
|
{
|
2010-08-11 01:02:58 +00:00
|
|
|
struct mem_cgroup *from;
|
|
|
|
struct mem_cgroup *to;
|
2010-08-11 01:02:57 +00:00
|
|
|
bool ret = false;
|
2010-08-11 01:02:58 +00:00
|
|
|
/*
|
|
|
|
* Unlike task_move routines, we access mc.to, mc.from not under
|
|
|
|
* mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
|
|
|
|
*/
|
|
|
|
spin_lock(&mc.lock);
|
|
|
|
from = mc.from;
|
|
|
|
to = mc.to;
|
|
|
|
if (!from)
|
|
|
|
goto unlock;
|
|
|
|
if (from == mem || to == mem
|
|
|
|
|| (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css))
|
|
|
|
|| (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css)))
|
|
|
|
ret = true;
|
|
|
|
unlock:
|
|
|
|
spin_unlock(&mc.lock);
|
2010-08-11 01:02:57 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
if (mc.moving_task && current != mc.moving_task) {
|
|
|
|
if (mem_cgroup_under_move(mem)) {
|
|
|
|
DEFINE_WAIT(wait);
|
|
|
|
prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
|
|
|
|
/* moving charge context might have finished. */
|
|
|
|
if (mc.moving_task)
|
|
|
|
schedule();
|
|
|
|
finish_wait(&mc.waitq, &wait);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:36 +00:00
|
|
|
static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data)
|
|
|
|
{
|
|
|
|
int *val = data;
|
|
|
|
(*val)++;
|
|
|
|
return 0;
|
|
|
|
}
|
2009-04-02 23:57:39 +00:00
|
|
|
|
|
|
|
/**
|
2010-03-10 23:22:25 +00:00
|
|
|
* mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
|
2009-04-02 23:57:39 +00:00
|
|
|
* @memcg: The memory cgroup that went over limit
|
|
|
|
* @p: Task that is going to be killed
|
|
|
|
*
|
|
|
|
* NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
|
|
|
|
* enabled
|
|
|
|
*/
|
|
|
|
void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
|
|
|
|
{
|
|
|
|
struct cgroup *task_cgrp;
|
|
|
|
struct cgroup *mem_cgrp;
|
|
|
|
/*
|
|
|
|
* Need a buffer in BSS, can't rely on allocations. The code relies
|
|
|
|
* on the assumption that OOM is serialized for memory controller.
|
|
|
|
* If this assumption is broken, revisit this code.
|
|
|
|
*/
|
|
|
|
static char memcg_name[PATH_MAX];
|
|
|
|
int ret;
|
|
|
|
|
2009-12-16 00:47:12 +00:00
|
|
|
if (!memcg || !p)
|
2009-04-02 23:57:39 +00:00
|
|
|
return;
|
|
|
|
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
|
|
|
|
mem_cgrp = memcg->css.cgroup;
|
|
|
|
task_cgrp = task_cgroup(p, mem_cgroup_subsys_id);
|
|
|
|
|
|
|
|
ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX);
|
|
|
|
if (ret < 0) {
|
|
|
|
/*
|
|
|
|
* Unfortunately, we are unable to convert to a useful name
|
|
|
|
* But we'll still print out the usage information
|
|
|
|
*/
|
|
|
|
rcu_read_unlock();
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
printk(KERN_INFO "Task in %s killed", memcg_name);
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX);
|
|
|
|
if (ret < 0) {
|
|
|
|
rcu_read_unlock();
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Continues from above, so we don't need an KERN_ level
|
|
|
|
*/
|
|
|
|
printk(KERN_CONT " as a result of limit of %s\n", memcg_name);
|
|
|
|
done:
|
|
|
|
|
|
|
|
printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n",
|
|
|
|
res_counter_read_u64(&memcg->res, RES_USAGE) >> 10,
|
|
|
|
res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10,
|
|
|
|
res_counter_read_u64(&memcg->res, RES_FAILCNT));
|
|
|
|
printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, "
|
|
|
|
"failcnt %llu\n",
|
|
|
|
res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10,
|
|
|
|
res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10,
|
|
|
|
res_counter_read_u64(&memcg->memsw, RES_FAILCNT));
|
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:36 +00:00
|
|
|
/*
|
|
|
|
* This function returns the number of memcg under hierarchy tree. Returns
|
|
|
|
* 1(self count) if no children.
|
|
|
|
*/
|
|
|
|
static int mem_cgroup_count_children(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
int num = 0;
|
|
|
|
mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb);
|
|
|
|
return num;
|
|
|
|
}
|
|
|
|
|
oom: badness heuristic rewrite
This a complete rewrite of the oom killer's badness() heuristic which is
used to determine which task to kill in oom conditions. The goal is to
make it as simple and predictable as possible so the results are better
understood and we end up killing the task which will lead to the most
memory freeing while still respecting the fine-tuning from userspace.
Instead of basing the heuristic on mm->total_vm for each task, the task's
rss and swap space is used instead. This is a better indication of the
amount of memory that will be freeable if the oom killed task is chosen
and subsequently exits. This helps specifically in cases where KDE or
GNOME is chosen for oom kill on desktop systems instead of a memory
hogging task.
The baseline for the heuristic is a proportion of memory that each task is
currently using in memory plus swap compared to the amount of "allowable"
memory. "Allowable," in this sense, means the system-wide resources for
unconstrained oom conditions, the set of mempolicy nodes, the mems
attached to current's cpuset, or a memory controller's limit. The
proportion is given on a scale of 0 (never kill) to 1000 (always kill),
roughly meaning that if a task has a badness() score of 500 that the task
consumes approximately 50% of allowable memory resident in RAM or in swap
space.
The proportion is always relative to the amount of "allowable" memory and
not the total amount of RAM systemwide so that mempolicies and cpusets may
operate in isolation; they shall not need to know the true size of the
machine on which they are running if they are bound to a specific set of
nodes or mems, respectively.
Root tasks are given 3% extra memory just like __vm_enough_memory()
provides in LSMs. In the event of two tasks consuming similar amounts of
memory, it is generally better to save root's task.
Because of the change in the badness() heuristic's baseline, it is also
necessary to introduce a new user interface to tune it. It's not possible
to redefine the meaning of /proc/pid/oom_adj with a new scale since the
ABI cannot be changed for backward compatability. Instead, a new tunable,
/proc/pid/oom_score_adj, is added that ranges from -1000 to +1000. It may
be used to polarize the heuristic such that certain tasks are never
considered for oom kill while others may always be considered. The value
is added directly into the badness() score so a value of -500, for
example, means to discount 50% of its memory consumption in comparison to
other tasks either on the system, bound to the mempolicy, in the cpuset,
or sharing the same memory controller.
/proc/pid/oom_adj is changed so that its meaning is rescaled into the
units used by /proc/pid/oom_score_adj, and vice versa. Changing one of
these per-task tunables will rescale the value of the other to an
equivalent meaning. Although /proc/pid/oom_adj was originally defined as
a bitshift on the badness score, it now shares the same linear growth as
/proc/pid/oom_score_adj but with different granularity. This is required
so the ABI is not broken with userspace applications and allows oom_adj to
be deprecated for future removal.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-10 00:19:46 +00:00
|
|
|
/*
|
|
|
|
* Return the memory (and swap, if configured) limit for a memcg.
|
|
|
|
*/
|
|
|
|
u64 mem_cgroup_get_limit(struct mem_cgroup *memcg)
|
|
|
|
{
|
|
|
|
u64 limit;
|
|
|
|
u64 memsw;
|
|
|
|
|
|
|
|
limit = res_counter_read_u64(&memcg->res, RES_LIMIT) +
|
|
|
|
total_swap_pages;
|
|
|
|
memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
|
|
|
|
/*
|
|
|
|
* If memsw is finite and limits the amount of swap space available
|
|
|
|
* to this memcg, return that limit.
|
|
|
|
*/
|
|
|
|
return min(limit, memsw);
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:06 +00:00
|
|
|
/*
|
2009-04-02 23:57:33 +00:00
|
|
|
* Visit the first child (need not be the first child as per the ordering
|
|
|
|
* of the cgroup list, since we track last_scanned_child) of @mem and use
|
|
|
|
* that to reclaim free pages from.
|
|
|
|
*/
|
|
|
|
static struct mem_cgroup *
|
|
|
|
mem_cgroup_select_victim(struct mem_cgroup *root_mem)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *ret = NULL;
|
|
|
|
struct cgroup_subsys_state *css;
|
|
|
|
int nextid, found;
|
|
|
|
|
|
|
|
if (!root_mem->use_hierarchy) {
|
|
|
|
css_get(&root_mem->css);
|
|
|
|
ret = root_mem;
|
|
|
|
}
|
|
|
|
|
|
|
|
while (!ret) {
|
|
|
|
rcu_read_lock();
|
|
|
|
nextid = root_mem->last_scanned_child + 1;
|
|
|
|
css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css,
|
|
|
|
&found);
|
|
|
|
if (css && css_tryget(css))
|
|
|
|
ret = container_of(css, struct mem_cgroup, css);
|
|
|
|
|
|
|
|
rcu_read_unlock();
|
|
|
|
/* Updates scanning parameter */
|
|
|
|
spin_lock(&root_mem->reclaim_param_lock);
|
|
|
|
if (!css) {
|
|
|
|
/* this means start scan from ID:1 */
|
|
|
|
root_mem->last_scanned_child = 0;
|
|
|
|
} else
|
|
|
|
root_mem->last_scanned_child = found;
|
|
|
|
spin_unlock(&root_mem->reclaim_param_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Scan the hierarchy if needed to reclaim memory. We remember the last child
|
|
|
|
* we reclaimed from, so that we don't end up penalizing one child extensively
|
|
|
|
* based on its position in the children list.
|
2009-01-08 02:08:06 +00:00
|
|
|
*
|
|
|
|
* root_mem is the original ancestor that we've been reclaim from.
|
2009-04-02 23:57:33 +00:00
|
|
|
*
|
|
|
|
* We give up and return to the caller when we visit root_mem twice.
|
|
|
|
* (other groups can be removed while we're walking....)
|
2009-04-02 23:57:36 +00:00
|
|
|
*
|
|
|
|
* If shrink==true, for avoiding to free too much, this returns immedieately.
|
2009-01-08 02:08:06 +00:00
|
|
|
*/
|
|
|
|
static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem,
|
2009-09-23 22:56:39 +00:00
|
|
|
struct zone *zone,
|
2009-09-23 22:56:38 +00:00
|
|
|
gfp_t gfp_mask,
|
|
|
|
unsigned long reclaim_options)
|
2009-01-08 02:08:06 +00:00
|
|
|
{
|
2009-04-02 23:57:33 +00:00
|
|
|
struct mem_cgroup *victim;
|
|
|
|
int ret, total = 0;
|
|
|
|
int loop = 0;
|
2009-09-23 22:56:38 +00:00
|
|
|
bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP;
|
|
|
|
bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK;
|
2009-09-23 22:56:39 +00:00
|
|
|
bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT;
|
|
|
|
unsigned long excess = mem_cgroup_get_excess(root_mem);
|
2009-04-02 23:57:33 +00:00
|
|
|
|
2009-06-17 23:27:19 +00:00
|
|
|
/* If memsw_is_minimum==1, swap-out is of-no-use. */
|
|
|
|
if (root_mem->memsw_is_minimum)
|
|
|
|
noswap = true;
|
|
|
|
|
2009-09-23 22:56:39 +00:00
|
|
|
while (1) {
|
2009-04-02 23:57:33 +00:00
|
|
|
victim = mem_cgroup_select_victim(root_mem);
|
2009-09-23 22:56:39 +00:00
|
|
|
if (victim == root_mem) {
|
2009-04-02 23:57:33 +00:00
|
|
|
loop++;
|
2009-12-16 00:47:08 +00:00
|
|
|
if (loop >= 1)
|
|
|
|
drain_all_stock_async();
|
2009-09-23 22:56:39 +00:00
|
|
|
if (loop >= 2) {
|
|
|
|
/*
|
|
|
|
* If we have not been able to reclaim
|
|
|
|
* anything, it might because there are
|
|
|
|
* no reclaimable pages under this hierarchy
|
|
|
|
*/
|
|
|
|
if (!check_soft || !total) {
|
|
|
|
css_put(&victim->css);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* We want to do more targetted reclaim.
|
|
|
|
* excess >> 2 is not to excessive so as to
|
|
|
|
* reclaim too much, nor too less that we keep
|
|
|
|
* coming back to reclaim from this cgroup
|
|
|
|
*/
|
|
|
|
if (total >= (excess >> 2) ||
|
|
|
|
(loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) {
|
|
|
|
css_put(&victim->css);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
if (!mem_cgroup_local_usage(victim)) {
|
2009-04-02 23:57:33 +00:00
|
|
|
/* this cgroup's local usage == 0 */
|
|
|
|
css_put(&victim->css);
|
2009-01-08 02:08:06 +00:00
|
|
|
continue;
|
|
|
|
}
|
2009-04-02 23:57:33 +00:00
|
|
|
/* we use swappiness of local cgroup */
|
2009-09-23 22:56:39 +00:00
|
|
|
if (check_soft)
|
|
|
|
ret = mem_cgroup_shrink_node_zone(victim, gfp_mask,
|
2010-08-11 01:03:05 +00:00
|
|
|
noswap, get_swappiness(victim), zone);
|
2009-09-23 22:56:39 +00:00
|
|
|
else
|
|
|
|
ret = try_to_free_mem_cgroup_pages(victim, gfp_mask,
|
|
|
|
noswap, get_swappiness(victim));
|
2009-04-02 23:57:33 +00:00
|
|
|
css_put(&victim->css);
|
2009-04-02 23:57:36 +00:00
|
|
|
/*
|
|
|
|
* At shrinking usage, we can't check we should stop here or
|
|
|
|
* reclaim more. It's depends on callers. last_scanned_child
|
|
|
|
* will work enough for keeping fairness under tree.
|
|
|
|
*/
|
|
|
|
if (shrink)
|
|
|
|
return ret;
|
2009-04-02 23:57:33 +00:00
|
|
|
total += ret;
|
2009-09-23 22:56:39 +00:00
|
|
|
if (check_soft) {
|
|
|
|
if (res_counter_check_under_soft_limit(&root_mem->res))
|
|
|
|
return total;
|
|
|
|
} else if (mem_cgroup_check_under_limit(root_mem))
|
2009-04-02 23:57:33 +00:00
|
|
|
return 1 + total;
|
2009-01-08 02:08:06 +00:00
|
|
|
}
|
2009-04-02 23:57:33 +00:00
|
|
|
return total;
|
2009-01-08 02:08:06 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:39 +00:00
|
|
|
static int mem_cgroup_oom_lock_cb(struct mem_cgroup *mem, void *data)
|
2009-01-08 02:08:08 +00:00
|
|
|
{
|
2010-03-10 23:22:39 +00:00
|
|
|
int *val = (int *)data;
|
|
|
|
int x;
|
|
|
|
/*
|
|
|
|
* Logically, we can stop scanning immediately when we find
|
|
|
|
* a memcg is already locked. But condidering unlock ops and
|
|
|
|
* creation/removal of memcg, scan-all is simple operation.
|
|
|
|
*/
|
|
|
|
x = atomic_inc_return(&mem->oom_lock);
|
|
|
|
*val = max(x, *val);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* Check OOM-Killer is already running under our hierarchy.
|
|
|
|
* If someone is running, return false.
|
|
|
|
*/
|
|
|
|
static bool mem_cgroup_oom_lock(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
int lock_count = 0;
|
2009-01-08 02:08:08 +00:00
|
|
|
|
2010-03-10 23:22:39 +00:00
|
|
|
mem_cgroup_walk_tree(mem, &lock_count, mem_cgroup_oom_lock_cb);
|
|
|
|
|
|
|
|
if (lock_count == 1)
|
|
|
|
return true;
|
|
|
|
return false;
|
2009-01-08 02:08:08 +00:00
|
|
|
}
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
|
2010-03-10 23:22:39 +00:00
|
|
|
static int mem_cgroup_oom_unlock_cb(struct mem_cgroup *mem, void *data)
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
{
|
2010-03-10 23:22:39 +00:00
|
|
|
/*
|
|
|
|
* When a new child is created while the hierarchy is under oom,
|
|
|
|
* mem_cgroup_oom_lock() may not be called. We have to use
|
|
|
|
* atomic_add_unless() here.
|
|
|
|
*/
|
|
|
|
atomic_add_unless(&mem->oom_lock, -1, 0);
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:39 +00:00
|
|
|
static void mem_cgroup_oom_unlock(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_unlock_cb);
|
|
|
|
}
|
|
|
|
|
|
|
|
static DEFINE_MUTEX(memcg_oom_mutex);
|
|
|
|
static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
|
|
|
|
|
2010-05-26 21:42:36 +00:00
|
|
|
struct oom_wait_info {
|
|
|
|
struct mem_cgroup *mem;
|
|
|
|
wait_queue_t wait;
|
|
|
|
};
|
|
|
|
|
|
|
|
static int memcg_oom_wake_function(wait_queue_t *wait,
|
|
|
|
unsigned mode, int sync, void *arg)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg;
|
|
|
|
struct oom_wait_info *oom_wait_info;
|
|
|
|
|
|
|
|
oom_wait_info = container_of(wait, struct oom_wait_info, wait);
|
|
|
|
|
|
|
|
if (oom_wait_info->mem == wake_mem)
|
|
|
|
goto wakeup;
|
|
|
|
/* if no hierarchy, no match */
|
|
|
|
if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy)
|
|
|
|
return 0;
|
|
|
|
/*
|
|
|
|
* Both of oom_wait_info->mem and wake_mem are stable under us.
|
|
|
|
* Then we can use css_is_ancestor without taking care of RCU.
|
|
|
|
*/
|
|
|
|
if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) &&
|
|
|
|
!css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
wakeup:
|
|
|
|
return autoremove_wake_function(wait, mode, sync, arg);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void memcg_wakeup_oom(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
/* for filtering, pass "mem" as argument. */
|
|
|
|
__wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem);
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:37 +00:00
|
|
|
static void memcg_oom_recover(struct mem_cgroup *mem)
|
|
|
|
{
|
2010-08-11 01:02:58 +00:00
|
|
|
if (mem && atomic_read(&mem->oom_lock))
|
2010-05-26 21:42:37 +00:00
|
|
|
memcg_wakeup_oom(mem);
|
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:39 +00:00
|
|
|
/*
|
|
|
|
* try to call OOM killer. returns false if we should exit memory-reclaim loop.
|
|
|
|
*/
|
|
|
|
bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask)
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
{
|
2010-05-26 21:42:36 +00:00
|
|
|
struct oom_wait_info owait;
|
2010-05-26 21:42:37 +00:00
|
|
|
bool locked, need_to_kill;
|
2010-03-10 23:22:39 +00:00
|
|
|
|
2010-05-26 21:42:36 +00:00
|
|
|
owait.mem = mem;
|
|
|
|
owait.wait.flags = 0;
|
|
|
|
owait.wait.func = memcg_oom_wake_function;
|
|
|
|
owait.wait.private = current;
|
|
|
|
INIT_LIST_HEAD(&owait.wait.task_list);
|
2010-05-26 21:42:37 +00:00
|
|
|
need_to_kill = true;
|
2010-03-10 23:22:39 +00:00
|
|
|
/* At first, try to OOM lock hierarchy under mem.*/
|
|
|
|
mutex_lock(&memcg_oom_mutex);
|
|
|
|
locked = mem_cgroup_oom_lock(mem);
|
|
|
|
/*
|
|
|
|
* Even if signal_pending(), we can't quit charge() loop without
|
|
|
|
* accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
|
|
|
|
* under OOM is always welcomed, use TASK_KILLABLE here.
|
|
|
|
*/
|
2010-05-26 21:42:37 +00:00
|
|
|
prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
|
|
|
|
if (!locked || mem->oom_kill_disable)
|
|
|
|
need_to_kill = false;
|
|
|
|
if (locked)
|
2010-05-26 21:42:36 +00:00
|
|
|
mem_cgroup_oom_notify(mem);
|
2010-03-10 23:22:39 +00:00
|
|
|
mutex_unlock(&memcg_oom_mutex);
|
|
|
|
|
2010-05-26 21:42:37 +00:00
|
|
|
if (need_to_kill) {
|
|
|
|
finish_wait(&memcg_oom_waitq, &owait.wait);
|
2010-03-10 23:22:39 +00:00
|
|
|
mem_cgroup_out_of_memory(mem, mask);
|
2010-05-26 21:42:37 +00:00
|
|
|
} else {
|
2010-03-10 23:22:39 +00:00
|
|
|
schedule();
|
2010-05-26 21:42:36 +00:00
|
|
|
finish_wait(&memcg_oom_waitq, &owait.wait);
|
2010-03-10 23:22:39 +00:00
|
|
|
}
|
|
|
|
mutex_lock(&memcg_oom_mutex);
|
|
|
|
mem_cgroup_oom_unlock(mem);
|
2010-05-26 21:42:36 +00:00
|
|
|
memcg_wakeup_oom(mem);
|
2010-03-10 23:22:39 +00:00
|
|
|
mutex_unlock(&memcg_oom_mutex);
|
|
|
|
|
|
|
|
if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current))
|
|
|
|
return false;
|
|
|
|
/* Give chance to dying process */
|
|
|
|
schedule_timeout(1);
|
|
|
|
return true;
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
}
|
|
|
|
|
2009-06-17 23:26:34 +00:00
|
|
|
/*
|
|
|
|
* Currently used to update mapped file statistics, but the routine can be
|
|
|
|
* generalized to update other statistics as well.
|
|
|
|
*/
|
2009-12-16 00:47:09 +00:00
|
|
|
void mem_cgroup_update_file_mapped(struct page *page, int val)
|
2009-06-17 23:26:34 +00:00
|
|
|
{
|
|
|
|
struct mem_cgroup *mem;
|
|
|
|
struct page_cgroup *pc;
|
|
|
|
|
|
|
|
pc = lookup_page_cgroup(page);
|
|
|
|
if (unlikely(!pc))
|
|
|
|
return;
|
|
|
|
|
|
|
|
lock_page_cgroup(pc);
|
|
|
|
mem = pc->mem_cgroup;
|
2010-04-06 21:35:05 +00:00
|
|
|
if (!mem || !PageCgroupUsed(pc))
|
2009-06-17 23:26:34 +00:00
|
|
|
goto done;
|
|
|
|
|
|
|
|
/*
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
* Preemption is already disabled. We can use __this_cpu_xxx
|
2009-06-17 23:26:34 +00:00
|
|
|
*/
|
2010-04-06 21:35:05 +00:00
|
|
|
if (val > 0) {
|
|
|
|
__this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
|
|
|
|
SetPageCgroupFileMapped(pc);
|
|
|
|
} else {
|
|
|
|
__this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
|
|
|
|
ClearPageCgroupFileMapped(pc);
|
|
|
|
}
|
2009-06-17 23:26:34 +00:00
|
|
|
|
|
|
|
done:
|
|
|
|
unlock_page_cgroup(pc);
|
|
|
|
}
|
memcg: fix OOM killer under memcg
This patch tries to fix OOM Killer problems caused by hierarchy.
Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to
kill a task in memcg.
But, when hierarchy is used, it's broken and correct task cannot
be killed. For example, in following cgroup
/groupA/ hierarchy=1, limit=1G,
01 nolimit
02 nolimit
All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to
groupA's 1Gbytes but OOM Killer just kills tasks in groupA.
This patch provides makes the bad process be selected from all tasks
under hierarchy. BTW, currently, oom_jiffies is updated against groupA
in above case. oom_jiffies of tree should be updated.
To see how oom_jiffies is used, please check mem_cgroup_oom_called()
callers.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: const fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-02 23:57:38 +00:00
|
|
|
|
2009-12-16 00:47:08 +00:00
|
|
|
/*
|
|
|
|
* size of first charge trial. "32" comes from vmscan.c's magic value.
|
|
|
|
* TODO: maybe necessary to use big numbers in big irons.
|
|
|
|
*/
|
|
|
|
#define CHARGE_SIZE (32 * PAGE_SIZE)
|
|
|
|
struct memcg_stock_pcp {
|
|
|
|
struct mem_cgroup *cached; /* this never be root cgroup */
|
|
|
|
int charge;
|
|
|
|
struct work_struct work;
|
|
|
|
};
|
|
|
|
static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
|
|
|
|
static atomic_t memcg_drain_count;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
|
|
|
|
* from local stock and true is returned. If the stock is 0 or charges from a
|
|
|
|
* cgroup which is not current target, returns false. This stock will be
|
|
|
|
* refilled.
|
|
|
|
*/
|
|
|
|
static bool consume_stock(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
struct memcg_stock_pcp *stock;
|
|
|
|
bool ret = true;
|
|
|
|
|
|
|
|
stock = &get_cpu_var(memcg_stock);
|
|
|
|
if (mem == stock->cached && stock->charge)
|
|
|
|
stock->charge -= PAGE_SIZE;
|
|
|
|
else /* need to call res_counter_charge */
|
|
|
|
ret = false;
|
|
|
|
put_cpu_var(memcg_stock);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Returns stocks cached in percpu to res_counter and reset cached information.
|
|
|
|
*/
|
|
|
|
static void drain_stock(struct memcg_stock_pcp *stock)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *old = stock->cached;
|
|
|
|
|
|
|
|
if (stock->charge) {
|
|
|
|
res_counter_uncharge(&old->res, stock->charge);
|
|
|
|
if (do_swap_account)
|
|
|
|
res_counter_uncharge(&old->memsw, stock->charge);
|
|
|
|
}
|
|
|
|
stock->cached = NULL;
|
|
|
|
stock->charge = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This must be called under preempt disabled or must be called by
|
|
|
|
* a thread which is pinned to local cpu.
|
|
|
|
*/
|
|
|
|
static void drain_local_stock(struct work_struct *dummy)
|
|
|
|
{
|
|
|
|
struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
|
|
|
|
drain_stock(stock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Cache charges(val) which is from res_counter, to local per_cpu area.
|
2010-03-15 14:27:28 +00:00
|
|
|
* This will be consumed by consume_stock() function, later.
|
2009-12-16 00:47:08 +00:00
|
|
|
*/
|
|
|
|
static void refill_stock(struct mem_cgroup *mem, int val)
|
|
|
|
{
|
|
|
|
struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
|
|
|
|
|
|
|
|
if (stock->cached != mem) { /* reset if necessary */
|
|
|
|
drain_stock(stock);
|
|
|
|
stock->cached = mem;
|
|
|
|
}
|
|
|
|
stock->charge += val;
|
|
|
|
put_cpu_var(memcg_stock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Tries to drain stocked charges in other cpus. This function is asynchronous
|
|
|
|
* and just put a work per cpu for draining localy on each cpu. Caller can
|
|
|
|
* expects some charges will be back to res_counter later but cannot wait for
|
|
|
|
* it.
|
|
|
|
*/
|
|
|
|
static void drain_all_stock_async(void)
|
|
|
|
{
|
|
|
|
int cpu;
|
|
|
|
/* This function is for scheduling "drain" in asynchronous way.
|
|
|
|
* The result of "drain" is not directly handled by callers. Then,
|
|
|
|
* if someone is calling drain, we don't have to call drain more.
|
|
|
|
* Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
|
|
|
|
* there is a race. We just do loose check here.
|
|
|
|
*/
|
|
|
|
if (atomic_read(&memcg_drain_count))
|
|
|
|
return;
|
|
|
|
/* Notify other cpus that system-wide "drain" is running */
|
|
|
|
atomic_inc(&memcg_drain_count);
|
|
|
|
get_online_cpus();
|
|
|
|
for_each_online_cpu(cpu) {
|
|
|
|
struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
|
|
|
|
schedule_work_on(cpu, &stock->work);
|
|
|
|
}
|
|
|
|
put_online_cpus();
|
|
|
|
atomic_dec(&memcg_drain_count);
|
|
|
|
/* We don't wait for flush_work */
|
|
|
|
}
|
|
|
|
|
|
|
|
/* This is a synchronous drain interface. */
|
|
|
|
static void drain_all_stock_sync(void)
|
|
|
|
{
|
|
|
|
/* called when force_empty is called */
|
|
|
|
atomic_inc(&memcg_drain_count);
|
|
|
|
schedule_on_each_cpu(drain_local_stock);
|
|
|
|
atomic_dec(&memcg_drain_count);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb,
|
|
|
|
unsigned long action,
|
|
|
|
void *hcpu)
|
|
|
|
{
|
|
|
|
int cpu = (unsigned long)hcpu;
|
|
|
|
struct memcg_stock_pcp *stock;
|
|
|
|
|
|
|
|
if (action != CPU_DEAD)
|
|
|
|
return NOTIFY_OK;
|
|
|
|
stock = &per_cpu(memcg_stock, cpu);
|
|
|
|
drain_stock(stock);
|
|
|
|
return NOTIFY_OK;
|
|
|
|
}
|
|
|
|
|
2010-08-11 01:02:57 +00:00
|
|
|
|
|
|
|
/* See __mem_cgroup_try_charge() for details */
|
|
|
|
enum {
|
|
|
|
CHARGE_OK, /* success */
|
|
|
|
CHARGE_RETRY, /* need to retry but retry is not bad */
|
|
|
|
CHARGE_NOMEM, /* we can't do more. return -ENOMEM */
|
|
|
|
CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */
|
|
|
|
CHARGE_OOM_DIE, /* the current is killed because of OOM */
|
|
|
|
};
|
|
|
|
|
|
|
|
static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
|
|
|
|
int csize, bool oom_check)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem_over_limit;
|
|
|
|
struct res_counter *fail_res;
|
|
|
|
unsigned long flags = 0;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = res_counter_charge(&mem->res, csize, &fail_res);
|
|
|
|
|
|
|
|
if (likely(!ret)) {
|
|
|
|
if (!do_swap_account)
|
|
|
|
return CHARGE_OK;
|
|
|
|
ret = res_counter_charge(&mem->memsw, csize, &fail_res);
|
|
|
|
if (likely(!ret))
|
|
|
|
return CHARGE_OK;
|
|
|
|
|
|
|
|
mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw);
|
|
|
|
flags |= MEM_CGROUP_RECLAIM_NOSWAP;
|
|
|
|
} else
|
|
|
|
mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
|
|
|
|
|
|
|
|
if (csize > PAGE_SIZE) /* change csize and retry */
|
|
|
|
return CHARGE_RETRY;
|
|
|
|
|
|
|
|
if (!(gfp_mask & __GFP_WAIT))
|
|
|
|
return CHARGE_WOULDBLOCK;
|
|
|
|
|
|
|
|
ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
|
|
|
|
gfp_mask, flags);
|
|
|
|
/*
|
|
|
|
* try_to_free_mem_cgroup_pages() might not give us a full
|
|
|
|
* picture of reclaim. Some pages are reclaimed and might be
|
|
|
|
* moved to swap cache or just unmapped from the cgroup.
|
|
|
|
* Check the limit again to see if the reclaim reduced the
|
|
|
|
* current usage of the cgroup before giving up
|
|
|
|
*/
|
|
|
|
if (ret || mem_cgroup_check_under_limit(mem_over_limit))
|
|
|
|
return CHARGE_RETRY;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* At task move, charge accounts can be doubly counted. So, it's
|
|
|
|
* better to wait until the end of task_move if something is going on.
|
|
|
|
*/
|
|
|
|
if (mem_cgroup_wait_acct_move(mem_over_limit))
|
|
|
|
return CHARGE_RETRY;
|
|
|
|
|
|
|
|
/* If we don't need to call oom-killer at el, return immediately */
|
|
|
|
if (!oom_check)
|
|
|
|
return CHARGE_NOMEM;
|
|
|
|
/* check OOM */
|
|
|
|
if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask))
|
|
|
|
return CHARGE_OOM_DIE;
|
|
|
|
|
|
|
|
return CHARGE_RETRY;
|
|
|
|
}
|
|
|
|
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
/*
|
|
|
|
* Unlike exported interface, "oom" parameter is added. if oom==true,
|
|
|
|
* oom-killer can be invoked.
|
2008-02-07 08:13:53 +00:00
|
|
|
*/
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
static int __mem_cgroup_try_charge(struct mm_struct *mm,
|
2010-08-11 01:02:57 +00:00
|
|
|
gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom)
|
2008-02-07 08:13:53 +00:00
|
|
|
{
|
2010-08-11 01:02:57 +00:00
|
|
|
int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
|
|
|
|
struct mem_cgroup *mem = NULL;
|
|
|
|
int ret;
|
2009-12-16 00:47:08 +00:00
|
|
|
int csize = CHARGE_SIZE;
|
2009-01-08 02:08:08 +00:00
|
|
|
|
2010-03-10 23:22:39 +00:00
|
|
|
/*
|
|
|
|
* Unlike gloval-vm's OOM-kill, we're not in memory shortage
|
|
|
|
* in system level. So, allow to go ahead dying process in addition to
|
|
|
|
* MEMDIE process.
|
|
|
|
*/
|
|
|
|
if (unlikely(test_thread_flag(TIF_MEMDIE)
|
|
|
|
|| fatal_signal_pending(current)))
|
|
|
|
goto bypass;
|
2009-01-08 02:08:08 +00:00
|
|
|
|
2008-02-07 08:13:53 +00:00
|
|
|
/*
|
2008-02-07 08:14:19 +00:00
|
|
|
* We always charge the cgroup the mm_struct belongs to.
|
|
|
|
* The mm_struct's mem_cgroup changes on task migration if the
|
2008-02-07 08:13:53 +00:00
|
|
|
* thread group leader migrates. It's possible that mm is not
|
|
|
|
* set, if so charge the init_mm (happens for pagecache usage).
|
|
|
|
*/
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
if (!*memcg && !mm)
|
|
|
|
goto bypass;
|
|
|
|
again:
|
|
|
|
if (*memcg) { /* css should be a valid one */
|
2010-08-11 01:02:57 +00:00
|
|
|
mem = *memcg;
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
VM_BUG_ON(css_is_removed(&mem->css));
|
|
|
|
if (mem_cgroup_is_root(mem))
|
|
|
|
goto done;
|
|
|
|
if (consume_stock(mem))
|
|
|
|
goto done;
|
2010-08-11 01:02:57 +00:00
|
|
|
css_get(&mem->css);
|
|
|
|
} else {
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
struct task_struct *p;
|
2009-01-08 02:08:33 +00:00
|
|
|
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
rcu_read_lock();
|
|
|
|
p = rcu_dereference(mm->owner);
|
|
|
|
VM_BUG_ON(!p);
|
|
|
|
/*
|
|
|
|
* because we don't have task_lock(), "p" can exit while
|
|
|
|
* we're here. In that case, "mem" can point to root
|
|
|
|
* cgroup but never be NULL. (and task_struct itself is freed
|
|
|
|
* by RCU, cgroup itself is RCU safe.) Then, we have small
|
|
|
|
* risk here to get wrong cgroup. But such kind of mis-account
|
|
|
|
* by race always happens because we don't have cgroup_mutex().
|
|
|
|
* It's overkill and we allow that small race, here.
|
|
|
|
*/
|
|
|
|
mem = mem_cgroup_from_task(p);
|
|
|
|
VM_BUG_ON(!mem);
|
|
|
|
if (mem_cgroup_is_root(mem)) {
|
|
|
|
rcu_read_unlock();
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
if (consume_stock(mem)) {
|
|
|
|
/*
|
|
|
|
* It seems dagerous to access memcg without css_get().
|
|
|
|
* But considering how consume_stok works, it's not
|
|
|
|
* necessary. If consume_stock success, some charges
|
|
|
|
* from this memcg are cached on this cpu. So, we
|
|
|
|
* don't need to call css_get()/css_tryget() before
|
|
|
|
* calling consume_stock().
|
|
|
|
*/
|
|
|
|
rcu_read_unlock();
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
/* after here, we may be blocked. we need to get refcnt */
|
|
|
|
if (!css_tryget(&mem->css)) {
|
|
|
|
rcu_read_unlock();
|
|
|
|
goto again;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
2008-02-07 08:13:53 +00:00
|
|
|
|
2010-08-11 01:02:57 +00:00
|
|
|
do {
|
|
|
|
bool oom_check;
|
2009-01-08 02:07:48 +00:00
|
|
|
|
2010-08-11 01:02:57 +00:00
|
|
|
/* If killed, bypass charge */
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
if (fatal_signal_pending(current)) {
|
|
|
|
css_put(&mem->css);
|
2010-08-11 01:02:57 +00:00
|
|
|
goto bypass;
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
}
|
2009-01-08 02:08:06 +00:00
|
|
|
|
2010-08-11 01:02:57 +00:00
|
|
|
oom_check = false;
|
|
|
|
if (oom && !nr_oom_retries) {
|
|
|
|
oom_check = true;
|
|
|
|
nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
|
2009-12-16 00:47:08 +00:00
|
|
|
}
|
2008-02-07 08:13:56 +00:00
|
|
|
|
2010-08-11 01:02:57 +00:00
|
|
|
ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check);
|
2010-03-10 23:22:16 +00:00
|
|
|
|
2010-08-11 01:02:57 +00:00
|
|
|
switch (ret) {
|
|
|
|
case CHARGE_OK:
|
|
|
|
break;
|
|
|
|
case CHARGE_RETRY: /* not in OOM situation but retry */
|
|
|
|
csize = PAGE_SIZE;
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
css_put(&mem->css);
|
|
|
|
mem = NULL;
|
|
|
|
goto again;
|
2010-08-11 01:02:57 +00:00
|
|
|
case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
css_put(&mem->css);
|
2010-08-11 01:02:57 +00:00
|
|
|
goto nomem;
|
|
|
|
case CHARGE_NOMEM: /* OOM routine works */
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
if (!oom) {
|
|
|
|
css_put(&mem->css);
|
2010-03-10 23:22:39 +00:00
|
|
|
goto nomem;
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
}
|
2010-08-11 01:02:57 +00:00
|
|
|
/* If oom, we never return -ENOMEM */
|
|
|
|
nr_oom_retries--;
|
|
|
|
break;
|
|
|
|
case CHARGE_OOM_DIE: /* Killed by OOM Killer */
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
css_put(&mem->css);
|
2010-03-10 23:22:39 +00:00
|
|
|
goto bypass;
|
2008-02-07 08:13:56 +00:00
|
|
|
}
|
2010-08-11 01:02:57 +00:00
|
|
|
} while (ret != CHARGE_OK);
|
|
|
|
|
2009-12-16 00:47:08 +00:00
|
|
|
if (csize > PAGE_SIZE)
|
|
|
|
refill_stock(mem, csize - PAGE_SIZE);
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
css_put(&mem->css);
|
2009-09-23 22:56:42 +00:00
|
|
|
done:
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
*memcg = mem;
|
2009-01-08 02:07:48 +00:00
|
|
|
return 0;
|
|
|
|
nomem:
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
*memcg = NULL;
|
2009-01-08 02:07:48 +00:00
|
|
|
return -ENOMEM;
|
2010-03-10 23:22:39 +00:00
|
|
|
bypass:
|
|
|
|
*memcg = NULL;
|
|
|
|
return 0;
|
2009-01-08 02:07:48 +00:00
|
|
|
}
|
2008-02-07 08:13:53 +00:00
|
|
|
|
2009-12-16 00:47:10 +00:00
|
|
|
/*
|
|
|
|
* Somemtimes we have to undo a charge we got by try_charge().
|
|
|
|
* This function is for that and do uncharge, put css's refcnt.
|
|
|
|
* gotten by try_charge().
|
|
|
|
*/
|
2010-03-10 23:22:15 +00:00
|
|
|
static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem,
|
|
|
|
unsigned long count)
|
2009-12-16 00:47:10 +00:00
|
|
|
{
|
|
|
|
if (!mem_cgroup_is_root(mem)) {
|
2010-03-10 23:22:15 +00:00
|
|
|
res_counter_uncharge(&mem->res, PAGE_SIZE * count);
|
2009-12-16 00:47:10 +00:00
|
|
|
if (do_swap_account)
|
2010-03-10 23:22:15 +00:00
|
|
|
res_counter_uncharge(&mem->memsw, PAGE_SIZE * count);
|
2009-12-16 00:47:10 +00:00
|
|
|
}
|
2010-03-10 23:22:15 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void mem_cgroup_cancel_charge(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
__mem_cgroup_cancel_charge(mem, 1);
|
2009-12-16 00:47:10 +00:00
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:45 +00:00
|
|
|
/*
|
|
|
|
* A helper function to get mem_cgroup from ID. must be called under
|
|
|
|
* rcu_read_lock(). The caller must check css_is_removed() or some if
|
|
|
|
* it's concern. (dropping refcnt from swap can be called against removed
|
|
|
|
* memcg.)
|
|
|
|
*/
|
|
|
|
static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
|
|
|
|
{
|
|
|
|
struct cgroup_subsys_state *css;
|
|
|
|
|
|
|
|
/* ID 0 is unused ID */
|
|
|
|
if (!id)
|
|
|
|
return NULL;
|
|
|
|
css = css_lookup(&mem_cgroup_subsys, id);
|
|
|
|
if (!css)
|
|
|
|
return NULL;
|
|
|
|
return container_of(css, struct mem_cgroup, css);
|
|
|
|
}
|
|
|
|
|
2009-12-16 11:19:59 +00:00
|
|
|
struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
|
2009-01-08 02:08:35 +00:00
|
|
|
{
|
2009-12-16 11:19:59 +00:00
|
|
|
struct mem_cgroup *mem = NULL;
|
2009-04-02 23:57:43 +00:00
|
|
|
struct page_cgroup *pc;
|
2009-04-02 23:57:45 +00:00
|
|
|
unsigned short id;
|
2009-01-08 02:08:35 +00:00
|
|
|
swp_entry_t ent;
|
|
|
|
|
2009-04-02 23:57:43 +00:00
|
|
|
VM_BUG_ON(!PageLocked(page));
|
|
|
|
|
|
|
|
pc = lookup_page_cgroup(page);
|
2009-04-30 22:08:11 +00:00
|
|
|
lock_page_cgroup(pc);
|
2009-04-02 23:57:45 +00:00
|
|
|
if (PageCgroupUsed(pc)) {
|
2009-04-02 23:57:43 +00:00
|
|
|
mem = pc->mem_cgroup;
|
2009-04-02 23:57:45 +00:00
|
|
|
if (mem && !css_tryget(&mem->css))
|
|
|
|
mem = NULL;
|
2009-12-16 11:19:59 +00:00
|
|
|
} else if (PageSwapCache(page)) {
|
2009-04-02 23:57:43 +00:00
|
|
|
ent.val = page_private(page);
|
2009-04-02 23:57:45 +00:00
|
|
|
id = lookup_swap_cgroup(ent);
|
|
|
|
rcu_read_lock();
|
|
|
|
mem = mem_cgroup_lookup(id);
|
|
|
|
if (mem && !css_tryget(&mem->css))
|
|
|
|
mem = NULL;
|
|
|
|
rcu_read_unlock();
|
2009-04-02 23:57:43 +00:00
|
|
|
}
|
2009-04-30 22:08:11 +00:00
|
|
|
unlock_page_cgroup(pc);
|
2009-01-08 02:08:35 +00:00
|
|
|
return mem;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:07:48 +00:00
|
|
|
/*
|
2009-01-08 02:08:28 +00:00
|
|
|
* commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
|
2009-01-08 02:07:48 +00:00
|
|
|
* USED state. If already USED, uncharge and return.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
|
|
|
|
struct page_cgroup *pc,
|
|
|
|
enum charge_type ctype)
|
|
|
|
{
|
|
|
|
/* try_charge() can return NULL to *memcg, taking care of it. */
|
|
|
|
if (!mem)
|
|
|
|
return;
|
2008-10-19 03:28:16 +00:00
|
|
|
|
|
|
|
lock_page_cgroup(pc);
|
|
|
|
if (unlikely(PageCgroupUsed(pc))) {
|
|
|
|
unlock_page_cgroup(pc);
|
2009-12-16 00:47:10 +00:00
|
|
|
mem_cgroup_cancel_charge(mem);
|
2009-01-08 02:07:48 +00:00
|
|
|
return;
|
2008-10-19 03:28:16 +00:00
|
|
|
}
|
2009-09-23 22:56:32 +00:00
|
|
|
|
2008-02-07 08:13:53 +00:00
|
|
|
pc->mem_cgroup = mem;
|
2009-09-23 22:56:33 +00:00
|
|
|
/*
|
|
|
|
* We access a page_cgroup asynchronously without lock_page_cgroup().
|
|
|
|
* Especially when a page_cgroup is taken from a page, pc->mem_cgroup
|
|
|
|
* is accessed after testing USED bit. To make pc->mem_cgroup visible
|
|
|
|
* before USED bit, we need memory barrier here.
|
|
|
|
* See mem_cgroup_add_lru_list(), etc.
|
|
|
|
*/
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
smp_wmb();
|
2009-09-23 22:56:32 +00:00
|
|
|
switch (ctype) {
|
|
|
|
case MEM_CGROUP_CHARGE_TYPE_CACHE:
|
|
|
|
case MEM_CGROUP_CHARGE_TYPE_SHMEM:
|
|
|
|
SetPageCgroupCache(pc);
|
|
|
|
SetPageCgroupUsed(pc);
|
|
|
|
break;
|
|
|
|
case MEM_CGROUP_CHARGE_TYPE_MAPPED:
|
|
|
|
ClearPageCgroupCache(pc);
|
|
|
|
SetPageCgroupUsed(pc);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
2008-02-07 08:14:19 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
mem_cgroup_charge_statistics(mem, pc, true);
|
2008-10-19 03:28:16 +00:00
|
|
|
|
|
|
|
unlock_page_cgroup(pc);
|
2010-03-10 23:22:30 +00:00
|
|
|
/*
|
|
|
|
* "charge_statistics" updated event counter. Then, check it.
|
|
|
|
* Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
|
|
|
|
* if they exceeds softlimit.
|
|
|
|
*/
|
2010-03-10 23:22:31 +00:00
|
|
|
memcg_check_events(mem, pc->page);
|
2009-01-08 02:07:48 +00:00
|
|
|
}
|
2008-02-07 08:13:56 +00:00
|
|
|
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
/**
|
2009-12-16 00:47:11 +00:00
|
|
|
* __mem_cgroup_move_account - move account of the page
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
* @pc: page_cgroup of the page.
|
|
|
|
* @from: mem_cgroup which the page is moved from.
|
|
|
|
* @to: mem_cgroup which the page is moved to. @from != @to.
|
2010-03-10 23:22:15 +00:00
|
|
|
* @uncharge: whether we should call uncharge and css_put against @from.
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
*
|
|
|
|
* The caller must confirm following.
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
* - page is not on LRU (isolate_page() is useful.)
|
2009-12-16 00:47:11 +00:00
|
|
|
* - the pc is locked, used, and ->mem_cgroup points to @from.
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
*
|
2010-03-10 23:22:15 +00:00
|
|
|
* This function doesn't do "charge" nor css_get to new cgroup. It should be
|
|
|
|
* done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
|
|
|
|
* true, this function does "uncharge" from old cgroup, but it doesn't if
|
|
|
|
* @uncharge is false, so a caller should do "uncharge".
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
*/
|
|
|
|
|
2009-12-16 00:47:11 +00:00
|
|
|
static void __mem_cgroup_move_account(struct page_cgroup *pc,
|
2010-03-10 23:22:15 +00:00
|
|
|
struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
{
|
|
|
|
VM_BUG_ON(from == to);
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
VM_BUG_ON(PageLRU(pc->page));
|
2009-12-16 00:47:11 +00:00
|
|
|
VM_BUG_ON(!PageCgroupLocked(pc));
|
|
|
|
VM_BUG_ON(!PageCgroupUsed(pc));
|
|
|
|
VM_BUG_ON(pc->mem_cgroup != from);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
|
2010-04-06 21:35:05 +00:00
|
|
|
if (PageCgroupFileMapped(pc)) {
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
/* Update mapped_file data for mem_cgroup */
|
|
|
|
preempt_disable();
|
|
|
|
__this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
|
|
|
|
__this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
|
|
|
|
preempt_enable();
|
2009-06-17 23:26:34 +00:00
|
|
|
}
|
2010-03-10 23:22:15 +00:00
|
|
|
mem_cgroup_charge_statistics(from, pc, false);
|
|
|
|
if (uncharge)
|
|
|
|
/* This is not "cancel", but cancel_charge does all we need. */
|
|
|
|
mem_cgroup_cancel_charge(from);
|
2009-06-17 23:26:34 +00:00
|
|
|
|
2010-03-10 23:22:15 +00:00
|
|
|
/* caller should have done css_get */
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
pc->mem_cgroup = to;
|
|
|
|
mem_cgroup_charge_statistics(to, pc, true);
|
2009-07-29 22:04:06 +00:00
|
|
|
/*
|
|
|
|
* We charges against "to" which may not have any tasks. Then, "to"
|
|
|
|
* can be under rmdir(). But in current implementation, caller of
|
2010-03-10 23:22:14 +00:00
|
|
|
* this function is just force_empty() and move charge, so it's
|
|
|
|
* garanteed that "to" is never removed. So, we don't check rmdir
|
|
|
|
* status here.
|
2009-07-29 22:04:06 +00:00
|
|
|
*/
|
2009-12-16 00:47:11 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* check whether the @pc is valid for moving account and call
|
|
|
|
* __mem_cgroup_move_account()
|
|
|
|
*/
|
|
|
|
static int mem_cgroup_move_account(struct page_cgroup *pc,
|
2010-03-10 23:22:15 +00:00
|
|
|
struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge)
|
2009-12-16 00:47:11 +00:00
|
|
|
{
|
|
|
|
int ret = -EINVAL;
|
|
|
|
lock_page_cgroup(pc);
|
|
|
|
if (PageCgroupUsed(pc) && pc->mem_cgroup == from) {
|
2010-03-10 23:22:15 +00:00
|
|
|
__mem_cgroup_move_account(pc, from, to, uncharge);
|
2009-12-16 00:47:11 +00:00
|
|
|
ret = 0;
|
|
|
|
}
|
|
|
|
unlock_page_cgroup(pc);
|
2010-03-10 23:22:31 +00:00
|
|
|
/*
|
|
|
|
* check events
|
|
|
|
*/
|
|
|
|
memcg_check_events(to, pc->page);
|
|
|
|
memcg_check_events(from, pc->page);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* move charges to its parent.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static int mem_cgroup_move_parent(struct page_cgroup *pc,
|
|
|
|
struct mem_cgroup *child,
|
|
|
|
gfp_t gfp_mask)
|
|
|
|
{
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
struct page *page = pc->page;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
struct cgroup *cg = child->css.cgroup;
|
|
|
|
struct cgroup *pcg = cg->parent;
|
|
|
|
struct mem_cgroup *parent;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/* Is ROOT ? */
|
|
|
|
if (!pcg)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2009-12-16 00:47:11 +00:00
|
|
|
ret = -EBUSY;
|
|
|
|
if (!get_page_unless_zero(page))
|
|
|
|
goto out;
|
|
|
|
if (isolate_lru_page(page))
|
|
|
|
goto put;
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
parent = mem_cgroup_from_cont(pcg);
|
2010-03-10 23:22:30 +00:00
|
|
|
ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false);
|
2009-01-08 02:08:08 +00:00
|
|
|
if (ret || !parent)
|
2009-12-16 00:47:11 +00:00
|
|
|
goto put_back;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
|
2010-03-10 23:22:15 +00:00
|
|
|
ret = mem_cgroup_move_account(pc, child, parent, true);
|
|
|
|
if (ret)
|
|
|
|
mem_cgroup_cancel_charge(parent);
|
2009-12-16 00:47:11 +00:00
|
|
|
put_back:
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
putback_lru_page(page);
|
2009-12-16 00:47:11 +00:00
|
|
|
put:
|
2009-01-15 21:51:12 +00:00
|
|
|
put_page(page);
|
2009-12-16 00:47:11 +00:00
|
|
|
out:
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:07:48 +00:00
|
|
|
/*
|
|
|
|
* Charge the memory controller for page usage.
|
|
|
|
* Return
|
|
|
|
* 0 if the charge was successful
|
|
|
|
* < 0 if the cgroup is over its limit
|
|
|
|
*/
|
|
|
|
static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
|
2010-08-11 01:02:59 +00:00
|
|
|
gfp_t gfp_mask, enum charge_type ctype)
|
2009-01-08 02:07:48 +00:00
|
|
|
{
|
2010-08-11 01:02:59 +00:00
|
|
|
struct mem_cgroup *mem = NULL;
|
2009-01-08 02:07:48 +00:00
|
|
|
struct page_cgroup *pc;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
pc = lookup_page_cgroup(page);
|
|
|
|
/* can happen at boot */
|
|
|
|
if (unlikely(!pc))
|
|
|
|
return 0;
|
|
|
|
prefetchw(pc);
|
|
|
|
|
2010-03-10 23:22:30 +00:00
|
|
|
ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true);
|
2009-01-08 02:08:08 +00:00
|
|
|
if (ret || !mem)
|
2009-01-08 02:07:48 +00:00
|
|
|
return ret;
|
|
|
|
|
|
|
|
__mem_cgroup_commit_charge(mem, pc, ctype);
|
2008-02-07 08:13:53 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:07:48 +00:00
|
|
|
int mem_cgroup_newpage_charge(struct page *page,
|
|
|
|
struct mm_struct *mm, gfp_t gfp_mask)
|
2008-02-07 08:14:17 +00:00
|
|
|
{
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2008-07-25 08:47:18 +00:00
|
|
|
return 0;
|
2008-10-19 03:28:16 +00:00
|
|
|
if (PageCompound(page))
|
|
|
|
return 0;
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
/*
|
|
|
|
* If already mapped, we don't have to account.
|
|
|
|
* If page cache, page->mapping has address_space.
|
|
|
|
* But page->mapping may have out-of-use anon_vma pointer,
|
|
|
|
* detecit it by PageAnon() check. newly-mapped-anon's page->mapping
|
|
|
|
* is NULL.
|
|
|
|
*/
|
|
|
|
if (page_mapped(page) || (page->mapping && !PageAnon(page)))
|
|
|
|
return 0;
|
|
|
|
if (unlikely(!mm))
|
|
|
|
mm = &init_mm;
|
2008-02-07 08:14:17 +00:00
|
|
|
return mem_cgroup_charge_common(page, mm, gfp_mask,
|
2010-08-11 01:02:59 +00:00
|
|
|
MEM_CGROUP_CHARGE_TYPE_MAPPED);
|
2008-02-07 08:14:17 +00:00
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:48 +00:00
|
|
|
static void
|
|
|
|
__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
|
|
|
|
enum charge_type ctype);
|
|
|
|
|
2008-02-07 08:14:02 +00:00
|
|
|
int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
|
|
|
|
gfp_t gfp_mask)
|
2008-02-07 08:13:59 +00:00
|
|
|
{
|
2009-01-08 02:08:35 +00:00
|
|
|
int ret;
|
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2008-07-25 08:47:18 +00:00
|
|
|
return 0;
|
2008-10-19 03:28:16 +00:00
|
|
|
if (PageCompound(page))
|
|
|
|
return 0;
|
2008-07-25 08:47:17 +00:00
|
|
|
/*
|
|
|
|
* Corner case handling. This is called from add_to_page_cache()
|
|
|
|
* in usual. But some FS (shmem) precharges this page before calling it
|
|
|
|
* and call add_to_page_cache() with GFP_NOWAIT.
|
|
|
|
*
|
|
|
|
* For GFP_NOWAIT case, the page may be pre-charged before calling
|
|
|
|
* add_to_page_cache(). (See shmem.c) check it here and avoid to call
|
|
|
|
* charge twice. (It works but has to pay a bit larger cost.)
|
2009-01-08 02:08:35 +00:00
|
|
|
* And when the page is SwapCache, it should take swap information
|
|
|
|
* into account. This is under lock_page() now.
|
2008-07-25 08:47:17 +00:00
|
|
|
*/
|
|
|
|
if (!(gfp_mask & __GFP_WAIT)) {
|
|
|
|
struct page_cgroup *pc;
|
|
|
|
|
2008-10-19 03:28:16 +00:00
|
|
|
pc = lookup_page_cgroup(page);
|
|
|
|
if (!pc)
|
|
|
|
return 0;
|
|
|
|
lock_page_cgroup(pc);
|
|
|
|
if (PageCgroupUsed(pc)) {
|
|
|
|
unlock_page_cgroup(pc);
|
2008-07-25 08:47:17 +00:00
|
|
|
return 0;
|
|
|
|
}
|
2008-10-19 03:28:16 +00:00
|
|
|
unlock_page_cgroup(pc);
|
2008-07-25 08:47:17 +00:00
|
|
|
}
|
|
|
|
|
2010-08-11 01:02:59 +00:00
|
|
|
if (unlikely(!mm))
|
2008-02-07 08:13:59 +00:00
|
|
|
mm = &init_mm;
|
2008-07-25 08:47:17 +00:00
|
|
|
|
2008-10-19 03:28:11 +00:00
|
|
|
if (page_is_file_cache(page))
|
|
|
|
return mem_cgroup_charge_common(page, mm, gfp_mask,
|
2010-08-11 01:02:59 +00:00
|
|
|
MEM_CGROUP_CHARGE_TYPE_CACHE);
|
2009-01-08 02:08:35 +00:00
|
|
|
|
2009-04-02 23:57:48 +00:00
|
|
|
/* shmem */
|
|
|
|
if (PageSwapCache(page)) {
|
2010-08-11 01:02:59 +00:00
|
|
|
struct mem_cgroup *mem = NULL;
|
|
|
|
|
2009-04-02 23:57:48 +00:00
|
|
|
ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
|
|
|
|
if (!ret)
|
|
|
|
__mem_cgroup_commit_charge_swapin(page, mem,
|
|
|
|
MEM_CGROUP_CHARGE_TYPE_SHMEM);
|
|
|
|
} else
|
|
|
|
ret = mem_cgroup_charge_common(page, mm, gfp_mask,
|
2010-08-11 01:02:59 +00:00
|
|
|
MEM_CGROUP_CHARGE_TYPE_SHMEM);
|
2009-01-08 02:08:35 +00:00
|
|
|
|
|
|
|
return ret;
|
2008-07-25 08:47:10 +00:00
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:33 +00:00
|
|
|
/*
|
|
|
|
* While swap-in, try_charge -> commit or cancel, the page is locked.
|
|
|
|
* And when try_charge() successfully returns, one refcnt to memcg without
|
2009-10-07 13:21:09 +00:00
|
|
|
* struct page_cgroup is acquired. This refcnt will be consumed by
|
2009-01-08 02:08:33 +00:00
|
|
|
* "commit()" or removed by "cancel()"
|
|
|
|
*/
|
2009-01-08 02:08:00 +00:00
|
|
|
int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
|
|
|
|
struct page *page,
|
|
|
|
gfp_t mask, struct mem_cgroup **ptr)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem;
|
2009-01-08 02:08:33 +00:00
|
|
|
int ret;
|
2009-01-08 02:08:00 +00:00
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2009-01-08 02:08:00 +00:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!do_swap_account)
|
|
|
|
goto charge_cur_mm;
|
|
|
|
/*
|
|
|
|
* A racing thread's fault, or swapoff, may have already updated
|
2009-12-15 01:59:30 +00:00
|
|
|
* the pte, and even removed page from swap cache: in those cases
|
|
|
|
* do_swap_page()'s pte_same() test will fail; but there's also a
|
|
|
|
* KSM case which does need to charge the page.
|
2009-01-08 02:08:00 +00:00
|
|
|
*/
|
|
|
|
if (!PageSwapCache(page))
|
2009-12-15 01:59:30 +00:00
|
|
|
goto charge_cur_mm;
|
2009-12-16 11:19:59 +00:00
|
|
|
mem = try_get_mem_cgroup_from_page(page);
|
2009-01-08 02:08:33 +00:00
|
|
|
if (!mem)
|
|
|
|
goto charge_cur_mm;
|
2009-01-08 02:08:00 +00:00
|
|
|
*ptr = mem;
|
2010-03-10 23:22:30 +00:00
|
|
|
ret = __mem_cgroup_try_charge(NULL, mask, ptr, true);
|
2009-01-08 02:08:33 +00:00
|
|
|
css_put(&mem->css);
|
|
|
|
return ret;
|
2009-01-08 02:08:00 +00:00
|
|
|
charge_cur_mm:
|
|
|
|
if (unlikely(!mm))
|
|
|
|
mm = &init_mm;
|
2010-03-10 23:22:30 +00:00
|
|
|
return __mem_cgroup_try_charge(mm, mask, ptr, true);
|
2009-01-08 02:08:00 +00:00
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:48 +00:00
|
|
|
static void
|
|
|
|
__mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
|
|
|
|
enum charge_type ctype)
|
2009-01-08 02:07:48 +00:00
|
|
|
{
|
|
|
|
struct page_cgroup *pc;
|
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2009-01-08 02:07:48 +00:00
|
|
|
return;
|
|
|
|
if (!ptr)
|
|
|
|
return;
|
2009-07-29 22:04:06 +00:00
|
|
|
cgroup_exclude_rmdir(&ptr->css);
|
2009-01-08 02:07:48 +00:00
|
|
|
pc = lookup_page_cgroup(page);
|
2009-01-08 02:08:34 +00:00
|
|
|
mem_cgroup_lru_del_before_commit_swapcache(page);
|
2009-04-02 23:57:48 +00:00
|
|
|
__mem_cgroup_commit_charge(ptr, pc, ctype);
|
2009-01-08 02:08:34 +00:00
|
|
|
mem_cgroup_lru_add_after_commit_swapcache(page);
|
2009-01-08 02:08:00 +00:00
|
|
|
/*
|
|
|
|
* Now swap is on-memory. This means this page may be
|
|
|
|
* counted both as mem and swap....double count.
|
2009-01-08 02:08:31 +00:00
|
|
|
* Fix it by uncharging from memsw. Basically, this SwapCache is stable
|
|
|
|
* under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
|
|
|
|
* may call delete_from_swap_cache() before reach here.
|
2009-01-08 02:08:00 +00:00
|
|
|
*/
|
2009-01-08 02:08:31 +00:00
|
|
|
if (do_swap_account && PageSwapCache(page)) {
|
2009-01-08 02:08:00 +00:00
|
|
|
swp_entry_t ent = {.val = page_private(page)};
|
2009-04-02 23:57:45 +00:00
|
|
|
unsigned short id;
|
2009-01-08 02:08:00 +00:00
|
|
|
struct mem_cgroup *memcg;
|
2009-04-02 23:57:45 +00:00
|
|
|
|
|
|
|
id = swap_cgroup_record(ent, 0);
|
|
|
|
rcu_read_lock();
|
|
|
|
memcg = mem_cgroup_lookup(id);
|
2009-01-08 02:08:00 +00:00
|
|
|
if (memcg) {
|
2009-04-02 23:57:45 +00:00
|
|
|
/*
|
|
|
|
* This recorded memcg can be obsolete one. So, avoid
|
|
|
|
* calling css_tryget
|
|
|
|
*/
|
2009-09-23 22:56:42 +00:00
|
|
|
if (!mem_cgroup_is_root(memcg))
|
2009-10-01 22:44:11 +00:00
|
|
|
res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
|
2009-09-23 22:56:42 +00:00
|
|
|
mem_cgroup_swap_statistics(memcg, false);
|
2009-01-08 02:08:00 +00:00
|
|
|
mem_cgroup_put(memcg);
|
|
|
|
}
|
2009-04-02 23:57:45 +00:00
|
|
|
rcu_read_unlock();
|
2009-01-08 02:08:00 +00:00
|
|
|
}
|
2009-07-29 22:04:06 +00:00
|
|
|
/*
|
|
|
|
* At swapin, we may charge account against cgroup which has no tasks.
|
|
|
|
* So, rmdir()->pre_destroy() can be called while we do this charge.
|
|
|
|
* In that case, we need to call pre_destroy() again. check it here.
|
|
|
|
*/
|
|
|
|
cgroup_release_and_wakeup_rmdir(&ptr->css);
|
2009-01-08 02:07:48 +00:00
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:48 +00:00
|
|
|
void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr)
|
|
|
|
{
|
|
|
|
__mem_cgroup_commit_charge_swapin(page, ptr,
|
|
|
|
MEM_CGROUP_CHARGE_TYPE_MAPPED);
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:07:48 +00:00
|
|
|
void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
|
|
|
|
{
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2009-01-08 02:07:48 +00:00
|
|
|
return;
|
|
|
|
if (!mem)
|
|
|
|
return;
|
2009-12-16 00:47:10 +00:00
|
|
|
mem_cgroup_cancel_charge(mem);
|
2009-01-08 02:07:48 +00:00
|
|
|
}
|
|
|
|
|
2009-12-16 00:47:03 +00:00
|
|
|
static void
|
|
|
|
__do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
|
|
|
|
{
|
|
|
|
struct memcg_batch_info *batch = NULL;
|
|
|
|
bool uncharge_memsw = true;
|
|
|
|
/* If swapout, usage of swap doesn't decrease */
|
|
|
|
if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
|
|
|
|
uncharge_memsw = false;
|
|
|
|
|
|
|
|
batch = ¤t->memcg_batch;
|
|
|
|
/*
|
|
|
|
* In usual, we do css_get() when we remember memcg pointer.
|
|
|
|
* But in this case, we keep res->usage until end of a series of
|
|
|
|
* uncharges. Then, it's ok to ignore memcg's refcnt.
|
|
|
|
*/
|
|
|
|
if (!batch->memcg)
|
|
|
|
batch->memcg = mem;
|
2010-05-26 21:42:37 +00:00
|
|
|
/*
|
|
|
|
* do_batch > 0 when unmapping pages or inode invalidate/truncate.
|
|
|
|
* In those cases, all pages freed continously can be expected to be in
|
|
|
|
* the same cgroup and we have chance to coalesce uncharges.
|
|
|
|
* But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
|
|
|
|
* because we want to do uncharge as soon as possible.
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
|
|
|
|
goto direct_uncharge;
|
|
|
|
|
2009-12-16 00:47:03 +00:00
|
|
|
/*
|
|
|
|
* In typical case, batch->memcg == mem. This means we can
|
|
|
|
* merge a series of uncharges to an uncharge of res_counter.
|
|
|
|
* If not, we uncharge res_counter ony by one.
|
|
|
|
*/
|
|
|
|
if (batch->memcg != mem)
|
|
|
|
goto direct_uncharge;
|
|
|
|
/* remember freed charge and uncharge it later */
|
|
|
|
batch->bytes += PAGE_SIZE;
|
|
|
|
if (uncharge_memsw)
|
|
|
|
batch->memsw_bytes += PAGE_SIZE;
|
|
|
|
return;
|
|
|
|
direct_uncharge:
|
|
|
|
res_counter_uncharge(&mem->res, PAGE_SIZE);
|
|
|
|
if (uncharge_memsw)
|
|
|
|
res_counter_uncharge(&mem->memsw, PAGE_SIZE);
|
2010-05-26 21:42:37 +00:00
|
|
|
if (unlikely(batch->memcg != mem))
|
|
|
|
memcg_oom_recover(mem);
|
2009-12-16 00:47:03 +00:00
|
|
|
return;
|
|
|
|
}
|
2009-01-08 02:07:48 +00:00
|
|
|
|
2008-02-07 08:13:53 +00:00
|
|
|
/*
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
* uncharge if !page_mapped(page)
|
2008-02-07 08:13:53 +00:00
|
|
|
*/
|
2009-01-08 02:08:00 +00:00
|
|
|
static struct mem_cgroup *
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
__mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
|
2008-02-07 08:13:53 +00:00
|
|
|
{
|
2008-03-04 22:29:08 +00:00
|
|
|
struct page_cgroup *pc;
|
2009-01-08 02:08:00 +00:00
|
|
|
struct mem_cgroup *mem = NULL;
|
2008-02-07 08:13:53 +00:00
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2009-01-08 02:08:00 +00:00
|
|
|
return NULL;
|
2008-04-04 21:29:59 +00:00
|
|
|
|
2009-01-08 02:07:56 +00:00
|
|
|
if (PageSwapCache(page))
|
2009-01-08 02:08:00 +00:00
|
|
|
return NULL;
|
2009-01-08 02:07:56 +00:00
|
|
|
|
2008-02-07 08:13:59 +00:00
|
|
|
/*
|
2008-02-07 08:14:41 +00:00
|
|
|
* Check if our page_cgroup is valid
|
2008-02-07 08:13:59 +00:00
|
|
|
*/
|
2008-10-19 03:28:16 +00:00
|
|
|
pc = lookup_page_cgroup(page);
|
|
|
|
if (unlikely(!pc || !PageCgroupUsed(pc)))
|
2009-01-08 02:08:00 +00:00
|
|
|
return NULL;
|
2008-03-04 22:29:11 +00:00
|
|
|
|
2008-10-19 03:28:16 +00:00
|
|
|
lock_page_cgroup(pc);
|
2009-01-08 02:07:56 +00:00
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
mem = pc->mem_cgroup;
|
|
|
|
|
2009-01-08 02:07:56 +00:00
|
|
|
if (!PageCgroupUsed(pc))
|
|
|
|
goto unlock_out;
|
|
|
|
|
|
|
|
switch (ctype) {
|
|
|
|
case MEM_CGROUP_CHARGE_TYPE_MAPPED:
|
2009-06-17 23:27:17 +00:00
|
|
|
case MEM_CGROUP_CHARGE_TYPE_DROP:
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
/* See mem_cgroup_prepare_migration() */
|
|
|
|
if (page_mapped(page) || PageCgroupMigration(pc))
|
2009-01-08 02:07:56 +00:00
|
|
|
goto unlock_out;
|
|
|
|
break;
|
|
|
|
case MEM_CGROUP_CHARGE_TYPE_SWAPOUT:
|
|
|
|
if (!PageAnon(page)) { /* Shared memory */
|
|
|
|
if (page->mapping && !page_is_file_cache(page))
|
|
|
|
goto unlock_out;
|
|
|
|
} else if (page_mapped(page)) /* Anon */
|
|
|
|
goto unlock_out;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
2008-10-19 03:28:16 +00:00
|
|
|
}
|
2009-01-08 02:07:56 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
mem_cgroup_charge_statistics(mem, pc, false);
|
2009-04-02 23:57:33 +00:00
|
|
|
|
2008-10-19 03:28:16 +00:00
|
|
|
ClearPageCgroupUsed(pc);
|
2009-01-08 02:08:34 +00:00
|
|
|
/*
|
|
|
|
* pc->mem_cgroup is not cleared here. It will be accessed when it's
|
|
|
|
* freed from LRU. This is safe because uncharged page is expected not
|
|
|
|
* to be reused (freed soon). Exception is SwapCache, it's handled by
|
|
|
|
* special functions.
|
|
|
|
*/
|
2008-03-04 22:29:11 +00:00
|
|
|
|
2008-10-19 03:28:16 +00:00
|
|
|
unlock_page_cgroup(pc);
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
/*
|
|
|
|
* even after unlock, we have mem->res.usage here and this memcg
|
|
|
|
* will never be freed.
|
|
|
|
*/
|
2010-03-10 23:22:31 +00:00
|
|
|
memcg_check_events(mem, page);
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) {
|
|
|
|
mem_cgroup_swap_statistics(mem, true);
|
|
|
|
mem_cgroup_get(mem);
|
|
|
|
}
|
|
|
|
if (!mem_cgroup_is_root(mem))
|
|
|
|
__do_uncharge(mem, ctype);
|
2008-02-07 08:14:31 +00:00
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
return mem;
|
2009-01-08 02:07:56 +00:00
|
|
|
|
|
|
|
unlock_out:
|
|
|
|
unlock_page_cgroup(pc);
|
2009-01-08 02:08:00 +00:00
|
|
|
return NULL;
|
2008-02-07 08:14:41 +00:00
|
|
|
}
|
|
|
|
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
void mem_cgroup_uncharge_page(struct page *page)
|
|
|
|
{
|
2008-10-19 03:28:16 +00:00
|
|
|
/* early check. */
|
|
|
|
if (page_mapped(page))
|
|
|
|
return;
|
|
|
|
if (page->mapping && !PageAnon(page))
|
|
|
|
return;
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
|
|
|
|
}
|
|
|
|
|
|
|
|
void mem_cgroup_uncharge_cache_page(struct page *page)
|
|
|
|
{
|
|
|
|
VM_BUG_ON(page_mapped(page));
|
2008-10-19 03:28:09 +00:00
|
|
|
VM_BUG_ON(page->mapping);
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
|
|
|
|
}
|
|
|
|
|
2009-12-16 00:47:03 +00:00
|
|
|
/*
|
|
|
|
* Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
|
|
|
|
* In that cases, pages are freed continuously and we can expect pages
|
|
|
|
* are in the same memcg. All these calls itself limits the number of
|
|
|
|
* pages freed at once, then uncharge_start/end() is called properly.
|
|
|
|
* This may be called prural(2) times in a context,
|
|
|
|
*/
|
|
|
|
|
|
|
|
void mem_cgroup_uncharge_start(void)
|
|
|
|
{
|
|
|
|
current->memcg_batch.do_batch++;
|
|
|
|
/* We can do nest. */
|
|
|
|
if (current->memcg_batch.do_batch == 1) {
|
|
|
|
current->memcg_batch.memcg = NULL;
|
|
|
|
current->memcg_batch.bytes = 0;
|
|
|
|
current->memcg_batch.memsw_bytes = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void mem_cgroup_uncharge_end(void)
|
|
|
|
{
|
|
|
|
struct memcg_batch_info *batch = ¤t->memcg_batch;
|
|
|
|
|
|
|
|
if (!batch->do_batch)
|
|
|
|
return;
|
|
|
|
|
|
|
|
batch->do_batch--;
|
|
|
|
if (batch->do_batch) /* If stacked, do nothing. */
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (!batch->memcg)
|
|
|
|
return;
|
|
|
|
/*
|
|
|
|
* This "batch->memcg" is valid without any css_get/put etc...
|
|
|
|
* bacause we hide charges behind us.
|
|
|
|
*/
|
|
|
|
if (batch->bytes)
|
|
|
|
res_counter_uncharge(&batch->memcg->res, batch->bytes);
|
|
|
|
if (batch->memsw_bytes)
|
|
|
|
res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
|
2010-05-26 21:42:37 +00:00
|
|
|
memcg_oom_recover(batch->memcg);
|
2009-12-16 00:47:03 +00:00
|
|
|
/* forget this pointer (for sanity check) */
|
|
|
|
batch->memcg = NULL;
|
|
|
|
}
|
|
|
|
|
2009-05-28 21:34:28 +00:00
|
|
|
#ifdef CONFIG_SWAP
|
2009-01-08 02:08:00 +00:00
|
|
|
/*
|
2009-05-28 21:34:28 +00:00
|
|
|
* called after __delete_from_swap_cache() and drop "page" account.
|
2009-01-08 02:08:00 +00:00
|
|
|
* memcg information is recorded to swap_cgroup of "ent"
|
|
|
|
*/
|
2009-06-17 23:27:17 +00:00
|
|
|
void
|
|
|
|
mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
|
2009-01-08 02:08:00 +00:00
|
|
|
{
|
|
|
|
struct mem_cgroup *memcg;
|
2009-06-17 23:27:17 +00:00
|
|
|
int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT;
|
|
|
|
|
|
|
|
if (!swapout) /* this was a swap cache but the swap is unused ! */
|
|
|
|
ctype = MEM_CGROUP_CHARGE_TYPE_DROP;
|
|
|
|
|
|
|
|
memcg = __mem_cgroup_uncharge_common(page, ctype);
|
2009-01-08 02:08:00 +00:00
|
|
|
|
memcg: avoid css_get()
Now, memory cgroup increments css(cgroup subsys state)'s reference count
per a charged page. And the reference count is kept until the page is
uncharged. But this has 2 bad effect.
1. Because css_get/put calls atomic_inc()/dec, heavy call of them
on large smp will not scale well.
2. Because css's refcnt cannot be in a state as "ready-to-release",
cgroup's notify_on_release handler can't work with memcg.
3. css's refcnt is atomic_t, it means smaller than 32bit. Maybe too small.
This has been a problem since the 1st merge of memcg.
This is a trial to remove css's refcnt per a page. Even if we remove
refcnt, pre_destroy() does enough synchronization as
- check res->usage == 0.
- check no pages on LRU.
This patch removes css's refcnt per page. Even after this patch, at the
1st look, it seems css_get() is still called in try_charge().
But the logic is.
- If a memcg of mm->owner is cached one, consume_stock() will work.
At success, return immediately.
- If consume_stock returns false, css_get() is called and go to
slow path which may be blocked. At the end of slow path,
css_put() is called and restart from the start if necessary.
So, in the fast path, we don't call css_get() and can avoid access to
shared counter. This patch can make the most possible case fast.
Here is a result of multi-threaded page fault benchmark.
[Before]
25.32% multi-fault-all [kernel.kallsyms] [k] clear_page_c
9.30% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
8.02% multi-fault-all [kernel.kallsyms] [k] try_get_mem_cgroup_from_mm <=====(*)
7.83% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
5.38% multi-fault-all [kernel.kallsyms] [k] __css_put
5.29% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
4.92% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
4.24% multi-fault-all [kernel.kallsyms] [k] up_read
3.53% multi-fault-all [kernel.kallsyms] [k] css_put
2.11% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
1.76% multi-fault-all [kernel.kallsyms] [k] __rmqueue
1.64% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
[After]
28.41% multi-fault-all [kernel.kallsyms] [k] clear_page_c
10.08% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irq
9.58% multi-fault-all [kernel.kallsyms] [k] down_read_trylock
9.38% multi-fault-all [kernel.kallsyms] [k] _raw_spin_lock_irqsave
5.86% multi-fault-all [kernel.kallsyms] [k] __alloc_pages_nodemask
5.65% multi-fault-all [kernel.kallsyms] [k] up_read
2.82% multi-fault-all [kernel.kallsyms] [k] handle_mm_fault
2.64% multi-fault-all [kernel.kallsyms] [k] mem_cgroup_add_lru_list
2.48% multi-fault-all [kernel.kallsyms] [k] __mem_cgroup_commit_charge
Then, 8.02% of try_get_mem_cgroup_from_mm() disappears because this patch
removes css_tryget() in it. (But yes, this is an extreme case.)
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-08-11 01:03:02 +00:00
|
|
|
/*
|
|
|
|
* record memcg information, if swapout && memcg != NULL,
|
|
|
|
* mem_cgroup_get() was called in uncharge().
|
|
|
|
*/
|
|
|
|
if (do_swap_account && swapout && memcg)
|
2009-04-02 23:57:45 +00:00
|
|
|
swap_cgroup_record(ent, css_id(&memcg->css));
|
2009-01-08 02:08:00 +00:00
|
|
|
}
|
2009-05-28 21:34:28 +00:00
|
|
|
#endif
|
2009-01-08 02:08:00 +00:00
|
|
|
|
|
|
|
#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
|
|
|
|
/*
|
|
|
|
* called from swap_entry_free(). remove record in swap_cgroup and
|
|
|
|
* uncharge "memsw" account.
|
|
|
|
*/
|
|
|
|
void mem_cgroup_uncharge_swap(swp_entry_t ent)
|
2009-01-08 02:07:56 +00:00
|
|
|
{
|
2009-01-08 02:08:00 +00:00
|
|
|
struct mem_cgroup *memcg;
|
2009-04-02 23:57:45 +00:00
|
|
|
unsigned short id;
|
2009-01-08 02:08:00 +00:00
|
|
|
|
|
|
|
if (!do_swap_account)
|
|
|
|
return;
|
|
|
|
|
2009-04-02 23:57:45 +00:00
|
|
|
id = swap_cgroup_record(ent, 0);
|
|
|
|
rcu_read_lock();
|
|
|
|
memcg = mem_cgroup_lookup(id);
|
2009-01-08 02:08:00 +00:00
|
|
|
if (memcg) {
|
2009-04-02 23:57:45 +00:00
|
|
|
/*
|
|
|
|
* We uncharge this because swap is freed.
|
|
|
|
* This memcg can be obsolete one. We avoid calling css_tryget
|
|
|
|
*/
|
2009-09-23 22:56:42 +00:00
|
|
|
if (!mem_cgroup_is_root(memcg))
|
2009-10-01 22:44:11 +00:00
|
|
|
res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
|
2009-09-23 22:56:42 +00:00
|
|
|
mem_cgroup_swap_statistics(memcg, false);
|
2009-01-08 02:08:00 +00:00
|
|
|
mem_cgroup_put(memcg);
|
|
|
|
}
|
2009-04-02 23:57:45 +00:00
|
|
|
rcu_read_unlock();
|
2009-01-08 02:07:56 +00:00
|
|
|
}
|
2010-03-10 23:22:17 +00:00
|
|
|
|
|
|
|
/**
|
|
|
|
* mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
|
|
|
|
* @entry: swap entry to be moved
|
|
|
|
* @from: mem_cgroup which the entry is moved from
|
|
|
|
* @to: mem_cgroup which the entry is moved to
|
2010-03-10 23:22:18 +00:00
|
|
|
* @need_fixup: whether we should fixup res_counters and refcounts.
|
2010-03-10 23:22:17 +00:00
|
|
|
*
|
|
|
|
* It succeeds only when the swap_cgroup's record for this entry is the same
|
|
|
|
* as the mem_cgroup's id of @from.
|
|
|
|
*
|
|
|
|
* Returns 0 on success, -EINVAL on failure.
|
|
|
|
*
|
|
|
|
* The caller must have charged to @to, IOW, called res_counter_charge() about
|
|
|
|
* both res and memsw, and called css_get().
|
|
|
|
*/
|
|
|
|
static int mem_cgroup_move_swap_account(swp_entry_t entry,
|
2010-03-10 23:22:18 +00:00
|
|
|
struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
|
2010-03-10 23:22:17 +00:00
|
|
|
{
|
|
|
|
unsigned short old_id, new_id;
|
|
|
|
|
|
|
|
old_id = css_id(&from->css);
|
|
|
|
new_id = css_id(&to->css);
|
|
|
|
|
|
|
|
if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
|
|
|
|
mem_cgroup_swap_statistics(from, false);
|
2010-03-10 23:22:18 +00:00
|
|
|
mem_cgroup_swap_statistics(to, true);
|
2010-03-10 23:22:17 +00:00
|
|
|
/*
|
2010-03-10 23:22:18 +00:00
|
|
|
* This function is only called from task migration context now.
|
|
|
|
* It postpones res_counter and refcount handling till the end
|
|
|
|
* of task migration(mem_cgroup_clear_mc()) for performance
|
|
|
|
* improvement. But we cannot postpone mem_cgroup_get(to)
|
|
|
|
* because if the process that has been moved to @to does
|
|
|
|
* swap-in, the refcount of @to might be decreased to 0.
|
2010-03-10 23:22:17 +00:00
|
|
|
*/
|
|
|
|
mem_cgroup_get(to);
|
2010-03-10 23:22:18 +00:00
|
|
|
if (need_fixup) {
|
|
|
|
if (!mem_cgroup_is_root(from))
|
|
|
|
res_counter_uncharge(&from->memsw, PAGE_SIZE);
|
|
|
|
mem_cgroup_put(from);
|
|
|
|
/*
|
|
|
|
* we charged both to->res and to->memsw, so we should
|
|
|
|
* uncharge to->res.
|
|
|
|
*/
|
|
|
|
if (!mem_cgroup_is_root(to))
|
|
|
|
res_counter_uncharge(&to->res, PAGE_SIZE);
|
|
|
|
}
|
2010-03-10 23:22:17 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
|
2010-03-10 23:22:18 +00:00
|
|
|
struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup)
|
2010-03-10 23:22:17 +00:00
|
|
|
{
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
2009-01-08 02:08:00 +00:00
|
|
|
#endif
|
2009-01-08 02:07:56 +00:00
|
|
|
|
2008-02-07 08:14:10 +00:00
|
|
|
/*
|
2009-01-08 02:07:50 +00:00
|
|
|
* Before starting migration, account PAGE_SIZE to mem_cgroup that the old
|
|
|
|
* page belongs to.
|
2008-02-07 08:14:10 +00:00
|
|
|
*/
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
int mem_cgroup_prepare_migration(struct page *page,
|
|
|
|
struct page *newpage, struct mem_cgroup **ptr)
|
2008-02-07 08:14:10 +00:00
|
|
|
{
|
|
|
|
struct page_cgroup *pc;
|
2008-07-25 08:47:10 +00:00
|
|
|
struct mem_cgroup *mem = NULL;
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
enum charge_type ctype;
|
2008-07-25 08:47:10 +00:00
|
|
|
int ret = 0;
|
2008-03-04 22:29:09 +00:00
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2008-04-04 21:29:59 +00:00
|
|
|
return 0;
|
|
|
|
|
2008-10-19 03:28:16 +00:00
|
|
|
pc = lookup_page_cgroup(page);
|
|
|
|
lock_page_cgroup(pc);
|
|
|
|
if (PageCgroupUsed(pc)) {
|
2008-07-25 08:47:10 +00:00
|
|
|
mem = pc->mem_cgroup;
|
|
|
|
css_get(&mem->css);
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
/*
|
|
|
|
* At migrating an anonymous page, its mapcount goes down
|
|
|
|
* to 0 and uncharge() will be called. But, even if it's fully
|
|
|
|
* unmapped, migration may fail and this page has to be
|
|
|
|
* charged again. We set MIGRATION flag here and delay uncharge
|
|
|
|
* until end_migration() is called
|
|
|
|
*
|
|
|
|
* Corner Case Thinking
|
|
|
|
* A)
|
|
|
|
* When the old page was mapped as Anon and it's unmap-and-freed
|
|
|
|
* while migration was ongoing.
|
|
|
|
* If unmap finds the old page, uncharge() of it will be delayed
|
|
|
|
* until end_migration(). If unmap finds a new page, it's
|
|
|
|
* uncharged when it make mapcount to be 1->0. If unmap code
|
|
|
|
* finds swap_migration_entry, the new page will not be mapped
|
|
|
|
* and end_migration() will find it(mapcount==0).
|
|
|
|
*
|
|
|
|
* B)
|
|
|
|
* When the old page was mapped but migraion fails, the kernel
|
|
|
|
* remaps it. A charge for it is kept by MIGRATION flag even
|
|
|
|
* if mapcount goes down to 0. We can do remap successfully
|
|
|
|
* without charging it again.
|
|
|
|
*
|
|
|
|
* C)
|
|
|
|
* The "old" page is under lock_page() until the end of
|
|
|
|
* migration, so, the old page itself will not be swapped-out.
|
|
|
|
* If the new page is swapped out before end_migraton, our
|
|
|
|
* hook to usual swap-out path will catch the event.
|
|
|
|
*/
|
|
|
|
if (PageAnon(page))
|
|
|
|
SetPageCgroupMigration(pc);
|
2008-07-25 08:47:10 +00:00
|
|
|
}
|
2008-10-19 03:28:16 +00:00
|
|
|
unlock_page_cgroup(pc);
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
/*
|
|
|
|
* If the page is not charged at this point,
|
|
|
|
* we return here.
|
|
|
|
*/
|
|
|
|
if (!mem)
|
|
|
|
return 0;
|
2009-01-08 02:07:50 +00:00
|
|
|
|
2010-04-23 17:17:39 +00:00
|
|
|
*ptr = mem;
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false);
|
|
|
|
css_put(&mem->css);/* drop extra refcnt */
|
|
|
|
if (ret || *ptr == NULL) {
|
|
|
|
if (PageAnon(page)) {
|
|
|
|
lock_page_cgroup(pc);
|
|
|
|
ClearPageCgroupMigration(pc);
|
|
|
|
unlock_page_cgroup(pc);
|
|
|
|
/*
|
|
|
|
* The old page may be fully unmapped while we kept it.
|
|
|
|
*/
|
|
|
|
mem_cgroup_uncharge_page(page);
|
|
|
|
}
|
|
|
|
return -ENOMEM;
|
2008-07-25 08:47:10 +00:00
|
|
|
}
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
/*
|
|
|
|
* We charge new page before it's used/mapped. So, even if unlock_page()
|
|
|
|
* is called before end_migration, we can catch all events on this new
|
|
|
|
* page. In the case new page is migrated but not remapped, new page's
|
|
|
|
* mapcount will be finally 0 and we call uncharge in end_migration().
|
|
|
|
*/
|
|
|
|
pc = lookup_page_cgroup(newpage);
|
|
|
|
if (PageAnon(page))
|
|
|
|
ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
|
|
|
|
else if (page_is_file_cache(page))
|
|
|
|
ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
|
|
|
|
else
|
|
|
|
ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
|
|
|
|
__mem_cgroup_commit_charge(mem, pc, ctype);
|
2008-07-25 08:47:10 +00:00
|
|
|
return ret;
|
2008-02-07 08:14:10 +00:00
|
|
|
}
|
2008-03-04 22:29:09 +00:00
|
|
|
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
/* remove redundant charge if migration failed*/
|
2009-01-08 02:07:50 +00:00
|
|
|
void mem_cgroup_end_migration(struct mem_cgroup *mem,
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
struct page *oldpage, struct page *newpage)
|
2008-02-07 08:14:10 +00:00
|
|
|
{
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
struct page *used, *unused;
|
2009-01-08 02:07:50 +00:00
|
|
|
struct page_cgroup *pc;
|
|
|
|
|
|
|
|
if (!mem)
|
|
|
|
return;
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
/* blocks rmdir() */
|
2009-07-29 22:04:06 +00:00
|
|
|
cgroup_exclude_rmdir(&mem->css);
|
2009-01-08 02:07:50 +00:00
|
|
|
/* at migration success, oldpage->mapping is NULL. */
|
|
|
|
if (oldpage->mapping) {
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
used = oldpage;
|
|
|
|
unused = newpage;
|
2009-01-08 02:07:50 +00:00
|
|
|
} else {
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
used = newpage;
|
2009-01-08 02:07:50 +00:00
|
|
|
unused = oldpage;
|
|
|
|
}
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
/*
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
* We disallowed uncharge of pages under migration because mapcount
|
|
|
|
* of the page goes down to zero, temporarly.
|
|
|
|
* Clear the flag and check the page should be charged.
|
2009-01-08 02:07:50 +00:00
|
|
|
*/
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
pc = lookup_page_cgroup(oldpage);
|
|
|
|
lock_page_cgroup(pc);
|
|
|
|
ClearPageCgroupMigration(pc);
|
|
|
|
unlock_page_cgroup(pc);
|
2009-01-08 02:07:50 +00:00
|
|
|
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
__mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE);
|
|
|
|
|
2009-01-08 02:07:50 +00:00
|
|
|
/*
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
* If a page is a file cache, radix-tree replacement is very atomic
|
|
|
|
* and we can skip this check. When it was an Anon page, its mapcount
|
|
|
|
* goes down to 0. But because we added MIGRATION flage, it's not
|
|
|
|
* uncharged yet. There are several case but page->mapcount check
|
|
|
|
* and USED bit check in mem_cgroup_uncharge_page() will do enough
|
|
|
|
* check. (see prepare_charge() also)
|
memcg: remove refcnt from page_cgroup
memcg: performance improvements
Patch Description
1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
2/5 ... swapcache handling patch
3/5 ... add helper function for shmem's memory reclaim patch
4/5 ... optimize by likely/unlikely ppatch
5/5 ... remove redundunt check patch (shmem handling is fixed.)
Unix bench result.
== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput 2915.4 lps (29.6 secs, 3 samples)
C Compiler Throughput 1019.3 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5796.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1097.7 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 565.3 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1022128.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 544057.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 346481.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 319325.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 148788.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 99051.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2058917.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1606109.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 854789.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 126145.2 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 2915.4 678.0
File Copy 1024 bufsize 2000 maxblocks 3960.0 346481.0 875.0
File Copy 256 bufsize 500 maxblocks 1655.0 99051.0 598.5
File Copy 4096 bufsize 8000 maxblocks 5800.0 854789.0 1473.8
Shell Scripts (8 concurrent) 6.0 1097.7 1829.5
=========
FINAL SCORE 991.3
== 2.6.26-rc2-mm1 + this set ==
Execl Throughput 3012.9 lps (29.9 secs, 3 samples)
C Compiler Throughput 981.0 lpm (60.0 secs, 3 samples)
Shell Scripts (1 concurrent) 5872.0 lpm (60.0 secs, 3 samples)
Shell Scripts (8 concurrent) 1120.3 lpm (60.0 secs, 3 samples)
Shell Scripts (16 concurrent) 578.0 lpm (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks 1003993.0 KBps (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks 550452.0 KBps (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks 347159.0 KBps (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks 314644.0 KBps (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks 151852.0 KBps (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks 101000.0 KBps (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks 2033256.0 KBps (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks 1611814.0 KBps (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks 847979.0 KBps (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places 128148.7 lpm (30.0 secs, 3 samples)
INDEX VALUES
TEST BASELINE RESULT INDEX
Execl Throughput 43.0 3012.9 700.7
File Copy 1024 bufsize 2000 maxblocks 3960.0 347159.0 876.7
File Copy 256 bufsize 500 maxblocks 1655.0 101000.0 610.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 847979.0 1462.0
Shell Scripts (8 concurrent) 6.0 1120.3 1867.2
=========
FINAL SCORE 1004.6
This patch:
Remove refcnt from page_cgroup().
After this,
* A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.
* A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.
There is no change in behavior from user's view.
This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.
[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-25 08:47:14 +00:00
|
|
|
*/
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
if (PageAnon(used))
|
|
|
|
mem_cgroup_uncharge_page(used);
|
2009-07-29 22:04:06 +00:00
|
|
|
/*
|
memcg: fix mis-accounting of file mapped racy with migration
FILE_MAPPED per memcg of migrated file cache is not properly updated,
because our hook in page_add_file_rmap() can't know to which memcg
FILE_MAPPED should be counted.
Basically, this patch is for fixing the bug but includes some big changes
to fix up other messes.
Now, at migrating mapped file, events happen in following sequence.
1. allocate a new page.
2. get memcg of an old page.
3. charge ageinst a new page before migration. But at this point,
no changes to new page's page_cgroup, no commit for the charge.
(IOW, PCG_USED bit is not set.)
4. page migration replaces radix-tree, old-page and new-page.
5. page migration remaps the new page if the old page was mapped.
6. Here, the new page is unlocked.
7. memcg commits the charge for newpage, Mark the new page's page_cgroup
as PCG_USED.
Because "commit" happens after page-remap, we can count FILE_MAPPED
at "5", because we should avoid to trust page_cgroup->mem_cgroup.
if PCG_USED bit is unset.
(Note: memcg's LRU removal code does that but LRU-isolation logic is used
for helping it. When we overwrite page_cgroup->mem_cgroup, page_cgroup is
not on LRU or page_cgroup->mem_cgroup is NULL.)
We can lose file_mapped accounting information at 5 because FILE_MAPPED
is updated only when mapcount changes 0->1. So we should catch it.
BTW, historically, above implemntation comes from migration-failure
of anonymous page. Because we charge both of old page and new page
with mapcount=0, we can't catch
- the page is really freed before remap.
- migration fails but it's freed before remap
or .....corner cases.
New migration sequence with memcg is:
1. allocate a new page.
2. mark PageCgroupMigration to the old page.
3. charge against a new page onto the old page's memcg. (here, new page's pc
is marked as PageCgroupUsed.)
4. page migration replaces radix-tree, page table, etc...
5. At remapping, new page's page_cgroup is now makrked as "USED"
We can catch 0->1 event and FILE_MAPPED will be properly updated.
And we can catch SWAPOUT event after unlock this and freeing this
page by unmap() can be caught.
7. Clear PageCgroupMigration of the old page.
So, FILE_MAPPED will be correctly updated.
Then, for what MIGRATION flag is ?
Without it, at migration failure, we may have to charge old page again
because it may be fully unmapped. "charge" means that we have to dive into
memory reclaim or something complated. So, it's better to avoid
charge it again. Before this patch, __commit_charge() was working for
both of the old/new page and fixed up all. But this technique has some
racy condtion around FILE_MAPPED and SWAPOUT etc...
Now, the kernel use MIGRATION flag and don't uncharge old page until
the end of migration.
I hope this change will make memcg's page migration much simpler. This
page migration has caused several troubles. Worth to add a flag for
simplification.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:42:46 +00:00
|
|
|
* At migration, we may charge account against cgroup which has no
|
|
|
|
* tasks.
|
2009-07-29 22:04:06 +00:00
|
|
|
* So, rmdir()->pre_destroy() can be called while we do this charge.
|
|
|
|
* In that case, we need to call pre_destroy() again. check it here.
|
|
|
|
*/
|
|
|
|
cgroup_release_and_wakeup_rmdir(&mem->css);
|
2008-02-07 08:14:10 +00:00
|
|
|
}
|
2008-02-07 08:13:51 +00:00
|
|
|
|
2008-07-25 08:47:15 +00:00
|
|
|
/*
|
2009-04-30 22:08:19 +00:00
|
|
|
* A call to try to shrink memory usage on charge failure at shmem's swapin.
|
|
|
|
* Calling hierarchical_reclaim is not enough because we should update
|
|
|
|
* last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
|
|
|
|
* Moreover considering hierarchy, we should reclaim from the mem_over_limit,
|
|
|
|
* not from the memcg which this page would be charged to.
|
|
|
|
* try_charge_swapin does all of these works properly.
|
2008-07-25 08:47:15 +00:00
|
|
|
*/
|
2009-04-30 22:08:19 +00:00
|
|
|
int mem_cgroup_shmem_charge_fallback(struct page *page,
|
2009-01-08 02:08:35 +00:00
|
|
|
struct mm_struct *mm,
|
|
|
|
gfp_t gfp_mask)
|
2008-07-25 08:47:15 +00:00
|
|
|
{
|
2009-01-08 02:08:35 +00:00
|
|
|
struct mem_cgroup *mem = NULL;
|
2009-04-30 22:08:19 +00:00
|
|
|
int ret;
|
2008-07-25 08:47:15 +00:00
|
|
|
|
2009-01-08 02:08:02 +00:00
|
|
|
if (mem_cgroup_disabled())
|
2008-07-25 08:47:18 +00:00
|
|
|
return 0;
|
2008-07-25 08:47:15 +00:00
|
|
|
|
2009-04-30 22:08:19 +00:00
|
|
|
ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem);
|
|
|
|
if (!ret)
|
|
|
|
mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */
|
2008-07-25 08:47:15 +00:00
|
|
|
|
2009-04-30 22:08:19 +00:00
|
|
|
return ret;
|
2008-07-25 08:47:15 +00:00
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
static DEFINE_MUTEX(set_limit_mutex);
|
|
|
|
|
2009-01-06 22:39:44 +00:00
|
|
|
static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
|
2009-01-08 02:08:00 +00:00
|
|
|
unsigned long long val)
|
2008-07-25 08:47:20 +00:00
|
|
|
{
|
2009-04-02 23:57:36 +00:00
|
|
|
int retry_count;
|
2010-05-26 21:42:37 +00:00
|
|
|
u64 memswlimit, memlimit;
|
2008-07-25 08:47:20 +00:00
|
|
|
int ret = 0;
|
2009-04-02 23:57:36 +00:00
|
|
|
int children = mem_cgroup_count_children(memcg);
|
|
|
|
u64 curusage, oldusage;
|
2010-05-26 21:42:37 +00:00
|
|
|
int enlarge;
|
2009-04-02 23:57:36 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* For keeping hierarchical_reclaim simple, how long we should retry
|
|
|
|
* is depends on callers. We set our retry-count to be function
|
|
|
|
* of # of children which we should visit in this loop.
|
|
|
|
*/
|
|
|
|
retry_count = MEM_CGROUP_RECLAIM_RETRIES * children;
|
|
|
|
|
|
|
|
oldusage = res_counter_read_u64(&memcg->res, RES_USAGE);
|
2008-07-25 08:47:20 +00:00
|
|
|
|
2010-05-26 21:42:37 +00:00
|
|
|
enlarge = 0;
|
2009-01-08 02:08:00 +00:00
|
|
|
while (retry_count) {
|
2008-07-25 08:47:20 +00:00
|
|
|
if (signal_pending(current)) {
|
|
|
|
ret = -EINTR;
|
|
|
|
break;
|
|
|
|
}
|
2009-01-08 02:08:00 +00:00
|
|
|
/*
|
|
|
|
* Rather than hide all in some function, I do this in
|
|
|
|
* open coded manner. You see what this really does.
|
|
|
|
* We have to guarantee mem->res.limit < mem->memsw.limit.
|
|
|
|
*/
|
|
|
|
mutex_lock(&set_limit_mutex);
|
|
|
|
memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
|
|
|
|
if (memswlimit < val) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
mutex_unlock(&set_limit_mutex);
|
2008-07-25 08:47:20 +00:00
|
|
|
break;
|
|
|
|
}
|
2010-05-26 21:42:37 +00:00
|
|
|
|
|
|
|
memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
|
|
|
|
if (memlimit < val)
|
|
|
|
enlarge = 1;
|
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
ret = res_counter_set_limit(&memcg->res, val);
|
2009-06-17 23:27:19 +00:00
|
|
|
if (!ret) {
|
|
|
|
if (memswlimit == val)
|
|
|
|
memcg->memsw_is_minimum = true;
|
|
|
|
else
|
|
|
|
memcg->memsw_is_minimum = false;
|
|
|
|
}
|
2009-01-08 02:08:00 +00:00
|
|
|
mutex_unlock(&set_limit_mutex);
|
|
|
|
|
|
|
|
if (!ret)
|
|
|
|
break;
|
|
|
|
|
2009-12-16 00:47:14 +00:00
|
|
|
mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
|
2009-09-23 22:56:39 +00:00
|
|
|
MEM_CGROUP_RECLAIM_SHRINK);
|
2009-04-02 23:57:36 +00:00
|
|
|
curusage = res_counter_read_u64(&memcg->res, RES_USAGE);
|
|
|
|
/* Usage is reduced ? */
|
|
|
|
if (curusage >= oldusage)
|
|
|
|
retry_count--;
|
|
|
|
else
|
|
|
|
oldusage = curusage;
|
2009-01-08 02:08:00 +00:00
|
|
|
}
|
2010-05-26 21:42:37 +00:00
|
|
|
if (!ret && enlarge)
|
|
|
|
memcg_oom_recover(memcg);
|
2009-01-08 02:08:18 +00:00
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-06-17 23:27:15 +00:00
|
|
|
static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
|
|
|
|
unsigned long long val)
|
2009-01-08 02:08:00 +00:00
|
|
|
{
|
2009-04-02 23:57:36 +00:00
|
|
|
int retry_count;
|
2010-05-26 21:42:37 +00:00
|
|
|
u64 memlimit, memswlimit, oldusage, curusage;
|
2009-04-02 23:57:36 +00:00
|
|
|
int children = mem_cgroup_count_children(memcg);
|
|
|
|
int ret = -EBUSY;
|
2010-05-26 21:42:37 +00:00
|
|
|
int enlarge = 0;
|
2009-01-08 02:08:00 +00:00
|
|
|
|
2009-04-02 23:57:36 +00:00
|
|
|
/* see mem_cgroup_resize_res_limit */
|
|
|
|
retry_count = children * MEM_CGROUP_RECLAIM_RETRIES;
|
|
|
|
oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
|
2009-01-08 02:08:00 +00:00
|
|
|
while (retry_count) {
|
|
|
|
if (signal_pending(current)) {
|
|
|
|
ret = -EINTR;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* Rather than hide all in some function, I do this in
|
|
|
|
* open coded manner. You see what this really does.
|
|
|
|
* We have to guarantee mem->res.limit < mem->memsw.limit.
|
|
|
|
*/
|
|
|
|
mutex_lock(&set_limit_mutex);
|
|
|
|
memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT);
|
|
|
|
if (memlimit > val) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
mutex_unlock(&set_limit_mutex);
|
|
|
|
break;
|
|
|
|
}
|
2010-05-26 21:42:37 +00:00
|
|
|
memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
|
|
|
|
if (memswlimit < val)
|
|
|
|
enlarge = 1;
|
2009-01-08 02:08:00 +00:00
|
|
|
ret = res_counter_set_limit(&memcg->memsw, val);
|
2009-06-17 23:27:19 +00:00
|
|
|
if (!ret) {
|
|
|
|
if (memlimit == val)
|
|
|
|
memcg->memsw_is_minimum = true;
|
|
|
|
else
|
|
|
|
memcg->memsw_is_minimum = false;
|
|
|
|
}
|
2009-01-08 02:08:00 +00:00
|
|
|
mutex_unlock(&set_limit_mutex);
|
|
|
|
|
|
|
|
if (!ret)
|
|
|
|
break;
|
|
|
|
|
2009-09-23 22:56:39 +00:00
|
|
|
mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL,
|
2009-09-23 22:56:38 +00:00
|
|
|
MEM_CGROUP_RECLAIM_NOSWAP |
|
|
|
|
MEM_CGROUP_RECLAIM_SHRINK);
|
2009-01-08 02:08:00 +00:00
|
|
|
curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE);
|
2009-04-02 23:57:36 +00:00
|
|
|
/* Usage is reduced ? */
|
2009-01-08 02:08:00 +00:00
|
|
|
if (curusage >= oldusage)
|
2008-07-25 08:47:20 +00:00
|
|
|
retry_count--;
|
2009-04-02 23:57:36 +00:00
|
|
|
else
|
|
|
|
oldusage = curusage;
|
2008-07-25 08:47:20 +00:00
|
|
|
}
|
2010-05-26 21:42:37 +00:00
|
|
|
if (!ret && enlarge)
|
|
|
|
memcg_oom_recover(memcg);
|
2008-07-25 08:47:20 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:39 +00:00
|
|
|
unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
|
2010-08-11 01:03:05 +00:00
|
|
|
gfp_t gfp_mask)
|
2009-09-23 22:56:39 +00:00
|
|
|
{
|
|
|
|
unsigned long nr_reclaimed = 0;
|
|
|
|
struct mem_cgroup_per_zone *mz, *next_mz = NULL;
|
|
|
|
unsigned long reclaimed;
|
|
|
|
int loop = 0;
|
|
|
|
struct mem_cgroup_tree_per_zone *mctz;
|
2009-10-01 22:44:12 +00:00
|
|
|
unsigned long long excess;
|
2009-09-23 22:56:39 +00:00
|
|
|
|
|
|
|
if (order > 0)
|
|
|
|
return 0;
|
|
|
|
|
2010-08-11 01:03:05 +00:00
|
|
|
mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
|
2009-09-23 22:56:39 +00:00
|
|
|
/*
|
|
|
|
* This loop can run a while, specially if mem_cgroup's continuously
|
|
|
|
* keep exceeding their soft limit and putting the system under
|
|
|
|
* pressure
|
|
|
|
*/
|
|
|
|
do {
|
|
|
|
if (next_mz)
|
|
|
|
mz = next_mz;
|
|
|
|
else
|
|
|
|
mz = mem_cgroup_largest_soft_limit_node(mctz);
|
|
|
|
if (!mz)
|
|
|
|
break;
|
|
|
|
|
|
|
|
reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone,
|
|
|
|
gfp_mask,
|
|
|
|
MEM_CGROUP_RECLAIM_SOFT);
|
|
|
|
nr_reclaimed += reclaimed;
|
|
|
|
spin_lock(&mctz->lock);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we failed to reclaim anything from this memory cgroup
|
|
|
|
* it is time to move on to the next cgroup
|
|
|
|
*/
|
|
|
|
next_mz = NULL;
|
|
|
|
if (!reclaimed) {
|
|
|
|
do {
|
|
|
|
/*
|
|
|
|
* Loop until we find yet another one.
|
|
|
|
*
|
|
|
|
* By the time we get the soft_limit lock
|
|
|
|
* again, someone might have aded the
|
|
|
|
* group back on the RB tree. Iterate to
|
|
|
|
* make sure we get a different mem.
|
|
|
|
* mem_cgroup_largest_soft_limit_node returns
|
|
|
|
* NULL if no other cgroup is present on
|
|
|
|
* the tree
|
|
|
|
*/
|
|
|
|
next_mz =
|
|
|
|
__mem_cgroup_largest_soft_limit_node(mctz);
|
|
|
|
if (next_mz == mz) {
|
|
|
|
css_put(&next_mz->mem->css);
|
|
|
|
next_mz = NULL;
|
|
|
|
} else /* next_mz == NULL or other memcg */
|
|
|
|
break;
|
|
|
|
} while (1);
|
|
|
|
}
|
|
|
|
__mem_cgroup_remove_exceeded(mz->mem, mz, mctz);
|
2009-10-01 22:44:12 +00:00
|
|
|
excess = res_counter_soft_limit_excess(&mz->mem->res);
|
2009-09-23 22:56:39 +00:00
|
|
|
/*
|
|
|
|
* One school of thought says that we should not add
|
|
|
|
* back the node to the tree if reclaim returns 0.
|
|
|
|
* But our reclaim could return 0, simply because due
|
|
|
|
* to priority we are exposing a smaller subset of
|
|
|
|
* memory to reclaim from. Consider this as a longer
|
|
|
|
* term TODO.
|
|
|
|
*/
|
2009-10-01 22:44:12 +00:00
|
|
|
/* If excess == 0, no tree ops */
|
|
|
|
__mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess);
|
2009-09-23 22:56:39 +00:00
|
|
|
spin_unlock(&mctz->lock);
|
|
|
|
css_put(&mz->mem->css);
|
|
|
|
loop++;
|
|
|
|
/*
|
|
|
|
* Could not reclaim anything and there are no more
|
|
|
|
* mem cgroups to try or we seem to be looping without
|
|
|
|
* reclaiming anything.
|
|
|
|
*/
|
|
|
|
if (!nr_reclaimed &&
|
|
|
|
(next_mz == NULL ||
|
|
|
|
loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
|
|
|
|
break;
|
|
|
|
} while (!nr_reclaimed);
|
|
|
|
if (next_mz)
|
|
|
|
css_put(&next_mz->mem->css);
|
|
|
|
return nr_reclaimed;
|
|
|
|
}
|
|
|
|
|
2008-02-07 08:14:16 +00:00
|
|
|
/*
|
|
|
|
* This routine traverse page_cgroup in given list and drop them all.
|
|
|
|
* *And* this routine doesn't reclaim page itself, just removes page_cgroup.
|
|
|
|
*/
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
static int mem_cgroup_force_empty_list(struct mem_cgroup *mem,
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
int node, int zid, enum lru_list lru)
|
2008-02-07 08:14:16 +00:00
|
|
|
{
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
struct zone *zone;
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
struct page_cgroup *pc, *busy;
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
unsigned long flags, loop;
|
2008-02-07 08:14:39 +00:00
|
|
|
struct list_head *list;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
int ret = 0;
|
2008-02-07 08:14:39 +00:00
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
zone = &NODE_DATA(node)->node_zones[zid];
|
|
|
|
mz = mem_cgroup_zoneinfo(mem, node, zid);
|
2008-10-19 03:26:14 +00:00
|
|
|
list = &mz->lists[lru];
|
2008-02-07 08:14:16 +00:00
|
|
|
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
loop = MEM_CGROUP_ZSTAT(mz, lru);
|
|
|
|
/* give some margin against EBUSY etc...*/
|
|
|
|
loop += 256;
|
|
|
|
busy = NULL;
|
|
|
|
while (loop--) {
|
|
|
|
ret = 0;
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
spin_lock_irqsave(&zone->lru_lock, flags);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
if (list_empty(list)) {
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
spin_unlock_irqrestore(&zone->lru_lock, flags);
|
2008-10-19 03:28:16 +00:00
|
|
|
break;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
}
|
|
|
|
pc = list_entry(list->prev, struct page_cgroup, lru);
|
|
|
|
if (busy == pc) {
|
|
|
|
list_move(&pc->lru, list);
|
2010-03-05 21:42:04 +00:00
|
|
|
busy = NULL;
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
spin_unlock_irqrestore(&zone->lru_lock, flags);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
continue;
|
|
|
|
}
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
spin_unlock_irqrestore(&zone->lru_lock, flags);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
|
2009-01-08 02:08:10 +00:00
|
|
|
ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
if (ret == -ENOMEM)
|
2008-10-19 03:28:16 +00:00
|
|
|
break;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
|
|
|
|
if (ret == -EBUSY || ret == -EINVAL) {
|
|
|
|
/* found lock contention or "pc" is obsolete. */
|
|
|
|
busy = pc;
|
|
|
|
cond_resched();
|
|
|
|
} else
|
|
|
|
busy = NULL;
|
2008-02-07 08:14:16 +00:00
|
|
|
}
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
if (!ret && !list_empty(list))
|
|
|
|
return -EBUSY;
|
|
|
|
return ret;
|
2008-02-07 08:14:16 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* make mem_cgroup's charge to be 0 if there is no task.
|
|
|
|
* This enables deleting this mem_cgroup.
|
|
|
|
*/
|
2009-01-08 02:07:55 +00:00
|
|
|
static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all)
|
2008-02-07 08:14:16 +00:00
|
|
|
{
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
int ret;
|
|
|
|
int node, zid, shrink;
|
|
|
|
int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
|
2009-01-08 02:07:55 +00:00
|
|
|
struct cgroup *cgrp = mem->css.cgroup;
|
2008-03-04 22:29:09 +00:00
|
|
|
|
2008-02-07 08:14:16 +00:00
|
|
|
css_get(&mem->css);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
|
|
|
|
shrink = 0;
|
2009-01-08 02:07:55 +00:00
|
|
|
/* should free all ? */
|
|
|
|
if (free_all)
|
|
|
|
goto try_to_free;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
move_account:
|
memcg: ensure list is empty at rmdir
Current mem_cgroup_force_empty() only ensures mem->res.usage == 0 on
success. But this doesn't guarantee memcg's LRU is really empty, because
there are some cases in which !PageCgrupUsed pages exist on memcg's LRU.
For example:
- Pages can be uncharged by its owner process while they are on LRU.
- race between mem_cgroup_add_lru_list() and __mem_cgroup_uncharge_common().
So there can be a case in which the usage is zero but some of the LRUs are not empty.
OTOH, mem_cgroup_del_lru_list(), which can be called asynchronously with
rmdir, accesses the mem_cgroup, so this access can cause a problem if it
races with rmdir because the mem_cgroup might have been freed by rmdir.
Actually, I saw a bug which seems to be caused by this race.
[1530745.949906] BUG: unable to handle kernel NULL pointer dereference at 0000000000000230
[1530745.950651] IP: [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] PGD 3863de067 PUD 3862c7067 PMD 0
[1530745.950651] Oops: 0002 [#1] SMP
[1530745.950651] last sysfs file: /sys/devices/system/cpu/cpu7/cache/index1/shared_cpu_map
[1530745.950651] CPU 3
[1530745.950651] Modules linked in: configs ipt_REJECT xt_tcpudp iptable_filter ip_tables x_tables bridge stp nfsd nfs_acl auth_rpcgss exportfs autofs4 hidp rfcomm l2cap crc16 bluetooth lockd sunrpc ib_iser rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr iscsi_tcp bnx2i cnic uio ipv6 cxgb3i cxgb3 mdio libiscsi_tcp libiscsi scsi_transport_iscsi dm_mirror dm_multipath scsi_dh video output sbs sbshc battery ac lp kvm_intel kvm sg ide_cd_mod cdrom serio_raw tpm_tis tpm tpm_bios acpi_memhotplug button parport_pc parport rtc_cmos rtc_core rtc_lib e1000 i2c_i801 i2c_core pcspkr dm_region_hash dm_log dm_mod ata_piix libata shpchp megaraid_mbox sd_mod scsi_mod megaraid_mm ext3 jbd uhci_hcd ohci_hcd ehci_hcd [last unloaded: freq_table]
[1530745.950651] Pid: 19653, comm: shmem_test_02 Tainted: G M 2.6.32-mm1-00701-g2b04386 #3 Express5800/140Rd-4 [N8100-1065]
[1530745.950651] RIP: 0010:[<ffffffff810fbc11>] [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] RSP: 0018:ffff8803863ddcb8 EFLAGS: 00010002
[1530745.950651] RAX: 00000000000001e0 RBX: ffff8803abc02238 RCX: 00000000000001e0
[1530745.950651] RDX: 0000000000000000 RSI: ffff88038611a000 RDI: ffff8803abc02238
[1530745.950651] RBP: ffff8803863ddcc8 R08: 0000000000000002 R09: ffff8803a04c8643
[1530745.950651] R10: 0000000000000000 R11: ffffffff810c7333 R12: 0000000000000000
[1530745.950651] R13: ffff880000017f00 R14: 0000000000000092 R15: ffff8800179d0310
[1530745.950651] FS: 0000000000000000(0000) GS:ffff880017800000(0000) knlGS:0000000000000000
[1530745.950651] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[1530745.950651] CR2: 0000000000000230 CR3: 0000000379d87000 CR4: 00000000000006e0
[1530745.950651] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[1530745.950651] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[1530745.950651] Process shmem_test_02 (pid: 19653, threadinfo ffff8803863dc000, task ffff88038612a8a0)
[1530745.950651] Stack:
[1530745.950651] ffffea00040c2fe8 0000000000000000 ffff8803863ddd98 ffffffff810c739a
[1530745.950651] <0> 00000000863ddd18 000000000000000c 0000000000000000 0000000000000000
[1530745.950651] <0> 0000000000000002 0000000000000000 ffff8803863ddd68 0000000000000046
[1530745.950651] Call Trace:
[1530745.950651] [<ffffffff810c739a>] release_pages+0x142/0x1e7
[1530745.950651] [<ffffffff810c778f>] ? pagevec_move_tail+0x6e/0x112
[1530745.950651] [<ffffffff810c781e>] pagevec_move_tail+0xfd/0x112
[1530745.950651] [<ffffffff810c78a9>] lru_add_drain+0x76/0x94
[1530745.950651] [<ffffffff810dba0c>] exit_mmap+0x6e/0x145
[1530745.950651] [<ffffffff8103f52d>] mmput+0x5e/0xcf
[1530745.950651] [<ffffffff81043ea8>] exit_mm+0x11c/0x129
[1530745.950651] [<ffffffff8108fb29>] ? audit_free+0x196/0x1c9
[1530745.950651] [<ffffffff81045353>] do_exit+0x1f5/0x6b7
[1530745.950651] [<ffffffff8106133f>] ? up_read+0x2b/0x2f
[1530745.950651] [<ffffffff8137d187>] ? lockdep_sys_exit_thunk+0x35/0x67
[1530745.950651] [<ffffffff81045898>] do_group_exit+0x83/0xb0
[1530745.950651] [<ffffffff810458dc>] sys_exit_group+0x17/0x1b
[1530745.950651] [<ffffffff81002c1b>] system_call_fastpath+0x16/0x1b
[1530745.950651] Code: 54 53 0f 1f 44 00 00 83 3d cc 29 7c 00 00 41 89 f4 75 63 eb 4e 48 83 7b 08 00 75 04 0f 0b eb fe 48 89 df e8 18 f3 ff ff 44 89 e2 <48> ff 4c d0 50 48 8b 05 2b 2d 7c 00 48 39 43 08 74 39 48 8b 4b
[1530745.950651] RIP [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] RSP <ffff8803863ddcb8>
[1530745.950651] CR2: 0000000000000230
[1530745.950651] ---[ end trace c3419c1bb8acc34f ]---
[1530745.950651] Fixing recursive fault but reboot is needed!
The problem here is pages on LRU may contain pointer to stale memcg. To
make res->usage to be 0, all pages on memcg must be uncharged or moved to
another(parent) memcg. Moved page_cgroup have already removed from
original LRU, but uncharged page_cgroup contains pointer to memcg withou
PCG_USED bit. (This asynchronous LRU work is for improving performance.)
If PCG_USED bit is not set, page_cgroup will never be added to memcg's
LRU. So, about pages not on LRU, they never access stale pointer. Then,
what we have to take care of is page_cgroup _on_ LRU list. This patch
fixes this problem by making mem_cgroup_force_empty() visit all LRUs
before exiting its loop and guarantee there are no pages on its LRU.
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-01-16 01:01:30 +00:00
|
|
|
do {
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
ret = -EBUSY;
|
2009-01-08 02:07:55 +00:00
|
|
|
if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children))
|
|
|
|
goto out;
|
|
|
|
ret = -EINTR;
|
|
|
|
if (signal_pending(current))
|
2008-02-07 08:14:16 +00:00
|
|
|
goto out;
|
2008-10-19 03:28:16 +00:00
|
|
|
/* This is for making all *used* pages to be on LRU. */
|
|
|
|
lru_add_drain_all();
|
2009-12-16 00:47:08 +00:00
|
|
|
drain_all_stock_sync();
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
ret = 0;
|
2009-01-29 22:25:17 +00:00
|
|
|
for_each_node_state(node, N_HIGH_MEMORY) {
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) {
|
2008-10-19 03:26:14 +00:00
|
|
|
enum lru_list l;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
for_each_lru(l) {
|
|
|
|
ret = mem_cgroup_force_empty_list(mem,
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
node, zid, l);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
}
|
2008-02-07 08:14:38 +00:00
|
|
|
}
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
}
|
2010-05-26 21:42:37 +00:00
|
|
|
memcg_oom_recover(mem);
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
/* it seems parent cgroup doesn't have enough mem */
|
|
|
|
if (ret == -ENOMEM)
|
|
|
|
goto try_to_free;
|
2008-10-19 03:28:16 +00:00
|
|
|
cond_resched();
|
memcg: ensure list is empty at rmdir
Current mem_cgroup_force_empty() only ensures mem->res.usage == 0 on
success. But this doesn't guarantee memcg's LRU is really empty, because
there are some cases in which !PageCgrupUsed pages exist on memcg's LRU.
For example:
- Pages can be uncharged by its owner process while they are on LRU.
- race between mem_cgroup_add_lru_list() and __mem_cgroup_uncharge_common().
So there can be a case in which the usage is zero but some of the LRUs are not empty.
OTOH, mem_cgroup_del_lru_list(), which can be called asynchronously with
rmdir, accesses the mem_cgroup, so this access can cause a problem if it
races with rmdir because the mem_cgroup might have been freed by rmdir.
Actually, I saw a bug which seems to be caused by this race.
[1530745.949906] BUG: unable to handle kernel NULL pointer dereference at 0000000000000230
[1530745.950651] IP: [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] PGD 3863de067 PUD 3862c7067 PMD 0
[1530745.950651] Oops: 0002 [#1] SMP
[1530745.950651] last sysfs file: /sys/devices/system/cpu/cpu7/cache/index1/shared_cpu_map
[1530745.950651] CPU 3
[1530745.950651] Modules linked in: configs ipt_REJECT xt_tcpudp iptable_filter ip_tables x_tables bridge stp nfsd nfs_acl auth_rpcgss exportfs autofs4 hidp rfcomm l2cap crc16 bluetooth lockd sunrpc ib_iser rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr iscsi_tcp bnx2i cnic uio ipv6 cxgb3i cxgb3 mdio libiscsi_tcp libiscsi scsi_transport_iscsi dm_mirror dm_multipath scsi_dh video output sbs sbshc battery ac lp kvm_intel kvm sg ide_cd_mod cdrom serio_raw tpm_tis tpm tpm_bios acpi_memhotplug button parport_pc parport rtc_cmos rtc_core rtc_lib e1000 i2c_i801 i2c_core pcspkr dm_region_hash dm_log dm_mod ata_piix libata shpchp megaraid_mbox sd_mod scsi_mod megaraid_mm ext3 jbd uhci_hcd ohci_hcd ehci_hcd [last unloaded: freq_table]
[1530745.950651] Pid: 19653, comm: shmem_test_02 Tainted: G M 2.6.32-mm1-00701-g2b04386 #3 Express5800/140Rd-4 [N8100-1065]
[1530745.950651] RIP: 0010:[<ffffffff810fbc11>] [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] RSP: 0018:ffff8803863ddcb8 EFLAGS: 00010002
[1530745.950651] RAX: 00000000000001e0 RBX: ffff8803abc02238 RCX: 00000000000001e0
[1530745.950651] RDX: 0000000000000000 RSI: ffff88038611a000 RDI: ffff8803abc02238
[1530745.950651] RBP: ffff8803863ddcc8 R08: 0000000000000002 R09: ffff8803a04c8643
[1530745.950651] R10: 0000000000000000 R11: ffffffff810c7333 R12: 0000000000000000
[1530745.950651] R13: ffff880000017f00 R14: 0000000000000092 R15: ffff8800179d0310
[1530745.950651] FS: 0000000000000000(0000) GS:ffff880017800000(0000) knlGS:0000000000000000
[1530745.950651] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[1530745.950651] CR2: 0000000000000230 CR3: 0000000379d87000 CR4: 00000000000006e0
[1530745.950651] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[1530745.950651] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[1530745.950651] Process shmem_test_02 (pid: 19653, threadinfo ffff8803863dc000, task ffff88038612a8a0)
[1530745.950651] Stack:
[1530745.950651] ffffea00040c2fe8 0000000000000000 ffff8803863ddd98 ffffffff810c739a
[1530745.950651] <0> 00000000863ddd18 000000000000000c 0000000000000000 0000000000000000
[1530745.950651] <0> 0000000000000002 0000000000000000 ffff8803863ddd68 0000000000000046
[1530745.950651] Call Trace:
[1530745.950651] [<ffffffff810c739a>] release_pages+0x142/0x1e7
[1530745.950651] [<ffffffff810c778f>] ? pagevec_move_tail+0x6e/0x112
[1530745.950651] [<ffffffff810c781e>] pagevec_move_tail+0xfd/0x112
[1530745.950651] [<ffffffff810c78a9>] lru_add_drain+0x76/0x94
[1530745.950651] [<ffffffff810dba0c>] exit_mmap+0x6e/0x145
[1530745.950651] [<ffffffff8103f52d>] mmput+0x5e/0xcf
[1530745.950651] [<ffffffff81043ea8>] exit_mm+0x11c/0x129
[1530745.950651] [<ffffffff8108fb29>] ? audit_free+0x196/0x1c9
[1530745.950651] [<ffffffff81045353>] do_exit+0x1f5/0x6b7
[1530745.950651] [<ffffffff8106133f>] ? up_read+0x2b/0x2f
[1530745.950651] [<ffffffff8137d187>] ? lockdep_sys_exit_thunk+0x35/0x67
[1530745.950651] [<ffffffff81045898>] do_group_exit+0x83/0xb0
[1530745.950651] [<ffffffff810458dc>] sys_exit_group+0x17/0x1b
[1530745.950651] [<ffffffff81002c1b>] system_call_fastpath+0x16/0x1b
[1530745.950651] Code: 54 53 0f 1f 44 00 00 83 3d cc 29 7c 00 00 41 89 f4 75 63 eb 4e 48 83 7b 08 00 75 04 0f 0b eb fe 48 89 df e8 18 f3 ff ff 44 89 e2 <48> ff 4c d0 50 48 8b 05 2b 2d 7c 00 48 39 43 08 74 39 48 8b 4b
[1530745.950651] RIP [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] RSP <ffff8803863ddcb8>
[1530745.950651] CR2: 0000000000000230
[1530745.950651] ---[ end trace c3419c1bb8acc34f ]---
[1530745.950651] Fixing recursive fault but reboot is needed!
The problem here is pages on LRU may contain pointer to stale memcg. To
make res->usage to be 0, all pages on memcg must be uncharged or moved to
another(parent) memcg. Moved page_cgroup have already removed from
original LRU, but uncharged page_cgroup contains pointer to memcg withou
PCG_USED bit. (This asynchronous LRU work is for improving performance.)
If PCG_USED bit is not set, page_cgroup will never be added to memcg's
LRU. So, about pages not on LRU, they never access stale pointer. Then,
what we have to take care of is page_cgroup _on_ LRU list. This patch
fixes this problem by making mem_cgroup_force_empty() visit all LRUs
before exiting its loop and guarantee there are no pages on its LRU.
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-01-16 01:01:30 +00:00
|
|
|
/* "ret" should also be checked to ensure all lists are empty. */
|
|
|
|
} while (mem->res.usage > 0 || ret);
|
2008-02-07 08:14:16 +00:00
|
|
|
out:
|
|
|
|
css_put(&mem->css);
|
|
|
|
return ret;
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
|
|
|
|
try_to_free:
|
2009-01-08 02:07:55 +00:00
|
|
|
/* returns EBUSY if there is a task or if we come here twice. */
|
|
|
|
if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) {
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
ret = -EBUSY;
|
|
|
|
goto out;
|
|
|
|
}
|
2009-01-08 02:07:55 +00:00
|
|
|
/* we call try-to-free pages for make this cgroup empty */
|
|
|
|
lru_add_drain_all();
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
/* try to free all pages in this cgroup */
|
|
|
|
shrink = 1;
|
|
|
|
while (nr_retries && mem->res.usage > 0) {
|
|
|
|
int progress;
|
2009-01-08 02:07:55 +00:00
|
|
|
|
|
|
|
if (signal_pending(current)) {
|
|
|
|
ret = -EINTR;
|
|
|
|
goto out;
|
|
|
|
}
|
2009-01-08 02:08:24 +00:00
|
|
|
progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL,
|
|
|
|
false, get_swappiness(mem));
|
2009-01-08 02:07:55 +00:00
|
|
|
if (!progress) {
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
nr_retries--;
|
2009-01-08 02:07:55 +00:00
|
|
|
/* maybe some writeback is necessary */
|
2009-07-09 12:52:32 +00:00
|
|
|
congestion_wait(BLK_RW_ASYNC, HZ/10);
|
2009-01-08 02:07:55 +00:00
|
|
|
}
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
|
|
|
|
}
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
lru_add_drain();
|
memcg: move all acccounting to parent at rmdir()
This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.
This moving of page_cgroup is done under
- lru_lock of source/destination mem_cgroup is held.
- lock_page_cgroup() is held.
Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not. Typical code can be
following.
(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);
Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);
But you should confirm the nest of lock and avoid deadlock.
If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.
Expected users of this routine is:
- force_empty (rmdir)
- moving tasks between cgroup (for moving account information.)
- hierarchy (maybe useful.)
force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)
If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.
Purpose of this behavior is
- Fix "forget all" behavior of force_empty and avoid leak of accounting.
- By "moving first, free if necessary", keep pages on memory as much as
possible.
Adding a switch to change behavior of force_empty to
- free first, move if necessary
- free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)
This patch also removes memory.force_empty file, a brutal debug-only interface.
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:07:53 +00:00
|
|
|
/* try move_account...there may be some *locked* pages. */
|
memcg: ensure list is empty at rmdir
Current mem_cgroup_force_empty() only ensures mem->res.usage == 0 on
success. But this doesn't guarantee memcg's LRU is really empty, because
there are some cases in which !PageCgrupUsed pages exist on memcg's LRU.
For example:
- Pages can be uncharged by its owner process while they are on LRU.
- race between mem_cgroup_add_lru_list() and __mem_cgroup_uncharge_common().
So there can be a case in which the usage is zero but some of the LRUs are not empty.
OTOH, mem_cgroup_del_lru_list(), which can be called asynchronously with
rmdir, accesses the mem_cgroup, so this access can cause a problem if it
races with rmdir because the mem_cgroup might have been freed by rmdir.
Actually, I saw a bug which seems to be caused by this race.
[1530745.949906] BUG: unable to handle kernel NULL pointer dereference at 0000000000000230
[1530745.950651] IP: [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] PGD 3863de067 PUD 3862c7067 PMD 0
[1530745.950651] Oops: 0002 [#1] SMP
[1530745.950651] last sysfs file: /sys/devices/system/cpu/cpu7/cache/index1/shared_cpu_map
[1530745.950651] CPU 3
[1530745.950651] Modules linked in: configs ipt_REJECT xt_tcpudp iptable_filter ip_tables x_tables bridge stp nfsd nfs_acl auth_rpcgss exportfs autofs4 hidp rfcomm l2cap crc16 bluetooth lockd sunrpc ib_iser rdma_cm ib_cm iw_cm ib_sa ib_mad ib_core ib_addr iscsi_tcp bnx2i cnic uio ipv6 cxgb3i cxgb3 mdio libiscsi_tcp libiscsi scsi_transport_iscsi dm_mirror dm_multipath scsi_dh video output sbs sbshc battery ac lp kvm_intel kvm sg ide_cd_mod cdrom serio_raw tpm_tis tpm tpm_bios acpi_memhotplug button parport_pc parport rtc_cmos rtc_core rtc_lib e1000 i2c_i801 i2c_core pcspkr dm_region_hash dm_log dm_mod ata_piix libata shpchp megaraid_mbox sd_mod scsi_mod megaraid_mm ext3 jbd uhci_hcd ohci_hcd ehci_hcd [last unloaded: freq_table]
[1530745.950651] Pid: 19653, comm: shmem_test_02 Tainted: G M 2.6.32-mm1-00701-g2b04386 #3 Express5800/140Rd-4 [N8100-1065]
[1530745.950651] RIP: 0010:[<ffffffff810fbc11>] [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] RSP: 0018:ffff8803863ddcb8 EFLAGS: 00010002
[1530745.950651] RAX: 00000000000001e0 RBX: ffff8803abc02238 RCX: 00000000000001e0
[1530745.950651] RDX: 0000000000000000 RSI: ffff88038611a000 RDI: ffff8803abc02238
[1530745.950651] RBP: ffff8803863ddcc8 R08: 0000000000000002 R09: ffff8803a04c8643
[1530745.950651] R10: 0000000000000000 R11: ffffffff810c7333 R12: 0000000000000000
[1530745.950651] R13: ffff880000017f00 R14: 0000000000000092 R15: ffff8800179d0310
[1530745.950651] FS: 0000000000000000(0000) GS:ffff880017800000(0000) knlGS:0000000000000000
[1530745.950651] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[1530745.950651] CR2: 0000000000000230 CR3: 0000000379d87000 CR4: 00000000000006e0
[1530745.950651] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[1530745.950651] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[1530745.950651] Process shmem_test_02 (pid: 19653, threadinfo ffff8803863dc000, task ffff88038612a8a0)
[1530745.950651] Stack:
[1530745.950651] ffffea00040c2fe8 0000000000000000 ffff8803863ddd98 ffffffff810c739a
[1530745.950651] <0> 00000000863ddd18 000000000000000c 0000000000000000 0000000000000000
[1530745.950651] <0> 0000000000000002 0000000000000000 ffff8803863ddd68 0000000000000046
[1530745.950651] Call Trace:
[1530745.950651] [<ffffffff810c739a>] release_pages+0x142/0x1e7
[1530745.950651] [<ffffffff810c778f>] ? pagevec_move_tail+0x6e/0x112
[1530745.950651] [<ffffffff810c781e>] pagevec_move_tail+0xfd/0x112
[1530745.950651] [<ffffffff810c78a9>] lru_add_drain+0x76/0x94
[1530745.950651] [<ffffffff810dba0c>] exit_mmap+0x6e/0x145
[1530745.950651] [<ffffffff8103f52d>] mmput+0x5e/0xcf
[1530745.950651] [<ffffffff81043ea8>] exit_mm+0x11c/0x129
[1530745.950651] [<ffffffff8108fb29>] ? audit_free+0x196/0x1c9
[1530745.950651] [<ffffffff81045353>] do_exit+0x1f5/0x6b7
[1530745.950651] [<ffffffff8106133f>] ? up_read+0x2b/0x2f
[1530745.950651] [<ffffffff8137d187>] ? lockdep_sys_exit_thunk+0x35/0x67
[1530745.950651] [<ffffffff81045898>] do_group_exit+0x83/0xb0
[1530745.950651] [<ffffffff810458dc>] sys_exit_group+0x17/0x1b
[1530745.950651] [<ffffffff81002c1b>] system_call_fastpath+0x16/0x1b
[1530745.950651] Code: 54 53 0f 1f 44 00 00 83 3d cc 29 7c 00 00 41 89 f4 75 63 eb 4e 48 83 7b 08 00 75 04 0f 0b eb fe 48 89 df e8 18 f3 ff ff 44 89 e2 <48> ff 4c d0 50 48 8b 05 2b 2d 7c 00 48 39 43 08 74 39 48 8b 4b
[1530745.950651] RIP [<ffffffff810fbc11>] mem_cgroup_del_lru_list+0x30/0x80
[1530745.950651] RSP <ffff8803863ddcb8>
[1530745.950651] CR2: 0000000000000230
[1530745.950651] ---[ end trace c3419c1bb8acc34f ]---
[1530745.950651] Fixing recursive fault but reboot is needed!
The problem here is pages on LRU may contain pointer to stale memcg. To
make res->usage to be 0, all pages on memcg must be uncharged or moved to
another(parent) memcg. Moved page_cgroup have already removed from
original LRU, but uncharged page_cgroup contains pointer to memcg withou
PCG_USED bit. (This asynchronous LRU work is for improving performance.)
If PCG_USED bit is not set, page_cgroup will never be added to memcg's
LRU. So, about pages not on LRU, they never access stale pointer. Then,
what we have to take care of is page_cgroup _on_ LRU list. This patch
fixes this problem by making mem_cgroup_force_empty() visit all LRUs
before exiting its loop and guarantee there are no pages on its LRU.
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-01-16 01:01:30 +00:00
|
|
|
goto move_account;
|
2008-02-07 08:14:16 +00:00
|
|
|
}
|
|
|
|
|
2009-01-08 02:07:55 +00:00
|
|
|
int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event)
|
|
|
|
{
|
|
|
|
return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2009-01-08 02:08:07 +00:00
|
|
|
static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft)
|
|
|
|
{
|
|
|
|
return mem_cgroup_from_cont(cont)->use_hierarchy;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
|
|
|
|
u64 val)
|
|
|
|
{
|
|
|
|
int retval = 0;
|
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
|
|
|
|
struct cgroup *parent = cont->parent;
|
|
|
|
struct mem_cgroup *parent_mem = NULL;
|
|
|
|
|
|
|
|
if (parent)
|
|
|
|
parent_mem = mem_cgroup_from_cont(parent);
|
|
|
|
|
|
|
|
cgroup_lock();
|
|
|
|
/*
|
tree-wide: fix assorted typos all over the place
That is "success", "unknown", "through", "performance", "[re|un]mapping"
, "access", "default", "reasonable", "[con]currently", "temperature"
, "channel", "[un]used", "application", "example","hierarchy", "therefore"
, "[over|under]flow", "contiguous", "threshold", "enough" and others.
Signed-off-by: André Goddard Rosa <andre.goddard@gmail.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2009-11-14 15:09:05 +00:00
|
|
|
* If parent's use_hierarchy is set, we can't make any modifications
|
2009-01-08 02:08:07 +00:00
|
|
|
* in the child subtrees. If it is unset, then the change can
|
|
|
|
* occur, provided the current cgroup has no children.
|
|
|
|
*
|
|
|
|
* For the root cgroup, parent_mem is NULL, we allow value to be
|
|
|
|
* set if there are no children.
|
|
|
|
*/
|
|
|
|
if ((!parent_mem || !parent_mem->use_hierarchy) &&
|
|
|
|
(val == 1 || val == 0)) {
|
|
|
|
if (list_empty(&cont->children))
|
|
|
|
mem->use_hierarchy = val;
|
|
|
|
else
|
|
|
|
retval = -EBUSY;
|
|
|
|
} else
|
|
|
|
retval = -EINVAL;
|
|
|
|
cgroup_unlock();
|
|
|
|
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
2009-09-23 22:56:42 +00:00
|
|
|
struct mem_cgroup_idx_data {
|
|
|
|
s64 val;
|
|
|
|
enum mem_cgroup_stat_index idx;
|
|
|
|
};
|
|
|
|
|
|
|
|
static int
|
|
|
|
mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_idx_data *d = data;
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
d->val += mem_cgroup_read_stat(mem, d->idx);
|
2009-09-23 22:56:42 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
|
|
|
|
enum mem_cgroup_stat_index idx, s64 *val)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_idx_data d;
|
|
|
|
d.idx = idx;
|
|
|
|
d.val = 0;
|
|
|
|
mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat);
|
|
|
|
*val = d.val;
|
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:21 +00:00
|
|
|
static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
|
|
|
|
{
|
|
|
|
u64 idx_val, val;
|
|
|
|
|
|
|
|
if (!mem_cgroup_is_root(mem)) {
|
|
|
|
if (!swap)
|
|
|
|
return res_counter_read_u64(&mem->res, RES_USAGE);
|
|
|
|
else
|
|
|
|
return res_counter_read_u64(&mem->memsw, RES_USAGE);
|
|
|
|
}
|
|
|
|
|
|
|
|
mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val);
|
|
|
|
val = idx_val;
|
|
|
|
mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val);
|
|
|
|
val += idx_val;
|
|
|
|
|
|
|
|
if (swap) {
|
|
|
|
mem_cgroup_get_recursive_idx_stat(mem,
|
|
|
|
MEM_CGROUP_STAT_SWAPOUT, &idx_val);
|
|
|
|
val += idx_val;
|
|
|
|
}
|
|
|
|
|
|
|
|
return val << PAGE_SHIFT;
|
|
|
|
}
|
|
|
|
|
2008-04-29 07:59:58 +00:00
|
|
|
static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
|
2008-02-07 08:13:50 +00:00
|
|
|
{
|
2009-01-08 02:08:00 +00:00
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
|
2010-03-10 23:22:21 +00:00
|
|
|
u64 val;
|
2009-01-08 02:08:00 +00:00
|
|
|
int type, name;
|
|
|
|
|
|
|
|
type = MEMFILE_TYPE(cft->private);
|
|
|
|
name = MEMFILE_ATTR(cft->private);
|
|
|
|
switch (type) {
|
|
|
|
case _MEM:
|
2010-03-10 23:22:21 +00:00
|
|
|
if (name == RES_USAGE)
|
|
|
|
val = mem_cgroup_usage(mem, false);
|
|
|
|
else
|
2009-09-23 22:56:42 +00:00
|
|
|
val = res_counter_read_u64(&mem->res, name);
|
2009-01-08 02:08:00 +00:00
|
|
|
break;
|
|
|
|
case _MEMSWAP:
|
2010-03-10 23:22:21 +00:00
|
|
|
if (name == RES_USAGE)
|
|
|
|
val = mem_cgroup_usage(mem, true);
|
|
|
|
else
|
2009-09-23 22:56:42 +00:00
|
|
|
val = res_counter_read_u64(&mem->memsw, name);
|
2009-01-08 02:08:00 +00:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
BUG();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return val;
|
2008-02-07 08:13:50 +00:00
|
|
|
}
|
2008-07-25 08:47:20 +00:00
|
|
|
/*
|
|
|
|
* The user of this function is...
|
|
|
|
* RES_LIMIT.
|
|
|
|
*/
|
2008-07-25 08:47:04 +00:00
|
|
|
static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
|
|
|
|
const char *buffer)
|
2008-02-07 08:13:50 +00:00
|
|
|
{
|
2008-07-25 08:47:20 +00:00
|
|
|
struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
|
2009-01-08 02:08:00 +00:00
|
|
|
int type, name;
|
2008-07-25 08:47:20 +00:00
|
|
|
unsigned long long val;
|
|
|
|
int ret;
|
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
type = MEMFILE_TYPE(cft->private);
|
|
|
|
name = MEMFILE_ATTR(cft->private);
|
|
|
|
switch (name) {
|
2008-07-25 08:47:20 +00:00
|
|
|
case RES_LIMIT:
|
2009-09-23 22:56:32 +00:00
|
|
|
if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
|
|
|
|
ret = -EINVAL;
|
|
|
|
break;
|
|
|
|
}
|
2008-07-25 08:47:20 +00:00
|
|
|
/* This function does all necessary parse...reuse it */
|
|
|
|
ret = res_counter_memparse_write_strategy(buffer, &val);
|
2009-01-08 02:08:00 +00:00
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
if (type == _MEM)
|
2008-07-25 08:47:20 +00:00
|
|
|
ret = mem_cgroup_resize_limit(memcg, val);
|
2009-01-08 02:08:00 +00:00
|
|
|
else
|
|
|
|
ret = mem_cgroup_resize_memsw_limit(memcg, val);
|
2008-07-25 08:47:20 +00:00
|
|
|
break;
|
2009-09-23 22:56:36 +00:00
|
|
|
case RES_SOFT_LIMIT:
|
|
|
|
ret = res_counter_memparse_write_strategy(buffer, &val);
|
|
|
|
if (ret)
|
|
|
|
break;
|
|
|
|
/*
|
|
|
|
* For memsw, soft limits are hard to implement in terms
|
|
|
|
* of semantics, for now, we support soft limits for
|
|
|
|
* control without swap
|
|
|
|
*/
|
|
|
|
if (type == _MEM)
|
|
|
|
ret = res_counter_set_soft_limit(&memcg->res, val);
|
|
|
|
else
|
|
|
|
ret = -EINVAL;
|
|
|
|
break;
|
2008-07-25 08:47:20 +00:00
|
|
|
default:
|
|
|
|
ret = -EINVAL; /* should be BUG() ? */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return ret;
|
2008-02-07 08:13:50 +00:00
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:26 +00:00
|
|
|
static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg,
|
|
|
|
unsigned long long *mem_limit, unsigned long long *memsw_limit)
|
|
|
|
{
|
|
|
|
struct cgroup *cgroup;
|
|
|
|
unsigned long long min_limit, min_memsw_limit, tmp;
|
|
|
|
|
|
|
|
min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT);
|
|
|
|
min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
|
|
|
|
cgroup = memcg->css.cgroup;
|
|
|
|
if (!memcg->use_hierarchy)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
while (cgroup->parent) {
|
|
|
|
cgroup = cgroup->parent;
|
|
|
|
memcg = mem_cgroup_from_cont(cgroup);
|
|
|
|
if (!memcg->use_hierarchy)
|
|
|
|
break;
|
|
|
|
tmp = res_counter_read_u64(&memcg->res, RES_LIMIT);
|
|
|
|
min_limit = min(min_limit, tmp);
|
|
|
|
tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT);
|
|
|
|
min_memsw_limit = min(min_memsw_limit, tmp);
|
|
|
|
}
|
|
|
|
out:
|
|
|
|
*mem_limit = min_limit;
|
|
|
|
*memsw_limit = min_memsw_limit;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2008-04-29 08:00:21 +00:00
|
|
|
static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
|
2008-04-29 08:00:17 +00:00
|
|
|
{
|
|
|
|
struct mem_cgroup *mem;
|
2009-01-08 02:08:00 +00:00
|
|
|
int type, name;
|
2008-04-29 08:00:17 +00:00
|
|
|
|
|
|
|
mem = mem_cgroup_from_cont(cont);
|
2009-01-08 02:08:00 +00:00
|
|
|
type = MEMFILE_TYPE(event);
|
|
|
|
name = MEMFILE_ATTR(event);
|
|
|
|
switch (name) {
|
2008-04-29 08:00:21 +00:00
|
|
|
case RES_MAX_USAGE:
|
2009-01-08 02:08:00 +00:00
|
|
|
if (type == _MEM)
|
|
|
|
res_counter_reset_max(&mem->res);
|
|
|
|
else
|
|
|
|
res_counter_reset_max(&mem->memsw);
|
2008-04-29 08:00:21 +00:00
|
|
|
break;
|
|
|
|
case RES_FAILCNT:
|
2009-01-08 02:08:00 +00:00
|
|
|
if (type == _MEM)
|
|
|
|
res_counter_reset_failcnt(&mem->res);
|
|
|
|
else
|
|
|
|
res_counter_reset_failcnt(&mem->memsw);
|
2008-04-29 08:00:21 +00:00
|
|
|
break;
|
|
|
|
}
|
2009-09-23 22:56:37 +00:00
|
|
|
|
2008-04-29 08:00:20 +00:00
|
|
|
return 0;
|
2008-04-29 08:00:17 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:13 +00:00
|
|
|
static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp,
|
|
|
|
struct cftype *cft)
|
|
|
|
{
|
|
|
|
return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate;
|
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:17 +00:00
|
|
|
#ifdef CONFIG_MMU
|
2010-03-10 23:22:13 +00:00
|
|
|
static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
|
|
|
|
struct cftype *cft, u64 val)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
|
|
|
|
|
|
|
|
if (val >= (1 << NR_MOVE_TYPE))
|
|
|
|
return -EINVAL;
|
|
|
|
/*
|
|
|
|
* We check this value several times in both in can_attach() and
|
|
|
|
* attach(), so we need cgroup lock to prevent this value from being
|
|
|
|
* inconsistent.
|
|
|
|
*/
|
|
|
|
cgroup_lock();
|
|
|
|
mem->move_charge_at_immigrate = val;
|
|
|
|
cgroup_unlock();
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2010-03-10 23:22:17 +00:00
|
|
|
#else
|
|
|
|
static int mem_cgroup_move_charge_write(struct cgroup *cgrp,
|
|
|
|
struct cftype *cft, u64 val)
|
|
|
|
{
|
|
|
|
return -ENOSYS;
|
|
|
|
}
|
|
|
|
#endif
|
2010-03-10 23:22:13 +00:00
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
|
|
|
|
/* For read statistics */
|
|
|
|
enum {
|
|
|
|
MCS_CACHE,
|
|
|
|
MCS_RSS,
|
2009-12-16 00:47:09 +00:00
|
|
|
MCS_FILE_MAPPED,
|
2009-04-02 23:57:35 +00:00
|
|
|
MCS_PGPGIN,
|
|
|
|
MCS_PGPGOUT,
|
2009-09-23 22:56:43 +00:00
|
|
|
MCS_SWAP,
|
2009-04-02 23:57:35 +00:00
|
|
|
MCS_INACTIVE_ANON,
|
|
|
|
MCS_ACTIVE_ANON,
|
|
|
|
MCS_INACTIVE_FILE,
|
|
|
|
MCS_ACTIVE_FILE,
|
|
|
|
MCS_UNEVICTABLE,
|
|
|
|
NR_MCS_STAT,
|
|
|
|
};
|
|
|
|
|
|
|
|
struct mcs_total_stat {
|
|
|
|
s64 stat[NR_MCS_STAT];
|
2008-02-07 08:14:25 +00:00
|
|
|
};
|
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
struct {
|
|
|
|
char *local_name;
|
|
|
|
char *total_name;
|
|
|
|
} memcg_stat_strings[NR_MCS_STAT] = {
|
|
|
|
{"cache", "total_cache"},
|
|
|
|
{"rss", "total_rss"},
|
2009-06-17 23:26:34 +00:00
|
|
|
{"mapped_file", "total_mapped_file"},
|
2009-04-02 23:57:35 +00:00
|
|
|
{"pgpgin", "total_pgpgin"},
|
|
|
|
{"pgpgout", "total_pgpgout"},
|
2009-09-23 22:56:43 +00:00
|
|
|
{"swap", "total_swap"},
|
2009-04-02 23:57:35 +00:00
|
|
|
{"inactive_anon", "total_inactive_anon"},
|
|
|
|
{"active_anon", "total_active_anon"},
|
|
|
|
{"inactive_file", "total_inactive_file"},
|
|
|
|
{"active_file", "total_active_file"},
|
|
|
|
{"unevictable", "total_unevictable"}
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data)
|
|
|
|
{
|
|
|
|
struct mcs_total_stat *s = data;
|
|
|
|
s64 val;
|
|
|
|
|
|
|
|
/* per cpu stat */
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
|
2009-04-02 23:57:35 +00:00
|
|
|
s->stat[MCS_CACHE] += val * PAGE_SIZE;
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
|
2009-04-02 23:57:35 +00:00
|
|
|
s->stat[MCS_RSS] += val * PAGE_SIZE;
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED);
|
2009-12-16 00:47:09 +00:00
|
|
|
s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE;
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT);
|
2009-04-02 23:57:35 +00:00
|
|
|
s->stat[MCS_PGPGIN] += val;
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT);
|
2009-04-02 23:57:35 +00:00
|
|
|
s->stat[MCS_PGPGOUT] += val;
|
2009-09-23 22:56:43 +00:00
|
|
|
if (do_swap_account) {
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
|
2009-09-23 22:56:43 +00:00
|
|
|
s->stat[MCS_SWAP] += val * PAGE_SIZE;
|
|
|
|
}
|
2009-04-02 23:57:35 +00:00
|
|
|
|
|
|
|
/* per zone stat */
|
|
|
|
val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON);
|
|
|
|
s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE;
|
|
|
|
val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON);
|
|
|
|
s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE;
|
|
|
|
val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE);
|
|
|
|
s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE;
|
|
|
|
val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE);
|
|
|
|
s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE;
|
|
|
|
val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE);
|
|
|
|
s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s)
|
|
|
|
{
|
|
|
|
mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat);
|
|
|
|
}
|
|
|
|
|
2008-04-29 08:00:02 +00:00
|
|
|
static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
|
|
|
|
struct cgroup_map_cb *cb)
|
2008-02-07 08:14:25 +00:00
|
|
|
{
|
|
|
|
struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
|
2009-04-02 23:57:35 +00:00
|
|
|
struct mcs_total_stat mystat;
|
2008-02-07 08:14:25 +00:00
|
|
|
int i;
|
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
memset(&mystat, 0, sizeof(mystat));
|
|
|
|
mem_cgroup_get_local_stat(mem_cont, &mystat);
|
2008-02-07 08:14:25 +00:00
|
|
|
|
2009-09-23 22:56:43 +00:00
|
|
|
for (i = 0; i < NR_MCS_STAT; i++) {
|
|
|
|
if (i == MCS_SWAP && !do_swap_account)
|
|
|
|
continue;
|
2009-04-02 23:57:35 +00:00
|
|
|
cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]);
|
2009-09-23 22:56:43 +00:00
|
|
|
}
|
2008-10-19 03:26:40 +00:00
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
/* Hierarchical information */
|
2009-01-08 02:08:26 +00:00
|
|
|
{
|
|
|
|
unsigned long long limit, memsw_limit;
|
|
|
|
memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit);
|
|
|
|
cb->fill(cb, "hierarchical_memory_limit", limit);
|
|
|
|
if (do_swap_account)
|
|
|
|
cb->fill(cb, "hierarchical_memsw_limit", memsw_limit);
|
|
|
|
}
|
2009-01-08 02:08:22 +00:00
|
|
|
|
2009-04-02 23:57:35 +00:00
|
|
|
memset(&mystat, 0, sizeof(mystat));
|
|
|
|
mem_cgroup_get_total_stat(mem_cont, &mystat);
|
2009-09-23 22:56:43 +00:00
|
|
|
for (i = 0; i < NR_MCS_STAT; i++) {
|
|
|
|
if (i == MCS_SWAP && !do_swap_account)
|
|
|
|
continue;
|
2009-04-02 23:57:35 +00:00
|
|
|
cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]);
|
2009-09-23 22:56:43 +00:00
|
|
|
}
|
2009-04-02 23:57:35 +00:00
|
|
|
|
2009-01-08 02:08:22 +00:00
|
|
|
#ifdef CONFIG_DEBUG_VM
|
2009-01-08 02:08:25 +00:00
|
|
|
cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL));
|
2009-01-08 02:08:22 +00:00
|
|
|
|
|
|
|
{
|
|
|
|
int nid, zid;
|
|
|
|
struct mem_cgroup_per_zone *mz;
|
|
|
|
unsigned long recent_rotated[2] = {0, 0};
|
|
|
|
unsigned long recent_scanned[2] = {0, 0};
|
|
|
|
|
|
|
|
for_each_online_node(nid)
|
|
|
|
for (zid = 0; zid < MAX_NR_ZONES; zid++) {
|
|
|
|
mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
|
|
|
|
|
|
|
|
recent_rotated[0] +=
|
|
|
|
mz->reclaim_stat.recent_rotated[0];
|
|
|
|
recent_rotated[1] +=
|
|
|
|
mz->reclaim_stat.recent_rotated[1];
|
|
|
|
recent_scanned[0] +=
|
|
|
|
mz->reclaim_stat.recent_scanned[0];
|
|
|
|
recent_scanned[1] +=
|
|
|
|
mz->reclaim_stat.recent_scanned[1];
|
|
|
|
}
|
|
|
|
cb->fill(cb, "recent_rotated_anon", recent_rotated[0]);
|
|
|
|
cb->fill(cb, "recent_rotated_file", recent_rotated[1]);
|
|
|
|
cb->fill(cb, "recent_scanned_anon", recent_scanned[0]);
|
|
|
|
cb->fill(cb, "recent_scanned_file", recent_scanned[1]);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2008-02-07 08:14:25 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:24 +00:00
|
|
|
static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
|
|
|
|
|
|
|
|
return get_swappiness(memcg);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft,
|
|
|
|
u64 val)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
|
|
|
|
struct mem_cgroup *parent;
|
2009-01-15 21:51:26 +00:00
|
|
|
|
2009-01-08 02:08:24 +00:00
|
|
|
if (val > 100)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (cgrp->parent == NULL)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
parent = mem_cgroup_from_cont(cgrp->parent);
|
2009-01-15 21:51:26 +00:00
|
|
|
|
|
|
|
cgroup_lock();
|
|
|
|
|
2009-01-08 02:08:24 +00:00
|
|
|
/* If under hierarchy, only empty-root can set this value */
|
|
|
|
if ((parent->use_hierarchy) ||
|
2009-01-15 21:51:26 +00:00
|
|
|
(memcg->use_hierarchy && !list_empty(&cgrp->children))) {
|
|
|
|
cgroup_unlock();
|
2009-01-08 02:08:24 +00:00
|
|
|
return -EINVAL;
|
2009-01-15 21:51:26 +00:00
|
|
|
}
|
2009-01-08 02:08:24 +00:00
|
|
|
|
|
|
|
spin_lock(&memcg->reclaim_param_lock);
|
|
|
|
memcg->swappiness = val;
|
|
|
|
spin_unlock(&memcg->reclaim_param_lock);
|
|
|
|
|
2009-01-15 21:51:26 +00:00
|
|
|
cgroup_unlock();
|
|
|
|
|
2009-01-08 02:08:24 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:24 +00:00
|
|
|
static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_threshold_ary *t;
|
|
|
|
u64 usage;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
if (!swap)
|
2010-05-26 21:42:47 +00:00
|
|
|
t = rcu_dereference(memcg->thresholds.primary);
|
2010-03-10 23:22:24 +00:00
|
|
|
else
|
2010-05-26 21:42:47 +00:00
|
|
|
t = rcu_dereference(memcg->memsw_thresholds.primary);
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
if (!t)
|
|
|
|
goto unlock;
|
|
|
|
|
|
|
|
usage = mem_cgroup_usage(memcg, swap);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* current_threshold points to threshold just below usage.
|
|
|
|
* If it's not true, a threshold was crossed after last
|
|
|
|
* call of __mem_cgroup_threshold().
|
|
|
|
*/
|
2010-05-26 21:42:42 +00:00
|
|
|
i = t->current_threshold;
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Iterate backward over array of thresholds starting from
|
|
|
|
* current_threshold and check if a threshold is crossed.
|
|
|
|
* If none of thresholds below usage is crossed, we read
|
|
|
|
* only one element of the array here.
|
|
|
|
*/
|
|
|
|
for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
|
|
|
|
eventfd_signal(t->entries[i].eventfd, 1);
|
|
|
|
|
|
|
|
/* i = current_threshold + 1 */
|
|
|
|
i++;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Iterate forward over array of thresholds starting from
|
|
|
|
* current_threshold+1 and check if a threshold is crossed.
|
|
|
|
* If none of thresholds above usage is crossed, we read
|
|
|
|
* only one element of the array here.
|
|
|
|
*/
|
|
|
|
for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
|
|
|
|
eventfd_signal(t->entries[i].eventfd, 1);
|
|
|
|
|
|
|
|
/* Update current_threshold */
|
2010-05-26 21:42:42 +00:00
|
|
|
t->current_threshold = i - 1;
|
2010-03-10 23:22:24 +00:00
|
|
|
unlock:
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
|
|
|
|
static void mem_cgroup_threshold(struct mem_cgroup *memcg)
|
|
|
|
{
|
|
|
|
__mem_cgroup_threshold(memcg, false);
|
|
|
|
if (do_swap_account)
|
|
|
|
__mem_cgroup_threshold(memcg, true);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int compare_thresholds(const void *a, const void *b)
|
|
|
|
{
|
|
|
|
const struct mem_cgroup_threshold *_a = a;
|
|
|
|
const struct mem_cgroup_threshold *_b = b;
|
|
|
|
|
|
|
|
return _a->threshold - _b->threshold;
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:36 +00:00
|
|
|
static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem, void *data)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_eventfd_list *ev;
|
|
|
|
|
|
|
|
list_for_each_entry(ev, &mem->oom_notify, list)
|
|
|
|
eventfd_signal(ev->eventfd, 1);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void mem_cgroup_oom_notify(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int mem_cgroup_usage_register_event(struct cgroup *cgrp,
|
|
|
|
struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
|
2010-03-10 23:22:24 +00:00
|
|
|
{
|
|
|
|
struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
|
2010-05-26 21:42:47 +00:00
|
|
|
struct mem_cgroup_thresholds *thresholds;
|
|
|
|
struct mem_cgroup_threshold_ary *new;
|
2010-03-10 23:22:24 +00:00
|
|
|
int type = MEMFILE_TYPE(cft->private);
|
|
|
|
u64 threshold, usage;
|
2010-05-26 21:42:47 +00:00
|
|
|
int i, size, ret;
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
ret = res_counter_memparse_write_strategy(args, &threshold);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
mutex_lock(&memcg->thresholds_lock);
|
2010-05-26 21:42:47 +00:00
|
|
|
|
2010-03-10 23:22:24 +00:00
|
|
|
if (type == _MEM)
|
2010-05-26 21:42:47 +00:00
|
|
|
thresholds = &memcg->thresholds;
|
2010-03-10 23:22:24 +00:00
|
|
|
else if (type == _MEMSWAP)
|
2010-05-26 21:42:47 +00:00
|
|
|
thresholds = &memcg->memsw_thresholds;
|
2010-03-10 23:22:24 +00:00
|
|
|
else
|
|
|
|
BUG();
|
|
|
|
|
|
|
|
usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
|
|
|
|
|
|
|
|
/* Check if a threshold crossed before adding a new one */
|
2010-05-26 21:42:47 +00:00
|
|
|
if (thresholds->primary)
|
2010-03-10 23:22:24 +00:00
|
|
|
__mem_cgroup_threshold(memcg, type == _MEMSWAP);
|
|
|
|
|
2010-05-26 21:42:47 +00:00
|
|
|
size = thresholds->primary ? thresholds->primary->size + 1 : 1;
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
/* Allocate memory for new array of thresholds */
|
2010-05-26 21:42:47 +00:00
|
|
|
new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
|
2010-03-10 23:22:24 +00:00
|
|
|
GFP_KERNEL);
|
2010-05-26 21:42:47 +00:00
|
|
|
if (!new) {
|
2010-03-10 23:22:24 +00:00
|
|
|
ret = -ENOMEM;
|
|
|
|
goto unlock;
|
|
|
|
}
|
2010-05-26 21:42:47 +00:00
|
|
|
new->size = size;
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
/* Copy thresholds (if any) to new array */
|
2010-05-26 21:42:47 +00:00
|
|
|
if (thresholds->primary) {
|
|
|
|
memcpy(new->entries, thresholds->primary->entries, (size - 1) *
|
2010-03-10 23:22:24 +00:00
|
|
|
sizeof(struct mem_cgroup_threshold));
|
2010-05-26 21:42:47 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:24 +00:00
|
|
|
/* Add new threshold */
|
2010-05-26 21:42:47 +00:00
|
|
|
new->entries[size - 1].eventfd = eventfd;
|
|
|
|
new->entries[size - 1].threshold = threshold;
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
/* Sort thresholds. Registering of new threshold isn't time-critical */
|
2010-05-26 21:42:47 +00:00
|
|
|
sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
|
2010-03-10 23:22:24 +00:00
|
|
|
compare_thresholds, NULL);
|
|
|
|
|
|
|
|
/* Find current threshold */
|
2010-05-26 21:42:47 +00:00
|
|
|
new->current_threshold = -1;
|
2010-03-10 23:22:24 +00:00
|
|
|
for (i = 0; i < size; i++) {
|
2010-05-26 21:42:47 +00:00
|
|
|
if (new->entries[i].threshold < usage) {
|
2010-03-10 23:22:24 +00:00
|
|
|
/*
|
2010-05-26 21:42:47 +00:00
|
|
|
* new->current_threshold will not be used until
|
|
|
|
* rcu_assign_pointer(), so it's safe to increment
|
2010-03-10 23:22:24 +00:00
|
|
|
* it here.
|
|
|
|
*/
|
2010-05-26 21:42:47 +00:00
|
|
|
++new->current_threshold;
|
2010-03-10 23:22:24 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:47 +00:00
|
|
|
/* Free old spare buffer and save old primary buffer as spare */
|
|
|
|
kfree(thresholds->spare);
|
|
|
|
thresholds->spare = thresholds->primary;
|
|
|
|
|
|
|
|
rcu_assign_pointer(thresholds->primary, new);
|
2010-03-10 23:22:24 +00:00
|
|
|
|
2010-05-26 21:42:46 +00:00
|
|
|
/* To be sure that nobody uses thresholds */
|
2010-03-10 23:22:24 +00:00
|
|
|
synchronize_rcu();
|
|
|
|
|
|
|
|
unlock:
|
|
|
|
mutex_unlock(&memcg->thresholds_lock);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:46 +00:00
|
|
|
static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp,
|
2010-05-26 21:42:36 +00:00
|
|
|
struct cftype *cft, struct eventfd_ctx *eventfd)
|
2010-03-10 23:22:24 +00:00
|
|
|
{
|
|
|
|
struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
|
2010-05-26 21:42:47 +00:00
|
|
|
struct mem_cgroup_thresholds *thresholds;
|
|
|
|
struct mem_cgroup_threshold_ary *new;
|
2010-03-10 23:22:24 +00:00
|
|
|
int type = MEMFILE_TYPE(cft->private);
|
|
|
|
u64 usage;
|
2010-05-26 21:42:47 +00:00
|
|
|
int i, j, size;
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
mutex_lock(&memcg->thresholds_lock);
|
|
|
|
if (type == _MEM)
|
2010-05-26 21:42:47 +00:00
|
|
|
thresholds = &memcg->thresholds;
|
2010-03-10 23:22:24 +00:00
|
|
|
else if (type == _MEMSWAP)
|
2010-05-26 21:42:47 +00:00
|
|
|
thresholds = &memcg->memsw_thresholds;
|
2010-03-10 23:22:24 +00:00
|
|
|
else
|
|
|
|
BUG();
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Something went wrong if we trying to unregister a threshold
|
|
|
|
* if we don't have thresholds
|
|
|
|
*/
|
|
|
|
BUG_ON(!thresholds);
|
|
|
|
|
|
|
|
usage = mem_cgroup_usage(memcg, type == _MEMSWAP);
|
|
|
|
|
|
|
|
/* Check if a threshold crossed before removing */
|
|
|
|
__mem_cgroup_threshold(memcg, type == _MEMSWAP);
|
|
|
|
|
|
|
|
/* Calculate new number of threshold */
|
2010-05-26 21:42:47 +00:00
|
|
|
size = 0;
|
|
|
|
for (i = 0; i < thresholds->primary->size; i++) {
|
|
|
|
if (thresholds->primary->entries[i].eventfd != eventfd)
|
2010-03-10 23:22:24 +00:00
|
|
|
size++;
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:47 +00:00
|
|
|
new = thresholds->spare;
|
2010-05-26 21:42:46 +00:00
|
|
|
|
2010-03-10 23:22:24 +00:00
|
|
|
/* Set thresholds array to NULL if we don't have thresholds */
|
|
|
|
if (!size) {
|
2010-05-26 21:42:47 +00:00
|
|
|
kfree(new);
|
|
|
|
new = NULL;
|
2010-05-26 21:42:46 +00:00
|
|
|
goto swap_buffers;
|
2010-03-10 23:22:24 +00:00
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:47 +00:00
|
|
|
new->size = size;
|
2010-03-10 23:22:24 +00:00
|
|
|
|
|
|
|
/* Copy thresholds and find current threshold */
|
2010-05-26 21:42:47 +00:00
|
|
|
new->current_threshold = -1;
|
|
|
|
for (i = 0, j = 0; i < thresholds->primary->size; i++) {
|
|
|
|
if (thresholds->primary->entries[i].eventfd == eventfd)
|
2010-03-10 23:22:24 +00:00
|
|
|
continue;
|
|
|
|
|
2010-05-26 21:42:47 +00:00
|
|
|
new->entries[j] = thresholds->primary->entries[i];
|
|
|
|
if (new->entries[j].threshold < usage) {
|
2010-03-10 23:22:24 +00:00
|
|
|
/*
|
2010-05-26 21:42:47 +00:00
|
|
|
* new->current_threshold will not be used
|
2010-03-10 23:22:24 +00:00
|
|
|
* until rcu_assign_pointer(), so it's safe to increment
|
|
|
|
* it here.
|
|
|
|
*/
|
2010-05-26 21:42:47 +00:00
|
|
|
++new->current_threshold;
|
2010-03-10 23:22:24 +00:00
|
|
|
}
|
|
|
|
j++;
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:46 +00:00
|
|
|
swap_buffers:
|
2010-05-26 21:42:47 +00:00
|
|
|
/* Swap primary and spare array */
|
|
|
|
thresholds->spare = thresholds->primary;
|
|
|
|
rcu_assign_pointer(thresholds->primary, new);
|
2010-03-10 23:22:24 +00:00
|
|
|
|
2010-05-26 21:42:46 +00:00
|
|
|
/* To be sure that nobody uses thresholds */
|
2010-03-10 23:22:24 +00:00
|
|
|
synchronize_rcu();
|
|
|
|
|
|
|
|
mutex_unlock(&memcg->thresholds_lock);
|
|
|
|
}
|
2009-01-08 02:07:55 +00:00
|
|
|
|
2010-05-26 21:42:36 +00:00
|
|
|
static int mem_cgroup_oom_register_event(struct cgroup *cgrp,
|
|
|
|
struct cftype *cft, struct eventfd_ctx *eventfd, const char *args)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp);
|
|
|
|
struct mem_cgroup_eventfd_list *event;
|
|
|
|
int type = MEMFILE_TYPE(cft->private);
|
|
|
|
|
|
|
|
BUG_ON(type != _OOM_TYPE);
|
|
|
|
event = kmalloc(sizeof(*event), GFP_KERNEL);
|
|
|
|
if (!event)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
mutex_lock(&memcg_oom_mutex);
|
|
|
|
|
|
|
|
event->eventfd = eventfd;
|
|
|
|
list_add(&event->list, &memcg->oom_notify);
|
|
|
|
|
|
|
|
/* already in OOM ? */
|
|
|
|
if (atomic_read(&memcg->oom_lock))
|
|
|
|
eventfd_signal(eventfd, 1);
|
|
|
|
mutex_unlock(&memcg_oom_mutex);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:46 +00:00
|
|
|
static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp,
|
2010-05-26 21:42:36 +00:00
|
|
|
struct cftype *cft, struct eventfd_ctx *eventfd)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
|
|
|
|
struct mem_cgroup_eventfd_list *ev, *tmp;
|
|
|
|
int type = MEMFILE_TYPE(cft->private);
|
|
|
|
|
|
|
|
BUG_ON(type != _OOM_TYPE);
|
|
|
|
|
|
|
|
mutex_lock(&memcg_oom_mutex);
|
|
|
|
|
|
|
|
list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) {
|
|
|
|
if (ev->eventfd == eventfd) {
|
|
|
|
list_del(&ev->list);
|
|
|
|
kfree(ev);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
mutex_unlock(&memcg_oom_mutex);
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:37 +00:00
|
|
|
static int mem_cgroup_oom_control_read(struct cgroup *cgrp,
|
|
|
|
struct cftype *cft, struct cgroup_map_cb *cb)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
|
|
|
|
|
|
|
|
cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable);
|
|
|
|
|
|
|
|
if (atomic_read(&mem->oom_lock))
|
|
|
|
cb->fill(cb, "under_oom", 1);
|
|
|
|
else
|
|
|
|
cb->fill(cb, "under_oom", 0);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int mem_cgroup_oom_control_write(struct cgroup *cgrp,
|
|
|
|
struct cftype *cft, u64 val)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp);
|
|
|
|
struct mem_cgroup *parent;
|
|
|
|
|
|
|
|
/* cannot set to root cgroup and only 0 and 1 are allowed */
|
|
|
|
if (!cgrp->parent || !((val == 0) || (val == 1)))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
parent = mem_cgroup_from_cont(cgrp->parent);
|
|
|
|
|
|
|
|
cgroup_lock();
|
|
|
|
/* oom-kill-disable is a flag for subhierarchy. */
|
|
|
|
if ((parent->use_hierarchy) ||
|
|
|
|
(mem->use_hierarchy && !list_empty(&cgrp->children))) {
|
|
|
|
cgroup_unlock();
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
mem->oom_kill_disable = val;
|
2010-06-29 22:05:18 +00:00
|
|
|
if (!val)
|
|
|
|
memcg_oom_recover(mem);
|
2010-05-26 21:42:37 +00:00
|
|
|
cgroup_unlock();
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-02-07 08:13:50 +00:00
|
|
|
static struct cftype mem_cgroup_files[] = {
|
|
|
|
{
|
2008-02-07 08:13:57 +00:00
|
|
|
.name = "usage_in_bytes",
|
2009-01-08 02:08:00 +00:00
|
|
|
.private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
|
2008-04-29 07:59:58 +00:00
|
|
|
.read_u64 = mem_cgroup_read,
|
2010-05-26 21:42:36 +00:00
|
|
|
.register_event = mem_cgroup_usage_register_event,
|
|
|
|
.unregister_event = mem_cgroup_usage_unregister_event,
|
2008-02-07 08:13:50 +00:00
|
|
|
},
|
2008-04-29 08:00:17 +00:00
|
|
|
{
|
|
|
|
.name = "max_usage_in_bytes",
|
2009-01-08 02:08:00 +00:00
|
|
|
.private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
|
2008-04-29 08:00:21 +00:00
|
|
|
.trigger = mem_cgroup_reset,
|
2008-04-29 08:00:17 +00:00
|
|
|
.read_u64 = mem_cgroup_read,
|
|
|
|
},
|
2008-02-07 08:13:50 +00:00
|
|
|
{
|
2008-02-07 08:13:57 +00:00
|
|
|
.name = "limit_in_bytes",
|
2009-01-08 02:08:00 +00:00
|
|
|
.private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
|
2008-07-25 08:47:04 +00:00
|
|
|
.write_string = mem_cgroup_write,
|
2008-04-29 07:59:58 +00:00
|
|
|
.read_u64 = mem_cgroup_read,
|
2008-02-07 08:13:50 +00:00
|
|
|
},
|
2009-09-23 22:56:36 +00:00
|
|
|
{
|
|
|
|
.name = "soft_limit_in_bytes",
|
|
|
|
.private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
|
|
|
|
.write_string = mem_cgroup_write,
|
|
|
|
.read_u64 = mem_cgroup_read,
|
|
|
|
},
|
2008-02-07 08:13:50 +00:00
|
|
|
{
|
|
|
|
.name = "failcnt",
|
2009-01-08 02:08:00 +00:00
|
|
|
.private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
|
2008-04-29 08:00:21 +00:00
|
|
|
.trigger = mem_cgroup_reset,
|
2008-04-29 07:59:58 +00:00
|
|
|
.read_u64 = mem_cgroup_read,
|
2008-02-07 08:13:50 +00:00
|
|
|
},
|
2008-02-07 08:14:25 +00:00
|
|
|
{
|
|
|
|
.name = "stat",
|
2008-04-29 08:00:02 +00:00
|
|
|
.read_map = mem_control_stat_show,
|
2008-02-07 08:14:25 +00:00
|
|
|
},
|
2009-01-08 02:07:55 +00:00
|
|
|
{
|
|
|
|
.name = "force_empty",
|
|
|
|
.trigger = mem_cgroup_force_empty_write,
|
|
|
|
},
|
2009-01-08 02:08:07 +00:00
|
|
|
{
|
|
|
|
.name = "use_hierarchy",
|
|
|
|
.write_u64 = mem_cgroup_hierarchy_write,
|
|
|
|
.read_u64 = mem_cgroup_hierarchy_read,
|
|
|
|
},
|
2009-01-08 02:08:24 +00:00
|
|
|
{
|
|
|
|
.name = "swappiness",
|
|
|
|
.read_u64 = mem_cgroup_swappiness_read,
|
|
|
|
.write_u64 = mem_cgroup_swappiness_write,
|
|
|
|
},
|
2010-03-10 23:22:13 +00:00
|
|
|
{
|
|
|
|
.name = "move_charge_at_immigrate",
|
|
|
|
.read_u64 = mem_cgroup_move_charge_read,
|
|
|
|
.write_u64 = mem_cgroup_move_charge_write,
|
|
|
|
},
|
2010-05-26 21:42:36 +00:00
|
|
|
{
|
|
|
|
.name = "oom_control",
|
2010-05-26 21:42:37 +00:00
|
|
|
.read_map = mem_cgroup_oom_control_read,
|
|
|
|
.write_u64 = mem_cgroup_oom_control_write,
|
2010-05-26 21:42:36 +00:00
|
|
|
.register_event = mem_cgroup_oom_register_event,
|
|
|
|
.unregister_event = mem_cgroup_oom_unregister_event,
|
|
|
|
.private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
|
|
|
|
},
|
2008-02-07 08:13:50 +00:00
|
|
|
};
|
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
|
|
|
|
static struct cftype memsw_cgroup_files[] = {
|
|
|
|
{
|
|
|
|
.name = "memsw.usage_in_bytes",
|
|
|
|
.private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
|
|
|
|
.read_u64 = mem_cgroup_read,
|
2010-05-26 21:42:36 +00:00
|
|
|
.register_event = mem_cgroup_usage_register_event,
|
|
|
|
.unregister_event = mem_cgroup_usage_unregister_event,
|
2009-01-08 02:08:00 +00:00
|
|
|
},
|
|
|
|
{
|
|
|
|
.name = "memsw.max_usage_in_bytes",
|
|
|
|
.private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
|
|
|
|
.trigger = mem_cgroup_reset,
|
|
|
|
.read_u64 = mem_cgroup_read,
|
|
|
|
},
|
|
|
|
{
|
|
|
|
.name = "memsw.limit_in_bytes",
|
|
|
|
.private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
|
|
|
|
.write_string = mem_cgroup_write,
|
|
|
|
.read_u64 = mem_cgroup_read,
|
|
|
|
},
|
|
|
|
{
|
|
|
|
.name = "memsw.failcnt",
|
|
|
|
.private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
|
|
|
|
.trigger = mem_cgroup_reset,
|
|
|
|
.read_u64 = mem_cgroup_read,
|
|
|
|
},
|
|
|
|
};
|
|
|
|
|
|
|
|
static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
|
|
|
|
{
|
|
|
|
if (!do_swap_account)
|
|
|
|
return 0;
|
|
|
|
return cgroup_add_files(cont, ss, memsw_cgroup_files,
|
|
|
|
ARRAY_SIZE(memsw_cgroup_files));
|
|
|
|
};
|
|
|
|
#else
|
|
|
|
static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss)
|
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2008-02-07 08:14:31 +00:00
|
|
|
static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_per_node *pn;
|
2008-02-07 08:14:38 +00:00
|
|
|
struct mem_cgroup_per_zone *mz;
|
2008-10-19 03:26:14 +00:00
|
|
|
enum lru_list l;
|
2008-04-09 00:41:54 +00:00
|
|
|
int zone, tmp = node;
|
2008-02-07 08:14:38 +00:00
|
|
|
/*
|
|
|
|
* This routine is called against possible nodes.
|
|
|
|
* But it's BUG to call kmalloc() against offline node.
|
|
|
|
*
|
|
|
|
* TODO: this routine can waste much memory for nodes which will
|
|
|
|
* never be onlined. It's better to use memory hotplug callback
|
|
|
|
* function.
|
|
|
|
*/
|
2008-04-09 00:41:54 +00:00
|
|
|
if (!node_state(node, N_NORMAL_MEMORY))
|
|
|
|
tmp = -1;
|
|
|
|
pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
|
2008-02-07 08:14:31 +00:00
|
|
|
if (!pn)
|
|
|
|
return 1;
|
2008-02-07 08:14:38 +00:00
|
|
|
|
2008-02-07 08:14:31 +00:00
|
|
|
mem->info.nodeinfo[node] = pn;
|
|
|
|
memset(pn, 0, sizeof(*pn));
|
2008-02-07 08:14:38 +00:00
|
|
|
|
|
|
|
for (zone = 0; zone < MAX_NR_ZONES; zone++) {
|
|
|
|
mz = &pn->zoneinfo[zone];
|
2008-10-19 03:26:14 +00:00
|
|
|
for_each_lru(l)
|
|
|
|
INIT_LIST_HEAD(&mz->lists[l]);
|
2009-09-23 22:56:37 +00:00
|
|
|
mz->usage_in_excess = 0;
|
2009-09-23 22:56:39 +00:00
|
|
|
mz->on_tree = false;
|
|
|
|
mz->mem = mem;
|
2008-02-07 08:14:38 +00:00
|
|
|
}
|
2008-02-07 08:14:31 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-02-07 08:14:38 +00:00
|
|
|
static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
|
|
|
|
{
|
|
|
|
kfree(mem->info.nodeinfo[node]);
|
|
|
|
}
|
|
|
|
|
2008-04-29 08:00:24 +00:00
|
|
|
static struct mem_cgroup *mem_cgroup_alloc(void)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem;
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
int size = sizeof(struct mem_cgroup);
|
2008-04-29 08:00:24 +00:00
|
|
|
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
/* Can be very big if MAX_NUMNODES is very big */
|
2009-01-08 02:07:53 +00:00
|
|
|
if (size < PAGE_SIZE)
|
|
|
|
mem = kmalloc(size, GFP_KERNEL);
|
2008-04-29 08:00:24 +00:00
|
|
|
else
|
2009-01-08 02:07:53 +00:00
|
|
|
mem = vmalloc(size);
|
2008-04-29 08:00:24 +00:00
|
|
|
|
2010-03-23 20:35:12 +00:00
|
|
|
if (!mem)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
memset(mem, 0, size);
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
|
|
|
|
if (!mem->stat) {
|
|
|
|
if (size < PAGE_SIZE)
|
|
|
|
kfree(mem);
|
|
|
|
else
|
|
|
|
vfree(mem);
|
|
|
|
mem = NULL;
|
|
|
|
}
|
2008-04-29 08:00:24 +00:00
|
|
|
return mem;
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
/*
|
|
|
|
* At destroying mem_cgroup, references from swap_cgroup can remain.
|
|
|
|
* (scanning all at force_empty is too costly...)
|
|
|
|
*
|
|
|
|
* Instead of clearing all references at force_empty, we remember
|
|
|
|
* the number of reference from swap_cgroup and free mem_cgroup when
|
|
|
|
* it goes down to 0.
|
|
|
|
*
|
|
|
|
* Removal of cgroup itself succeeds regardless of refs from swap.
|
|
|
|
*/
|
|
|
|
|
2009-01-08 02:08:32 +00:00
|
|
|
static void __mem_cgroup_free(struct mem_cgroup *mem)
|
2008-04-29 08:00:24 +00:00
|
|
|
{
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
int node;
|
|
|
|
|
2009-09-23 22:56:37 +00:00
|
|
|
mem_cgroup_remove_from_trees(mem);
|
2009-04-02 23:57:33 +00:00
|
|
|
free_css_id(&mem_cgroup_subsys, &mem->css);
|
|
|
|
|
memcg: synchronized LRU
A big patch for changing memcg's LRU semantics.
Now,
- page_cgroup is linked to mem_cgroup's its own LRU (per zone).
- LRU of page_cgroup is not synchronous with global LRU.
- page and page_cgroup is one-to-one and statically allocated.
- To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
- lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);
- SwapCache is handled.
And, when we handle LRU list of page_cgroup, we do following.
pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);
But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.
This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
1. When pc->mem_cgroup can be modified.
- at charge.
- at account_move().
2. at charge
the PCG_USED bit is not set before pc->mem_cgroup is fixed.
3. at account_move()
the page is isolated and not on LRU.
Pros.
- easy for maintenance.
- memcg can make use of laziness of pagevec.
- we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
- LRU status of memcg will be synchronized with global LRU's one.
- # of locks are reduced.
- account_move() is simplified very much.
Cons.
- may increase cost of LRU rotation.
(no impact if memcg is not configured.)
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 02:08:01 +00:00
|
|
|
for_each_node_state(node, N_POSSIBLE)
|
|
|
|
free_mem_cgroup_per_zone_info(mem, node);
|
|
|
|
|
memcg: use generic percpu instead of private implementation
When per-cpu counter for memcg was implemneted, dynamic percpu allocator
was not very good. But now, we have good one and useful macros. This
patch replaces memcg's private percpu counter implementation with generic
dynamic percpu allocator.
The benefits are
- We can remove private implementation.
- The counters will be NUMA-aware. (Current one is not...)
- This patch makes sizeof struct mem_cgroup smaller. Then,
struct mem_cgroup may be fit in page size on small config.
- About basic performance aspects, see below.
[Before]
# size mm/memcontrol.o
text data bss dec hex filename
24373 2528 4132 31033 7939 mm/memcontrol.o
[page-fault-throuput test on 8cpu/SMP in root cgroup]
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
45878618 page-faults ( +- 0.110% )
602635826 cache-misses ( +- 0.105% )
61.005373262 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 13.14
[After]
#size mm/memcontrol.o
text data bss dec hex filename
23913 2528 4132 30573 776d mm/memcontrol.o
# /root/bin/perf stat -a -e page-faults,cache-misses --repeat 5 ./multi-fault-fork 8
Performance counter stats for './multi-fault-fork 8' (5 runs):
48179400 page-faults ( +- 0.271% )
588628407 cache-misses ( +- 0.136% )
61.004615021 seconds time elapsed ( +- 0.004% )
Then cache-miss/page fault = 12.22
Text size is reduced.
This performance improvement is not big and will be invisible in real world
applications. But this result shows this patch has some good effect even
on (small) SMP.
Here is a test program I used.
1. fork() processes on each cpus.
2. do page fault repeatedly on each process.
3. after 60secs, kill all childredn and exit.
(3 is necessary for getting stable data, this is improvement from previous one.)
#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <signal.h>
#include <stdlib.h>
/*
* For avoiding contention in page table lock, FAULT area is
* sparse. If FAULT_LENGTH is too large for your cpus, decrease it.
*/
#define FAULT_LENGTH (2 * 1024 * 1024)
#define PAGE_SIZE 4096
#define MAXNUM (128)
void alarm_handler(int sig)
{
}
void *worker(int cpu, int ppid)
{
void *start, *end;
char *c;
cpu_set_t set;
int i;
CPU_ZERO(&set);
CPU_SET(cpu, &set);
sched_setaffinity(0, sizeof(set), &set);
start = mmap(NULL, FAULT_LENGTH, PROT_READ|PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
if (start == MAP_FAILED) {
perror("mmap");
exit(1);
}
end = start + FAULT_LENGTH;
pause();
//fprintf(stderr, "run%d", cpu);
while (1) {
for (c = (char*)start; (void *)c < end; c += PAGE_SIZE)
*c = 0;
madvise(start, FAULT_LENGTH, MADV_DONTNEED);
}
return NULL;
}
int main(int argc, char *argv[])
{
int num, i, ret, pid, status;
int pids[MAXNUM];
if (argc < 2)
return 0;
setpgid(0, 0);
signal(SIGALRM, alarm_handler);
num = atoi(argv[1]);
pid = getpid();
for (i = 0; i < num; ++i) {
ret = fork();
if (!ret) {
worker(i, pid);
exit(0);
}
pids[i] = ret;
}
sleep(1);
kill(-pid, SIGALRM);
sleep(60);
for (i = 0; i < num; i++)
kill(pids[i], SIGKILL);
for (i = 0; i < num; i++)
waitpid(pids[i], &status, 0);
return 0;
}
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-10 23:22:29 +00:00
|
|
|
free_percpu(mem->stat);
|
|
|
|
if (sizeof(struct mem_cgroup) < PAGE_SIZE)
|
2008-04-29 08:00:24 +00:00
|
|
|
kfree(mem);
|
|
|
|
else
|
|
|
|
vfree(mem);
|
|
|
|
}
|
|
|
|
|
2009-01-08 02:08:00 +00:00
|
|
|
static void mem_cgroup_get(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
atomic_inc(&mem->refcnt);
|
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:18 +00:00
|
|
|
static void __mem_cgroup_put(struct mem_cgroup *mem, int count)
|
2009-01-08 02:08:00 +00:00
|
|
|
{
|
2010-03-10 23:22:18 +00:00
|
|
|
if (atomic_sub_and_test(count, &mem->refcnt)) {
|
2009-01-29 22:25:11 +00:00
|
|
|
struct mem_cgroup *parent = parent_mem_cgroup(mem);
|
2009-01-08 02:08:32 +00:00
|
|
|
__mem_cgroup_free(mem);
|
2009-01-29 22:25:11 +00:00
|
|
|
if (parent)
|
|
|
|
mem_cgroup_put(parent);
|
|
|
|
}
|
2009-01-08 02:08:00 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:18 +00:00
|
|
|
static void mem_cgroup_put(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
__mem_cgroup_put(mem, 1);
|
|
|
|
}
|
|
|
|
|
2009-01-29 22:25:11 +00:00
|
|
|
/*
|
|
|
|
* Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
|
|
|
|
*/
|
|
|
|
static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem)
|
|
|
|
{
|
|
|
|
if (!mem->res.parent)
|
|
|
|
return NULL;
|
|
|
|
return mem_cgroup_from_res_counter(mem->res.parent, res);
|
|
|
|
}
|
2008-04-29 08:00:24 +00:00
|
|
|
|
2009-01-08 02:07:57 +00:00
|
|
|
#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
|
|
|
|
static void __init enable_swap_cgroup(void)
|
|
|
|
{
|
2009-01-08 02:08:02 +00:00
|
|
|
if (!mem_cgroup_disabled() && really_do_swap_account)
|
2009-01-08 02:07:57 +00:00
|
|
|
do_swap_account = 1;
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
static void __init enable_swap_cgroup(void)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2009-09-23 22:56:37 +00:00
|
|
|
static int mem_cgroup_soft_limit_tree_init(void)
|
|
|
|
{
|
|
|
|
struct mem_cgroup_tree_per_node *rtpn;
|
|
|
|
struct mem_cgroup_tree_per_zone *rtpz;
|
|
|
|
int tmp, node, zone;
|
|
|
|
|
|
|
|
for_each_node_state(node, N_POSSIBLE) {
|
|
|
|
tmp = node;
|
|
|
|
if (!node_state(node, N_NORMAL_MEMORY))
|
|
|
|
tmp = -1;
|
|
|
|
rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp);
|
|
|
|
if (!rtpn)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
soft_limit_tree.rb_tree_per_node[node] = rtpn;
|
|
|
|
|
|
|
|
for (zone = 0; zone < MAX_NR_ZONES; zone++) {
|
|
|
|
rtpz = &rtpn->rb_tree_per_zone[zone];
|
|
|
|
rtpz->rb_root = RB_ROOT;
|
|
|
|
spin_lock_init(&rtpz->lock);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-01-15 21:51:25 +00:00
|
|
|
static struct cgroup_subsys_state * __ref
|
2008-02-07 08:13:50 +00:00
|
|
|
mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
|
|
|
|
{
|
2009-01-08 02:08:05 +00:00
|
|
|
struct mem_cgroup *mem, *parent;
|
2009-04-02 23:57:33 +00:00
|
|
|
long error = -ENOMEM;
|
2008-02-07 08:14:31 +00:00
|
|
|
int node;
|
2008-02-07 08:13:50 +00:00
|
|
|
|
2009-01-08 02:07:53 +00:00
|
|
|
mem = mem_cgroup_alloc();
|
|
|
|
if (!mem)
|
2009-04-02 23:57:33 +00:00
|
|
|
return ERR_PTR(error);
|
2008-02-07 08:13:51 +00:00
|
|
|
|
2008-02-07 08:14:31 +00:00
|
|
|
for_each_node_state(node, N_POSSIBLE)
|
|
|
|
if (alloc_mem_cgroup_per_zone_info(mem, node))
|
|
|
|
goto free_out;
|
2009-09-23 22:56:37 +00:00
|
|
|
|
2009-01-08 02:07:57 +00:00
|
|
|
/* root ? */
|
2009-01-08 02:08:05 +00:00
|
|
|
if (cont->parent == NULL) {
|
2009-12-16 00:47:08 +00:00
|
|
|
int cpu;
|
2009-01-08 02:07:57 +00:00
|
|
|
enable_swap_cgroup();
|
2009-01-08 02:08:05 +00:00
|
|
|
parent = NULL;
|
2009-09-23 22:56:32 +00:00
|
|
|
root_mem_cgroup = mem;
|
2009-09-23 22:56:37 +00:00
|
|
|
if (mem_cgroup_soft_limit_tree_init())
|
|
|
|
goto free_out;
|
2009-12-16 00:47:08 +00:00
|
|
|
for_each_possible_cpu(cpu) {
|
|
|
|
struct memcg_stock_pcp *stock =
|
|
|
|
&per_cpu(memcg_stock, cpu);
|
|
|
|
INIT_WORK(&stock->work, drain_local_stock);
|
|
|
|
}
|
|
|
|
hotcpu_notifier(memcg_stock_cpu_callback, 0);
|
2009-01-08 02:08:07 +00:00
|
|
|
} else {
|
2009-01-08 02:08:05 +00:00
|
|
|
parent = mem_cgroup_from_cont(cont->parent);
|
2009-01-08 02:08:07 +00:00
|
|
|
mem->use_hierarchy = parent->use_hierarchy;
|
2010-05-26 21:42:37 +00:00
|
|
|
mem->oom_kill_disable = parent->oom_kill_disable;
|
2009-01-08 02:08:07 +00:00
|
|
|
}
|
2009-01-08 02:08:05 +00:00
|
|
|
|
2009-01-08 02:08:07 +00:00
|
|
|
if (parent && parent->use_hierarchy) {
|
|
|
|
res_counter_init(&mem->res, &parent->res);
|
|
|
|
res_counter_init(&mem->memsw, &parent->memsw);
|
2009-01-29 22:25:11 +00:00
|
|
|
/*
|
|
|
|
* We increment refcnt of the parent to ensure that we can
|
|
|
|
* safely access it on res_counter_charge/uncharge.
|
|
|
|
* This refcnt will be decremented when freeing this
|
|
|
|
* mem_cgroup(see mem_cgroup_put).
|
|
|
|
*/
|
|
|
|
mem_cgroup_get(parent);
|
2009-01-08 02:08:07 +00:00
|
|
|
} else {
|
|
|
|
res_counter_init(&mem->res, NULL);
|
|
|
|
res_counter_init(&mem->memsw, NULL);
|
|
|
|
}
|
2009-04-02 23:57:33 +00:00
|
|
|
mem->last_scanned_child = 0;
|
2009-01-08 02:08:23 +00:00
|
|
|
spin_lock_init(&mem->reclaim_param_lock);
|
2010-05-26 21:42:36 +00:00
|
|
|
INIT_LIST_HEAD(&mem->oom_notify);
|
2009-01-08 02:08:06 +00:00
|
|
|
|
2009-01-08 02:08:24 +00:00
|
|
|
if (parent)
|
|
|
|
mem->swappiness = get_swappiness(parent);
|
2009-01-08 02:08:32 +00:00
|
|
|
atomic_set(&mem->refcnt, 1);
|
2010-03-10 23:22:13 +00:00
|
|
|
mem->move_charge_at_immigrate = 0;
|
2010-03-10 23:22:24 +00:00
|
|
|
mutex_init(&mem->thresholds_lock);
|
2008-02-07 08:13:50 +00:00
|
|
|
return &mem->css;
|
2008-02-07 08:14:31 +00:00
|
|
|
free_out:
|
2009-01-08 02:08:32 +00:00
|
|
|
__mem_cgroup_free(mem);
|
2009-09-23 22:56:32 +00:00
|
|
|
root_mem_cgroup = NULL;
|
2009-04-02 23:57:33 +00:00
|
|
|
return ERR_PTR(error);
|
2008-02-07 08:13:50 +00:00
|
|
|
}
|
|
|
|
|
2009-04-02 23:57:26 +00:00
|
|
|
static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
|
2008-02-07 08:14:28 +00:00
|
|
|
struct cgroup *cont)
|
|
|
|
{
|
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
|
2009-04-02 23:57:26 +00:00
|
|
|
|
|
|
|
return mem_cgroup_force_empty(mem, false);
|
2008-02-07 08:14:28 +00:00
|
|
|
}
|
|
|
|
|
2008-02-07 08:13:50 +00:00
|
|
|
static void mem_cgroup_destroy(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cont)
|
|
|
|
{
|
2009-01-15 21:51:13 +00:00
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
|
|
|
|
|
|
|
|
mem_cgroup_put(mem);
|
2008-02-07 08:13:50 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static int mem_cgroup_populate(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cont)
|
|
|
|
{
|
2009-01-08 02:08:00 +00:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = cgroup_add_files(cont, ss, mem_cgroup_files,
|
|
|
|
ARRAY_SIZE(mem_cgroup_files));
|
|
|
|
|
|
|
|
if (!ret)
|
|
|
|
ret = register_memsw_files(cont, ss);
|
|
|
|
return ret;
|
2008-02-07 08:13:50 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:17 +00:00
|
|
|
#ifdef CONFIG_MMU
|
2010-03-10 23:22:13 +00:00
|
|
|
/* Handlers for move charge at task migration. */
|
2010-03-10 23:22:15 +00:00
|
|
|
#define PRECHARGE_COUNT_AT_ONCE 256
|
|
|
|
static int mem_cgroup_do_precharge(unsigned long count)
|
2010-03-10 23:22:13 +00:00
|
|
|
{
|
2010-03-10 23:22:15 +00:00
|
|
|
int ret = 0;
|
|
|
|
int batch_count = PRECHARGE_COUNT_AT_ONCE;
|
2010-03-10 23:22:14 +00:00
|
|
|
struct mem_cgroup *mem = mc.to;
|
|
|
|
|
2010-03-10 23:22:15 +00:00
|
|
|
if (mem_cgroup_is_root(mem)) {
|
|
|
|
mc.precharge += count;
|
|
|
|
/* we don't need css_get for root */
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
/* try to charge at once */
|
|
|
|
if (count > 1) {
|
|
|
|
struct res_counter *dummy;
|
|
|
|
/*
|
|
|
|
* "mem" cannot be under rmdir() because we've already checked
|
|
|
|
* by cgroup_lock_live_cgroup() that it is not removed and we
|
|
|
|
* are still under the same cgroup_mutex. So we can postpone
|
|
|
|
* css_get().
|
|
|
|
*/
|
|
|
|
if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy))
|
|
|
|
goto one_by_one;
|
|
|
|
if (do_swap_account && res_counter_charge(&mem->memsw,
|
|
|
|
PAGE_SIZE * count, &dummy)) {
|
|
|
|
res_counter_uncharge(&mem->res, PAGE_SIZE * count);
|
|
|
|
goto one_by_one;
|
|
|
|
}
|
|
|
|
mc.precharge += count;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
one_by_one:
|
|
|
|
/* fall back to one by one charge */
|
|
|
|
while (count--) {
|
|
|
|
if (signal_pending(current)) {
|
|
|
|
ret = -EINTR;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (!batch_count--) {
|
|
|
|
batch_count = PRECHARGE_COUNT_AT_ONCE;
|
|
|
|
cond_resched();
|
|
|
|
}
|
2010-03-10 23:22:30 +00:00
|
|
|
ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
|
2010-03-10 23:22:15 +00:00
|
|
|
if (ret || !mem)
|
|
|
|
/* mem_cgroup_clear_mc() will do uncharge later */
|
|
|
|
return -ENOMEM;
|
|
|
|
mc.precharge++;
|
|
|
|
}
|
2010-03-10 23:22:14 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* is_target_pte_for_mc - check a pte whether it is valid for move charge
|
|
|
|
* @vma: the vma the pte to be checked belongs
|
|
|
|
* @addr: the address corresponding to the pte to be checked
|
|
|
|
* @ptent: the pte to be checked
|
2010-03-10 23:22:17 +00:00
|
|
|
* @target: the pointer the target page or swap ent will be stored(can be NULL)
|
2010-03-10 23:22:14 +00:00
|
|
|
*
|
|
|
|
* Returns
|
|
|
|
* 0(MC_TARGET_NONE): if the pte is not a target for move charge.
|
|
|
|
* 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
|
|
|
|
* move charge. if @target is not NULL, the page is stored in target->page
|
|
|
|
* with extra refcnt got(Callers should handle it).
|
2010-03-10 23:22:17 +00:00
|
|
|
* 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
|
|
|
|
* target for charge migration. if @target is not NULL, the entry is stored
|
|
|
|
* in target->ent.
|
2010-03-10 23:22:14 +00:00
|
|
|
*
|
|
|
|
* Called with pte lock held.
|
|
|
|
*/
|
|
|
|
union mc_target {
|
|
|
|
struct page *page;
|
2010-03-10 23:22:17 +00:00
|
|
|
swp_entry_t ent;
|
2010-03-10 23:22:14 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
enum mc_target_type {
|
|
|
|
MC_TARGET_NONE, /* not used */
|
|
|
|
MC_TARGET_PAGE,
|
2010-03-10 23:22:17 +00:00
|
|
|
MC_TARGET_SWAP,
|
2010-03-10 23:22:14 +00:00
|
|
|
};
|
|
|
|
|
2010-05-26 21:42:38 +00:00
|
|
|
static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
|
|
|
|
unsigned long addr, pte_t ptent)
|
2010-03-10 23:22:14 +00:00
|
|
|
{
|
2010-05-26 21:42:38 +00:00
|
|
|
struct page *page = vm_normal_page(vma, addr, ptent);
|
2010-03-10 23:22:14 +00:00
|
|
|
|
2010-05-26 21:42:38 +00:00
|
|
|
if (!page || !page_mapped(page))
|
|
|
|
return NULL;
|
|
|
|
if (PageAnon(page)) {
|
|
|
|
/* we don't move shared anon */
|
|
|
|
if (!move_anon() || page_mapcount(page) > 2)
|
|
|
|
return NULL;
|
2010-05-26 21:42:39 +00:00
|
|
|
} else if (!move_file())
|
|
|
|
/* we ignore mapcount for file pages */
|
2010-05-26 21:42:38 +00:00
|
|
|
return NULL;
|
|
|
|
if (!get_page_unless_zero(page))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return page;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
|
|
|
|
unsigned long addr, pte_t ptent, swp_entry_t *entry)
|
|
|
|
{
|
|
|
|
int usage_count;
|
|
|
|
struct page *page = NULL;
|
|
|
|
swp_entry_t ent = pte_to_swp_entry(ptent);
|
|
|
|
|
|
|
|
if (!move_anon() || non_swap_entry(ent))
|
|
|
|
return NULL;
|
|
|
|
usage_count = mem_cgroup_count_swap_user(ent, &page);
|
|
|
|
if (usage_count > 1) { /* we don't move shared anon */
|
2010-03-10 23:22:17 +00:00
|
|
|
if (page)
|
|
|
|
put_page(page);
|
2010-05-26 21:42:38 +00:00
|
|
|
return NULL;
|
2010-03-10 23:22:17 +00:00
|
|
|
}
|
2010-05-26 21:42:38 +00:00
|
|
|
if (do_swap_account)
|
|
|
|
entry->val = ent.val;
|
|
|
|
|
|
|
|
return page;
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:39 +00:00
|
|
|
static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
|
|
|
|
unsigned long addr, pte_t ptent, swp_entry_t *entry)
|
|
|
|
{
|
|
|
|
struct page *page = NULL;
|
|
|
|
struct inode *inode;
|
|
|
|
struct address_space *mapping;
|
|
|
|
pgoff_t pgoff;
|
|
|
|
|
|
|
|
if (!vma->vm_file) /* anonymous vma */
|
|
|
|
return NULL;
|
|
|
|
if (!move_file())
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
inode = vma->vm_file->f_path.dentry->d_inode;
|
|
|
|
mapping = vma->vm_file->f_mapping;
|
|
|
|
if (pte_none(ptent))
|
|
|
|
pgoff = linear_page_index(vma, addr);
|
|
|
|
else /* pte_file(ptent) is true */
|
|
|
|
pgoff = pte_to_pgoff(ptent);
|
|
|
|
|
|
|
|
/* page is moved even if it's not RSS of this task(page-faulted). */
|
|
|
|
if (!mapping_cap_swap_backed(mapping)) { /* normal file */
|
|
|
|
page = find_get_page(mapping, pgoff);
|
|
|
|
} else { /* shmem/tmpfs file. we should take account of swap too. */
|
|
|
|
swp_entry_t ent;
|
|
|
|
mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent);
|
|
|
|
if (do_swap_account)
|
|
|
|
entry->val = ent.val;
|
|
|
|
}
|
|
|
|
|
|
|
|
return page;
|
|
|
|
}
|
|
|
|
|
2010-05-26 21:42:38 +00:00
|
|
|
static int is_target_pte_for_mc(struct vm_area_struct *vma,
|
|
|
|
unsigned long addr, pte_t ptent, union mc_target *target)
|
|
|
|
{
|
|
|
|
struct page *page = NULL;
|
|
|
|
struct page_cgroup *pc;
|
|
|
|
int ret = 0;
|
|
|
|
swp_entry_t ent = { .val = 0 };
|
|
|
|
|
|
|
|
if (pte_present(ptent))
|
|
|
|
page = mc_handle_present_pte(vma, addr, ptent);
|
|
|
|
else if (is_swap_pte(ptent))
|
|
|
|
page = mc_handle_swap_pte(vma, addr, ptent, &ent);
|
2010-05-26 21:42:39 +00:00
|
|
|
else if (pte_none(ptent) || pte_file(ptent))
|
|
|
|
page = mc_handle_file_pte(vma, addr, ptent, &ent);
|
2010-05-26 21:42:38 +00:00
|
|
|
|
|
|
|
if (!page && !ent.val)
|
|
|
|
return 0;
|
2010-03-10 23:22:17 +00:00
|
|
|
if (page) {
|
|
|
|
pc = lookup_page_cgroup(page);
|
|
|
|
/*
|
|
|
|
* Do only loose check w/o page_cgroup lock.
|
|
|
|
* mem_cgroup_move_account() checks the pc is valid or not under
|
|
|
|
* the lock.
|
|
|
|
*/
|
|
|
|
if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
|
|
|
|
ret = MC_TARGET_PAGE;
|
|
|
|
if (target)
|
|
|
|
target->page = page;
|
|
|
|
}
|
|
|
|
if (!ret || !target)
|
|
|
|
put_page(page);
|
|
|
|
}
|
2010-05-26 21:42:38 +00:00
|
|
|
/* There is a swap entry and a page doesn't exist or isn't charged */
|
|
|
|
if (ent.val && !ret &&
|
2010-05-11 21:06:58 +00:00
|
|
|
css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
|
|
|
|
ret = MC_TARGET_SWAP;
|
|
|
|
if (target)
|
|
|
|
target->ent = ent;
|
2010-03-10 23:22:14 +00:00
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
|
|
|
|
unsigned long addr, unsigned long end,
|
|
|
|
struct mm_walk *walk)
|
|
|
|
{
|
|
|
|
struct vm_area_struct *vma = walk->private;
|
|
|
|
pte_t *pte;
|
|
|
|
spinlock_t *ptl;
|
|
|
|
|
|
|
|
pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
|
|
|
|
for (; addr != end; pte++, addr += PAGE_SIZE)
|
|
|
|
if (is_target_pte_for_mc(vma, addr, *pte, NULL))
|
|
|
|
mc.precharge++; /* increment precharge temporarily */
|
|
|
|
pte_unmap_unlock(pte - 1, ptl);
|
|
|
|
cond_resched();
|
|
|
|
|
2010-03-10 23:22:13 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:14 +00:00
|
|
|
static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
|
|
|
|
{
|
|
|
|
unsigned long precharge;
|
|
|
|
struct vm_area_struct *vma;
|
|
|
|
|
|
|
|
down_read(&mm->mmap_sem);
|
|
|
|
for (vma = mm->mmap; vma; vma = vma->vm_next) {
|
|
|
|
struct mm_walk mem_cgroup_count_precharge_walk = {
|
|
|
|
.pmd_entry = mem_cgroup_count_precharge_pte_range,
|
|
|
|
.mm = mm,
|
|
|
|
.private = vma,
|
|
|
|
};
|
|
|
|
if (is_vm_hugetlb_page(vma))
|
|
|
|
continue;
|
|
|
|
walk_page_range(vma->vm_start, vma->vm_end,
|
|
|
|
&mem_cgroup_count_precharge_walk);
|
|
|
|
}
|
|
|
|
up_read(&mm->mmap_sem);
|
|
|
|
|
|
|
|
precharge = mc.precharge;
|
|
|
|
mc.precharge = 0;
|
|
|
|
|
|
|
|
return precharge;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int mem_cgroup_precharge_mc(struct mm_struct *mm)
|
|
|
|
{
|
2010-03-10 23:22:15 +00:00
|
|
|
return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm));
|
2010-03-10 23:22:14 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void mem_cgroup_clear_mc(void)
|
|
|
|
{
|
2010-08-11 01:02:58 +00:00
|
|
|
struct mem_cgroup *from = mc.from;
|
|
|
|
struct mem_cgroup *to = mc.to;
|
|
|
|
|
2010-03-10 23:22:14 +00:00
|
|
|
/* we must uncharge all the leftover precharges from mc.to */
|
2010-03-10 23:22:15 +00:00
|
|
|
if (mc.precharge) {
|
|
|
|
__mem_cgroup_cancel_charge(mc.to, mc.precharge);
|
|
|
|
mc.precharge = 0;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* we didn't uncharge from mc.from at mem_cgroup_move_account(), so
|
|
|
|
* we must uncharge here.
|
|
|
|
*/
|
|
|
|
if (mc.moved_charge) {
|
|
|
|
__mem_cgroup_cancel_charge(mc.from, mc.moved_charge);
|
|
|
|
mc.moved_charge = 0;
|
2010-03-10 23:22:14 +00:00
|
|
|
}
|
2010-03-10 23:22:18 +00:00
|
|
|
/* we must fixup refcnts and charges */
|
|
|
|
if (mc.moved_swap) {
|
|
|
|
/* uncharge swap account from the old cgroup */
|
|
|
|
if (!mem_cgroup_is_root(mc.from))
|
|
|
|
res_counter_uncharge(&mc.from->memsw,
|
|
|
|
PAGE_SIZE * mc.moved_swap);
|
|
|
|
__mem_cgroup_put(mc.from, mc.moved_swap);
|
|
|
|
|
|
|
|
if (!mem_cgroup_is_root(mc.to)) {
|
|
|
|
/*
|
|
|
|
* we charged both to->res and to->memsw, so we should
|
|
|
|
* uncharge to->res.
|
|
|
|
*/
|
|
|
|
res_counter_uncharge(&mc.to->res,
|
|
|
|
PAGE_SIZE * mc.moved_swap);
|
|
|
|
}
|
|
|
|
/* we've already done mem_cgroup_get(mc.to) */
|
|
|
|
|
|
|
|
mc.moved_swap = 0;
|
|
|
|
}
|
2010-08-11 01:02:58 +00:00
|
|
|
spin_lock(&mc.lock);
|
2010-03-10 23:22:14 +00:00
|
|
|
mc.from = NULL;
|
|
|
|
mc.to = NULL;
|
2010-03-10 23:22:16 +00:00
|
|
|
mc.moving_task = NULL;
|
2010-08-11 01:02:58 +00:00
|
|
|
spin_unlock(&mc.lock);
|
|
|
|
memcg_oom_recover(from);
|
|
|
|
memcg_oom_recover(to);
|
2010-03-10 23:22:16 +00:00
|
|
|
wake_up_all(&mc.waitq);
|
2010-03-10 23:22:14 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:13 +00:00
|
|
|
static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cgroup,
|
|
|
|
struct task_struct *p,
|
|
|
|
bool threadgroup)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup);
|
|
|
|
|
|
|
|
if (mem->move_charge_at_immigrate) {
|
|
|
|
struct mm_struct *mm;
|
|
|
|
struct mem_cgroup *from = mem_cgroup_from_task(p);
|
|
|
|
|
|
|
|
VM_BUG_ON(from == mem);
|
|
|
|
|
|
|
|
mm = get_task_mm(p);
|
|
|
|
if (!mm)
|
|
|
|
return 0;
|
|
|
|
/* We move charges only when we move a owner of the mm */
|
2010-03-10 23:22:14 +00:00
|
|
|
if (mm->owner == p) {
|
|
|
|
VM_BUG_ON(mc.from);
|
|
|
|
VM_BUG_ON(mc.to);
|
|
|
|
VM_BUG_ON(mc.precharge);
|
2010-03-10 23:22:15 +00:00
|
|
|
VM_BUG_ON(mc.moved_charge);
|
2010-03-10 23:22:18 +00:00
|
|
|
VM_BUG_ON(mc.moved_swap);
|
2010-03-10 23:22:16 +00:00
|
|
|
VM_BUG_ON(mc.moving_task);
|
2010-08-11 01:02:58 +00:00
|
|
|
spin_lock(&mc.lock);
|
2010-03-10 23:22:14 +00:00
|
|
|
mc.from = from;
|
|
|
|
mc.to = mem;
|
|
|
|
mc.precharge = 0;
|
2010-03-10 23:22:15 +00:00
|
|
|
mc.moved_charge = 0;
|
2010-03-10 23:22:18 +00:00
|
|
|
mc.moved_swap = 0;
|
2010-03-10 23:22:16 +00:00
|
|
|
mc.moving_task = current;
|
2010-08-11 01:02:58 +00:00
|
|
|
spin_unlock(&mc.lock);
|
2010-03-10 23:22:14 +00:00
|
|
|
|
|
|
|
ret = mem_cgroup_precharge_mc(mm);
|
|
|
|
if (ret)
|
|
|
|
mem_cgroup_clear_mc();
|
|
|
|
}
|
2010-03-10 23:22:13 +00:00
|
|
|
mmput(mm);
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cgroup,
|
|
|
|
struct task_struct *p,
|
|
|
|
bool threadgroup)
|
|
|
|
{
|
2010-03-10 23:22:14 +00:00
|
|
|
mem_cgroup_clear_mc();
|
2010-03-10 23:22:13 +00:00
|
|
|
}
|
|
|
|
|
2010-03-10 23:22:14 +00:00
|
|
|
static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
|
|
|
|
unsigned long addr, unsigned long end,
|
|
|
|
struct mm_walk *walk)
|
2010-03-10 23:22:13 +00:00
|
|
|
{
|
2010-03-10 23:22:14 +00:00
|
|
|
int ret = 0;
|
|
|
|
struct vm_area_struct *vma = walk->private;
|
|
|
|
pte_t *pte;
|
|
|
|
spinlock_t *ptl;
|
|
|
|
|
|
|
|
retry:
|
|
|
|
pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
|
|
|
|
for (; addr != end; addr += PAGE_SIZE) {
|
|
|
|
pte_t ptent = *(pte++);
|
|
|
|
union mc_target target;
|
|
|
|
int type;
|
|
|
|
struct page *page;
|
|
|
|
struct page_cgroup *pc;
|
2010-03-10 23:22:17 +00:00
|
|
|
swp_entry_t ent;
|
2010-03-10 23:22:14 +00:00
|
|
|
|
|
|
|
if (!mc.precharge)
|
|
|
|
break;
|
|
|
|
|
|
|
|
type = is_target_pte_for_mc(vma, addr, ptent, &target);
|
|
|
|
switch (type) {
|
|
|
|
case MC_TARGET_PAGE:
|
|
|
|
page = target.page;
|
|
|
|
if (isolate_lru_page(page))
|
|
|
|
goto put;
|
|
|
|
pc = lookup_page_cgroup(page);
|
2010-03-10 23:22:15 +00:00
|
|
|
if (!mem_cgroup_move_account(pc,
|
|
|
|
mc.from, mc.to, false)) {
|
2010-03-10 23:22:14 +00:00
|
|
|
mc.precharge--;
|
2010-03-10 23:22:15 +00:00
|
|
|
/* we uncharge from mc.from later. */
|
|
|
|
mc.moved_charge++;
|
2010-03-10 23:22:14 +00:00
|
|
|
}
|
|
|
|
putback_lru_page(page);
|
|
|
|
put: /* is_target_pte_for_mc() gets the page */
|
|
|
|
put_page(page);
|
|
|
|
break;
|
2010-03-10 23:22:17 +00:00
|
|
|
case MC_TARGET_SWAP:
|
|
|
|
ent = target.ent;
|
2010-03-10 23:22:18 +00:00
|
|
|
if (!mem_cgroup_move_swap_account(ent,
|
|
|
|
mc.from, mc.to, false)) {
|
2010-03-10 23:22:17 +00:00
|
|
|
mc.precharge--;
|
2010-03-10 23:22:18 +00:00
|
|
|
/* we fixup refcnts and charges later. */
|
|
|
|
mc.moved_swap++;
|
|
|
|
}
|
2010-03-10 23:22:17 +00:00
|
|
|
break;
|
2010-03-10 23:22:14 +00:00
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
pte_unmap_unlock(pte - 1, ptl);
|
|
|
|
cond_resched();
|
|
|
|
|
|
|
|
if (addr != end) {
|
|
|
|
/*
|
|
|
|
* We have consumed all precharges we got in can_attach().
|
|
|
|
* We try charge one by one, but don't do any additional
|
|
|
|
* charges to mc.to if we have failed in charge once in attach()
|
|
|
|
* phase.
|
|
|
|
*/
|
2010-03-10 23:22:15 +00:00
|
|
|
ret = mem_cgroup_do_precharge(1);
|
2010-03-10 23:22:14 +00:00
|
|
|
if (!ret)
|
|
|
|
goto retry;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void mem_cgroup_move_charge(struct mm_struct *mm)
|
|
|
|
{
|
|
|
|
struct vm_area_struct *vma;
|
|
|
|
|
|
|
|
lru_add_drain_all();
|
|
|
|
down_read(&mm->mmap_sem);
|
|
|
|
for (vma = mm->mmap; vma; vma = vma->vm_next) {
|
|
|
|
int ret;
|
|
|
|
struct mm_walk mem_cgroup_move_charge_walk = {
|
|
|
|
.pmd_entry = mem_cgroup_move_charge_pte_range,
|
|
|
|
.mm = mm,
|
|
|
|
.private = vma,
|
|
|
|
};
|
|
|
|
if (is_vm_hugetlb_page(vma))
|
|
|
|
continue;
|
|
|
|
ret = walk_page_range(vma->vm_start, vma->vm_end,
|
|
|
|
&mem_cgroup_move_charge_walk);
|
|
|
|
if (ret)
|
|
|
|
/*
|
|
|
|
* means we have consumed all precharges and failed in
|
|
|
|
* doing additional charge. Just abandon here.
|
|
|
|
*/
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
up_read(&mm->mmap_sem);
|
2010-03-10 23:22:13 +00:00
|
|
|
}
|
|
|
|
|
2008-02-07 08:13:54 +00:00
|
|
|
static void mem_cgroup_move_task(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cont,
|
|
|
|
struct cgroup *old_cont,
|
2009-09-23 22:56:31 +00:00
|
|
|
struct task_struct *p,
|
|
|
|
bool threadgroup)
|
2008-02-07 08:13:54 +00:00
|
|
|
{
|
2010-03-10 23:22:14 +00:00
|
|
|
struct mm_struct *mm;
|
|
|
|
|
|
|
|
if (!mc.to)
|
|
|
|
/* no need to move charge */
|
|
|
|
return;
|
|
|
|
|
|
|
|
mm = get_task_mm(p);
|
|
|
|
if (mm) {
|
|
|
|
mem_cgroup_move_charge(mm);
|
|
|
|
mmput(mm);
|
|
|
|
}
|
|
|
|
mem_cgroup_clear_mc();
|
2008-02-07 08:13:54 +00:00
|
|
|
}
|
2010-03-23 20:35:11 +00:00
|
|
|
#else /* !CONFIG_MMU */
|
|
|
|
static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cgroup,
|
|
|
|
struct task_struct *p,
|
|
|
|
bool threadgroup)
|
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cgroup,
|
|
|
|
struct task_struct *p,
|
|
|
|
bool threadgroup)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
static void mem_cgroup_move_task(struct cgroup_subsys *ss,
|
|
|
|
struct cgroup *cont,
|
|
|
|
struct cgroup *old_cont,
|
|
|
|
struct task_struct *p,
|
|
|
|
bool threadgroup)
|
|
|
|
{
|
|
|
|
}
|
|
|
|
#endif
|
2008-02-07 08:13:54 +00:00
|
|
|
|
2008-02-07 08:13:50 +00:00
|
|
|
struct cgroup_subsys mem_cgroup_subsys = {
|
|
|
|
.name = "memory",
|
|
|
|
.subsys_id = mem_cgroup_subsys_id,
|
|
|
|
.create = mem_cgroup_create,
|
2008-02-07 08:14:28 +00:00
|
|
|
.pre_destroy = mem_cgroup_pre_destroy,
|
2008-02-07 08:13:50 +00:00
|
|
|
.destroy = mem_cgroup_destroy,
|
|
|
|
.populate = mem_cgroup_populate,
|
2010-03-10 23:22:13 +00:00
|
|
|
.can_attach = mem_cgroup_can_attach,
|
|
|
|
.cancel_attach = mem_cgroup_cancel_attach,
|
2008-02-07 08:13:54 +00:00
|
|
|
.attach = mem_cgroup_move_task,
|
2008-02-07 08:14:31 +00:00
|
|
|
.early_init = 0,
|
2009-04-02 23:57:33 +00:00
|
|
|
.use_id = 1,
|
2008-02-07 08:13:50 +00:00
|
|
|
};
|
2009-01-08 02:07:57 +00:00
|
|
|
|
|
|
|
#ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
|
|
|
|
|
|
|
|
static int __init disable_swap_account(char *s)
|
|
|
|
{
|
|
|
|
really_do_swap_account = 0;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
__setup("noswapaccount", disable_swap_account);
|
|
|
|
#endif
|