Commit 49854662 authored by Liu Shixin's avatar Liu Shixin Committed by Ma Wupeng
Browse files

mm/page_alloc.c: add sysctl to revise the batch and high of percpu pageset 

hulk inclusion
category: performance
bugzilla: 187468, https://gitee.com/openeuler/kernel/issues/I61HVC



--------------------------------

Patch d8a759b5 ("mm, page_alloc: double zone's batchsize") change the
default batchsize. For some machines with large memory, the value seems to
be too large. Although percpu_pagelist_fraction can be used to revise the
batchsize but needs to be adjusted based on managed memory.
So add a new sysctl percpu_max_batchsize to revise the batchsize to adapt
different scenarios.

Signed-off-by: default avatarLiu Shixin <liushixin2@huawei.com>
parent 1e44290e

Documentation/admin-guide/sysctl/vm.rst

+10 −0
Original line number Diff line number Diff line
@@ -65,6 +65,7 @@ Currently, these files are in /proc/sys/vm: 
- page-cluster

- panic_on_oom

- percpu_pagelist_fraction

- percpu_max_batchsize

- stat_interval

- stat_refresh

- numa_stat
@@ -856,6 +857,15 @@ the high water marks for each per cpu page list. If the user writes '0' to this 
sysctl, it will revert to this default behavior.





percpu_max_batchsize

========================



This is used to setup the max batch and high size of percpu in each zone.

The default value is set to (256 * 1024) / PAGE_SIZE.

The max value is limited to (512 * 1024) / PAGE_SIZE.

The min value is limited to (64 * 1024) / PAGE_SIZE.





stat_interval

=============

include/linux/mmzone.h

+3 −0
Original line number Diff line number Diff line
@@ -1009,6 +1009,8 @@ int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *, 
		size_t *, loff_t *);

int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,

		void *, size_t *, loff_t *);

int percpu_max_batchsize_sysctl_handler(struct ctl_table *, int,

		void *, size_t *, loff_t *);

int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,

		void *, size_t *, loff_t *);

int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
@@ -1016,6 +1018,7 @@ int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, 
int numa_zonelist_order_handler(struct ctl_table *, int,

		void *, size_t *, loff_t *);

extern int percpu_pagelist_fraction;

extern int percpu_max_batchsize;

extern char numa_zonelist_order[];

#define NUMA_ZONELIST_ORDER_LEN	16

kernel/sysctl.c

+8 −0
Original line number Diff line number Diff line
@@ -2993,6 +2993,14 @@ static struct ctl_table vm_table[] = { 
		.proc_handler	= percpu_pagelist_fraction_sysctl_handler,

		.extra1		= SYSCTL_ZERO,

	},

	{

		.procname	= "percpu_max_batchsize",

		.data		= &percpu_max_batchsize,

		.maxlen		= sizeof(percpu_max_batchsize),

		.mode		= 0644,

		.proc_handler	= percpu_max_batchsize_sysctl_handler,

		.extra1		= SYSCTL_ZERO,

	},

	{

		.procname	= "page_lock_unfairness",

		.data		= &sysctl_page_lock_unfairness,

mm/page_alloc.c

+39 −3
Original line number Diff line number Diff line
@@ -112,6 +112,8 @@ typedef int __bitwise fpi_t; 
/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */

static DEFINE_MUTEX(pcp_batch_high_lock);

#define MIN_PERCPU_PAGELIST_FRACTION	(8)

#define MAX_PERCPU_MAX_BATCHSIZE	((512 * 1024) / PAGE_SIZE)

#define MIN_PERCPU_MAX_BATCHSIZE	(MAX_PERCPU_MAX_BATCHSIZE / 8)



#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID

DEFINE_PER_CPU(int, numa_node);
@@ -167,6 +169,8 @@ unsigned long totalreserve_pages __read_mostly; 
unsigned long totalcma_pages __read_mostly;



int percpu_pagelist_fraction;

int percpu_max_batchsize = MAX_PERCPU_MAX_BATCHSIZE / 2;



gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;

#ifdef CONFIG_INIT_ON_ALLOC_DEFAULT_ON

DEFINE_STATIC_KEY_TRUE(init_on_alloc);
@@ -6757,10 +6761,9 @@ static int zone_batchsize(struct zone *zone) 
	 * size of the zone.

	 */

	batch = zone_managed_pages(zone) / 1024;

	/* But no more than a meg. */

	if (batch * PAGE_SIZE > 1024 * 1024)

		batch = (1024 * 1024) / PAGE_SIZE;

	batch /= 4;		/* We effectively *= 4 below */

	if (batch > percpu_max_batchsize)

		batch = percpu_max_batchsize;

	if (batch < 1)

		batch = 1;
@@ -8615,6 +8618,39 @@ int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write, 
	return ret;

}



int percpu_max_batchsize_sysctl_handler(struct ctl_table *table, int write,

		void *buffer, size_t *length, loff_t *ppos)

{

	struct zone *zone;

	int old_percpu_max_batchsize;

	int ret;



	mutex_lock(&pcp_batch_high_lock);

	old_percpu_max_batchsize = percpu_max_batchsize;



	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);

	if (!write || ret < 0)

		goto out;



	/* Sanity checking to avoid pcp imbalance */

	if (percpu_max_batchsize > MAX_PERCPU_MAX_BATCHSIZE ||

	    percpu_max_batchsize < MIN_PERCPU_MAX_BATCHSIZE) {

		percpu_max_batchsize = old_percpu_max_batchsize;

		ret = -EINVAL;

		goto out;

	}



	/* No change? */

	if (percpu_max_batchsize == old_percpu_max_batchsize)

		goto out;



	for_each_populated_zone(zone)

		zone_set_pageset_high_and_batch(zone);

out:

	mutex_unlock(&pcp_batch_high_lock);

	return ret;

}



#ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES

/*

 * Returns the number of pages that arch has reserved but