Commit 0f6fb357 authored by Nanyong Sun's avatar Nanyong Sun Committed by Ma Wupeng
Browse files

memcg: support ksm merge any mode per cgroup 

hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I72R0B


CVE: NA

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

Add control file "memory.ksm" to enable ksm per cgroup.
Echo to 1 will set all tasks currently in the cgroup to ksm merge
any mode, which means ksm gets enabled for all vma's of a process.
Meanwhile echo to 0 will disable ksm for them and unmerge the
merged pages.
Cat the file will show the above state and ksm related profits
of this cgroup.

Signed-off-by: default avatarNanyong Sun <sunnanyong@huawei.com>
parent 351ceedb

Documentation/admin-guide/cgroup-v1/memory.rst

+1 −0
Original line number Diff line number Diff line
@@ -99,6 +99,7 @@ Brief summary of control files. 
 memory.kmem.tcp.failcnt             show the number of tcp buf memory usage

				     hits limits

 memory.kmem.tcp.max_usage_in_bytes  show max tcp buf memory usage recorded

 memory.ksm                          set/show ksm merge any mode

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



1. History

include/linux/memcontrol.h

+4 −0
Original line number Diff line number Diff line
@@ -392,7 +392,11 @@ struct mem_cgroup { 
	KABI_RESERVE(3)

	KABI_RESERVE(4)

#endif

#ifdef CONFIG_KSM

	KABI_USE(5, bool ksm_merge_any)

#else

	KABI_RESERVE(5)

#endif

	KABI_RESERVE(6)

	KABI_RESERVE(7)

	KABI_RESERVE(8)

mm/memcontrol.c

+119 −3
Original line number Diff line number Diff line
@@ -71,6 +71,9 @@ 
#include <linux/uaccess.h>



#include <trace/events/vmscan.h>

#ifndef __GENKSYMS__

#include <linux/ksm.h>

#endif



struct cgroup_subsys memory_cgrp_subsys __read_mostly;

EXPORT_SYMBOL(memory_cgrp_subsys);
@@ -250,10 +253,15 @@ enum res_type { 
	     iter != NULL;				\

	     iter = mem_cgroup_iter(NULL, iter, NULL))



static inline bool __task_is_dying(struct task_struct *task)

{

	return tsk_is_oom_victim(task) || fatal_signal_pending(task) ||

		(task->flags & PF_EXITING);

}



static inline bool task_is_dying(void)

{

	return tsk_is_oom_victim(current) || fatal_signal_pending(current) ||

		(current->flags & PF_EXITING);

	return __task_is_dying(current);

}



/* Some nice accessors for the vmpressure. */
@@ -5331,6 +5339,104 @@ static ssize_t memcg_high_async_ratio_write(struct kernfs_open_file *of, 
	return nbytes;

}



#ifdef CONFIG_KSM

static int memcg_set_ksm_for_tasks(struct mem_cgroup *memcg, bool enable)

{

	struct task_struct *task;

	struct mm_struct *mm;

	struct css_task_iter it;

	int ret = 0;



	if (enable == READ_ONCE(memcg->ksm_merge_any))

		return 0;



	css_task_iter_start(&memcg->css, CSS_TASK_ITER_PROCS, &it);

	while (!ret && (task = css_task_iter_next(&it))) {

		if (__task_is_dying(task))

			continue;



		mm = get_task_mm(task);

		if (!mm)

			continue;



		if (mmap_write_lock_killable(mm)) {

			mmput(mm);

			continue;

		}



		if (enable)

			ret = ksm_enable_merge_any(mm);

		else

			ret = ksm_disable_merge_any(mm);



		mmap_write_unlock(mm);

		mmput(mm);

	}

	css_task_iter_end(&it);



	return ret;

}



static int memory_ksm_show(struct seq_file *m, void *v)

{

	unsigned long ksm_merging_pages = 0;

	unsigned long ksm_rmap_items = 0;

	long ksm_process_profits = 0;

	unsigned int tasks = 0;

	struct task_struct *task;

	struct mm_struct *mm;

	struct css_task_iter it;

	struct mem_cgroup *memcg = mem_cgroup_from_seq(m);



	css_task_iter_start(&memcg->css, CSS_TASK_ITER_PROCS, &it);

	while ((task = css_task_iter_next(&it))) {

		mm = get_task_mm(task);

		if (!mm)

			continue;



		if (test_bit(MMF_VM_MERGE_ANY, &mm->flags))

			tasks++;



		ksm_rmap_items += mm->ksm_rmap_items;

		ksm_merging_pages += mm->ksm_merging_pages;

		ksm_process_profits += ksm_process_profit(mm);

		mmput(mm);

	}

	css_task_iter_end(&it);



	seq_printf(m, "merge any state: %d\n", READ_ONCE(memcg->ksm_merge_any));

	seq_printf(m, "merge any tasks: %u\n", tasks);

	seq_printf(m, "ksm_rmap_items %lu\n", ksm_rmap_items);

	seq_printf(m, "ksm_merging_pages %lu\n", ksm_merging_pages);

	seq_printf(m, "ksm_process_profits %ld\n", ksm_process_profits);

	return 0;

}



static ssize_t memory_ksm_write(struct kernfs_open_file *of, char *buf,

					size_t nbytes, loff_t off)

{

	bool enable;

	int err;

	struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));



	buf = strstrip(buf);

	if (!buf)

		return -EINVAL;



	err = kstrtobool(buf, &enable);

	if (err)

		return err;



	err = memcg_set_ksm_for_tasks(memcg, enable);

	if (err)

		return err;



	WRITE_ONCE(memcg->ksm_merge_any, enable);



	return nbytes;

}

#endif /* CONFIG_KSM */



#ifdef CONFIG_CGROUP_V1_WRITEBACK

#include "../kernel/cgroup/cgroup-internal.h"
@@ -5615,6 +5721,14 @@ static struct cftype mem_cgroup_legacy_files[] = { 
		.seq_show = wb_blkio_show,

		.write = wb_blkio_write,

	},

#endif

#ifdef CONFIG_KSM

	{

		.name = "ksm",

		.flags = CFTYPE_NOT_ON_ROOT,

		.write = memory_ksm_write,

		.seq_show = memory_ksm_show,

	},

#endif

	{ },	/* terminate */

};
@@ -5858,7 +5972,9 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) 
		if (parent != root_mem_cgroup)

			memory_cgrp_subsys.broken_hierarchy = true;

	}



#ifdef CONFIG_KSM

	memcg->ksm_merge_any = false;

#endif

	/* The following stuff does not apply to the root */

	if (!parent) {

		root_mem_cgroup = memcg;