Re: [RFC][PATCH 5/8] memcg: migrate charge of mapped page

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Thu, 24 Sep 2009 14:48:08 +0900
Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote:

> This patch is the core part of this charge migration feature.
> It adds functions to migrate charge of pages mapped by the task.
> 
> Implementation:
> - define struct migrate_charge and a valuable of it(mc) to remember the count
>   of pre-charges and other information.
> - At can_attach(), parse the page table of the task and count the number of
>   mapped pages which are charged to the source mem_cgroup, and call
>   __mem_cgroup_try_charge() repeatedly and count up mc->precharge.
> - At attach(), parse the page table again, find a target page as we did in
>   can_attach(), and call mem_cgroup_move_account() about the page.
> - Cancel all charges if mc->precharge > 0 on failure or at the end of charge
>   migration.
> 

At first, thank you for hearing my request :)


> Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
> ---
>  mm/memcontrol.c |  270 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-
>  1 files changed, 268 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 30499d9..fbcc195 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -21,6 +21,8 @@
>  #include <linux/memcontrol.h>
>  #include <linux/cgroup.h>
>  #include <linux/mm.h>
> +#include <linux/migrate.h>
> +#include <linux/hugetlb.h>
>  #include <linux/pagemap.h>
>  #include <linux/smp.h>
>  #include <linux/page-flags.h>
> @@ -274,6 +276,18 @@ enum charge_type {
>  #define MEM_CGROUP_RECLAIM_SOFT_BIT	0x2
>  #define MEM_CGROUP_RECLAIM_SOFT		(1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
>  
> +/*
> + * Variables for charge migration at task move.
> + * mc and its members are protected by cgroup_lock
> + */
> +struct migrate_charge {
> +	struct task_struct *tsk;
> +	struct mem_cgroup *from;
> +	struct mem_cgroup *to;
> +	unsigned long precharge;
> +};
> +static struct migrate_charge *mc;
> +

I associate migrate with "page migration".
Then, hmm, recharge or move_charge or some other good word, I like.

BTW, why "mc" is a global vairable ?

IIUC, this all migration is done under cgroup_lock, mc can be
global variable...right ?

But ah..ok, this is a similar thing around cpuset's migration..
When you removing RFC, I'd like to see some performance cost
information around this recharge..



>  static void mem_cgroup_get(struct mem_cgroup *mem);
>  static void mem_cgroup_put(struct mem_cgroup *mem);
>  static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem);
> @@ -1362,7 +1376,7 @@ static int __mem_cgroup_try_charge(struct mm_struct *mm,
>  	if (soft_fail_res) {
>  		mem_over_soft_limit =
>  			mem_cgroup_from_res_counter(soft_fail_res, res);
> -		if (mem_cgroup_soft_limit_check(mem_over_soft_limit))
> +		if (page && mem_cgroup_soft_limit_check(mem_over_soft_limit))
>  			mem_cgroup_update_tree(mem_over_soft_limit, page);
>  	}
>  done:
> @@ -3197,10 +3211,167 @@ static int mem_cgroup_populate(struct cgroup_subsys *ss,
>  	return ret;
>  }
>  
> +/* Handlers for charge migration at task move. */
> +/**
> + * is_target_pte_for_migration - check a pte whether it is target for migration
> + * @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
> + * @target: the pointer the target page will be stored(can be NULL)
> + *
> + * Returns
> + *   0(MIGRATION_TARGET_NONE): if the pte is not a target for charge migration.
> + *   1(MIGRATION_TARGET_PAGE): if the page corresponding to this pte is a target
> + *     for charge migration. if @target is not NULL, the page is stored in
> + *     target->page with extra refcnt got(Callers should handle it).

Will these type incrase more ? If not, bool value is enough.



> + *
> + * Called with pte lock held.
> + */
> +union migration_target {
> +	struct page	*page;
> +};
> +
> +enum migration_target_type {
> +	MIGRATION_TARGET_NONE,	/* not used */
> +	MIGRATION_TARGET_PAGE,
> +};
> +
> +static int is_target_pte_for_migration(struct vm_area_struct *vma,
> +		unsigned long addr, pte_t ptent, union migration_target *target)
> +{
> +	struct page *page;
> +	struct page_cgroup *pc;
> +	int ret = 0;
> +
> +	if (!pte_present(ptent))
> +		return 0;
> +
> +	page = vm_normal_page(vma, addr, ptent);
> +	if (!page || !page_mapped(page))
> +		return 0;
> +	if (!get_page_unless_zero(page))
> +		return 0;

Is this necessary ? We're udner page table lock.
Then, no one can unmap this.

> +
> +	pc = lookup_page_cgroup(page);
> +	lock_page_cgroup(pc);

Hmm...we may even avoid this lock. sounds tricy but we're
under page_table_lock, no one can numap this.


> +	if (PageCgroupUsed(pc) && pc->mem_cgroup == mc->from) {
> +		ret = MIGRATION_TARGET_PAGE;
> +		if (target)
> +			target->page = page;
> +	}
> +	unlock_page_cgroup(pc);
> +
> +	if (!ret || !target)
> +		put_page(page);
> +
> +	return ret;
> +}
> +
> +static int migrate_charge_do_precharge(void)
> +{
> +	int ret = -ENOMEM;
> +	struct mem_cgroup *mem = mc->to;



> +
> +	ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, NULL);
> +	if (ret || !mem)
> +		return -ENOMEM;
> +
> +	mc->precharge++;
> +	return ret;
> +}
> +
> +static int migrate_charge_prepare_pte_range(pmd_t *pmd,
> +					unsigned long addr, unsigned long end,
> +					struct mm_walk *walk)
> +{
> +	int ret = 0;
> +	unsigned long count = 0;
> +	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_migration(vma, addr, *pte, NULL))
> +			count++;
> +	pte_unmap_unlock(pte - 1, ptl);
> +
> +	while (count-- && !ret)
> +		ret = migrate_charge_do_precharge();
> +
> +	return ret;
> +}
> +
> +static int migrate_charge_prepare(void)
> +{
> +	int ret = 0;
> +	struct mm_struct *mm;
> +	struct vm_area_struct *vma;
> +
> +	mm = get_task_mm(mc->tsk);
> +	if (!mm)
> +		return 0;
> +
> +	down_read(&mm->mmap_sem);
> +	for (vma = mm->mmap; vma; vma = vma->vm_next) {
> +		struct mm_walk migrate_charge_prepare_walk = {
> +			.pmd_entry = migrate_charge_prepare_pte_range,
> +			.mm = mm,
> +			.private = vma,
> +		};
> +		if (signal_pending(current)) {
> +			ret = -EINTR;
> +			break;
> +		}
> +		if (is_vm_hugetlb_page(vma))
> +			continue;
> +		ret = walk_page_range(vma->vm_start, vma->vm_end,
> +						&migrate_charge_prepare_walk);
> +		if (ret)
> +			break;
> +		cond_resched();
> +	}
> +	up_read(&mm->mmap_sem);
> +
> +	mmput(mm);
> +	return ret;
> +}
> +
> +static void mem_cgroup_clear_migrate_charge(void)
> +{
> +	VM_BUG_ON(!mc);
> +
> +	while (mc->precharge--)
> +		__mem_cgroup_cancel_charge(mc->to);
> +	kfree(mc);
> +	mc = NULL;
> +}
> +
>  static int mem_cgroup_can_migrate_charge(struct mem_cgroup *mem,
>  					struct task_struct *p)
>  {
> -	return 0;
> +	int ret;
> +	struct mem_cgroup *from = mem_cgroup_from_task(p);
> +
> +	VM_BUG_ON(mc);
> +
> +	if (from == mem)
> +		return 0;
> +
please VM_BUG_ON(!mc);

> +	mc = kmalloc(sizeof(struct migrate_charge), GFP_KERNEL);
> +	if (!mc)
> +		return -ENOMEM;
> +
> +	mc->tsk = p;
> +	mc->from = from;
> +	mc->to = mem;
> +	mc->precharge = 0;
> +
> +	ret = migrate_charge_prepare();
> +
> +	if (ret)
> +		mem_cgroup_clear_migrate_charge();
> +	return ret;
>  }
>  
>  static int mem_cgroup_can_attach(struct cgroup_subsys *ss,
> @@ -3220,10 +3391,105 @@ static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss,
>  				struct task_struct *p,
>  				bool threadgroup)
>  {
> +	if (mc)
> +		mem_cgroup_clear_migrate_charge();
> +}
> +
> +static int migrate_charge_pte_range(pmd_t *pmd,
> +				unsigned long addr, unsigned long end,
> +				struct mm_walk *walk)
> +{
> +	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 migration_target target;
> +		int type;
> +		struct page *page;
> +		struct page_cgroup *pc;
> +
> +		if (!mc->precharge)
> +			break;
> +
> +		type = is_target_pte_for_migration(vma, addr, ptent, &target);
> +		switch (type) {
> +		case MIGRATION_TARGET_PAGE:
> +			page = target.page;
> +			if (isolate_lru_page(page))
> +				goto put;
> +			pc = lookup_page_cgroup(page);
> +			if (!mem_cgroup_move_account(pc, mc->from, mc->to)) {
> +				css_put(&mc->to->css);
> +				mc->precharge--;
> +			}
> +			putback_lru_page(page);
> +put:			/* is_target_pte_for_migration() gets the page */
> +			put_page(page);
> +			break;
> +		default:
> +			continue;
> +		}
> +	}
> +	pte_unmap_unlock(pte - 1, ptl);
> +
> +	if (addr != end) {
> +		/*
> +		 * We have consumed all precharges we got in can_attach().
> +		 * We try precharge one by one, but don't do any additional
> +		 * precharges nor charge migration if we have failed in
> +		 * precharge once in attach() phase.
> +		 */
> +		ret = migrate_charge_do_precharge();
> +		if (!ret)
> +			goto retry;

I think this is a nice handling.



> +	}
> +
> +	return ret;
>  }
>  
>  static void mem_cgroup_migrate_charge(void)
>  {
> +	struct mm_struct *mm;
> +	struct vm_area_struct *vma;
> +
> +	if (!mc)
> +		return;
> +
> +	mm = get_task_mm(mc->tsk);
> +	if (!mm)
> +		goto out;
> +
> +	lru_add_drain_all();
> +	down_read(&mm->mmap_sem);
> +	for (vma = mm->mmap; vma; vma = vma->vm_next) {
> +		int ret;
> +		struct mm_walk migrate_charge_walk = {
> +			.pmd_entry = migrate_charge_pte_range,
> +			.mm = mm,
> +			.private = vma,
> +		};
> +		if (is_vm_hugetlb_page(vma))
> +			continue;
> +		ret = walk_page_range(vma->vm_start, vma->vm_end,
> +							&migrate_charge_walk);
> +		if (ret)
> +			/*
> +			 * means we have consumed all precharges and failed in
> +			 * doing additional precharge. Just abandon here.
> +			 */
> +			break;
> +		cond_resched();
> +	}
> +	up_read(&mm->mmap_sem);
> +
> +	mmput(mm);
> +out:
> +	mem_cgroup_clear_migrate_charge();
>  }
>  
>  static void mem_cgroup_move_task(struct cgroup_subsys *ss,
> -- 

Hmm, I don't complain to this patch itself but cgroup_lock() will be the
last wall to be overcomed for production use...

Can't we just prevent rmdir/mkdir on a hierarchy and move a task ?
fork() etc..can be stopped by this and cpuset's code is not very good.
 

Thanks,
-Kame





--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>