Re: [RFC][PATCH] memcg: avoid THP split in task migration

[Hillf Danton <dhillf@gmail.com>]

On Wed, Feb 29, 2012 at 8:28 AM, KAMEZAWA Hiroyuki
<kamezawa.hiroyu@jp.fujitsu.com> wrote:
> On Tue, 28 Feb 2012 16:12:32 -0500
> Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> wrote:
>
>> Currently we can't do task migration among memory cgroups without THP split,
>> which means processes heavily using THP experience large overhead in task
>> migration. This patch introduce the code for moving charge of THP and makes
>> THP more valuable.
>>
>> Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
>> Cc: Hillf Danton <dhillf@gmail.com>
>
>
> Thank you!

   ++hd;

>
> A comment below.
>
>> ---
>>  mm/memcontrol.c |   76 ++++++++++++++++++++++++++++++++++++++++++++++++++----
>>  1 files changed, 70 insertions(+), 6 deletions(-)
>>
>> diff --git linux-next-20120228.orig/mm/memcontrol.c linux-next-20120228/mm/memcontrol.c
>> index c83aeb5..e97c041 100644
>> --- linux-next-20120228.orig/mm/memcontrol.c
>> +++ linux-next-20120228/mm/memcontrol.c
>> @@ -5211,6 +5211,42 @@ static int is_target_pte_for_mc(struct vm_area_struct *vma,
>>       return ret;
>>  }
>>
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +/*
>> + * We don't consider swapping or file mapped pages because THP does not
>> + * support them for now.
>> + */
>> +static int is_target_huge_pmd_for_mc(struct vm_area_struct *vma,

static int is_target_thp_for_mc(struct vm_area_struct *vma,
or
static int is_target_pmd_for_mc(struct vm_area_struct *vma,
sounds better?

>> +             unsigned long addr, pmd_t pmd, union mc_target *target)
>> +{
>> +     struct page *page = NULL;
>> +     struct page_cgroup *pc;
>> +     int ret = 0;
>> +
>> +     if (pmd_present(pmd))
>> +             page = pmd_page(pmd);
>> +     if (!page)
>> +             return 0;
>> +     VM_BUG_ON(!PageHead(page));

With a huge and stable pmd, the above operations on page could be
compacted into one line?

	page = pmd_page(pmd);

>> +     get_page(page);
>> +     pc = lookup_page_cgroup(page);
>> +     if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
>> +             ret = MC_TARGET_PAGE;
>> +             if (target)

After checking target, looks only get_page() needed?

>> +                     target->page = page;
>> +     }
>> +     if (!ret || !target)
>> +             put_page(page);
>> +     return ret;
>> +}
>> +#else
>> +static inline int is_target_huge_pmd_for_mc(struct vm_area_struct *vma,
>> +             unsigned long addr, pmd_t pmd, union mc_target *target)
>> +{
>> +     return 0;
>> +}
>> +#endif
>> +
>>  static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
>>                                       unsigned long addr, unsigned long end,
>>                                       struct mm_walk *walk)
>> @@ -5219,7 +5255,13 @@ static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
>>       pte_t *pte;
>>       spinlock_t *ptl;
>>
>> -     split_huge_page_pmd(walk->mm, pmd);
>> +     if (pmd_trans_huge_lock(pmd, vma) == 1) {
>> +             if (is_target_huge_pmd_for_mc(vma, addr, *pmd, NULL))

		if (is_target_huge_pmd_for_mc(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
looks clearer

>> +                     mc.precharge += HPAGE_PMD_NR;

As HPAGE_PMD_NR is directly used, compiler beeps if THP disabled, I guess.

If yes, please cleanup huge_mm.h with s/BUG()/BUILD_BUG()/ and with
both HPAGE_PMD_ORDER and HPAGE_PMD_NR also defined,
to easy others a bit.

>> +             spin_unlock(&walk->mm->page_table_lock);

		spin_unlock(&vma->mm->page_table_lock);
looks clearer
>> +             cond_resched();
>> +             return 0;
>> +     }
>>
>>       pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
>>       for (; addr != end; pte++, addr += PAGE_SIZE)
>> @@ -5378,16 +5420,38 @@ static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
>>       struct vm_area_struct *vma = walk->private;
>>       pte_t *pte;
>>       spinlock_t *ptl;
>> +     int type;
>> +     union mc_target target;
>> +     struct page *page;
>> +     struct page_cgroup *pc;
>> +
>> +     if (pmd_trans_huge_lock(pmd, vma) == 1) {
>> +             if (!mc.precharge)
>> +                     return 0;

Bang, return without page table lock released.

>> +             type = is_target_huge_pmd_for_mc(vma, addr, *pmd, &target);
>> +             if (type == MC_TARGET_PAGE) {
>> +                     page = target.page;
>> +                     if (!isolate_lru_page(page)) {
>> +                             pc = lookup_page_cgroup(page);
>
> Here is a diffuclut point. Please see mem_cgroup_split_huge_fixup(). It splits

Hard and hard point IMO.

> updates memcg's status of splitted pages under lru_lock and compound_lock
> but not under mm->page_table_lock.
>
> Looking into split_huge_page()
>
>        split_huge_page()  # take anon_vma lock
>                __split_huge_page()
>                        __split_huge_page_refcount() # take lru_lock, compound_lock.
>                                mem_cgroup_split_huge_fixup()
>                        __split_huge_page_map() # take page table lock.
>
[copied from Naoya-san's reply]

> I'm afraid this callchain is not correct.

s/correct/complete/

> Page table lock seems to be taken before we enter the main split work.
>
>    split_huge_page
>        take anon_vma lock
>        __split_huge_page
>            __split_huge_page_splitting
>                lock page_table_lock     <--- *1
>                page_check_address_pmd
>                unlock page_table_lock

Yeah, splitters are blocked.
Plus from the *ugly* documented lock function(another
cleanup needed), the embedded mmap_sem also blocks splitters.

That said, could we simply wait and see results of test cases?

-hd

/* mmap_sem must be held on entry */
static inline int pmd_trans_huge_lock(pmd_t *pmd,
				      struct vm_area_struct *vma)
{
	VM_BUG_ON(!rwsem_is_locked(&vma->vm_mm->mmap_sem));
	if (pmd_trans_huge(*pmd))
		return __pmd_trans_huge_lock(pmd, vma);
	else
		return 0;
}

>            __split_huge_page_refcount
>                lock lru_lock
>                compound_lock
>                mem_cgroup_split_huge_fixup
>                compound_unlock
>                unlock lru_lock
>            __split_huge_page_map
>                lock page_table_lock
>                ... some work
>                unlock page_table_lock
>        unlock anon_vma lock
>

--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>