On Fri, Mar 2, 2012 at 8:31 AM, Naoya Horiguchi
<n-horiguchi@xxxxxxxxxxxxx> 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.
>
> Changes from v1:
> - rename is_target_huge_pmd_for_mc() to is_target_thp_for_mc()
> - remove pmd_present() check (it's buggy when pmd_trans_huge(pmd) is true)
> - is_target_thp_for_mc() calls get_page() only when checks are passed
> - unlock page table lock if !mc.precharge
> - compare return value of is_target_thp_for_mc() explicitly to MC_TARGET_TYPE
> - clean up &walk->mm->page_table_lock to &vma->vm_mm->page_table_lock
> - add comment about why race with split_huge_page() does not happen
>
> Signed-off-by: Naoya Horiguchi <n-horiguchi@xxxxxxxxxxxxx>
> Cc: Hillf Danton <dhillf@xxxxxxxxx>
> ---
> mm/memcontrol.c | 87 +++++++++++++++++++++++++++++++++++++++++++++++++++----
> 1 files changed, 81 insertions(+), 6 deletions(-)
>
> diff --git linux-next-20120228.orig/mm/memcontrol.c linux-next-20120228/mm/memcontrol.c
> index c83aeb5..d45b21c 100644
> --- linux-next-20120228.orig/mm/memcontrol.c
> +++ linux-next-20120228/mm/memcontrol.c
> @@ -5211,6 +5211,39 @@ 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.
> + * Caller should make sure that pmd_trans_huge(pmd) is true.
> + */
> +static int is_target_thp_for_mc(struct vm_area_struct *vma,
> + unsigned long addr, pmd_t pmd, union mc_target *target)
> +{
> + struct page *page = NULL;
> + struct page_cgroup *pc;
> + int ret = 0;
> +
Need to check move_anon() ?
Other than that,
Acked-by: Hillf Danton <dhillf@xxxxxxxxx>
> + page = pmd_page(pmd);
> + VM_BUG_ON(!page || !PageHead(page));
> + pc = lookup_page_cgroup(page);
> + if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) {
> + ret = MC_TARGET_PAGE;
> + if (target) {
> + get_page(page);
> + target->page = page;
> + }
> + }
> + return ret;
> +}
> +#else
> +static inline int is_target_thp_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 +5252,14 @@ 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_thp_for_mc(vma, addr, *pmd, NULL)
> + == MC_TARGET_PAGE)
> + mc.precharge += HPAGE_PMD_NR;
> + spin_unlock(&vma->vm_mm->page_table_lock);
> + cond_resched();
> + return 0;
> + }
>
> pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
> for (; addr != end; pte++, addr += PAGE_SIZE)
> @@ -5378,16 +5418,51 @@ 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;
> +
> + /*
> + * We don't take compound_lock() here but no race with splitting thp
> + * happens because:
> + * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
> + * under splitting, which means there's no concurrent thp split,
> + * - if another thread runs into split_huge_page() just after we
> + * entered this if-block, the thread must wait for page table lock
> + * to be unlocked in __split_huge_page_splitting(), where the main
> + * part of thp split is not executed yet.
> + */
> + if (pmd_trans_huge_lock(pmd, vma) == 1) {
> + if (!mc.precharge) {
> + spin_unlock(&vma->vm_mm->page_table_lock);
> + cond_resched();
> + return 0;
> + }
> + type = is_target_thp_for_mc(vma, addr, *pmd, &target);
> + if (type == MC_TARGET_PAGE) {
> + page = target.page;
> + if (!isolate_lru_page(page)) {
> + pc = lookup_page_cgroup(page);
> + if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
> + pc, mc.from, mc.to,
> + false)) {
> + mc.precharge -= HPAGE_PMD_NR;
> + mc.moved_charge += HPAGE_PMD_NR;
> + }
> + putback_lru_page(page);
> + }
> + put_page(page);
> + }
> + spin_unlock(&vma->vm_mm->page_table_lock);
> + cond_resched();
> + return 0;
> + }
>
> - split_huge_page_pmd(walk->mm, pmd);
> 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;
> swp_entry_t ent;
>
> if (!mc.precharge)
> --
> 1.7.7.6
>
��.n������g��� �a��� �&ޖ)���)��h���&������梷���� �Ǟ�m������)�����b�n���y��{^�w�r���&�i��('����춊m�鞵��â����چ�����i�������$����