[PATCH] mm/hugetlb: avoid looping to the same hugepage if !pages and !vmas

[PATCH] mm/hugetlb: avoid looping to the same hugepage if !pages and !vmas

[zhigang lu]

From: Zhigang Lu <tonnylu@tencent.com>

This change greatly decrease the time of mmaping a file in hugetlbfs.
With MAP_POPULATE flag, it takes about 50 milliseconds to mmap a
existing 128GB file in hugetlbfs. With this change, it takes less
then 1 millisecond.

Signed-off-by: Zhigang Lu <tonnylu@tencent.com>
Reviewed-by: Haozhong Zhang <hzhongzhang@tencent.com>
Reviewed-by: Zongming Zhang <knightzhang@tencent.com>
---
 mm/hugetlb.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 6d7296d..2df941a 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -4391,6 +4391,17 @@ long follow_hugetlb_page(struct mm_struct *mm,
struct vm_area_struct *vma,
  break;
  }
  }
+
+ if (!pages && !vmas && !pfn_offset &&
+     (vaddr + huge_page_size(h) < vma->vm_end) &&
+     (remainder >= pages_per_huge_page(h))) {
+ vaddr += huge_page_size(h);
+ remainder -= pages_per_huge_page(h);
+ i += pages_per_huge_page(h);
+ spin_unlock(ptl);
+ continue;
+ }
+
 same_page:
  if (pages) {
  pages[i] = mem_map_offset(page, pfn_offset);
-- 
1.8.3.1

Re: [PATCH] mm/hugetlb: avoid looping to the same hugepage if !pages and !vmas

[Matthew Wilcox]

On Thu, Aug 29, 2019 at 07:37:22PM +0800, zhigang lu wrote:
> This change greatly decrease the time of mmaping a file in hugetlbfs.
> With MAP_POPULATE flag, it takes about 50 milliseconds to mmap a
> existing 128GB file in hugetlbfs. With this change, it takes less
> then 1 millisecond.

You're going to need to find a new way of sending patches; this patch is
mangled by your mail system.

> @@ -4391,6 +4391,17 @@ long follow_hugetlb_page(struct mm_struct *mm,
> struct vm_area_struct *vma,
>   break;
>   }
>   }
> +
> + if (!pages && !vmas && !pfn_offset &&
> +     (vaddr + huge_page_size(h) < vma->vm_end) &&
> +     (remainder >= pages_per_huge_page(h))) {
> + vaddr += huge_page_size(h);
> + remainder -= pages_per_huge_page(h);
> + i += pages_per_huge_page(h);
> + spin_unlock(ptl);
> + continue;
> + }

The concept seems good to me.  The description above could do with some
better explanation though.

答复: [PATCH] mm/hugetlb: avoid looping to the same hugepage if !pages and !vmas(Internet mail)

[tonnylu(陆志刚)]

-----邮件原件-----
发件人: Matthew Wilcox <willy@infradead.org> 
发送时间: 2019年8月29日 19:55
收件人: zhigang lu <luzhigang001@gmail.com>
抄送: mike.kravetz@oracle.com; linux-mm@kvack.org; linux-kernel@vger.kernel.org; tonnylu(陆志刚) <tonnylu@tencent.com>; hzhongzhang(张昊中) <hzhongzhang@tencent.com>; knightzhang(张宗明) <knightzhang@tencent.com>
主题: Re: [PATCH] mm/hugetlb: avoid looping to the same hugepage if !pages and !vmas(Internet mail)

On Thu, Aug 29, 2019 at 07:37:22PM +0800, zhigang lu wrote:
> This change greatly decrease the time of mmaping a file in hugetlbfs.
> With MAP_POPULATE flag, it takes about 50 milliseconds to mmap a
> existing 128GB file in hugetlbfs. With this change, it takes less
> then 1 millisecond.

You're going to need to find a new way of sending patches; this patch is
mangled by your mail system.


> @@ -4391,6 +4391,17 @@ long follow_hugetlb_page(struct mm_struct *mm,
> struct vm_area_struct *vma,
>   break;
>   }
>   }
> +
> + if (!pages && !vmas && !pfn_offset &&
> +     (vaddr + huge_page_size(h) < vma->vm_end) &&
> +     (remainder >= pages_per_huge_page(h))) {
> + vaddr += huge_page_size(h);
> + remainder -= pages_per_huge_page(h);
> + i += pages_per_huge_page(h);
> + spin_unlock(ptl);
> + continue;
> + }

The concept seems good to me.  The description above could do with some
better explanation though.

Thanks, Willy. I will add more explanation and resend the patches in plain text mode.