lkml.org 
[lkml]   [2015]   [Nov]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH 4/4] mm: prepare page_referenced() and page_idle to new THP refcounting
On Thu, Nov 05, 2015 at 04:32:11PM -0800, Andrew Morton wrote:
> On Tue, 3 Nov 2015 17:26:15 +0200 "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> wrote:
>
> > I've missed two simlar codepath which need some preparation to work well
> > with reworked THP refcounting.
> >
> > Both page_referenced() and page_idle_clear_pte_refs_one() assume that
> > THP can only be mapped with PMD, so there's no reason to look on PTEs
> > for PageTransHuge() pages. That's no true anymore: THP can be mapped
> > with PTEs too.
> >
> > The patch removes PageTransHuge() test from the functions and opencode
> > page table check.
>
> x86_64 allnoconfig:
>
> In file included from mm/rmap.c:47:
> include/linux/mm.h: In function 'page_referenced':
> include/linux/mm.h:448: error: call to '__compiletime_assert_448' declared with attribute error: BUILD_BUG failed
> make[1]: *** [mm/rmap.o] Error 1
> make: *** [mm/rmap.o] Error 2
>
> because
>
> #else /* CONFIG_TRANSPARENT_HUGEPAGE */
> #define HPAGE_PMD_SHIFT ({ BUILD_BUG(); 0; })
>
>
> btw, total_mapcount() is far too large to be inlined and

The patch below is my propsal to fix this.

> page_mapcount() is getting pretty bad too.

Do you want me to uninline slow path (PageCompound())?

diff --git a/include/linux/mm.h b/include/linux/mm.h
index a36f9fa4e4cd..f874d2a1d1a6 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -432,24 +432,14 @@ static inline int page_mapcount(struct page *page)
return ret;
}

+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+int total_mapcount(struct page *page);
+#else
static inline int total_mapcount(struct page *page)
{
- int i, ret;
-
- VM_BUG_ON_PAGE(PageTail(page), page);
-
- if (likely(!PageCompound(page)))
- return atomic_read(&page->_mapcount) + 1;
-
- ret = compound_mapcount(page);
- if (PageHuge(page))
- return ret;
- for (i = 0; i < HPAGE_PMD_NR; i++)
- ret += atomic_read(&page[i]._mapcount) + 1;
- if (PageDoubleMap(page))
- ret -= HPAGE_PMD_NR;
- return ret;
+ return page_mapcount(page);
}
+#endif

static inline int page_count(struct page *page)
{
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 14cbbad54a3e..287bc009bc10 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -3236,6 +3236,25 @@ static void __split_huge_page(struct page *page, struct list_head *list)
}
}

+int total_mapcount(struct page *page)
+{
+ int i, ret;
+
+ VM_BUG_ON_PAGE(PageTail(page), page);
+
+ if (likely(!PageCompound(page)))
+ return atomic_read(&page->_mapcount) + 1;
+
+ ret = compound_mapcount(page);
+ if (PageHuge(page))
+ return ret;
+ for (i = 0; i < HPAGE_PMD_NR; i++)
+ ret += atomic_read(&page[i]._mapcount) + 1;
+ if (PageDoubleMap(page))
+ ret -= HPAGE_PMD_NR;
+ return ret;
+}
+
/*
* This function splits huge page into normal pages. @page can point to any
* subpage of huge page to split. Split doesn't change the position of @page.
--
Kirill A. Shutemov

\
 
 \ /
  Last update: 2015-11-06 11:41    [W:0.182 / U:0.032 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site