This patch introduce fake VM_VRANGE flag in vma->vm_flags.
Actually, vma->vm_flags doesn't have such flag and it is just
used to detect a page is volatile page or not in page_referenced.

For it, page_referenced's vm_flags argument semantic is changed so that
caller should specify what kinds of flags he has interest.
It could make to avoid unnecessary volatile range lookup in
page_referenced_one.

Cc: Mel Gorman <m...@csn.ul.ie>
Cc: Hugh Dickins <hu...@google.com>
Cc: Dave Hansen <dave.han...@intel.com>
Cc: Rik van Riel <r...@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motoh...@gmail.com>
Cc: Michel Lespinasse <wal...@google.com>
Cc: Johannes Weiner <han...@cmpxchg.org>
Cc: John Stultz <john.stu...@linaro.org>
Signed-off-by: Minchan Kim <minc...@kernel.org>
---
 include/linux/mm.h |    9 +++++++++
 mm/rmap.c          |   17 +++++++++++++----
 mm/vmscan.c        |    4 ++--
 3 files changed, 24 insertions(+), 6 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 8b6e55ee8855..3dec30154f96 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -103,6 +103,15 @@ extern unsigned int kobjsize(const void *objp);
 #define VM_IO           0x00004000     /* Memory mapped I/O or similar */
 
                                        /* Used by sys_madvise() */
+/*
+ * VM_VRANGE is rather special. Actually, vma->vm_flags doesn't have such flag.
+ * It is used to identify whether a page put on volatile range or not
+ * by page_referenced. So, if we are lack of new bit in vmflags, we could
+ * replace it with assembling exclusive flags.
+ *
+ * ex) VM_HUGEPAGE|VM_NOHUGEPAGE
+ */
+#define VM_VRANGE      0x00001000
 #define VM_SEQ_READ    0x00008000      /* App will access data sequentially */
 #define VM_RAND_READ   0x00010000      /* App will not benefit from clustered 
reads */
 
diff --git a/mm/rmap.c b/mm/rmap.c
index fd3ee7a54a13..9220f12deb93 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -57,6 +57,7 @@
 #include <linux/migrate.h>
 #include <linux/hugetlb.h>
 #include <linux/backing-dev.h>
+#include <linux/vrange.h>
 
 #include <asm/tlbflush.h>
 
@@ -685,7 +686,7 @@ int page_referenced_one(struct page *page, struct 
vm_area_struct *vma,
                if (vma->vm_flags & VM_LOCKED) {
                        spin_unlock(&mm->page_table_lock);
                        *mapcount = 0;  /* break early from loop */
-                       *vm_flags |= VM_LOCKED;
+                       *vm_flags &= VM_LOCKED;
                        goto out;
                }
 
@@ -708,7 +709,7 @@ int page_referenced_one(struct page *page, struct 
vm_area_struct *vma,
                if (vma->vm_flags & VM_LOCKED) {
                        pte_unmap_unlock(pte, ptl);
                        *mapcount = 0;  /* break early from loop */
-                       *vm_flags |= VM_LOCKED;
+                       *vm_flags &= VM_LOCKED;
                        goto out;
                }
 
@@ -724,12 +725,18 @@ int page_referenced_one(struct page *page, struct 
vm_area_struct *vma,
                                referenced++;
                }
                pte_unmap_unlock(pte, ptl);
+               if (*vm_flags & VM_VRANGE &&
+                               vrange_addr_volatile(vma, address)) {
+                       *mapcount = 0; /* break ealry from loop */
+                       *vm_flags &= VM_VRANGE;
+                       goto out;
+               }
        }
 
        (*mapcount)--;
 
        if (referenced)
-               *vm_flags |= vma->vm_flags;
+               *vm_flags &= vma->vm_flags;
 out:
        return referenced;
 }
@@ -844,6 +851,9 @@ static int page_referenced_file(struct page *page,
  *
  * Quick test_and_clear_referenced for all mappings to a page,
  * returns the number of ptes which referenced the page.
+ *
+ * NOTE: caller should pass interested flags in vm_flags to collect
+ * vma->vm_flags.
  */
 int page_referenced(struct page *page,
                    int is_locked,
@@ -853,7 +863,6 @@ int page_referenced(struct page *page,
        int referenced = 0;
        int we_locked = 0;
 
-       *vm_flags = 0;
        if (page_mapped(page) && page_rmapping(page)) {
                if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
                        we_locked = trylock_page(page);
diff --git a/mm/vmscan.c b/mm/vmscan.c
index eea668d9cff6..8bff386e65a0 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -672,7 +672,7 @@ static enum page_references page_check_references(struct 
page *page,
                                                  struct scan_control *sc)
 {
        int referenced_ptes, referenced_page;
-       unsigned long vm_flags;
+       unsigned long vm_flags = VM_EXEC|VM_LOCKED|VM_VRANGE;
 
        referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup,
                                          &vm_flags);
@@ -1619,7 +1619,6 @@ static void shrink_active_list(unsigned long nr_to_scan,
 {
        unsigned long nr_taken;
        unsigned long nr_scanned;
-       unsigned long vm_flags;
        LIST_HEAD(l_hold);      /* The pages which were snipped off */
        LIST_HEAD(l_active);
        LIST_HEAD(l_inactive);
@@ -1652,6 +1651,7 @@ static void shrink_active_list(unsigned long nr_to_scan,
        spin_unlock_irq(&zone->lru_lock);
 
        while (!list_empty(&l_hold)) {
+               unsigned long vm_flags = VM_EXEC;
                cond_resched();
                page = lru_to_page(&l_hold);
                list_del(&page->lru);
-- 
1.7.9.5

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to