mm/rmap: extend rmap_walk_xxx() to cope with different cases
authorJoonsoo Kim <iamjoonsoo.kim@lge.com>
Tue, 21 Jan 2014 23:49:49 +0000 (15:49 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 22 Jan 2014 00:19:45 +0000 (16:19 -0800)
There are a lot of common parts in traversing functions, but there are
also a little of uncommon parts in it.  By assigning proper function
pointer on each rmap_walker_control, we can handle these difference
correctly.

Following are differences we should handle.

1. difference of lock function in anon mapping case
2. nonlinear handling in file mapping case
3. prechecked condition:
checking memcg in page_referenced(),
checking VM_SHARE in page_mkclean()
checking temporary vma in try_to_unmap()
4. exit condition:
checking page_mapped() in try_to_unmap()

So, in this patch, I introduce 4 function pointers to handle above
differences.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Hillf Danton <dhillf@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/rmap.h
mm/ksm.c
mm/rmap.c

index 6a456ce..616aa4d 100644 (file)
@@ -235,10 +235,25 @@ struct anon_vma *page_lock_anon_vma_read(struct page *page);
 void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
 
+/*
+ * rmap_walk_control: To control rmap traversing for specific needs
+ *
+ * arg: passed to rmap_one() and invalid_vma()
+ * rmap_one: executed on each vma where page is mapped
+ * done: for checking traversing termination condition
+ * file_nonlinear: for handling file nonlinear mapping
+ * anon_lock: for getting anon_lock by optimized way rather than default
+ * invalid_vma: for skipping uninterested vma
+ */
 struct rmap_walk_control {
        void *arg;
        int (*rmap_one)(struct page *page, struct vm_area_struct *vma,
                                        unsigned long addr, void *arg);
+       int (*done)(struct page *page);
+       int (*file_nonlinear)(struct page *, struct address_space *,
+                                       struct vm_area_struct *vma);
+       struct anon_vma *(*anon_lock)(struct page *page);
+       bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
 };
 
 /*
index c3035fe..91b8cb3 100644 (file)
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -2032,12 +2032,19 @@ again:
                        if ((rmap_item->mm == vma->vm_mm) == search_new_forks)
                                continue;
 
+                       if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
+                               continue;
+
                        ret = rwc->rmap_one(page, vma,
                                        rmap_item->address, rwc->arg);
                        if (ret != SWAP_AGAIN) {
                                anon_vma_unlock_read(anon_vma);
                                goto out;
                        }
+                       if (rwc->done && rwc->done(page)) {
+                               anon_vma_unlock_read(anon_vma);
+                               goto out;
+                       }
                }
                anon_vma_unlock_read(anon_vma);
        }
index f8f10ad..97bf8f0 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1684,10 +1684,14 @@ void __put_anon_vma(struct anon_vma *anon_vma)
 }
 
 #ifdef CONFIG_MIGRATION
-static struct anon_vma *rmap_walk_anon_lock(struct page *page)
+static struct anon_vma *rmap_walk_anon_lock(struct page *page,
+                                       struct rmap_walk_control *rwc)
 {
        struct anon_vma *anon_vma;
 
+       if (rwc->anon_lock)
+               return rwc->anon_lock(page);
+
        /*
         * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
         * because that depends on page_mapped(); but not all its usages
@@ -1713,16 +1717,22 @@ static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
        struct anon_vma_chain *avc;
        int ret = SWAP_AGAIN;
 
-       anon_vma = rmap_walk_anon_lock(page);
+       anon_vma = rmap_walk_anon_lock(page, rwc);
        if (!anon_vma)
                return ret;
 
        anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
                struct vm_area_struct *vma = avc->vma;
                unsigned long address = vma_address(page, vma);
+
+               if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
+                       continue;
+
                ret = rwc->rmap_one(page, vma, address, rwc->arg);
                if (ret != SWAP_AGAIN)
                        break;
+               if (rwc->done && rwc->done(page))
+                       break;
        }
        anon_vma_unlock_read(anon_vma);
        return ret;
@@ -1740,15 +1750,26 @@ static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
        mutex_lock(&mapping->i_mmap_mutex);
        vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
                unsigned long address = vma_address(page, vma);
+
+               if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
+                       continue;
+
                ret = rwc->rmap_one(page, vma, address, rwc->arg);
                if (ret != SWAP_AGAIN)
-                       break;
+                       goto done;
+               if (rwc->done && rwc->done(page))
+                       goto done;
        }
-       /*
-        * No nonlinear handling: being always shared, nonlinear vmas
-        * never contain migration ptes.  Decide what to do about this
-        * limitation to linear when we need rmap_walk() on nonlinear.
-        */
+
+       if (!rwc->file_nonlinear)
+               goto done;
+
+       if (list_empty(&mapping->i_mmap_nonlinear))
+               goto done;
+
+       ret = rwc->file_nonlinear(page, mapping, vma);
+
+done:
        mutex_unlock(&mapping->i_mmap_mutex);
        return ret;
 }