ksm: remove old stable nodes more thoroughly
[platform/adaptation/renesas_rcar/renesas_kernel.git] / mm / ksm.c
index e02430f..4c22cdf 100644 (file)
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -644,6 +644,57 @@ static int unmerge_ksm_pages(struct vm_area_struct *vma,
 /*
  * Only called through the sysfs control interface:
  */
+static int remove_stable_node(struct stable_node *stable_node)
+{
+       struct page *page;
+       int err;
+
+       page = get_ksm_page(stable_node, true);
+       if (!page) {
+               /*
+                * get_ksm_page did remove_node_from_stable_tree itself.
+                */
+               return 0;
+       }
+
+       if (WARN_ON_ONCE(page_mapped(page)))
+               err = -EBUSY;
+       else {
+               /*
+                * This page might be in a pagevec waiting to be freed,
+                * or it might be PageSwapCache (perhaps under writeback),
+                * or it might have been removed from swapcache a moment ago.
+                */
+               set_page_stable_node(page, NULL);
+               remove_node_from_stable_tree(stable_node);
+               err = 0;
+       }
+
+       unlock_page(page);
+       put_page(page);
+       return err;
+}
+
+static int remove_all_stable_nodes(void)
+{
+       struct stable_node *stable_node;
+       int nid;
+       int err = 0;
+
+       for (nid = 0; nid < nr_node_ids; nid++) {
+               while (root_stable_tree[nid].rb_node) {
+                       stable_node = rb_entry(root_stable_tree[nid].rb_node,
+                                               struct stable_node, node);
+                       if (remove_stable_node(stable_node)) {
+                               err = -EBUSY;
+                               break;  /* proceed to next nid */
+                       }
+                       cond_resched();
+               }
+       }
+       return err;
+}
+
 static int unmerge_and_remove_all_rmap_items(void)
 {
        struct mm_slot *mm_slot;
@@ -691,6 +742,8 @@ static int unmerge_and_remove_all_rmap_items(void)
                }
        }
 
+       /* Clean up stable nodes, but don't worry if some are still busy */
+       remove_all_stable_nodes();
        ksm_scan.seqnr = 0;
        return 0;
 
@@ -1586,11 +1639,19 @@ int __ksm_enter(struct mm_struct *mm)
        spin_lock(&ksm_mmlist_lock);
        insert_to_mm_slots_hash(mm, mm_slot);
        /*
-        * Insert just behind the scanning cursor, to let the area settle
+        * When KSM_RUN_MERGE (or KSM_RUN_STOP),
+        * insert just behind the scanning cursor, to let the area settle
         * down a little; when fork is followed by immediate exec, we don't
         * want ksmd to waste time setting up and tearing down an rmap_list.
+        *
+        * But when KSM_RUN_UNMERGE, it's important to insert ahead of its
+        * scanning cursor, otherwise KSM pages in newly forked mms will be
+        * missed: then we might as well insert at the end of the list.
         */
-       list_add_tail(&mm_slot->mm_list, &ksm_scan.mm_slot->mm_list);
+       if (ksm_run & KSM_RUN_UNMERGE)
+               list_add_tail(&mm_slot->mm_list, &ksm_mm_head.mm_list);
+       else
+               list_add_tail(&mm_slot->mm_list, &ksm_scan.mm_slot->mm_list);
        spin_unlock(&ksm_mmlist_lock);
 
        set_bit(MMF_VM_MERGEABLE, &mm->flags);
@@ -1640,11 +1701,25 @@ void __ksm_exit(struct mm_struct *mm)
        }
 }
 
-struct page *ksm_does_need_to_copy(struct page *page,
+struct page *ksm_might_need_to_copy(struct page *page,
                        struct vm_area_struct *vma, unsigned long address)
 {
+       struct anon_vma *anon_vma = page_anon_vma(page);
        struct page *new_page;
 
+       if (PageKsm(page)) {
+               if (page_stable_node(page) &&
+                   !(ksm_run & KSM_RUN_UNMERGE))
+                       return page;    /* no need to copy it */
+       } else if (!anon_vma) {
+               return page;            /* no need to copy it */
+       } else if (anon_vma->root == vma->anon_vma->root &&
+                page->index == linear_page_index(vma, address)) {
+               return page;            /* still no need to copy it */
+       }
+       if (!PageUptodate(page))
+               return page;            /* let do_swap_page report the error */
+
        new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
        if (new_page) {
                copy_user_highpage(new_page, page, address, vma);
@@ -2024,7 +2099,7 @@ static ssize_t merge_across_nodes_store(struct kobject *kobj,
 
        mutex_lock(&ksm_thread_mutex);
        if (ksm_merge_across_nodes != knob) {
-               if (ksm_pages_shared)
+               if (ksm_pages_shared || remove_all_stable_nodes())
                        err = -EBUSY;
                else
                        ksm_merge_across_nodes = knob;