((node = hstate_next_node_to_free(hs, mask)) || 1);     \
                nr_nodes--)
 
-#ifdef CONFIG_ARCH_HAS_GIGANTIC_PAGE
+/* used to demote non-gigantic_huge pages as well */
 static void __destroy_compound_gigantic_page(struct page *page,
                                        unsigned int order, bool demote)
 {
        __ClearPageHead(page);
 }
 
+static void destroy_compound_hugetlb_page_for_demote(struct page *page,
+                                       unsigned int order)
+{
+       __destroy_compound_gigantic_page(page, order, true);
+}
+
+#ifdef CONFIG_ARCH_HAS_GIGANTIC_PAGE
 static void destroy_compound_gigantic_page(struct page *page,
                                        unsigned int order)
 {
        __remove_hugetlb_page(h, page, adjust_surplus, false);
 }
 
+static void remove_hugetlb_page_for_demote(struct hstate *h, struct page *page,
+                                                       bool adjust_surplus)
+{
+       __remove_hugetlb_page(h, page, adjust_surplus, true);
+}
+
 static void add_hugetlb_page(struct hstate *h, struct page *page,
                             bool adjust_surplus)
 {
        return __prep_compound_gigantic_page(page, order, false);
 }
 
+static bool prep_compound_gigantic_page_for_demote(struct page *page,
+                                                       unsigned int order)
+{
+       return __prep_compound_gigantic_page(page, order, true);
+}
+
 /*
  * PageHuge() only returns true for hugetlbfs pages, but not for normal or
  * transparent huge pages.  See the PageTransHuge() documentation for more
        return 0;
 }
 
+static int demote_free_huge_page(struct hstate *h, struct page *page)
+{
+       int i, nid = page_to_nid(page);
+       struct hstate *target_hstate;
+       int rc = 0;
+
+       target_hstate = size_to_hstate(PAGE_SIZE << h->demote_order);
+
+       remove_hugetlb_page_for_demote(h, page, false);
+       spin_unlock_irq(&hugetlb_lock);
+
+       rc = alloc_huge_page_vmemmap(h, page);
+       if (rc) {
+               /* Allocation of vmemmmap failed, we can not demote page */
+               spin_lock_irq(&hugetlb_lock);
+               set_page_refcounted(page);
+               add_hugetlb_page(h, page, false);
+               return rc;
+       }
+
+       /*
+        * Use destroy_compound_hugetlb_page_for_demote for all huge page
+        * sizes as it will not ref count pages.
+        */
+       destroy_compound_hugetlb_page_for_demote(page, huge_page_order(h));
+
+       /*
+        * Taking target hstate mutex synchronizes with set_max_huge_pages.
+        * Without the mutex, pages added to target hstate could be marked
+        * as surplus.
+        *
+        * Note that we already hold h->resize_lock.  To prevent deadlock,
+        * use the convention of always taking larger size hstate mutex first.
+        */
+       mutex_lock(&target_hstate->resize_lock);
+       for (i = 0; i < pages_per_huge_page(h);
+                               i += pages_per_huge_page(target_hstate)) {
+               if (hstate_is_gigantic(target_hstate))
+                       prep_compound_gigantic_page_for_demote(page + i,
+                                                       target_hstate->order);
+               else
+                       prep_compound_page(page + i, target_hstate->order);
+               set_page_private(page + i, 0);
+               set_page_refcounted(page + i);
+               prep_new_huge_page(target_hstate, page + i, nid);
+               put_page(page + i);
+       }
+       mutex_unlock(&target_hstate->resize_lock);
+
+       spin_lock_irq(&hugetlb_lock);
+
+       /*
+        * Not absolutely necessary, but for consistency update max_huge_pages
+        * based on pool changes for the demoted page.
+        */
+       h->max_huge_pages--;
+       target_hstate->max_huge_pages += pages_per_huge_page(h);
+
+       return rc;
+}
+
 static int demote_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed)
        __must_hold(&hugetlb_lock)
 {
+       int nr_nodes, node;
+       struct page *page;
        int rc = 0;
 
        lockdep_assert_held(&hugetlb_lock);
                return -EINVAL;         /* internal error */
        }
 
-       /*
-        * TODO - demote fucntionality will be added in subsequent patch
-        */
+       for_each_node_mask_to_free(h, nr_nodes, node, nodes_allowed) {
+               if (!list_empty(&h->hugepage_freelists[node])) {
+                       page = list_entry(h->hugepage_freelists[node].next,
+                                       struct page, lru);
+                       rc = demote_free_huge_page(h, page);
+                       break;
+               }
+       }
+
        return rc;
 }
 
                /*
                 * Check for available pages to demote each time thorough the
                 * loop as demote_pool_huge_page will drop hugetlb_lock.
-                *
-                * NOTE: demote_pool_huge_page does not yet drop hugetlb_lock
-                * but will when full demote functionality is added in a later
-                * patch.
                 */
                if (nid != NUMA_NO_NODE)
                        nr_available = h->free_huge_pages_node[nid];