mm/hugetlb: change hugetlb allocation functions to return a folio
authorSidhartha Kumar <sidhartha.kumar@oracle.com>
Tue, 29 Nov 2022 22:50:39 +0000 (14:50 -0800)
committerAndrew Morton <akpm@linux-foundation.org>
Mon, 12 Dec 2022 02:12:15 +0000 (18:12 -0800)
Many hugetlb allocation helper functions have now been converting to
folios, update their higher level callers to be compatible with folios.
alloc_pool_huge_page is reorganized to avoid a smatch warning reporting
the folio variable is uninitialized.

[sidhartha.kumar@oracle.com: update alloc_and_dissolve_hugetlb_folio comments]
Link: https://lkml.kernel.org/r/20221206233512.146535-1-sidhartha.kumar@oracle.com
Link: https://lkml.kernel.org/r/20221129225039.82257-11-sidhartha.kumar@oracle.com
Signed-off-by: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Reported-by: Wei Chen <harperchen1110@gmail.com>
Suggested-by: John Hubbard <jhubbard@nvidia.com>
Suggested-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Tarun Sahu <tsahu@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/hugetlb.c

index 903ee75..8c6fe22 100644 (file)
@@ -1378,23 +1378,23 @@ static void free_gigantic_folio(struct folio *folio, unsigned int order)
 }
 
 #ifdef CONFIG_CONTIG_ALLOC
-static struct page *alloc_gigantic_page(struct hstate *h, gfp_t gfp_mask,
+static struct folio *alloc_gigantic_folio(struct hstate *h, gfp_t gfp_mask,
                int nid, nodemask_t *nodemask)
 {
+       struct page *page;
        unsigned long nr_pages = pages_per_huge_page(h);
        if (nid == NUMA_NO_NODE)
                nid = numa_mem_id();
 
 #ifdef CONFIG_CMA
        {
-               struct page *page;
                int node;
 
                if (hugetlb_cma[nid]) {
                        page = cma_alloc(hugetlb_cma[nid], nr_pages,
                                        huge_page_order(h), true);
                        if (page)
-                               return page;
+                               return page_folio(page);
                }
 
                if (!(gfp_mask & __GFP_THISNODE)) {
@@ -1405,17 +1405,18 @@ static struct page *alloc_gigantic_page(struct hstate *h, gfp_t gfp_mask,
                                page = cma_alloc(hugetlb_cma[node], nr_pages,
                                                huge_page_order(h), true);
                                if (page)
-                                       return page;
+                                       return page_folio(page);
                        }
                }
        }
 #endif
 
-       return alloc_contig_pages(nr_pages, gfp_mask, nid, nodemask);
+       page = alloc_contig_pages(nr_pages, gfp_mask, nid, nodemask);
+       return page ? page_folio(page) : NULL;
 }
 
 #else /* !CONFIG_CONTIG_ALLOC */
-static struct page *alloc_gigantic_page(struct hstate *h, gfp_t gfp_mask,
+static struct folio *alloc_gigantic_folio(struct hstate *h, gfp_t gfp_mask,
                                        int nid, nodemask_t *nodemask)
 {
        return NULL;
@@ -1423,7 +1424,7 @@ static struct page *alloc_gigantic_page(struct hstate *h, gfp_t gfp_mask,
 #endif /* CONFIG_CONTIG_ALLOC */
 
 #else /* !CONFIG_ARCH_HAS_GIGANTIC_PAGE */
-static struct page *alloc_gigantic_page(struct hstate *h, gfp_t gfp_mask,
+static struct folio *alloc_gigantic_folio(struct hstate *h, gfp_t gfp_mask,
                                        int nid, nodemask_t *nodemask)
 {
        return NULL;
@@ -1950,7 +1951,7 @@ pgoff_t hugetlb_basepage_index(struct page *page)
        return (index << compound_order(page_head)) + compound_idx;
 }
 
-static struct page *alloc_buddy_huge_page(struct hstate *h,
+static struct folio *alloc_buddy_hugetlb_folio(struct hstate *h,
                gfp_t gfp_mask, int nid, nodemask_t *nmask,
                nodemask_t *node_alloc_noretry)
 {
@@ -1988,11 +1989,6 @@ retry:
                page = NULL;
        }
 
-       if (page)
-               __count_vm_event(HTLB_BUDDY_PGALLOC);
-       else
-               __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
-
        /*
         * If we did not specify __GFP_RETRY_MAYFAIL, but still got a page this
         * indicates an overall state change.  Clear bit so that we resume
@@ -2009,7 +2005,13 @@ retry:
        if (node_alloc_noretry && !page && alloc_try_hard)
                node_set(nid, *node_alloc_noretry);
 
-       return page;
+       if (!page) {
+               __count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
+               return NULL;
+       }
+
+       __count_vm_event(HTLB_BUDDY_PGALLOC);
+       return page_folio(page);
 }
 
 /*
@@ -2019,23 +2021,21 @@ retry:
  * Note that returned page is 'frozen':  ref count of head page and all tail
  * pages is zero.
  */
-static struct page *alloc_fresh_huge_page(struct hstate *h,
+static struct folio *alloc_fresh_hugetlb_folio(struct hstate *h,
                gfp_t gfp_mask, int nid, nodemask_t *nmask,
                nodemask_t *node_alloc_noretry)
 {
-       struct page *page;
        struct folio *folio;
        bool retry = false;
 
 retry:
        if (hstate_is_gigantic(h))
-               page = alloc_gigantic_page(h, gfp_mask, nid, nmask);
+               folio = alloc_gigantic_folio(h, gfp_mask, nid, nmask);
        else
-               page = alloc_buddy_huge_page(h, gfp_mask,
+               folio = alloc_buddy_hugetlb_folio(h, gfp_mask,
                                nid, nmask, node_alloc_noretry);
-       if (!page)
+       if (!folio)
                return NULL;
-       folio = page_folio(page);
        if (hstate_is_gigantic(h)) {
                if (!prep_compound_gigantic_folio(folio, huge_page_order(h))) {
                        /*
@@ -2052,7 +2052,7 @@ retry:
        }
        prep_new_hugetlb_folio(h, folio, folio_nid(folio));
 
-       return page;
+       return folio;
 }
 
 /*
@@ -2062,23 +2062,20 @@ retry:
 static int alloc_pool_huge_page(struct hstate *h, nodemask_t *nodes_allowed,
                                nodemask_t *node_alloc_noretry)
 {
-       struct page *page;
+       struct folio *folio;
        int nr_nodes, node;
        gfp_t gfp_mask = htlb_alloc_mask(h) | __GFP_THISNODE;
 
        for_each_node_mask_to_alloc(h, nr_nodes, node, nodes_allowed) {
-               page = alloc_fresh_huge_page(h, gfp_mask, node, nodes_allowed,
-                                               node_alloc_noretry);
-               if (page)
-                       break;
+               folio = alloc_fresh_hugetlb_folio(h, gfp_mask, node,
+                                       nodes_allowed, node_alloc_noretry);
+               if (folio) {
+                       free_huge_page(&folio->page); /* free it into the hugepage allocator */
+                       return 1;
+               }
        }
 
-       if (!page)
-               return 0;
-
-       free_huge_page(page); /* free it into the hugepage allocator */
-
-       return 1;
+       return 0;
 }
 
 /*
@@ -2237,7 +2234,7 @@ int dissolve_free_huge_pages(unsigned long start_pfn, unsigned long end_pfn)
 static struct page *alloc_surplus_huge_page(struct hstate *h, gfp_t gfp_mask,
                                                int nid, nodemask_t *nmask)
 {
-       struct page *page = NULL;
+       struct folio *folio = NULL;
 
        if (hstate_is_gigantic(h))
                return NULL;
@@ -2247,8 +2244,8 @@ static struct page *alloc_surplus_huge_page(struct hstate *h, gfp_t gfp_mask,
                goto out_unlock;
        spin_unlock_irq(&hugetlb_lock);
 
-       page = alloc_fresh_huge_page(h, gfp_mask, nid, nmask, NULL);
-       if (!page)
+       folio = alloc_fresh_hugetlb_folio(h, gfp_mask, nid, nmask, NULL);
+       if (!folio)
                return NULL;
 
        spin_lock_irq(&hugetlb_lock);
@@ -2260,43 +2257,42 @@ static struct page *alloc_surplus_huge_page(struct hstate *h, gfp_t gfp_mask,
         * codeflow
         */
        if (h->surplus_huge_pages >= h->nr_overcommit_huge_pages) {
-               SetHPageTemporary(page);
+               folio_set_hugetlb_temporary(folio);
                spin_unlock_irq(&hugetlb_lock);
-               free_huge_page(page);
+               free_huge_page(&folio->page);
                return NULL;
        }
 
        h->surplus_huge_pages++;
-       h->surplus_huge_pages_node[page_to_nid(page)]++;
+       h->surplus_huge_pages_node[folio_nid(folio)]++;
 
 out_unlock:
        spin_unlock_irq(&hugetlb_lock);
 
-       return page;
+       return &folio->page;
 }
 
 static struct page *alloc_migrate_huge_page(struct hstate *h, gfp_t gfp_mask,
                                     int nid, nodemask_t *nmask)
 {
-       struct page *page;
+       struct folio *folio;
 
        if (hstate_is_gigantic(h))
                return NULL;
 
-       page = alloc_fresh_huge_page(h, gfp_mask, nid, nmask, NULL);
-       if (!page)
+       folio = alloc_fresh_hugetlb_folio(h, gfp_mask, nid, nmask, NULL);
+       if (!folio)
                return NULL;
 
        /* fresh huge pages are frozen */
-       set_page_refcounted(page);
-
+       folio_ref_unfreeze(folio, 1);
        /*
         * We do not account these pages as surplus because they are only
         * temporary and will be released properly on the last reference
         */
-       SetHPageTemporary(page);
+       folio_set_hugetlb_temporary(folio);
 
-       return page;
+       return &folio->page;
 }
 
 /*
@@ -2745,54 +2741,52 @@ void restore_reserve_on_error(struct hstate *h, struct vm_area_struct *vma,
 }
 
 /*
- * alloc_and_dissolve_huge_page - Allocate a new page and dissolve the old one
+ * alloc_and_dissolve_hugetlb_folio - Allocate a new folio and dissolve
+ * the old one
  * @h: struct hstate old page belongs to
- * @old_page: Old page to dissolve
+ * @old_folio: Old folio to dissolve
  * @list: List to isolate the page in case we need to
  * Returns 0 on success, otherwise negated error.
  */
-static int alloc_and_dissolve_huge_page(struct hstate *h, struct page *old_page,
-                                       struct list_head *list)
+static int alloc_and_dissolve_hugetlb_folio(struct hstate *h,
+                       struct folio *old_folio, struct list_head *list)
 {
        gfp_t gfp_mask = htlb_alloc_mask(h) | __GFP_THISNODE;
-       struct folio *old_folio = page_folio(old_page);
        int nid = folio_nid(old_folio);
-       struct page *new_page;
        struct folio *new_folio;
        int ret = 0;
 
        /*
-        * Before dissolving the page, we need to allocate a new one for the
-        * pool to remain stable.  Here, we allocate the page and 'prep' it
+        * Before dissolving the folio, we need to allocate a new one for the
+        * pool to remain stable.  Here, we allocate the folio and 'prep' it
         * by doing everything but actually updating counters and adding to
         * the pool.  This simplifies and let us do most of the processing
         * under the lock.
         */
-       new_page = alloc_buddy_huge_page(h, gfp_mask, nid, NULL, NULL);
-       if (!new_page)
+       new_folio = alloc_buddy_hugetlb_folio(h, gfp_mask, nid, NULL, NULL);
+       if (!new_folio)
                return -ENOMEM;
-       new_folio = page_folio(new_page);
        __prep_new_hugetlb_folio(h, new_folio);
 
 retry:
        spin_lock_irq(&hugetlb_lock);
        if (!folio_test_hugetlb(old_folio)) {
                /*
-                * Freed from under us. Drop new_page too.
+                * Freed from under us. Drop new_folio too.
                 */
                goto free_new;
        } else if (folio_ref_count(old_folio)) {
                /*
-                * Someone has grabbed the page, try to isolate it here.
+                * Someone has grabbed the folio, try to isolate it here.
                 * Fail with -EBUSY if not possible.
                 */
                spin_unlock_irq(&hugetlb_lock);
-               ret = isolate_hugetlb(old_page, list);
+               ret = isolate_hugetlb(&old_folio->page, list);
                spin_lock_irq(&hugetlb_lock);
                goto free_new;
        } else if (!folio_test_hugetlb_freed(old_folio)) {
                /*
-                * Page's refcount is 0 but it has not been enqueued in the
+                * Folio's refcount is 0 but it has not been enqueued in the
                 * freelist yet. Race window is small, so we can succeed here if
                 * we retry.
                 */
@@ -2801,7 +2795,7 @@ retry:
                goto retry;
        } else {
                /*
-                * Ok, old_page is still a genuine free hugepage. Remove it from
+                * Ok, old_folio is still a genuine free hugepage. Remove it from
                 * the freelist and decrease the counters. These will be
                 * incremented again when calling __prep_account_new_huge_page()
                 * and enqueue_hugetlb_folio() for new_folio. The counters will
@@ -2810,14 +2804,14 @@ retry:
                remove_hugetlb_folio(h, old_folio, false);
 
                /*
-                * Ref count on new page is already zero as it was dropped
+                * Ref count on new_folio is already zero as it was dropped
                 * earlier.  It can be directly added to the pool free list.
                 */
                __prep_account_new_huge_page(h, nid);
                enqueue_hugetlb_folio(h, new_folio);
 
                /*
-                * Pages have been replaced, we can safely free the old one.
+                * Folio has been replaced, we can safely free the old one.
                 */
                spin_unlock_irq(&hugetlb_lock);
                update_and_free_hugetlb_folio(h, old_folio, false);
@@ -2827,7 +2821,7 @@ retry:
 
 free_new:
        spin_unlock_irq(&hugetlb_lock);
-       /* Page has a zero ref count, but needs a ref to be freed */
+       /* Folio has a zero ref count, but needs a ref to be freed */
        folio_ref_unfreeze(new_folio, 1);
        update_and_free_hugetlb_folio(h, new_folio, false);
 
@@ -2865,7 +2859,7 @@ int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list)
        if (folio_ref_count(folio) && !isolate_hugetlb(&folio->page, list))
                ret = 0;
        else if (!folio_ref_count(folio))
-               ret = alloc_and_dissolve_huge_page(h, &folio->page, list);
+               ret = alloc_and_dissolve_hugetlb_folio(h, folio, list);
 
        return ret;
 }
@@ -3083,14 +3077,14 @@ static void __init hugetlb_hstate_alloc_pages_onenode(struct hstate *h, int nid)
                        if (!alloc_bootmem_huge_page(h, nid))
                                break;
                } else {
-                       struct page *page;
+                       struct folio *folio;
                        gfp_t gfp_mask = htlb_alloc_mask(h) | __GFP_THISNODE;
 
-                       page = alloc_fresh_huge_page(h, gfp_mask, nid,
+                       folio = alloc_fresh_hugetlb_folio(h, gfp_mask, nid,
                                        &node_states[N_MEMORY], NULL);
-                       if (!page)
+                       if (!folio)
                                break;
-                       free_huge_page(page); /* free it into the hugepage allocator */
+                       free_huge_page(&folio->page); /* free it into the hugepage allocator */
                }
                cond_resched();
        }