[3/3] mm/page_alloc: Split context in free_area_init_node

Commit Message

Oscar Salvador July 18, 2018, 12:47 p.m. UTC

From: Oscar Salvador <[e-mail]>

If free_area_init_node gets called from memhotplug code,
we do not need to call calculate_node_totalpages(),
as the node has no pages.

The same goes for the deferred initialization, as
memmap_init_zone skips that when the context is MEMMAP_HOTPLUG.

Signed-off-by: Oscar Salvador <[e-mail]>
---
 mm/page_alloc.c | 37 +++++++++++++++++++++++++------------
 1 file changed, 25 insertions(+), 12 deletions(-)

Comments

Pavel Tatashin July 18, 2018, 2:34 p.m. UTC | #1

On Wed, Jul 18, 2018 at 8:47 AM <[e-mail]> wrote:
>
> From: Oscar Salvador <[e-mail]>
>
> If free_area_init_node gets called from memhotplug code,
> we do not need to call calculate_node_totalpages(),
> as the node has no pages.

I am not positive this is safe. Some pgdat fields in
calculate_node_totalpages() are set. Even if those fields are always
set to zeros, pgdat may be reused (i.e. node went offline and later
came back online), so we might still need to set those fields to
zeroes.

Pavel

Patch

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index d652a3ad720c..99c342eeb5db 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6388,12 +6388,28 @@  static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
 static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
 
+#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
+static void pgdat_set_deferred_range(pg_data_t *pgdat)
+{
+	/*
+	 * We start only with one section of pages, more pages are added as
+	 * needed until the rest of deferred pages are initialized.
+	 */
+	pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION,
+						pgdat->node_spanned_pages);
+	pgdat->first_deferred_pfn = ULONG_MAX;
+}
+#else
+static void pgdat_set_deferred_range(pg_data_t *pgdat) {}
+#endif
+
 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
 		unsigned long node_start_pfn, unsigned long *zholes_size)
 {
 	pg_data_t *pgdat = NODE_DATA(nid);
 	unsigned long start_pfn = 0;
 	unsigned long end_pfn = 0;
+	bool no_hotplug_context = node_online(nid);
 
 	/* pg_data_t should be reset to zero when it's allocated */
 	WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
@@ -6409,20 +6425,17 @@  void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
 #else
 	start_pfn = node_start_pfn;
 #endif
-	calculate_node_totalpages(pgdat, start_pfn, end_pfn,
-				  zones_size, zholes_size);
 
-	alloc_node_mem_map(pgdat);
-
-#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
-	/*
-	 * We start only with one section of pages, more pages are added as
-	 * needed until the rest of deferred pages are initialized.
+	/* Memhotplug is the only place where free_area_init_node gets called
+	 * with the node being still offline.
 	 */
-	pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION,
-					 pgdat->node_spanned_pages);
-	pgdat->first_deferred_pfn = ULONG_MAX;
-#endif
+	if (no_hotplug_context) {
+		calculate_node_totalpages(pgdat, start_pfn, end_pfn,
+					  zones_size, zholes_size);
+		alloc_node_mem_map(pgdat);
+		pgdat_set_deferred_range(pgdat);
+	}
+
 	free_area_init_core(pgdat);
 }