mm: memcontrol: convert NR_FILE_THPS account to pages

Currently we use struct per_cpu_nodestat to cache the vmstat counters,
which leads to inaccurate statistics especially THP vmstat counters.  In
the systems with if hundreds of processors it can be GBs of memory.  For
example, for a 96 CPUs system, the threshold is the maximum number of 125.
And the per cpu counters can cache 23.4375 GB in total.

The THP page is already a form of batched addition (it will add 512 worth
of memory in one go) so skipping the batching seems like sensible.
Although every THP stats update overflows the per-cpu counter, resorting
to atomic global updates.  But it can make the statistics more accuracy
for the THP vmstat counters.

So we convert the NR_FILE_THPS account to pages.  This patch is consistent
with 8f18227 ("mm/swap.c: flush lru pvecs on compound page arrival").
Doing this also can make the unit of vmstat counters more unified.
Finally, the unit of the vmstat counters are pages, kB and bytes.  The
B/KB suffix can tell us that the unit is bytes or kB.  The rest which is
without suffix are pages.

Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Muchun Song <[email protected]>
Cc: Alexey Dobriyan <[email protected]>
Cc: Feng Tang <[email protected]>
Cc: Greg Kroah-Hartman <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Joonsoo Kim <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: NeilBrown <[email protected]>
Cc: Pankaj Gupta <[email protected]>
Cc: Rafael. J. Wysocki <[email protected]>
Cc: Randy Dunlap <[email protected]>
Cc: Roman Gushchin <[email protected]>
Cc: Sami Tolvanen <[email protected]>
Cc: Shakeel Butt <[email protected]>
Cc: Vladimir Davydov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

drivers/base/node.c

@@ -466,8 +466,7 @@ static ssize_t node_read_meminfo(struct device *dev,
 				    HPAGE_PMD_NR),
 			     nid, K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED) *
 				    HPAGE_PMD_NR),
-			     nid, K(node_page_state(pgdat, NR_FILE_THPS) *
-				    HPAGE_PMD_NR),
+			     nid, K(node_page_state(pgdat, NR_FILE_THPS)),
 			     nid, K(node_page_state(pgdat, NR_FILE_PMDMAPPED) *
 				    HPAGE_PMD_NR)
 #endif

fs/proc/meminfo.c

@@ -135,7 +135,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v)
 	show_val_kb(m, "ShmemPmdMapped: ",
 		    global_node_page_state(NR_SHMEM_PMDMAPPED) * HPAGE_PMD_NR);
 	show_val_kb(m, "FileHugePages:  ",
-		    global_node_page_state(NR_FILE_THPS) * HPAGE_PMD_NR);
+		    global_node_page_state(NR_FILE_THPS));
 	show_val_kb(m, "FilePmdMapped:  ",
 		    global_node_page_state(NR_FILE_PMDMAPPED) * HPAGE_PMD_NR);
 #endif

include/linux/mmzone.h

@@ -219,7 +219,8 @@ static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
 	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
 		return false;
 
-	return item == NR_ANON_THPS;
+	return item == NR_ANON_THPS ||
+	       item == NR_FILE_THPS;
 }
 
 /*

mm/filemap.c

@@ -208,7 +208,7 @@ static void unaccount_page_cache_page(struct address_space *mapping,
 		if (PageTransHuge(page))
 			__dec_lruvec_page_state(page, NR_SHMEM_THPS);
 	} else if (PageTransHuge(page)) {
-		__dec_lruvec_page_state(page, NR_FILE_THPS);
+		__mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
 		filemap_nr_thps_dec(mapping);
 	}
 

mm/huge_memory.c

@@ -2752,10 +2752,13 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
 		}
 		spin_unlock(&ds_queue->split_queue_lock);
 		if (mapping) {
+			int nr = thp_nr_pages(head);
+
 			if (PageSwapBacked(head))
 				__dec_lruvec_page_state(head, NR_SHMEM_THPS);
 			else
-				__dec_lruvec_page_state(head, NR_FILE_THPS);
+				__mod_lruvec_page_state(head, NR_FILE_THPS,
+							-nr);
 		}
 
 		__split_huge_page(page, list, end);

mm/khugepaged.c

@@ -1643,6 +1643,7 @@ static void collapse_file(struct mm_struct *mm,
 	XA_STATE_ORDER(xas, &mapping->i_pages, start, HPAGE_PMD_ORDER);
 	int nr_none = 0, result = SCAN_SUCCEED;
 	bool is_shmem = shmem_file(file);
+	int nr;
 
 	VM_BUG_ON(!IS_ENABLED(CONFIG_READ_ONLY_THP_FOR_FS) && !is_shmem);
 	VM_BUG_ON(start & (HPAGE_PMD_NR - 1));
@@ -1854,11 +1855,12 @@ static void collapse_file(struct mm_struct *mm,
 		put_page(page);
 		goto xa_unlocked;
 	}
+	nr = thp_nr_pages(new_page);
 
 	if (is_shmem)
 		__inc_lruvec_page_state(new_page, NR_SHMEM_THPS);
 	else {
-		__inc_lruvec_page_state(new_page, NR_FILE_THPS);
+		__mod_lruvec_page_state(new_page, NR_FILE_THPS, nr);
 		filemap_nr_thps_inc(mapping);
 	}
 

mm/memcontrol.c

@@ -1534,7 +1534,7 @@ static struct memory_stat memory_stats[] = {
 	 * constant(e.g. powerpc).
 	 */
 	{ "anon_thp", PAGE_SIZE, NR_ANON_THPS },
-	{ "file_thp", 0, NR_FILE_THPS },
+	{ "file_thp", PAGE_SIZE, NR_FILE_THPS },
 	{ "shmem_thp", 0, NR_SHMEM_THPS },
 #endif
 	{ "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON },
@@ -1566,8 +1566,7 @@ static int __init memory_stats_init(void)
 
 	for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-		if (memory_stats[i].idx == NR_FILE_THPS ||
-		    memory_stats[i].idx == NR_SHMEM_THPS)
+		if (memory_stats[i].idx == NR_SHMEM_THPS)
 			memory_stats[i].ratio = HPAGE_PMD_SIZE;
 #endif
 		VM_BUG_ON(!memory_stats[i].ratio);