[PATCH v2] mm/vmstat: annotate data race for zone->free_area[order].nr_free

From: Liu Shixin
Date: Sat Sep 18 2021 - 04:18:36 EST


KCSAN reports a data-race on v5.10 which also exists on mainline:

==================================================================
BUG: KCSAN: data-race in extfrag_for_order+0x33/0x2d0

race at unknown origin, with read to 0xffff9ee9bfffab48 of 8 bytes by task 34 on cpu 1:
extfrag_for_order+0x33/0x2d0
kcompactd+0x5f0/0xce0
kthread+0x1f9/0x220
ret_from_fork+0x22/0x30

Reported by Kernel Concurrency Sanitizer on:
CPU: 1 PID: 34 Comm: kcompactd0 Not tainted 5.10.0+ #2
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014
==================================================================

Access to zone->free_area[order].nr_free in extfrag_for_order() and
frag_show_print() is lockless. That's intentional and the stats are
a rough estimate anyway. Annotate them with data_race().

Signed-off-by: Liu Shixin <liushixin2@xxxxxxxxxx>
---
mm/vmstat.c | 15 ++++++++++++---
1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/mm/vmstat.c b/mm/vmstat.c
index 8ce2620344b2..8525849a48f0 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -1070,8 +1070,13 @@ static void fill_contig_page_info(struct zone *zone,
for (order = 0; order < MAX_ORDER; order++) {
unsigned long blocks;

- /* Count number of free blocks */
- blocks = zone->free_area[order].nr_free;
+ /*
+ * Count number of free blocks.
+ *
+ * Access to nr_free is lockless as nr_free is used only for
+ * diagnostic purposes. Use data_race to avoid KCSAN warning.
+ */
+ blocks = data_race(zone->free_area[order].nr_free);
info->free_blocks_total += blocks;

/* Count free base pages */
@@ -1445,7 +1450,11 @@ static void frag_show_print(struct seq_file *m, pg_data_t *pgdat,

seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
for (order = 0; order < MAX_ORDER; ++order)
- seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
+ /*
+ * Access to nr_free is lockless as nr_free is used only for
+ * printing purposes. Use data_race to avoid KCSAN warning.
+ */
+ seq_printf(m, "%6lu ", data_race(zone->free_area[order].nr_free));
seq_putc(m, '\n');
}

--
2.18.0.huawei.25