blk-cgroup: Make 64bit per cpu stats safe on 32bit arch

Some of the stats are 64bit and updation will be non atomic on 32bit
architecture. Use sequence counters on 32bit arch to make reading
of stats safe.

Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index 34bfcef..3622518e 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -400,14 +400,25 @@
 				uint64_t bytes, bool direction, bool sync)
 {
 	struct blkio_group_stats_cpu *stats_cpu;
+	unsigned long flags;
+
+	/*
+	 * Disabling interrupts to provide mutual exclusion between two
+	 * writes on same cpu. It probably is not needed for 64bit. Not
+	 * optimizing that case yet.
+	 */
+	local_irq_save(flags);
 
 	stats_cpu = this_cpu_ptr(blkg->stats_cpu);
 
+	u64_stats_update_begin(&stats_cpu->syncp);
 	stats_cpu->sectors += bytes >> 9;
 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
 			1, direction, sync);
 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
 			bytes, direction, sync);
+	u64_stats_update_end(&stats_cpu->syncp);
+	local_irq_restore(flags);
 }
 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
 
@@ -622,15 +633,21 @@
 {
 	int cpu;
 	struct blkio_group_stats_cpu *stats_cpu;
-	uint64_t val = 0;
+	u64 val = 0, tval;
 
 	for_each_possible_cpu(cpu) {
+		unsigned int start;
 		stats_cpu  = per_cpu_ptr(blkg->stats_cpu, cpu);
 
-		if (type == BLKIO_STAT_CPU_SECTORS)
-			val += stats_cpu->sectors;
-		else
-			val += stats_cpu->stat_arr_cpu[type][sub_type];
+		do {
+			start = u64_stats_fetch_begin(&stats_cpu->syncp);
+			if (type == BLKIO_STAT_CPU_SECTORS)
+				tval = stats_cpu->sectors;
+			else
+				tval = stats_cpu->stat_arr_cpu[type][sub_type];
+		} while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
+
+		val += tval;
 	}
 
 	return val;