The traditional single queue block device has no problem. Following is
the output when I writes to sda3 on version v3.10.

$cat /sys/block/sda/sda3/inflight
       0       33
$cat /sys/block/sda/inflight
       0       33


On the other hand, we can analyze this from the code. Following code
path for single-queue block device is from v4.19.

1. When reading '/sys/block/sda/inflight', the statistics is actually
fetched from part 0.

part_inflight_show
  part_in_flight_rw
	inflight[0] = atomic_read(&part->in_flight[0]);
	inflight[1] = atomic_read(&part->in_flight[1]);


2. part 0 will always be updated whenever sub partition is updated.

blk_queue_bio
    add_acct_request
        blk_account_io_start
            part_inc_in_flight
                atomic_inc(&part->in_flight[rw])
                if (part->partno)
			atomic_inc(part0.in_flight[rw]);


On 12/1/20 1:05 AM, Christoph Hellwig wrote:

On Thu, Nov 26, 2020 at 05:48:33PM +0800, Jeffle Xu wrote:

quoted

The inflight of partition 0 doesn't include inflight IOs to all
sub-partitions, since currently mq calculates inflight of specific
partition by simply camparing the value of the partition pointer.

Thus the following case is possible:

$ cat /sys/block/vda/inflight
?? ?? ?? ??0 ?? ?? ?? ??0
$ cat /sys/block/vda/vda1/inflight
?? ?? ?? ??0 ?? ?? ??128

Partition 0 should be specially handled since it represents the whole
disk.

I'm not sure and can see arguments for either side.  In doubt we should
stick to historic behavior, can you check what old kernels (especially
before blk-mq) did?

-- 
Thanks,
Jeffle