Re: [PATCH 7/7] blk-mq: don't use the requeue list to queue flush commands

[Bart Van Assche <bvanassche@xxxxxxx>]

On 4/16/23 13:09, Christoph Hellwig wrote:
> diff --git a/block/blk-flush.c b/block/blk-flush.c
> index 69e9806f575455..231d3780e74ad1 100644
> --- a/block/blk-flush.c
> +++ b/block/blk-flush.c
> @@ -188,7 +188,9 @@ static void blk_flush_complete_seq(struct request *rq,
>   
>   	case REQ_FSEQ_DATA:
>   		list_move_tail(&rq->flush.list, &fq->flush_data_in_flight);
> -		blk_mq_add_to_requeue_list(rq, 0);
> +		spin_lock(&q->requeue_lock);
> +		list_add_tail(&rq->queuelist, &q->flush_list);
> +		spin_unlock(&q->requeue_lock);
>   		blk_mq_kick_requeue_list(q);
>   		break;

At least the SCSI core can call blk_flush_complete_seq() from interrupt 
context so I don't think the above code is correct. The call chain is as 
follows:

LLD interrupt handler
  scsi_done()
    scsi_done_internal()
      blk_mq_complete_request()
        scsi_complete()
          scsi_finish_command()
            scsi_io_completion()
              scsi_end_request()
                __blk_mq_end_request()
                  flush_end_io()
                    blk_flush_complete_seq()

Bart.