Re: [PATCH v3 1/1] nvme: multipath: Implemented new iopolicy "queue-depth"

[Keith Busch <kbusch@kernel.org>]

On Tue, May 21, 2024 at 02:18:09PM +0530, Nilay Shroff wrote:
> On 5/21/24 01:50, John Meneghini wrote:
> > @@ -140,8 +148,12 @@ void nvme_mpath_end_request(struct request *rq)
> >  {
> >  	struct nvme_ns *ns = rq->q->queuedata;
> >  
> > +	if ((nvme_req(rq)->flags & NVME_MPATH_CNT_ACTIVE))
> > +		atomic_dec_if_positive(&ns->ctrl->nr_active);
> > +
> >  	if (!(nvme_req(rq)->flags & NVME_MPATH_IO_STATS))
> >  		return;
> > +
> >  	bdev_end_io_acct(ns->head->disk->part0, req_op(rq),
> >  			 blk_rq_bytes(rq) >> SECTOR_SHIFT,
> >  			 nvme_req(rq)->start_time);
> > @@ -330,6 +342,40 @@ static struct nvme_ns *nvme_round_robin_path(struct nvme_ns_head *head,
> >  	return found;
> >  }
> >  
> I think you may also want to reset nr_active counter if in case, in-flight nvme request 
> is cancelled. If the request is cancelled then nvme_mpath_end_request() wouldn't be invoked.
> So you may want to reset nr_active counter from nvme_cancel_request() as below:
> 
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index bf7615cb36ee..4fea7883ce8e 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -497,8 +497,9 @@ EXPORT_SYMBOL_GPL(nvme_host_path_error);
>  
>  bool nvme_cancel_request(struct request *req, void *data)
>  {
> -       dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
> -                               "Cancelling I/O %d", req->tag);
> +       struct nvme_ctrl *ctrl = (struct nvme_ctrl *)data;
> +
> +       dev_dbg_ratelimited(ctrl->device, "Cancelling I/O %d", req->tag);
>  
>         /* don't abort one completed or idle request */
>         if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
> @@ -506,6 +507,8 @@ bool nvme_cancel_request(struct request *req, void *data)
>  
>         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
>         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
> +       if ((nvme_req(rq)->flags & NVME_MPATH_CNT_ACTIVE))
> +               atomic_dec(&ctrl->nr_active);
>         blk_mq_complete_request(req);
>         return true;
>  }

The io stats wouldn't be right if that happened. And maybe it isn't
right on a failover, but it needs to be. Would it work if
nvme_failover_req() calls nvme_end_req() instead of directly calling
blk_mq_end_req()?