RE: [PATCH V2] Consider inflight IO in io accounting for high latency devices

[Gulam Mohamed <gulam.mohamed@xxxxxxxxxx>]

Hi Bart,

   Thanks for your review. Can you please see my inline comments?

Regards,
Gulam Mohamed.

-----Original Message-----
From: Bart Van Assche <bvanassche@xxxxxxx> 
Sent: Saturday, October 14, 2023 2:43 AM
To: Gulam Mohamed <gulam.mohamed@xxxxxxxxxx>; axboe@xxxxxxxxx; linux-block@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
Subject: Re: [PATCH V2] Consider inflight IO in io accounting for high latency devices

On 10/13/23 12:55, Gulam Mohamed wrote:
> @@ -1015,7 +1018,13 @@ static inline void blk_account_io_start(struct request *req)
>   			req->part = req->q->disk->part0;
>   
>   		part_stat_lock();
> -		update_io_ticks(req->part, jiffies, false);
> +
> +		if (req->q->nr_hw_queues == 1) {
> +			hctx = xa_load(&req->q->hctx_table, 0);
> +			inflight = blk_mq_hctx_has_tags(hctx);
> +		}
> +
> +		update_io_ticks(req->part, jiffies, inflight);
>   		part_stat_unlock();
>   	}
>   }

blk_account_io_start() is called by blk_mq_bio_to_request(). So if I/O statistics are enabled and if there is only a single hardware queue,
blk_mq_hctx_has_tags() will be called every time a bio is submitted?
The blk_mq_hctx_has_tags() function iterates over all tags. I would be surprised if anyone would consider the overhead of this approach acceptable.

[GULAM]: Yes, it will be called for every submitted bio but for the high latency devices it will not have much impact. This is indicated by the latency figures I provided in the review mail, with and without our patch.

Thanks,

Bart.