From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <axboe@kernel.dk>
Subject: Re: WARNING: CPU: 2 PID: 207 at drivers/nvme/host/core.c:527
 nvme_setup_cmd+0x3d3
To: Keith Busch <keith.busch@intel.com>
Cc: "jianchao.wang" <jianchao.w.wang@oracle.com>,
 "linux-block@vger.kernel.org" <linux-block@vger.kernel.org>,
 Christoph Hellwig <hch@lst.de>,
 "linux-nvme@lists.infradead.org" <linux-nvme@lists.infradead.org>,
 Ming Lei <ming.lei@redhat.com>
References: <45f93661-da0d-94c5-1740-85242df8776e@kernel.dk>
 <0872b361-157b-a876-20af-3d7a4ee7ff31@kernel.dk>
 <8fd916ab-42d7-c654-5a01-8f1eb4be730e@oracle.com>
 <0b7686b3-f716-49ba-c7c4-929d84905569@kernel.dk>
 <7459ffed-c63c-38a9-84f5-456c2a5c4fe0@oracle.com>
 <0f6c248b-eb67-9b02-a4b9-0366d476e70d@kernel.dk>
 <20180201045645.GF27735@localhost.localdomain>
From: Jens Axboe <axboe@kernel.dk>
Message-ID: <576ba7dc-78cf-b5d2-d1de-446d1f4b2732@kernel.dk>
Date: Thu, 1 Feb 2018 08:26:11 -0700
MIME-Version: 1.0
In-Reply-To: <20180201045645.GF27735@localhost.localdomain>
Content-Type: text/plain; charset=utf-8
List-ID: <linux-block@vger.kernel.org>

On 1/31/18 9:56 PM, Keith Busch wrote:
> On Wed, Jan 31, 2018 at 08:07:41PM -0700, Jens Axboe wrote:
>>  	if (total_phys_segments > queue_max_segments(q))
>> -		return 0;
>> +			return 0;
> 
> This perhaps unintended change happens to point out another problem:
> queue_max_segments is the wrong limit for discards, which require
> queue_max_discard_segments.
> 
> It might be easier to merge discard requests special, like how merging
> a discard bio is handled (untested).

Yeah agreed, we should just split it up completely instead instead of
special casing it in the read/write path.


> diff --git a/block/blk-merge.c b/block/blk-merge.c
> index 8452fc7164cc..01671e1373ff 100644
> --- a/block/blk-merge.c
> +++ b/block/blk-merge.c
> @@ -550,6 +550,28 @@ static bool req_no_special_merge(struct request *req)
>  	return !q->mq_ops && req->special;
>  }
>  
> +static bool req_attempt_discard_merge(struct request_queue *q, struct request *req,
> +		struct request *next)
> +{
> +	unsigned short segments = blk_rq_nr_discard_segments(req);
> +
> +	if (segments >= queue_max_discard_segments(q))
> +		goto no_merge;
> +	if (blk_rq_sectors(req) + bio_sectors(next->bio) >
> +	    blk_rq_get_max_sectors(req, blk_rq_pos(req)))
> +		goto no_merge;
> +
> +	req->biotail->bi_next = next->bio;
> +	req->biotail = next->biotail;
> +	req->nr_phys_segments = segments + blk_rq_nr_discard_segments(next);
> +	next->bio = NULL;
> +	return true;
> +
> +no_merge:
> +	req_set_nomerge(q, req);
> +	return false;
> +}
> +
>  static int ll_merge_requests_fn(struct request_queue *q, struct request *req,
>  				struct request *next)
>  {
> @@ -679,6 +701,15 @@ static struct request *attempt_merge(struct request_queue *q,
>  	if (req->write_hint != next->write_hint)
>  		return NULL;
>  
> +	/*
> +	 * Discards are ... special.
> +	 */
> +	if (req_op(req) == REQ_OP_DISCARD) {
> +		if (req_attempt_discard_merge(q, req, next))
> +			return next;
> +		return NULL;
> +	}
> +
>  	/*
>  	 * If we are allowed to merge, then append bio list
>  	 * from next to rq and release next. merge_requests_fn

This looks fine to me, the bio-to-request merge path already looks correct.
Care to send a properly formatted patch?

-- 
Jens Axboe