Date:   Tue, 11 Jan 2022 16:15:04 +0200
From:   Andy Shevchenko <andriy.shevchenko@linux.intel.com>
To:     Laibin Qiu <qiulaibin@huawei.com>
Cc:     axboe@kernel.dk, ming.lei@redhat.com, john.garry@huawei.com,
        martin.petersen@oracle.com, hare@suse.de,
        johannes.thumshirn@wdc.com, bvanassche@acm.org,
        linux-block@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH -next v4] blk-mq: fix tag_get wait task can't be awakened
Message-ID: <Yd2Q6LyJUDAU54Dt@smile.fi.intel.com>
References: <20220111140216.1858823-1-qiulaibin@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220111140216.1858823-1-qiulaibin@huawei.com>
Organization: Intel Finland Oy - BIC 0357606-4 - Westendinkatu 7, 02160 Espoo
Precedence: bulk

On Tue, Jan 11, 2022 at 10:02:16PM +0800, Laibin Qiu wrote:
> In case of shared tags, there might be more than one hctx which
> allocates from the same tags, and each hctx is limited to allocate at
> most:
>         hctx_max_depth = max((bt->sb.depth + users - 1) / users, 4U);
> 
> tag idle detection is lazy, and may be delayed for 30sec, so there
> could be just one real active hctx(queue) but all others are actually
> idle and still accounted as active because of the lazy idle detection.
> Then if wake_batch is > hctx_max_depth, driver tag allocation may wait
> forever on this real active hctx.
> 
> Fix this by recalculating wake_batch when inc or dec active_queues.

...

>  {
> +	unsigned int users;

Missed blank line here.

>  	if (blk_mq_is_shared_tags(hctx->flags)) {
>  		struct request_queue *q = hctx->queue;
>  
> +		if (test_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags) ||
> +		    test_and_set_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags)) {

Whoever wrote this code did too much defensive programming, because the first
conditional doesn't make much sense here. Am I right?

> +			return true;
> +		}
>  	} else {

> +		if (test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state) ||
> +		    test_and_set_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state)) {

Ditto.

> +			return true;
> +		}
>  	}

...

> +	unsigned int wake_batch = clamp_t(unsigned int,
> +			(sbq->sb.depth + users - 1) / users, 4U, SBQ_WAKE_BATCH);


	unsigned int wake_batch;

	wake_batch = clamp_val((sbq->sb.depth + users - 1) / users, 4, SBQ_WAKE_BATCH);
	...

is easier to read, no?

-- 
With Best Regards,
Andy Shevchenko