Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Sender: Tejun Heo <htejun@gmail.com>
Date:   Mon, 14 Nov 2022 12:07:45 -1000
From:   Tejun Heo <tj@kernel.org>
To:     Yu Kuai <yukuai1@huaweicloud.com>
Cc:     hch@lst.de, josef@toxicpanda.com, axboe@kernel.dk,
        yukuai3@huawei.com, cgroups@vger.kernel.org,
        linux-block@vger.kernel.org, linux-kernel@vger.kernel.org,
        yi.zhang@huawei.com
Subject: Re: [PATCH v2 4/5] blk-iocost: fix sleeping in atomic context
 warnning
Message-ID: <Y3K8MSFWw8eTnxtm@slm.duckdns.org>
References: <20221104023938.2346986-1-yukuai1@huaweicloud.com>
 <20221104023938.2346986-5-yukuai1@huaweicloud.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20221104023938.2346986-5-yukuai1@huaweicloud.com>
Precedence: bulk

On Fri, Nov 04, 2022 at 10:39:37AM +0800, Yu Kuai wrote:
> From: Yu Kuai <yukuai3@huawei.com>
> 
> match_u64() is called inside ioc->lock, which causes smatch static
> checker warnings:
> 
> block/blk-iocost.c:3211 ioc_qos_write() warn: sleeping in atomic context
> block/blk-iocost.c:3240 ioc_qos_write() warn: sleeping in atomic context
> block/blk-iocost.c:3407 ioc_cost_model_write() warn: sleeping in atomic
> context
> 
> Fix the problem by introducing a mutex and using it while prasing input
> params.

It bothers me that parsing an u64 string requires a GFP_KERNEL memory
allocation.

> @@ -2801,9 +2806,11 @@ static void ioc_rqos_queue_depth_changed(struct rq_qos *rqos)
>  {
>  	struct ioc *ioc = rqos_to_ioc(rqos);
>  
> +	mutex_lock(&ioc->params_mutex);
>  	spin_lock_irq(&ioc->lock);
>  	ioc_refresh_params(ioc, false);
>  	spin_unlock_irq(&ioc->lock);
> +	mutex_unlock(&ioc->params_mutex);
>  }

Aren't the params still protected by ioc->lock? Why do we need to grab both?

Any chance I can persuade you into updating match_NUMBER() helpers to not
use match_strdup()? They can easily disable irq/preemption and use percpu
buffers and we won't need most of this patchset.

Thanks.

-- 
tejun