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;
Message-ID: <6b2fc148-3bf9-83d5-fd5e-242ff51c9c96@kernel.dk>
Date:   Fri, 9 Jun 2023 11:41:51 -0600
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux aarch64; rv:102.0) Gecko/20100101
 Thunderbird/102.11.0
Subject: Re: [PATCH -next] blk-mq: fix potential io hang by wrong 'wake_batch'
Content-Language: en-US
To:     Yu Kuai <yukuai1@huaweicloud.com>, jack@suse.cz,
        andriy.shevchenko@linux.intel.com, qiulaibin@huawei.com
Cc:     linux-block@vger.kernel.org, linux-kernel@vger.kernel.org,
        yukuai3@huawei.com, yi.zhang@huawei.com, yangerkun@huawei.com
References: <20230609085130.2320859-1-yukuai1@huaweicloud.com>
From:   Jens Axboe <axboe@kernel.dk>
In-Reply-To: <20230609085130.2320859-1-yukuai1@huaweicloud.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 6/9/23 2:51?AM, Yu Kuai wrote:
> From: Yu Kuai <yukuai3@huawei.com>
> 
> In __blk_mq_tag_busy/idle(), updating 'active_queues' and calculating
> 'wake_batch' is not atomic:
> 
> t1:			t2:
> _blk_mq_tag_busy	blk_mq_tag_busy
> inc active_queues
> // assume 1->2
> 			inc active_queues
> 			// 2 -> 3
> 			blk_mq_update_wake_batch
> 			// calculate based on 3
> blk_mq_update_wake_batch
> /* calculate based on 2, while active_queues is actually 3. */
> 
> Fix this problem by protecting them wih 'tags->lock', this is not a hot
> path, so performance should not be concerned.
> 
> Fixes: 180dccb0dba4 ("blk-mq: fix tag_get wait task can't be awakened")
> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
> ---
>  block/blk-mq-tag.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
> index dfd81cab5788..43fe523f39c7 100644
> --- a/block/blk-mq-tag.c
> +++ b/block/blk-mq-tag.c
> @@ -55,9 +55,10 @@ void __blk_mq_tag_busy(struct blk_mq_hw_ctx *hctx)
>  			return;
>  	}
>  
> +	spin_lock_irq(&hctx->tags->lock);
>  	users = atomic_inc_return(&hctx->tags->active_queues);
> -
>  	blk_mq_update_wake_batch(hctx->tags, users);
> +	spin_unlock_irq(&hctx->tags->lock);
>  }
>  
>  /*
> @@ -90,9 +91,10 @@ void __blk_mq_tag_idle(struct blk_mq_hw_ctx *hctx)
>  			return;
>  	}
>  
> +	spin_lock_irq(&tags->lock);
>  	users = atomic_dec_return(&tags->active_queues);
> -
>  	blk_mq_update_wake_batch(tags, users);
> +	spin_unlock_irq(&tags->lock);
>  
>  	blk_mq_tag_wakeup_all(tags, false);
>  }

From a quick look, these are the only manipulators of active_queues.
If we're under the tags lock, why do they still need to be atomics?

-- 
Jens Axboe