Re: [PATCH v7 12/16] blk-iocost: fix incorrect lock order for rq_qos_mutex and freeze queue

[Ming Lei <ming.lei@redhat.com>]

On Wed, Dec 31, 2025 at 04:51:22PM +0800, Yu Kuai wrote:
> Like wbt, rq_qos_add() can be called from two path and the lock order
> are inversely:
> 
> - From ioc_qos_write(), queue is already frozen before rq_qos_add();
> - From ioc_cost_model_write(), rq_qos_add() is called directly;
> 
> Fix this problem by converting to use blkg_conf_open_bdev_frozen()
> from ioc_cost_model_write(), then since all rq_qos_add() callers
> already freeze queue, convert to use rq_qos_add_frozen().
> 
> Signed-off-by: Yu Kuai <yukuai@fnnas.com>
> ---
>  block/blk-iocost.c | 17 ++++++++---------
>  1 file changed, 8 insertions(+), 9 deletions(-)
> 
> diff --git a/block/blk-iocost.c b/block/blk-iocost.c
> index a0416927d33d..929fc1421d7e 100644
> --- a/block/blk-iocost.c
> +++ b/block/blk-iocost.c
> @@ -2925,7 +2925,7 @@ static int blk_iocost_init(struct gendisk *disk)
>  	 * called before policy activation completion, can't assume that the
>  	 * target bio has an iocg associated and need to test for NULL iocg.
>  	 */
> -	ret = rq_qos_add(&ioc->rqos, disk, RQ_QOS_COST, &ioc_rqos_ops);
> +	ret = rq_qos_add_frozen(&ioc->rqos, disk, RQ_QOS_COST, &ioc_rqos_ops);
>  	if (ret)
>  		goto err_free_ioc;
>  
> @@ -3408,7 +3408,7 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
>  {
>  	struct blkg_conf_ctx ctx;
>  	struct request_queue *q;
> -	unsigned int memflags;
> +	unsigned long memflags;
>  	struct ioc *ioc;
>  	u64 u[NR_I_LCOEFS];
>  	bool user;
> @@ -3417,9 +3417,11 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
>  
>  	blkg_conf_init(&ctx, input);
>  
> -	ret = blkg_conf_open_bdev(&ctx);
> -	if (ret)
> +	memflags = blkg_conf_open_bdev_frozen(&ctx);
> +	if (IS_ERR_VALUE(memflags)) {
> +		ret = memflags;
>  		goto err;
> +	}
>  

The following blk_iocost_init() runs into percpu allocation, so it will
create new lockdep warning.



Thanks,
Ming