Re: [RFC v2 5/5] block: revert back to synchronous request_queue removal

[Bart Van Assche <bvanassche@acm.org>]

On 2020-04-09 14:45, Luis Chamberlain wrote:
> blk_put_queue() puts decrements the refcount for the request_queue
                  ^^^^
        can this word be left out?

> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 8b1cab52cef9..46fee1ef92e3 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -614,6 +614,7 @@ struct request_queue {
>  #define QUEUE_FLAG_PCI_P2PDMA	25	/* device supports PCI p2p requests */
>  #define QUEUE_FLAG_ZONE_RESETALL 26	/* supports Zone Reset All */
>  #define QUEUE_FLAG_RQ_ALLOC_TIME 27	/* record rq->alloc_time_ns */
> +#define QUEUE_FLAG_DEFER_REMOVAL 28	/* defer queue removal */
>  
>  #define QUEUE_FLAG_MQ_DEFAULT	((1 << QUEUE_FLAG_IO_STAT) |		\
>  				 (1 << QUEUE_FLAG_SAME_COMP))
> @@ -648,6 +649,8 @@ bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q);
>  #else
>  #define blk_queue_rq_alloc_time(q)	false
>  #endif
> +#define blk_queue_defer_removal(q) \
> +	test_bit(QUEUE_FLAG_DEFER_REMOVAL, &(q)->queue_flags)

Since blk_queue_defer_removal() has no callers the code that depends on
QUEUE_FLAG_DEFER_REMOVAL to be set will be subject to bitrot. It would
make me happy if the QUEUE_FLAG_DEFER_REMOVAL flag and the code that
depends on that flag would be removed.

Please add a might_sleep() call in blk_put_queue() since with this patch
applied it is no longer allowed to call blk_put_queue() from atomic context.

Thanks,

Bart.