Re: [PATCH] block: directly insert blk-mq request from blk_insert_cloned_request()

[Jens Axboe <axboe@kernel.dk>]

On 09/08/2017 03:42 PM, Mike Snitzer wrote:
> diff --git a/block/blk-core.c b/block/blk-core.c
> index d709c0e..7a06b2b 100644
> --- a/block/blk-core.c
> +++ b/block/blk-core.c
> @@ -2342,7 +2342,7 @@ blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *
>  	if (q->mq_ops) {
>  		if (blk_queue_io_stat(q))
>  			blk_account_io_start(rq, true);
> -		blk_mq_sched_insert_request(rq, false, true, false, false);
> +		blk_mq_insert_request(rq);
>  		return BLK_STS_OK;
>  	}

I think this is fine, since only dm uses this function. Would be nice to
have some check though, to ensure it doesn't get misused in the future.

> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 3f18cff..5c5bb3f 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -1401,6 +1401,24 @@ void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
>  	blk_mq_hctx_mark_pending(hctx, ctx);
>  }
>  
> +static inline void blk_mq_queue_io(struct blk_mq_hw_ctx *hctx,
> +				   struct blk_mq_ctx *ctx,
> +				   struct request *rq)
> +{
> +	spin_lock(&ctx->lock);
> +	__blk_mq_insert_request(hctx, rq, false);
> +	spin_unlock(&ctx->lock);
> +}

Any particular reason it isn't just added to the dispatch queue?

> +void blk_mq_insert_request(struct request *rq)
> +{
> +	struct blk_mq_ctx *ctx = rq->mq_ctx;
> +	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(rq->q, ctx->cpu);
> +
> +	blk_mq_queue_io(hctx, ctx, rq);
> +	blk_mq_run_hw_queue(hctx, false);
> +}

Would probably be cleaner as blk_mq_insert_and_run_request() or
something, to make sure it's understood that it also runs the queue.

-- 
Jens Axboe