Re: [PATCH] block: Limit number of items taken from the I/O scheduler in one go

From: Ming Lei <ming.lei@redhat.com>
To: Salman Qazi <sqazi@google.com>
Cc: Jens Axboe <axboe@kernel.dk>,
	Bart Van Assche <bvanassche@acm.org>,
	linux-block@vger.kernel.org, linux-kernel@vger.kernel.org,
	Jesse Barnes <jsbarnes@google.com>,
	Gwendal Grignou <gwendal@google.com>,
	Hannes Reinecke <hare@suse.com>, Christoph Hellwig <hch@lst.de>
Subject: Re: [PATCH] block: Limit number of items taken from the I/O scheduler in one go
Date: Wed, 5 Feb 2020 12:55:26 +0800	[thread overview]
Message-ID: <20200205045526.GA15286@ming.t460p> (raw)
In-Reply-To: <20200204193711.257285-1-sqazi@google.com>

On Tue, Feb 04, 2020 at 11:37:11AM -0800, Salman Qazi wrote:
> Flushes bypass the I/O scheduler and get added to hctx->dispatch
> in blk_mq_sched_bypass_insert.  This can happen while a kworker is running
> hctx->run_work work item and is past the point in
> blk_mq_sched_dispatch_requests where hctx->dispatch is checked.
> 
> The blk_mq_do_dispatch_sched call is not guaranteed to end in bounded time,
> because the I/O scheduler can feed an arbitrary number of commands.
> 
> Since we have only one hctx->run_work, the commands waiting in
> hctx->dispatch will wait an arbitrary length of time for run_work to be
> rerun.
> 
> A similar phenomenon exists with dispatches from the software queue.
> 
> The solution is to poll hctx->dispatch in blk_mq_do_dispatch_sched and
> blk_mq_do_dispatch_ctx and return from the run_work handler and let it
> rerun.
> 
> Signed-off-by: Salman Qazi <sqazi@google.com>
> ---
>  block/blk-mq-sched.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
> index ca22afd47b3d..d1b8b31bc3d4 100644
> --- a/block/blk-mq-sched.c
> +++ b/block/blk-mq-sched.c
> @@ -97,6 +97,9 @@ static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
>  		if (e->type->ops.has_work && !e->type->ops.has_work(hctx))
>  			break;
>  
> +		if (!list_empty_careful(&hctx->dispatch))
> +			break;
> +
>  		if (!blk_mq_get_dispatch_budget(hctx))
>  			break;
>  
> @@ -140,6 +143,9 @@ static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx)
>  	do {
>  		struct request *rq;
>  
> +		if (!list_empty_careful(&hctx->dispatch))
> +			break;
> +
>  		if (!sbitmap_any_bit_set(&hctx->ctx_map))
>  			break;

This approach looks good, given actually we retrieved request this way in
legacy IO request path, see __elv_next_request().

However,  blk_mq_request_bypass_insert() may be run at the same time, so
this patch may cause requests stalled in scheduler queue.

How about returning if there is request available in hctx->dispatch from
the two helpers, then re-dispatch requests in blk_mq_sched_dispatch_requests()
if yes.

Thanks,
Ming