Received-SPF: pass (google.com: domain of linux-kernel+bounces-25126-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Sender: Tejun Heo <htejun@gmail.com>
Date: Fri, 12 Jan 2024 14:18:02 -1000
From: Tejun Heo <tj@kernel.org>
To: jiangshanlai@gmail.com
Cc: linux-kernel@vger.kernel.org, Naohiro.Aota@wdc.com,
	kernel-team@meta.com
Subject: Re: [PATCH 10/10] workqueue: Reimplement ordered workqueue using
 shared nr_active
Message-ID: <ZaHWukHxfndSWAOK@slm.duckdns.org>
References: <20231220072529.1036099-1-tj@kernel.org>
 <20231220072529.1036099-11-tj@kernel.org>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20231220072529.1036099-11-tj@kernel.org>

On Wed, Dec 20, 2023 at 04:24:41PM +0900, Tejun Heo wrote:
> Because nr_active used to be tied to pwq, an ordered workqueue had to have a
> single pwq to guarantee strict ordering. This led to several contortions to
> avoid creating multiple pwqs.
> 
> Now that nr_active can be shared across multiple pwqs, we can simplify
> ordered workqueue implementation. All that's necessary is ensuring that a
> single wq_node_nr_active is shared across all pwqs, which is achieved by
> making wq_node_nr_active() always return wq->node_nr_active[nr_node_ids] for
> ordered workqueues.
> 
> The new implementation is simpler and allows ordered workqueues to share
> locality aware worker_pools with other unbound workqueues which should
> improve execution locality.
> 
> Signed-off-by: Tejun Heo <tj@kernel.org>

This patch breaks ordered workqueues as the inactive pwq RR logic doesn't
follow work item queueing order. I could reproduce severe perf degradations
and outright hangs. I'm dropping this patch.

Thanks.

-- 
tejun