Received-SPF: pass (google.com: domain of linux-kernel+bounces-65871-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Sender: Tejun Heo <htejun@gmail.com>
Date: Wed, 14 Feb 2024 09:16:57 -1000
From: Tejun Heo <tj@kernel.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH wq/for-6.9] workqueue: Fix queue_work_on() with BH
 workqueues
Message-ID: <Zc0Rqaq5DXcJLTta@slm.duckdns.org>
References: <Zc0IzeM6tAvm1NTl@mtj.duckdns.org>
 <CAHk-=wiVWmTfaCUjZYrV6pVh3O6hYksKsWx4bJY0EhNNat9QDg@mail.gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAHk-=wiVWmTfaCUjZYrV6pVh3O6hYksKsWx4bJY0EhNNat9QDg@mail.gmail.com>

Hello,

On Wed, Feb 14, 2024 at 11:03:46AM -0800, Linus Torvalds wrote:
> On Wed, 14 Feb 2024 at 10:39, Tejun Heo <tj@kernel.org> wrote:
> >
> > When queue_work_on() is used to queue a BH work item on a remote CPU, the
> > work item is queued on that CPU but kick_pool() raises softirq on the local
> > CPU.
> 
> Now, does it make a lot of sense to ask to queue a BH work on another
> CPU in the first place?
> 
> I don't think tasklets supported that. And while the workqueues
> obviously do - and you fix that case - I wonder if we shouldn't say
> "that operation makes no sense, please don't do it" rather than
> actually support it?
> 
> What made you notice this issue?

It's test code I'm using to verify new features to cover
tasklet_disable/enable(), so not a real use case. For tasklet migration,
this isn't necessary but at the same time all the mechanics are already
there, so it's nice to keep the API orthogonal and one can conceive
plausible use cases - e.g. "I'm using per-cpu work items to collect per-cpu
states but that comes with really high tail latencies, lemme switch to
per-cpu BH work items".

Thanks.

-- 
tejun