Received-SPF: pass (google.com: domain of linux-kernel+bounces-41356-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Sender: Tejun Heo <htejun@gmail.com>
Date: Sat, 27 Jan 2024 08:37:45 -1000
From: Tejun Heo <tj@kernel.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mikulas Patocka <mpatocka@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>, linux-kernel@vger.kernel.org,
	dm-devel@lists.linux.dev, Mike Snitzer <msnitzer@redhat.com>,
	Ignat Korchagin <ignat@cloudflare.com>,
	Damien Le Moal <damien.lemoal@wdc.com>,
	Bob Liu <bob.liu@oracle.com>, Hou Tao <houtao1@huawei.com>,
	Nathan Huckleberry <nhuck@google.com>,
	Peter Zijlstra <peterz@infradead.org>, Ingo Molnar <mingo@elte.hu>
Subject: Re: [PATCH] softirq: fix memory corruption when freeing
 tasklet_struct
Message-ID: <ZbVNeVkGItt1KTan@slm.duckdns.org>
References: <82b964f0-c2c8-a2c6-5b1f-f3145dc2c8e5@redhat.com>
 <CAHk-=wjDW53w4-YcSmgKC5RruiRLHmJ1sXeYdp_ZgVoBw=5byA@mail.gmail.com>
 <ZbRDnSyiSE93puQL@slm.duckdns.org>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZbRDnSyiSE93puQL@slm.duckdns.org>

On Fri, Jan 26, 2024 at 01:43:25PM -1000, Tejun Heo wrote:
> Hello,
> 
> The following is a draft patch which implements atomic workqueues and
> convert dm-crypt to use it instead of tasklet. It's an early draft and very
> lightly tested but seems to work more or less. It's on top of wq/for6.9 + a
> pending patchset. The following git branch can be used for testing.
> 
>   git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git wq-atomic-draft
> 
> I'll go over it to make sure all the pieces work. While it adds some
> complications, it doesn't seem too bad and conversion from tasklet should be
> straightforward too.
> 
> - It hooks into tasklet[_hi] for now but if we get to update all of tasklet
>   users, we can just repurpose the tasklet softirq slots directly.
> 
> - I thought about allowing busy-waits for flushes and cancels but it didn't
>   seem necessary. Keeping them blocking has the benefit of avoiding possible
>   nasty deadlocks. We can revisit if there's need.
> 
> - Compared to tasklet, each work item goes through a bit more management
>   code because I wanted to keep the code as unified as possible to regular
>   threaded workqueues. That said, it's not a huge amount and my bet is that
>   the difference is unlikely to be noticeable.

Should have known when it worked too well on the first try but I missed a
part in init and this was just running them on per-cpu workqueues. Will post
an actually working version later.

Thanks.

-- 
tejun