Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 17 May 2023 22:22:19 +0000
In-Reply-To: <ZGVLo1V4kjx_lep5@slm.duckdns.org>
Mime-Version: 1.0
References: <ZGVLo1V4kjx_lep5@slm.duckdns.org>
Message-ID: <20230517222219.3191560-1-aliceryhl@google.com>
Subject: Re: [PATCH v1 0/7] Bindings for the workqueue
From:   Alice Ryhl <aliceryhl@google.com>
To:     tj@kernel.org
Cc:     alex.gaynor@gmail.com, aliceryhl@google.com,
        benno.lossin@proton.me, bjorn3_gh@protonmail.com,
        boqun.feng@gmail.com, gary@garyguo.net, jiangshanlai@gmail.com,
        linux-kernel@vger.kernel.org, ojeda@kernel.org,
        patches@lists.linux.dev, rust-for-linux@vger.kernel.org,
        wedsonaf@gmail.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, 17 May 2023 11:48:19 -1000, Tejun Heo wrote:
> I tried to read the patches but am too dumb to understand much.

The patch is more complicated than I would have liked, unfortunately.
However, as I mentioned in the cover letter, simplifications should be
on their way.

Luckily, using the workqueue bindings is simpler than the bindings
themselves.

> Any chance you can provide some examples so that I can at least
> imagine how workqueue would be used from rust side?

Yes, of course!

The simplest way to use the workqueue is to use the `try_spawn` method
introduced by the last patch in the series. With this function, you just
pass a function pointer to the `try_spawn` method, and it schedules the
function for execution. Unfortunately this allocates memory, making it
a fallible operation.

To avoid allocation memory, we do something else. As an example, we can
look at the Rust binder driver that I am currently working on. Here is
how it will be used in the binder driver: First, the `Process` struct
will be given a `work_struct` field:

#[pin_data]
pub(crate) struct Process {
    // Work node for deferred work item.
    #[pin]
    defer_work: Work<Arc<Process>>,

    // Other fields follow...
}

Here, we use the type `Work<Arc<Process>>` for our field. This type is
the Rust wrapper for `work_struct`. The generic parameter to `Work`
should be the pointer type used to access `Process`, and in this case it
is `Arc<Process>`. The pointer type `Arc` is used for reference
counting, and its a pointer type that owns a ref-count to the inner
value. (So e.g., it decrements the ref-cout when the arc goes out of
scope.) Arc is an abbreviation of "atomic reference count". This means
that while it is enqueued in the workqueue, the workqueue owns a
ref-count to the process.

Next, binder will use the `impl_has_work!` macro to declare that it
wants to use `defer_work` as its `work_struct` field. That looks like
this:

kernel::impl_has_work! {
    impl HasWork<Arc<Process>> for Process { self.defer_work }
}

To define the code that should run when the work item is executed on the
workqueue, binder does the following:

impl workqueue::ArcWorkItem for Process {
    fn run(self: Arc<Process>) {
        // this runs when the work item is executed
    }
}

Finally to schedule it to the system workqueue, it does the following:

let _ = workqueue::system().enqueue(process);

Here, the `enqueue` call is fallible, since it might fail if the process
has already been enqueued to a work queue. However, binder just uses
`let _ =` to ignore the failure, since it doesn't need to do anything
special in that case.

I hope that helps, and let me know if you have any further questions.

Alice