Message-ID: <20260203150855.77c93e22@fedora>
Date: Tue, 3 Feb 2026 15:08:55 +0100
From: Boris Brezillon <boris.brezillon@...labora.com>
To: Daniel Almeida <daniel.almeida@...labora.com>
Cc: Alice Ryhl <aliceryhl@...gle.com>, Maxime Ripard <mripard@...nel.org>,
 "Rafael J. Wysocki" <rafael@...nel.org>, Viresh Kumar
 <viresh.kumar@...aro.org>, Danilo Krummrich <dakr@...nel.org>, Maarten
 Lankhorst <maarten.lankhorst@...ux.intel.com>, Thomas Zimmermann
 <tzimmermann@...e.de>, David Airlie <airlied@...il.com>, Simona Vetter
 <simona@...ll.ch>, Drew Fustini <fustini@...nel.org>, Guo Ren
 <guoren@...nel.org>, Fu Wei <wefu@...hat.com>, Uwe Kleine-König <ukleinek@...nel.org>, Michael Turquette
 <mturquette@...libre.com>, Stephen Boyd <sboyd@...nel.org>, Miguel Ojeda
 <ojeda@...nel.org>, Boqun Feng <boqun.feng@...il.com>, Gary Guo
 <gary@...yguo.net>, Björn Roy Baron
 <bjorn3_gh@...tonmail.com>, Benno Lossin <lossin@...nel.org>, Andreas
 Hindborg <a.hindborg@...nel.org>, Trevor Gross <tmgross@...ch.edu>,
 linux-pm@...r.kernel.org, linux-kernel@...r.kernel.org,
 dri-devel@...ts.freedesktop.org, linux-riscv@...ts.infradead.org,
 linux-pwm@...r.kernel.org, linux-clk@...r.kernel.org,
 rust-for-linux@...r.kernel.org
Subject: Re: [PATCH v3 1/3] rust: clk: use the type-state pattern

On Tue, 3 Feb 2026 10:33:34 -0300
Daniel Almeida <daniel.almeida@...labora.com> wrote:

> Hi Boris,
> 
> > On 3 Feb 2026, at 07:39, Boris Brezillon <boris.brezillon@...labora.com> wrote:
> > 
> > On Mon, 19 Jan 2026 12:35:21 +0000
> > Alice Ryhl <aliceryhl@...gle.com> wrote:
> >   
> >> On Mon, Jan 19, 2026 at 11:45:57AM +0100, Maxime Ripard wrote:  
> >>> On Thu, Jan 08, 2026 at 11:14:37AM -0300, Daniel Almeida wrote:    
> >>>>> For example, it's quite typical to have (at least) one clock for the bus
> >>>>> interface that drives the register, and one that drives the main
> >>>>> component logic. The former needs to be enabled only when you're
> >>>>> accessing the registers (and can be abstracted with
> >>>>> regmap_mmio_attach_clk for example), and the latter needs to be enabled
> >>>>> only when the device actually starts operating.
> >>>>> 
> >>>>> You have a similar thing for the prepare vs enable thing. The difference
> >>>>> between the two is that enable can be called into atomic context but
> >>>>> prepare can't.
> >>>>> 
> >>>>> So for drivers that would care about this, you would create your device
> >>>>> with an unprepared clock, and then at various times during the driver
> >>>>> lifetime, you would mutate that state.    
> >> 
> >> The case where you're doing it only while accessing registers is
> >> interesting, because that means the Enable bit may be owned by a local
> >> variable. We may imagine an:
> >> 
> >>    let enabled = self.prepared_clk.enable_scoped();
> >>    ... use registers
> >>    drop(enabled);
> >> 
> >> Now ... this doesn't quite work with the current API - the current
> >> Enabled stated owns both a prepare and enable count, but the above keeps
> >> the prepare count in `self` and the enabled count in a local variable.
> >> But it could be done with a fourth state, or by a closure method:
> >> 
> >>    self.prepared_clk.with_enabled(|| {
> >>        ... use registers
> >>    });
> >> 
> >> All of this would work with an immutable variable of type Clk<Prepared>.  
> > 
> > Hm, maybe it'd make sense to implement Clone so we can have a temporary
> > clk variable that has its own prepare/enable refs and releases them
> > as it goes out of scope. This implies wrapping *mut bindings::clk in an
> > Arc<> because bindings::clk is not ARef, but should be relatively easy
> > to do. Posting the quick experiment I did with this approach, in case
> > you're interested [1]
> > 
> > [1]https://gitlab.freedesktop.org/bbrezillon/linux/-/commit/d5d04da4f4f6192b6e6760d5f861c69596c7d837  
> 
> The problem with what you have suggested is that the previous state is not
> consumed if you can clone it, and consuming the previous state is a pretty key
> element in ensuring you cannot misuse it. For example, here:
> 
> let enabled_clk = prepared_clk.clone().enable()?;
> // do stuff
> // enabled_clk goes out of scope and releases the enable
> // ref it had
> 
> prepared_clk is still alive.

That was intentional in this example. Think about a prepared_clk that's
stored in some driver-internal object, because you want to keep the clk
prepared at all times between the probe() and unbind(). Then you have
some sections where you want to briefly enable the clk to access
registers, and immediately disable it when you're done. The clone()
here guarantees that the initial prepared_clk stays valid.

If you were to disable, unprepare and put the clk when enabled_clk goes
out of scope, you'd go

	let enabled_clk = prepared_clk.enable()?;

and that would still work, it's just not the same use-case.

> Now, this may not be the end of the world in this
> particular case, but for API consistency, I'd say we should probably avoid this
> behavior.
> 
> I see that Alice suggested a closure approach. IMHO, we should use that
> instead.

The closure, while being useful for the above local clk-enablement
example, doesn't allow for passing some Clk<Enabled> guard around, like
you would do with a lock Guard, and I believe that's a useful thing to
have.

Powered by blists - more mailing lists