Re: [PATCH v3 1/3] rust: clk: use the type-state pattern

["Gary Guo" <gary@garyguo.net>]

On Mon Jan 19, 2026 at 10:45 AM GMT, Maxime Ripard wrote:
> On Thu, Jan 08, 2026 at 11:14:37AM -0300, Daniel Almeida wrote:
>> Hi Maxime :)
>> 
>> > 
>> > I don't know the typestate pattern that well, but I wonder if we don't
>> > paint ourselves into a corner by introducing it.
>> > 
>> > While it's pretty common to get your clock from the get go into a state,
>> > and then don't modify it (like what devm_clk_get_enabled provides for
>> > example), and the typestate pattern indeed works great for those, we
>> 
>> Minor correction, devm_clk_get_enabled is not handled by the typestate
>> pattern. The next patch does include this function for convenience, but
>> you get a Result<()>. The typestate pattern is used when you want more
>> control.
>>
>> > also have a significant number of drivers that will have a finer-grained
>> > control over the clock enablement for PM.
>> > 
>> > 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.
>> > 
>> > AFAIU, encoding the state of the clock into the Clk type (and thus
>> > forcing the structure that holds it) prevents that mutation. If not, we
>> > should make it clearer (by expanding the doc maybe?) how such a pattern
>> > can be supported.
>> > 
>> > Maxime
>> 
>> IIUC, your main point seems to be about mutating the state at runtime? This is
>> possible with this code. You can just have an enum, for example:
>> 
>> enum MyClocks {
>> 	Unprepared(Clk<Unprepared>),
>>         Prepared(Clk<Prepared>),
>> 	Enabled(Clk<Enabled>), 
>> }
>> 
>> In fact, I specifically wanted to ensure that this was possible when writing
>> these patches, as it’s needed by drivers. If you want to, I can cover that in
>> the examples, no worries.
>
> Yes, that would be great. I do wonder though if it wouldn't make sense
> to turn it the other way around. It creates a fair share of boilerplate
> for a number of drivers. Can't we keep Clk the way it is as a
> lower-level type, and crate a ManagedClk (or whatever name you prefer)
> that drivers can use, and would be returned by higher-level helpers, if
> they so choose?
>
> That way, we do have the typestate API for whoever wants to, without
> creating too much boilerplate for everybody else.

One solution is to have a new typestate `Dynamic` which opts to track things
using variables.

struct Dynamic {
    enabled: bool,
    prepared: bool,
}

trait ClkState {
    // Change to methods
    fn disable_on_drop(&self) -> bool;
}

struct Clk<State> {
    ...
    // Keep an instance, which is zero-sized for everything except `Dynamic`
    state: State,
}

this way we can have runtime-checked state conversions.

Best,
Gary