Re: [PATCH v3 1/2] rust: irq: add support for request_irq()

[Danilo Krummrich <dakr@kernel.org>]

On Thu, May 15, 2025 at 09:27:51AM -0300, Daniel Almeida wrote:
> 
> 
> > On 15 May 2025, at 09:04, Danilo Krummrich <dakr@kernel.org> wrote:
> > 
> > On Thu, May 15, 2025 at 08:54:35AM -0300, Daniel Almeida wrote:
> >> Hi Danilo,
> >> 
> >>> On 14 May 2025, at 18:53, Danilo Krummrich <dakr@kernel.org> wrote:
> >>> 
> >>> On Wed, May 14, 2025 at 04:20:51PM -0300, Daniel Almeida wrote:
> >>>> +/// // This is running in process context.
> >>>> +/// fn register_irq(irq: u32, handler: Handler) -> Result<Arc<Registration<Handler>>> {
> >>>> +///     let registration = Registration::register(irq, flags::SHARED, c_str!("my-device"), handler);
> >>>> +///
> >>>> +///     // You can have as many references to the registration as you want, so
> >>>> +///     // multiple parts of the driver can access it.
> >>>> +///     let registration = Arc::pin_init(registration, GFP_KERNEL)?;
> >>> 
> >>> This makes it possible to arbitrarily extend the lifetime of an IRQ
> >>> registration. However, we must guarantee that the IRQ is unregistered when the
> >>> corresponding device is unbound. We can't allow drivers to hold on to device
> >>> resources after the corresponding device has been unbound.
> >>> 
> >>> Why does the data need to be part of the IRQ registration itself? Why can't we
> >>> pass in an Arc<T> instance already when we register the IRQ?
> >>> 
> >>> This way we'd never have a reason to ever access the Registration instance
> >>> itself ever again and we can easily wrap it as Devres<irq::Registration> -
> >>> analogously to devm_request_irq() on the C side - without any penalties.
> >>> 
> >>>> +///     // The handler may be called immediately after the function above
> >>>> +///     // returns, possibly in a different CPU.
> >>>> +///
> >>>> +///     {
> >>>> +///         // The data can be accessed from the process context too.
> >>>> +///         let mut data = registration.handler().0.lock();
> >>>> +///         *data = 42;
> >>>> +///     }
> >>>> +///
> >>>> +///     Ok(registration)
> >>>> +/// }
> >>> 
> >> 
> >> Up until this point, there was no need for the data to not be inline with the
> >> registration. This new design would force an Arc, which, apart from the
> >> heap-allocation, is restrictive for users.
> > 
> > Does the current design not also imply a heap allocation heap allocation? With
> > my proposal irq::Registration::new() can just return an irq::Registration
> > instance, not an impl PinInit that you need to stuff into a Box or Arc instead.
> > Hence, there shouldn't be a difference.
> 
> Well, not really, because this impl PinInit can be assigned to something larger
> that is already pinned, like drm::Device::Data for example, which is (or was)
> already behind an Arc, or any other private data in other subsystems.
> 
> IIUC what you proposed has yet another indirection. If we reuse the example
> from above, that would be an Arc for the drm Data, and another Arc for the
> handler itself?

Can't you implement Handler for drm::Device::Data and e.g. make Registration
take an Arc<T: Handler>?

The irq::Registration itself doesn't need to be allocated dynamically, so it'd
still be a single allocation, no?

> I definitely see your point here, I am just trying to brainstorm another way of
> doing this.
> > 
> >> Can’t we use Devres with the current implementation?
> >> 
> >> IIUC from a very cursory glance, all that would mean is that you'd have to call
> >> try_access() on your handler, which should be fine?
> > 
> > Well, that would work indeed.
> > 
> > But people will - with good reason - be upset that every access to the handler's
> > data needs to be guarded with the RCU read side critical section implied by
> > Revocable and hence Devres.
> 
> True, I totally missed that.
> 
> > 
> > We can easily avoid that in this case, hence we should do it.
> 
> — Daniel
>