[PATCH 0/3] Documentation for Device / Driver infrastructure

[PATCH 0/3] Documentation for Device / Driver infrastructure

[Danilo Krummrich]

This patch series expands the documentation for the generic device and driver
infrastructure and provides a guideline on how to implement bus and class
specific devices as well as bus specific driver infrastructure, i.e. bus
abstractions. It also expands on the existing documentation of device context
types.

Danilo Krummrich (3):
  device: rust: documentation for DeviceContext
  device: rust: expand documentation for Device
  driver: rust: expand documentation for driver infrastructure

 rust/kernel/device.rs | 198 +++++++++++++++++++++++++++++++++++++-----
 rust/kernel/driver.rs |  82 ++++++++++++++++-
 2 files changed, 254 insertions(+), 26 deletions(-)


base-commit: 9a8682f0875b8cedad42bdfe601e6ab204fad06d
-- 
2.50.0

[PATCH 1/3] device: rust: documentation for DeviceContext

[Danilo Krummrich]

Expand the documentation around DeviceContext states and types, in order
to provide detailed information about their purpose and relationship
with each other.

Signed-off-by: Danilo Krummrich <dakr@kernel.org>
---
 rust/kernel/device.rs | 63 +++++++++++++++++++++++++++++++++++--------
 1 file changed, 52 insertions(+), 11 deletions(-)

diff --git a/rust/kernel/device.rs b/rust/kernel/device.rs
index ca82926fd67f..d7ac56628fe5 100644
--- a/rust/kernel/device.rs
+++ b/rust/kernel/device.rs
@@ -311,28 +311,69 @@ unsafe impl Send for Device {}
 // synchronization in `struct device`.
 unsafe impl Sync for Device {}
 
-/// Marker trait for the context of a bus specific device.
+/// Marker trait for the context or scope of a bus specific device.
 ///
-/// Some functions of a bus specific device should only be called from a certain context, i.e. bus
-/// callbacks, such as `probe()`.
+/// [`DeviceContext`] is a marker trait for structures representing the context of a bus specific
+/// [`Device`].
 ///
-/// This is the marker trait for structures representing the context of a bus specific device.
+/// The specific device context types are: [`CoreInternal`], [`Core`], [`Bound`] and [`Normal`].
+///
+/// [`DeviceContext`] types are hierarchical, which means that there is a strict hierarchy that
+/// defines which [`DeviceContext`] type can be derived from another. For instance, any
+/// [`Device<Core>`] can dereference to a [`Device<Bound>`].
+///
+/// The following enunumeration illustrates the dereference hierarchy of [`DeviceContext`] types.
+///
+/// - [`CoreInternal`] => [`Core`] => [`Bound`] => [`Normal`]
+/// - [`Core`] => [`Bound`] => [`Normal`]
+/// - [`Bound`] => [`Normal`]
+/// - [`Normal`]
+///
+/// Bus devices can automatically implement the dereference hierarchy by using
+/// [`impl_device_context_deref`](kernel::impl_device_context_deref).
 pub trait DeviceContext: private::Sealed {}
 
-/// The [`Normal`] context is the context of a bus specific device when it is not an argument of
-/// any bus callback.
+/// The [`Normal`] context is the default [`DeviceContext`] of any [`Device`].
+///
+/// The normal context does not indicate any specific scope. Any `Device<Ctx>` is also a valid
+/// [`Device<Normal>`]. It is the only [`DeviceContext`] for which it is valid to implement
+/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) for.
 pub struct Normal;
 
-/// The [`Core`] context is the context of a bus specific device when it is supplied as argument of
-/// any of the bus callbacks, such as `probe()`.
+/// The [`Core`] context is the context of a bus specific device when it appears as argument of
+/// any bus specific callback, such as `probe()`.
+///
+/// The core context indicates that the [`Device<Core>`] reference's scope is limited to the bus
+/// callback it appears in. It is intended to be used for synchronization purposes. Bus device
+/// implementations can implement methods for [`Device<Core>`], such that they can only be called
+/// from bus callbacks.
 pub struct Core;
 
-/// Semantically the same as [`Core`] but reserved for internal usage of the corresponding bus
+/// Semantically the same as [`Core`], but reserved for internal usage of the corresponding bus
 /// abstraction.
+///
+/// The internal core context is intended to be used in exactly the same way as the [Core] context,
+/// with the difference that this [`DeviceContext`] is internal to the corresponding bus
+/// abstraction.
+///
+/// This context mainly exists to share generic [`Device`] infrastructure that should only be called
+/// from bus callbacks with bus abstractions, but without making them accessible for drivers.
 pub struct CoreInternal;
 
-/// The [`Bound`] context is the context of a bus specific device reference when it is guaranteed to
-/// be bound for the duration of its lifetime.
+/// The [`Bound`] context is the [`DeviceContext`] of a bus specific device when it is guaranteed to
+/// be bound to a driver.
+///
+/// The bound context indicates that for the entire duration of the lifetime of a [`Device<Bound>`]
+/// reference, the [`Device`] is guaranteed to be bound to a driver.
+///
+/// Some APIs, such as [`dma::CoherentAllocation`](kernel::dma::CoherentAllocation) or
+/// [`Devres`](kernel::devres::Devres) rely on the [`Device`] to be bound, which can be proven with
+/// the [`Bound`] device context.
+///
+/// Any abstraction that can guarantee a scope where the corresponding bus device is bound, should
+/// provide a [`Device<Bound>`] reference to its users for this scope. This allows users to benefit
+/// from optimizations for accessing device resources, see also
+/// [Devres::access](kernel::devres::Devres::access).
 pub struct Bound;
 
 mod private {
-- 
2.50.0

[PATCH 2/3] device: rust: expand documentation for Device

[Danilo Krummrich]

The documentation for the generic Device type is outdated and deserves
much more detail.

Hence, expand the documentation and cover topics such as device types,
device contexts, as well as information on how to use the generic device
infrastructure to implement bus and class specific device types.

Signed-off-by: Danilo Krummrich <dakr@kernel.org>
---
 rust/kernel/device.rs | 135 ++++++++++++++++++++++++++++++++++++++----
 1 file changed, 122 insertions(+), 13 deletions(-)

diff --git a/rust/kernel/device.rs b/rust/kernel/device.rs
index d7ac56628fe5..1a2ae16d8f11 100644
--- a/rust/kernel/device.rs
+++ b/rust/kernel/device.rs
@@ -15,23 +15,132 @@
 
 pub mod property;
 
-/// A reference-counted device.
+/// The core representation of a device in the kernel's driver model.
 ///
-/// This structure represents the Rust abstraction for a C `struct device`. This implementation
-/// abstracts the usage of an already existing C `struct device` within Rust code that we get
-/// passed from the C side.
+/// This structure represents the Rust abstraction for a C `struct device`. A [`Device`] can either
+/// exist as temporary reference (see also [`Device::from_raw`]), which is only valid within a
+/// certain scope or as [`ARef<Device>`], owning a dedicated reference count.
 ///
-/// An instance of this abstraction can be obtained temporarily or permanent.
+/// # Device Types
 ///
-/// A temporary one is bound to the lifetime of the C `struct device` pointer used for creation.
-/// A permanent instance is always reference-counted and hence not restricted by any lifetime
-/// boundaries.
+/// A [`Device`] can represent either a bus device or a class device.
 ///
-/// For subsystems it is recommended to create a permanent instance to wrap into a subsystem
-/// specific device structure (e.g. `pci::Device`). This is useful for passing it to drivers in
-/// `T::probe()`, such that a driver can store the `ARef<Device>` (equivalent to storing a
-/// `struct device` pointer in a C driver) for arbitrary purposes, e.g. allocating DMA coherent
-/// memory.
+/// ## Bus Devices
+///
+/// A bus device is a [`Device`] that is associated with a physical or virtual bus. Examples of
+/// buses include PCI, USB, I2C, and SPI. Devices attached to a bus are registered with a specific
+/// bus type, which facilitates matching devices with appropriate drivers based on IDs or other
+/// identifying information. Bus devices are visible in sysfs under `/sys/bus/<bus-name>/devices/`.
+///
+/// ## Class Devices
+///
+/// A class device is a [`Device`] that is associated with a logical category of functionality
+/// rather than a physical bus. Examples of classes include block devices, network interfaces, sound
+/// cards, and input devices. Class devices are grouped under a common class and exposed to
+/// userspace via entries in `/sys/class/<class-name>/`.
+///
+/// # Device Context
+///
+/// [`Device`] references are generic over a [`DeviceContext`], which represents the type state of
+/// a [`Device`].
+///
+/// As the name indicates, this type state represents the context of the scope the [`Device`]
+/// reference is valid in. For instance, the [`Bound`] context guarantees that the [`Device`] is
+/// bound to a driver for the entire duration of the existence of a [`Device<Bound>`] reference.
+///
+/// Other [`DeviceContext`] types besides [`Bound`] are [`Normal`], [`Core`] and [`CoreInternal`].
+///
+/// Unless selected otherwise [`Device`] defaults to the [`Normal`] [`DeviceContext`], which by
+/// itself has no additional requirements.
+///
+/// It is always up to the caller of [`Device::from_raw`] to select the correct [`DeviceContext`]
+/// type for the corresponding scope the [`Device`] reference is created in.
+///
+/// All [`DeviceContext`] types other than [`Normal`] are intended to be used with
+/// [bus devices](#bus-devices) only.
+///
+/// # Implementing Bus Devices
+///
+/// This section provides a guideline to implement bus specific devices, such as
+/// [`pci::Device`](kernel::pci::Device) or [`platform::Device`](kernel::platform::Device).
+///
+/// A bus specific device should be defined as follows.
+///
+/// ```ignore
+/// #[repr(transparent)]
+/// pub struct Device<Ctx: device::DeviceContext = device::Normal>(
+///     Opaque<bindings::bus_device_type>,
+///     PhantomData<Ctx>,
+/// );
+/// ```
+///
+/// Since devices are reference counted, [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted)
+/// should be implemented for `Device` (i.e. `Device<Normal>`). Note that
+/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) must not be implemented for any other
+/// [`DeviceContext`], since all other device context types are only valid in a certain scope.
+///
+/// In order to be able to implement the [`DeviceContext`] dereference hierarchy, bus device
+/// implementations should call the [`impl_device_context_deref`](kernel::impl_device_context_deref)
+/// macro as shown below.
+///
+/// ```ignore
+/// // SAFETY: `Device` is a transparent wrapper of a type that doesn't depend on `Device`'s
+/// // generic argument.
+/// kernel::impl_device_context_deref!(unsafe { Device });
+/// ```
+/// In order to convert from a any [`Device<Ctx>`] to [`ARef<Device>`], bus devices can implement
+/// the following macro call.
+///
+/// ```ignore
+/// kernel::impl_device_context_into_aref!(Device);
+/// ```
+/// Bus devices should also implement the following [`AsRef`] implementation, such that users can
+/// easily derive a generic [`Device`] reference.
+///
+/// ```ignore
+/// impl<Ctx: device::DeviceContext> AsRef<device::Device<Ctx>> for Device<Ctx> {
+///     fn as_ref(&self) -> &device::Device<Ctx> {
+///         ...
+///     }
+/// }
+/// ```
+///
+/// # Implementing Class Devices
+///
+/// Class device implementations require less infrastructure and depend slightly more on the
+/// specific subsystem.
+///
+/// An example implementation for a class device could look like this.
+///
+/// ```ignore
+/// #[repr(C)]
+/// #[pin_data]
+/// pub struct Device<T: class::Driver> {
+///     dev: Opaque<bindings::class_device_type>,
+///     #[pin]
+///     data: T::Data,
+/// }
+/// ```
+///
+/// This class device uses the sub-classing pattern to embed the driver's private data within the
+/// allocation of the class device. For this to be possible the class device is generic over the
+/// class specific `Driver` trait implementation.
+///
+/// Just like any device, class devices are reference counted and should hence implement
+/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) for `Device`.
+///
+/// Class devices should also implement the following [`AsRef`] implementation, such that users can
+/// easily derive a generic [`Device`] reference.
+///
+/// ```ignore
+/// impl<T: class::Driver> AsRef<device::Device> for Device<T> {
+///     fn as_ref(&self) -> &device::Device {
+///         ...
+///     }
+/// }
+/// ```
+///
+/// An example for a class device implementation is [`drm::Device`](kernel::drm::Device).
 ///
 /// # Invariants
 ///
-- 
2.50.0

[PATCH 3/3] driver: rust: expand documentation for driver infrastructure

[Danilo Krummrich]

Add documentation about generic driver infrastructure, representing a
guideline on how the generic driver infrastructure is intended to be
used to implement bus specific driver APIs.

This covers aspects such as the bus specific driver trait, adapter
implementation, driver registration and custom device ID types.

Signed-off-by: Danilo Krummrich <dakr@kernel.org>
---
 rust/kernel/driver.rs | 82 +++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 80 insertions(+), 2 deletions(-)

diff --git a/rust/kernel/driver.rs b/rust/kernel/driver.rs
index a8f2675ba7a7..0bb80ad7aea4 100644
--- a/rust/kernel/driver.rs
+++ b/rust/kernel/driver.rs
@@ -2,8 +2,86 @@
 
 //! Generic support for drivers of different buses (e.g., PCI, Platform, Amba, etc.).
 //!
-//! Each bus / subsystem is expected to implement [`RegistrationOps`], which allows drivers to
-//! register using the [`Registration`] class.
+//! This documentation describes how to implement a bus specific driver API and how to align it with
+//! the design of (bus specific) devices.
+//!
+//! Note: Readers are expected to know the content of the documentation of
+//! [`Device`](device::Device) and [`device::DeviceContext`].
+//!
+//! # Driver Trait
+//!
+//! The main driver interface is defined by a bus specific driver trait. For instance:
+//!
+//! ```ignore
+//! pub trait Driver: Send {
+//!     /// The type holding information about each device ID supported by the driver.
+//!     type IdInfo: 'static;
+//!
+//!     /// The table of OF device ids supported by the driver.
+//!     const OF_ID_TABLE: Option<of::IdTable<Self::IdInfo>> = None;
+//!
+//!     /// The table of ACPI device ids supported by the driver.
+//!     const ACPI_ID_TABLE: Option<acpi::IdTable<Self::IdInfo>> = None;
+//!
+//!     /// Driver probe.
+//!     fn probe(dev: &Device<device::Core>, id_info: &Self::IdInfo) -> Result<Pin<KBox<Self>>>;
+//!
+//!     /// Driver unbind (optional).
+//!     fn unbind(dev: &Device<device::Core>, this: Pin<&Self>) {
+//!         let _ = (dev, this);
+//!     }
+//! }
+//! ```
+//!
+//! For specific examples see [`pci::Driver`](kernel::pci::Driver),
+//! [`platform::Driver`](kernel::platform::Driver) and
+//! [`auxiliary::Driver`](kernel::auxiliary::Driver).
+//!
+//! The `probe()` callback should return a `Result<Pin<KBox<Self>>>`, i.e. the driver's private
+//! data. The bus abstraction should store the pointer in the corresponding bus device. The generic
+//! [`Device`](device::Device) infrastructure provides common helpers for this purpose on its
+//! [`Device<CoreInternal>`](device::Device<device::CoreInternal>) implementation.
+//!
+//! All driver callbacks should provide a reference to the driver's private data. Once the driver
+//! is unbound from the device, the bus abstraction should take back the ownership of the driver's
+//! private data from the corresponding [`Device`](device::Device) and [`drop`] it.
+//!
+//! All driver callbacks should provide a [`Device<Core>`](device::Device<device::Core>) reference
+//! (see also [`device::Core`]).
+//!
+//! # Adapter
+//!
+//! The adapter implementation of a bus represents the abstraction layer between the C bus
+//! callbacks and the Rust bus callbacks. It therefore has to be generic over an implementation of
+//! the [driver trait](#driver-trait).
+//!
+//! ```ignore
+//! pub struct Adapter<T: Driver>;
+//! ```
+//!
+//! There's a common [`Adapter`] trait that can be implemented to inherit common driver
+//! infrastructure, such as finding the ID info from an [`of::IdTable`] or [`acpi::IdTable`].
+//!
+//! # Driver Registration
+//!
+//! In order to register C driver types (such as `struct platform_driver`) the [adapter](#adapter)
+//! should implement the [`RegistrationOps`] trait.
+//!
+//! This trait implementation can be used to create the actual registration with the common
+//! [`Registration`] type.
+//!
+//! Typically, bus abstractions want to provide a bus specific `module_bus_driver!` macro, which
+//! creates a kernel module with exactly one [`Registration`] for the bus specific adapter.
+//!
+//! The generic driver infrastructure provides a helper for this with the
+//! [`module_driver`](kernel::module_driver) macro.
+//!
+//! # Device IDs
+//!
+//! Besides the common device ID types, such as [`of::DeviceId`] and [`acpi::DeviceId`], most buses
+//! may need to implement their own device ID types.
+//!
+//! For this purpose the generic infrastructure in [`device_id`](kernel::device_id) should be used.
 
 use crate::error::{Error, Result};
 use crate::{acpi, device, of, str::CStr, try_pin_init, types::Opaque, ThisModule};
-- 
2.50.0

Re: [PATCH 1/3] device: rust: documentation for DeviceContext

[Alexandre Courbot]

On Fri Jul 18, 2025 at 7:45 AM JST, Danilo Krummrich wrote:
> Expand the documentation around DeviceContext states and types, in order
> to provide detailed information about their purpose and relationship
> with each other.
>
> Signed-off-by: Danilo Krummrich <dakr@kernel.org>

Thanks, that's a welcome clarification and I think I finally understand
the Rust device model after going through this series!

A few minor nits/questions below.

> ---
>  rust/kernel/device.rs | 63 +++++++++++++++++++++++++++++++++++--------
>  1 file changed, 52 insertions(+), 11 deletions(-)
>
> diff --git a/rust/kernel/device.rs b/rust/kernel/device.rs
> index ca82926fd67f..d7ac56628fe5 100644
> --- a/rust/kernel/device.rs
> +++ b/rust/kernel/device.rs
> @@ -311,28 +311,69 @@ unsafe impl Send for Device {}
>  // synchronization in `struct device`.
>  unsafe impl Sync for Device {}
>  
> -/// Marker trait for the context of a bus specific device.
> +/// Marker trait for the context or scope of a bus specific device.
>  ///
> -/// Some functions of a bus specific device should only be called from a certain context, i.e. bus
> -/// callbacks, such as `probe()`.
> +/// [`DeviceContext`] is a marker trait for structures representing the context of a bus specific
> +/// [`Device`].
>  ///
> -/// This is the marker trait for structures representing the context of a bus specific device.

Shall we say `types` instead of `structures`, since these are ZSTs?
`structures` carries the hint that they will contain data, when they
don't (but maybe that's only me :)).

> +/// The specific device context types are: [`CoreInternal`], [`Core`], [`Bound`] and [`Normal`].
> +///
> +/// [`DeviceContext`] types are hierarchical, which means that there is a strict hierarchy that
> +/// defines which [`DeviceContext`] type can be derived from another. For instance, any
> +/// [`Device<Core>`] can dereference to a [`Device<Bound>`].
> +///
> +/// The following enunumeration illustrates the dereference hierarchy of [`DeviceContext`] types.
> +///
> +/// - [`CoreInternal`] => [`Core`] => [`Bound`] => [`Normal`]
> +/// - [`Core`] => [`Bound`] => [`Normal`]
> +/// - [`Bound`] => [`Normal`]
> +/// - [`Normal`]

That graph is super helpful. The last 3 lines look redundant though,
since the graph can be followed from any node.

> +///
> +/// Bus devices can automatically implement the dereference hierarchy by using
> +/// [`impl_device_context_deref`](kernel::impl_device_context_deref).
>  pub trait DeviceContext: private::Sealed {}
>  
> -/// The [`Normal`] context is the context of a bus specific device when it is not an argument of
> -/// any bus callback.
> +/// The [`Normal`] context is the default [`DeviceContext`] of any [`Device`].
> +///
> +/// The normal context does not indicate any specific scope. Any `Device<Ctx>` is also a valid
> +/// [`Device<Normal>`]. It is the only [`DeviceContext`] for which it is valid to implement
> +/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) for.
>  pub struct Normal;

`Normal` as a name can be interpreted in many different ways, and in the
case of a device context it is not clear what the "normal" state is. I
think it would be helpful if we can elaborate a bit more on what this
implies (i.e. what concretely speaking are the limitations), and if
possible why this name has been chosen.

Re: [PATCH 1/3] device: rust: documentation for DeviceContext

[Daniel Almeida]

Hi Danilo,

> On 17 Jul 2025, at 19:45, Danilo Krummrich <dakr@kernel.org> wrote:
> 
> Expand the documentation around DeviceContext states and types, in order
> to provide detailed information about their purpose and relationship
> with each other.
> 
> Signed-off-by: Danilo Krummrich <dakr@kernel.org>
> ---
> rust/kernel/device.rs | 63 +++++++++++++++++++++++++++++++++++--------
> 1 file changed, 52 insertions(+), 11 deletions(-)
> 
> diff --git a/rust/kernel/device.rs b/rust/kernel/device.rs
> index ca82926fd67f..d7ac56628fe5 100644
> --- a/rust/kernel/device.rs
> +++ b/rust/kernel/device.rs
> @@ -311,28 +311,69 @@ unsafe impl Send for Device {}
> // synchronization in `struct device`.
> unsafe impl Sync for Device {}
> 
> -/// Marker trait for the context of a bus specific device.
> +/// Marker trait for the context or scope of a bus specific device.
> ///
> -/// Some functions of a bus specific device should only be called from a certain context, i.e. bus
> -/// callbacks, such as `probe()`.
> +/// [`DeviceContext`] is a marker trait for structures representing the context of a bus specific
> +/// [`Device`].
> ///
> -/// This is the marker trait for structures representing the context of a bus specific device.
> +/// The specific device context types are: [`CoreInternal`], [`Core`], [`Bound`] and [`Normal`].
> +///
> +/// [`DeviceContext`] types are hierarchical, which means that there is a strict hierarchy that
> +/// defines which [`DeviceContext`] type can be derived from another. For instance, any
> +/// [`Device<Core>`] can dereference to a [`Device<Bound>`].
> +///
> +/// The following enunumeration illustrates the dereference hierarchy of [`DeviceContext`] types.
> +///
> +/// - [`CoreInternal`] => [`Core`] => [`Bound`] => [`Normal`]
> +/// - [`Core`] => [`Bound`] => [`Normal`]
> +/// - [`Bound`] => [`Normal`]
> +/// - [`Normal`]
> +///
> +/// Bus devices can automatically implement the dereference hierarchy by using
> +/// [`impl_device_context_deref`](kernel::impl_device_context_deref).
> pub trait DeviceContext: private::Sealed {}

Overall this looks good to me. I think that one point you could perhaps
consider is that, to me at least, it wasn't clear that the contexts were only
valid for a given scope. Or what was precisely meant by “scope”.


I.e.: I thought that once you saw Device<Bound>, for example, that would be
valid indefinitely. If we retrieve one of our past conversations at [0]:

> 
> > Fine, but can’t you get a &Device<Bound> from a ARef<drm::Device>, for example?
> > Perhaps a nicer solution would be to offer this capability instead?
> 
> I think you're confusing quite some things here.
> 
>   [...]
> 
>   (2) Owning a reference count of a device (i.e. ARef<Device>) does *not*
>       guarantee that the device is bound. You can own a reference count to the
>       device object way beyond it being bound. Instead, the guarantee comes from
>       the scope.
> 
>       In this case, the scope is the IRQ callback, since the irq::Registration
>       guarantees to call and complete free_irq() before the underlying bus
>       device is unbound.


I see that you mention the word "scope" a few times, but perhaps it would be
more instructional if you say a few more things on this topic.

For example, when you mention probe(), it would be useful to emphasize that the
Core state would only be guaranteed for the _scope of that function_, and that
it wouldn't mean that "the state Core is active from now on", or "I can assume
that we have a Device<Core> from now on in other parts of the driver".

Kind of like you do here:

> +/// The core context indicates that the [`Device<Core>`] reference's scope is limited to the bus
> +/// callback it appears in.

But generalizing to all states if possible.

The difference is very subtle so this can sound a bit confusing. Let me know if
you want me to clarify this further.


[0] https://lore.kernel.org/rust-for-linux/DBB0NXU86D6G.2M3WZMS2NUV10@kernel.org/

— Daniel

Re: [PATCH 1/3] device: rust: documentation for DeviceContext

[Danilo Krummrich]

On Fri Jul 18, 2025 at 2:32 PM CEST, Alexandre Courbot wrote:
> On Fri Jul 18, 2025 at 7:45 AM JST, Danilo Krummrich wrote:
>> Expand the documentation around DeviceContext states and types, in order
>> to provide detailed information about their purpose and relationship
>> with each other.
>>
>> Signed-off-by: Danilo Krummrich <dakr@kernel.org>
>
> Thanks, that's a welcome clarification and I think I finally understand
> the Rust device model after going through this series!
>
> A few minor nits/questions below.
>
>> ---
>>  rust/kernel/device.rs | 63 +++++++++++++++++++++++++++++++++++--------
>>  1 file changed, 52 insertions(+), 11 deletions(-)
>>
>> diff --git a/rust/kernel/device.rs b/rust/kernel/device.rs
>> index ca82926fd67f..d7ac56628fe5 100644
>> --- a/rust/kernel/device.rs
>> +++ b/rust/kernel/device.rs
>> @@ -311,28 +311,69 @@ unsafe impl Send for Device {}
>>  // synchronization in `struct device`.
>>  unsafe impl Sync for Device {}
>>  
>> -/// Marker trait for the context of a bus specific device.
>> +/// Marker trait for the context or scope of a bus specific device.
>>  ///
>> -/// Some functions of a bus specific device should only be called from a certain context, i.e. bus
>> -/// callbacks, such as `probe()`.
>> +/// [`DeviceContext`] is a marker trait for structures representing the context of a bus specific
>> +/// [`Device`].
>>  ///
>> -/// This is the marker trait for structures representing the context of a bus specific device.
>
> Shall we say `types` instead of `structures`, since these are ZSTs?
> `structures` carries the hint that they will contain data, when they
> don't (but maybe that's only me :)).

I agree, 'types' is better.

>> +/// The specific device context types are: [`CoreInternal`], [`Core`], [`Bound`] and [`Normal`].
>> +///
>> +/// [`DeviceContext`] types are hierarchical, which means that there is a strict hierarchy that
>> +/// defines which [`DeviceContext`] type can be derived from another. For instance, any
>> +/// [`Device<Core>`] can dereference to a [`Device<Bound>`].
>> +///
>> +/// The following enunumeration illustrates the dereference hierarchy of [`DeviceContext`] types.
>> +///
>> +/// - [`CoreInternal`] => [`Core`] => [`Bound`] => [`Normal`]
>> +/// - [`Core`] => [`Bound`] => [`Normal`]
>> +/// - [`Bound`] => [`Normal`]
>> +/// - [`Normal`]
>
> That graph is super helpful. The last 3 lines look redundant though,
> since the graph can be followed from any node.

Yeah, it's indeed unnecessarily redundant.

>> +///
>> +/// Bus devices can automatically implement the dereference hierarchy by using
>> +/// [`impl_device_context_deref`](kernel::impl_device_context_deref).
>>  pub trait DeviceContext: private::Sealed {}
>>  
>> -/// The [`Normal`] context is the context of a bus specific device when it is not an argument of
>> -/// any bus callback.
>> +/// The [`Normal`] context is the default [`DeviceContext`] of any [`Device`].
>> +///
>> +/// The normal context does not indicate any specific scope. Any `Device<Ctx>` is also a valid
>> +/// [`Device<Normal>`]. It is the only [`DeviceContext`] for which it is valid to implement
>> +/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) for.
>>  pub struct Normal;
>
> `Normal` as a name can be interpreted in many different ways, and in the
> case of a device context it is not clear what the "normal" state is. I
> think it would be helpful if we can elaborate a bit more on what this
> implies (i.e. what concretely speaking are the limitations), and if
> possible why this name has been chosen.

It's the context that does not guarantee any specific scope. But that's also
what the documentation says.

I also wouldn't speak of limitations, it's just that it doesn't allow to make
*additional* assumptions compared to other device context types.

Yet, if you have suggestions on what to add specifically, please let me know
(maybe simply my previous sentence?).

Regarding the name, "Normal" seems reasonable for the device context that does
not guarantee any specific scope. We could have also named it just "Default".

I think "Normal" is fine, as in "it's just a normal device reference, no
specific scope guaranteed".

Re: [PATCH 1/3] device: rust: documentation for DeviceContext

[Danilo Krummrich]

On Fri Jul 18, 2025 at 3:09 PM CEST, Daniel Almeida wrote:
> Hi Danilo,
>
>> On 17 Jul 2025, at 19:45, Danilo Krummrich <dakr@kernel.org> wrote:
>> 
>> Expand the documentation around DeviceContext states and types, in order
>> to provide detailed information about their purpose and relationship
>> with each other.
>> 
>> Signed-off-by: Danilo Krummrich <dakr@kernel.org>
>> ---
>> rust/kernel/device.rs | 63 +++++++++++++++++++++++++++++++++++--------
>> 1 file changed, 52 insertions(+), 11 deletions(-)
>> 
>> diff --git a/rust/kernel/device.rs b/rust/kernel/device.rs
>> index ca82926fd67f..d7ac56628fe5 100644
>> --- a/rust/kernel/device.rs
>> +++ b/rust/kernel/device.rs
>> @@ -311,28 +311,69 @@ unsafe impl Send for Device {}
>> // synchronization in `struct device`.
>> unsafe impl Sync for Device {}
>> 
>> -/// Marker trait for the context of a bus specific device.
>> +/// Marker trait for the context or scope of a bus specific device.
>> ///
>> -/// Some functions of a bus specific device should only be called from a certain context, i.e. bus
>> -/// callbacks, such as `probe()`.
>> +/// [`DeviceContext`] is a marker trait for structures representing the context of a bus specific
>> +/// [`Device`].
>> ///
>> -/// This is the marker trait for structures representing the context of a bus specific device.
>> +/// The specific device context types are: [`CoreInternal`], [`Core`], [`Bound`] and [`Normal`].
>> +///
>> +/// [`DeviceContext`] types are hierarchical, which means that there is a strict hierarchy that
>> +/// defines which [`DeviceContext`] type can be derived from another. For instance, any
>> +/// [`Device<Core>`] can dereference to a [`Device<Bound>`].
>> +///
>> +/// The following enunumeration illustrates the dereference hierarchy of [`DeviceContext`] types.
>> +///
>> +/// - [`CoreInternal`] => [`Core`] => [`Bound`] => [`Normal`]
>> +/// - [`Core`] => [`Bound`] => [`Normal`]
>> +/// - [`Bound`] => [`Normal`]
>> +/// - [`Normal`]
>> +///
>> +/// Bus devices can automatically implement the dereference hierarchy by using
>> +/// [`impl_device_context_deref`](kernel::impl_device_context_deref).
>> pub trait DeviceContext: private::Sealed {}
>
> Overall this looks good to me. I think that one point you could perhaps
> consider is that, to me at least, it wasn't clear that the contexts were only
> valid for a given scope. Or what was precisely meant by “scope”.

Scope really means scope in the sense of programming languages, which is why I
didn't define it more specifically.

So, a reference to a Device<Bound> (i.e. &Device<Bound>) indicates that the
device is guaranteed to be bound for the scope the reference is valid in.

Please also note that the added documentation on Device already says:

"This structure represents the Rust abstraction for a C `struct device`. A
[`Device`] can either  exist as temporary reference (see also
[`Device::from_raw`]), which is only valid within a certain scope or as
[`ARef<Device>`], owning a dedicated reference count."

I think this should clarify it already, or are you looking for something else?

> I.e.: I thought that once you saw Device<Bound>, for example, that would be
> valid indefinitely. If we retrieve one of our past conversations at [0]:

You can't create or have a Device<Bound>, but a &Device<Bound>, which has a
defined lifetime that can't be extended arbitrarily with safe code.

>> 
>> > Fine, but can’t you get a &Device<Bound> from a ARef<drm::Device>, for example?
>> > Perhaps a nicer solution would be to offer this capability instead?
>> 
>> I think you're confusing quite some things here.
>> 
>>   [...]
>> 
>>   (2) Owning a reference count of a device (i.e. ARef<Device>) does *not*
>>       guarantee that the device is bound. You can own a reference count to the
>>       device object way beyond it being bound. Instead, the guarantee comes from
>>       the scope.
>> 
>>       In this case, the scope is the IRQ callback, since the irq::Registration
>>       guarantees to call and complete free_irq() before the underlying bus
>>       device is unbound.
>
>
> I see that you mention the word "scope" a few times, but perhaps it would be
> more instructional if you say a few more things on this topic.
>
> For example, when you mention probe(), it would be useful to emphasize that the
> Core state would only be guaranteed for the _scope of that function_, and that
> it wouldn't mean that "the state Core is active from now on", or "I can assume
> that we have a Device<Core> from now on in other parts of the driver".
>
> Kind of like you do here:
>
>> +/// The core context indicates that the [`Device<Core>`] reference's scope is limited to the bus
>> +/// callback it appears in.
>
> But generalizing to all states if possible.

That's not possible. Core is specific as in that it's really meant to be the
context of a device when it appears in a bus callback.

But the Bound context may appear whereever we it can be proven that within a
certain scope (e.g. the IRQ callback) the device is guaranteed to be bound.

So the generalization really is to say "scope".

> The difference is very subtle so this can sound a bit confusing. Let me know if
> you want me to clarify this further.
>
>
> [0] https://lore.kernel.org/rust-for-linux/DBB0NXU86D6G.2M3WZMS2NUV10@kernel.org/
>
> — Daniel

Re: [PATCH 0/3] Documentation for Device / Driver infrastructure

[Greg KH]

On Fri, Jul 18, 2025 at 12:45:36AM +0200, Danilo Krummrich wrote:
> This patch series expands the documentation for the generic device and driver
> infrastructure and provides a guideline on how to implement bus and class
> specific devices as well as bus specific driver infrastructure, i.e. bus
> abstractions. It also expands on the existing documentation of device context
> types.

Looks great to me, thanks for writing this all up:

Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Re: [PATCH 1/3] device: rust: documentation for DeviceContext

[Daniel Almeida]

Hi Danilo,


[…]

>>> unsafe impl Sync for Device {}
>>> 
>>> -/// Marker trait for the context of a bus specific device.
>>> +/// Marker trait for the context or scope of a bus specific device.
>>> ///
>>> -/// Some functions of a bus specific device should only be called from a certain context, i.e. bus
>>> -/// callbacks, such as `probe()`.
>>> +/// [`DeviceContext`] is a marker trait for structures representing the context of a bus specific
>>> +/// [`Device`].
>>> ///
>>> -/// This is the marker trait for structures representing the context of a bus specific device.
>>> +/// The specific device context types are: [`CoreInternal`], [`Core`], [`Bound`] and [`Normal`].
>>> +///
>>> +/// [`DeviceContext`] types are hierarchical, which means that there is a strict hierarchy that
>>> +/// defines which [`DeviceContext`] type can be derived from another. For instance, any
>>> +/// [`Device<Core>`] can dereference to a [`Device<Bound>`].
>>> +///
>>> +/// The following enunumeration illustrates the dereference hierarchy of [`DeviceContext`] types.
>>> +///
>>> +/// - [`CoreInternal`] => [`Core`] => [`Bound`] => [`Normal`]
>>> +/// - [`Core`] => [`Bound`] => [`Normal`]
>>> +/// - [`Bound`] => [`Normal`]
>>> +/// - [`Normal`]
>>> +///
>>> +/// Bus devices can automatically implement the dereference hierarchy by using
>>> +/// [`impl_device_context_deref`](kernel::impl_device_context_deref).
>>> pub trait DeviceContext: private::Sealed {}
>> 
>> Overall this looks good to me. I think that one point you could perhaps
>> consider is that, to me at least, it wasn't clear that the contexts were only
>> valid for a given scope. Or what was precisely meant by “scope”.
> 
> Scope really means scope in the sense of programming languages, which is why I
> didn't define it more specifically.
> 
> So, a reference to a Device<Bound> (i.e. &Device<Bound>) indicates that the
> device is guaranteed to be bound for the scope the reference is valid in.


Maybe this is the piece of information that I missed, i.e.: that scope was
being used in the usual sense in a programming language, and that the term
wasn't being borrowed to describe some other sort of device behavior. This
might look obvious now, but wasn't when I saw it the first time.

Anyway, this is starting to look like I was the only one to get confused, so
scratch what I said :)

Also, looking at this patch again, it looks like you already explain this well
enough, for example in the docs for Device<Core>, where "reference's scope" is
mentioned:

/// The core context indicates that the [`Device<Core>`] reference's scope is limited to the bus
/// callback it appears in. 

[…]

> 
>>> 
>>>> Fine, but can’t you get a &Device<Bound> from a ARef<drm::Device>, for example?
>>>> Perhaps a nicer solution would be to offer this capability instead?
>>> 
>>> I think you're confusing quite some things here.
>>> 
>>>  [...]
>>> 
>>>  (2) Owning a reference count of a device (i.e. ARef<Device>) does *not*
>>>      guarantee that the device is bound. You can own a reference count to the
>>>      device object way beyond it being bound. Instead, the guarantee comes from
>>>      the scope.
>>> 

I wonder if it would be helpful to specifically state this in the docs somehow?
Perhaps as an extra "Note that ...". It doesn't hurt to repeat ourselves a bit
here, IMHO.

— Daniel

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Daniel Almeida]

This looks good to me!

> On 17 Jul 2025, at 19:45, Danilo Krummrich <dakr@kernel.org> wrote:
> 
> The documentation for the generic Device type is outdated and deserves
> much more detail.
> 
> Hence, expand the documentation and cover topics such as device types,
> device contexts, as well as information on how to use the generic device
> infrastructure to implement bus and class specific device types.
> 
> Signed-off-by: Danilo Krummrich <dakr@kernel.org>


Reviewed-by: Daniel Almeida <daniel.almeida@collabora.com>

Re: [PATCH 3/3] driver: rust: expand documentation for driver infrastructure

[Daniel Almeida]

This also looks good to me.

> On 17 Jul 2025, at 19:45, Danilo Krummrich <dakr@kernel.org> wrote:
> 
> Add documentation about generic driver infrastructure, representing a
> guideline on how the generic driver infrastructure is intended to be
> used to implement bus specific driver APIs.
> 
> This covers aspects such as the bus specific driver trait, adapter
> implementation, driver registration and custom device ID types.
> 
> Signed-off-by: Danilo Krummrich <dakr@kernel.org>

Reviewed-by: Daniel Almeida <daniel.almeida@collabora.com>

— Daniel

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Alice Ryhl]

On Fri, Jul 18, 2025 at 12:45:38AM +0200, Danilo Krummrich wrote:
> The documentation for the generic Device type is outdated and deserves
> much more detail.
> 
> Hence, expand the documentation and cover topics such as device types,
> device contexts, as well as information on how to use the generic device
> infrastructure to implement bus and class specific device types.
> 
> Signed-off-by: Danilo Krummrich <dakr@kernel.org>

Overall I think this series is pretty great. It also clarifies some
things for me, particularly the difference between bus and class
devices.

> +/// # Device Types
>  ///
> +/// A [`Device`] can represent either a bus device or a class device.
>  ///
> +/// ## Bus Devices
> +///
> +/// A bus device is a [`Device`] that is associated with a physical or virtual bus. Examples of
> +/// buses include PCI, USB, I2C, and SPI. Devices attached to a bus are registered with a specific
> +/// bus type, which facilitates matching devices with appropriate drivers based on IDs or other
> +/// identifying information. Bus devices are visible in sysfs under `/sys/bus/<bus-name>/devices/`.
> +///
> +/// ## Class Devices
> +///
> +/// A class device is a [`Device`] that is associated with a logical category of functionality
> +/// rather than a physical bus. Examples of classes include block devices, network interfaces, sound
> +/// cards, and input devices. Class devices are grouped under a common class and exposed to
> +/// userspace via entries in `/sys/class/<class-name>/`.
> +///
> +/// # Device Context
> +///
> +/// [`Device`] references are generic over a [`DeviceContext`], which represents the type state of
> +/// a [`Device`].
> +///
> +/// As the name indicates, this type state represents the context of the scope the [`Device`]
> +/// reference is valid in. For instance, the [`Bound`] context guarantees that the [`Device`] is
> +/// bound to a driver for the entire duration of the existence of a [`Device<Bound>`] reference.
> +///
> +/// Other [`DeviceContext`] types besides [`Bound`] are [`Normal`], [`Core`] and [`CoreInternal`].
> +///
> +/// Unless selected otherwise [`Device`] defaults to the [`Normal`] [`DeviceContext`], which by
> +/// itself has no additional requirements.
> +///
> +/// It is always up to the caller of [`Device::from_raw`] to select the correct [`DeviceContext`]
> +/// type for the corresponding scope the [`Device`] reference is created in.
> +///
> +/// All [`DeviceContext`] types other than [`Normal`] are intended to be used with
> +/// [bus devices](#bus-devices) only.

This raises a few questions for me.

The first one is "why"? On other series I have been told that interrupts
must be registered and deregistered before the device is unbound. Does
the same not apply to interrupts for an input device such as a USB
keyboard?

The second one is why we use the same `Device` type for both cases?
Would it not make more sense to have a BusDevice and ClassDevice type?

> +/// # Implementing Bus Devices
> +///
> +/// This section provides a guideline to implement bus specific devices, such as
> +/// [`pci::Device`](kernel::pci::Device) or [`platform::Device`](kernel::platform::Device).
> +///
> +/// A bus specific device should be defined as follows.
> +///
> +/// ```ignore
> +/// #[repr(transparent)]
> +/// pub struct Device<Ctx: device::DeviceContext = device::Normal>(
> +///     Opaque<bindings::bus_device_type>,
> +///     PhantomData<Ctx>,
> +/// );
> +/// ```
> +///
> +/// Since devices are reference counted, [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted)
> +/// should be implemented for `Device` (i.e. `Device<Normal>`). Note that
> +/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) must not be implemented for any other
> +/// [`DeviceContext`], since all other device context types are only valid in a certain scope.

As a general comment to all three patches, I would suggest separating
out the link locations.

/// Since devices are reference counted, [`AlwaysRefCounted`] should be
/// implemented for `Device` (i.e. `Device<Normal>`). Note that
/// [`AlwaysRefCounted`] must not be implemented for any other
/// [`DeviceContext`], since all other device context types are only
/// valid in a certain scope.

and then at the end:

/// [`AlwaysRefCounted`]: kernel::types::AlwaysRefCounted

I think it's a lot easier to read the markdown version when links are
separated out like this.

Alice

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Greg KH]

On Mon, Jul 21, 2025 at 11:26:23AM +0000, Alice Ryhl wrote:
> On Fri, Jul 18, 2025 at 12:45:38AM +0200, Danilo Krummrich wrote:
> > The documentation for the generic Device type is outdated and deserves
> > much more detail.
> > 
> > Hence, expand the documentation and cover topics such as device types,
> > device contexts, as well as information on how to use the generic device
> > infrastructure to implement bus and class specific device types.
> > 
> > Signed-off-by: Danilo Krummrich <dakr@kernel.org>
> 
> Overall I think this series is pretty great. It also clarifies some
> things for me, particularly the difference between bus and class
> devices.
> 
> > +/// # Device Types
> >  ///
> > +/// A [`Device`] can represent either a bus device or a class device.
> >  ///
> > +/// ## Bus Devices
> > +///
> > +/// A bus device is a [`Device`] that is associated with a physical or virtual bus. Examples of
> > +/// buses include PCI, USB, I2C, and SPI. Devices attached to a bus are registered with a specific
> > +/// bus type, which facilitates matching devices with appropriate drivers based on IDs or other
> > +/// identifying information. Bus devices are visible in sysfs under `/sys/bus/<bus-name>/devices/`.
> > +///
> > +/// ## Class Devices
> > +///
> > +/// A class device is a [`Device`] that is associated with a logical category of functionality
> > +/// rather than a physical bus. Examples of classes include block devices, network interfaces, sound
> > +/// cards, and input devices. Class devices are grouped under a common class and exposed to
> > +/// userspace via entries in `/sys/class/<class-name>/`.
> > +///
> > +/// # Device Context
> > +///
> > +/// [`Device`] references are generic over a [`DeviceContext`], which represents the type state of
> > +/// a [`Device`].
> > +///
> > +/// As the name indicates, this type state represents the context of the scope the [`Device`]
> > +/// reference is valid in. For instance, the [`Bound`] context guarantees that the [`Device`] is
> > +/// bound to a driver for the entire duration of the existence of a [`Device<Bound>`] reference.
> > +///
> > +/// Other [`DeviceContext`] types besides [`Bound`] are [`Normal`], [`Core`] and [`CoreInternal`].
> > +///
> > +/// Unless selected otherwise [`Device`] defaults to the [`Normal`] [`DeviceContext`], which by
> > +/// itself has no additional requirements.
> > +///
> > +/// It is always up to the caller of [`Device::from_raw`] to select the correct [`DeviceContext`]
> > +/// type for the corresponding scope the [`Device`] reference is created in.
> > +///
> > +/// All [`DeviceContext`] types other than [`Normal`] are intended to be used with
> > +/// [bus devices](#bus-devices) only.
> 
> This raises a few questions for me.
> 
> The first one is "why"? On other series I have been told that interrupts
> must be registered and deregistered before the device is unbound. Does
> the same not apply to interrupts for an input device such as a USB
> keyboard?

USB drivers don't have interrupts :)

Seriously, generally, yes, you should always unregister your irq handler
BEFORE you unbind the driver from the device, otherwise the callback
logic normally assumes that the driver is bound to the device and the
pointers are almost always "stale" at that time.

But I don't understand what interrupts have to do with the above
documentation, what's the connection?

> The second one is why we use the same `Device` type for both cases?
> Would it not make more sense to have a BusDevice and ClassDevice type?

We used to have both types, years/decades ago, and in the end, just gave
up and merged them together as it really didn't make much sense as the
structure really was the same type of thing everywhere.  So let's not go
back to that mess if we can prevent it please.

We also used to have a out-of-tree patchset that just merged class and
bus together, as they are almost the same thing as well, but that just
caused too much confusion so we never merged it.

Hope this helps,

greg k-h

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Alice Ryhl]

On Mon, Jul 21, 2025 at 01:42:17PM +0200, Greg KH wrote:
> On Mon, Jul 21, 2025 at 11:26:23AM +0000, Alice Ryhl wrote:
> > On Fri, Jul 18, 2025 at 12:45:38AM +0200, Danilo Krummrich wrote:
> > > The documentation for the generic Device type is outdated and deserves
> > > much more detail.
> > > 
> > > Hence, expand the documentation and cover topics such as device types,
> > > device contexts, as well as information on how to use the generic device
> > > infrastructure to implement bus and class specific device types.
> > > 
> > > Signed-off-by: Danilo Krummrich <dakr@kernel.org>
> > 
> > Overall I think this series is pretty great. It also clarifies some
> > things for me, particularly the difference between bus and class
> > devices.
> > 
> > > +/// # Device Types
> > >  ///
> > > +/// A [`Device`] can represent either a bus device or a class device.
> > >  ///
> > > +/// ## Bus Devices
> > > +///
> > > +/// A bus device is a [`Device`] that is associated with a physical or virtual bus. Examples of
> > > +/// buses include PCI, USB, I2C, and SPI. Devices attached to a bus are registered with a specific
> > > +/// bus type, which facilitates matching devices with appropriate drivers based on IDs or other
> > > +/// identifying information. Bus devices are visible in sysfs under `/sys/bus/<bus-name>/devices/`.
> > > +///
> > > +/// ## Class Devices
> > > +///
> > > +/// A class device is a [`Device`] that is associated with a logical category of functionality
> > > +/// rather than a physical bus. Examples of classes include block devices, network interfaces, sound
> > > +/// cards, and input devices. Class devices are grouped under a common class and exposed to
> > > +/// userspace via entries in `/sys/class/<class-name>/`.
> > > +///
> > > +/// # Device Context
> > > +///
> > > +/// [`Device`] references are generic over a [`DeviceContext`], which represents the type state of
> > > +/// a [`Device`].
> > > +///
> > > +/// As the name indicates, this type state represents the context of the scope the [`Device`]
> > > +/// reference is valid in. For instance, the [`Bound`] context guarantees that the [`Device`] is
> > > +/// bound to a driver for the entire duration of the existence of a [`Device<Bound>`] reference.
> > > +///
> > > +/// Other [`DeviceContext`] types besides [`Bound`] are [`Normal`], [`Core`] and [`CoreInternal`].
> > > +///
> > > +/// Unless selected otherwise [`Device`] defaults to the [`Normal`] [`DeviceContext`], which by
> > > +/// itself has no additional requirements.
> > > +///
> > > +/// It is always up to the caller of [`Device::from_raw`] to select the correct [`DeviceContext`]
> > > +/// type for the corresponding scope the [`Device`] reference is created in.
> > > +///
> > > +/// All [`DeviceContext`] types other than [`Normal`] are intended to be used with
> > > +/// [bus devices](#bus-devices) only.
> > 
> > This raises a few questions for me.
> > 
> > The first one is "why"? On other series I have been told that interrupts
> > must be registered and deregistered before the device is unbound. Does
> > the same not apply to interrupts for an input device such as a USB
> > keyboard?
> 
> USB drivers don't have interrupts :)

I guess that if class drivers never have interrupts, then that is an
answer to my question.

> Seriously, generally, yes, you should always unregister your irq handler
> BEFORE you unbind the driver from the device, otherwise the callback
> logic normally assumes that the driver is bound to the device and the
> pointers are almost always "stale" at that time.
> 
> But I don't understand what interrupts have to do with the above
> documentation, what's the connection?

The connection is that to request an irq you must have a &Device<Bound>,
so if you can only obtain a &Device<Bound> to a bus device, then that
means that you can never request an irq for a class device.

> > The second one is why we use the same `Device` type for both cases?
> > Would it not make more sense to have a BusDevice and ClassDevice type?
> 
> We used to have both types, years/decades ago, and in the end, just gave
> up and merged them together as it really didn't make much sense as the
> structure really was the same type of thing everywhere.  So let's not go
> back to that mess if we can prevent it please.
> 
> We also used to have a out-of-tree patchset that just merged class and
> bus together, as they are almost the same thing as well, but that just
> caused too much confusion so we never merged it.

Okay, thanks!

Alice

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Danilo Krummrich]

On Mon Jul 21, 2025 at 1:26 PM CEST, Alice Ryhl wrote:
> On Fri, Jul 18, 2025 at 12:45:38AM +0200, Danilo Krummrich wrote:
>> +/// All [`DeviceContext`] types other than [`Normal`] are intended to be used with
>> +/// [bus devices](#bus-devices) only.
>
> This raises a few questions for me.
>
> The first one is "why"? On other series I have been told that interrupts
> must be registered and deregistered before the device is unbound. Does
> the same not apply to interrupts for an input device such as a USB
> keyboard?

In your example there would be a USB device *and* an input device, where the
former is a bus device and the latter a class device.

Any resources from the "real" device on the bus are on the USB device, not the
input device.

Or in other words, class devices do not own resources of a "real" device and
consequently are never bound to or unbound from a "real" device on the bus.

> The second one is why we use the same `Device` type for both cases?
> Would it not make more sense to have a BusDevice and ClassDevice type?

Not really, the generic struct device isn't one or the other until it's used by
an actual higher level bus or class device implementation.

There isn't really a difference between the two for a base device.

Regarding the device context, a base device can inherit a device context from
the higher level bus or class device. In case of a class device, it's just that
there's nothing to inherit other than Normal.

>> +/// # Implementing Bus Devices
>> +///
>> +/// This section provides a guideline to implement bus specific devices, such as
>> +/// [`pci::Device`](kernel::pci::Device) or [`platform::Device`](kernel::platform::Device).
>> +///
>> +/// A bus specific device should be defined as follows.
>> +///
>> +/// ```ignore
>> +/// #[repr(transparent)]
>> +/// pub struct Device<Ctx: device::DeviceContext = device::Normal>(
>> +///     Opaque<bindings::bus_device_type>,
>> +///     PhantomData<Ctx>,
>> +/// );
>> +/// ```
>> +///
>> +/// Since devices are reference counted, [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted)
>> +/// should be implemented for `Device` (i.e. `Device<Normal>`). Note that
>> +/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) must not be implemented for any other
>> +/// [`DeviceContext`], since all other device context types are only valid in a certain scope.
>
> As a general comment to all three patches, I would suggest separating
> out the link locations.
>
> /// Since devices are reference counted, [`AlwaysRefCounted`] should be
> /// implemented for `Device` (i.e. `Device<Normal>`). Note that
> /// [`AlwaysRefCounted`] must not be implemented for any other
> /// [`DeviceContext`], since all other device context types are only
> /// valid in a certain scope.
>
> and then at the end:
>
> /// [`AlwaysRefCounted`]: kernel::types::AlwaysRefCounted
>
> I think it's a lot easier to read the markdown version when links are
> separated out like this.

That's a good suggestion, thanks!

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Danilo Krummrich]

On Mon Jul 21, 2025 at 2:07 PM CEST, Alice Ryhl wrote:
> The connection is that to request an irq you must have a &Device<Bound>,
> so if you can only obtain a &Device<Bound> to a bus device, then that
> means that you can never request an irq for a class device.

As mentioned in my other reply, a class device never owns resources of a "real"
device such as an IRQ.

A USB device, which represents a real device on a bus, is a bus device, in your
example the class device is the input device.

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Greg KH]

On Mon, Jul 21, 2025 at 02:13:14PM +0200, Danilo Krummrich wrote:
> On Mon Jul 21, 2025 at 2:07 PM CEST, Alice Ryhl wrote:
> > The connection is that to request an irq you must have a &Device<Bound>,
> > so if you can only obtain a &Device<Bound> to a bus device, then that
> > means that you can never request an irq for a class device.
> 
> As mentioned in my other reply, a class device never owns resources of a "real"
> device such as an IRQ.
> 
> A USB device, which represents a real device on a bus, is a bus device, in your
> example the class device is the input device.
>

Just to confuse things a bit more (sorry), there are also "USB class
devices" that represent USB devices that use the "generic" USB interface
to userspace api.  You can find these by searching for
usb_register_dev().

Note, this is different from the drivers/usb/class/ drivers, which
represent various "USB class protocol" that the USB.org group defines,
and those talk to userspace through the various common class apis
depending on the specific device type (input, network, etc.) and are USB
bus drivers.

Naming is hard, wait until you learn about "usb gadget" a word we had to
invent to describe the thing :)

thanks,

greg k-h

Re: [PATCH 2/3] device: rust: expand documentation for Device

[Danilo Krummrich]

On Mon Jul 21, 2025 at 3:17 PM CEST, Greg KH wrote:
> On Mon, Jul 21, 2025 at 02:13:14PM +0200, Danilo Krummrich wrote:
>> On Mon Jul 21, 2025 at 2:07 PM CEST, Alice Ryhl wrote:
>> > The connection is that to request an irq you must have a &Device<Bound>,
>> > so if you can only obtain a &Device<Bound> to a bus device, then that
>> > means that you can never request an irq for a class device.
>> 
>> As mentioned in my other reply, a class device never owns resources of a "real"
>> device such as an IRQ.
>> 
>> A USB device, which represents a real device on a bus, is a bus device, in your
>> example the class device is the input device.
>>
>
> Just to confuse things a bit more (sorry), there are also "USB class
> devices" that represent USB devices that use the "generic" USB interface
> to userspace api.  You can find these by searching for
> usb_register_dev().

In this case the struct usb_interface represents a bus device, that can become
associated with e.g. a class device of the usbmisc class. :)

So, yeah, there are multiple USB devices, some of them are bus devices, some of
them are class devices.

I'm not sure about the exact topology, but AFAIK USB has different bus device
types representing the hierarchy, i.e. a device structure for the entire "real"
USB device and another one for its interfaces.

> Note, this is different from the drivers/usb/class/ drivers, which
> represent various "USB class protocol" that the USB.org group defines,
> and those talk to userspace through the various common class apis
> depending on the specific device type (input, network, etc.) and are USB
> bus drivers.
>
> Naming is hard, wait until you learn about "usb gadget" a word we had to
> invent to describe the thing :)
>
> thanks,
>
> greg k-h

Re: [PATCH 1/3] device: rust: documentation for DeviceContext

[Benno Lossin]

On Fri Jul 18, 2025 at 3:14 PM CEST, Danilo Krummrich wrote:
> On Fri Jul 18, 2025 at 2:32 PM CEST, Alexandre Courbot wrote:
>> On Fri Jul 18, 2025 at 7:45 AM JST, Danilo Krummrich wrote:
>>> +///
>>> +/// Bus devices can automatically implement the dereference hierarchy by using
>>> +/// [`impl_device_context_deref`](kernel::impl_device_context_deref).
>>>  pub trait DeviceContext: private::Sealed {}
>>>  
>>> -/// The [`Normal`] context is the context of a bus specific device when it is not an argument of
>>> -/// any bus callback.
>>> +/// The [`Normal`] context is the default [`DeviceContext`] of any [`Device`].
>>> +///
>>> +/// The normal context does not indicate any specific scope. Any `Device<Ctx>` is also a valid
>>> +/// [`Device<Normal>`]. It is the only [`DeviceContext`] for which it is valid to implement
>>> +/// [`AlwaysRefCounted`](kernel::types::AlwaysRefCounted) for.
>>>  pub struct Normal;
>>
>> `Normal` as a name can be interpreted in many different ways, and in the
>> case of a device context it is not clear what the "normal" state is. I
>> think it would be helpful if we can elaborate a bit more on what this
>> implies (i.e. what concretely speaking are the limitations), and if
>> possible why this name has been chosen.
>
> It's the context that does not guarantee any specific scope. But that's also
> what the documentation says.
>
> I also wouldn't speak of limitations, it's just that it doesn't allow to make
> *additional* assumptions compared to other device context types.
>
> Yet, if you have suggestions on what to add specifically, please let me know
> (maybe simply my previous sentence?).
>
> Regarding the name, "Normal" seems reasonable for the device context that does
> not guarantee any specific scope. We could have also named it just "Default".
>
> I think "Normal" is fine, as in "it's just a normal device reference, no
> specific scope guaranteed".

Not sure if this helps, but `Plain` might carry a slightly different
meaning from `Normal` that is better in this case?

(but you probably can't infer much from the name anyways)

---
Cheers,
Benno