Message-ID: <87r02xhgoa.fsf@protonmail.com>
Date: Sun, 16 Mar 2025 02:07:23 +0000
From: Rahul Rameshbabu <sergeantsagara@...tonmail.com>
To: Danilo Krummrich <dakr@...nel.org>
Cc: linux-kernel@...r.kernel.org, rust-for-linux@...r.kernel.org, linux-input@...r.kernel.org, dri-devel@...ts.freedesktop.org, Jiri Kosina <jikos@...nel.org>, Benjamin Tissoires <bentiss@...nel.org>, Miguel Ojeda <ojeda@...nel.org>, Alex Gaynor <alex.gaynor@...il.com>, Boqun Feng <boqun.feng@...il.com>, Gary Guo <gary@...yguo.net>, Björn Roy Baron <bjorn3_gh@...tonmail.com>, Benno Lossin <benno.lossin@...ton.me>, Andreas Hindborg <a.hindborg@...nel.org>, Alice Ryhl <aliceryhl@...gle.com>, Trevor Gross <tmgross@...ch.edu>
Subject: Re: [PATCH RFC 1/3] rust: core abstractions for HID drivers

On Thu, 13 Mar, 2025 20:25:23 +0100 "Danilo Krummrich" <dakr@...nel.org> wrote:
> On Thu, Mar 13, 2025 at 04:03:35PM +0000, Rahul Rameshbabu wrote:
>> These abstractions enable the development of HID drivers in Rust by binding
>> with the HID core C API. They provide Rust types that map to the
>> equivalents in C. In this initial draft, only hid_device and hid_device_id
>> are provided direct Rust type equivalents. hid_driver is specially wrapped
>> with a custom Driver type. The module_hid_driver! macro provides analogous
>> functionality to its C equivalent.
>>
>> Future work for these abstractions would include more bindings for common
>> HID-related types, such as hid_field, hid_report_enum, and hid_report.
>> Providing Rust equivalents to useful core HID functions will also be
>> necessary for HID driver development in Rust.
>>
>> Some concerns with this initial draft
>>   - The need for a DeviceId and DeviceIdShallow type.
>>     + DeviceIdShallow is used to guarantee the safety requirement for the
>>       Sync trait.
>>   - The create_hid_driver call in the module_hid_driver! macro does not use
>>     Pin semantics for passing the ID_TABLE. I could not get Pin semantics
>>     to work in a const fn. I get a feeling this might be safe but need help
>>     reviewing this.
>
> For a lot of things in this patch we have common infrastructure, please see
> rust/kernel/{device.rs, driver.rs, device_id.rs}. I think you should make use of
> the common infrastructure that solves the corresponding problems already.

Absolutely! 9b90864bb42b ("rust: implement `IdArray`, `IdTable` and
`RawDeviceId`"). The types here look really useful, so I am sorry for
missing this. Will refactor my next revision to make use of this common
infrastructure.

>
> It provides generic infrastructure for handling device IDs, a generalized
> Registration type, based on InPlaceModule with a common module_driver!
> implementation for busses to implement their corresponding module macro, etc.

Wow, module_driver! is very nice. On one hand, out of all of Rust's
semantics, writing macro_rules! and proc macros were difficult for me
before I started working on this. I think I am a bit better with them
now after this. On the other hand looking at how module_pci_driver is
implemented, I love how much this common infrastructure simplifies the
effort for various buses. Will be moving to this in my next revision as
well.

>
> There are two busses upstream, which are based on this infrastructure:
> rust/kernel/{pci.rs, platform.rs}.

Thanks for the pointers.

>
> There is a patch series that improves soundness of those two bus abstractions
> [1], which should be taken into consideration too. Even though your
> implementation isn't prone to the addressed issue, it would be good to align
> things accordingly.
>
> There is a third bus abstraction (auxiliary) on the mailing list [2], which
> already implements the mentioned improvements, which you can use as canonical
> example too.
>
> [1] https://lore.kernel.org/rust-for-linux/20250313021550.133041-1-dakr@kernel.org/
> [2] https://lore.kernel.org/rust-for-linux/20250313022454.147118-1-dakr@kernel.org/

Will base my work on top of the patches in
https://web.git.kernel.org/pub/scm/linux/kernel/git/dakr/linux.git/log/?h=rust/device.

Haven't had a chance yet to go through the patches in detail but the
main one of importance seems to be "rust: device: implement device
context marker" which adds the Normal and Core contexts for restricting
certain device functions from being called in certain bus callbacks
only. Will look at how the auxiliary bus abstraction makes use of this.

>

I did have a question with regards to a pattern of "getters" that I have
been implementing in my abstractions. Benjamin Tissoires brought up that
the pattern looked repetitive and was wondering if it can be generated
dynamically. I replied that I was of the opinion that it could not[1].

[1] https://lore.kernel.org/rust-for-linux/Z9MxI0u2yCfSzTvD@cassiopeiae/T/#m1b5b1ca96503a542c0da3089ef3f97e606032240

Thanks for the review,
Rahul Rameshbabu

>> Signed-off-by: Rahul Rameshbabu <sergeantsagara@...tonmail.com>
>> ---
>>  drivers/hid/Kconfig             |   8 ++
>>  rust/bindings/bindings_helper.h |   1 +
>>  rust/kernel/hid.rs              | 245 ++++++++++++++++++++++++++++++++
>>  rust/kernel/lib.rs              |   2 +
>>  4 files changed, 256 insertions(+)
>>  create mode 100644 rust/kernel/hid.rs

Powered by blists - more mailing lists