Re: [PATCH v4 2/4] samples: rust: Provide example using the new Rust MiscDevice abstraction

[Danilo Krummrich <dakr@kernel.org>]

On Fri, Dec 06, 2024 at 10:40:23AM +0100, Alice Ryhl wrote:
> On Fri, Dec 6, 2024 at 10:29 AM Danilo Krummrich <dakr@kernel.org> wrote:
> >
> > On Fri, Dec 06, 2024 at 09:05:06AM +0000, Lee Jones wrote:
> > > This sample driver demonstrates the following basic operations:
> > >
> > > * Register a Misc Device
> > > * Create /dev/rust-misc-device
> > > * Provide open call-back for the aforementioned character device
> > > * Operate on the character device via a simple ioctl()
> > > * Provide close call-back for the character device
> > >
> > > Signed-off-by: Lee Jones <lee@kernel.org>
> > > ---
> > >  samples/rust/Kconfig             | 10 ++++
> > >  samples/rust/Makefile            |  1 +
> > >  samples/rust/rust_misc_device.rs | 80 ++++++++++++++++++++++++++++++++
> > >  3 files changed, 91 insertions(+)
> > >  create mode 100644 samples/rust/rust_misc_device.rs
> > >
> > > diff --git a/samples/rust/Kconfig b/samples/rust/Kconfig
> > > index b0f74a81c8f9..df384e679901 100644
> > > --- a/samples/rust/Kconfig
> > > +++ b/samples/rust/Kconfig
> > > @@ -20,6 +20,16 @@ config SAMPLE_RUST_MINIMAL
> > >
> > >         If unsure, say N.
> > >
> > > +config SAMPLE_RUST_MISC_DEVICE
> > > +     tristate "Misc device"
> > > +     help
> > > +       This option builds the Rust misc device.
> > > +
> > > +       To compile this as a module, choose M here:
> > > +       the module will be called rust_misc_device.
> > > +
> > > +       If unsure, say N.
> > > +
> > >  config SAMPLE_RUST_PRINT
> > >       tristate "Printing macros"
> > >       help
> > > diff --git a/samples/rust/Makefile b/samples/rust/Makefile
> > > index c1a5c1655395..ad4b97a98580 100644
> > > --- a/samples/rust/Makefile
> > > +++ b/samples/rust/Makefile
> > > @@ -2,6 +2,7 @@
> > >  ccflags-y += -I$(src)                                # needed for trace events
> > >
> > >  obj-$(CONFIG_SAMPLE_RUST_MINIMAL)            += rust_minimal.o
> > > +obj-$(CONFIG_SAMPLE_RUST_MISC_DEVICE)                += rust_misc_device.o
> > >  obj-$(CONFIG_SAMPLE_RUST_PRINT)                      += rust_print.o
> > >
> > >  rust_print-y := rust_print_main.o rust_print_events.o
> > > diff --git a/samples/rust/rust_misc_device.rs b/samples/rust/rust_misc_device.rs
> > > new file mode 100644
> > > index 000000000000..3837532d259e
> > > --- /dev/null
> > > +++ b/samples/rust/rust_misc_device.rs
> > > @@ -0,0 +1,80 @@
> > > +// SPDX-License-Identifier: GPL-2.0
> > > +
> > > +// Copyright (C) 2024 Google LLC.
> > > +
> > > +//! Rust misc device sample.
> > > +
> > > +use kernel::{
> > > +    c_str,
> > > +    ioctl::_IO,
> > > +    miscdevice::{MiscDevice, MiscDeviceOptions, MiscDeviceRegistration},
> > > +    prelude::*,
> > > +};
> > > +
> > > +const RUST_MISC_DEV_HELLO: u32 = _IO('|' as u32, 9);
> > > +
> > > +module! {
> > > +    type: RustMiscDeviceModule,
> > > +    name: "rust_misc_device",
> > > +    author: "Lee Jones",
> > > +    description: "Rust misc device sample",
> > > +    license: "GPL",
> > > +}
> > > +
> > > +struct RustMiscDeviceModule {
> > > +    _miscdev: Pin<KBox<MiscDeviceRegistration<RustMiscDevice>>>,
> > > +}
> > > +
> > > +impl kernel::Module for RustMiscDeviceModule {
> > > +    fn init(_module: &'static ThisModule) -> Result<Self> {
> > > +        pr_info!("Initialising Rust Misc Device Sample\n");
> > > +
> > > +        let options = MiscDeviceOptions {
> > > +            name: c_str!("rust-misc-device"),
> > > +        };
> > > +
> > > +        Ok(Self {
> > > +            _miscdev: KBox::pin_init(
> > > +                MiscDeviceRegistration::<RustMiscDevice>::register(options),
> > > +                GFP_KERNEL,
> > > +            )?,
> > > +        })
> >
> > Since this v4 was sent just when I was commenting on v3:
> >
> > Why do we add examples where we ask people to allocate those structures with
> > kmalloc()?
> >
> > `MiscDevice` should switch to the generic `Registration` type in [1] and use
> > `InPlaceModule`, such that those structures land in the .data section of the
> > binary.
> >
> > [1] https://lore.kernel.org/rust-for-linux/20241205141533.111830-3-dakr@kernel.org/
> >
> > > +    }
> > > +}
> > > +
> > > +struct RustMiscDevice;
> > > +
> > > +#[vtable]
> > > +impl MiscDevice for RustMiscDevice {
> > > +    type Ptr = KBox<Self>;
> > > +
> > > +    fn open() -> Result<KBox<Self>> {
> > > +        pr_info!("Opening Rust Misc Device Sample\n");
> >
> > This should be `dev_info!`, but I see why you don't have access to the device
> > structure here...
> >
> > @Greg: How do miscdev drivers do this in C? I looked at a couple of them, but
> > all of those seem to use pr_* macros. They can't get the device pointer from
> > the inode or file pointer.
> >
> > However, C drivers could refer to the struct miscdevice directly since it's
> > almost always a static variable in the file scope.
> >
> > In Rust we do the static allocation part with `InPlaceModule` as mentioned
> > above. However, this still doesn't let us refer to the underlying struct
> > miscdevice.
> >
> > This all would be much cleaner if there'd be a "fake" probe() callback for a
> > struct miscdevice. This way we could perfectly align the miscdevice abstraction
> > with the abstractions for all other drivers, such as PCI, platform, etc.
> 
> It turns out that the file private data is a pointer to the `struct
> miscdevice` in fops->open(), so we can access it in open. To access it
> in other fops hooks, we need to take a refcount on the device in open
> and stash it.

Even better! I suggest to represent it as `misc::Device` and align it with
`platform::Device`, `pci::Device`, etc.

> 
> Alice
>