Switch to CoercePointee macro, with examples #1130

dingxiangfei2009 · 2024-11-20T11:34:18Z

Target audience: first timers to build kernel with Rust support

This patch demonstrates the new capability from the current Rust Nightly, which now supports the #[derive(CoercePointee)] macro to derive the key trait implementations, in place of use of unstable language features such as Unsize and DispatchFromDyn.

An illustrative example is attached for more details in the Rust sample kernel module.

How to test this

As usual, you shall follow the guide to set up the toolchain: https://docs.kernel.org/rust/quick-start.html. This demonstration can be executed through virtualization. In this case, during configuration through make LLVM=1 menuconfig, be sure to select Enable Paravirtualization code and KVM Guest support under Processor type and features/Linux guest support; also select Printing Macros as modularised under Kernel hacking/Sample kernel code/Rust samples. Initiate build with make LLVM=1 optionally with -j for parallelism.

Meanwhile, you will need busybox from https://www.busybox.net/downloads/binaries/1.35.0-x86_64-linux-musl/busybox. I am running a x86_64 box, so the demo uses qemu-system-x86_64 as the emulator. I will use the usr/gen_init_cpio script for generating the initramfs image initram.img. Here is the description I used.

dir     /bin                                          0755 0 0
dir     /sys                                          0755 0 0
dir     /dev                                          0755 0 0
file    /bin/busybox  <path to busybox>                         0755 0 0
slink   /bin/sh       /bin/busybox                    0755 0 0
slink   /bin/mount    /bin/busybox                    0755 0 0
slink   /bin/insmod   /bin/busybox                    0755 0 0
slink   /bin/rmmod    /bin/busybox                    0755 0 0
slink   /bin/cut      /bin/busybox                    0755 0 0
slink   /bin/cat      /bin/busybox                    0755 0 0
slink   /bin/cd       /bin/busybox                    0755 0 0
slink   /bin/ls       /bin/busybox                    0755 0 0
file    /init         init                            0755 0 0
file    /rust_print.ko <path to linux>/linux/samples/rust/rust_print.ko 0755 0 0

I used this command to launch the emulation and it dropped to a shell.

qemu-system-x86_64 \
    -kernel arch/x86_64/boot/bzImage \
    -initrd <path to initram.img> \
    -M pc \
    -m 4G \
    -cpu Cascadelake-Server \
    -smp 4 \
    -nographic \
    -vga none \
    -no-reboot \
    -accel kvm \
    -append 'console=ttyS0'

In the shell I executed insmod rust_print.ko.

What would you see

Upon successful launch, a quick insmod rust_print.ko will generate some kernel log. You may watch out for the following log stanza, especially the lines containing the text Arc<dyn Display>.

...
[   36.255440] rust_print: "hello, world"
[   36.255627] rust_print: [samples/rust/rust_print.rs:35:5] c = "hello, world"
[   36.256245] rust_print: Arc<dyn Display> says 42
[   36.256267] rust_print: Arc<dyn Display> says hello, world
[   36.256579] rust_print: "hello, world"
...

TODO(@wieDasDing): work on the patch epilogue a bit more

Darksonn · 2024-11-20T11:51:16Z

rust/kernel/list/arc.rs

-// This is to allow coercion from `ListArc<T>` to `ListArc<U>` if `T` can be converted to the
-// dynamically-sized type (DST) `U`.
-impl<T, U, const ID: u64> core::ops::CoerceUnsized<ListArc<U, ID>> for ListArc<T, ID>
-where
-    T: ListArcSafe<ID> + Unsize<U> + ?Sized,
-    U: ListArcSafe<ID> + ?Sized,
-{
-}
-
-// This is to allow `ListArc<U>` to be dispatched on when `ListArc<T>` can be coerced into
-// `ListArc<U>`.
-impl<T, U, const ID: u64> core::ops::DispatchFromDyn<ListArc<U, ID>> for ListArc<T, ID>
-where
-    T: ListArcSafe<ID> + Unsize<U> + ?Sized,
-    U: ListArcSafe<ID> + ?Sized,
-{
-}


We can't delete these due to support for older rustc releases. Please see this doc for info on conditional compilation.

For reference, our current minimum version is 1.78.0, so you can use that to test.

Okay, I will revert those chunks and wrap it with a conditional

dingxiangfei2009 · 2024-11-22T12:57:37Z

From now I am going to squash the commits a bit and attach proper commit messages. I will run the checks over the commit as per instructions.

ojeda · 2024-11-22T13:16:45Z

I will run the checks over the commit as per instructions.

Sounds great, thanks for doing that -- please also take a look at: https://rust-for-linux.com/contributing#submit-checklist-addendum (i.e. our own list, mostly about running rustfmt, Clippy...).

By the way, for linking to kernel docs, I recommend docs.kernel.org, e.g. the latest version would be at: https://docs.kernel.org/process/submitting-patches.html

nbdd0121 · 2024-11-23T17:20:56Z

rust/kernel/sync/arc.rs

@@ -471,13 +479,16 @@ impl<T: ?Sized> From<Pin<UniqueArc<T>>> for Arc<T> {
 /// obj.as_arc_borrow().use_reference();
 /// # Ok::<(), Error>(())
 /// ```
+#[cfg_attr(CONFIG_RUST_COERCE_POINTEE, repr(transparent))]


This can be unconditional

Darksonn · 2024-11-25T13:47:59Z

samples/rust/rust_print.rs

+    #[cfg(not(CONFIG_RUST_COERCE_POINTEE))]
+    pr_info!(
+        "The demonstration for dynamic dispatching through Arc is skipped. "
+        "To see the print-out from this example, please upgrade your Rust compiler to 1.83.0 or higher.\n"
+    );


Why is this example limited. Does the logic we have pre-1.83 not work for the sample?

Darksonn · 2024-11-25T13:49:08Z

rust/kernel/list/arc.rs

+/// `ListArc` is well-behaved so that its dereferencing operation does not mutate.
+#[cfg(CONFIG_RUST_COERCE_POINTEE)]
+unsafe impl<T: ?Sized + ListArcSafe<ID>, const ID: u64> core::pin::PinCoerceUnsized
+    for ListArc<T, ID>
+{
+}


I don't think we need PinCoerceUnsized.

Gotcha. Dropped.

Darksonn · 2024-11-25T13:49:49Z

rust/kernel/lib.rs

+#![cfg_attr(
+    CONFIG_RUST_COERCE_POINTEE,
+    feature(derive_coerce_pointee, pin_coerce_unsized_trait)
+)]
+#![cfg_attr(
+    not(CONFIG_RUST_COERCE_POINTEE),
+    feature(coerce_unsized, dispatch_from_dyn, unsize)
+)]


Does this format better if we split each feature out on its own line?

I have splitted them in the final patch. I will push the results shortly.

Darksonn

Overall looks good and ready for submission. Just one nit:

Darksonn · 2024-11-27T20:31:33Z

rust/kernel/sync/arc.rs

+#[cfg(CONFIG_RUST_COERCE_POINTEE)]
+use core::marker::CoercePointee;
+#[cfg(not(CONFIG_RUST_COERCE_POINTEE))]
+use core::marker::Unsize;


Maybe use full paths for these instead of the imports?

Yes, fully qualified paths looks much better and easier to reason given all the cfg conditional.

Since Rust 1.84.0 the macro `CoercePointee` has been made public on Nightly, so that it answers the some usability questions. If one wants to equip generic types with the ability to weaken itself and to work with unsized types with dynamic dispatching. This feature is useful such that Rust code are enabled to work with a family of types satisfying the same protocol or Rust traits, while the same safety guarantees are still uphold [1]. Examples of this weakening include those from *[u8; 8]* to *[u8]*, eliding the concrete size of the array; and a concrete type *T* to *dyn Trait* where *T* implements the trait or traits *Trait*. As of date, the exact language features to enable this type weakening is still under stabilization effort. Nevertheless, Alice Ryhl has proposed [2] a very valuable combination of them such that a user can enable this feature via a procedural macro `CoercePointee` without much verbosity and without declaring dependence on the relevant unstable language features. Alice has previously filed a patch [3] to demonstrate the capability of this macro. This patch provides further updates to incorporate recent changes to the proposal in [2] and paves the way for the final stabilization of the feature in the Rust language. A minimal demostration code is added to the *samples/rust/rust_print_main.rs* module. The use of the macro is now gated behind the available Rust version *1.83.0*. The *kernel* crate will still be as functional on the prior Rust toolchains. Link: https://doc.rust-lang.org/stable/nomicon/exotic-sizes.html?highlight=dynamic#dynamically-sized-types-dsts [1] Link: https://rust-lang.github.io/rfcs/3621-derive-smart-pointer.html [2] Link: https://lore.kernel.org/all/[email protected]/ [3] Signed-off-by: Xiangfei Ding <[email protected]>

ojeda mentioned this pull request Nov 20, 2024

Rust unstable features needed for the kernel #2

Open

90 tasks

Darksonn reviewed Nov 20, 2024

View reviewed changes

dingxiangfei2009 force-pushed the coerce-pointee-example branch 2 times, most recently from 27e8103 to 36e9396 Compare November 22, 2024 12:56

nbdd0121 reviewed Nov 23, 2024

View reviewed changes

Darksonn reviewed Nov 25, 2024

View reviewed changes

Darksonn reviewed Nov 27, 2024

View reviewed changes

dingxiangfei2009 force-pushed the coerce-pointee-example branch from 507c541 to 06e4e20 Compare November 29, 2024 16:54

ojeda force-pushed the rust-next branch from 92797e9 to 9b98be7 Compare December 16, 2024 20:49

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Switch to CoercePointee macro, with examples #1130

Switch to CoercePointee macro, with examples #1130

dingxiangfei2009 commented Nov 20, 2024 •

edited

Loading

Darksonn Nov 20, 2024

ojeda Nov 20, 2024

dingxiangfei2009 Nov 20, 2024 •

edited

Loading

dingxiangfei2009 commented Nov 22, 2024

ojeda commented Nov 22, 2024

nbdd0121 Nov 23, 2024

Darksonn Nov 25, 2024

Darksonn Nov 25, 2024

dingxiangfei2009 Nov 29, 2024

Darksonn Nov 25, 2024

dingxiangfei2009 Nov 29, 2024

Darksonn left a comment

Darksonn Nov 27, 2024

dingxiangfei2009 Nov 29, 2024

Switch to CoercePointee macro, with examples #1130

Are you sure you want to change the base?

Switch to CoercePointee macro, with examples #1130

Conversation

dingxiangfei2009 commented Nov 20, 2024 • edited Loading

How to test this

What would you see

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dingxiangfei2009 Nov 20, 2024 • edited Loading

Choose a reason for hiding this comment

dingxiangfei2009 commented Nov 22, 2024

ojeda commented Nov 22, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Darksonn left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dingxiangfei2009 commented Nov 20, 2024 •

edited

Loading

dingxiangfei2009 Nov 20, 2024 •

edited

Loading