[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <DC0INEKNZPSE.WLEQH24SLW1I@nvidia.com>
Date: Tue, 12 Aug 2025 23:24:25 +0900
From: "Alexandre Courbot" <acourbot@...dia.com>
To: "Christian S. Lima" <christiansantoslima21@...il.com>, "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>,
"Danilo Krummrich" <dakr@...nel.org>, <rust-for-linux@...r.kernel.org>,
<linux-kernel@...r.kernel.org>, <~lkcamp/patches@...ts.sr.ht>,
<richard120310@...il.com>
Subject: Re: [PATCH v9] rust: transmute: Add methods for FromBytes trait
On Tue Aug 12, 2025 at 6:38 AM JST, Christian S. Lima wrote:
> The two methods added take a slice of bytes and return those bytes in a
> specific type. These methods are useful when we need to transform the
> stream of bytes into specific type.
>
> The `Frombytessized` trait was added to make it easier to implement other
typo: FromBytesSized.
> user defined types within the codebase. With the current implementation,
> there's no way to interact without implementing `from_bytes` and
> `from_mut_bytes` for every new type, and this would end up generating a lot
> of duplicate code. By using FromBytesSized as a proxy trait, we can avoid
> this without generating a direct dependency. If necessary, the user can
> simply implement `FromBytes` if needed. For more context, please check the
> [1] and [2].
>
> [1] https://lore.kernel.org/rust-for-linux/DANSZ6Q476EC.3GY00K717QVUL@nvidia.com/
> [2] https://lore.kernel.org/rust-for-linux/DAOESYD6F287.3U3M64X0S1WN5@nvidia.com/
>
> Link: https://github.com/Rust-for-Linux/linux/issues/1119
> Suggested-by: Alexandre Courbot <acourbot@...dia.com>
> Signed-off-by: Christian S. Lima <christiansantoslima21@...il.com>
I mentioned it on v8 [1] and v7 [2], but the tests that break due to
this change need to be updated by this patch as well. This includes:
* The doctests in `rust/kernel/dma.rs`,
* The `samples/rust/rust_dma.rs` sample,
* The example for `FromBytes` introduced by this patch which uses `?` without
defining a function.
Please make sure to compile with `CONFIG_RUST_KERNEL_DOCTESTS` and
`CONFIG_SAMPLE_RUST_DMA` to see the failures.
Also, now that we are on 6.17-rc1, the types in `nova-core` that implement
`FromBytes` will also fail to build unless they are switched to
`FromBytesSized`. Which I guess speaks in favor of taking this into the Nova
tree if there are not other planned user for this cycle?
[1]
https://lore.kernel.org/rust-for-linux/DB5KEWX9EJ2Q.3CX5EGS66OVHH@nvidia.com/
[2]
https://lore.kernel.org/rust-for-linux/DANSZ6Q476EC.3GY00K717QVUL@nvidia.com/
In case this helps, here is the diff I needed to apply to get the tests
to build:
diff --git a/rust/kernel/dma.rs b/rust/kernel/dma.rs
index 1801836f31455..3797c70c13040 100644
--- a/rust/kernel/dma.rs
+++ b/rust/kernel/dma.rs
@@ -597,7 +597,7 @@ unsafe impl<T: AsBytes + FromBytes + Send> Send for CoherentAllocation<T> {}
/// struct MyStruct { field: u32, }
///
/// // SAFETY: All bit patterns are acceptable values for `MyStruct`.
-/// unsafe impl kernel::transmute::FromBytes for MyStruct{};
+/// unsafe impl kernel::transmute::FromBytesSized for MyStruct{};
/// // SAFETY: Instances of `MyStruct` have no uninitialized portions.
/// unsafe impl kernel::transmute::AsBytes for MyStruct{};
///
@@ -641,7 +641,7 @@ macro_rules! dma_read {
/// struct MyStruct { member: u32, }
///
/// // SAFETY: All bit patterns are acceptable values for `MyStruct`.
-/// unsafe impl kernel::transmute::FromBytes for MyStruct{};
+/// unsafe impl kernel::transmute::FromBytesSized for MyStruct{};
/// // SAFETY: Instances of `MyStruct` have no uninitialized portions.
/// unsafe impl kernel::transmute::AsBytes for MyStruct{};
///
diff --git a/rust/kernel/transmute.rs b/rust/kernel/transmute.rs
index ba21fe49e4f07..e7bd698ec99cd 100644
--- a/rust/kernel/transmute.rs
+++ b/rust/kernel/transmute.rs
@@ -16,6 +16,7 @@
/// ```
/// use kernel::transmute::FromBytes;
///
+/// # fn test() -> Option<()> {
/// let foo = [1, 2, 3, 4];
///
/// let result = u32::from_bytes(&foo)?;
@@ -25,6 +26,7 @@
///
/// #[cfg(target_endian = "big")]
/// assert_eq!(*result, 0x1020304);
+/// # Some(()) }
/// ```
///
/// # Safety
diff --git a/samples/rust/rust_dma.rs b/samples/rust/rust_dma.rs
index 9fb3ae1dd8587..36ad877d4804d 100644
--- a/samples/rust/rust_dma.rs
+++ b/samples/rust/rust_dma.rs
@@ -65,7 +65,7 @@ fn new(h: u32, b: u32) -> Self {
// SAFETY: All bit patterns are acceptable values for `MyStruct`.
unsafe impl kernel::transmute::AsBytes for MyStruct {}
// SAFETY: Instances of `MyStruct` have no uninitialized portions.
-unsafe impl kernel::transmute::FromBytes for MyStruct {}
+unsafe impl kernel::transmute::FromBytesSized for MyStruct {}
kernel::pci_device_table!(
PCI_TABLE,
-- end of diff --
Otherwise the patch in itself looks good to me. It is used in nova-core (and a
critical dependency for this cycle :)), so although I cannot give my
Reviewed-by until the tests are fixed, it is at least
Tested-by: Alexandre Courbot <acourbot@...dia.com>
Powered by blists - more mailing lists