[PATCH] rust: iommu: replace core::mem::zeroed with pin_init::zeroed

[PATCH] rust: iommu: replace core::mem::zeroed with pin_init::zeroed

[Nicolás Antinori]

All types in `bindings` implement `Zeroable` if they can. This enables
using `pin_init::zeroed()` for `io_pgtable_cfg` initialization instead
of relying on `..unsafe { core::mem::zeroed() }`.

This change improves readability and removes an unnecessary unsafe
block.

Link: https://github.com/Rust-for-Linux/linux/issues/1189
Suggested-by: Benno Lossin <lossin@kernel.org>
Signed-off-by: Nicolás Antinori <nico.antinori.7@gmail.com>
---
 rust/kernel/iommu/pgtable.rs | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/rust/kernel/iommu/pgtable.rs b/rust/kernel/iommu/pgtable.rs
index c88e38fd938a..8faa1d8d9fad 100644
--- a/rust/kernel/iommu/pgtable.rs
+++ b/rust/kernel/iommu/pgtable.rs
@@ -102,8 +102,7 @@ pub unsafe fn new_raw(dev: &Device<Bound>, config: Config) -> Result<IoPageTable
             coherent_walk: config.coherent_walk,
             tlb: &raw const NOOP_FLUSH_OPS,
             iommu_dev: dev.as_raw(),
-            // SAFETY: All zeroes is a valid value for `struct io_pgtable_cfg`.
-            ..unsafe { core::mem::zeroed() }
+            ..pin_init::zeroed()
         };

         // SAFETY:
--
2.47.3

Re: [PATCH] rust: iommu: replace core::mem::zeroed with pin_init::zeroed

[Alexandre Courbot]

On Sat Jun 27, 2026 at 5:31 AM JST, Nicolás Antinori wrote:
> All types in `bindings` implement `Zeroable` if they can. This enables
> using `pin_init::zeroed()` for `io_pgtable_cfg` initialization instead
> of relying on `..unsafe { core::mem::zeroed() }`.
>
> This change improves readability and removes an unnecessary unsafe
> block.
>
> Link: https://github.com/Rust-for-Linux/linux/issues/1189
> Suggested-by: Benno Lossin <lossin@kernel.org>
> Signed-off-by: Nicolás Antinori <nico.antinori.7@gmail.com>
> ---
>  rust/kernel/iommu/pgtable.rs | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/rust/kernel/iommu/pgtable.rs b/rust/kernel/iommu/pgtable.rs
> index c88e38fd938a..8faa1d8d9fad 100644
> --- a/rust/kernel/iommu/pgtable.rs
> +++ b/rust/kernel/iommu/pgtable.rs
> @@ -102,8 +102,7 @@ pub unsafe fn new_raw(dev: &Device<Bound>, config: Config) -> Result<IoPageTable
>              coherent_walk: config.coherent_walk,
>              tlb: &raw const NOOP_FLUSH_OPS,
>              iommu_dev: dev.as_raw(),
> -            // SAFETY: All zeroes is a valid value for `struct io_pgtable_cfg`.
> -            ..unsafe { core::mem::zeroed() }
> +            ..pin_init::zeroed()

Note that you can also use `..Zeroable::zeroed()` here since we are not
in const context. Not sure which one we prefer in this case, but I'd
personally err on the side of not referring to `pin_init` unless we are
dealing with a pinned object or need to because of a const requirement.

Re: [PATCH] rust: iommu: replace core::mem::zeroed with pin_init::zeroed

[Miguel Ojeda]

On Sat, Jun 27, 2026 at 10:35 AM Alexandre Courbot <acourbot@nvidia.com> wrote:
>
> Note that you can also use `..Zeroable::zeroed()` here since we are not
> in const context. Not sure which one we prefer in this case, but I'd
> personally err on the side of not referring to `pin_init` unless we are
> dealing with a pinned object or need to because of a const requirement.

That sounds reasonable -- Benno/Gary, any comments?

By the way, speaking of that: the docs of `zeroed()` (both of them)
could mention the existence of the other (and the `const fn`
difference).

I guess this will eventually go through IOMMU, but I can also take it
via `rust-next` if needed.

Thanks!

Cheers,
Miguel

Re: [PATCH] rust: iommu: replace core::mem::zeroed with pin_init::zeroed

[Gary Guo]

On Sun Jun 28, 2026 at 11:31 AM BST, Miguel Ojeda wrote:
> On Sat, Jun 27, 2026 at 10:35 AM Alexandre Courbot <acourbot@nvidia.com> wrote:
>>
>> Note that you can also use `..Zeroable::zeroed()` here since we are not
>> in const context. Not sure which one we prefer in this case, but I'd
>> personally err on the side of not referring to `pin_init` unless we are
>> dealing with a pinned object or need to because of a const requirement.
>
> That sounds reasonable -- Benno/Gary, any comments?

`Zeroable::zeroed()` should be preferred where possible. Currently init macros
also recognize and turn them into zero-init if used inside init macros --
outside init macros they don't matter, but in a future where const trait methods
are stable we would want to prefer `Zeroable::zeroed()` anyway.

Best,
Gary

Re: [PATCH] rust: iommu: replace core::mem::zeroed with pin_init::zeroed

[Nicolás Antinori]

On Sun Jun 28, 2026 at 7:31 AM -03, Miguel Ojeda wrote:
> On Sat, Jun 27, 2026 at 10:35 AM Alexandre Courbot <acourbot@nvidia.com> wrote:
>>
>> Note that you can also use `..Zeroable::zeroed()` here since we are not
>> in const context. Not sure which one we prefer in this case, but I'd
>> personally err on the side of not referring to `pin_init` unless we are
>> dealing with a pinned object or need to because of a const requirement.
>
> That sounds reasonable -- Benno/Gary, any comments?
>
> By the way, speaking of that: the docs of `zeroed()` (both of them)
> could mention the existence of the other (and the `const fn`
> difference).

If it is ok, I can send a patch modifying the doc comments to link both 
functions, noting that the const version should be used in const 
contexts/pinned objects and that `Zeroable::zeroed()` should be 
preferred when possible (as Gary commented in [1]).

I'll send a v2 with `..Zeroable::zeroed()` shortly.

Thank you all for the comments!

[1] https://lore.kernel.org/rust-for-linux/DJKV3IQ0SXA1.30T5Y71AL6KV7@garyguo.net/

Re: [PATCH] rust: iommu: replace core::mem::zeroed with pin_init::zeroed

[Miguel Ojeda]

On Mon, Jun 29, 2026 at 9:46 PM Nicolás Antinori
<nico.antinori.7@gmail.com> wrote:
>
> If it is ok, I can send a patch modifying the doc comments to link both
> functions, noting that the const version should be used in const
> contexts/pinned objects and that `Zeroable::zeroed()` should be
> preferred when possible (as Gary commented in [1]).

Of course, please always feel free to send a patch (it depends if the
maintainers end up accepting it, in this case Benno/Gary, but sending
is OK :)

You can also send it as a PR to the pin-init repository -- I think
they still accept both patches here and PRs there, but a PR there may
be easier for them.

Thanks!

Cheers,
Miguel