Re: [PATCH v6 RESEND 6/7] rust: pci: add config space read/write support

[Zhi Wang <zhiw@nvidia.com>]

On Thu, 13 Nov 2025 19:20:05 -0500
Joel Fernandes <joelagnelf@nvidia.com> wrote:

> Hi Zhi,
> 
> On Mon, Nov 10, 2025 at 10:41:18PM +0200, Zhi Wang wrote:
> [..]  
> >  impl Device<device::Core> {
> > diff --git a/rust/kernel/pci/io.rs b/rust/kernel/pci/io.rs
> > index 2bbb3261198d..bb78a83fe92c 100644
> > --- a/rust/kernel/pci/io.rs
> > +++ b/rust/kernel/pci/io.rs
> > @@ -2,12 +2,19 @@
> >  
> >  //! PCI memory-mapped I/O infrastructure.
> >  
> > -use super::Device;
> > +use super::{
> > +    ConfigSpaceSize,
> > +    Device, //
> > +};
> >  use crate::{
> >      bindings,
> >      device,
> >      devres::Devres,
> >      io::{
> > +        define_read,
> > +        define_write,
> > +        Io,
> > +        IoInfallible,
> >          Mmio,
> >          MmioRaw, //
> >      },
> > @@ -16,6 +23,58 @@
> >  };
> >  use core::ops::Deref;
> >  
> > +/// The PCI configuration space of a device.
> > +///
> > +/// Provides typed read and write accessors for configuration
> > registers +/// using the standard `pci_read_config_*` and
> > `pci_write_config_*` helpers. +///
> > +/// The generic const parameter `SIZE` can be used to indicate the
> > +/// maximum size of the configuration space (e.g. 256 bytes for
> > legacy, +/// 4096 bytes for extended config space).
> > +pub struct ConfigSpace<'a, const SIZE: usize = {
> > ConfigSpaceSize::Extended as usize }> {
> > +    pub(crate) pdev: &'a Device<device::Bound>,
> > +}
> > +
> > +macro_rules! call_config_read {
> > +    (infallible, $c_fn:ident, $self:ident, $ty:ty, $addr:expr) =>
> > {{
> > +        let mut val: $ty = 0;
> > +        let _ret = unsafe { bindings::$c_fn($self.pdev.as_raw(),
> > $addr as i32, &mut val) };
> > +        val
> > +    }};
> > +}
> > +
> > +macro_rules! call_config_write {
> > +    (infallible, $c_fn:ident, $self:ident, $ty:ty, $addr:expr,
> > $value:expr) => {
> > +        let _ret = unsafe { bindings::$c_fn($self.pdev.as_raw(),
> > $addr as i32, $value) };
> 
> unsafe block needs safety comments, also I understand 'as' to convert

snip

> is generally forbidden without a CAST: comment or using ::from() for
> conversion because it can by a lossy conversion.
> 

Let me take a look, basically this was similar with the define_{read,
mmio} macros, which has originally been from the mainline. Might have to
fix that part as well.

> Also we should have a comment on why its safe for _ret to be ignored.
> Basically what guarantees that the call is really infallible?
> Anything we can do to ensure errors are not silently ignored? Let me
> know if I missed something.
> 

This was discussed with Danilo on Zulip. The driver will observe a
!0 value on read when an error occurs in the infallible accessors. For
writes, I think the driver should read the value back. (similar with
MMIO cases)

Besides those, I think the driver needs to use fallible ones if it
really cares about the return.

Z.

> [..]
> 
> > +
> > +    /// Return an initialized config space object.
> > +    pub fn config_space_exteneded<'a>(
> 
> typo in func name.
> 
> thanks,
> 
>  - Joel
> 
> > +        &'a self,
> > +    ) -> Result<ConfigSpace<'a, {
> > ConfigSpaceSize::Extended.as_raw() }>> {
> > +        Ok(ConfigSpace { pdev: self })
> > +    }
> >  }
> > -- 
> > 2.51.0
> >