Re: [PATCH v3 1/3] rust: pci: provide access to PCI Class, subclass, implementation values

[John Hubbard <jhubbard@nvidia.com>]

On 8/19/25 2:09 AM, Danilo Krummrich wrote:
> On Tue Aug 19, 2025 at 5:11 AM CEST, John Hubbard wrote:
>> +/// PCI device class codes. Each entry contains the full 24-bit PCI
>> +/// class code (base class in bits 23-16, subclass in bits 15-8,
>> +/// programming interface in bits 7-0).
>> +///
>> +/// # Examples
>> +///
>> +/// ```
>> +/// # use kernel::{device::Core, pci::{self, Class}, prelude::*};
>> +/// fn probe_device(pdev: &pci::Device<Core>) -> Result<()> {
>> +///     // Get the PCI class for this device
>> +///     let pci_class = pdev.pci_class();
>> +///     dev_info!(
>> +///         pdev.as_ref(),
>> +///         "Detected PCI class: (0x{:06x})\n",
>> +///         pci_class.as_u32()
>> +///     );
> 
> Maybe a bit cleaner to implement Display for pci::Class?

OK, will do.

> 
>> +///     Ok(())
>> +/// }
>> +/// ```
>> +#[derive(Debug, Clone, Copy, PartialEq, Eq)]
>> +#[repr(transparent)]
>> +pub struct Class(u32);
> 
> [ Class impl and lots of pci class ids... ]
> 
> I think we should move all this to a new Rust module (rust/kernel/pci/class.rs)
> to keep this file reasonably small.
> 
> You can add
> 
> 	use self::class::Class;
> 	use self::class::ClassMask;
> 
> in this file to make it appear as e.g. kernel::pci::Class.

OK. In patch #3 you suggested combining with Vendor into an id.rs,
which seems like a good layout and naming system, I'll do that.

> 
> Sorry I didn't mention this in the previous version.
> 
>>   /// An adapter for the registration of PCI drivers.
>>   pub struct Adapter<T: Driver>(T);
>>   
>> @@ -157,6 +355,23 @@ pub const fn from_class(class: u32, class_mask: u32) -> Self {
>>               override_only: 0,
>>           })
>>       }
>> +
>> +    /// Create a new `pci::DeviceId` from a class number, mask, and specific vendor.
>> +    ///
>> +    /// This is more targeted than [`DeviceId::from_class`]: in addition to matching by Vendor, it
>> +    /// also matches the PCI Class (up to the entire 24 bits, depending on the mask).
>> +    pub const fn from_class_and_vendor(class: Class, class_mask: u32, vendor: u32) -> Self {
> 
> I think it would be good if class_mask would be a new type ClassMask that only
> has the constants that are applicable for this field, i.e. MASK_FULL and
> MASK_CLASS_SUBCLASS.

Yes, good idea, that will lock it down. We only ever want those two
mask choices here.

> 
>> +        Self(bindings::pci_device_id {
>> +            vendor,
>> +            device: DeviceId::PCI_ANY_ID,
>> +            subvendor: DeviceId::PCI_ANY_ID,
>> +            subdevice: DeviceId::PCI_ANY_ID,
>> +            class: class.as_u32(),
>> +            class_mask,
>> +            driver_data: 0,
>> +            override_only: 0,
>> +        })
>> +    }
>>   }
>>   
>>   // SAFETY: `DeviceId` is a `#[repr(transparent)]` wrapper of `pci_device_id` and does not add
>> @@ -410,6 +625,18 @@ pub fn resource_len(&self, bar: u32) -> Result<bindings::resource_size_t> {
>>           // - by its type invariant `self.as_raw` is always a valid pointer to a `struct pci_dev`.
>>           Ok(unsafe { bindings::pci_resource_len(self.as_raw(), bar.try_into()?) })
>>       }
>> +
>> +    /// Returns the full 24-bit PCI class code as stored in hardware.
>> +    /// This includes base class, subclass, and programming interface.
>> +    pub fn pci_class_code_raw(&self) -> u32 {
>> +        // SAFETY: `self.as_raw` is a valid pointer to a `struct pci_dev`.
>> +        unsafe { (*self.as_raw()).class }
>> +    }
> 
> Do we need this method? I think drivers can just call pdev.pci_class().as_u32()
> instead (which we could also name as_raw()).

Sounds good.

> 
>> +    /// Returns the PCI class as a `Class` struct.
>> +    pub fn pci_class(&self) -> Class {
>> +        Class(self.pci_class_code_raw())
>> +    }
> 
> This is good! At a first glance the name looks a bit odd or redundant, but
> people would clearly expect something different when this is called as
> pdev.class() (i.e. a struct class representation).


Thanks, that's exactly the reasoning I used, too.


thanks,
-- 
John Hubbard