Add pci I/O bar access and use bitfields

[hecatia-elegua]

• edition update/clippy as well

I added some TODOs because I'm not done with pci, will do more PRs. I actually stated a question there too: Do we / does the device author always know what kind of address a determine_mem_base call returns?
I can look that up in a pci spec, just asking if anybody knows.

[kevinaboos]

Overall, i like this PR, but am concerned that the representation of the PCI BAR is just way too complex with the multiple enums and bitfields.

Given the fact that the register layout is so dynamic (e.g., address field changes width and type based on bit 0), I think a design like this would be much simpler (pseudocode):

/// A PCI Base Address Register (BAR) ... finish docs
#[repr(transparent)]
pub struct BaseAddressRegister(u32);
impl BaseAddressRegister {
    
    /// TODO: here, add const defs for the various magic number

    /// Docs here
    pub fn get(&self) -> PciBar {
        match self.0 & 0x1 {
            0 => {
                let typ = self.0.get_bits(1..=2);
                if typ != 0x0 {
                    panic!("PCI BAR of type {typ:#X} is is unsupported")
                }
                PciBar::MemorySpace {
                    base_addr: PhysicalAddress::new_canonical((self.0 >> 4) as usize),
                    prefetchable: self.get_bit(3),
                }
            }
            1 => PciBar::IoSpace(self.0 >> 2)
        }
    }
}

/// Docs here
pub enum PciBar {
    /// Docs here
    MemorySpace {
        base_addr: PhysicalAddress,
        /// If `true`, reading from this PCI memory can be cached,
        /// and should thus be mapped as write-through, not uncacheable.
        prefetchable: bool,
    }
    /// Docs here
    IoSpace {
        addr: u32,
    }
}

and then the PciDevice struct would use the BaseAddressRegister type in its bars field.

[kevinaboos]

this type can now be deleted since you've replaced it with the two enums

[kevinaboos]

is this better? I don't necessarily think it's any clearer

[hecatia-elegua]

this is temporary since I wanted to remove the bitfield crate and this was the only place left

[kevinaboos]

To answer your question:

Do we / does the device author always know what kind of address a determine_mem_base call returns?

Yes, but we should verify that by actually checking the PCI BAR content. Your approach in the network device drivers of returning an error if it's not a memory space BAR is correct.

[hecatia-elegua]

Thanks for the comments, I'll think about this some more when I'm stuck on my main tasks.
I do think that your version is more readable, though.