Initial import to v1a, does not compile

5 files changed, 928 insertions, 0 deletions

diff --git a/embassy-stm32/src/eth/mod.rs b/embassy-stm32/src/eth/mod.rs
index 1e304b789..ef5424e59 100644
--- a/embassy-stm32/src/eth/mod.rs
+++ b/embassy-stm32/src/eth/mod.rs
@@ -1,5 +1,6 @@
 #![macro_use]
 
+#[cfg_attr(eth_v1a, path = "v1a/mod.rs")]
 #[cfg_attr(eth_v1c, path = "v1c/mod.rs")]
 #[cfg_attr(eth_v2, path = "v2/mod.rs")]
 #[cfg_attr(eth_v1, path = "v1.rs")]
diff --git a/embassy-stm32/src/eth/v1a/descriptors.rs b/embassy-stm32/src/eth/v1a/descriptors.rs
new file mode 100644
index 000000000..25f21ce19
--- /dev/null
+++ b/embassy-stm32/src/eth/v1a/descriptors.rs
@@ -0,0 +1,21 @@
+use crate::eth::_version::rx_desc::RDesRing;
+use crate::eth::_version::tx_desc::TDesRing;
+
+pub struct DescriptorRing<const T: usize, const R: usize> {
+    pub(crate) tx: TDesRing<T>,
+    pub(crate) rx: RDesRing<R>,
+}
+
+impl<const T: usize, const R: usize> DescriptorRing<T, R> {
+    pub const fn new() -> Self {
+        Self {
+            tx: TDesRing::new(),
+            rx: RDesRing::new(),
+        }
+    }
+
+    pub fn init(&mut self) {
+        self.tx.init();
+        self.rx.init();
+    }
+}
diff --git a/embassy-stm32/src/eth/v1a/mod.rs b/embassy-stm32/src/eth/v1a/mod.rs
new file mode 100644
index 000000000..8abe2e172
--- /dev/null
+++ b/embassy-stm32/src/eth/v1a/mod.rs
@@ -0,0 +1,359 @@
+// The v1c ethernet driver was ported to embassy from the awesome stm32-eth project (https://github.com/stm32-rs/stm32-eth).
+
+use core::marker::PhantomData;
+use core::sync::atomic::{fence, Ordering};
+use core::task::Waker;
+
+use embassy::util::Unborrow;
+use embassy::waitqueue::AtomicWaker;
+use embassy_hal_common::peripheral::{PeripheralMutex, PeripheralState, StateStorage};
+use embassy_hal_common::unborrow;
+use embassy_net::{Device, DeviceCapabilities, LinkState, PacketBuf, MTU};
+
+use crate::gpio::sealed::Pin as __GpioPin;
+use crate::gpio::{sealed::AFType, AnyPin, Speed};
+use crate::pac::{ETH, RCC, SYSCFG};
+
+mod descriptors;
+mod rx_desc;
+mod tx_desc;
+
+use super::*;
+use descriptors::DescriptorRing;
+use stm32_metapac::eth::vals::{
+    Apcs, Cr, Dm, DmaomrSr, Fes, Ftf, Ifg, MbProgress, Mw, Pbl, Rsf, St, Tsf,
+};
+
+pub struct State<'d, T: Instance, const TX: usize, const RX: usize>(
+    StateStorage<Inner<'d, T, TX, RX>>,
+);
+impl<'d, T: Instance, const TX: usize, const RX: usize> State<'d, T, TX, RX> {
+    pub fn new() -> Self {
+        Self(StateStorage::new())
+    }
+}
+
+pub struct Ethernet<'d, T: Instance, P: PHY, const TX: usize, const RX: usize> {
+    state: PeripheralMutex<'d, Inner<'d, T, TX, RX>>,
+    pins: [AnyPin; 9],
+    _phy: P,
+    clock_range: Cr,
+    phy_addr: u8,
+    mac_addr: [u8; 6],
+}
+
+macro_rules! config_pins {
+    ($($pin:ident),*) => {
+        // NOTE(unsafe) Exclusive access to the registers
+        critical_section::with(|_| {
+            $(
+                $pin.set_as_af($pin.af_num(), AFType::OutputPushPull);
+                $pin.set_speed(Speed::VeryHigh);
+            )*
+        })
+    };
+}
+
+impl<'d, T: Instance, P: PHY, const TX: usize, const RX: usize> Ethernet<'d, T, P, TX, RX> {
+    /// safety: the returned instance is not leak-safe
+    pub unsafe fn new(
+        state: &'d mut State<'d, T, TX, RX>,
+        peri: impl Unborrow<Target = T> + 'd,
+        interrupt: impl Unborrow<Target = crate::interrupt::ETH> + 'd,
+        ref_clk: impl Unborrow<Target = impl RefClkPin<T>> + 'd,
+        mdio: impl Unborrow<Target = impl MDIOPin<T>> + 'd,
+        mdc: impl Unborrow<Target = impl MDCPin<T>> + 'd,
+        crs: impl Unborrow<Target = impl CRSPin<T>> + 'd,
+        rx_d0: impl Unborrow<Target = impl RXD0Pin<T>> + 'd,
+        rx_d1: impl Unborrow<Target = impl RXD1Pin<T>> + 'd,
+        tx_d0: impl Unborrow<Target = impl TXD0Pin<T>> + 'd,
+        tx_d1: impl Unborrow<Target = impl TXD1Pin<T>> + 'd,
+        tx_en: impl Unborrow<Target = impl TXEnPin<T>> + 'd,
+        phy: P,
+        mac_addr: [u8; 6],
+        phy_addr: u8,
+    ) -> Self {
+        unborrow!(interrupt, ref_clk, mdio, mdc, crs, rx_d0, rx_d1, tx_d0, tx_d1, tx_en);
+
+        // Enable the necessary Clocks
+        // NOTE(unsafe) We have exclusive access to the registers
+        critical_section::with(|_| {
+            RCC.apb2enr().modify(|w| w.set_syscfgen(true));
+            RCC.ahb1enr().modify(|w| {
+                w.set_ethen(true);
+                w.set_ethtxen(true);
+                w.set_ethrxen(true);
+            });
+
+            // RMII (Reduced Media Independent Interface)
+            SYSCFG.pmc().modify(|w| w.set_mii_rmii_sel(true));
+        });
+
+        config_pins!(ref_clk, mdio, mdc, crs, rx_d0, rx_d1, tx_d0, tx_d1, tx_en);
+
+        // NOTE(unsafe) We are ourselves not leak-safe.
+        let state = PeripheralMutex::new_unchecked(interrupt, &mut state.0, || Inner::new(peri));
+
+        // NOTE(unsafe) We have exclusive access to the registers
+        let dma = ETH.ethernet_dma();
+        let mac = ETH.ethernet_mac();
+
+        // Reset and wait
+        dma.dmabmr().modify(|w| w.set_sr(true));
+        while dma.dmabmr().read().sr() {}
+
+        mac.maccr().modify(|w| {
+            w.set_ifg(Ifg::IFG96); // inter frame gap 96 bit times
+            w.set_apcs(Apcs::STRIP); // automatic padding and crc stripping
+            w.set_fes(Fes::FES100); // fast ethernet speed
+            w.set_dm(Dm::FULLDUPLEX); // full duplex
+                                      // TODO: Carrier sense ? ECRSFD
+        });
+
+        // Note: Writing to LR triggers synchronisation of both LR and HR into the MAC core,
+        // so the LR write must happen after the HR write.
+        mac.maca0hr()
+            .modify(|w| w.set_maca0h(u16::from(mac_addr[4]) | (u16::from(mac_addr[5]) << 8)));
+        mac.maca0lr().write(|w| {
+            w.set_maca0l(
+                u32::from(mac_addr[0])
+                    | (u32::from(mac_addr[1]) << 8)
+                    | (u32::from(mac_addr[2]) << 16)
+                    | (u32::from(mac_addr[3]) << 24),
+            )
+        });
+
+        // pause time
+        mac.macfcr().modify(|w| w.set_pt(0x100));
+
+        // Transfer and Forward, Receive and Forward
+        dma.dmaomr().modify(|w| {
+            w.set_tsf(Tsf::STOREFORWARD);
+            w.set_rsf(Rsf::STOREFORWARD);
+        });
+
+        dma.dmabmr().modify(|w| {
+            w.set_pbl(Pbl::PBL32) // programmable burst length - 32 ?
+        });
+
+        // TODO MTU size setting not found for v1 ethernet, check if correct
+
+        // NOTE(unsafe) We got the peripheral singleton, which means that `rcc::init` was called
+        let hclk = crate::rcc::get_freqs().ahb1;
+        let hclk_mhz = hclk.0 / 1_000_000;
+
+        // Set the MDC clock frequency in the range 1MHz - 2.5MHz
+        let clock_range = match hclk_mhz {
+            0..=24 => panic!("Invalid HCLK frequency - should be at least 25 MHz."),
+            25..=34 => Cr::CR_20_35,     // Divide by 16
+            35..=59 => Cr::CR_35_60,     // Divide by 26
+            60..=99 => Cr::CR_60_100,    // Divide by 42
+            100..=149 => Cr::CR_100_150, // Divide by 62
+            150..=216 => Cr::CR_150_168, // Divide by 102
+            _ => {
+                panic!("HCLK results in MDC clock > 2.5MHz even for the highest CSR clock divider")
+            }
+        };
+
+        let pins = [
+            ref_clk.degrade(),
+            mdio.degrade(),
+            mdc.degrade(),
+            crs.degrade(),
+            rx_d0.degrade(),
+            rx_d1.degrade(),
+            tx_d0.degrade(),
+            tx_d1.degrade(),
+            tx_en.degrade(),
+        ];
+
+        let mut this = Self {
+            state,
+            pins,
+            _phy: phy,
+            clock_range,
+            phy_addr,
+            mac_addr,
+        };
+
+        this.state.with(|s| {
+            s.desc_ring.init();
+
+            fence(Ordering::SeqCst);
+
+            let mac = ETH.ethernet_mac();
+            let dma = ETH.ethernet_dma();
+
+            mac.maccr().modify(|w| {
+                w.set_re(true);
+                w.set_te(true);
+            });
+            dma.dmaomr().modify(|w| {
+                w.set_ftf(Ftf::FLUSH); // flush transmit fifo (queue)
+                w.set_st(St::STARTED); // start transmitting channel
+                w.set_sr(DmaomrSr::STARTED); // start receiving channel
+            });
+
+            // Enable interrupts
+            dma.dmaier().modify(|w| {
+                w.set_nise(true);
+                w.set_rie(true);
+                w.set_tie(true);
+            });
+        });
+        P::phy_reset(&mut this);
+        P::phy_init(&mut this);
+
+        this
+    }
+}
+
+unsafe impl<'d, T: Instance, P: PHY, const TX: usize, const RX: usize> StationManagement
+    for Ethernet<'d, T, P, TX, RX>
+{
+    fn smi_read(&mut self, reg: u8) -> u16 {
+        // NOTE(unsafe) These registers aren't used in the interrupt and we have `&mut self`
+        unsafe {
+            let mac = ETH.ethernet_mac();
+
+            mac.macmiiar().modify(|w| {
+                w.set_pa(self.phy_addr);
+                w.set_mr(reg);
+                w.set_mw(Mw::READ); // read operation
+                w.set_cr(self.clock_range);
+                w.set_mb(MbProgress::BUSY); // indicate that operation is in progress
+            });
+            while mac.macmiiar().read().mb() == MbProgress::BUSY {}
+            mac.macmiidr().read().md()
+        }
+    }
+
+    fn smi_write(&mut self, reg: u8, val: u16) {
+        // NOTE(unsafe) These registers aren't used in the interrupt and we have `&mut self`
+        unsafe {
+            let mac = ETH.ethernet_mac();
+
+            mac.macmiidr().write(|w| w.set_md(val));
+            mac.macmiiar().modify(|w| {
+                w.set_pa(self.phy_addr);
+                w.set_mr(reg);
+                w.set_mw(Mw::WRITE); // write
+                w.set_cr(self.clock_range);
+                w.set_mb(MbProgress::BUSY);
+            });
+            while mac.macmiiar().read().mb() == MbProgress::BUSY {}
+        }
+    }
+}
+
+impl<'d, T: Instance, P: PHY, const TX: usize, const RX: usize> Device
+    for Ethernet<'d, T, P, TX, RX>
+{
+    fn is_transmit_ready(&mut self) -> bool {
+        self.state.with(|s| s.desc_ring.tx.available())
+    }
+
+    fn transmit(&mut self, pkt: PacketBuf) {
+        self.state.with(|s| unwrap!(s.desc_ring.tx.transmit(pkt)));
+    }
+
+    fn receive(&mut self) -> Option<PacketBuf> {
+        self.state.with(|s| s.desc_ring.rx.pop_packet())
+    }
+
+    fn register_waker(&mut self, waker: &Waker) {
+        WAKER.register(waker);
+    }
+
+    fn capabilities(&mut self) -> DeviceCapabilities {
+        let mut caps = DeviceCapabilities::default();
+        caps.max_transmission_unit = MTU;
+        caps.max_burst_size = Some(TX.min(RX));
+        caps
+    }
+
+    fn link_state(&mut self) -> LinkState {
+        if P::poll_link(self) {
+            LinkState::Up
+        } else {
+            LinkState::Down
+        }
+    }
+
+    fn ethernet_address(&mut self) -> [u8; 6] {
+        self.mac_addr
+    }
+}
+
+impl<'d, T: Instance, P: PHY, const TX: usize, const RX: usize> Drop
+    for Ethernet<'d, T, P, TX, RX>
+{
+    fn drop(&mut self) {
+        // NOTE(unsafe) We have `&mut self` and the interrupt doesn't use this registers
+        unsafe {
+            let dma = ETH.ethernet_dma();
+            let mac = ETH.ethernet_mac();
+
+            // Disable the TX DMA and wait for any previous transmissions to be completed
+            dma.dmaomr().modify(|w| w.set_st(St::STOPPED));
+
+            // Disable MAC transmitter and receiver
+            mac.maccr().modify(|w| {
+                w.set_re(false);
+                w.set_te(false);
+            });
+
+            dma.dmaomr().modify(|w| w.set_sr(DmaomrSr::STOPPED));
+        }
+
+        // NOTE(unsafe) Exclusive access to the regs
+        critical_section::with(|_| unsafe {
+            for pin in self.pins.iter_mut() {
+                pin.set_as_disconnected();
+            }
+        })
+    }
+}
+
+//----------------------------------------------------------------------
+
+struct Inner<'d, T: Instance, const TX: usize, const RX: usize> {
+    _peri: PhantomData<&'d mut T>,
+    desc_ring: DescriptorRing<TX, RX>,
+}
+
+impl<'d, T: Instance, const TX: usize, const RX: usize> Inner<'d, T, TX, RX> {
+    pub fn new(_peri: impl Unborrow<Target = T> + 'd) -> Self {
+        Self {
+            _peri: PhantomData,
+            desc_ring: DescriptorRing::new(),
+        }
+    }
+}
+
+impl<'d, T: Instance, const TX: usize, const RX: usize> PeripheralState for Inner<'d, T, TX, RX> {
+    type Interrupt = crate::interrupt::ETH;
+
+    fn on_interrupt(&mut self) {
+        unwrap!(self.desc_ring.tx.on_interrupt());
+        self.desc_ring.rx.on_interrupt();
+
+        WAKER.wake();
+
+        // TODO: Check and clear more flags
+        unsafe {
+            let dma = ETH.ethernet_dma();
+
+            dma.dmasr().modify(|w| {
+                w.set_ts(true);
+                w.set_rs(true);
+                w.set_nis(true);
+            });
+            // Delay two peripheral's clock
+            dma.dmasr().read();
+            dma.dmasr().read();
+        }
+    }
+}
+
+static WAKER: AtomicWaker = AtomicWaker::new();
diff --git a/embassy-stm32/src/eth/v1a/rx_desc.rs b/embassy-stm32/src/eth/v1a/rx_desc.rs
new file mode 100644
index 000000000..6164f2975
--- /dev/null
+++ b/embassy-stm32/src/eth/v1a/rx_desc.rs
@@ -0,0 +1,309 @@
+use core::sync::atomic::{compiler_fence, fence, Ordering};
+
+use embassy_net::{Packet, PacketBox, PacketBoxExt, PacketBuf};
+use stm32_metapac::eth::vals::{DmaomrSr, Rpd, Rps};
+use vcell::VolatileCell;
+
+use crate::pac::ETH;
+
+mod rx_consts {
+    /// Owned by DMA engine
+    pub const RXDESC_0_OWN: u32 = 1 << 31;
+    /// First descriptor
+    pub const RXDESC_0_FS: u32 = 1 << 9;
+    /// Last descriptor
+    pub const RXDESC_0_LS: u32 = 1 << 8;
+    /// Error summary
+    pub const RXDESC_0_ES: u32 = 1 << 15;
+    /// Frame length
+    pub const RXDESC_0_FL_MASK: u32 = 0x3FFF;
+    pub const RXDESC_0_FL_SHIFT: usize = 16;
+
+    pub const RXDESC_1_RBS_MASK: u32 = 0x0fff;
+    /// Second address chained
+    pub const RXDESC_1_RCH: u32 = 1 << 14;
+    /// End Of Ring
+    pub const RXDESC_1_RER: u32 = 1 << 15;
+}
+
+use rx_consts::*;
+
+/// Receive Descriptor representation
+///
+/// * rdes0: OWN and Status
+/// * rdes1: allocated buffer length
+/// * rdes2: data buffer address
+/// * rdes3: next descriptor address
+#[repr(C)]
+struct RDes {
+    rdes0: VolatileCell<u32>,
+    rdes1: VolatileCell<u32>,
+    rdes2: VolatileCell<u32>,
+    rdes3: VolatileCell<u32>,
+}
+
+impl RDes {
+    pub const fn new() -> Self {
+        Self {
+            rdes0: VolatileCell::new(0),
+            rdes1: VolatileCell::new(0),
+            rdes2: VolatileCell::new(0),
+            rdes3: VolatileCell::new(0),
+        }
+    }
+
+    /// Return true if this RDes is acceptable to us
+    #[inline(always)]
+    pub fn valid(&self) -> bool {
+        // Write-back descriptor is valid if:
+        //
+        // Contains first buffer of packet AND contains last buf of
+        // packet AND no errors
+        (self.rdes0.get() & (RXDESC_0_ES | RXDESC_0_FS | RXDESC_0_LS))
+            == (RXDESC_0_FS | RXDESC_0_LS)
+    }
+
+    /// Return true if this RDes is not currently owned by the DMA
+    #[inline(always)]
+    pub fn available(&self) -> bool {
+        self.rdes0.get() & RXDESC_0_OWN == 0 // Owned by us
+    }
+
+    /// Configures the reception buffer address and length and passed descriptor ownership to the DMA
+    #[inline(always)]
+    pub fn set_ready(&mut self, buf_addr: u32, buf_len: usize) {
+        self.rdes1
+            .set(self.rdes1.get() | (buf_len as u32) & RXDESC_1_RBS_MASK);
+        self.rdes2.set(buf_addr);
+
+        // "Preceding reads and writes cannot be moved past subsequent writes."
+        fence(Ordering::Release);
+
+        compiler_fence(Ordering::Release);
+
+        self.rdes0.set(self.rdes0.get() | RXDESC_0_OWN);
+
+        // Used to flush the store buffer as fast as possible to make the buffer available for the
+        // DMA.
+        fence(Ordering::SeqCst);
+    }
+
+    // points to next descriptor (RCH)
+    #[inline(always)]
+    fn set_buffer2(&mut self, buffer: *const u8) {
+        self.rdes3.set(buffer as u32);
+    }
+
+    #[inline(always)]
+    fn set_end_of_ring(&mut self) {
+        self.rdes1.set(self.rdes1.get() | RXDESC_1_RER);
+    }
+
+    #[inline(always)]
+    fn packet_len(&self) -> usize {
+        ((self.rdes0.get() >> RXDESC_0_FL_SHIFT) & RXDESC_0_FL_MASK) as usize
+    }
+
+    pub fn setup(&mut self, next: Option<&Self>) {
+        // Defer this initialization to this function, so we can have `RingEntry` on bss.
+        self.rdes1.set(self.rdes1.get() | RXDESC_1_RCH);
+
+        match next {
+            Some(next) => self.set_buffer2(next as *const _ as *const u8),
+            None => {
+                self.set_buffer2(0 as *const u8);
+                self.set_end_of_ring();
+            }
+        }
+    }
+}
+/// Running state of the `RxRing`
+#[derive(PartialEq, Eq, Debug)]
+pub enum RunningState {
+    Unknown,
+    Stopped,
+    Running,
+}
+
+impl RunningState {
+    /// whether self equals to `RunningState::Running`
+    pub fn is_running(&self) -> bool {
+        *self == RunningState::Running
+    }
+}
+
+/// Rx ring of descriptors and packets
+///
+/// This ring has three major locations that work in lock-step. The DMA will never write to the tail
+/// index, so the `read_index` must never pass the tail index. The `next_tail_index` is always 1
+/// slot ahead of the real tail index, and it must never pass the `read_index` or it could overwrite
+/// a packet still to be passed to the application.
+///
+///                                                                   nt can't pass r (no alloc)
+/// +---+---+---+---+  Read ok       +---+---+---+---+ No Read       +---+---+---+---+
+/// |   |   |   |   |  ------------> |   |   |   |   | ------------> |   |   |   |   |
+/// +---+---+---+---+  Allocation ok +---+---+---+---+               +---+---+---+---+
+///   ^           ^t                   ^t  ^                           ^t  ^
+///   |r                                   |r                              |r
+///   |nt                                  |nt                             |nt
+///
+///
+/// +---+---+---+---+  Read ok         +---+---+---+---+ Can't read    +---+---+---+---+
+/// |   |   |   |   |  ------------>   |   |   |   |   | ------------> |   |   |   |   |
+/// +---+---+---+---+  Allocation fail +---+---+---+---+ Allocation ok +---+---+---+---+
+///       ^   ^t  ^                              ^t  ^                   ^       ^   ^t
+///       |r      |                              |r  |                   |       |r
+///               |nt                                |nt                 |nt
+///
+pub(crate) struct RDesRing<const N: usize> {
+    descriptors: [RDes; N],
+    buffers: [Option<PacketBox>; N],
+    read_index: usize,
+    next_tail_index: usize,
+}
+
+impl<const N: usize> RDesRing<N> {
+    pub const fn new() -> Self {
+        const RDES: RDes = RDes::new();
+        const BUFFERS: Option<PacketBox> = None;
+
+        Self {
+            descriptors: [RDES; N],
+            buffers: [BUFFERS; N],
+            read_index: 0,
+            next_tail_index: 0,
+        }
+    }
+
+    pub(crate) fn init(&mut self) {
+        assert!(N > 1);
+        let mut last_index = 0;
+        for (index, buf) in self.buffers.iter_mut().enumerate() {
+            let pkt = match PacketBox::new(Packet::new()) {
+                Some(p) => p,
+                None => {
+                    if index == 0 {
+                        panic!("Could not allocate at least one buffer for Ethernet receiving");
+                    } else {
+                        break;
+                    }
+                }
+            };
+            self.descriptors[index].set_ready(pkt.as_ptr() as u32, pkt.len());
+            *buf = Some(pkt);
+            last_index = index;
+        }
+        self.next_tail_index = (last_index + 1) % N;
+
+        // not sure if this is supposed to span all of the descriptor or just those that contain buffers
+        {
+            let mut previous: Option<&mut RDes> = None;
+            for entry in self.descriptors.iter_mut() {
+                if let Some(prev) = &mut previous {
+                    prev.setup(Some(entry));
+                }
+                previous = Some(entry);
+            }
+
+            if let Some(entry) = &mut previous {
+                entry.setup(None);
+            }
+        }
+
+        // Register txdescriptor start
+        // NOTE (unsafe) Used for atomic writes
+        unsafe {
+            ETH.ethernet_dma()
+                .dmardlar()
+                .write(|w| w.0 = &self.descriptors as *const _ as u32);
+        };
+        // We already have fences in `set_owned`, which is called in `setup`
+
+        // Start receive
+        unsafe {
+            ETH.ethernet_dma()
+                .dmaomr()
+                .modify(|w| w.set_sr(DmaomrSr::STARTED))
+        };
+
+        self.demand_poll();
+    }
+
+    fn demand_poll(&self) {
+        unsafe { ETH.ethernet_dma().dmarpdr().write(|w| w.set_rpd(Rpd::POLL)) };
+    }
+
+    pub(crate) fn on_interrupt(&mut self) {
+        // XXX: Do we need to do anything here ? Maybe we should try to advance the tail ptr, but it
+        // would soon hit the read ptr anyway, and we will wake smoltcp's stack on the interrupt
+        // which should try to pop a packet...
+    }
+
+    /// Get current `RunningState`
+    fn running_state(&self) -> RunningState {
+        match unsafe { ETH.ethernet_dma().dmasr().read().rps() } {
+            //  Reset or Stop Receive Command issued
+            Rps::STOPPED => RunningState::Stopped,
+            //  Fetching receive transfer descriptor
+            Rps::RUNNINGFETCHING => RunningState::Running,
+            //  Waiting for receive packet
+            Rps::RUNNINGWAITING => RunningState::Running,
+            //  Receive descriptor unavailable
+            Rps::SUSPENDED => RunningState::Stopped,
+            //  Closing receive descriptor
+            Rps(0b101) => RunningState::Running,
+            //  Transferring the receive packet data from receive buffer to host memory
+            Rps::RUNNINGWRITING => RunningState::Running,
+            _ => RunningState::Unknown,
+        }
+    }
+
+    pub(crate) fn pop_packet(&mut self) -> Option<PacketBuf> {
+        if !self.running_state().is_running() {
+            self.demand_poll();
+        }
+        // Not sure if the contents of the write buffer on the M7 can affects reads, so we are using
+        // a DMB here just in case, it also serves as a hint to the compiler that we're syncing the
+        // buffer (I think .-.)
+        fence(Ordering::SeqCst);
+
+        let read_available = self.descriptors[self.read_index].available();
+        let tail_index = (self.next_tail_index + N - 1) % N;
+
+        let pkt = if read_available && self.read_index != tail_index {
+            let pkt = self.buffers[self.read_index].take();
+            let len = self.descriptors[self.read_index].packet_len();
+
+            assert!(pkt.is_some());
+            let valid = self.descriptors[self.read_index].valid();
+
+            self.read_index = (self.read_index + 1) % N;
+            if valid {
+                pkt.map(|p| p.slice(0..len))
+            } else {
+                None
+            }
+        } else {
+            None
+        };
+
+        // Try to advance the tail_index
+        if self.next_tail_index != self.read_index {
+            match PacketBox::new(Packet::new()) {
+                Some(b) => {
+                    let addr = b.as_ptr() as u32;
+                    let buffer_len = b.len();
+                    self.buffers[self.next_tail_index].replace(b);
+                    self.descriptors[self.next_tail_index].set_ready(addr, buffer_len);
+
+                    // "Preceding reads and writes cannot be moved past subsequent writes."
+                    fence(Ordering::Release);
+
+                    self.next_tail_index = (self.next_tail_index + 1) % N;
+                }
+                None => {}
+            }
+        }
+        pkt
+    }
+}
diff --git a/embassy-stm32/src/eth/v1a/tx_desc.rs b/embassy-stm32/src/eth/v1a/tx_desc.rs
new file mode 100644
index 000000000..f253ab19a
--- /dev/null
+++ b/embassy-stm32/src/eth/v1a/tx_desc.rs
@@ -0,0 +1,238 @@
+use core::sync::atomic::{compiler_fence, fence, Ordering};
+
+use embassy_net::PacketBuf;
+use stm32_metapac::eth::vals::St;
+use vcell::VolatileCell;
+
+use crate::pac::ETH;
+
+#[non_exhaustive]
+#[derive(Debug, Copy, Clone)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum Error {
+    NoBufferAvailable,
+    // TODO: Break down this error into several others
+    TransmissionError,
+}
+
+/// Transmit and Receive Descriptor fields
+#[allow(dead_code)]
+mod tx_consts {
+    pub const TXDESC_0_OWN: u32 = 1 << 31;
+    pub const TXDESC_0_IOC: u32 = 1 << 30;
+    // First segment of frame
+    pub const TXDESC_0_FS: u32 = 1 << 28;
+    // Last segment of frame
+    pub const TXDESC_0_LS: u32 = 1 << 29;
+    // Transmit end of ring
+    pub const TXDESC_0_TER: u32 = 1 << 21;
+    // Second address chained
+    pub const TXDESC_0_TCH: u32 = 1 << 20;
+    // Error status
+    pub const TXDESC_0_ES: u32 = 1 << 15;
+
+    // Transmit buffer size
+    pub const TXDESC_1_TBS_SHIFT: usize = 0;
+    pub const TXDESC_1_TBS_MASK: u32 = 0x0fff << TXDESC_1_TBS_SHIFT;
+}
+use tx_consts::*;
+
+/// Transmit Descriptor representation
+///
+/// * tdes0: control
+/// * tdes1: buffer lengths
+/// * tdes2: data buffer address
+/// * tdes3: next descriptor address
+#[repr(C)]
+struct TDes {
+    tdes0: VolatileCell<u32>,
+    tdes1: VolatileCell<u32>,
+    tdes2: VolatileCell<u32>,
+    tdes3: VolatileCell<u32>,
+}
+
+impl TDes {
+    pub const fn new() -> Self {
+        Self {
+            tdes0: VolatileCell::new(0),
+            tdes1: VolatileCell::new(0),
+            tdes2: VolatileCell::new(0),
+            tdes3: VolatileCell::new(0),
+        }
+    }
+
+    /// Return true if this TDes is not currently owned by the DMA
+    pub fn available(&self) -> bool {
+        (self.tdes0.get() & TXDESC_0_OWN) == 0
+    }
+
+    /// Pass ownership to the DMA engine
+    fn set_owned(&mut self) {
+        // "Preceding reads and writes cannot be moved past subsequent writes."
+        fence(Ordering::Release);
+
+        compiler_fence(Ordering::Release);
+        self.tdes0.set(self.tdes0.get() | TXDESC_0_OWN);
+
+        // Used to flush the store buffer as fast as possible to make the buffer available for the
+        // DMA.
+        fence(Ordering::SeqCst);
+    }
+
+    fn set_buffer1(&mut self, buffer: *const u8) {
+        self.tdes2.set(buffer as u32);
+    }
+
+    fn set_buffer1_len(&mut self, len: usize) {
+        self.tdes1
+            .set((self.tdes1.get() & !TXDESC_1_TBS_MASK) | ((len as u32) << TXDESC_1_TBS_SHIFT));
+    }
+
+    // points to next descriptor (RCH)
+    fn set_buffer2(&mut self, buffer: *const u8) {
+        self.tdes3.set(buffer as u32);
+    }
+
+    fn set_end_of_ring(&mut self) {
+        self.tdes0.set(self.tdes0.get() | TXDESC_0_TER);
+    }
+
+    // set up as a part fo the ring buffer - configures the tdes
+    pub fn setup(&mut self, next: Option<&Self>) {
+        // Defer this initialization to this function, so we can have `RingEntry` on bss.
+        self.tdes0
+            .set(TXDESC_0_TCH | TXDESC_0_IOC | TXDESC_0_FS | TXDESC_0_LS);
+        match next {
+            Some(next) => self.set_buffer2(next as *const TDes as *const u8),
+            None => {
+                self.set_buffer2(0 as *const u8);
+                self.set_end_of_ring();
+            }
+        }
+    }
+}
+
+pub(crate) struct TDesRing<const N: usize> {
+    descriptors: [TDes; N],
+    buffers: [Option<PacketBuf>; N],
+    next_entry: usize,
+}
+
+impl<const N: usize> TDesRing<N> {
+    pub const fn new() -> Self {
+        const TDES: TDes = TDes::new();
+        const BUFFERS: Option<PacketBuf> = None;
+
+        Self {
+            descriptors: [TDES; N],
+            buffers: [BUFFERS; N],
+            next_entry: 0,
+        }
+    }
+
+    /// Initialise this TDesRing. Assume TDesRing is corrupt
+    ///
+    /// The current memory address of the buffers inside this TDesRing
+    /// will be stored in the descriptors, so ensure the TDesRing is
+    /// not moved after initialisation.
+    pub(crate) fn init(&mut self) {
+        assert!(N > 0);
+
+        {
+            let mut previous: Option<&mut TDes> = None;
+            for entry in self.descriptors.iter_mut() {
+                if let Some(prev) = &mut previous {
+                    prev.setup(Some(entry));
+                }
+                previous = Some(entry);
+            }
+
+            if let Some(entry) = &mut previous {
+                entry.setup(None);
+            }
+        }
+        self.next_entry = 0;
+
+        // Register txdescriptor start
+        // NOTE (unsafe) Used for atomic writes
+        unsafe {
+            ETH.ethernet_dma()
+                .dmatdlar()
+                .write(|w| w.0 = &self.descriptors as *const _ as u32);
+        }
+
+        // "Preceding reads and writes cannot be moved past subsequent writes."
+        #[cfg(feature = "fence")]
+        fence(Ordering::Release);
+
+        // We don't need a compiler fence here because all interactions with `Descriptor` are
+        // volatiles
+
+        // Start transmission
+        unsafe {
+            ETH.ethernet_dma()
+                .dmaomr()
+                .modify(|w| w.set_st(St::STARTED))
+        };
+    }
+
+    /// Return true if a TDes is available for use
+    pub(crate) fn available(&self) -> bool {
+        self.descriptors[self.next_entry].available()
+    }
+
+    pub(crate) fn transmit(&mut self, pkt: PacketBuf) -> Result<(), Error> {
+        if !self.available() {
+            return Err(Error::NoBufferAvailable);
+        }
+
+        let descriptor = &mut self.descriptors[self.next_entry];
+
+        let pkt_len = pkt.len();
+        let address = pkt.as_ptr() as *const u8;
+
+        descriptor.set_buffer1(address);
+        descriptor.set_buffer1_len(pkt_len);
+
+        self.buffers[self.next_entry].replace(pkt);
+
+        descriptor.set_owned();
+
+        // Ensure changes to the descriptor are committed before DMA engine sees tail pointer store.
+        // This will generate an DMB instruction.
+        // "Preceding reads and writes cannot be moved past subsequent writes."
+        fence(Ordering::Release);
+
+        // Move the tail pointer (TPR) to the next descriptor
+        self.next_entry = (self.next_entry + 1) % N;
+
+        // Request the DMA engine to poll the latest tx descriptor
+        unsafe { ETH.ethernet_dma().dmatpdr().modify(|w| w.0 = 1) }
+        Ok(())
+    }
+
+    pub(crate) fn on_interrupt(&mut self) -> Result<(), Error> {
+        let previous = (self.next_entry + N - 1) % N;
+        let td = &self.descriptors[previous];
+
+        // DMB to ensure that we are reading an updated value, probably not needed at the hardware
+        // level, but this is also a hint to the compiler that we're syncing on the buffer.
+        fence(Ordering::SeqCst);
+
+        let tdes0 = td.tdes0.get();
+
+        if tdes0 & TXDESC_0_OWN != 0 {
+            // Transmission isn't done yet, probably a receive interrupt that fired this
+            return Ok(());
+        }
+
+        // Release the buffer
+        self.buffers[previous].take();
+
+        if tdes0 & TXDESC_0_ES != 0 {
+            Err(Error::TransmissionError)
+        } else {
+            Ok(())
+        }
+    }
+}