forked from M-Labs/zynq-rs
1
0
Fork 0
zynq-rs/libboard_zynq/src/eth/mod.rs

621 lines
18 KiB
Rust

use core::ops::{Deref, DerefMut};
use log::{error, info, warn};
use libregister::*;
use super::slcr;
use super::clocks::Clocks;
pub mod phy;
use phy::{Phy, PhyAccess};
mod regs;
pub mod rx;
pub mod tx;
/// Size of all the buffers
pub const MTU: usize = 1536;
/// Maximum MDC clock
const MAX_MDC: u32 = 2_500_000;
const TX_10: u32 = 10_000_000;
const TX_100: u32 = 25_000_000;
/// Clock for GbE
const TX_1000: u32 = 125_000_000;
#[derive(Clone)]
#[repr(C, align(0x20))]
pub struct Buffer(pub [u8; MTU]);
impl Buffer {
pub fn new() -> Self {
Buffer([0; MTU])
}
}
impl Deref for Buffer {
type Target = [u8];
fn deref(&self) -> &Self::Target {
&self.0
}
}
impl DerefMut for Buffer {
fn deref_mut(&mut self) -> &mut <Self as Deref>::Target {
&mut self.0
}
}
pub struct Eth<'r, RX, TX> {
rx: RX,
tx: TX,
inner: EthInner<'r>,
phy: Phy,
}
impl<'r> Eth<'r, (), ()> {
pub fn default(macaddr: [u8; 6]) -> Self {
slcr::RegisterBlock::unlocked(|slcr| {
// Manual example: 0x0000_1280
// MDIO
slcr.mio_pin_53.write(
slcr::MioPin53::zeroed()
.l3_sel(0b100)
.io_type(slcr::IoBufferType::Lvcmos18)
.pullup(true)
);
// MDC
slcr.mio_pin_52.write(
slcr::MioPin52::zeroed()
.l3_sel(0b100)
.io_type(slcr::IoBufferType::Lvcmos18)
.pullup(true)
);
// Manual example: 0x0000_3902
// TX_CLK
slcr.mio_pin_16.write(
slcr::MioPin16::zeroed()
.l0_sel(true)
.speed(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
.disable_rcvr(true)
);
// TX_CTRL
slcr.mio_pin_21.write(
slcr::MioPin21::zeroed()
.l0_sel(true)
.speed(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
.disable_rcvr(true)
);
// TXD3
slcr.mio_pin_20.write(
slcr::MioPin20::zeroed()
.l0_sel(true)
.speed(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
.disable_rcvr(true)
);
// TXD2
slcr.mio_pin_19.write(
slcr::MioPin19::zeroed()
.l0_sel(true)
.speed(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
.disable_rcvr(true)
);
// TXD1
slcr.mio_pin_18.write(
slcr::MioPin18::zeroed()
.l0_sel(true)
.speed(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
.disable_rcvr(true)
);
// TXD0
slcr.mio_pin_17.write(
slcr::MioPin17::zeroed()
.l0_sel(true)
.speed(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
.disable_rcvr(true)
);
// Manual example: 0x0000_1903
// RX_CLK
slcr.mio_pin_22.write(
slcr::MioPin22::zeroed()
.tri_enable(true)
.l0_sel(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
);
// RX_CTRL
slcr.mio_pin_27.write(
slcr::MioPin27::zeroed()
.tri_enable(true)
.l0_sel(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
);
// RXD3
slcr.mio_pin_26.write(
slcr::MioPin26::zeroed()
.tri_enable(true)
.l0_sel(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
);
// RXD2
slcr.mio_pin_25.write(
slcr::MioPin25::zeroed()
.tri_enable(true)
.l0_sel(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
);
// RXD1
slcr.mio_pin_24.write(
slcr::MioPin24::zeroed()
.tri_enable(true)
.l0_sel(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
);
// RXD0
slcr.mio_pin_23.write(
slcr::MioPin23::zeroed()
.tri_enable(true)
.l0_sel(true)
.io_type(slcr::IoBufferType::Hstl)
.pullup(true)
);
// VREF internal generator
slcr.gpiob_ctrl.write(
slcr::GpiobCtrl::zeroed()
.vref_en(true)
);
});
Self::gem0(macaddr)
}
pub fn gem0(macaddr: [u8; 6]) -> Self {
Self::setup_gem0_clock(TX_1000);
let regs = regs::RegisterBlock::gem0();
Self::from_regs(regs, macaddr)
}
pub fn gem1(macaddr: [u8; 6]) -> Self {
Self::setup_gem1_clock(TX_1000);
let regs = regs::RegisterBlock::gem1();
Self::from_regs(regs, macaddr)
}
fn from_regs(regs: &'r mut regs::RegisterBlock, macaddr: [u8; 6]) -> Self {
let mut inner = EthInner {
regs,
link: None,
};
inner.init();
inner.configure(macaddr);
let phy = Phy::find(&mut inner).expect("phy");
phy.reset(&mut inner);
phy.restart_autoneg(&mut inner);
Eth {
rx: (),
tx: (),
inner,
phy,
}
}
}
impl<'r, RX, TX> Eth<'r, RX, TX> {
pub fn setup_gem0_clock(tx_clock: u32) {
let io_pll = Clocks::get().io;
let d0 = ((tx_clock - 1 + io_pll) / tx_clock).max(1).min(63);
let d1 = (io_pll / tx_clock / d0).max(1).min(63);
slcr::RegisterBlock::unlocked(|slcr| {
slcr.gem0_clk_ctrl.write(
// 0x0050_0801: 8, 5: 100 Mb/s
// ...: 8, 1: 1000 Mb/s
slcr::GemClkCtrl::zeroed()
.clkact(true)
.srcsel(slcr::PllSource::IoPll)
.divisor(d0 as u8)
.divisor1(d1 as u8)
);
// Enable gem0 recv clock
slcr.gem0_rclk_ctrl.write(
// 0x0000_0801
slcr::RclkCtrl::zeroed()
.clkact(true)
);
});
}
pub fn setup_gem1_clock(tx_clock: u32) {
let io_pll = Clocks::get().io;
let d0 = ((tx_clock - 1 + io_pll) / tx_clock).max(1).min(63);
let d1 = (io_pll / tx_clock / d0).max(1).min(63);
slcr::RegisterBlock::unlocked(|slcr| {
slcr.gem1_clk_ctrl.write(
slcr::GemClkCtrl::zeroed()
.clkact(true)
.srcsel(slcr::PllSource::IoPll)
.divisor(d0 as u8)
.divisor1(d1 as u8)
);
// Enable gem1 recv clock
slcr.gem1_rclk_ctrl.write(
// 0x0000_0801
slcr::RclkCtrl::zeroed()
.clkact(true)
);
});
}
pub fn start_rx(self, rx_size: usize) -> Eth<'r, rx::DescList, TX> {
let new_self = Eth {
rx: rx::DescList::new(rx_size),
tx: self.tx,
inner: self.inner,
phy: self.phy,
};
let list_addr = new_self.rx.list_addr();
assert!(list_addr & 0b11 == 0);
new_self.inner.regs.rx_qbar.write(
regs::RxQbar::zeroed()
.rx_q_baseaddr(list_addr >> 2)
);
new_self.inner.regs.net_ctrl.modify(|_, w|
w.rx_en(true)
);
new_self
}
pub fn start_tx(self, tx_size: usize) -> Eth<'r, RX, tx::DescList> {
let new_self = Eth {
rx: self.rx,
tx: tx::DescList::new(tx_size),
inner: self.inner,
phy: self.phy,
};
let list_addr = &new_self.tx.list_addr();
assert!(list_addr & 0b11 == 0);
new_self.inner.regs.tx_qbar.write(
regs::TxQbar::zeroed()
.tx_q_baseaddr(list_addr >> 2)
);
new_self.inner.regs.net_ctrl.modify(|_, w|
w.tx_en(true)
);
new_self
}
}
impl<'r, TX> Eth<'r, rx::DescList, TX> {
pub fn recv_next<'s: 'p, 'p>(&'s mut self) -> Result<Option<rx::PktRef<'p>>, rx::Error> {
let status = self.inner.regs.rx_status.read();
if status.hresp_not_ok() {
// Clear
self.inner.regs.rx_status.write(
regs::RxStatus::zeroed()
.hresp_not_ok(true)
);
return Err(rx::Error::HrespNotOk);
}
if status.rx_overrun() {
// Clear
self.inner.regs.rx_status.write(
regs::RxStatus::zeroed()
.rx_overrun(true)
);
return Err(rx::Error::RxOverrun);
}
if status.buffer_not_avail() {
// Clear
self.inner.regs.rx_status.write(
regs::RxStatus::zeroed()
.buffer_not_avail(true)
);
return Err(rx::Error::BufferNotAvail);
}
if status.frame_recd() {
let result = self.rx.recv_next();
match result {
Ok(None) => {
// No packet, clear status bit
self.inner.regs.rx_status.write(
regs::RxStatus::zeroed()
.frame_recd(true)
);
}
_ => {}
}
result
} else {
self.inner.check_link_change(&self.phy);
Ok(None)
}
}
}
impl<'r, RX> Eth<'r, RX, tx::DescList> {
pub fn send<'s: 'p, 'p>(&'s mut self, length: usize) -> Option<tx::PktRef<'p>> {
self.tx.send(self.inner.regs, length)
}
}
impl<'r, 'a> smoltcp::phy::Device<'a> for &mut Eth<'r, rx::DescList, tx::DescList> {
type RxToken = rx::PktRef<'a>;
type TxToken = tx::Token<'a>;
fn capabilities(&self) -> smoltcp::phy::DeviceCapabilities {
use smoltcp::phy::{DeviceCapabilities, ChecksumCapabilities, Checksum};
let mut checksum_caps = ChecksumCapabilities::default();
checksum_caps.ipv4 = Checksum::Both;
checksum_caps.tcp = Checksum::Both;
checksum_caps.udp = Checksum::Both;
let mut caps = DeviceCapabilities::default();
caps.max_transmission_unit = MTU;
caps.max_burst_size = Some(self.rx.len().min(self.tx.len()));
caps.checksum = checksum_caps;
caps
}
fn receive(&'a mut self) -> Option<(Self::RxToken, Self::TxToken)> {
match self.rx.recv_next() {
Ok(Some(pktref)) => {
let tx_token = tx::Token {
regs: self.inner.regs,
desc_list: &mut self.tx,
};
Some((pktref, tx_token))
}
Ok(None) => {
self.inner.check_link_change(&self.phy);
None
}
Err(e) => {
error!("eth recv error: {:?}", e);
None
}
}
}
fn transmit(&'a mut self) -> Option<Self::TxToken> {
Some(tx::Token {
regs: self.inner.regs,
desc_list: &mut self.tx,
})
}
}
struct EthInner<'r> {
regs: &'r mut regs::RegisterBlock,
link: Option<phy::Link>,
}
impl<'r> EthInner<'r> {
fn init(&mut self) {
// Clear the Network Control register.
self.regs.net_ctrl.write(regs::NetCtrl::zeroed());
self.regs.net_ctrl.write(regs::NetCtrl::zeroed().clear_stat_regs(true));
// Clear the Status registers.
self.regs.rx_status.write(
regs::RxStatus::zeroed()
.buffer_not_avail(true)
.frame_recd(true)
.rx_overrun(true)
.hresp_not_ok(true)
);
self.regs.tx_status.write(
regs::TxStatus::zeroed()
.used_bit_read(true)
.collision(true)
.retry_limit_exceeded(true)
.tx_go(true)
.tx_corr_ahb_err(true)
.tx_complete(true)
.tx_under_run(true)
.late_collision(true)
// not in the manual:
.hresp_not_ok(true)
);
// Disable all interrupts.
self.regs.intr_dis.write(
regs::IntrDis::zeroed()
.mgmt_done(true)
.rx_complete(true)
.rx_used_read(true)
.tx_used_read(true)
.tx_underrun(true)
.retry_ex_late_collisn(true)
.tx_corrupt_ahb_err(true)
.tx_complete(true)
.link_chng(true)
.rx_overrun(true)
.hresp_not_ok(true)
.pause_nonzeroq(true)
.pause_zero(true)
.pause_tx(true)
.ex_intr(true)
.autoneg_complete(true)
.partner_pg_rx(true)
.delay_req_rx(true)
.sync_rx(true)
.delay_req_tx(true)
.sync_tx(true)
.pdelay_req_rx(true)
.pdelay_resp_rx(true)
.pdelay_req_tx(true)
.pdelay_resp_tx(true)
.tsu_sec_incr(true)
);
// Clear the buffer queues.
self.regs.rx_qbar.write(
regs::RxQbar::zeroed()
);
self.regs.tx_qbar.write(
regs::TxQbar::zeroed()
);
}
fn configure(&mut self, macaddr: [u8; 6]) {
let clocks = Clocks::get();
let mdc_clk_div = (clocks.cpu_1x() / MAX_MDC) + 1;
self.regs.net_cfg.write(
regs::NetCfg::zeroed()
.full_duplex(true)
.gige_en(true)
.speed(true)
.no_broadcast(false)
.multi_hash_en(true)
// Promiscuous mode (TODO?)
.copy_all(true)
// Remove 4-byte Frame CheckSum
.fcs_remove(true)
// RX checksum offload
.rx_chksum_offld_en(true)
// One of the slower speeds
.mdc_clk_div((mdc_clk_div >> 4).min(0b111) as u8)
);
let macaddr_msbs =
(u16::from(macaddr[0]) << 8) |
u16::from(macaddr[1]);
let macaddr_lsbs =
(u32::from(macaddr[2]) << 24) |
(u32::from(macaddr[3]) << 16) |
(u32::from(macaddr[4]) << 8) |
u32::from(macaddr[5]);
self.regs.spec_addr1_top.write(
regs::SpecAddrTop::zeroed()
.addr_msbs(macaddr_msbs)
);
self.regs.spec_addr1_bot.write(
regs::SpecAddrBot::zeroed()
.addr_lsbs(macaddr_lsbs)
);
self.regs.dma_cfg.write(
regs::DmaCfg::zeroed()
// 1536 bytes
.ahb_mem_rx_buf_size((MTU >> 6) as u8)
// 8 KB
.rx_pktbuf_memsz_sel(0x3)
// 4 KB
.tx_pktbuf_memsz_sel(true)
// TX checksum offload
.csum_gen_offload_en(true)
// Little-endian
.ahb_endian_swp_mgmt_en(false)
// INCR16 AHB burst
.ahb_fixed_burst_len(0x10)
);
self.regs.net_ctrl.write(
regs::NetCtrl::zeroed()
.mgmt_port_en(true)
);
}
fn wait_phy_idle(&self) {
while !self.regs.net_status.read().phy_mgmt_idle() {}
}
fn check_link_change(&mut self, phy: &Phy) {
// As the PHY access takes some time, exit early if there was
// already a link. TODO: check once per second.
if self.link.is_some() {
return
}
let link = phy.get_link(self);
// Check link state transition
if self.link != link {
match &link {
Some(link) => {
info!("eth: got {:?}", link);
use phy::LinkSpeed::*;
let txclock = match link.speed {
S10 => TX_10,
S100 => TX_100,
S1000 => TX_1000,
};
Eth::<(), ()>::setup_gem0_clock(txclock);
/* .full_duplex(false) doesn't work even if
half duplex has been negotiated. */
self.regs.net_cfg.modify(|_, w| w
.full_duplex(true)
.gige_en(link.speed == S1000)
.speed(link.speed != S10)
);
}
None => {
warn!("eth: link lost");
phy.modify_control(self, |control|
control.set_autoneg_enable(true)
.set_restart_autoneg(true)
);
}
}
self.link = link;
}
}
}
impl<'r> PhyAccess for EthInner<'r> {
fn read_phy(&mut self, addr: u8, reg: u8) -> u16 {
self.wait_phy_idle();
self.regs.phy_maint.write(
regs::PhyMaint::zeroed()
.clause_22(true)
.operation(regs::PhyOperation::Read)
.phy_addr(addr)
.reg_addr(reg)
.must_10(0b10)
);
self.wait_phy_idle();
self.regs.phy_maint.read().data()
}
fn write_phy(&mut self, addr: u8, reg: u8, data: u16) {
self.wait_phy_idle();
self.regs.phy_maint.write(
regs::PhyMaint::zeroed()
.clause_22(true)
.operation(regs::PhyOperation::Write)
.phy_addr(addr)
.reg_addr(reg)
.must_10(0b10)
.data(data)
);
self.wait_phy_idle();
}
}