embassy/embassy-nrf/src/qspi.rs

569 lines
17 KiB
Rust

#![macro_use]
use core::future::poll_fn;
use core::ptr;
use core::task::Poll;
use embassy_hal_common::drop::DropBomb;
use embassy_hal_common::{into_ref, PeripheralRef};
use crate::gpio::{self, Pin as GpioPin};
use crate::interrupt::{Interrupt, InterruptExt};
pub use crate::pac::qspi::ifconfig0::{
ADDRMODE_A as AddressMode, PPSIZE_A as WritePageSize, READOC_A as ReadOpcode, WRITEOC_A as WriteOpcode,
};
pub use crate::pac::qspi::ifconfig1::SPIMODE_A as SpiMode;
use crate::{pac, Peripheral};
pub struct DeepPowerDownConfig {
/// Time required for entering DPM, in units of 16us
pub enter_time: u16,
/// Time required for exiting DPM, in units of 16us
pub exit_time: u16,
}
pub enum Frequency {
M32 = 0,
M16 = 1,
M10_7 = 2,
M8 = 3,
M6_4 = 4,
M5_3 = 5,
M4_6 = 6,
M4 = 7,
M3_6 = 8,
M3_2 = 9,
M2_9 = 10,
M2_7 = 11,
M2_5 = 12,
M2_3 = 13,
M2_1 = 14,
M2 = 15,
}
#[non_exhaustive]
pub struct Config {
pub xip_offset: u32,
pub read_opcode: ReadOpcode,
pub write_opcode: WriteOpcode,
pub write_page_size: WritePageSize,
pub deep_power_down: Option<DeepPowerDownConfig>,
pub frequency: Frequency,
/// Value is specified in number of 16 MHz periods (62.5 ns)
pub sck_delay: u8,
/// Whether data is captured on the clock rising edge and data is output on a falling edge (MODE0) or vice-versa (MODE3)
pub spi_mode: SpiMode,
pub address_mode: AddressMode,
}
impl Default for Config {
fn default() -> Self {
Self {
read_opcode: ReadOpcode::READ4IO,
write_opcode: WriteOpcode::PP4IO,
xip_offset: 0,
write_page_size: WritePageSize::_256BYTES,
deep_power_down: None,
frequency: Frequency::M8,
sck_delay: 80,
spi_mode: SpiMode::MODE0,
address_mode: AddressMode::_24BIT,
}
}
}
#[derive(Debug, Copy, Clone, Eq, PartialEq)]
#[cfg_attr(feature = "defmt", derive(defmt::Format))]
#[non_exhaustive]
pub enum Error {
OutOfBounds,
// TODO add "not in data memory" error and check for it
}
pub struct Qspi<'d, T: Instance, const FLASH_SIZE: usize> {
irq: PeripheralRef<'d, T::Interrupt>,
dpm_enabled: bool,
}
impl<'d, T: Instance, const FLASH_SIZE: usize> Qspi<'d, T, FLASH_SIZE> {
pub fn new(
_qspi: impl Peripheral<P = T> + 'd,
irq: impl Peripheral<P = T::Interrupt> + 'd,
sck: impl Peripheral<P = impl GpioPin> + 'd,
csn: impl Peripheral<P = impl GpioPin> + 'd,
io0: impl Peripheral<P = impl GpioPin> + 'd,
io1: impl Peripheral<P = impl GpioPin> + 'd,
io2: impl Peripheral<P = impl GpioPin> + 'd,
io3: impl Peripheral<P = impl GpioPin> + 'd,
config: Config,
) -> Qspi<'d, T, FLASH_SIZE> {
into_ref!(irq, sck, csn, io0, io1, io2, io3);
let r = T::regs();
sck.set_high();
csn.set_high();
io0.set_high();
io1.set_high();
io2.set_high();
io3.set_high();
sck.conf().write(|w| w.dir().output().drive().h0h1());
csn.conf().write(|w| w.dir().output().drive().h0h1());
io0.conf().write(|w| w.dir().output().drive().h0h1());
io1.conf().write(|w| w.dir().output().drive().h0h1());
io2.conf().write(|w| w.dir().output().drive().h0h1());
io3.conf().write(|w| w.dir().output().drive().h0h1());
r.psel.sck.write(|w| unsafe { w.bits(sck.psel_bits()) });
r.psel.csn.write(|w| unsafe { w.bits(csn.psel_bits()) });
r.psel.io0.write(|w| unsafe { w.bits(io0.psel_bits()) });
r.psel.io1.write(|w| unsafe { w.bits(io1.psel_bits()) });
r.psel.io2.write(|w| unsafe { w.bits(io2.psel_bits()) });
r.psel.io3.write(|w| unsafe { w.bits(io3.psel_bits()) });
r.ifconfig0.write(|w| {
w.addrmode().variant(config.address_mode);
w.dpmenable().bit(config.deep_power_down.is_some());
w.ppsize().variant(config.write_page_size);
w.readoc().variant(config.read_opcode);
w.writeoc().variant(config.write_opcode);
w
});
if let Some(dpd) = &config.deep_power_down {
r.dpmdur.write(|w| unsafe {
w.enter().bits(dpd.enter_time);
w.exit().bits(dpd.exit_time);
w
})
}
r.ifconfig1.write(|w| unsafe {
w.sckdelay().bits(config.sck_delay);
w.dpmen().exit();
w.spimode().variant(config.spi_mode);
w.sckfreq().bits(config.frequency as u8);
w
});
r.xipoffset.write(|w| unsafe {
w.xipoffset().bits(config.xip_offset);
w
});
irq.set_handler(Self::on_interrupt);
irq.unpend();
irq.enable();
// Enable it
r.enable.write(|w| w.enable().enabled());
let mut res = Self {
dpm_enabled: config.deep_power_down.is_some(),
irq,
};
r.events_ready.reset();
r.intenset.write(|w| w.ready().set());
r.tasks_activate.write(|w| w.tasks_activate().bit(true));
res.blocking_wait_ready();
res
}
fn on_interrupt(_: *mut ()) {
let r = T::regs();
let s = T::state();
if r.events_ready.read().bits() != 0 {
s.ready_waker.wake();
r.intenclr.write(|w| w.ready().clear());
}
}
pub async fn custom_instruction(&mut self, opcode: u8, req: &[u8], resp: &mut [u8]) -> Result<(), Error> {
let bomb = DropBomb::new();
let len = core::cmp::max(req.len(), resp.len()) as u8;
self.custom_instruction_start(opcode, req, len)?;
self.wait_ready().await;
self.custom_instruction_finish(resp)?;
bomb.defuse();
Ok(())
}
pub fn blocking_custom_instruction(&mut self, opcode: u8, req: &[u8], resp: &mut [u8]) -> Result<(), Error> {
let len = core::cmp::max(req.len(), resp.len()) as u8;
self.custom_instruction_start(opcode, req, len)?;
self.blocking_wait_ready();
self.custom_instruction_finish(resp)?;
Ok(())
}
fn custom_instruction_start(&mut self, opcode: u8, req: &[u8], len: u8) -> Result<(), Error> {
assert!(req.len() <= 8);
let mut dat0: u32 = 0;
let mut dat1: u32 = 0;
for i in 0..4 {
if i < req.len() {
dat0 |= (req[i] as u32) << (i * 8);
}
}
for i in 0..4 {
if i + 4 < req.len() {
dat1 |= (req[i + 4] as u32) << (i * 8);
}
}
let r = T::regs();
r.cinstrdat0.write(|w| unsafe { w.bits(dat0) });
r.cinstrdat1.write(|w| unsafe { w.bits(dat1) });
r.events_ready.reset();
r.intenset.write(|w| w.ready().set());
r.cinstrconf.write(|w| {
let w = unsafe { w.opcode().bits(opcode) };
let w = unsafe { w.length().bits(len + 1) };
let w = w.lio2().bit(true);
let w = w.lio3().bit(true);
let w = w.wipwait().bit(true);
let w = w.wren().bit(true);
let w = w.lfen().bit(false);
let w = w.lfstop().bit(false);
w
});
Ok(())
}
fn custom_instruction_finish(&mut self, resp: &mut [u8]) -> Result<(), Error> {
let r = T::regs();
let dat0 = r.cinstrdat0.read().bits();
let dat1 = r.cinstrdat1.read().bits();
for i in 0..4 {
if i < resp.len() {
resp[i] = (dat0 >> (i * 8)) as u8;
}
}
for i in 0..4 {
if i + 4 < resp.len() {
resp[i] = (dat1 >> (i * 8)) as u8;
}
}
Ok(())
}
async fn wait_ready(&mut self) {
poll_fn(move |cx| {
let r = T::regs();
let s = T::state();
s.ready_waker.register(cx.waker());
if r.events_ready.read().bits() != 0 {
return Poll::Ready(());
}
Poll::Pending
})
.await
}
fn blocking_wait_ready(&mut self) {
loop {
let r = T::regs();
if r.events_ready.read().bits() != 0 {
break;
}
}
}
fn start_read(&mut self, address: usize, data: &mut [u8]) -> Result<(), Error> {
assert_eq!(data.as_ptr() as u32 % 4, 0);
assert_eq!(data.len() as u32 % 4, 0);
assert_eq!(address as u32 % 4, 0);
if address > FLASH_SIZE {
return Err(Error::OutOfBounds);
}
let r = T::regs();
r.read.src.write(|w| unsafe { w.src().bits(address as u32) });
r.read.dst.write(|w| unsafe { w.dst().bits(data.as_ptr() as u32) });
r.read.cnt.write(|w| unsafe { w.cnt().bits(data.len() as u32) });
r.events_ready.reset();
r.intenset.write(|w| w.ready().set());
r.tasks_readstart.write(|w| w.tasks_readstart().bit(true));
Ok(())
}
fn start_write(&mut self, address: usize, data: &[u8]) -> Result<(), Error> {
assert_eq!(data.as_ptr() as u32 % 4, 0);
assert_eq!(data.len() as u32 % 4, 0);
assert_eq!(address as u32 % 4, 0);
if address > FLASH_SIZE {
return Err(Error::OutOfBounds);
}
let r = T::regs();
r.write.src.write(|w| unsafe { w.src().bits(data.as_ptr() as u32) });
r.write.dst.write(|w| unsafe { w.dst().bits(address as u32) });
r.write.cnt.write(|w| unsafe { w.cnt().bits(data.len() as u32) });
r.events_ready.reset();
r.intenset.write(|w| w.ready().set());
r.tasks_writestart.write(|w| w.tasks_writestart().bit(true));
Ok(())
}
fn start_erase(&mut self, address: usize) -> Result<(), Error> {
assert_eq!(address as u32 % 4096, 0);
if address > FLASH_SIZE {
return Err(Error::OutOfBounds);
}
let r = T::regs();
r.erase.ptr.write(|w| unsafe { w.ptr().bits(address as u32) });
r.erase.len.write(|w| w.len()._4kb());
r.events_ready.reset();
r.intenset.write(|w| w.ready().set());
r.tasks_erasestart.write(|w| w.tasks_erasestart().bit(true));
Ok(())
}
pub async fn read(&mut self, address: usize, data: &mut [u8]) -> Result<(), Error> {
let bomb = DropBomb::new();
self.start_read(address, data)?;
self.wait_ready().await;
bomb.defuse();
Ok(())
}
pub async fn write(&mut self, address: usize, data: &[u8]) -> Result<(), Error> {
let bomb = DropBomb::new();
self.start_write(address, data)?;
self.wait_ready().await;
bomb.defuse();
Ok(())
}
pub async fn erase(&mut self, address: usize) -> Result<(), Error> {
let bomb = DropBomb::new();
self.start_erase(address)?;
self.wait_ready().await;
bomb.defuse();
Ok(())
}
pub fn blocking_read(&mut self, address: usize, data: &mut [u8]) -> Result<(), Error> {
self.start_read(address, data)?;
self.blocking_wait_ready();
Ok(())
}
pub fn blocking_write(&mut self, address: usize, data: &[u8]) -> Result<(), Error> {
self.start_write(address, data)?;
self.blocking_wait_ready();
Ok(())
}
pub fn blocking_erase(&mut self, address: usize) -> Result<(), Error> {
self.start_erase(address)?;
self.blocking_wait_ready();
Ok(())
}
}
impl<'d, T: Instance, const FLASH_SIZE: usize> Drop for Qspi<'d, T, FLASH_SIZE> {
fn drop(&mut self) {
let r = T::regs();
if self.dpm_enabled {
trace!("qspi: doing deep powerdown...");
r.ifconfig1.modify(|_, w| w.dpmen().enter());
// Wait for DPM enter.
// Unfortunately we must spin. There's no way to do this interrupt-driven.
// The READY event does NOT fire on DPM enter (but it does fire on DPM exit :shrug:)
while r.status.read().dpm().is_disabled() {}
// Wait MORE for DPM enter.
// I have absolutely no idea why, but the wait above is not enough :'(
// Tested with mx25r64 in nrf52840-dk, and with mx25r16 in custom board
cortex_m::asm::delay(4096);
}
// it seems events_ready is not generated in response to deactivate. nrfx doesn't wait for it.
r.tasks_deactivate.write(|w| w.tasks_deactivate().set_bit());
// Workaround https://infocenter.nordicsemi.com/topic/errata_nRF52840_Rev1/ERR/nRF52840/Rev1/latest/anomaly_840_122.html?cp=4_0_1_2_1_7
// Note that the doc has 2 register writes, but the first one is really the write to tasks_deactivate,
// so we only do the second one here.
unsafe { ptr::write_volatile(0x40029054 as *mut u32, 1) }
r.enable.write(|w| w.enable().disabled());
self.irq.disable();
// Note: we do NOT deconfigure CSN here. If DPM is in use and we disconnect CSN,
// leaving it floating, the flash chip might read it as zero which would cause it to
// spuriously exit DPM.
gpio::deconfigure_pin(r.psel.sck.read().bits());
gpio::deconfigure_pin(r.psel.io0.read().bits());
gpio::deconfigure_pin(r.psel.io1.read().bits());
gpio::deconfigure_pin(r.psel.io2.read().bits());
gpio::deconfigure_pin(r.psel.io3.read().bits());
trace!("qspi: dropped");
}
}
use embedded_storage::nor_flash::{ErrorType, NorFlash, NorFlashError, NorFlashErrorKind, ReadNorFlash};
impl<'d, T: Instance, const FLASH_SIZE: usize> ErrorType for Qspi<'d, T, FLASH_SIZE> {
type Error = Error;
}
impl NorFlashError for Error {
fn kind(&self) -> NorFlashErrorKind {
NorFlashErrorKind::Other
}
}
impl<'d, T: Instance, const FLASH_SIZE: usize> ReadNorFlash for Qspi<'d, T, FLASH_SIZE> {
const READ_SIZE: usize = 4;
fn read(&mut self, offset: u32, bytes: &mut [u8]) -> Result<(), Self::Error> {
self.blocking_read(offset as usize, bytes)?;
Ok(())
}
fn capacity(&self) -> usize {
FLASH_SIZE
}
}
impl<'d, T: Instance, const FLASH_SIZE: usize> NorFlash for Qspi<'d, T, FLASH_SIZE> {
const WRITE_SIZE: usize = 4;
const ERASE_SIZE: usize = 4096;
fn erase(&mut self, from: u32, to: u32) -> Result<(), Self::Error> {
for address in (from as usize..to as usize).step_by(<Self as NorFlash>::ERASE_SIZE) {
self.blocking_erase(address)?;
}
Ok(())
}
fn write(&mut self, offset: u32, bytes: &[u8]) -> Result<(), Self::Error> {
self.blocking_write(offset as usize, bytes)?;
Ok(())
}
}
cfg_if::cfg_if! {
if #[cfg(feature = "nightly")]
{
use embedded_storage_async::nor_flash::{AsyncNorFlash, AsyncReadNorFlash};
use core::future::Future;
impl<'d, T: Instance, const FLASH_SIZE: usize> AsyncNorFlash for Qspi<'d, T, FLASH_SIZE> {
const WRITE_SIZE: usize = <Self as NorFlash>::WRITE_SIZE;
const ERASE_SIZE: usize = <Self as NorFlash>::ERASE_SIZE;
type WriteFuture<'a> = impl Future<Output = Result<(), Self::Error>> + 'a where Self: 'a;
fn write<'a>(&'a mut self, offset: u32, data: &'a [u8]) -> Self::WriteFuture<'a> {
async move { self.write(offset as usize, data).await }
}
type EraseFuture<'a> = impl Future<Output = Result<(), Self::Error>> + 'a where Self: 'a;
fn erase<'a>(&'a mut self, from: u32, to: u32) -> Self::EraseFuture<'a> {
async move {
for address in (from as usize..to as usize).step_by(<Self as AsyncNorFlash>::ERASE_SIZE) {
self.erase(address).await?
}
Ok(())
}
}
}
impl<'d, T: Instance, const FLASH_SIZE: usize> AsyncReadNorFlash for Qspi<'d, T, FLASH_SIZE> {
const READ_SIZE: usize = 4;
type ReadFuture<'a> = impl Future<Output = Result<(), Self::Error>> + 'a where Self: 'a;
fn read<'a>(&'a mut self, address: u32, data: &'a mut [u8]) -> Self::ReadFuture<'a> {
async move { self.read(address as usize, data).await }
}
fn capacity(&self) -> usize {
FLASH_SIZE
}
}
}
}
pub(crate) mod sealed {
use embassy_sync::waitqueue::AtomicWaker;
use super::*;
pub struct State {
pub ready_waker: AtomicWaker,
}
impl State {
pub const fn new() -> Self {
Self {
ready_waker: AtomicWaker::new(),
}
}
}
pub trait Instance {
fn regs() -> &'static pac::qspi::RegisterBlock;
fn state() -> &'static State;
}
}
pub trait Instance: Peripheral<P = Self> + sealed::Instance + 'static {
type Interrupt: Interrupt;
}
macro_rules! impl_qspi {
($type:ident, $pac_type:ident, $irq:ident) => {
impl crate::qspi::sealed::Instance for peripherals::$type {
fn regs() -> &'static pac::qspi::RegisterBlock {
unsafe { &*pac::$pac_type::ptr() }
}
fn state() -> &'static crate::qspi::sealed::State {
static STATE: crate::qspi::sealed::State = crate::qspi::sealed::State::new();
&STATE
}
}
impl crate::qspi::Instance for peripherals::$type {
type Interrupt = crate::interrupt::$irq;
}
};
}