DMA experimentation

examples/simple/examples/dma.rs

@@ -0,0 +1,269 @@
+//! Simple DMA example
+#![no_main]
+#![no_std]
+
+use core::cell::Cell;
+
+use cortex_m::interrupt::Mutex;
+use cortex_m_rt::entry;
+use embedded_hal::delay::DelayNs;
+use panic_rtt_target as _;
+use rtt_target::{rprintln, rtt_init_print};
+use simple_examples::peb1;
+use va416xx_hal::dma::{Dma, DmaCfg, DmaChannel, DmaCtrlBlock};
+use va416xx_hal::pwm::CountdownTimer;
+use va416xx_hal::{
+    pac::{self, interrupt},
+    prelude::*,
+};
+
+static DMA_DONE_FLAG: Mutex<Cell<bool>> = Mutex::new(Cell::new(false));
+static DMA_ACTIVE_FLAG: Mutex<Cell<bool>> = Mutex::new(Cell::new(false));
+
+// Place the DMA control block into SRAM1 statically. This section needs to be defined in
+// memory.x
+#[link_section = ".sram1"]
+static mut DMA_CTRL_BLOCK: DmaCtrlBlock = DmaCtrlBlock::new();
+
+#[entry]
+fn main() -> ! {
+    rtt_init_print!();
+    rprintln!("VA416xx DMA example");
+
+    let mut dp = pac::Peripherals::take().unwrap();
+    // Use the external clock connected to XTAL_N.
+    let clocks = dp
+        .clkgen
+        .constrain()
+        .xtal_n_clk_with_src_freq(peb1::EXTCLK_FREQ)
+        .freeze(&mut dp.sysconfig)
+        .unwrap();
+    // Safety: The DMA control block has an alignment rule of 128 and we constructed it directly
+    // statically.
+    let dma = Dma::new(&mut dp.sysconfig, dp.dma, DmaCfg::default(), unsafe {
+        core::ptr::addr_of_mut!(DMA_CTRL_BLOCK)
+    })
+    .expect("error creating DMA");
+    let (mut dma0, _, _, _) = dma.split();
+    let mut delay_ms = CountdownTimer::new(&mut dp.sysconfig, dp.tim0, &clocks);
+    let mut src_buf_8_bit: [u8; 65] = [0; 65];
+    let mut dest_buf_8_bit: [u8; 65] = [0; 65];
+    let mut src_buf_16_bit: [u16; 33] = [0; 33];
+    let mut dest_buf_16_bit: [u16; 33] = [0; 33];
+    let mut src_buf_32_bit: [u32; 17] = [0; 17];
+    let mut dest_buf_32_bit: [u32; 17] = [0; 17];
+    loop {
+        transfer_example_8_bit(
+            &mut src_buf_8_bit,
+            &mut dest_buf_8_bit,
+            &mut dma0,
+            &mut delay_ms,
+        );
+        delay_ms.delay_ms(500);
+        transfer_example_16_bit(
+            &mut src_buf_16_bit,
+            &mut dest_buf_16_bit,
+            &mut dma0,
+            &mut delay_ms,
+        );
+        delay_ms.delay_ms(500);
+        transfer_example_32_bit(
+            &mut src_buf_32_bit,
+            &mut dest_buf_32_bit,
+            &mut dma0,
+            &mut delay_ms,
+        );
+        delay_ms.delay_ms(500);
+    }
+}
+
+fn transfer_example_8_bit(
+    src_buf: &mut [u8; 65],
+    dest_buf: &mut [u8; 65],
+    dma0: &mut DmaChannel,
+    delay_ms: &mut CountdownTimer<pac::Tim0>,
+) {
+    (0..64).for_each(|i| {
+        src_buf[i] = i as u8;
+    });
+    cortex_m::interrupt::free(|cs| {
+        DMA_DONE_FLAG.borrow(cs).set(false);
+    });
+    cortex_m::interrupt::free(|cs| {
+        DMA_ACTIVE_FLAG.borrow(cs).set(false);
+    });
+    let dma_transfer = dma0
+        .prepare_mem_to_mem_transfer_8_bit(src_buf, dest_buf)
+        .expect("error preparing transfer");
+    // Enable all interrupts.
+    // Safety: Not using mask based critical sections.
+    unsafe {
+        dma0.enable_done_interrupt();
+        dma0.enable_active_interrupt();
+    };
+    // Enable the individual channel.
+    dma0.enable();
+    // We still need to manually trigger the DMA request.
+    dma0.trigger_with_sw_request();
+    let dest_buf;
+    // Use polling for completion status.
+    loop {
+        let mut dma_done = false;
+        cortex_m::interrupt::free(|cs| {
+            if DMA_ACTIVE_FLAG.borrow(cs).get() {
+                rprintln!("DMA0 is active with 8 bit transfer");
+                DMA_ACTIVE_FLAG.borrow(cs).set(false);
+            }
+            if DMA_DONE_FLAG.borrow(cs).get() {
+                dma_done = true;
+            }
+        });
+        if dma_done {
+            rprintln!("8-bit transfer done");
+            // Safety: We checked for transfer completion.
+            dest_buf = unsafe { dma_transfer.release() };
+            break;
+        }
+        delay_ms.delay_ms(1);
+    }
+    (0..64).for_each(|i| {
+        assert_eq!(dest_buf[i], i as u8);
+    });
+    // Sentinel value, should be 0.
+    assert_eq!(dest_buf[64], 0);
+    dest_buf.fill(0);
+}
+
+fn transfer_example_16_bit(
+    src_buf: &mut [u16; 33],
+    dest_buf: &mut [u16; 33],
+    dma0: &mut DmaChannel,
+    delay_ms: &mut CountdownTimer<pac::Tim0>,
+) {
+    // Set values scaled from 0 to 65535 to verify this is really a 16-bit transfer.
+    (0..32).for_each(|i| {
+        src_buf[i] = (i as u32 * u16::MAX as u32 / (src_buf.len() - 1) as u32) as u16;
+    });
+    cortex_m::interrupt::free(|cs| {
+        DMA_DONE_FLAG.borrow(cs).set(false);
+    });
+    cortex_m::interrupt::free(|cs| {
+        DMA_ACTIVE_FLAG.borrow(cs).set(false);
+    });
+    let dma_transfer = dma0
+        .prepare_mem_to_mem_transfer_16_bit(src_buf, dest_buf)
+        .expect("error preparing transfer");
+    dest_buf[5] = 2;
+    // Enable all interrupts.
+    // Safety: Not using mask based critical sections.
+    unsafe {
+        dma0.enable_done_interrupt();
+        dma0.enable_active_interrupt();
+    };
+    // Enable the individual channel.
+    dma0.enable();
+    // We still need to manually trigger the DMA request.
+    dma0.trigger_with_sw_request();
+    // Use polling for completion status.
+    loop {
+        let mut dma_done = false;
+        cortex_m::interrupt::free(|cs| {
+            if DMA_ACTIVE_FLAG.borrow(cs).get() {
+                rprintln!("DMA0 is active with 16-bit transfer");
+                DMA_ACTIVE_FLAG.borrow(cs).set(false);
+            }
+            if DMA_DONE_FLAG.borrow(cs).get() {
+                dma_done = true;
+            }
+        });
+        if dma_done {
+            rprintln!("16-bit transfer done");
+            break;
+        }
+        delay_ms.delay_ms(1);
+    }
+    (0..32).for_each(|i| {
+        assert_eq!(
+            dest_buf[i],
+            (i as u32 * u16::MAX as u32 / (src_buf.len() - 1) as u32) as u16
+        );
+    });
+    // Sentinel value, should be 0.
+    assert_eq!(dest_buf[32], 0);
+    dest_buf.fill(0);
+}
+
+fn transfer_example_32_bit(
+    src_buf: &mut [u32; 17],
+    dest_buf: &mut [u32; 17],
+    dma0: &mut DmaChannel,
+    delay_ms: &mut CountdownTimer<pac::Tim0>,
+) {
+    // Set values scaled from 0 to 65535 to verify this is really a 16-bit transfer.
+    (0..16).for_each(|i| {
+        src_buf[i] = (i as u64 * u32::MAX as u64 / (src_buf.len() - 1) as u64) as u32;
+    });
+    cortex_m::interrupt::free(|cs| {
+        DMA_DONE_FLAG.borrow(cs).set(false);
+    });
+    cortex_m::interrupt::free(|cs| {
+        DMA_ACTIVE_FLAG.borrow(cs).set(false);
+    });
+    dma0.prepare_mem_to_mem_transfer_32_bit(src_buf, dest_buf)
+        .expect("error preparing transfer");
+    // Enable all interrupts.
+    // Safety: Not using mask based critical sections.
+    unsafe {
+        dma0.enable_done_interrupt();
+        dma0.enable_active_interrupt();
+    };
+    // Enable the individual channel.
+    dma0.enable();
+    // We still need to manually trigger the DMA request.
+    dma0.trigger_with_sw_request();
+    // Use polling for completion status.
+    loop {
+        let mut dma_done = false;
+        cortex_m::interrupt::free(|cs| {
+            if DMA_ACTIVE_FLAG.borrow(cs).get() {
+                rprintln!("DMA0 is active with 32-bit transfer");
+                DMA_ACTIVE_FLAG.borrow(cs).set(false);
+            }
+            if DMA_DONE_FLAG.borrow(cs).get() {
+                dma_done = true;
+            }
+        });
+        if dma_done {
+            rprintln!("32-bit transfer done");
+            break;
+        }
+        delay_ms.delay_ms(1);
+    }
+    (0..16).for_each(|i| {
+        assert_eq!(
+            dest_buf[i],
+            (i as u64 * u32::MAX as u64 / (src_buf.len() - 1) as u64) as u32
+        );
+    });
+    // Sentinel value, should be 0.
+    assert_eq!(dest_buf[16], 0);
+    dest_buf.fill(0);
+}
+
+#[interrupt]
+#[allow(non_snake_case)]
+fn DMA_DONE0() {
+    // Notify the main loop that the DMA transfer is finished.
+    cortex_m::interrupt::free(|cs| {
+        DMA_DONE_FLAG.borrow(cs).set(true);
+    });
+}
+
+#[interrupt]
+#[allow(non_snake_case)]
+fn DMA_ACTIVE0() {
+    // Notify the main loop that the DMA 0 is active now.
+    cortex_m::interrupt::free(|cs| {
+        DMA_ACTIVE_FLAG.borrow(cs).set(true);
+    });
+}

va416xx-hal/Cargo.toml

@@ -17,9 +17,11 @@ paste = "1"
 embedded-hal-nb = "1"
 embedded-hal = "1"
 embedded-io = "0.6"
+embedded-dma = "0.2"
 num_enum = { version = "0.7", default-features = false }
 typenum = "1"
 bitflags = "2"
+bitfield = "0.15"
 defmt = { version = "0.3", optional = true }
 fugit = "0.3"
 delegate = "0.12"

va416xx-hal/memory.x

@@ -1,14 +0,0 @@
-MEMORY
-{
-	FLASH : ORIGIN = 0x00000000, LENGTH = 256K
-	/* RAM is a mandatory region. This RAM refers to the SRAM_0 */
-	RAM : ORIGIN = 0x1FFF8000, LENGTH = 32K
-	SRAM_1 : ORIGIN = 0x20000000, LENGTH = 32K
-}
-
-/* This is where the call stack will be allocated. */
-/* The stack is of the full descending type. */
-/* NOTE Do NOT modify `_stack_start` unless you know what you are doing */
-/* SRAM_0 can be used for all busses: Instruction, Data and System */
-/* SRAM_1 only supports the system bus */
-_stack_start = ORIGIN(RAM) + LENGTH(RAM) - 4;

va416xx-hal/src/dma.rs

@@ -0,0 +1,515 @@
+//! API for the DMA peripheral
+//!
+//! ## Examples
+//!
+//! - [Simple DMA example](https://egit.irs.uni-stuttgart.de/rust/va416xx-rs/src/branch/main/examples/simple/examples/dma.rs)
+use embedded_dma::WriteBuffer;
+
+use crate::{
+    clock::{PeripheralClock, PeripheralSelect},
+    enable_interrupt, pac,
+    prelude::*,
+};
+
+const MAX_DMA_TRANSFERS_PER_CYCLE: usize = 1024;
+const BASE_PTR_ADDR_MASK: u32 = 0b1111111;
+
+/// DMA cycle control values.
+///
+/// Refer to chapter 6.3.1 and 6.6.3 of the datasheet for more details.
+#[repr(u8)]
+#[derive(Debug, Clone, Copy)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum CycleControl {
+    /// Indicates that the data structure is invalid.
+    Stop = 0b000,
+    /// The controller must receive a new request prior to entering the arbitration
+    /// process, to enable the DMA cycle to complete. This means that the DMA will only
+    /// continue to do transfers as long as a trigger signal is still active. Therefore,
+    /// this should not be used for momentary triggers like a timer.
+    Basic = 0b001,
+    /// The controller automatically inserts a request for the appropriate channel during the
+    /// arbitration process. This means that the initial request is sufficient to enable the
+    /// DMA cycle to complete.
+    Auto = 0b010,
+    /// This is used to support continuous data flow. Both primary and alternate data structure
+    /// are used. The primary data structure is used first. When the first transfer is complete, an
+    /// interrupt can be generated, and the DMA switches to the alternate data structure. When the
+    /// second transfer is complete, the primary data structure is used. This pattern continues
+    /// until software disables the channel.
+    PingPong = 0b011,
+    MemScatterGatherPrimary = 0b100,
+    MemScatterGatherAlternate = 0b101,
+    PeriphScatterGatherPrimary = 0b110,
+    PeriphScatterGatherAlternate = 0b111,
+}
+
+#[derive(Debug, Clone, Copy)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum AddrIncrement {
+    Byte = 0b00,
+    Halfword = 0b01,
+    Word = 0b10,
+    None = 0b11,
+}
+
+#[derive(Debug, Clone, Copy)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum DataSize {
+    Byte = 0b00,
+    Halfword = 0b01,
+    Word = 0b10,
+}
+
+/// This configuration controls how many DMA transfers can occur before the controller arbitrates.
+#[derive(Debug, Clone, Copy)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum RPower {
+    EachTransfer = 0b0000,
+    Every2 = 0b0001,
+    Every4 = 0b0010,
+    Every8 = 0b0011,
+    Every16 = 0b0100,
+    Every32 = 0b0101,
+    Every64 = 0b0110,
+    Every128 = 0b0111,
+    Every256 = 0b1000,
+    Every512 = 0b1001,
+    Every1024Min = 0b1010,
+    Every1024 = 0b1111,
+}
+
+#[derive(Debug, PartialEq, Eq)]
+pub struct InvalidCtrlBlockAddr;
+
+bitfield::bitfield! {
+    #[repr(transparent)]
+    #[derive(Clone, Copy)]
+    pub struct ChannelConfig(u32);
+    impl Debug;
+    u32;
+    pub raw, set_raw: 31,0;
+    u8;
+    pub dst_inc, set_dst_inc: 31, 30;
+    u8;
+    pub dst_size, set_dst_size: 29, 28;
+    u8;
+    pub src_inc, set_src_inc: 27, 26;
+    u8;
+    pub src_size, set_src_size: 25, 24;
+    u8;
+    pub dest_prot_ctrl, set_dest_prot_ctrl: 23, 21;
+    u8;
+    pub src_prot_ctrl, set_src_prot_ctrl: 20, 18;
+    u8;
+    pub r_power, set_r_power: 17, 14;
+    u16;
+    pub n_minus_1, set_n_minus_1: 13, 4;
+    bool;
+    pub next_useburst, set_next_useburst: 3;
+    u8;
+    pub cycle_ctrl, set_cycle_ctr: 2, 0;
+}
+
+#[repr(C)]
+#[derive(Debug, Copy, Clone)]
+pub struct DmaChannelControl {
+    pub src_end_ptr: u32,
+    pub dest_end_ptr: u32,
+    pub cfg: ChannelConfig,
+    padding: u32,
+}
+
+impl DmaChannelControl {
+    const fn new() -> Self {
+        Self {
+            src_end_ptr: 0,
+            dest_end_ptr: 0,
+            cfg: ChannelConfig(0),
+            padding: 0,
+        }
+    }
+}
+impl Default for DmaChannelControl {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+#[repr(C)]
+#[repr(align(128))]
+pub struct DmaCtrlBlock {
+    pub pri: [DmaChannelControl; 4],
+    pub alt: [DmaChannelControl; 4],
+}
+
+impl DmaCtrlBlock {
+    pub const fn new() -> Self {
+        Self {
+            pri: [DmaChannelControl::new(); 4],
+            alt: [DmaChannelControl::new(); 4],
+        }
+    }
+}
+impl Default for DmaCtrlBlock {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl DmaCtrlBlock {
+    /// This function creates a DMA control block at the specified memory address.
+    ///
+    /// The passed address must be 128-byte aligned. The user must also take care of specifying
+    /// a valid memory address for the DMA control block which is accessible by the system as well.
+    /// For example, the control block can be placed in the SRAM1.
+    pub fn new_at_addr(addr: u32) -> Result<*mut DmaCtrlBlock, InvalidCtrlBlockAddr> {
+        if addr & BASE_PTR_ADDR_MASK > 0 {
+            return Err(InvalidCtrlBlockAddr);
+        }
+        let ctrl_block_ptr = addr as *mut DmaCtrlBlock;
+        unsafe { core::ptr::write(ctrl_block_ptr, DmaCtrlBlock::default()) }
+        Ok(ctrl_block_ptr)
+    }
+}
+
+pub struct Dma {
+    dma: pac::Dma,
+    ctrl_block: *mut DmaCtrlBlock,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub enum DmaTransferInitError {
+    SourceDestLenMissmatch {
+        src_len: usize,
+        dest_len: usize,
+    },
+    /// Overflow when calculating the source or destination end address.
+    AddrOverflow,
+    /// Transfer size larger than 1024 units.
+    TransferSizeTooLarge(usize),
+}
+
+#[derive(Debug, Clone, Copy, Default)]
+pub struct DmaCfg {
+    pub bufferable: bool,
+    pub cacheable: bool,
+    pub privileged: bool,
+}
+
+pub struct DmaChannel {
+    channel: u8,
+    done_interrupt: pac::Interrupt,
+    active_interrupt: pac::Interrupt,
+    pub dma: pac::Dma,
+    pub ch_ctrl_pri: &'static mut DmaChannelControl,
+    pub ch_ctrl_alt: &'static mut DmaChannelControl,
+}
+
+/// This transfer structure takes ownership of the mutable destination slice.
+///
+/// This avoids accidental violation of the ownership rules because the DMA now has mutable
+/// access to that slice as well. The mutable slice can be retrieved after DMA transfer completion
+/// by using the [Self::release] method.
+pub struct DmaTransfer<W> {
+    buf: W,
+    //ch: DmaChannel
+}
+
+impl<W: WriteBuffer> DmaTransfer<W> {
+    /// Retrieve the mutable destination slice once the DMA transfer has completed.
+    ///
+    /// # Safety
+    ///
+    /// - The user MUST ensure that the DMA transfer has completed, for example by polling a
+    ///   completion flag set by the DMA_DONE ISR.
+    pub unsafe fn release(self) -> W {
+        self.buf
+    }
+}
+
+impl DmaChannel {
+    #[inline(always)]
+    pub fn channel(&self) -> u8 {
+        self.channel
+    }
+
+    #[inline(always)]
+    pub fn enable(&mut self) {
+        self.dma
+            .chnl_enable_set()
+            .write(|w| unsafe { w.bits(1 << self.channel) });
+    }
+
+    #[inline(always)]
+    pub fn is_enabled(&mut self) -> bool {
+        ((self.dma.chnl_enable_set().read().bits() >> self.channel) & 0b1) != 0
+    }
+
+    #[inline(always)]
+    pub fn disable(&mut self) {
+        self.dma
+            .chnl_enable_clr()
+            .write(|w| unsafe { w.bits(1 << self.channel) });
+    }
+
+    #[inline(always)]
+    pub fn trigger_with_sw_request(&mut self) {
+        self.dma
+            .chnl_sw_request()
+            .write(|w| unsafe { w.bits(1 << self.channel) });
+    }
+
+    #[inline(always)]
+    pub fn state_raw(&self) -> u8 {
+        self.dma.status().read().state().bits()
+    }
+
+    #[inline(always)]
+    pub fn select_primary_structure(&self) {
+        self.dma
+            .chnl_pri_alt_clr()
+            .write(|w| unsafe { w.bits(1 << self.channel) });
+    }
+
+    #[inline(always)]
+    pub fn select_alternate_structure(&self) {
+        self.dma
+            .chnl_pri_alt_set()
+            .write(|w| unsafe { w.bits(1 << self.channel) });
+    }
+
+    /// Enables the DMA_DONE interrupt for the DMA channel.
+    ///
+    /// # Safety
+    ///
+    /// This function is `unsafe` because it can break mask-based critical sections.
+    pub unsafe fn enable_done_interrupt(&mut self) {
+        enable_interrupt(self.done_interrupt);
+    }
+
+    /// Enables the DMA_ACTIVE interrupt for the DMA channel.
+    ///
+    /// # Safety
+    ///
+    /// This function is `unsafe` because it can break mask-based critical sections.
+    pub unsafe fn enable_active_interrupt(&mut self) {
+        enable_interrupt(self.active_interrupt);
+    }
+
+    /// Prepares a 8-bit DMA transfer from memory to memory.
+    ///
+    /// This function does not enable the DMA channel and interrupts and only prepares
+    /// the DMA control block parameters for the transfer. It configures the primary channel control
+    /// structure to perform the transfer.
+    ///
+    /// You can use [Self::enable], [Self::enable_done_interrupt], [Self::enable_active_interrupt]
+    /// to finish the transfer preparation and then use [Self::trigger_with_sw_request] to
+    /// start the DMA transfer.
+    pub fn prepare_mem_to_mem_transfer_8_bit<W: WriteBuffer<Word = u8>>(
+        &mut self,
+        source: &[u8],
+        mut dest: W,
+    ) -> Result<DmaTransfer<W>, DmaTransferInitError> {
+        let (write_ptr, len) = unsafe { dest.write_buffer() };
+        let len = Self::common_mem_transfer_checks(source.len(), len)?;
+        self.generic_mem_to_mem_transfer_init(
+            len,
+            (source.as_ptr() as u32)
+                .checked_add(len as u32)
+                .ok_or(DmaTransferInitError::AddrOverflow)?,
+            (write_ptr as u32)
+                .checked_add(len as u32)
+                .ok_or(DmaTransferInitError::AddrOverflow)?,
+            DataSize::Byte,
+            AddrIncrement::Byte,
+        );
+        Ok(DmaTransfer { buf: dest })
+    }
+
+    /// Prepares a 16-bit DMA transfer from memory to memory.
+    ///
+    /// This function does not enable the DMA channel and interrupts and only prepares
+    /// the DMA control block parameters for the transfer. It configures the primary channel control
+    /// structure to perform the transfer.
+    ///
+    /// You can use [Self::enable], [Self::enable_done_interrupt], [Self::enable_active_interrupt]
+    /// to finish the transfer preparation and then use [Self::trigger_with_sw_request] to
+    /// start the DMA transfer.
+    pub fn prepare_mem_to_mem_transfer_16_bit<'dest>(
+        &mut self,
+        source: &[u16],
+        dest: &'dest mut [u16],
+    ) -> Result<(), DmaTransferInitError> {
+        let len = Self::common_mem_transfer_checks(source.len(), dest.len())?;
+        self.generic_mem_to_mem_transfer_init(
+            len,
+            (source.as_ptr() as u32)
+                .checked_add(len as u32 * core::mem::size_of::<u16>() as u32)
+                .ok_or(DmaTransferInitError::AddrOverflow)?,
+            (dest.as_ptr() as u32)
+                .checked_add(len as u32 * core::mem::size_of::<u16>() as u32)
+                .ok_or(DmaTransferInitError::AddrOverflow)?,
+            DataSize::Halfword,
+            AddrIncrement::Halfword,
+        );
+        Ok(())
+    }
+
+    /// Prepares a 32-bit DMA transfer from memory to memory.
+    ///
+    /// This function does not enable the DMA channel and interrupts and only prepares
+    /// the DMA control block parameters for the transfer. It configures the primary channel control
+    /// structure to perform the transfer.
+    ///
+    /// You can use [Self::enable], [Self::enable_done_interrupt], [Self::enable_active_interrupt]
+    /// to finish the transfer preparation and then use [Self::trigger_with_sw_request] to
+    /// start the DMA transfer.
+    pub fn prepare_mem_to_mem_transfer_32_bit<'dest>(
+        &mut self,
+        source: &[u32],
+        dest: &'dest mut [u32],
+    ) -> Result<(), DmaTransferInitError> {
+        let len = Self::common_mem_transfer_checks(source.len(), dest.len())?;
+        self.generic_mem_to_mem_transfer_init(
+            len,
+            (source.as_ptr() as u32)
+                .checked_add(len as u32 * core::mem::size_of::<u32>() as u32)
+                .ok_or(DmaTransferInitError::AddrOverflow)?,
+            (dest.as_ptr() as u32)
+                .checked_add(len as u32 * core::mem::size_of::<u32>() as u32)
+                .ok_or(DmaTransferInitError::AddrOverflow)?,
+            DataSize::Word,
+            AddrIncrement::Word,
+        );
+        Ok(())
+    }
+
+    // This function performs common checks and returns the source length minus one which is
+    // relevant for further configuration of the DMA. This is because the DMA API expects N minus
+    // 1 and the source and end pointer need to point to the last transfer address.
+    fn common_mem_transfer_checks(
+        src_len: usize,
+        dest_len: usize,
+    ) -> Result<usize, DmaTransferInitError> {
+        if src_len != dest_len {
+            return Err(DmaTransferInitError::SourceDestLenMissmatch { src_len, dest_len });
+        }
+        if src_len > MAX_DMA_TRANSFERS_PER_CYCLE {
+            return Err(DmaTransferInitError::TransferSizeTooLarge(src_len));
+        }
+        Ok(src_len - 1)
+    }
+
+    fn generic_mem_to_mem_transfer_init(
+        &mut self,
+        n_minus_one: usize,
+        src_end_ptr: u32,
+        dest_end_ptr: u32,
+        data_size: DataSize,
+        addr_incr: AddrIncrement,
+    ) {
+        self.ch_ctrl_pri.cfg.set_raw(0);
+        self.ch_ctrl_pri.src_end_ptr = src_end_ptr;
+        self.ch_ctrl_pri.dest_end_ptr = dest_end_ptr;
+        self.ch_ctrl_pri.cfg.set_cycle_ctr(CycleControl::Auto as u8);
+        self.ch_ctrl_pri.cfg.set_src_size(data_size as u8);
+        self.ch_ctrl_pri.cfg.set_src_inc(addr_incr as u8);
+        self.ch_ctrl_pri.cfg.set_dst_size(data_size as u8);
+        self.ch_ctrl_pri.cfg.set_dst_inc(addr_incr as u8);
+        self.ch_ctrl_pri.cfg.set_n_minus_1(n_minus_one as u16);
+        self.ch_ctrl_pri.cfg.set_r_power(RPower::Every4 as u8);
+        self.select_primary_structure();
+    }
+}
+
+impl Dma {
+    /// Create a new DMA instance.
+    ///
+    /// You can also place the [DmaCtrlBlock] statically using a global static mutable
+    /// instance and the [DmaCtrlBlock::new] const constructor This also allows to place the control
+    /// block in a memory section using the [link_section](https://doc.rust-lang.org/reference/abi.html#the-link_section-attribute)
+    /// attribute and then creating a mutable pointer to it using [core::ptr::addr_of_mut].
+    ///
+    /// Alternatively, the [DmaCtrlBlock::new_at_addr] function can be used to create the DMA
+    /// control block at a specific address.
+    pub fn new(
+        syscfg: &mut pac::Sysconfig,
+        dma: pac::Dma,
+        cfg: DmaCfg,
+        ctrl_block: *mut DmaCtrlBlock,
+    ) -> Result<Self, InvalidCtrlBlockAddr> {
+        // The conversion to u32 is safe here because we are on a 32-bit system.
+        let raw_addr = ctrl_block as u32;
+        if raw_addr & BASE_PTR_ADDR_MASK > 0 {
+            return Err(InvalidCtrlBlockAddr);
+        }
+        syscfg.enable_peripheral_clock(PeripheralClock::Dma);
+        syscfg.assert_periph_reset_for_two_cycles(PeripheralSelect::Dma);
+        let dma = Dma { dma, ctrl_block };
+        dma.dma
+            .ctrl_base_ptr()
+            .write(|w| unsafe { w.bits(raw_addr) });
+        dma.set_protection_bits(&cfg);
+        dma.enable();
+        Ok(dma)
+    }
+
+    #[inline(always)]
+    pub fn enable(&self) {
+        self.dma.cfg().write(|w| w.master_enable().set_bit());
+    }
+
+    #[inline(always)]
+    pub fn disable(&self) {
+        self.dma.cfg().write(|w| w.master_enable().clear_bit());
+    }
+
+    #[inline(always)]
+    pub fn set_protection_bits(&self, cfg: &DmaCfg) {
+        self.dma.cfg().write(|w| unsafe {
+            w.chnl_prot_ctrl().bits(
+                cfg.privileged as u8 | ((cfg.bufferable as u8) << 1) | ((cfg.cacheable as u8) << 2),
+            )
+        });
+    }
+
+    /// Split the DMA instance into four DMA channels which can be used individually. This allows
+    /// using the inidividual DMA channels in separate tasks.
+    pub fn split(self) -> (DmaChannel, DmaChannel, DmaChannel, DmaChannel) {
+        // Safety: The DMA channel API only operates on its respective channels.
+        (
+            DmaChannel {
+                channel: 0,
+                done_interrupt: pac::Interrupt::DMA_DONE0,
+                active_interrupt: pac::Interrupt::DMA_ACTIVE0,
+                dma: unsafe { pac::Dma::steal() },
+                ch_ctrl_pri: unsafe { &mut (*self.ctrl_block).pri[0] },
+                ch_ctrl_alt: unsafe { &mut (*self.ctrl_block).alt[0] },
+            },
+            DmaChannel {
+                channel: 1,
+                done_interrupt: pac::Interrupt::DMA_DONE1,
+                active_interrupt: pac::Interrupt::DMA_ACTIVE1,
+                dma: unsafe { pac::Dma::steal() },
+                ch_ctrl_pri: unsafe { &mut (*self.ctrl_block).pri[1] },
+                ch_ctrl_alt: unsafe { &mut (*self.ctrl_block).alt[1] },
+            },
+            DmaChannel {
+                channel: 2,
+                done_interrupt: pac::Interrupt::DMA_DONE2,
+                active_interrupt: pac::Interrupt::DMA_ACTIVE2,
+                dma: unsafe { pac::Dma::steal() },
+                ch_ctrl_pri: unsafe { &mut (*self.ctrl_block).pri[2] },
+                ch_ctrl_alt: unsafe { &mut (*self.ctrl_block).alt[2] },
+            },
+            DmaChannel {
+                channel: 3,
+                done_interrupt: pac::Interrupt::DMA_DONE3,
+                active_interrupt: pac::Interrupt::DMA_ACTIVE3,
+                dma: unsafe { pac::Dma::steal() },
+                ch_ctrl_pri: unsafe { &mut (*self.ctrl_block).pri[3] },
+                ch_ctrl_alt: unsafe { &mut (*self.ctrl_block).alt[3] },
+            },
+        )
+    }
+}

va416xx-hal/src/lib.rs

@@ -11,6 +11,7 @@ pub mod prelude;
 pub mod adc;
 pub mod clock;
 pub mod dac;
+pub mod dma;
 pub mod gpio;
 pub mod i2c;
 pub mod pwm;