update for usbd-uac2 0.1.0, dma pipeline

Cargo.lock

@@ -181,7 +181,7 @@ dependencies = [
  "bare-metal",
  "bitfield 0.13.2",
  "critical-section",
- "embedded-hal 0.2.7",
+ "embedded-hal",
  "volatile-register",
 ]
 
@@ -292,12 +292,6 @@ dependencies = [
  "void",
 ]
 
-[[package]]
-name = "embedded-hal"
-version = "1.0.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "361a90feb7004eca4019fb28352a9465666b24f840f5c3cddf0ff13920590b89"
-
 [[package]]
 name = "embedded-io"
 version = "0.7.1"
@@ -365,7 +359,7 @@ dependencies = [
  "cortex-m-rt",
  "defmt 1.0.1",
  "defmt-rtt",
- "embedded-hal 1.0.0",
+ "embedded-hal",
  "embedded-io",
  "log-to-defmt",
  "lpc55-hal",
@@ -491,12 +485,13 @@ dependencies = [
 [[package]]
 name = "lpc55-hal"
 version = "0.5.0"
+source = "git+https://github.com/ktims/lpc55-hal?branch=main#8dfefd62aff4abd2de535f23107812dda68437be"
 dependencies = [
  "block-buffer",
  "cipher",
  "cortex-m",
  "digest",
- "embedded-hal 0.2.7",
+ "embedded-hal",
  "embedded-time",
  "generic-array 1.4.1",
  "lpc55-pac",
@@ -1084,6 +1079,8 @@ dependencies = [
 [[package]]
 name = "usbd-uac2"
 version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "117c289dcd316caa7aca9c0909117d8cf9d35f3ed2e7a5739067f0bcedc93e35"
 dependencies = [
  "byteorder-embedded-io",
  "defmt 1.0.1",

Cargo.toml

@@ -18,17 +18,18 @@ cortex-m = { version = "0.7.7", features = ["critical-section-single-core"] }
 cortex-m-rt = "0.7.5"
 defmt = "1.0.1"
 defmt-rtt = "1.1.0"
-embedded-hal = "1.0.0"
+embedded-hal = "0.2.7"
 embedded-io = "0.7.1"
 log-to-defmt = "0.1.0"
-lpc55-hal = { version = "0.5.0", path = "../usbd_uac2/examples/lpc55-hal" }
+# Includes update to usb-device 0.3, fix for isochronous and smaller critical sections
+lpc55-hal = { git = "https://github.com/ktims/lpc55-hal", branch = "main" }
 nb = "1.1.0"
 panic-halt = "1.0.0"
 panic-probe = { version = "1.0.0", features = ["print-defmt"] }
 static_cell = "2.1.1"
-usb-device = "0.3"
+usb-device = { version = "0.3", features = ["control-buffer-256"] }
 usbd-hid = { version = "0.10.0", optional = true }
-usbd-uac2 = { version = "0.1.0", path = "../usbd_uac2", features = ["defmt"]}
+usbd-uac2 = { version = "0.1.0", features = ["defmt"]}
 
 [profile.release]
 opt-level = "z"

src/dma.rs

@@ -0,0 +1,427 @@
+use hal::Syscon;
+use hal::peripherals::syscon::ClockControl;
+
+use crate::{hal, pac};
+use core::cell::UnsafeCell;
+use core::convert::Infallible;
+use core::ptr::copy_nonoverlapping;
+use core::sync::atomic::{AtomicUsize, Ordering, compiler_fence};
+
+pub const DMA0_FLEXCOMM7_TX: u8 = 19;
+
+#[repr(C)]
+#[derive(Copy, Clone)]
+pub struct DmaDescriptor {
+    pub xfercfg: u32,
+    pub src_end: *const u8,
+    pub dst_end: *mut u32,
+    pub next: *const DmaDescriptor,
+}
+
+impl defmt::Format for DmaDescriptor {
+    fn format(&self, fmt: defmt::Formatter) {
+        defmt::write!(
+            fmt,
+            "xfercfg={:x} src_end={:x} dst_end={:x} next={:x}",
+            self.xfercfg,
+            self.src_end,
+            self.dst_end,
+            self.next
+        )
+    }
+}
+
+// Channel descriptor table; linked from SRAMBASE
+#[repr(C, align(512))]
+pub struct DescriptorTable {
+    pub d: [DmaDescriptor; 32],
+}
+// Our ring that we will transition to once the transfer begins
+#[repr(C)]
+pub struct RingDescriptors<const N: usize> {
+    pub d: [DmaDescriptor; N],
+}
+
+#[derive(Debug, Copy, Clone, Eq, PartialEq)]
+pub struct PushResult {
+    pub written: usize,
+    pub dropped: usize,
+}
+
+#[derive(Debug, Copy, Clone, Eq, PartialEq)]
+pub enum ConfigError {
+    SlotTooLarge,
+    SlotTooSmall,
+    SlotNotAligned,
+    UnsupportedWidth,
+}
+
+#[derive(Debug)]
+pub enum DmaError {
+    Underrun,
+}
+impl core::error::Error for DmaError {}
+impl core::fmt::Display for DmaError {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        f.write_str("DmaUnderrun")
+    }
+}
+
+/// Slot-based DMA ring
+pub struct DmaRing<const N: usize, const MAX_SLOT_BYTES: usize> {
+    dma: pac::DMA0,
+
+    /// Destination peripheral register (FIFO write register)
+    dst_reg: *mut u32,
+
+    // SAFETY: only written by USB task (on start)
+    pub(crate) channel_desc: UnsafeCell<DescriptorTable>,
+    // SAFETY: only written by USB task (on start)
+    pub(crate) desc: UnsafeCell<RingDescriptors<N>>,
+    slots: UnsafeCell<[[u8; MAX_SLOT_BYTES]; N]>,
+
+    /// Effective bytes per slot. Maybe be smaller than MAX_SLOT_BYTES (e.g. at lower sample rates), as the setup is designed for constant rate not constant size.
+    slot_bytes: usize,
+    /// How many bytes to transfer to the FIFO
+    word_bytes: usize,
+
+    // SAFETY: producer only
+    write_slot: UnsafeCell<usize>,
+    write_off: UnsafeCell<usize>,
+
+    produced: AtomicUsize,
+    consumed: AtomicUsize,
+
+    /// Leave at least one slot empty so producer never overwrites a slot DMA may still read.
+    safety_gap: usize,
+    pub produced_bytes: AtomicUsize,
+    pub consumed_bytes: AtomicUsize,
+}
+
+impl<const N: usize, const MAX_SLOT_BYTES: usize> DmaRing<N, MAX_SLOT_BYTES> {
+    /// Construct using PAC DMA0 + &mut SYSCON + a destination FIFO register.
+    pub fn new(
+        dma: pac::DMA0,
+        syscon: &mut Syscon,
+        dst_reg: *mut u32,
+        word_bytes: usize,
+    ) -> Result<Self, ConfigError> {
+        if word_bytes != 1 && word_bytes != 2 && word_bytes != 4 {
+            return Err(ConfigError::UnsupportedWidth);
+        }
+        // Start the DMA0 clock
+        dma.enable_clock(syscon);
+
+        Ok(Self {
+            dma,
+            dst_reg: dst_reg,
+            channel_desc: UnsafeCell::new(DescriptorTable {
+                d: [DmaDescriptor {
+                    xfercfg: 0,
+                    src_end: core::ptr::null(),
+                    dst_end: core::ptr::null_mut(),
+                    next: core::ptr::null(),
+                }; 32],
+            }),
+            desc: UnsafeCell::new(RingDescriptors {
+                d: [DmaDescriptor {
+                    xfercfg: 0,
+                    src_end: core::ptr::null(),
+                    dst_end: core::ptr::null_mut(),
+                    next: core::ptr::null(),
+                }; N],
+            }),
+            slots: UnsafeCell::new([[0u8; MAX_SLOT_BYTES]; N]),
+            slot_bytes: MAX_SLOT_BYTES,
+            word_bytes,
+            write_slot: UnsafeCell::new(0),
+            write_off: UnsafeCell::new(0),
+            produced: AtomicUsize::new(0),
+            consumed: AtomicUsize::new(0),
+            safety_gap: 1,
+            produced_bytes: AtomicUsize::new(0),
+            consumed_bytes: AtomicUsize::new(0),
+        })
+    }
+
+    /// Optional: adjust safety gap (defaults to 1 empty slot).
+    pub fn set_safety_gap(&mut self, gap_slots: usize) {
+        self.safety_gap = gap_slots.min(N);
+    }
+    pub fn slot_size(&self) -> usize {
+        self.slot_bytes
+    }
+    pub fn set_slot_size(&mut self, slot_bytes: usize) -> Result<(), ConfigError> {
+        if slot_bytes == 0 {
+            return Err(ConfigError::SlotTooSmall);
+        }
+        if slot_bytes > MAX_SLOT_BYTES {
+            return Err(ConfigError::SlotTooLarge);
+        }
+        if slot_bytes % self.word_bytes != 0 {
+            return Err(ConfigError::SlotNotAligned);
+        }
+        self.slot_bytes = slot_bytes;
+        self.reset_producer();
+        Ok(())
+    }
+
+    /// Producer: copy into ring; commits whole slots; reports overflow by returning dropped bytes.
+    pub fn push(&self, mut data: &[u8]) -> PushResult {
+        let mut written = 0usize;
+
+        let write_slot = unsafe { &mut *self.write_slot.get() };
+        let write_off = unsafe { &mut *self.write_off.get() };
+
+        let slots = unsafe { &mut *self.slots.get() };
+        defmt::debug!(
+            "produced={} consumed={} fill={}",
+            self.produced(),
+            self.consumed(),
+            self.fill_slots()
+        );
+        while !data.is_empty() {
+            if self.is_full_for_producer() {
+                break;
+            }
+
+            let cap = self.slot_bytes - *write_off;
+            let n = core::cmp::min(cap, data.len());
+
+            unsafe {
+                let dst = slots[*write_slot].as_mut_ptr().add(*write_off);
+                copy_nonoverlapping(data.as_ptr(), dst, n);
+            }
+
+            *write_off += n;
+            written += n;
+            data = &data[n..];
+
+            if *write_off == self.slot_bytes {
+                // publish completed slot
+                compiler_fence(Ordering::Release);
+                self.produced.fetch_add(1, Ordering::Release);
+
+                *write_slot = (*write_slot + 1) % N;
+                *write_off = 0;
+            }
+        }
+
+        self.produced_bytes.fetch_add(written, Ordering::Release);
+
+        PushResult {
+            written,
+            dropped: data.len(),
+        }
+    }
+
+    /// Call from DMA IRQ bookkeeping when a slot has been consumed.
+    pub fn advance_consumed(&self, slots: usize) -> Result<(), DmaError> {
+        let produced = self.produced.load(Ordering::Acquire);
+        let consumed = self.consumed.load(Ordering::Relaxed);
+        if consumed < produced {
+            self.consumed.fetch_add(slots, Ordering::Release);
+            self.consumed_bytes
+                .fetch_add(slots * self.slot_bytes, Ordering::Relaxed);
+            Ok(())
+        } else {
+            defmt::error!("DMA underrun!");
+            Err(DmaError::Underrun)
+        }
+    }
+
+    pub fn produced(&self) -> usize {
+        self.produced.load(Ordering::Acquire)
+    }
+    pub fn produced_bytes(&self) -> usize {
+        self.produced_bytes.load(Ordering::Acquire)
+    }
+    pub fn consumed(&self) -> usize {
+        self.consumed.load(Ordering::Acquire)
+    }
+    pub fn consumed_bytes(&self) -> usize {
+        loop {
+            let consumed_start = self.consumed.load(Ordering::Acquire);
+
+            let reg_1 = self.dma.channel19.xfercfg.read().bits() as usize >> 16 & 0x3ff;
+            let reg_2 = self.dma.channel19.xfercfg.read().bits() as usize >> 16 & 0x3ff;
+
+            let consumed_end = self.consumed.load(Ordering::Acquire);
+
+            if consumed_start == consumed_end && reg_1 == reg_2 {
+                // 1. Map the hardware remaining countdown into a clean byte count
+                let remaining_bytes = if reg_1 == 0x3ff {
+                    0 // 0x3FF means all transfers completed, 0 bytes remaining
+                } else {
+                    // Formula from NXP manual: (XFERCOUNT + 1) * Data Width
+                    (reg_1 + 1) * self.word_bytes
+                };
+
+                // 2. Total bytes consumed in this specific active slot
+                let active_slot_consumed = self.slot_bytes - remaining_bytes;
+
+                // 3. Combine with your software index history accumulator
+                return consumed_start * self.slot_bytes + active_slot_consumed;
+            }
+        }
+    }
+
+    pub fn fill_slots(&self) -> usize {
+        self.produced().wrapping_sub(self.consumed())
+    }
+
+    pub fn init(&self) {
+        self.init_descriptors();
+
+        // Descriptor table base
+        let desc = unsafe { &*self.desc.get() };
+        let base = self.channel_desc.get() as u32;
+        self.dma.srambase.write(|w| unsafe { w.bits(base) });
+        self.dma
+            .channel19
+            .cfg
+            .write(|w| w.periphreqen().enabled().hwtrigen().disabled());
+        self.dma
+            .channel19
+            .xfercfg
+            .write(|w| unsafe { w.bits(desc.d[0].xfercfg) });
+
+        self.dma.enableclr0.write(|w| unsafe { w.bits(1 << 19) });
+        self.dma.ctrl.write(|w| w.enable().enabled());
+        self.dma.setvalid0.write(|w| unsafe { w.bits(1 << 19) });
+        self.dma.intenset0.write(|w| unsafe { w.bits(1 << 19) });
+
+        self.dma.settrig0.write(|w| unsafe { w.bits(1 << 19) });
+    }
+
+    pub fn run(&self) {
+        self.dma.enableset0.write(|w| unsafe { w.bits(1 << 19) });
+    }
+
+    pub fn stop(&self) {
+        self.dma.enableclr0.write(|w| unsafe { w.bits(1 << 19) });
+        nb::block!(if (self.dma.busy0.read().bits() & 1 << 19) == 0 {
+            Ok(())
+        } else {
+            Err(nb::Error::<Infallible>::WouldBlock)
+        });
+        self.dma.abort0.write(|w| unsafe { w.bits(1 << 19) });
+        self.reset_producer();
+    }
+
+    fn reset_producer(&self) {
+        unsafe {
+            *(&mut *self.write_slot.get()) = 0;
+            *(&mut *self.write_off.get()) = 0;
+        }
+        self.produced.store(0, Ordering::Relaxed);
+        self.produced_bytes.store(0, Ordering::Relaxed);
+        self.consumed.store(0, Ordering::Relaxed);
+        self.consumed_bytes.store(0, Ordering::Relaxed);
+    }
+
+    fn is_full_for_producer(&self) -> bool {
+        let fill = self.fill_slots();
+        fill >= N.wrapping_sub(self.safety_gap)
+    }
+    fn reset_producer_init_only(&self) {
+        unsafe {
+            *self.write_slot.get() = 0;
+        }
+        unsafe {
+            *self.write_off.get() = 0;
+        }
+
+        self.produced.store(0, Ordering::Relaxed);
+        self.consumed.store(0, Ordering::Relaxed);
+
+        self.produced_bytes.store(0, Ordering::Relaxed);
+        self.consumed_bytes.store(0, Ordering::Relaxed);
+    }
+
+    fn init_descriptors(&self) {
+        let slots = unsafe { &mut *self.slots.get() };
+        let desc = unsafe { &mut *self.desc.get() };
+        let chan_desc = unsafe { &mut *self.channel_desc.get() };
+        defmt::debug!("slots base: &{:x}", self.slots.get());
+
+        // Pre-fill with silence so underrun replays silence.
+        for i in 0..N {
+            slots[i][..self.slot_bytes].fill(0);
+        }
+
+        let transfers = (self.slot_bytes / self.word_bytes) as u32;
+
+        for i in 0..N {
+            let src_start = slots[i].as_ptr() as usize;
+            let src_end = (src_start + self.slot_bytes - self.word_bytes) as *const u8;
+
+            let next = &desc.d[(i + 1) % N] as *const DmaDescriptor;
+
+            desc.d[i] = DmaDescriptor {
+                xfercfg: encode_xfercfg(
+                    true,  // valid
+                    true,  // reload
+                    false, // swtrig (we use XFERCFG SWTRIG kick)
+                    false, // clrtrig
+                    true,  // intA
+                    false, // intB
+                    self.word_bytes as u32,
+                    1, // src_inc
+                    0, // dst_inc
+                    transfers,
+                ),
+                src_end,
+                dst_end: self.dst_reg,
+                next,
+            };
+        }
+        chan_desc.d[19] = desc.d[0];
+        chan_desc.d[19].xfercfg = 0;
+
+        // reset producer indices + counters (init-only action)
+        self.reset_producer_init_only();
+    }
+}
+
+unsafe impl<const N: usize, const MAX_SLOT_BYTES: usize> Sync for DmaRing<N, MAX_SLOT_BYTES> {}
+
+/// XFERCFG encoding follows the common LPC DMA layout:
+/// - SETINTA at bit4, SETINTB at bit5
+/// - WIDTH at bits 9:8
+/// - SRCINC at bits 13:12
+/// - DSTINC at bits 15:14
+/// - XFERCOUNT at bits 25:16
+/// This layout is shown in LPC DMA examples. [5](https://www.kernel.org/doc/html/latest/core-api/dma-api-howto.html)
+fn encode_xfercfg(
+    cfgvalid: bool,
+    reload: bool,
+    swtrig: bool,
+    clrtrig: bool,
+    inta: bool,
+    intb: bool,
+    width_bytes: u32,
+    src_inc: u32,
+    dst_inc: u32,
+    transfers: u32,
+) -> u32 {
+    let width_code = match width_bytes {
+        1 => 0,
+        2 => 1,
+        4 => 2,
+        _ => 0,
+    };
+
+    let count_field = transfers.saturating_sub(1) & 0x3FF;
+
+    ((cfgvalid as u32) << 0)
+        | ((reload as u32) << 1)
+        | ((swtrig as u32) << 2)
+        | ((clrtrig as u32) << 3)
+        | ((inta as u32) << 4)
+        | ((intb as u32) << 5)
+        | ((width_code & 0x3) << 8)
+        | ((src_inc & 0x3) << 12)
+        | ((dst_inc & 0x3) << 14)
+        | (count_field << 16)
+}

src/main.rs

@@ -8,12 +8,8 @@ fn panic() -> ! {
 }
 
 use atomic::Atomic;
-use bbqueue::nicknames::Churrasco;
-use bbqueue::prod_cons::stream::{StreamConsumer, StreamProducer};
-use bbqueue::traits::bbqhdl::BbqHandle;
-use bbqueue::traits::coordination::ReadGrantError;
 use bytemuck::NoUninit;
-use core::sync::atomic::{AtomicBool, AtomicI32, AtomicU32, AtomicUsize, Ordering};
+use core::sync::atomic::{AtomicBool, AtomicI32, AtomicUsize, Ordering};
 use cortex_m_rt::entry;
 use defmt;
 use defmt::debug;
@@ -26,24 +22,23 @@ use hal::raw as pac;
 use hal::time::{Hertz, Microseconds};
 use hal::typestates::pin::state::Gpio;
 use lpc55_hal as hal;
-use lpc55_hal::raw::NVIC;
-use lpc55_hal::raw::sdif::FIFO;
 use pac::interrupt;
+use static_cell::StaticCell;
 use usb_device::{
     bus::{self},
-    device::{StringDescriptors, UsbDeviceBuilder, UsbVidPid},
-    endpoint::IsochronousSynchronizationType,
+    device::{StringDescriptors, UsbVidPid},
 };
 #[cfg(feature = "hid")]
 use usbd_hid::{descriptor::SerializedDescriptor, hid_class::HIDClass};
-use usbd_uac2::UsbIsochronousFeedback;
 use usbd_uac2::{
-    self, AudioClassConfig, RangeEntry, TerminalConfig, UsbAudioClass, UsbAudioClockImpl, UsbSpeed,
+    self, AudioHandler, ClockSource, RangeEntry, TerminalConfig, UsbAudioClassConfig,
+    UsbAudioClassError, UsbIsochronousFeedback, UsbSpeed,
     constants::{FunctionCode, TerminalType},
-    descriptors::{ChannelConfig, ClockType, FormatType1, LockDelay},
+    descriptors::ClockType,
 };
 
 use crate::dac::DacImpl;
+use crate::dma::DmaRing;
 use crate::hid::AudioTelemetryReport;
 use crate::traits::Dac;
 
@@ -63,20 +58,26 @@ pub mod dac {
     pub use self::noop::NoopDac as DacImpl;
 }
 
+mod dma;
 #[cfg(feature = "hid")]
 mod hid;
 mod hw;
 mod traits;
 
-// Fo = M/(N*2*P) * Fin
-// Fo = 3072/(125*2*8) * 16MHz = 24.576MHz
-//
-const FIFO_LENGTH: usize = 256; // frames
-const QUEUE_RUNNING_UP: usize = (FIFO_LENGTH * 4) / 10; // 40%
-const QUEUE_RUNNING_DOWN: usize = (FIFO_LENGTH * 2) / 10; // 20%
+const BYTES_PER_SAMPLE: usize = 4; // 32 bit samples
+const BYTES_PER_FRAME: usize = BYTES_PER_SAMPLE * 2; // 2 channels
+const FRAMES_PER_SLOT: usize = SAMPLE_RATE as usize / 2000; // run the DMA at 2khz
+const BYTES_PER_SLOT: usize = FRAMES_PER_SLOT * BYTES_PER_FRAME;
+const N_SLOTS: usize = 8;
+const FILL_TARGET_BYTES: i32 = (BYTES_PER_SLOT * N_SLOTS) as i32 / 2;
+const USB_FRAME_RATE: u32 = 8000; // microframe rate: 8000 for HS, 1000 for FS
+
+// In frames
+const QUEUE_RUNNING_UP: usize = ((FRAMES_PER_SLOT * N_SLOTS) * 4) / 10; // 40%
+const QUEUE_RUNNING_DOWN: usize = ((FRAMES_PER_SLOT * N_SLOTS) * 2) / 10; // 20%
 const NODATA_TIMEOUT_FRAMES: usize = SAMPLE_RATE as usize / 100; // ~100ms
 const MCLK_FREQ: u32 = 24576000;
-const SAMPLE_RATE: u32 = 88200;
+const SAMPLE_RATE: u32 = 192000;
 const HID_INTERVAL_MS: u8 = 100;
 
 struct CodecPins {
@@ -88,63 +89,6 @@ struct ClockSelPins {
     sel_22m: Pin<pins::Pio0_31, Gpio<Output>>,
 }
 
-struct Clock {
-    pins: ClockSelPins,
-    cur_rate: u32,
-}
-impl Clock {
-    const RATES: [RangeEntry<u32>; 1] = [RangeEntry::new_fixed(SAMPLE_RATE)];
-}
-impl UsbAudioClockImpl for Clock {
-    const CLOCK_TYPE: usbd_uac2::descriptors::ClockType = ClockType::InternalFixed;
-    const SOF_SYNC: bool = false;
-    fn get_sample_rate(&self) -> u32 {
-        self.cur_rate
-    }
-    fn set_sample_rate(
-        &mut self,
-        sample_rate: u32,
-    ) -> core::result::Result<(), usbd_uac2::UsbAudioClassError> {
-        if 24_576_000u32.is_multiple_of(sample_rate) {
-            defmt::info!("[clock] 24M clock selected");
-            self.pins.sel_22m.set_low().ok();
-            // hal::wait_at_least(1);
-            self.pins.sel_24m.set_high().ok();
-        } else {
-            defmt::info!("[clock] 22M clock selected");
-            self.pins.sel_24m.set_low().ok();
-            // hal::wait_at_least(1);
-            self.pins.sel_22m.set_high().ok();
-        };
-        self.cur_rate = sample_rate;
-        Ok(())
-    }
-    fn get_rates(
-        &self,
-    ) -> core::result::Result<&[usbd_uac2::RangeEntry<u32>], usbd_uac2::UsbAudioClassError> {
-        Ok(&Clock::RATES)
-    }
-    fn get_clock_validity(&self) -> core::result::Result<bool, usbd_uac2::UsbAudioClassError> {
-        Ok(true)
-    }
-    fn alt_setting(
-        &mut self,
-        alt_setting: u8,
-    ) -> core::result::Result<(), usbd_uac2::UsbAudioClassError> {
-        match alt_setting {
-            0 => {
-                self.pins.sel_22m.set_low().ok();
-                self.pins.sel_24m.set_low().ok();
-            }
-            1 => {
-                self.set_sample_rate(self.cur_rate).ok();
-            }
-            _ => {}
-        };
-        Ok(())
-    }
-}
-
 #[derive(Default)]
 struct PerfCounters {
     received_frames: AtomicUsize,
@@ -160,8 +104,10 @@ impl PerfCounters {
     fn reset(&self) {
         self.received_frames.store(0, Ordering::Relaxed);
         self.played_frames.store(0, Ordering::Relaxed);
-        self.min_fill.store(FIFO_LENGTH, Ordering::Relaxed);
-        self.avg_fill.store(FIFO_LENGTH / 2, Ordering::Relaxed);
+        self.min_fill
+            .store(N_SLOTS * BYTES_PER_SLOT, Ordering::Relaxed);
+        self.avg_fill
+            .store(FILL_TARGET_BYTES as usize, Ordering::Relaxed);
         self.queue_underflows.store(0, Ordering::Relaxed);
         self.queue_overflows.store(0, Ordering::Relaxed);
         self.audio_underflows.store(0, Ordering::Relaxed);
@@ -193,91 +139,69 @@ impl defmt::Format for PerfCounters {
     }
 }
 
-const BYTES_PER_FRAME: usize = 8;
-const QUEUE_BYTES: usize = FIFO_LENGTH * BYTES_PER_FRAME;
-// We use bbqueue here for performance in the USB driver that runs almost entirely in interrupt free critical section.
-static QUEUE: Churrasco<QUEUE_BYTES> = Churrasco::new();
-// Used for feedback calculation of current fifo state
-static PRODUCED: AtomicU32 = AtomicU32::new(0);
-static CONSUMED: AtomicU32 = AtomicU32::new(0);
-
 static PERF: PerfCounters = PerfCounters {
     received_frames: AtomicUsize::new(0), // received from USB
     played_frames: AtomicUsize::new(0),   // played audio frames
-    min_fill: AtomicUsize::new(FIFO_LENGTH), // not recording this for now, need to figure out how to make it meaningful, since the queue starts empty
-    avg_fill: AtomicUsize::new(FIFO_LENGTH / 2),
+    min_fill: AtomicUsize::new(0), // not recording this for now, need to figure out how to make it meaningful, since the queue starts empty
+    avg_fill: AtomicUsize::new(FILL_TARGET_BYTES as usize),
     queue_underflows: AtomicUsize::new(0), // ditto here, since we underflow at startup, but we record this one as it can be trended
     queue_overflows: AtomicUsize::new(0),
     audio_underflows: AtomicUsize::new(0),
 };
 
-fn cur_fill() -> u32 {
-    PRODUCED
-        .load(Ordering::Acquire)
-        .wrapping_sub(CONSUMED.load(Ordering::Acquire))
-        / BYTES_PER_FRAME as u32
+static DMA_RING: StaticCell<DmaRing<N_SLOTS, BYTES_PER_SLOT>> = StaticCell::new();
+static mut DMA_RING_REF: Option<&'static DmaRing<N_SLOTS, BYTES_PER_SLOT>> = None;
+#[inline]
+fn dma_ring() -> &'static DmaRing<N_SLOTS, BYTES_PER_SLOT> {
+    unsafe { DMA_RING_REF.unwrap() }
 }
 
-#[inline]
-fn try_write_one_frame<T: BbqHandle>(
-    cons: &mut StreamConsumer<T>,
-    i2s: &pac::i2s7::RegisterBlock,
-) -> bool {
-    match cons.read() {
-        Ok(rgr) => {
-            // TODO: Fix this to handle the case where frame lands on a ring buffer boundary (if it is possible)
-            if rgr.len() >= BYTES_PER_FRAME {
-                let l = u32::from_le_bytes(rgr[0..4].try_into().unwrap());
-                let r = u32::from_le_bytes(rgr[4..8].try_into().unwrap());
+fn cur_fill() -> usize {
+    let produced_bytes = dma_ring().produced_bytes() as u32;
+    let consumed_bytes = dma_ring().consumed_bytes() as u32;
 
-                i2s.fifowr.write(|w| unsafe { w.bits(l) });
-                i2s.fifowr.write(|w| unsafe { w.bits(r) });
-
-                // consume exactly one frame (8 bytes)
-                rgr.release(BYTES_PER_FRAME);
-                PERF.played_frames.fetch_add(1, Ordering::Relaxed);
-                CONSUMED.fetch_add(BYTES_PER_FRAME as u32, Ordering::Relaxed);
-                return true;
-            } else {
-                // Not enough bytes for a full frame: leave it in the queue.
-                return false;
-            }
-        }
-        Err(ReadGrantError::Empty) => {
-            return false;
-        }
-        Err(e) => {
-            defmt::error!("Unexpected queue read error")
-        }
-    }
-    false
+    // Handle rollover properly
+    produced_bytes.wrapping_sub(consumed_bytes) as usize
 }
 
 #[interrupt]
-fn FLEXCOMM7() {
-    let i2s = unsafe { &*pac::I2S7::ptr() };
-    defmt::info!("isr");
+fn DMA0() {
+    defmt::debug!("dma0");
+    let dma = unsafe { &*pac::DMA0::ptr() };
 
-    if i2s.fifostat.read().txlvl().bits() == 0 {
-        // ISR was not serviced before the FIFO drained
-        PERF.audio_underflows.fetch_add(1, Ordering::Relaxed);
+    let inta = dma.inta0.read().bits();
+    let err = dma.errint0.read().bits();
+
+    // TODO: figure out how to track underflows properly
+    if (err & (1 << 19)) != 0 {
+        let live = dma.channel19.xfercfg.read().bits();
+
+        let desc = unsafe { &*dma_ring().channel_desc.get() };
+        let mem = desc.d[19];
+        defmt::error!(
+            "DMA error ch19: live={=u32:08x} INTA={=u32:x} ERR={=u32:x}\n         desc: {}",
+            live,
+            inta,
+            err,
+            mem
+        );
+        // red_led().on();
+        dma.errint0.write(|w| unsafe { w.bits(1 << 19) });
     }
 
-    // refil the buffer to 4 frames / 8 samples
-    let mut cons = QUEUE.stream_consumer();
-    while i2s.fifostat.read().txlvl().bits() <= 6 {
-        if !try_write_one_frame(&mut cons, i2s) {
-            // No complete frame available: write silence to keep FIFO above threshold or we will
-            // get stuck in the ISR.
-            PERF.queue_underflows.fetch_add(1, Ordering::Relaxed);
-            i2s.fifowr.write(|w| unsafe { w.bits(0) });
-            i2s.fifowr.write(|w| unsafe { w.bits(0) });
-            break;
+    if (inta & (1 << 19)) != 0 {
+        dma.inta0.write(|w| unsafe { w.bits(1 << 19) });
+        if dma_ring().advance_consumed(1).is_err() {
+            // red_led().on();
+        } else {
+            PERF.played_frames
+                .fetch_add(FRAMES_PER_SLOT, Ordering::Relaxed);
         }
     }
 }
+
 #[repr(u8)]
-#[derive(Clone, Copy, NoUninit)]
+#[derive(Clone, Copy, NoUninit, Eq, PartialEq)]
 enum AudioState {
     /// Knowingly stopped, ie. AltSetting=0. DAC muted, I2S disabled.
     ///  
@@ -343,7 +267,7 @@ impl FeedbackState {
         self.correction_enabled.store(false, Ordering::Relaxed);
         self.integrator.store(0, Ordering::Relaxed);
         self.filtered_fill
-            .store(FIFO_LENGTH as i32 / 2, Ordering::Relaxed);
+            .store(FILL_TARGET_BYTES, Ordering::Relaxed);
     }
 }
 impl Default for FeedbackState {
@@ -351,22 +275,25 @@ impl Default for FeedbackState {
         Self {
             correction_enabled: AtomicBool::new(false),
             integrator: AtomicI32::new(0),
-            filtered_fill: AtomicI32::new(FIFO_LENGTH as i32 / 2),
+            filtered_fill: AtomicI32::new(FILL_TARGET_BYTES),
         }
     }
 }
 
-struct Audio<T: BbqHandle, D: Dac<I>, I> {
+struct Audio<'a, D: Dac<I>, I> {
     state: Atomic<AudioState>,
     alt_setting: u8,
     i2s: I2sTx,
     dac: D,
-    producer: StreamProducer<T>,
+    dma: &'a DmaRing<N_SLOTS, BYTES_PER_SLOT>,
     fb: FeedbackState,
     nodata_timeout_frame: AtomicUsize,
+    cur_rate: u32,
+    clock_pins: ClockSelPins,
     _marker: core::marker::PhantomData<I>,
 }
-impl<T: BbqHandle, D: Dac<I>, I> Audio<T, D, I> {
+impl<D: Dac<I>, I> Audio<'_, D, I> {
+    const RATES: [RangeEntry<u32>; 1] = [RangeEntry::new_fixed(SAMPLE_RATE)];
     /// Perform a state transition to `state`
     fn transition(&mut self, state: AudioState) {
         defmt::info!(
@@ -425,14 +352,13 @@ impl<T: BbqHandle, D: Dac<I>, I> Audio<T, D, I> {
                 .normal()
         });
         self.dac.init();
-        self.dac.change_rate(SAMPLE_RATE);
     }
     ///Transition -> Stopped:
     ///clear queue, mute DAC, mask I2S ISR, stop I2S peripheral, disable & reset feedback and performance queues
     fn stop(&mut self) {
+        dma_ring().stop();
+        pac::NVIC::mask(pac::Interrupt::DMA0);
         self.dac.mute();
-        // Disable level interrupt on I2S
-        self.i2s.i2s.fifointenclr.write(|w| w.txlvl().set_bit());
         // Clear any samples in the FIFO
         self.i2s.i2s.fifocfg.modify(|_, w| w.emptytx().set_bit());
         // Disable I2S
@@ -441,17 +367,17 @@ impl<T: BbqHandle, D: Dac<I>, I> Audio<T, D, I> {
         self.fb.reset();
         // reset performance counters
         PERF.reset();
-        // Drain anything left in the queue
-        while let Ok(d) = QUEUE.stream_consumer().read() {
-            let len = d.len();
-            d.release(len)
-        }
+        // Stop the clocks
+        self.clock_pins.sel_22m.set_low().ok();
+        self.clock_pins.sel_24m.set_low().ok();
     }
     ///Transition -> Armed
-    /// Start I2S peripheral. Since we assume we have interrupts disabled at
+    /// Start I2S peripheral and MCLK. Since we assume we have interrupts disabled at
     /// this point (as we came from Stopped), and the FIFO is empty, this will
     /// play out 0s.
     fn arm(&mut self) {
+        dma_ring().init();
+        self.set_sample_rate(self.cur_rate).ok();
         self.i2s.i2s.cfg1.modify(|_, w| w.mainenable().enabled());
     }
     ///Transition -> Prefill
@@ -468,8 +394,14 @@ impl<T: BbqHandle, D: Dac<I>, I> Audio<T, D, I> {
             .i2s
             .fifotrig
             .modify(|_, w| unsafe { w.txlvl().bits(6).txlvlena().enabled() });
-        // FIFO level interrupt enable
-        self.i2s.i2s.fifointenset.modify(|_, w| w.txlvl().enabled());
+        self.i2s
+            .i2s
+            .fifocfg
+            .modify(|_, w| w.enabletx().enabled().dmatx().enabled());
+        dma_ring().run();
+        unsafe {
+            pac::NVIC::unmask(pac::Interrupt::DMA0);
+        }
     }
     ///Transition->NoData
     ///store framecount at transition so we can time out recovery
@@ -480,14 +412,49 @@ impl<T: BbqHandle, D: Dac<I>, I> Audio<T, D, I> {
         );
     }
 }
-impl<T: BbqHandle, D: Dac<I>, I, B: bus::UsbBus> UsbAudioClass<'_, B> for Audio<T, D, I> {
+impl<D: Dac<I>, I> ClockSource for Audio<'_, D, I> {
+    const CLOCK_TYPE: usbd_uac2::descriptors::ClockType = ClockType::InternalFixed;
+    const SOF_SYNC: bool = false;
+
+    fn sample_rate(&self) -> u32 {
+        self.cur_rate
+    }
+    fn set_sample_rate(
+        &mut self,
+        sample_rate: u32,
+    ) -> core::result::Result<(), usbd_uac2::UsbAudioClassError> {
+        if 24_576_000u32.is_multiple_of(sample_rate) {
+            defmt::info!("[clock] 24M clock selected");
+            self.clock_pins.sel_22m.set_low().ok();
+            // hal::wait_at_least(1);
+            self.clock_pins.sel_24m.set_high().ok();
+        } else {
+            defmt::info!("[clock] 22M clock selected");
+            self.clock_pins.sel_24m.set_low().ok();
+            // hal::wait_at_least(1);
+            self.clock_pins.sel_22m.set_high().ok();
+        };
+        self.dac.change_rate(sample_rate);
+        self.cur_rate = sample_rate;
+        Ok(())
+    }
+    fn sample_rates(
+        &self,
+    ) -> core::result::Result<&[usbd_uac2::RangeEntry<u32>], usbd_uac2::UsbAudioClassError> {
+        Ok(&Self::RATES)
+    }
+    fn clock_validity(&self) -> Result<bool, UsbAudioClassError> {
+        Ok(true)
+    }
+}
+impl<D: Dac<I>, I, B: bus::UsbBus> AudioHandler<'_, B> for Audio<'_, D, I> {
     fn alternate_setting_changed(&mut self, _terminal: usb_device::UsbDirection, alt_setting: u8) {
         let state = self.state.load(Ordering::Relaxed);
         match (alt_setting, state) {
             (0, AudioState::Armed | AudioState::Prefill | AudioState::NoData) => {
                 self.transition(AudioState::Stopped)
             }
-            (0, AudioState::Running) => {} // noop, we naturally transition through LowData to Stopped
+            (0, AudioState::Running | AudioState::Stopped) => {} // noop, we naturally transition through LowData to Stopped
             (1, AudioState::Stopped) => self.transition(AudioState::Armed),
             (1, _) => {} // altSetting 1 in any other state is a no-op
             (_, _) => {
@@ -501,27 +468,32 @@ impl<T: BbqHandle, D: Dac<I>, I, B: bus::UsbBus> UsbAudioClass<'_, B> for Audio<
         ep: &usb_device::endpoint::Endpoint<'_, B, usb_device::endpoint::Out>,
     ) {
         let state = self.state.load(Ordering::Relaxed);
-        let mut buf = [0; SAMPLE_RATE as usize / 1000 * 64];
+        let mut buf = [0; (SAMPLE_RATE.div_ceil(USB_FRAME_RATE) + 1) as usize * BYTES_PER_FRAME];
         let len = match ep.read(&mut buf) {
             Ok(len) => len,
-            Err(e) => {
-                defmt::error!("usb error in rx callback {:?}", e);
+            Err(_) => {
+                defmt::error!("usb error in rx callback");
                 return;
             }
         };
         let buf = &buf[..len];
+        let res = self.dma.push(buf);
 
-        if let Ok(mut wg) = self.producer.grant_exact(buf.len()) {
-            wg.copy_from_slice(buf);
-            wg.commit(buf.len());
-            PRODUCED.fetch_add(buf.len() as u32, Ordering::Relaxed);
-            PERF.received_frames
-                .fetch_add(buf.len() / BYTES_PER_FRAME, Ordering::Relaxed);
-        } else {
-            PERF.queue_overflows.fetch_add(1, Ordering::Relaxed);
-            // defmt::error!("overflowed bbq, asked {}", buf.len());
+        if res.dropped != 0 {
+            // Overflow: some or all bytes couldn't be queued.
+            defmt::error!(
+                "overflowed dma ring, asked {}, wrote {}, dropped {}",
+                buf.len(),
+                res.written,
+                res.dropped
+            );
+            PERF.queue_overflows
+                .fetch_add(res.dropped / BYTES_PER_FRAME, Ordering::Relaxed);
         }
 
+        PERF.received_frames
+            .fetch_add(res.written / BYTES_PER_FRAME, Ordering::Relaxed);
+
         // Valid states here are Armed, Prefill, Running, Draining and NoData
         match state {
             AudioState::Stopped => {
@@ -553,47 +525,44 @@ impl<T: BbqHandle, D: Dac<I>, I, B: bus::UsbBus> UsbAudioClass<'_, B> for Audio<
             AudioState::NoData => self.transition(AudioState::LowData),
         }
     }
-    fn feedback(&mut self, nominal_rate: UsbIsochronousFeedback) -> Option<UsbIsochronousFeedback> {
-        let target = FIFO_LENGTH as i32 / 2 - nominal_rate.int as i32;
-
-        let fill = cur_fill() as i32;
-        let prev = self.fb.filtered_fill.load(Ordering::Relaxed);
-        let filtered = prev + ((fill - prev) >> 4); // ~1/16 smoothing
-        self.fb.filtered_fill.store(filtered, Ordering::Relaxed);
-
-        let error = filtered - target;
-
-        // Clamp startup excursions.
-        let error = error.clamp(-(nominal_rate.int as i32 * 4), nominal_rate.int as i32 * 4);
-
-        // Reset integrator when the error is small
-        if error.abs() < 2 {
-            self.fb.integrator.store(0, Ordering::Relaxed);
+    fn audio_data_tx(
+        &mut self,
+        _ep: &usb_device::endpoint::Endpoint<'_, B, usb_device::endpoint::In>,
+    ) {
+    }
+    fn feedback(&mut self, nominal_rate: UsbIsochronousFeedback) -> Option<UsbIsochronousFeedback> {
+        if !self.fb.correction_enabled.load(Ordering::Relaxed) {
+            return Some(nominal_rate);
         }
-        let mut integrator = self.fb.integrator.load(Ordering::Relaxed);
-        integrator = integrator - (integrator >> 6); // ~1/64 leak, reduce windup
 
-        integrator = integrator.clamp(-256, 256);
-        self.fb.integrator.store(integrator, Ordering::Relaxed);
+        let current_bytes = cur_fill() as i32;
 
-        // gains
-        let p = error << 3;
-        let i = integrator << 2;
+        if current_bytes == 0 {
+            defmt::error!("[fb] dma underrun detected!");
+            PERF.queue_underflows.fetch_add(1, Ordering::Relaxed);
+            return Some(nominal_rate);
+        }
+        PERF.avg_fill
+            .fetch_update(Ordering::SeqCst, Ordering::SeqCst, |v| {
+                Some(((v << 6) - v + current_bytes as usize) >> 6)
+            })
+            .ok();
+
+        // normalize error wrt. frame size etc.
+        let error_permille = ((current_bytes - FILL_TARGET_BYTES) * 1000) / FILL_TARGET_BYTES;
 
-        let correction = -((p + i) >> 2);
         let nominal_v = nominal_rate.to_u32_12_13() as i32;
 
-        let mut v = nominal_v + correction;
+        // 0.2% which is a huge clock error
+        let max_allowed_deviation = nominal_v / 500;
 
-        // Tight clamp around nominal.
-        v = v.clamp(nominal_v - (1 << 12), nominal_v + (1 << 12));
+        let p_term = -(error_permille * nominal_v) / 256000; // this works reasonably well to keep the buffer
+        let i_term = 0; // placeholder
 
-        defmt::debug!(
-            "fill:{} err:{} int:{} fb:{=i32:x}",
-            fill,
-            error,
-            integrator,
-            v
+        let mut v = nominal_v + p_term + i_term;
+        v = v.clamp(
+            nominal_v - max_allowed_deviation,
+            nominal_v + max_allowed_deviation,
         );
 
         Some(UsbIsochronousFeedback::new(v as u32))
@@ -640,8 +609,6 @@ pub fn init_i2s(mut fc7: pac::FLEXCOMM7, i2s7: pac::I2S7, syscon: &mut Syscon) -
     // Select I2S TX function
     fc7.pselid.write(|w| w.persel().i2s_transmit());
 
-    unsafe { NVIC::unmask(interrupt::FLEXCOMM7) }
-
     let regs = i2s7;
     I2sTx { i2s: regs }
 }
@@ -740,55 +707,49 @@ fn main() -> ! {
         clocks.support_usbhs_token().unwrap(),
     );
 
+    defmt::info!("dma init");
+    let i2s_dma_addr = &i2s_peripheral.i2s.fifowr as *const _ as *mut u32;
+    let dma =
+        DmaRing::<N_SLOTS, BYTES_PER_SLOT>::new(hal.dma.release(), &mut syscon, i2s_dma_addr, 4)
+            .unwrap();
+    let dma_ref = DMA_RING.init(dma);
+    unsafe { DMA_RING_REF = Some(dma_ref) };
+
     defmt::info!("audio init");
     let mut audio = Audio {
         state: Atomic::new(AudioState::Stopped),
         i2s: i2s_peripheral,
         dac: dac_impl,
-        producer: QUEUE.stream_producer(),
+        dma: dma_ring(),
         fb: FeedbackState::default(),
         alt_setting: 0,
         nodata_timeout_frame: AtomicUsize::new(0),
+        cur_rate: SAMPLE_RATE,
+        clock_pins: clock_sel_pins,
         _marker: core::marker::PhantomData,
     };
     audio.init();
 
     let usb_bus = UsbBus::new(usb_peripheral, usb0_vbus_pin);
-    let mut clock = Clock {
-        pins: clock_sel_pins,
-        cur_rate: SAMPLE_RATE,
-    };
-    let config = AudioClassConfig::new(UsbSpeed::High, FunctionCode::Other, &mut clock, &mut audio)
-        .with_output_config(TerminalConfig::new(
-            2,
-            1,
-            2,
-            FormatType1 {
-                bit_resolution: 32,
-                bytes_per_sample: 4,
-            },
-            TerminalType::OutHeadphones,
-            ChannelConfig::default_chans(2),
-            IsochronousSynchronizationType::Asynchronous,
-            LockDelay::Undefined(0),
-            None,
-        ));
+
+    let config = UsbAudioClassConfig::new(UsbSpeed::High, FunctionCode::Other, &mut audio)
+        .with_output_config(
+            TerminalConfig::builder()
+                .base_id(2)
+                .terminal_type(TerminalType::UsbUndefined)
+                .build(),
+        );
     let mut uac2 = config.build(&usb_bus).unwrap();
     #[cfg(feature = "hid")]
     let mut hid = HIDClass::new_ep_in(&usb_bus, AudioTelemetryReport::desc(), HID_INTERVAL_MS);
 
-    let mut usb_dev = UsbDeviceBuilder::new(&usb_bus, UsbVidPid(0x1209, 0xcc1d))
-        .composite_with_iads()
+    let mut usb_dev = usbd_uac2::builder(&usb_bus, UsbVidPid(0x1209, 0xcc1d))
         .strings(&[StringDescriptors::default()
             .manufacturer("VE7XEN")
-            .product("Guac Tortilla")
-            .serial_number("123456789")])
+            .product("Guac Tortilla")])
         .unwrap()
         .max_packet_size_0(64)
         .unwrap()
-        .device_class(0xef)
-        .device_sub_class(0x02)
-        .device_protocol(0x01)
         .build();
 
     #[cfg(feature = "hid")]