eth: use UncachedSlice for both desc and buffers

eth: latter tuning
eth: fix cache maintenance
2020-07-29 22:25:08 +02:00 · 2020-07-29 22:12:28 +02:00 · 2020-07-29 22:12:28 +02:00 · 2020-07-29 22:12:28 +02:00 · 2020-07-29 22:09:48 +02:00 · 2020-07-29 22:09:48 +02:00
37 changed files with 500 additions and 1267 deletions
--- a/Cargo.lock
+++ b/Cargo.lock
@ -79,6 +79,7 @@ version = "0.0.0"
 dependencies = [
 "bit_field 0.10.0 (registry+https://github.com/rust-lang/crates.io-index)",
 "libregister 0.0.0",
+ "volatile-register 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)",
 ]

 [[package]]
@ -98,13 +99,13 @@ dependencies = [
 "libboard_zynq 0.0.0",
 "libcortex_a9 0.0.0",
 "libregister 0.0.0",
- "linked_list_allocator 0.8.4 (registry+https://github.com/rust-lang/crates.io-index)",
+ "linked_list_allocator 0.8.3 (registry+https://github.com/rust-lang/crates.io-index)",
 "r0 1.0.0 (registry+https://github.com/rust-lang/crates.io-index)",
 ]

 [[package]]
 name = "linked_list_allocator"
-version = "0.8.4"
+version = "0.8.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"

 [[package]]
@ -170,7 +171,7 @@ dependencies = [
 "checksum cfg-if 0.1.10 (registry+https://github.com/rust-lang/crates.io-index)" = "4785bdd1c96b2a846b2bd7cc02e86b6b3dbf14e7e53446c4f54c92a361040822"
 "checksum compiler_builtins 0.1.27 (registry+https://github.com/rust-lang/crates.io-index)" = "38f18416546abfbf8d801c555a0e99524453e7214f9cc9107ad49de3d5948ccc"
 "checksum embedded-hal 0.2.3 (registry+https://github.com/rust-lang/crates.io-index)" = "ee4908a155094da7723c2d60d617b820061e3b4efcc3d9e293d206a5a76c170b"
-"checksum linked_list_allocator 0.8.4 (registry+https://github.com/rust-lang/crates.io-index)" = "e70e46c13c0e8374c26cec5752e3347ca1087d9711de8f45aa513a7700efd73d"
+"checksum linked_list_allocator 0.8.3 (registry+https://github.com/rust-lang/crates.io-index)" = "d6b60501dd4c850950bb43f970d544f6ce04e0ca021da2db2538fbe9d923f19e"
 "checksum log 0.4.8 (registry+https://github.com/rust-lang/crates.io-index)" = "14b6052be84e6b71ab17edffc2eeabf5c2c3ae1fdb464aae35ac50c67a44e1f7"
 "checksum managed 0.7.1 (registry+https://github.com/rust-lang/crates.io-index)" = "fdcec5e97041c7f0f1c5b7d93f12e57293c831c646f4cc7a5db59460c7ea8de6"
 "checksum nb 0.1.2 (registry+https://github.com/rust-lang/crates.io-index)" = "b1411551beb3c11dedfb0a90a0fa256b47d28b9ec2cdff34c25a2fa59e45dbdc"
--- a/README.md
+++ b/README.md
@ -1,7 +1,7 @@
 # Build

 ```shell
-nix-shell --command "cargo xbuild --release -p experiments"
+nix-shell --command "cargo xbuild --release"
 ```

 Currently the ELF output is placed at `target/armv7-none-eabihf/release/experiments`
@ -41,7 +41,7 @@ Proceed using gdb with `load`, `c`
 ### Running on the ZC706

 ```shell
-nix-shell --command "cargo xbuild --release -p experiments"
+nix-shell --command "cargo xbuild --release"
 cd openocd
 openocd -f zc706.cfg
 ```
@ -64,19 +64,16 @@ openocd -f zc706.cfg -c "pld load 0 blinker_migen.bit; exit"

 Clone this repo onto your development/build machine and the raspberry pi that controls the Xilinx 7000 board

-On the dev machine, the below script builds zc706 and secure copies it to the target pi (in your pi $HOME directory):
+On the dev machine, the below script builds zc706 and secure copies it to the target pi (in your pi $HOME directory)
 ```shell
-cd ~/zynq-rs
-./build.sh $your_user_or_ssh_id
+cd ~/zc706
+./build.sh $your_user/ssh_id
 ```

 On the pi, we need an information rich environment that includes a relatively reliable `gdb` experience (that includes `ctrl-p` and `ctrl-n` command history that persists across `cgdb` executions), run:
 ```shell
 ssh pi4
-cd zynq-rs
-# For ZC706, run:
-./tmux.sh 0
-# For Cora Z7, run:
+cd zc706
 ./tmux.sh
 ```

--- a/armv7-none-eabihf.json
+++ b/armv7-none-eabihf.json
@ -12,7 +12,7 @@
  "emit-debug-gdb-scripts": false,
  "env": "",
  "executables": true,
-  "features": "+v7,+vfp3,-d32,+thumb2,-neon",
+  "features": "+v7,+vfp3,-d32,+thumb2,-neon,+strict-align",
  "is-builtin": false,
  "linker": "rust-lld",
  "linker-flavor": "ld.lld",
--- a/build.sh
+++ b/build.sh
@ -1 +1 @@
-nix-shell --command "cargo xbuild --release -p experiments" && scp -C target/armv7-none-eabihf/release/experiments $1@rpi-4.m-labs.hk:/home/$1/zc706/zc706.elf
+nix-shell --command "cargo xbuild --release" && scp -P 2204 -C target/armv7-none-eabihf/release/zc706-experiments $1@nixbld.m-labs.hk:/home/$1/zc706/zc706.elf
--- a/default.nix
+++ b/default.nix
@ -49,7 +49,7 @@ let
    name = "${crate}";
    src = ./.;
    crateSubdir = crate;
-    cargoSha256 = "1c1qpg9by8bg93yhgllb5xs155g27qmh99pbrb681wazm8k7nwim";
+    cargoSha256 = "0xlynsr94dyv0g41qwk5490w3wnzd5g70msaih6mcbgr3v4s2q34";
    cargoFeatures = features;
    doCheck = false;
    dontFixup = true;
--- a/experiments/Cargo.toml
+++ b/experiments/Cargo.toml
@ -16,5 +16,5 @@ embedded-hal = "0.2"
 libregister = { path = "../libregister" }
 libcortex_a9 = { path = "../libcortex_a9" }
 libboard_zynq = { path = "../libboard_zynq" }
-libsupport_zynq = { path = "../libsupport_zynq", default-features = false, features = ["panic_handler"]}
+libsupport_zynq = { path = "../libsupport_zynq" }
 libasync = { path = "../libasync" }
--- a/experiments/link.x
+++ b/experiments/link.x
@ -32,23 +32,19 @@ SECTIONS
        *(.bss .bss.*);
        . = ALIGN(4);
        __bss_end = .;
-    } > OCM3
+    } > OCM

    .stack1 (NOLOAD) : ALIGN(8) {
        __stack1_end = .;
        . += 0x200;
        __stack1_start = .;
-    } > OCM3
+    } > OCM

    .stack0 (NOLOAD) : ALIGN(8) {
        __stack0_end = .;
-        . = ORIGIN(OCM3) + LENGTH(OCM3) - 8;
+        . = ORIGIN(OCM) + LENGTH(OCM) - 8;
        __stack0_start = .;
-
-        /* unused heap0 to prevent the linker from complaining*/
-        __heap0_start = .;
-        __heap0_end = .;
-    } > OCM3
+    } > OCM

    /DISCARD/ :
    {
--- a/experiments/src/main.rs
+++ b/experiments/src/main.rs
@ -1,7 +1,6 @@
 #![no_std]
 #![no_main]
 #![feature(const_in_array_repeat_expressions)]
-#![feature(naked_functions)]

 extern crate alloc;

@ -15,9 +14,9 @@ use libboard_zynq::{
    self as zynq,
    clocks::source::{ArmPll, ClockSource, IoPll},
    clocks::Clocks,
-    print, println, stdio,
-    mpcore,
-    gic,
+    print, println,
+    setup_l2cache,
+    sdio::sd_card::SdCard,
    smoltcp::{
        iface::{EthernetInterfaceBuilder, NeighborCache, Routes},
        time::Instant,
@ -25,70 +24,24 @@ use libboard_zynq::{
    },
    time::Milliseconds,
 };
+#[cfg(feature = "target_zc706")]
+use libboard_zynq::ps7_init;
 use libcortex_a9::{
    mutex::Mutex,
-    sync_channel::{Sender, Receiver},
    sync_channel,
-    regs::{MPIDR, SP},
-    spin_lock_yield, notify_spin_lock,
-    asm
 };
-use libregister::{RegisterR, RegisterW};
+use libregister::RegisterR;
 use libsupport_zynq::{
    boot, ram,
 };
 use log::{info, warn};
-use core::sync::atomic::{AtomicBool, Ordering};

 const HWADDR: [u8; 6] = [0, 0x23, 0xde, 0xea, 0xbe, 0xef];

-static mut CORE1_REQ: (Sender<usize>, Receiver<usize>) = sync_channel!(usize, 10);
-static mut CORE1_RES: (Sender<usize>, Receiver<usize>) = sync_channel!(usize, 10);
-
-extern "C" {
-    static mut __stack1_start: u32;
-}
-
-static CORE1_RESTART: AtomicBool = AtomicBool::new(false);
-
-#[link_section = ".text.boot"]
-#[no_mangle]
-#[naked]
-pub unsafe extern "C" fn IRQ() {
-    if MPIDR.read().cpu_id() == 1{
-        let mpcore = mpcore::RegisterBlock::new();
-        let mut gic = gic::InterruptController::new(mpcore);
-        let id = gic.get_interrupt_id();
-        if id.0 == 0 {
-            gic.end_interrupt(id);
-            asm::exit_irq();
-            SP.write(&mut __stack1_start as *mut _ as u32);
-            asm::enable_irq();
-            CORE1_RESTART.store(false, Ordering::Relaxed);
-            notify_spin_lock();
-            main_core1();
-        }
-    }
-    stdio::drop_uart();
-    println!("IRQ");
-    loop {}
-}
-
-pub fn restart_core1() {
-    let mut interrupt_controller = gic::InterruptController::new(mpcore::RegisterBlock::new());
-    CORE1_RESTART.store(true, Ordering::Relaxed);
-    interrupt_controller.send_sgi(gic::InterruptId(0), gic::CPUCore::Core1.into());
-    while CORE1_RESTART.load(Ordering::Relaxed) {
-        spin_lock_yield();
-    }
-}
-
 #[no_mangle]
 pub fn main_core0() {
    // zynq::clocks::CpuClocks::enable_io(1_250_000_000);
    println!("\nzc706 main");
-    let mut interrupt_controller = gic::InterruptController::new(mpcore::RegisterBlock::new());
-    interrupt_controller.enable_interrupts();
    // ps7_init::apply();
    libboard_zynq::stdio::drop_uart();

@ -130,6 +83,29 @@ pub fn main_core0() {
        clocks.cpu_2x(),
        clocks.cpu_1x()
    );
+    info!("Setup L2Cache");
+    setup_l2cache();
+    info!("L2Cache done");
+
+    // commented out due to OCM full
+    // let sd = libboard_zynq::sdio::SDIO::sdio0(true);
+    // // only test SD card if it is inserted
+    // if sd.is_card_inserted() {
+    //     let result = SdCard::from_sdio(sd);
+    //     match &result {
+    //         Ok(_) => info!("OK!"),
+    //         Err(a) => info!("{}", a),
+    //     };
+    //     const SIZE: usize = 512 * 2 + 1;
+    //     let mut sd_card = result.unwrap();
+    //     if false {
+    //         let buffer: [u8; SIZE] = [5; SIZE];
+    //         sd_card.write_block(0x0, 2, &buffer).unwrap();
+    //     }
+    //     let mut buffer: [u8; SIZE] = [0; SIZE];
+    //     sd_card.read_block(0 /*0x1*/, 2, &mut buffer[1..]).unwrap();
+    //     info!("buffer = {:?}", &buffer[..]);
+    // }

    let mut flash = zynq::flash::Flash::new(200_000_000).linear_addressing_mode();
    let flash_ram: &[u8] = unsafe { core::slice::from_raw_parts(flash.ptr(), flash.size()) };
@ -149,6 +125,7 @@ pub fn main_core0() {
    ddr.memtest();
    ram::init_alloc_ddr(&mut ddr);

+if false {
    #[cfg(dev)]
    for i in 0..=1 {
        let mut flash_io = flash.manual_mode(i);
@ -188,62 +165,29 @@ pub fn main_core0() {
        flash = flash_io.stop();
    }

-    boot::Core1::start(false);
+    let core1 = boot::Core1::start(false);

-    let core1_req = unsafe { &mut CORE1_REQ.0 };
-    let core1_res = unsafe { &mut CORE1_RES.1 };
+    let (mut core1_req, rx) = sync_channel!(usize, 10);
+    *CORE1_REQ.lock() = Some(rx);
+    let (tx, mut core1_res) = sync_channel!(usize, 10);
+    *CORE1_RES.lock() = Some(tx);
    task::block_on(async {
        for i in 0..10 {
-            restart_core1();
            core1_req.async_send(i).await;
            let j = core1_res.async_recv().await;
            println!("{} -> {}", i, j);
        }
    });
-    unsafe {
-        core1_req.drop_elements();
-    }
-    
-    // Test I2C
-    #[cfg(feature = "target_zc706")]
-    {
-        let mut i2c = zynq::i2c::I2C::i2c();
-        i2c.init();
-        println!("I2C bit-banging enabled");
-        let mut eeprom = zynq::i2c::eeprom::EEPROM::new(&mut i2c, 16);
-        // Write to 0x00 and 0x08
-        let eeprom_buffer: [u8; 22] = [
-            0x66, 0x77, 0x88, 0x99, 0xaa, 0xbb, 
-            0x77, 0x88, 0x99, 0xaa, 0xbb, 0xcc, 0xdd, 0xee, 
-            0xef, 0xcd, 0xab, 0x89, 0x67, 0x45, 0x23, 0x01,
-        ];
-        eeprom.write(0x00, &eeprom_buffer[0..6]);
-        eeprom.write(0x08, &eeprom_buffer[6..22]);
-        println!("Data written to EEPROM");
-        let mut eeprom_buffer = [0u8; 24];
-        // Read from 0x00
-        eeprom.read(0x00, &mut eeprom_buffer);
-        print!("Data read from EEPROM @ 0x00: (hex) ");
-        for i in 0..6 {
-            print!("{:02x} ", eeprom_buffer[i]);
-        }
-        println!("");
-        // Read from 0x08
-        eeprom.read(0x08, &mut eeprom_buffer);
-        print!("Data read from EEPROM @ 0x08: (hex) ");
-        for i in 0..16 {
-            print!("{:02x} ", eeprom_buffer[i]);
-        }
-        println!("");
+    core1.disable();
 }

    let eth = zynq::eth::Eth::default(HWADDR.clone());
    println!("Eth on");

-    const RX_LEN: usize = 4096;
+    const RX_LEN: usize = 8192;
    // Number of transmission buffers (minimum is two because with
    // one, duplicate packet transmission occurs)
-    const TX_LEN: usize = 4096;
+    const TX_LEN: usize = 8192;
    let eth = eth.start_rx(RX_LEN);
    let mut eth = eth.start_tx(TX_LEN);

@ -260,6 +204,9 @@ pub fn main_core0() {
        .neighbor_cache(neighbor_cache)
        .finalize();

+    #[cfg(feature = "target_zc706")]
+    ps7_init::report_differences();
+
    Sockets::init(32);

    const TCP_PORT: u16 = 19;
@ -270,7 +217,7 @@ pub fn main_core0() {
        while let Ok(stream) = TcpStream::accept(TCP_PORT, 0x10_0000, 0x10_0000).await {
            let stats_tx = stats_tx.clone();
            task::spawn(async move {
-                let tx_data = (0..=255).take(4096).collect::<alloc::vec::Vec<u8>>();
+                let tx_data = (0..=255).take(8192).collect::<alloc::vec::Vec<u8>>();
                loop {
                    // const CHUNK_SIZE: usize = 65536;
                    // match stream.send((0..=255).cycle().take(CHUNK_SIZE)).await {
@ -326,15 +273,24 @@ pub fn main_core0() {
    })
 }

+static CORE1_REQ: Mutex<Option<sync_channel::Receiver<usize>>> = Mutex::new(None);
+static CORE1_RES: Mutex<Option<sync_channel::Sender<usize>>> = Mutex::new(None);
 static DONE: Mutex<bool> = Mutex::new(false);

 #[no_mangle]
 pub fn main_core1() {
    println!("Hello from core1!");
-    let mut interrupt_controller = gic::InterruptController::new(mpcore::RegisterBlock::new());
-    interrupt_controller.enable_interrupts();
-    let req = unsafe { &mut CORE1_REQ.1 };
-    let res = unsafe { &mut CORE1_RES.0 };
+
+    let mut req = None;
+    while req.is_none() {
+        req = CORE1_REQ.lock().take();
+    }
+    let req = req.unwrap();
+    let mut res = None;
+    while res.is_none() {
+        res = CORE1_RES.lock().take();
+    }
+    let mut res = res.unwrap();

    for i in req {
        res.send(i * i);
--- a/libboard_zynq/Cargo.toml
+++ b/libboard_zynq/Cargo.toml
@ -21,5 +21,5 @@ libcortex_a9 = { path = "../libcortex_a9" }

 [dependencies.smoltcp]
 version = "0.6"
-features = ["ethernet", "proto-ipv4", "proto-ipv6", "socket-tcp"]
+features = ["ethernet", "proto-ipv4", "socket-tcp"]
 default-features = false
--- a/libboard_zynq/src/eth/mod.rs
+++ b/libboard_zynq/src/eth/mod.rs
@ -2,7 +2,7 @@ use core::{
    marker::PhantomData,
    ops::{Deref, DerefMut},
 };
-use log::{debug, info, warn, error};
+use log::{error, info, warn};
 use libregister::*;
 use super::slcr;
 use super::clocks::Clocks;
@ -132,7 +132,7 @@ fn calculate_tx_divisors(tx_clock: u32) -> (u8, u8) {
        }
    }
    let result = best.unwrap();
-    debug!("Eth TX clock for {}: {} / {} / {} = {}",
+    info!("Eth TX clock for {}: {} / {} / {} = {}",
          tx_clock, io_pll,
          result.0, result.1,
          io_pll / result.0 as u32 / result.1 as u32
--- a/libboard_zynq/src/eth/rx.rs
+++ b/libboard_zynq/src/eth/rx.rs
@ -2,6 +2,8 @@ use core::ops::Deref;
 use alloc::{vec, vec::Vec};
 use libcortex_a9::{asm::*, cache::*, UncachedSlice};
 use libregister::*;
+use log::debug;
+use crate::l2cache;
 use super::Buffer;

 #[derive(Debug)]
@ -57,7 +59,7 @@ register_bit!(desc_word1, global_broadcast, 31);
 #[repr(C)]
 pub struct DescList {
    list: UncachedSlice<DescEntry>,
-    buffers: Vec<Buffer>,
+    buffers: UncachedSlice<Buffer>,
    next: usize,
 }

@ -65,7 +67,8 @@ impl DescList {
    pub fn new(size: usize) -> Self {
        let mut list = UncachedSlice::new(size, || DescEntry::zeroed())
            .unwrap();
-        let mut buffers = vec![Buffer::new(); size];
+        let mut buffers = UncachedSlice::new(size, || Buffer::new())
+            .unwrap();

        let last = list.len().min(buffers.len()) - 1;
        for (i, (entry, buffer)) in list.iter_mut().zip(buffers.iter_mut()).enumerate() {
@ -81,9 +84,8 @@ impl DescList {
            entry.word1.write(
                DescWord1::zeroed()
            );
-            // Flush buffer from cache, to be filled by the peripheral
-            // before next read
-            dcci_slice(&buffer[..]);
+            // l2cache().invalidate_slice(&mut buffer[..]);
+            // dcci_slice(&buffer[..]);
        }

        DescList {
@ -109,6 +111,9 @@ impl DescList {
            let word1 = entry.word1.read();
            let len = word1.frame_length_lsbs().into();
            let buffer = &mut self.buffers[self.next][0..len];
+            // // Invalidate caches for packet buffer
+            // l2cache().invalidate_slice(&mut buffer[..]);
+            // dcci_slice(&buffer[..]);

            self.next += 1;
            if self.next >= list_len {
@ -117,8 +122,10 @@ impl DescList {

            let pkt = PktRef { entry, buffer };
            if word1.start_of_frame() && word1.end_of_frame() {
+                // debug!("pkt {}: {:08X}..{:08X}", len, &pkt.buffer[0] as *const _ as usize, &pkt.buffer[pkt.len()-1] as *const _ as usize);
                Ok(Some(pkt))
            } else {
+                debug!("pkt trunc");
                Err(Error::Truncated)
            }
        } else {
@ -135,9 +142,10 @@ pub struct PktRef<'a> {

 impl<'a> Drop for PktRef<'a> {
    fn drop(&mut self) {
-        // Flush buffer from cache, to be filled by the peripheral
-        // before next read
-        dcci_slice(self.buffer);
+        // // Flush buffer from cache, to be filled by the peripheral
+        // // before next read
+        // l2cache().invalidate_slice(self.buffer);
+        // dcci_slice(self.buffer);

        self.entry.word0.modify(|_, w| w.used(false));
        dmb();
--- a/libboard_zynq/src/eth/tx.rs
+++ b/libboard_zynq/src/eth/tx.rs
@ -1,7 +1,10 @@
 use core::ops::{Deref, DerefMut};
 use alloc::{vec, vec::Vec};
-use libcortex_a9::{cache::dcc_slice, UncachedSlice};
+use libcortex_a9::{asm::dmb, cache::dcc_slice, UncachedSlice};
+//use libcortex_a9::{asm::*, cache::*, UncachedSlice};
 use libregister::*;
+use log::{debug, warn};
+use crate::l2cache;
 use super::{Buffer, regs};

 /// Descriptor entry
@ -44,7 +47,7 @@ pub const DESCS: usize = 8;
 #[repr(C)]
 pub struct DescList {
    list: UncachedSlice<DescEntry>,
-    buffers: Vec<Buffer>,
+    buffers: UncachedSlice<Buffer>,
    next: usize,
 }

@ -52,7 +55,8 @@ impl DescList {
    pub fn new(size: usize) -> Self {
        let mut list = UncachedSlice::new(size, || DescEntry::zeroed())
            .unwrap();
-        let mut buffers = vec![Buffer::new(); size];
+        let mut buffers = UncachedSlice::new(size, || Buffer::new())
+            .unwrap();

        let last = list.len().min(buffers.len()) - 1;
        // Sending seems to not work properly with only one packet
@ -94,8 +98,10 @@ impl DescList {
    }

    pub fn send<'s: 'p, 'p>(&'s mut self, regs: &'s mut regs::RegisterBlock, length: usize) -> Option<PktRef<'p>> {
+        // debug!("send {}", length);
        let list_len = self.list.len();
        let entry = &mut self.list[self.next];
+        dmb();
        if entry.word1.read().used() {
            let buffer = &mut self.buffers[self.next][0..length];
            entry.word1.write(DescWord1::zeroed()
@ -113,6 +119,7 @@ impl DescList {
            Some(PktRef { entry, buffer, regs })
        } else {
            // Still in use by HW (sending too fast, ring exceeded)
+            warn!("tx ring overflow");
            None
        }
    }
@ -128,10 +135,13 @@ pub struct PktRef<'a> {

 impl<'a> Drop for PktRef<'a> {
    fn drop(&mut self) {
-        // Write back all dirty cachelines of this buffer
-        dcc_slice(self.buffer);
+        // // Write back all dirty cachelines of packet buffer
+        // dcc_slice(self.buffer);
+        // l2cache().clean_slice(self.buffer);

        self.entry.word1.modify(|_, w| w.used(false));
+        dmb();
+        // dsb();
        if ! self.regs.tx_status.read().tx_go() {
            // Start TX if not already running
            self.regs.net_ctrl.modify(|_, w| w.start_tx(true));
--- a/libboard_zynq/src/gic.rs
+++ b/libboard_zynq/src/gic.rs
@ -1,150 +0,0 @@
-//! ARM Generic Interrupt Controller
-
-use bit_field::BitField;
-use libregister::{RegisterW, RegisterRW, RegisterR};
-use super::mpcore;
-
-#[derive(Debug, Clone, Copy)]
-pub struct InterruptId(pub u8);
-
-#[derive(Debug, Clone, Copy)]
-#[repr(u8)]
-pub enum CPUCore {
-    Core0 = 0b01,
-    Core1 = 0b10
-}
-
-#[derive(Debug, Clone, Copy)]
-pub struct TargetCPU(u8);
-
-impl TargetCPU {
-    pub const fn none() -> TargetCPU {
-        TargetCPU(0)
-    }
-
-    pub const fn and(self, other: TargetCPU) -> TargetCPU {
-        TargetCPU(self.0 | other.0)
-    }
-}
-
-impl From<CPUCore> for TargetCPU {
-    fn from(core: CPUCore) -> Self {
-        TargetCPU(core as u8)
-    }
-}
-
-pub enum TargetList {
-    CPUList(TargetCPU),
-    Others,
-    This
-}
-
-impl From<CPUCore> for TargetList {
-    fn from(core: CPUCore) -> Self {
-        TargetList::CPUList(TargetCPU(core as u8))
-    }
-}
-
-impl From<TargetCPU> for TargetList {
-    fn from(cpu: TargetCPU) -> Self {
-        TargetList::CPUList(cpu)
-    }
-}
-
-#[derive(Debug, Clone, Copy)]
-pub enum InterruptSensitivity {
-    Level,
-    Edge,
-}
-
-pub struct InterruptController {
-    mpcore: &'static mut mpcore::RegisterBlock,
-}
-
-impl InterruptController {
-    pub fn new(mpcore: &'static mut mpcore::RegisterBlock) -> Self {
-        InterruptController { mpcore }
-    }
-
-    pub fn disable_interrupts(&mut self) {
-        self.mpcore.iccicr.modify(|_, w| w.enable_ns(false)
-                                          .enable_s(false));
-        // FIXME: Should we disable the distributor globally when we disable interrupt (for a single
-        // core)?
-        // self.mpcore.icddcr.modify(|_, w| w.enable_secure(false)
-        //                                   .enable_non_secure(false));
-    }
-
-    /// enable interrupt signaling
-    pub fn enable_interrupts(&mut self) {
-        self.mpcore.iccicr.modify(|_, w| w.enable_ns(true)
-                                          .enable_s(true));
-        self.mpcore.icddcr.modify(|_, w| w.enable_secure(true));
-
-        // Enable all interrupts except those of the lowest priority.
-        self.mpcore.iccpmr.write(mpcore::ICCPMR::zeroed().priority(0xFF));
-    }
-
-    /// send software generated interrupt
-    pub fn send_sgi(&mut self, id: InterruptId, targets: TargetList) {
-        assert!(id.0 < 16);
-        self.mpcore.icdsgir.modify(|_, w| match targets {
-            TargetList::CPUList(list) => w.target_list_filter(0).cpu_target_list(list.0),
-            TargetList::Others => w.target_list_filter(0b01),
-            TargetList::This => w.target_list_filter(0b10)
-        }.sgiintid(id.0).satt(false));
-    }
-
-    /// enable the interrupt *for this core*.
-    /// Not needed for SGI.
-    pub fn enable(&mut self, id: InterruptId, target_cpu: CPUCore, sensitivity: InterruptSensitivity, priority: u8) {
-        // only 5 bits of the priority is useful
-        assert!(priority < 32);
-
-        self.disable_interrupts();
-
-        // enable
-        let m = (id.0 >> 5) as usize;
-        let n = (id.0 & 0x1F) as usize;
-        assert!(m < 3);
-        unsafe {
-            self.mpcore.icdiser[m].modify(|mut icdiser| *icdiser.set_bit(n, true));
-        }
-
-        // target cpu
-        let m = (id.0 >> 2) as usize;
-        let n = (8 * (id.0 & 3)) as usize;
-        unsafe {
-            self.mpcore.icdiptr[m].modify(|mut icdiptr| *icdiptr.set_bits(n..=n+1, target_cpu as u32 + 1));
-        }
-
-        // sensitivity
-        let m = (id.0 >> 4) as usize;
-        let n = (2 * (id.0 & 0xF)) as usize;
-        unsafe {
-            self.mpcore.icdicfr[m].modify(|mut icdicfr| *icdicfr.set_bits(n..=n+1, match sensitivity {
-                InterruptSensitivity::Level => 0b00,
-                InterruptSensitivity::Edge  => 0b10,
-            }));
-        }
-
-        // priority
-        let offset = (id.0 % 4) * 8;
-        let priority: u32 = (priority as u32) << (offset + 3);
-        let mask: u32 = 0xFFFFFFFF ^ (0xFF << offset);
-        unsafe {
-            self.mpcore.icdipr[id.0 as usize / 4].modify(|v| (v & mask) | priority);
-        }
-
-        self.enable_interrupts();
-    }
-
-    pub fn end_interrupt(&mut self, id: InterruptId) {
-        self.mpcore.icceoir.modify(|_, w| w.eoiintid(id.0 as u32));
-    }
-
-    pub fn get_interrupt_id(&self) -> InterruptId {
-        InterruptId(self.mpcore.icciar.read().ackintid() as u8)
-    }
-
-}
--- a/libboard_zynq/src/i2c/eeprom.rs
+++ b/libboard_zynq/src/i2c/eeprom.rs
@ -1,107 +0,0 @@
-use super::I2C;
-use crate::time::Milliseconds;
-use embedded_hal::timer::CountDown;
-
-pub struct EEPROM<'a> {
-    i2c: &'a mut I2C,
-    port: u8,
-    address: u8,
-    page_size: u8,
-    count_down: crate::timer::global::CountDown<Milliseconds>
-}
-
-impl<'a> EEPROM<'a> {
-    #[cfg(feature = "target_zc706")]
-    pub fn new(i2c: &'a mut I2C, page_size: u8) -> Self {
-        EEPROM {
-            i2c: i2c,
-            port: 2,
-            address: 0b1010100,
-            page_size: page_size,
-            count_down: unsafe { crate::timer::GlobalTimer::get() }.countdown()
-        }
-    }
-
-    #[cfg(feature = "target_zc706")]
-    fn select(&mut self) -> Result<(), &'static str> {
-        let mask: u16 = 1 << self.port;
-        self.i2c.pca9548_select(0b1110100, mask as u8)?;
-        Ok(())
-    }
-
-    /// Random read
-    pub fn read<'r>(&mut self, addr: u8, buf: &'r mut [u8]) -> Result<(), &'static str> {
-        self.select()?;
-
-        self.i2c.start()?;
-        self.i2c.write(self.address << 1)?;
-        self.i2c.write(addr)?;
-
-        self.i2c.restart()?;
-        self.i2c.write((self.address << 1) | 1)?;
-        let buf_len = buf.len();
-        for (i, byte) in buf.iter_mut().enumerate() {
-            *byte = self.i2c.read(i < buf_len - 1)?;
-        }
-
-        self.i2c.stop()?;
-
-        Ok(())
-    }
-
-    /// Smart multi-page writing
-    /// Using the "Page Write" function of an EEPROM, the memory region for each transaction
-    /// (i.e. from byte `addr` to byte `addr+buf.len()`) should fit under each page 
-    /// (i.e. `addr+buf.len()` < `addr/self.page_size+1`); otherwise, a roll-oever occurs, 
-    /// where bytes beyond the page end. This smart function takes care of the scenario to avoid
-    /// any roll-over when writing ambiguous memory regions.
-    pub fn write(&mut self, addr: u8, buf: &[u8]) -> Result<(), &'static str> {
-        self.select()?;
-
-        let buf_len = buf.len();
-        let mut pb: u8 = addr % self.page_size;
-        for (i, byte) in buf.iter().enumerate() {
-            if (i == 0) || (pb == 0) {
-                self.i2c.start()?;
-                self.i2c.write(self.address << 1)?;
-                self.i2c.write(addr + (i as u8))?;
-            }
-            self.i2c.write(*byte)?;
-            pb += 1;
-
-            if (i == buf_len-1) || (pb == self.page_size) {
-                self.i2c.stop()?;
-                self.poll(1_000)?;
-                pb = 0;
-            }
-        }
-
-        Ok(())
-    }
-
-    /// Poll
-    pub fn poll(&mut self, timeout_ms: u64) -> Result<(), &'static str> {
-        self.select()?;
-
-        self.count_down.start(Milliseconds(timeout_ms));
-        loop {
-            self.i2c.start()?;
-            let ack = self.i2c.write(self.address << 1)?;
-            self.i2c.stop()?;
-            if ack {
-                break
-            };
-            if !self.count_down.waiting() {
-                return Err("I2C polling timeout")
-            }
-        }
-
-        Ok(())
-    }
-
-    pub fn read_eui48<'r>(&mut self) -> Result<[u8; 6], &'static str> {
-        let mut buffer = [0u8; 6];
-        self.read(0xFA, &mut buffer)?;
-        Ok(buffer)
-    }
-}
--- a/libboard_zynq/src/i2c/mod.rs
+++ b/libboard_zynq/src/i2c/mod.rs
@ -1,235 +0,0 @@
-//! I2C Bit-banging Controller
-
-mod regs;
-pub mod eeprom;
-use super::clocks::Clocks;
-use super::slcr;
-use super::time::Microseconds;
-use embedded_hal::timer::CountDown;
-use libregister::{RegisterR, RegisterRW, RegisterW};
-
-const INVALID_BUS: &'static str = "Invalid I2C bus";
-
-pub struct I2C {
-    regs: regs::RegisterWrapper,
-    count_down: super::timer::global::CountDown<Microseconds>
-}
-
-impl I2C {
-    #[cfg(feature = "target_zc706")]
-    pub fn i2c() -> Self {
-        // Route I2C 0 SCL / SDA Signals to MIO Pins 50 / 51
-        slcr::RegisterBlock::unlocked(|slcr| {
-            // SCL
-            slcr.mio_pin_50.write(
-                slcr::MioPin50::zeroed()
-                    .l3_sel(0b000)  // as GPIO 50
-                    .io_type(slcr::IoBufferType::Lvcmos18)
-                    .pullup(true)
-                    .disable_rcvr(true)
-            );
-            // SDA
-            slcr.mio_pin_51.write(
-                slcr::MioPin51::zeroed()
-                    .l3_sel(0b000)  // as GPIO 51
-                    .io_type(slcr::IoBufferType::Lvcmos18)
-                    .pullup(true)
-                    .disable_rcvr(true)
-            );
-            // Reset
-            slcr.gpio_rst_ctrl.reset_gpio();
-        });
-
-        Self::ctor_common(0xFFFF - 0x000C)
-    }
-
-    fn ctor_common(gpio_output_mask: u16) -> Self {
-        // Setup register block
-        let clocks = Clocks::get();
-        let self_ = Self {
-            regs: regs::RegisterWrapper::new(),
-            count_down: unsafe { super::timer::GlobalTimer::get() }.countdown()
-        };
-
-        // Setup GPIO output mask
-        self_.regs.gpio_output_mask.modify(|_, w| {
-            w.mask(gpio_output_mask)
-        });
-        // Setup GPIO driver direction
-        self_.regs.gpio_direction.modify(|_, w| {
-            w.scl(true).sda(true)
-        });
-
-        self_
-    }
-
-    /// Delay for I2C operations, simple wrapper for nb.
-    fn delay_us(&mut self, us: u64) {
-        self.count_down.start(Microseconds(us));
-        nb::block!(self.count_down.wait()).unwrap();
-    }
-
-    fn half_period(&mut self) { self.delay_us(100) }
-
-    fn sda_i(&mut self) -> bool {
-        self.regs.gpio_input.read().sda()
-    }
-
-    fn scl_i(&mut self) -> bool {
-        self.regs.gpio_input.read().scl()
-    }
-
-    fn sda_oe(&mut self, oe: bool) {
-        self.regs.gpio_output_enable.modify(|_, w| {
-             w.sda(oe)
-        })
-    }
-
-    fn sda_o(&mut self, o: bool) {
-        self.regs.gpio_output_mask.modify(|_, w| {
-             w.sda_o(o)
-        })
-    }
-
-    fn scl_oe(&mut self, oe: bool) {
-        self.regs.gpio_output_enable.modify(|_, w| {
-             w.scl(oe)
-        })
-    }
-
-    fn scl_o(&mut self, o: bool) {
-        self.regs.gpio_output_mask.modify(|_, w| {
-             w.scl_o(o)
-        })
-    }
-
-    pub fn init(&mut self) -> Result<(), &'static str> {
-        self.scl_oe(false);
-        self.sda_oe(false);
-        self.scl_o(false);
-        self.sda_o(false);
-
-        // Check the I2C bus is ready
-        self.half_period();
-        self.half_period();
-        if !self.sda_i() {
-            // Try toggling SCL a few times
-            for _bit in 0..8 {
-                self.scl_oe(true);
-                self.half_period();
-                self.scl_oe(false);
-                self.half_period();
-            }
-        }
-
-        if !self.sda_i() {
-            return Err("SDA is stuck low and doesn't get unstuck");
-        }
-        if !self.scl_i() {
-            return Err("SCL is stuck low and doesn't get unstuck");
-        }
-        // postcondition: SCL and SDA high
-        Ok(())
-    }
-
-    pub fn start(&mut self) -> Result<(), &'static str> {
-        // precondition: SCL and SDA high
-        if !self.scl_i() {
-            return Err("SCL is stuck low and doesn't get unstuck");
-        }
-        if !self.sda_i() {
-            return Err("SDA arbitration lost");
-        }
-        self.sda_oe(true);
-        self.half_period();
-        self.scl_oe(true);
-        // postcondition: SCL and SDA low
-        Ok(())
-    }
-
-    pub fn restart(&mut self) -> Result<(), &'static str> {
-        // precondition SCL and SDA low
-        self.sda_oe(false);
-        self.half_period();
-        self.scl_oe(false);
-        self.half_period();
-        self.start()?;
-        // postcondition: SCL and SDA low
-        Ok(())
-    }
-
-    pub fn stop(&mut self) -> Result<(), &'static str> {
-        // precondition: SCL and SDA low
-        self.half_period();
-        self.scl_oe(false);
-        self.half_period();
-        self.sda_oe(false);
-        self.half_period();
-        if !self.sda_i() {
-            return Err("SDA arbitration lost");
-        }
-        // postcondition: SCL and SDA high
-        Ok(())
-    }
-
-    pub fn write(&mut self, data: u8) -> Result<bool, &'static str> {
-        // precondition: SCL and SDA low
-        // MSB first
-        for bit in (0..8).rev() {
-            self.sda_oe(data & (1 << bit) == 0);
-            self.half_period();
-            self.scl_oe(false);
-            self.half_period();
-            self.scl_oe(true);
-        }
-        self.sda_oe(false);
-        self.half_period();
-        self.scl_oe(false);
-        self.half_period();
-        // Read ack/nack
-        let ack = !self.sda_i();
-        self.scl_oe(true);
-        self.sda_oe(true);
-        // postcondition: SCL and SDA low
-
-        Ok(ack)
-    }
-
-    pub fn read(&mut self, ack: bool) -> Result<u8, &'static str> {
-        // precondition: SCL and SDA low
-        self.sda_oe(false);
-
-        let mut data: u8 = 0;
-
-        // MSB first
-        for bit in (0..8).rev() {
-            self.half_period();
-            self.scl_oe(false);
-            self.half_period();
-            if self.sda_i() { data |= 1 << bit }
-            self.scl_oe(true);
-        }
-        // Send ack/nack
-        self.sda_oe(ack);
-        self.half_period();
-        self.scl_oe(false);
-        self.half_period();
-        self.scl_oe(true);
-        self.sda_oe(true);
-        // postcondition: SCL and SDA low
-
-        Ok(data)
-    }
-
-    pub fn pca9548_select(&mut self, address: u8, channels: u8) -> Result<(), &'static str> {
-        self.start()?;
-        if !self.write(address << 1)? {
-            return Err("PCA9548 failed to ack write address")
-        }
-        if !self.write(channels)? {
-            return Err("PCA9548 failed to ack control word")
-        }
-        self.stop()?;
-        Ok(())
-    }
-}
--- a/libboard_zynq/src/i2c/regs.rs
+++ b/libboard_zynq/src/i2c/regs.rs
@ -1,93 +0,0 @@
-use volatile_register::{RO, WO, RW};
-
-use libregister::{
-    register, register_at,
-    register_bit, register_bits
-};
-
-// With reference to:
-//
-// artiq:artiq/gateware/targets/kasli.py:
-// self.submodules.i2c = gpio.GPIOTristate([i2c.scl, i2c.sda])
-//
-// misoc:misoc/cores/gpio.py:
-// class GPIOTristate(Module, AutoCSR):
-//    def __init__(self, signals, reset_out=0, reset_oe=0):
-//        l = len(signals)
-//        self._in = CSRStatus(l)
-//        self._out = CSRStorage(l, reset=reset_out)
-//        self._oe = CSRStorage(l, reset=reset_oe)
-// 
-// Hence, using GPIOs as SCL and SDA GPIOs respectively.
-//
-// Current compatibility:
-// zc706: GPIO 50, 51 == SCL, SDA
-
-pub struct RegisterWrapper {
-    pub gpio_output_mask: &'static mut GPIOOutputMask,
-    pub gpio_input: &'static mut GPIOInput,
-    pub gpio_direction: &'static mut GPIODirection,
-    pub gpio_output_enable: &'static mut GPIOOutputEnable,
-}
-
-impl RegisterWrapper {
-    pub fn new() -> Self {
-        Self {
-            gpio_output_mask: GPIOOutputMask::new(),
-            gpio_input: GPIOInput::new(),
-            gpio_direction: GPIODirection::new(),
-            gpio_output_enable: GPIOOutputEnable::new()
-        }
-    }
-}
-
-// MASK_DATA_1_MSW:
-// Maskable output data for MIO[53:48]
-register!(gpio_output_mask, GPIOOutputMask, RW, u32);
-#[cfg(feature = "target_zc706")]
-register_at!(GPIOOutputMask, 0xE000A00C, new);
-// Output for SCL
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_output_mask, scl_o, 2);
-// Output for SDA
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_output_mask, sda_o, 3);
-// Mask for keeping bits except SCL and SDA unchanged
-#[cfg(feature = "target_zc706")]
-register_bits!(gpio_output_mask, mask, u16, 16, 31);
-
-// DATA_1_RO:
-// Input data for MIO[53:32]
-register!(gpio_input, GPIOInput, RO, u32);
-#[cfg(feature = "target_zc706")]
-register_at!(GPIOInput, 0xE000A064, new);
-// Input for SCL
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_input, scl, 18);
-// Input for SDA
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_input, sda, 19);
-
-// DIRM_1:
-// Direction mode for MIO[53:32]; 0/1 = in/out
-register!(gpio_direction, GPIODirection, RW, u32);
-#[cfg(feature = "target_zc706")]
-register_at!(GPIODirection, 0xE000A244, new);
-// Direction for SCL
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_direction, scl, 18);
-// Direction for SDA
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_direction, sda, 19);
-
-// OEN_1:
-// Output enable for MIO[53:32]
-register!(gpio_output_enable, GPIOOutputEnable, RW, u32);
-#[cfg(feature = "target_zc706")]
-register_at!(GPIOOutputEnable, 0xE000A248, new);
-// Output enable for SCL
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_output_enable, scl, 18);
-// Output enable for SDA
-#[cfg(feature = "target_zc706")]
-register_bit!(gpio_output_enable, sda, 19);
--- a/libboard_zynq/src/lib.rs
+++ b/libboard_zynq/src/lib.rs
@ -15,12 +15,37 @@ pub mod axi_hp;
 pub mod axi_gp;
 pub mod ddr;
 pub mod mpcore;
-pub mod gic;
 pub mod flash;
 pub mod time;
 pub mod timer;
 pub mod sdio;
-#[cfg(feature = "target_zc706")]
-pub mod i2c;
 pub mod logger;
 pub mod ps7_init;
+
+pub use libcortex_a9::pl310::L2Cache;
+
+pub fn l2cache() -> L2Cache {
+    const PL310_BASEADDR: usize = 0xF8F02000;
+    L2Cache::new(PL310_BASEADDR)
+}
+
+pub fn setup_l2cache() {
+    slcr::RegisterBlock::unlocked(|slcr| {
+        assert_eq!(&slcr.unnamed1 as *const _ as u32, 0xF8000A1C);
+        unsafe { slcr.unnamed1.write(0x020202); }
+    });
+    
+    let mut l2 = l2cache();
+    use log::info;
+    info!("l2 aux={:08X}", l2.regs.aux_control.read());
+    // TODO: set prefetch
+
+    // Configure ZYNQ-specific latency
+    l2.set_tag_ram_latencies(1, 1, 1);
+    l2.set_data_ram_latencies(1, 2, 1);
+
+    l2.disable_interrupts();
+    l2.reset_interrupts();
+    l2.invalidate_all();
+    l2.enable();
+}
--- a/libboard_zynq/src/mpcore.rs
+++ b/libboard_zynq/src/mpcore.rs
@ -8,141 +8,48 @@ use libregister::{

 #[repr(C)]
 pub struct RegisterBlock {
-    /// SCU Control Register
    pub scu_control: ScuControl,
-    /// SCU Configuration Register
-    pub scu_config: ScuConfig,
-    /// SCU CPU Power Status Register
-    pub scu_cpu_power_status: SCUCPUPowerStatusRegister,
-    /// SCU Invalidate All Registers in Secure State
+    pub scu_config: RO<u32>,
+    pub scu_cpu_power: RW<u32>,
    pub scu_invalidate: ScuInvalidate,
-    unused0: [u32; 12],
-    /// Filtering Start Address Register
-    pub filtering_start_address: FilteringStartAddressRegister,
-    /// Defined by FILTEREND input
-    pub filtering_end_address: FilteringEndAddressRegister,
-    unused1: [u32; 2],
-    /// SCU Access Control (SAC) Register
-    pub scu_access_control_sac: SCUAccessControlRegisterSAC,
-    /// SCU Non-secure Access Control Register SNSAC
-    pub scu_non_secure_access_control: SCUNonSecureAccessControlRegister,
-    unused2: [u32; 42],
-    /// CPU Interface Control Register
-    pub iccicr: ICCICR,
-    /// Interrupt Priority Mask Register
-    pub iccpmr: ICCPMR,
-    /// Binary Point Register
-    pub iccbpr: ICCBPR,
-    /// Interrupt Acknowledge Register
-    pub icciar: ICCIAR,
-    /// End Of Interrupt Register
-    pub icceoir: ICCEOIR,
-    /// Running Priority Register
-    pub iccrpr: ICCRPR,
-    /// Highest Pending Interrupt Register
-    pub icchpir: ICCHPIR,
-    /// Aliased Non-secure Binary Point Register
-    pub iccabpr: ICCABPR,
-    unused3: [u32; 55],
-    /// CPU Interface Implementer Identification Register
-    pub iccidr: ICCIDR,
-    /// Global Timer Counter Register 0
+    reserved0: [u32; 12],
+    pub filter_start: RW<u32>,
+    pub filter_end: RW<u32>,
+    reserved1: [u32; 2],
+    pub scu_access_control: RW<u32>,
+    pub scu_non_secure_access_control: RW<u32>,
+    reserved2: [u32; 42],
+    pub iccicr: RW<u32>,
+    pub iccpmw: RW<u32>,
+    pub iccbpr: RW<u32>,
+    pub icciar: RW<u32>,
+    pub icceoir: RW<u32>,
+    pub iccrpr: RW<u32>,
+    pub icchpir: RW<u32>,
+    pub iccabpr: RW<u32>,
+    reserved3: [u32; 55],
+    pub iccidr: RW<u32>,
    pub global_timer_counter0: ValueRegister,
    pub global_timer_counter1: ValueRegister,
-    /// Global Timer Control Register
    pub global_timer_control: GlobalTimerControl,
-    /// Global Timer Interrupt Status Register
-    pub global_timer_interrupt_status: GlobalTimerInterruptStatusRegister,
-    /// Comparator Value Register_0
+    pub global_timer_interrupt_status: RW<u32>,
    pub comparator_value0: ValueRegister,
    pub comparator_value1: ValueRegister,
-    /// Auto-increment Register
-    pub auto_increment: RW<u32>,
-    unused4: [u32; 249],
-    /// Private Timer Load Register
-    pub private_timer_load: RW<u32>,
-    /// Private Timer Counter Register
-    pub private_timer_counter: RW<u32>,
-    /// Private Timer Control Register
-    pub private_timer_control: PrivateTimerControlRegister,
-    /// Private Timer Interrupt Status Register
-    pub private_timer_interrupt_status: PrivateTimerInterruptStatusRegister,
-    unused5: [u32; 4],
-    /// Watchdog Load Register
-    pub watchdog_load: RW<u32>,
-    /// Watchdog Counter Register
-    pub watchdog_counter: RW<u32>,
-    /// Watchdog Control Register
-    pub watchdog_control: WatchdogControlRegister,
-    /// Watchdog Interrupt Status Register
-    pub watchdog_interrupt_status: WatchdogInterruptStatusRegister,
-    /// Watchdog Reset Status Register
-    pub watchdog_reset_status: WatchdogResetStatusRegister,
-    /// Watchdog Disable Register
-    pub watchdog_disable: RW<u32>,
-    unused6: [u32; 626],
-    /// Distributor Control Register
-    pub icddcr: ICDDCR,
-    /// Interrupt Controller Type Register
-    pub icdictr: ICDICTR,
-    /// Distributor Implementer Identification Register
-    pub icdiidr: ICDIIDR,
-    unused7: [u32; 29],
-    /// Interrupt Security Register
-    pub icdisr0: RW<u32>,
-    pub icdisr1: RW<u32>,
-    pub icdisr2: RW<u32>,
-    unused8: [u32; 29],
-    /// Interrupt Set-enable Registers
-    pub icdiser: [RW<u32>; 3],
-    unused9: [u32; 29],
-    /// Interrupt Clear-Enable Register 0
-    pub icdicer0: RW<u32>,
-    /// Interrupt Clear-Enable Register 1
-    pub icdicer1: RW<u32>,
-    /// Interrupt Clear-Enable Register 2
-    pub icdicer2: RW<u32>,
-    unused10: [u32; 29],
-    /// Interrupt Set-pending Register
-    pub icdispr0: RW<u32>,
-    pub icdispr1: RW<u32>,
-    pub icdispr2: RW<u32>,
-    unused11: [u32; 29],
-    /// Interrupt Clear-Pending Register
-    pub icdicpr0: RW<u32>,
-    pub icdicpr1: RW<u32>,
-    pub icdicpr2: RW<u32>,
-    unused12: [u32; 29],
-    /// Active Bit register
-    pub icdabr0: RW<u32>,
-    pub icdabr1: RW<u32>,
-    pub icdabr2: RW<u32>,
-    unused13: [u32; 61],
-    /// Interrupt Priority Register
-    pub icdipr: [RW<u32>; 24],
-    unused14: [u32; 232],
-    /// Interrupt Processor Targets Registers
-    pub icdiptr: [RW<u32>; 24],
-    unused15: [u32; 232],
-    /// Interrupt Configuration Registers
-    pub icdicfr: [RW<u32>; 6],
-    unused16: [u32; 58],
-    /// PPI Status Register
-    pub ppi_status: PpiStatus,
-    /// SPI Status Register 0
-    pub spi_status_0: RO<u32>,
-    /// SPI Status Register 1
-    pub spi_status_1: RO<u32>,
-    unused17: [u32; 125],
-    /// Software Generated Interrupt Register
-    pub icdsgir: ICDSGIR,
+    pub auto_increment: ValueRegister,
+    reserved4: [u32; 249],
+    pub private_timer_load: ValueRegister,
+    pub private_timer_counter: ValueRegister,
+    pub private_timer_control: RW<u32>,
+    pub private_timer_interrupt_status: RW<u32>,
+    reserved5: [u32; 4],
+    pub watchdog_load: ValueRegister,
+    pub watchdog_counter: ValueRegister,
+    pub watchdog_control: RW<u32>,
+    pub watchdog_interrupt_status: RW<u32>,
+    // there is plenty more (unimplemented)
 }
-
 register_at!(RegisterBlock, 0xF8F00000, new);

-register!(value_register, ValueRegister, RW, u32);
-register_bits!(value_register, value, u32, 0, 31);
-
 register!(scu_control, ScuControl, RW, u32);
 register_bit!(scu_control, ic_standby_enable, 6);
 register_bit!(scu_control, scu_standby_enable, 5);
@ -158,17 +65,6 @@ impl ScuControl {
    }
 }

-register!(scu_config, ScuConfig, RO, u32);
-register_bits!(scu_config, tag_ram_sizes, u8, 8, 15);
-register_bits!(scu_config, cpus_smp, u8, 4, 7);
-register_bits!(scu_config, cpu_number, u8, 0, 1);
-
-register!(scu_cpu_power_status, SCUCPUPowerStatusRegister, RW, u32);
-register_bits!(scu_cpu_power_status, cpu3_status, u8, 24, 25);
-register_bits!(scu_cpu_power_status, cpu2_status, u8, 16, 17);
-register_bits!(scu_cpu_power_status, cpu1_status, u8, 8, 9);
-register_bits!(scu_cpu_power_status, cpu0_status, u8, 0, 1);
-
 register!(scu_invalidate, ScuInvalidate, WO, u32);
 register_bits!(scu_invalidate, cpu0_ways, u8, 0, 3);
 register_bits!(scu_invalidate, cpu1_ways, u8, 4, 7);
@ -192,71 +88,8 @@ impl ScuInvalidate {
    }
 }

-register!(filtering_start_address, FilteringStartAddressRegister, RW, u32);
-register_bits!(filtering_start_address, filtering_start_address, u32, 20, 31);
-register_bits!(filtering_start_address, sbz, u32, 0, 19);
-
-register!(filtering_end_address, FilteringEndAddressRegister, RW, u32);
-register_bits!(filtering_end_address, filtering_end_address, u32, 20, 31);
-register_bits!(filtering_end_address, sbz, u32, 0, 19);
-
-register!(scu_access_control_sac, SCUAccessControlRegisterSAC, RW, u32);
-register_bit!(scu_access_control_sac, cp_u3, 3);
-register_bit!(scu_access_control_sac, cp_u2, 2);
-register_bit!(scu_access_control_sac, cp_u1, 1);
-register_bit!(scu_access_control_sac, cp_u0, 0);
-
-register!(scu_non_secure_access_control, SCUNonSecureAccessControlRegister, RO, u32);
-register_bits!(scu_non_secure_access_control, sbz, u32, 12, 31);
-register_bit!(scu_non_secure_access_control, cpu3_global_timer, 11);
-register_bit!(scu_non_secure_access_control, cpu2_global_timer, 10);
-register_bit!(scu_non_secure_access_control, cpu1_global_timer, 9);
-register_bit!(scu_non_secure_access_control, cpu0_global_timer, 8);
-register_bit!(scu_non_secure_access_control, private_timers_for_cpu3, 7);
-register_bit!(scu_non_secure_access_control, private_timers_for_cpu2, 6);
-register_bit!(scu_non_secure_access_control, private_timers_for_cpu1, 5);
-register_bit!(scu_non_secure_access_control, private_timers_for_cpu0, 4);
-register_bit!(scu_non_secure_access_control, component_access_for_cpu3, 3);
-register_bit!(scu_non_secure_access_control, component_access_for_cpu2, 2);
-register_bit!(scu_non_secure_access_control, component_access_for_cpu1, 1);
-register_bit!(scu_non_secure_access_control, component_access_for_cpu0, 0);
-
-register!(iccicr, ICCICR, RW, u32);
-register_bit!(iccicr, sbpr, 4);
-register_bit!(iccicr, fiq_en, 3);
-register_bit!(iccicr, ack_ctl, 2);
-register_bit!(iccicr, enable_ns, 1);
-register_bit!(iccicr, enable_s, 0);
-
-register!(iccpmr, ICCPMR, RW, u32);
-register_bits!(iccpmr, priority, u8, 0, 7);
-
-register!(iccbpr, ICCBPR, RW, u32);
-register_bits!(iccbpr, binary_point, u8, 0, 2);
-
-register!(icciar, ICCIAR, RW, u32);
-register_bits!(icciar, cpuid, u8, 10, 12);
-register_bits!(icciar, ackintid, u32, 0, 9);
-
-register!(icceoir, ICCEOIR, RW, u32);
-register_bits!(icceoir, cpuid, u8, 10, 12);
-register_bits!(icceoir, eoiintid, u32, 0, 9);
-
-register!(iccrpr, ICCRPR, RW, u32);
-register_bits!(iccrpr, priority, u8, 0, 7);
-
-register!(icchpir, ICCHPIR, RW, u32);
-register_bits!(icchpir, cpuid, u8, 10, 12);
-register_bits!(icchpir, pendintid, u32, 0, 9);
-
-register!(iccabpr, ICCABPR, RW, u32);
-register_bits!(iccabpr, binary_point, u8, 0, 2);
-
-register!(iccidr, ICCIDR, RO, u32);
-register_bits!(iccidr, part_number, u32, 20, 31);
-register_bits!(iccidr, architecture_version, u8, 16, 19);
-register_bits!(iccidr, revision_number, u8, 12, 15);
-register_bits!(iccidr, implementer, u32, 0, 11);
+register!(value_register, ValueRegister, RW, u32);
+register_bits!(value_register, value, u32, 0, 31);

 register!(global_timer_control, GlobalTimerControl, RW, u32);
 register_bits!(global_timer_control, prescaler, u8, 8, 15);
@ -264,58 +97,3 @@ register_bit!(global_timer_control, auto_increment_mode, 3);
 register_bit!(global_timer_control, irq_enable, 2);
 register_bit!(global_timer_control, comp_enablea, 1);
 register_bit!(global_timer_control, timer_enable, 0);
-
-register!(global_timer_interrupt_status, GlobalTimerInterruptStatusRegister, RW, u32);
-register_bit!(global_timer_interrupt_status, event_flag, 0);
-
-register!(private_timer_control, PrivateTimerControlRegister, RW, u32);
-register_bits!(private_timer_control, sbzp, u32, 16, 31);
-register_bits!(private_timer_control, prescaler, u8, 8, 15);
-register_bits!(private_timer_control, unk_sbzp, u8, 3, 7);
-register_bit!(private_timer_control, irq_enable, 2);
-register_bit!(private_timer_control, auto_reload, 1);
-register_bit!(private_timer_control, timer_enable, 0);
-
-register!(private_timer_interrupt_status, PrivateTimerInterruptStatusRegister, RW, u32);
-register_bits!(private_timer_interrupt_status, unk_sbzp, u32, 1, 31);
-
-register!(watchdog_control, WatchdogControlRegister, RW, u32);
-register_bits!(watchdog_control, prescaler, u8, 8, 15);
-register_bit!(watchdog_control, watchdog_mode, 3);
-register_bit!(watchdog_control, it_enable, 2);
-register_bit!(watchdog_control, auto_reload, 1);
-register_bit!(watchdog_control, watchdog_enable, 0);
-
-register!(watchdog_interrupt_status, WatchdogInterruptStatusRegister, RW, u32);
-register_bit!(watchdog_interrupt_status, event_flag, 0);
-
-register!(watchdog_reset_status, WatchdogResetStatusRegister, RW, u32);
-register_bit!(watchdog_reset_status, reset_flag, 0);
-
-register!(icddcr, ICDDCR, RW, u32);
-register_bit!(icddcr, enable_non_secure, 1);
-register_bit!(icddcr, enable_secure, 0);
-
-register!(icdictr, ICDICTR, RO, u32);
-register_bits!(icdictr, lspi, u8, 11, 15);
-register_bit!(icdictr, security_extn, 10);
-register_bits!(icdictr, sbz, u8, 8, 9);
-register_bits!(icdictr, cpu_number, u8, 5, 7);
-register_bits!(icdictr, it_lines_number, u8, 0, 4);
-
-register!(icdiidr, ICDIIDR, RO, u32);
-register_bits!(icdiidr, implementation_version, u8, 24, 31);
-register_bits!(icdiidr, revision_number, u32, 12, 23);
-register_bits!(icdiidr, implementer, u32, 0, 11);
-
-register!(ppi_status, PpiStatus, RO, u32);
-register_bits!(ppi_status, ppi_status, u8, 11, 15);
-register_bits!(ppi_status, sbz, u32, 0, 10);
-
-register!(icdsgir, ICDSGIR, RW, u32);
-register_bits!(icdsgir, target_list_filter, u8, 24, 25);
-register_bits!(icdsgir, cpu_target_list, u8, 16, 23);
-register_bit!(icdsgir, satt, 15);
-register_bits!(icdsgir, sbz, u32, 4, 14);
-register_bits!(icdsgir, sgiintid, u8, 0, 3);
-
--- a/libboard_zynq/src/slcr.rs
+++ b/libboard_zynq/src/slcr.rs
@ -132,7 +132,7 @@ pub struct RegisterBlock {
    pub can_rst_ctrl: RW<u32>,
    pub i2c_rst_ctrl: RW<u32>,
    pub uart_rst_ctrl: UartRstCtrl,
-    pub gpio_rst_ctrl: GpioRstCtrl,
+    pub gpio_rst_ctrl: RW<u32>,
    pub lqspi_rst_ctrl: LqspiRstCtrl,
    pub smc_rst_ctrl: RW<u32>,
    pub ocm_rst_ctrl: RW<u32>,
@ -229,15 +229,18 @@ pub struct RegisterBlock {
    pub lvl_shftr_en: LvlShftr,
    reserved18: [u32; 3],
    pub ocm_cfg: RW<u32>,
-    reserved19: [u32; 123],
+    reserved19: [u32; 66],
+    /// barely documented unnamed register to prepare L2 cache setup
+    pub unnamed1: RW<u32>,
+    reserved120: [u32; 56],
    pub gpiob_ctrl: GpiobCtrl,
    pub gpiob_cfg_cmos18: RW<u32>,
    pub gpiob_cfg_cmos25: RW<u32>,
    pub gpiob_cfg_cmos33: RW<u32>,
-    reserved20: [u32; 1],
+    reserved21: [u32; 1],
    pub gpiob_cfg_hstl: RW<u32>,
    pub gpiob_drvr_bias_ctrl: RW<u32>,
-    reserved21: [u32; 9],
+    reserved22: [u32; 9],
    pub ddriob_addr0: DdriobConfig,
    pub ddriob_addr1: DdriobConfig,
    pub ddriob_data0: DdriobConfig,
@ -531,20 +534,6 @@ impl UartRstCtrl {
    }
 }

-register!(gpio_rst_ctrl, GpioRstCtrl, RW, u32);
-register_bit!(gpio_rst_ctrl, gpio_cpu1x_rst, 0);
-register_at!(GpioRstCtrl, 0xF800022C, new);
-impl GpioRstCtrl {
-    pub fn reset_gpio(&mut self) {
-        self.modify(|_, w|
-            w.gpio_cpu1x_rst(true)
-        );
-        self.modify(|_, w|
-            w.gpio_cpu1x_rst(false)
-        );
-    }
-}
-
 register!(lqspi_clk_ctrl, LqspiClkCtrl, RW, u32);
 register_bit!(lqspi_clk_ctrl, clkact, 0);
 register_bits_typed!(lqspi_clk_ctrl, src_sel, u8, PllSource, 4, 5);
--- a/libboard_zynq/src/timer/global.rs
+++ b/libboard_zynq/src/timer/global.rs
@ -138,15 +138,6 @@ where
    }
 }

-impl<U: PartialOrd> CountDown<U>
-where
-    GlobalTimer: TimeSource<U>,
-{
-    pub fn waiting(&self) -> bool {
-        self.timer.now() <= self.timeout
-    }
-}
-
 /// embedded-hal sync API
 impl embedded_hal::blocking::delay::DelayMs<u64> for GlobalTimer {
    fn delay_ms(&mut self, ms: u64) {
--- a/libcortex_a9/Cargo.toml
+++ b/libcortex_a9/Cargo.toml
@ -7,9 +7,9 @@ edition = "2018"
 [features]
 target_zc706 = []
 target_cora_z7_10 = []
-power_saving = []
 default = ["target_zc706"]

 [dependencies]
 bit_field = "0.10"
+volatile-register = "0.2"
 libregister = { path = "../libregister" }
--- a/libcortex_a9/src/asm.rs
+++ b/libcortex_a9/src/asm.rs
@ -33,44 +33,3 @@ pub fn dsb() {
 pub fn isb() {
    unsafe { llvm_asm!("isb" :::: "volatile") }
 }
-
-/// Enable IRQ
-#[inline]
-pub unsafe fn enable_irq() {
-    llvm_asm!("cpsie i":::: "volatile");
-}
-
-/// Disable IRQ, return if IRQ was originally enabled.
-#[inline]
-pub unsafe fn enter_critical() -> bool {
-    let mut cpsr: u32;
-    llvm_asm!(
-        "mrs $0, cpsr
-         cpsid i"
-         : "=r"(cpsr) ::: "volatile");
-    (cpsr & (1 << 7)) == 0
-}
-
-#[inline]
-pub unsafe fn exit_critical(enable: bool) {
-    // https://stackoverflow.com/questions/40019929/temporarily-disable-interrupts-on-arm
-    let mask: u32 = if enable {
-        1 << 7
-    } else {
-        0
-    };
-    llvm_asm!(
-        "mrs r1, cpsr
-         bic r1, r1, $0
-         msr cpsr_c, r1"
-         :: "r"(mask) : "r1");
-}
-
-/// Exiting IRQ
-#[inline]
-pub unsafe fn exit_irq() {
-    llvm_asm!("
-        mrs r0, SPSR
-        msr CPSR, r0
-        " ::: "r0");
-}
--- a/libcortex_a9/src/cache.rs
+++ b/libcortex_a9/src/cache.rs
@ -197,7 +197,7 @@ pub unsafe fn dcimvac(addr: usize) {
    llvm_asm!("mcr p15, 0, $0, c7, c6, 1" :: "r" (addr) :: "volatile");
 }

-/// Data cache clean and invalidate for an object.
+/// Data cache invalidate for an object.
 pub unsafe fn dci<T>(object: &mut T) {
    let first_addr = object as *const _ as usize;
    let beyond_addr = (object as *const _ as usize) + core::mem::size_of_val(object);
--- a/libcortex_a9/src/lib.rs
+++ b/libcortex_a9/src/lib.rs
@ -11,26 +11,10 @@ pub mod cache;
 pub mod mmu;
 pub mod mutex;
 pub mod sync_channel;
-pub mod semaphore;
 mod uncached;
 mod fpu;
 pub use uncached::UncachedSlice;
 pub use fpu::enable_fpu;
+pub mod pl310;

 global_asm!(include_str!("exceptions.s"));
-
-#[inline]
-pub fn spin_lock_yield() {
-    #[cfg(feature = "power_saving")]
-    asm::wfe();
-}
-
-#[inline]
-pub fn notify_spin_lock() {
-    #[cfg(feature = "power_saving")]
-    {
-        asm::dsb();
-        asm::sev();
-    }
-}
-
--- a/libcortex_a9/src/mmu.rs
+++ b/libcortex_a9/src/mmu.rs
@ -158,7 +158,7 @@ impl L1Table {
                global: true,
                shareable: true,
                access: AccessPermissions::FullAccess,
-                tex: 0b101,
+                tex: 0b111,
                domain: 0b1111,
                exec: true,
                cacheable: true,
@ -338,7 +338,7 @@ impl L1Table {
        /* 0xfff00000 - 0xffffffff (256K OCM when mapped to high address space) */
        self.direct_mapped_section(0xfff, L1Section {
            global: true,
-            shareable: true,
+            shareable: false,
            access: AccessPermissions::FullAccess,
            tex: 0b100,
            domain: 0,
--- a/libcortex_a9/src/mutex.rs
+++ b/libcortex_a9/src/mutex.rs
@ -1,10 +1,20 @@
 use core::ops::{Deref, DerefMut};
 use core::sync::atomic::{AtomicU32, Ordering};
 use core::cell::UnsafeCell;
-use super::{
-    spin_lock_yield, notify_spin_lock,
-    asm::{dmb, enter_critical, exit_critical}
-};
+use super::asm::*;
+
+/// [Power-saving features](http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dht0008a/ch01s03s02.html)
+#[inline]
+fn wait_for_update() {
+    wfe();
+}
+
+/// [Power-saving features](http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dht0008a/ch01s03s02.html)
+#[inline]
+fn signal_update() {
+    dsb();
+    sev();
+}

 const LOCKED: u32 = 1;
 const UNLOCKED: u32 = 0;
@ -31,26 +41,19 @@ impl<T> Mutex<T> {

    /// Lock the Mutex, blocks when already locked
    pub fn lock(&self) -> MutexGuard<T> {
-        let mut irq = unsafe { enter_critical() };
        while self.locked.compare_and_swap(UNLOCKED, LOCKED, Ordering::Acquire) != UNLOCKED {
-            unsafe {
-                exit_critical(irq);
-                spin_lock_yield();
-                irq = enter_critical();
-            }
+            wait_for_update();
        }
        dmb();
-        MutexGuard { mutex: self, irq }
+        MutexGuard { mutex: self }
    }

    pub fn try_lock(&self) -> Option<MutexGuard<T>> {
-        let irq = unsafe { enter_critical() };
        if self.locked.compare_and_swap(UNLOCKED, LOCKED, Ordering::Acquire) != UNLOCKED {
-            unsafe { exit_critical(irq) };
            None
        } else {
            dmb();
-            Some(MutexGuard { mutex: self, irq })
+            Some(MutexGuard { mutex: self })
        }
    }

@ -58,7 +61,7 @@ impl<T> Mutex<T> {
        dmb();
        self.locked.store(UNLOCKED, Ordering::Release);

-        notify_spin_lock();
+        signal_update();
    }
 }

@ -66,7 +69,6 @@ impl<T> Mutex<T> {
 /// `Deref`/`DerefMutx`
 pub struct MutexGuard<'a, T> {
    mutex: &'a Mutex<T>,
-    irq: bool,
 }

 impl<'a, T> Deref for MutexGuard<'a, T> {
@ -86,6 +88,5 @@ impl<'a, T> DerefMut for MutexGuard<'a, T> {
 impl<'a, T> Drop for MutexGuard<'a, T> {
    fn drop(&mut self) {
        self.mutex.unlock();
-        unsafe { exit_critical(self.irq) };
    }
 }
--- a/libcortex_a9/src/pl310/mod.rs
+++ b/libcortex_a9/src/pl310/mod.rs
@ -0,0 +1,166 @@
+//! L2 cache controller
+
+use libregister::RegisterW;
+use crate::asm::*;
+
+mod regs;
+
+const CACHE_LINE: usize = 0x20;
+const CACHE_LINE_MASK: usize = CACHE_LINE - 1;
+
+#[inline]
+fn cache_line_addrs(first_addr: usize, beyond_addr: usize) -> impl Iterator<Item = usize> {
+    let first_addr = first_addr & !CACHE_LINE_MASK;
+    let beyond_addr = (beyond_addr | CACHE_LINE_MASK) + 1;
+
+    (first_addr..beyond_addr).step_by(CACHE_LINE)
+}
+
+fn object_cache_line_addrs<T>(object: &T) -> impl Iterator<Item = usize> {
+    let first_addr = object as *const _ as usize;
+    let beyond_addr = (object as *const _ as usize) + core::mem::size_of_val(object);
+    cache_line_addrs(first_addr, beyond_addr)
+}
+
+fn slice_cache_line_addrs<T>(slice: &[T]) -> impl Iterator<Item = usize> {
+    let first_addr = &slice[0] as *const _ as usize;
+    let beyond_addr = (&slice[slice.len() - 1] as *const _ as usize) +
+        core::mem::size_of_val(&slice[slice.len() - 1]);
+    cache_line_addrs(first_addr, beyond_addr)
+}
+
+pub struct L2Cache {
+    pub regs: &'static mut regs::RegisterBlock,
+}
+
+impl L2Cache {
+    pub fn new(register_baseaddr: usize) -> Self {
+        let regs = unsafe {
+            regs::RegisterBlock::new_at(register_baseaddr)
+        };
+        L2Cache { regs }
+    }
+
+    pub fn set_tag_ram_latencies(&mut self, setup_lat: u8, rd_access_lat: u8, wr_access_lat: u8) {
+        self.regs.tag_ram_control.write(
+            regs::RamControl::zeroed()
+                .setup_lat(setup_lat)
+                .rd_access_lat(rd_access_lat)
+                .wr_access_lat(wr_access_lat)
+        );
+    }
+
+    pub fn set_data_ram_latencies(&mut self, setup_lat: u8, rd_access_lat: u8, wr_access_lat: u8) {
+        self.regs.data_ram_control.write(
+            regs::RamControl::zeroed()
+                .setup_lat(setup_lat)
+                .rd_access_lat(rd_access_lat)
+                .wr_access_lat(wr_access_lat)
+        );
+    }
+
+    pub fn disable_interrupts(&mut self) {
+        self.regs.int_mask.write(
+            regs::Interrupts::zeroed()
+                .ecntr(true)
+                .parrt(true)
+                .parrd(true)
+                .errwt(true)
+                .errwd(true)
+                .errrt(true)
+                .errrd(true)
+                .slverr(true)
+                .decerr(true)
+        );
+    }
+
+    pub fn reset_interrupts(&mut self) {
+        self.regs.int_clear.write(
+            regs::Interrupts::zeroed()
+                .ecntr(true)
+                .parrt(true)
+                .parrd(true)
+                .errwt(true)
+                .errwd(true)
+                .errrt(true)
+                .errrd(true)
+                .slverr(true)
+                .decerr(true)
+        );
+    }
+
+    pub fn invalidate_all(&mut self) {
+        unsafe { self.regs.inv_way.write(0xFFFF); }
+        unsafe { self.regs.cache_sync.write(1); }
+        while self.regs.cache_sync.read() != 0 {}
+    }
+
+    pub fn enable(&mut self) {
+        dmb();
+        self.regs.control.write(
+            regs::Control::zeroed()
+                .l2_enable(true)
+        );
+        dsb();
+    }
+
+    pub fn clean_invalidate<T>(&mut self, obj: &T) {
+        dmb();
+        for addr in object_cache_line_addrs(obj) {
+            unsafe {
+                self.regs.clean_inv_pa.write(addr as u32);
+            }
+        }
+        dsb();
+        unsafe { self.regs.cache_sync.write(1); }
+        while self.regs.cache_sync.read() != 0 {}
+    }
+
+    pub fn clean_invalidate_slice<T>(&mut self, slice: &[T]) {
+        dmb();
+        for addr in slice_cache_line_addrs(slice) {
+            unsafe {
+                self.regs.clean_inv_pa.write(addr as u32);
+            }
+        }
+        dsb();
+        unsafe { self.regs.cache_sync.write(1); }
+        while self.regs.cache_sync.read() != 0 {}
+    }
+
+    pub fn clean_slice<T>(&mut self, slice: &[T]) {
+        dmb();
+        for addr in slice_cache_line_addrs(slice) {
+            unsafe {
+                self.regs.clean_pa.write(addr as u32);
+            }
+        }
+        dsb();
+        unsafe { self.regs.cache_sync.write(1); }
+        while self.regs.cache_sync.read() != 0 {}
+    }
+
+    pub fn invalidate<T>(&mut self, obj: &mut T) {
+        dmb();
+        for addr in object_cache_line_addrs(obj) {
+            unsafe {
+                self.regs.inv_pa.write(addr as u32);
+            }
+        }
+        dsb();
+        unsafe { self.regs.cache_sync.write(1); }
+        while self.regs.cache_sync.read() != 0 {}
+    }
+
+    pub fn invalidate_slice<T>(&mut self, slice: &mut [T]) {
+        dmb();
+        for addr in slice_cache_line_addrs(slice) {
+            unsafe {
+                self.regs.inv_pa.write(addr as u32);
+            }
+        }
+        dsb();
+        unsafe { self.regs.cache_sync.write(1); }
+        while self.regs.cache_sync.read() != 0 {}
+    }
+}
--- a/libcortex_a9/src/pl310/regs.rs
+++ b/libcortex_a9/src/pl310/regs.rs
@ -0,0 +1,93 @@
+use volatile_register::{RO, WO, RW};
+use libregister::{register, register_bit, register_bits, RegisterW};
+
+#[repr(C)]
+pub struct RegisterBlock {
+    pub cache_id: RW<u32>,
+    pub cache_type: RW<u32>,
+    pub _unused1: [RO<u32>; 62],
+    pub control: Control,
+    pub aux_control: RW<u32>,
+    pub tag_ram_control: RamControl,
+    pub data_ram_control: RamControl,
+    pub _unused2: [RO<u32>; 60],
+    pub ev_counter_ctrl: RW<u32>,
+    pub ev_counter1_cfg: RW<u32>,
+    pub ev_counter2_cfg: RW<u32>,
+    pub ev_counter1: RW<u32>,
+    pub ev_counter2: RW<u32>,
+    pub int_mask: Interrupts,
+    pub int_mask_status: Interrupts,
+    pub int_raw_status: Interrupts,
+    pub int_clear: Interrupts,
+    pub _unused3: [RO<u32>; 323],
+    pub cache_sync: RW<u32>,
+    pub _unused4: [RO<u32>; 15],
+    pub inv_pa: RW<u32>,
+    pub _unused5: [RO<u32>; 2],
+    pub inv_way: RW<u32>,
+    pub _unused6: [RO<u32>; 12],
+    pub clean_pa: RW<u32>,
+    pub _unused7: [RO<u32>; 1],
+    pub clean_index: RW<u32>,
+    pub clean_way: RW<u32>,
+    pub _unused8: [RO<u32>; 12],
+    pub clean_inv_pa: RW<u32>,
+    pub _unused9: [RO<u32>; 1],
+    pub clean_inv_index: RW<u32>,
+    pub clean_inv_way: RW<u32>,
+    pub _unused10: [RO<u32>; 64],
+    pub d_lockdown0: RW<u32>,
+    pub i_lockdown0: RW<u32>,
+    pub d_lockdown1: RW<u32>,
+    pub i_lockdown1: RW<u32>,
+    pub d_lockdown2: RW<u32>,
+    pub i_lockdown2: RW<u32>,
+    pub d_lockdown3: RW<u32>,
+    pub i_lockdown3: RW<u32>,
+    pub d_lockdown4: RW<u32>,
+    pub i_lockdown4: RW<u32>,
+    pub d_lockdown5: RW<u32>,
+    pub i_lockdown5: RW<u32>,
+    pub d_lockdown6: RW<u32>,
+    pub i_lockdown6: RW<u32>,
+    pub d_lockdown7: RW<u32>,
+    pub i_lockdown7: RW<u32>,
+    pub _unused11: [RO<u32>; 4],
+    pub lock_line_en: RW<u32>,
+    pub unlock_way: RW<u32>,
+    pub _unused12: [RO<u32>; 170],
+    pub addr_filtering_start: RW<u32>,
+    pub addr_filtering_end: RW<u32>,
+    pub _unused13: [RO<u32>; 206],
+    pub debug_ctrl: RW<u32>,
+    pub _unused14: [RO<u32>; 7],
+    pub prefetch_ctrl: RW<u32>,
+    pub _unused15: [RO<u32>; 7],
+    pub power_ctrl: RW<u32>,
+}
+
+impl RegisterBlock {
+    pub unsafe fn new_at(baseaddr: usize) -> &'static mut Self {
+        &mut *(baseaddr as *mut _)
+    }
+}
+
+register!(control, Control, RW, u32);
+register_bit!(control, l2_enable, 0);
+
+register!(ram_control, RamControl, RW, u32);
+register_bits!(ram_control, setup_lat, u8, 0, 2);
+register_bits!(ram_control, rd_access_lat, u8, 4, 6);
+register_bits!(ram_control, wr_access_lat, u8, 8, 10);
+
+register!(interrupts, Interrupts, RW, u32);
+register_bit!(interrupts, ecntr, 0);
+register_bit!(interrupts, parrt, 1);
+register_bit!(interrupts, parrd, 2);
+register_bit!(interrupts, errwt, 3);
+register_bit!(interrupts, errwd, 4);
+register_bit!(interrupts, errrt, 5);
+register_bit!(interrupts, errrd, 6);
+register_bit!(interrupts, slverr, 7);
+register_bit!(interrupts, decerr, 8);
--- a/libcortex_a9/src/regs.rs
+++ b/libcortex_a9/src/regs.rs
@ -149,7 +149,6 @@ pub struct ACTLR;
 wrap_reg!(actlr);
 def_reg_r!(ACTLR, actlr::Read, "mrc p15, 0, $0, c1, c0, 1");
 def_reg_w!(ACTLR, actlr::Write, "mcr p15, 0, $0, c1, c0, 1");
-// SMP bit
 register_bit!(actlr, parity_on, 9);
 register_bit!(actlr, alloc_one_way, 8);
 register_bit!(actlr, excl, 7);
--- a/libcortex_a9/src/semaphore.rs
+++ b/libcortex_a9/src/semaphore.rs
@ -1,71 +0,0 @@
-use super::{spin_lock_yield, notify_spin_lock};
-use core::{
-    task::{Context, Poll},
-    pin::Pin,
-    future::Future,
-    sync::atomic::{AtomicI32, Ordering}
-};
-
-pub struct Semaphore {
-    value: AtomicI32,
-    max: i32
-}
-
-impl Semaphore {
-    pub fn new(value: i32, max: i32) -> Self {
-        Semaphore { value: AtomicI32::new(value), max}
-    }
-
-    pub fn try_wait(&self) -> Option<()> {
-        loop {
-            let value = self.value.load(Ordering::Relaxed);
-            if value > 0 {
-                if self.value.compare_and_swap(value, value - 1, Ordering::SeqCst) == value {
-                    return Some(());
-                }
-            } else {
-                return None;
-            }
-        }
-    }
-
-    pub fn wait(&self) {
-        while self.try_wait().is_none() {
-            spin_lock_yield();
-        }
-    }
-
-    pub async fn async_wait(&self) {
-        struct Fut<'a>(&'a Semaphore);
-
-        impl Future for Fut<'_> {
-            type Output = ();
-            fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
-                match self.0.try_wait() {
-                    Some(_) => Poll::Ready(()),
-                    None => {
-                        cx.waker().wake_by_ref();
-                        Poll::Pending
-                    }
-                }
-            }
-        }
-
-        Fut(&self).await
-    }
-
-    pub fn signal(&self) {
-        loop {
-            let value = self.value.load(Ordering::Relaxed);
-            if value < self.max {
-                if self.value.compare_and_swap(value, value + 1, Ordering::SeqCst) == value {
-                    notify_spin_lock();
-                    return;
-                }
-            } else {
-                return;
-            }
-        }
-    }
-}
-
--- a/libcortex_a9/src/sync_channel.rs
+++ b/libcortex_a9/src/sync_channel.rs
@ -6,7 +6,7 @@ use core::{
    task::{Context, Poll},
 };
 use alloc::boxed::Box;
-use super::{spin_lock_yield, notify_spin_lock};
+use super::asm::*;

 pub struct Sender<'a, T> where T: Clone {
    list: &'a [AtomicPtr<T>],
@ -35,7 +35,9 @@ impl<'a, T> Sender<'a, T> where T: Clone {
            let prev = entry.swap(ptr, Ordering::Relaxed);
            // we allow other end get it first
            self.write.store((write + 1) % self.list.len(), Ordering::Release);
-            notify_spin_lock();
+            // wake up other core, actually I wonder if the dsb is really needed...
+            dsb();
+            sev();
            if !prev.is_null() {
                unsafe {
                    drop_in_place(prev);
@ -49,7 +51,7 @@ impl<'a, T> Sender<'a, T> where T: Clone {
        let mut content = content;
        while let Err(back) = self.try_send(content) {
            content = back;
-            spin_lock_yield();
+            wfe();
        }
    }

@ -85,27 +87,6 @@ impl<'a, T> Sender<'a, T> where T: Clone {
            content: Err(content.into()),
        }.await
    }
-
-    /// free all items in the queue. It is the user's responsibility to
-    /// ensure no reader is trying to copy the data.
-    pub unsafe fn drop_elements(&mut self) {
-        for v in self.list.iter() {
-            let original = v.swap(core::ptr::null_mut(), Ordering::Relaxed);
-            if !original.is_null() {
-                drop_in_place(original);
-            }
-        }
-    }
-
-    /// Reset the `sync_channel`, *forget* all items in the queue. Affects both the sender and
-    /// receiver.
-    pub unsafe fn reset(&mut self) {
-        self.write.store(0, Ordering::Relaxed);
-        self.read.store(0, Ordering::Relaxed);
-        for v in self.list.iter() {
-            v.store(core::ptr::null_mut(), Ordering::Relaxed);
-        }
-    }
 }

 impl<'a, T> Receiver<'a, T> where T: Clone {
@ -125,7 +106,9 @@ impl<'a, T> Receiver<'a, T> where T: Clone {
            };
            let result = data.clone();
            self.read.store((read + 1) % self.list.len(), Ordering::Release);
-            notify_spin_lock();
+            // wake up other core, still idk if the dsb is needed...
+            dsb();
+            sev();
            Ok(result)
        }
    }
@ -135,7 +118,7 @@ impl<'a, T> Receiver<'a, T> where T: Clone {
            if let Ok(data) = self.try_recv() {
                return data;
            }
-            spin_lock_yield();
+            wfe();
        }
    }

--- a/libcortex_a9/src/uncached.rs
+++ b/libcortex_a9/src/uncached.rs
@ -25,9 +25,10 @@ impl<T> UncachedSlice<T> {
        for page_start in (start..(start + size)).step_by(L1_PAGE_SIZE) {
            L1Table::get()
                .update(page_start as *const (), |l1_section| {
-                    l1_section.tex = 0b100;
+                    // Shareable Device
+                    l1_section.tex = 0b000;
                    l1_section.cacheable = false;
-                    l1_section.bufferable = false;
+                    l1_section.bufferable = true;
                });
        }

--- a/libsupport_zynq/Cargo.toml
+++ b/libsupport_zynq/Cargo.toml
@ -9,10 +9,8 @@ edition = "2018"
 target_zc706 = ["libboard_zynq/target_zc706"]
 target_cora_z7_10 = ["libboard_zynq/target_cora_z7_10"]
 panic_handler = []
-dummy_irq_handler = []
-alloc_core = []

-default = ["panic_handler", "dummy_irq_handler"]
+default = ["panic_handler"]

 [dependencies]
 r0 = "1"
--- a/libsupport_zynq/src/abort.rs
+++ b/libsupport_zynq/src/abort.rs
@ -53,7 +53,6 @@ pub unsafe extern "C" fn ReservedException() {
 #[link_section = ".text.boot"]
 #[no_mangle]
 #[naked]
-#[cfg(feature = "dummy_irq_handler")]
 pub unsafe extern "C" fn IRQ() {
    stdio::drop_uart();
    println!("IRQ");
--- a/libsupport_zynq/src/boot.rs
+++ b/libsupport_zynq/src/boot.rs
@ -4,7 +4,7 @@ use libregister::{
    VolatileCell,
    RegisterR, RegisterW, RegisterRW,
 };
-use libcortex_a9::{asm, regs::*, cache, mmu, spin_lock_yield, notify_spin_lock};
+use libcortex_a9::{asm, regs::*, cache, mmu};
 use libboard_zynq::{slcr, mpcore};

 extern "C" {
@ -29,7 +29,7 @@ pub unsafe extern "C" fn Reset() -> ! {
        }
        1 => {
            while !CORE1_ENABLED.get() {
-                spin_lock_yield();
+                asm::wfe();
            }
            SP.write(&mut __stack1_start as *mut _ as u32);
            boot_core1();
@ -57,7 +57,6 @@ unsafe fn boot_core0() -> ! {
        asm::dmb();
        asm::dsb();

-        asm::enable_irq();
        main_core0();
        panic!("return from main");
    });
@ -78,7 +77,6 @@ unsafe fn boot_core1() -> ! {
        asm::dmb();
        asm::dsb();

-        asm::enable_irq();
        main_core1();
        panic!("return from main_core1");
    });
@ -144,7 +142,6 @@ impl Core1 {
            slcr.a9_cpu_rst_ctrl.modify(|_, w| w.a9_rst1(false));
            slcr.a9_cpu_rst_ctrl.modify(|_, w| w.a9_clkstop1(false));
        });
-        notify_spin_lock();

        Core1 {}
    }
--- a/libsupport_zynq/src/ram.rs
+++ b/libsupport_zynq/src/ram.rs
@ -1,97 +1,55 @@
-use alloc::alloc::Layout;
 use core::alloc::GlobalAlloc;
 use core::ptr::NonNull;
-use libcortex_a9::{
-    mutex::Mutex,
-    regs::MPIDR
-};
-use libregister::RegisterR;
+use alloc::alloc::Layout;
 use linked_list_allocator::Heap;
-#[cfg(not(feature = "alloc_core"))]
+use libcortex_a9::mutex::Mutex;
 use libboard_zynq::ddr::DdrRam;

 #[global_allocator]
-static ALLOCATOR: CortexA9Alloc = CortexA9Alloc(
-    Mutex::new(Heap::empty()),
-    Mutex::new(Heap::empty()),
-);
+static ALLOCATOR: CortexA9Alloc = CortexA9Alloc(Mutex::new(Heap::empty()));

-struct CortexA9Alloc(Mutex<Heap>, Mutex<Heap>);
+/// LockedHeap doesn't lock properly
+struct CortexA9Alloc(Mutex<Heap>);

 unsafe impl Sync for CortexA9Alloc {}

 unsafe impl GlobalAlloc for CortexA9Alloc {
    unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
-        if cfg!(not(feature = "alloc_core")) || MPIDR.read().cpu_id() == 0 {
-            &self.0
-        } else {
-            &self.1
-        }
-        .lock()
+        self.0.lock()
            .allocate_first_fit(layout)
            .ok()
            .map_or(0 as *mut u8, |allocation| allocation.as_ptr())
    }

    unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
-        if cfg!(not(feature = "alloc_core")) || MPIDR.read().cpu_id() == 0 {
-            &self.0
-        } else {
-            &self.1
-        }
-        .lock()
+        self.0.lock()
            .deallocate(NonNull::new_unchecked(ptr), layout)
    }
 }

-#[cfg(not(feature = "alloc_core"))]
 pub fn init_alloc_ddr(ddr: &mut DdrRam) {
    unsafe {
-        ALLOCATOR
-            .0
-            .lock()
+        ALLOCATOR.0.lock()
            .init(ddr.ptr::<u8>() as usize, ddr.size());
    }
 }

 extern "C" {
-    static __heap0_start: usize;
-    static __heap0_end: usize;
-    #[cfg(feature = "alloc_core")]
-    static __heap1_start: usize;
-    #[cfg(feature = "alloc_core")]
-    static __heap1_end: usize;
+    static __heap_start: usize;
+    static __heap_end: usize;
 }

-pub fn init_alloc_core0() {
+pub fn init_alloc_linker() {
    unsafe {
-        let start = &__heap0_start as *const usize as usize;
-        let end = &__heap0_end as *const usize as usize;
-        ALLOCATOR.0.lock().init(start, end - start);
+        let start = &__heap_start as *const usize as usize;
+        let end = &__heap_end as *const usize as usize;
+        ALLOCATOR.0.lock()
+            .init(start, end - start);
    }
 }

-#[cfg(feature = "alloc_core")]
-pub fn init_alloc_core1() {
-    unsafe {
-        let start = &__heap1_start as *const usize as usize;
-        let end = &__heap1_end as *const usize as usize;
-        ALLOCATOR.1.lock().init(start, end - start);
-    }
-}

 #[alloc_error_handler]
-fn alloc_error(layout: core::alloc::Layout) -> ! {
-    let id = MPIDR.read().cpu_id();
-    let used = if cfg!(not(feature = "alloc_core")) || id == 0 {
-        ALLOCATOR.0.lock().used()
-    } else {
-        ALLOCATOR.1.lock().used()
-    };
-    panic!(
-        "Core {} alloc_error, layout: {:?}, used memory: {}",
-        id,
-        layout,
-        used
-    );
+fn alloc_error(_: core::alloc::Layout) -> ! {
+    panic!("alloc_error")
 }
--- a/shell.nix
+++ b/shell.nix
@ -19,6 +19,6 @@ stdenv.mkDerivation {
  XARGO_RUST_SRC = "${rustcSrc}/src";

  shellHook = ''
-    echo "Run 'cargo xbuild --release -p experiments' to build."
+    echo "Run 'cargo xbuild --release' to build."
  '';
 }
Author	SHA1	Message	Date
Astro	7c8ec46c8f	eth: use UncachedSlice for both desc and buffers	2020-07-29 22:25:08 +02:00
Astro	8267acfcba	eth: latter tuning	2020-07-29 22:12:28 +02:00
Astro	7ae8be58cf	eth: fix cache maintenance	2020-07-29 22:12:28 +02:00
Astro	7f3e75e20c	mmu: set L2-bufferable for DDR	2020-07-29 22:12:28 +02:00
Astro	8c26974816	eth: warn on tx ring overflow	2020-07-29 22:09:48 +02:00
Astro	c86d8b2af2	experiments: setup_l2cache(), add bandwidth tester	2020-07-29 22:09:48 +02:00
Astro	ac89367f8f	uncached: setup Shareable Device memory	2020-07-29 22:06:07 +02:00
Astro	0b99b0a864	armv7-none-eabihf.json: forbid unaligned access	2020-07-29 22:05:24 +02:00
Astro	4cb71e4f3d	libcortex_a9: implement pl310 l2cache	2020-07-29 22:05:24 +02:00