BLOG

Home Blog Videos Research LispBM About Privacy Policy
Bo Joel Svensson
blog (dot) joel (dot) svensson (at) gmail (dot) com

Example Implementation of a REPL Running on Zephyr OS

An earlier text here showed an example of how to implement a lispBM REPL in a ChibiOS thread. In this text we instead take a look at Zephyr. I have only used Zephyr with the NRF52 Microcontroller, the nrf52840 to be precise. This is quite a powerful microcontroller featuring an ARM Cortex m4 at 64MHz, 1MB of flash and 256K of RAM. The nrf52840 also comes with Bluetooth. Some background on what lispBM is can be found here.

When I first tried Zephyr OS, I felt that setting it up and getting to the point where you can build some programs with it was quite complicated. There is a video here about the steps involved in getting to a point where you can just edit your code and run make.

The code in the listings here below, is all available within the lispBM repository on github.

Introduction to the Code

I will start by going over the code involved and later in the text go into the configuration files and make scripts.

All IO will be done over USB and when hooking the nrf52 based development board up to a Linux computer it will appear as a CDC ACM device and will be accessible through /dev/ttyACMX where X is a number. Which ttyACM it gets attached as can be seen by running the command dmesg after attaching the device (note that it has to run firmware that configures the USB to appear as a device when connected in this way). As I understand it CDC ACM stands for "Communication Device Class" and "Abstract Control Model" but the exact implications of those words are a bit beyond me. So, that is how the device will appear as viewed from the Linux machine connecting to it, on the firmware side commucation will look like a UART connection and we will interact with that UART using interrupts.

#include <device.h>
#include <drivers/uart.h>
#include <zephyr.h>
#include <sys/ring_buffer.h>

#include "heap.h"
#include "symrepr.h"
#include "eval_cps.h"
#include "print.h"
#include "tokpar.h"
#include "prelude.h"


#define LISPBM_HEAP_SIZE 2048
#define LISPBM_OUTPUT_BUFFER_SIZE 4096
#define LISPBM_INPUT_BUFFER_SIZE 1024

The code starts out including the parts needed form Zephyr OS. device and uart are involved in the low level aspects of communication, the ring_buffer is used to store data received and buffered to be sent over the uart.

The the header files from lispBM are also included and some sizes that are going to be used further down are defined. The heap size is measured in number of cons-cells (so size in bytes is 8 * LISPBM_HEAP_SIZE) while the input and output (used for text interaction with the user) buffer sizes are in bytes.

Input and Output over USB on the NRF52

The interaction with the uart will be interrupt based. The interrupt will occur when data is arriving on the uart or when we trigger an interrupt because we have added data to be sent. A couple of slightly higher level functions, get_char, put_char and usb_printf are implemented as well. The interface between the low-level interrupt routine and the higher level functions are a pair of ringbuffers that are defined below. For example, when the interrupt routine goes off because there is data coming in, the data is added to the in_ringbuf.

#define RING_BUF_SIZE 1024
u8_t in_ring_buffer[RING_BUF_SIZE];
u8_t out_ring_buffer[RING_BUF_SIZE];

struct device *dev;

struct ring_buf in_ringbuf;
struct ring_buf out_ringbuf;

The in_ring_buffer and out_ring_buffer is the data storage arrays for input and output data and the rest of the state that is needed for ringbuffer functionality is in the struct ring_buf. This ringbuffer implementation is a part of the Zephyr OS.

Above, a device is also declared. These device structs are how Zephyr handles perpiherals (even down to a single GPIO pin seems to need one of these device structs declared). This device will represent the USB peripheral.

The interrupt_handler defined below runs for as long as there are uart interrupts pending and for each pending interrupt there is a check to see if there is data to read or to send.

If data is comming in it is added to the in_ringbuf. If more data is arriving than there is space in the buffer, those bytes are just ignored.

The case where there is data to send, data is read from the out_ringbuf and sent over the uart.

static void interrupt_handler(struct device *dev)
{
  while (uart_irq_update(dev) && uart_irq_is_pending(dev)) {
    if (uart_irq_rx_ready(dev)) {
      int recv_len, rb_len;
      u8_t buffer[64];
      size_t len = MIN(ring_buf_space_get(&in_ringbuf),
                       sizeof(buffer));

      recv_len = uart_fifo_read(dev, buffer, len);

      rb_len = ring_buf_put(&in_ringbuf, buffer, recv_len);
      if (rb_len < recv_len) {
        //silently dropping bytes
      }
    }

    if (uart_irq_tx_ready(dev)) {
      u8_t buffer[64];
      int rb_len, send_len;

      rb_len = ring_buf_get(&out_ringbuf, buffer, sizeof(buffer));
      if (!rb_len) {
        uart_irq_tx_disable(dev);
        continue;
      }

      send_len = uart_fifo_fill(dev, buffer, rb_len);
    }
  }
}

If I remember correcly, the interrupt routine above is just a tiny tweak of code from the CDC ACM example provided with Zephyr.

Now, get_char. This function returns an integer which is -1 on failure to get a character, otherwise the result is the character read. In this function interrupts are postponed while interacting with the ringbuffer, I'm not sure this is actually required (since we are operating at a single byte at a time). The function tries to read one byte from the input ringbuffer and if that is successful, the byte is returned as result.

int get_char() {

  int n;
  u8_t c;
  unsigned int key = irq_lock();
  n = ring_buf_get(&in_ringbuf, &c, 1);
  irq_unlock(key);
  if (n == 1) {
    return c;
  }
  return -1;
}

The put_char function takes an integer as input (not a char) this is so that it can be directly hooked up to get_char if one would like (put_char(get_char())). So putchar checks if the int represents a character and in that case adds it to the output ringbuffer. The same postponing of interrupts occur here.

void put_char(int i) {
  if (i >= 0 && i < 256) {

    u8_t c = (u8_t)i;
    unsigned int key = irq_lock();
    ring_buf_put(&out_ringbuf, &c, 1);
    uart_irq_tx_enable(dev);
    irq_unlock(key);
  }
}

A printf-like function is always useful! Here it is called usb_printf. One idea would be to implement this using put_char, but it feels more efficient to implement it by writing a whole bunch of bytes into the output ringbuffer at a time instead. The print_buffer used internally in this function can hold up to at 4096 bytes, which is more than the rinbuffers can hold at once. Because of this there is a loop that in each iteration adds chunks from the print_buffer to the ringbuffer (as many bytes as it can), and then triggers the interrupt that should free up new space in the ringbuffer.

void usb_printf(char *format, ...) {

  va_list arg;
  va_start(arg, format);
  int len;
  static char print_buffer[4096];

  len = vsnprintf(print_buffer, 4096,format, arg);
  va_end(arg);

  int num_written = 0;
  while (len - num_written > 0) {
    unsigned int key = irq_lock();
    num_written +=
      ring_buf_put(&out_ringbuf,
                   (print_buffer + num_written),
                   (len - num_written));
    irq_unlock(key);
    uart_irq_tx_enable(dev);
  }
}

To read lines of text from the user an inputline function is used. This function should be more or less identical to the same function shown in the ChibiOS REPL. So I wont go into any detail of this here.

int inputline(char *buffer, int size) {
  int n = 0;
  int c;
  for (n = 0; n < size - 1; n++) {

    c = get_char();
    switch (c) {
    case 127: /* fall through to below */
    case '\b': /* backspace character received */
      if (n > 0)
        n--;
      buffer[n] = 0;
      put_char('\b'); /* output backspace character */
      n--; /* set up next iteration to deal with preceding char location */
      break;
    case '\n': /* fall through to \r */
    case '\r':
      buffer[n] = 0;
      return n;
    default:
      if (c != -1 && c < 256) {
        put_char(c);
        buffer[n] = c;
      } else {
        n --;
      }

      break;
    }
  }
  buffer[size - 1] = 0;
  return 0; // Filled up buffer without reading a linebreak
}

Setting up Communication and Implementing the REPL

The main function starts out exactly like the CDC ACM example that comes with Zephyr. This example code can be found within your Zephyr directory zephyr/samples/subsys/usb/cdc_acm/src/main.c

Here the device is created, ringbuffers initialized, the uart configured and the interrupt_handler function is set to be called upon an uart interrupt.

void main(void)
{

  u32_t baudrate, dtr = 0U;

  dev = device_get_binding("CDC_ACM_0");
  if (!dev) {
    return;
  }

  ring_buf_init(&in_ringbuf, sizeof(in_ring_buffer), in_ring_buffer);
  ring_buf_init(&out_ringbuf, sizeof(out_ring_buffer), out_ring_buffer);

  while (true) {
    uart_line_ctrl_get(dev, LINE_CTRL_DTR, &dtr);
    if (dtr) {
      break;
    } else {
      k_sleep(100);
    }
  }

  uart_line_ctrl_set(dev, LINE_CTRL_DCD, 1);
  uart_line_ctrl_set(dev, LINE_CTRL_DSR, 1); 

  k_busy_wait(1000000);

  uart_line_ctrl_get(dev, LINE_CTRL_BAUD_RATE, &baudrate);
  
  uart_irq_callback_set(dev, interrupt_handler);
  
  uart_irq_rx_enable(dev);

In this example the REPL will run in the main thread, unlike the ChibiOS example where a second thread was created.

After allocating space for input and output of text exchanged with the user, the lispBM subsystems are started up.

  usb_printf("Allocating input/output buffers\n\r");
  char *str = malloc(LISPBM_INPUT_BUFFER_SIZE);
  char *outbuf = malloc(LISPBM_OUTPUT_BUFFER_SIZE);
  int res = 0;

  heap_state_t heap_state;

  res = symrepr_init();
  if (res)
    usb_printf("Symrepr initialized.\n\r");
  else {
    usb_printf("Error initializing symrepr!\n\r");
    return;
  }

  res = heap_init(LISPBM_HEAP_SIZE);
  if (res)
    usb_printf("Heap initialized. Free cons cells: %u\n\r", heap_num_free());
  else {
    usb_printf("Error initializing heap!\n\r");
    return;
  }

  res = eval_cps_init(false);
  if (res)
    usb_printf("Evaluator initialized.\n\r");
  else {
    usb_printf("Error initializing evaluator.\n\r");
  }
    
  VALUE prelude = prelude_load();
  eval_cps_program(prelude);

At the end here a small library is loaded, called the prelude, and evaluated. The prelude consists of a series of definitions of functions and evaluating this prelude primes the environment with this set of functions.

After this setup, the REPL loop begins. It prints a prompt. Clears input and output buffers and then reads a line from the user. If the line read is a command for the REPL, :info or quit those are processed otherwise the input is parsed by the tokpar_parse function and the result of that parsing (a heap structure) is evaluated.

  usb_printf("Lisp REPL started (ZephyrOS)!\n\r");
    
  while (1) {
    k_sleep(100);
    usb_printf("# ");
    memset(str,0,LISPBM_INPUT_BUFFER_SIZE);
    memset(outbuf,0, LISPBM_OUTPUT_BUFFER_SIZE);
    inputline(str, LISPBM_INPUT_BUFFER_SIZE);
    usb_printf("\n\r");

    if (strncmp(str, ":info", 5) == 0) {
      usb_printf("##(REPL - ZephyrOS)#########################################\n\r");
      usb_printf("Used cons cells: %lu \n\r", LISPBM_HEAP_SIZE - heap_num_free());
      usb_printf("ENV: "); simple_snprint(outbuf, LISPBM_OUTPUT_BUFFER_SIZE, eval_cps_get_env()); usb_printf("%s \n\r", outbuf);
      heap_get_state(&heap_state);
      usb_printf("GC counter: %lu\n\r", heap_state.gc_num);
      usb_printf("Recovered: %lu\n\r", heap_state.gc_recovered);
      usb_printf("Marked: %lu\n\r", heap_state.gc_marked);
      usb_printf("Free cons cells: %lu\n\r", heap_num_free());
      usb_printf("############################################################\n\r");
      memset(outbuf,0, LISPBM_OUTPUT_BUFFER_SIZE);
    } else if (strncmp(str, ":quit", 5) == 0) {
      break;
    } else {

      VALUE t;
      t = tokpar_parse(str);

      t = eval_cps_program(t);

      if (dec_sym(t) == symrepr_eerror()) {
        usb_printf("Error\n");
      } else {
        usb_printf("> "); simple_snprint(outbuf, LISPBM_OUTPUT_BUFFER_SIZE, t); usb_printf("%s \n\r", outbuf);
      }
    }
  }

  symrepr_del();
  heap_del();
}

If the evaluation is successful, the result is printed and the loop is executed again.

Building Zephyr OS

The code and other files are stored in a directory structure that looks as follows.

repl-zephyr
├── CMakeLists.txt
├── Kconfig
├── prj.conf
└── src
    └── main.c

We will now look at those files that have been altered for this example. The Kconfig file is unchanged from the CDC ACM example from Zephyr.

Starting with the CMakeLists.txt file that has been altered to also pull in and build the source files from lispBM. The repl-zephyr directory is located inside of the lispBM source tree directly under its root so when the CMakeLists.txt file refers to lispBM those paths start out with ../. The CMakeLists.txt is used by cmake to create a traditional Makefile that can be processed with the command make. The contents of the CMakeLists.txt file is shown below.

cmake_minimum_required(VERSION 3.13.1)
include($ENV{ZEPHYR_BASE}/cmake/app/boilerplate.cmake NO_POLICY_SCOPE)
project(repl)

add_definitions(-D_32_BIT_ -D_PRELUDE -DTINY_SYMTAB)

add_custom_command(OUTPUT ../src/prelude.xxd 
                   COMMAND xxd
                   ARGS -i < ../src/prelude.lisp > ../src/prelude.xxd
                   DEPENDS ../src/prelude.lisp
                  )

FILE(GLOB app_sources src/*.c)
FILE(GLOB lisp_sources ../src/*.c)
target_sources(app PRIVATE ${app_sources}
                   PRIVATE ${lisp_sources}
                   PRIVATE ../src/prelude.xxd)
target_include_directories(app PRIVATE ../include
                               PRIVATE ../src)

The add_definitions command means the generated Makefile should when building c source use the flags -D_32_BIT_, -D_PRELUDE and -DTINY_SYMTAB. This needs some polishing I see as the naming convention here is not that uniform. Also, the _32_BIT_ flag is pretty pointless as I have now decided that I will not put any effort into making it compatible with architectures that are not 32bit. However, the remaining defines mean that the prelude library will be built into the binary and that the tiny version of the symbol table will be used. See this blog post for more information on the symbol representation table.

To build the prelude into the binary, the lispBM source code of the library is "compiled" into a C array of bytes in the file prelude.xxd that will be baked into the executable and later parsed when starting up the REPL. This slightly special compilation step is added in the CMakeLists.txt file as a custom command that will be executed as part of the make procedure.

The CMakeLists.txt file is also told where the additional source and header files can be found from the lispBM source tree. This is done with the target_include_directories command for headers and the target_sources for source code.

Next up is the prj.conf file that allows us to configure various aspects of the Zephyr system. This is also derived from the CDC ACM example but with some additions.

For one, the CONFIG_NEWLIB_LIBC flag is set to y to build with newlib functionality. Newlib is C library (like the standard library) intended for embedded platforms and provides many of the functions one may be familiar with. I use it mainly to have access to a malloc function.

CONFIG_GPIO=y
CONFIG_USB=y
CONFIG_USB_DEVICE_STACK=y
CONFIG_USB_DEVICE_PRODUCT="Zephyr CDC ACM sample"
CONFIG_USB_CDC_ACM=y
CONFIG_SERIAL=y
CONFIG_UART_INTERRUPT_DRIVEN=y
CONFIG_UART_LINE_CTRL=y
CONFIG_NEWLIB_LIBC=y
CONFIG_MAIN_STACK_SIZE=8192


CONFIG_SYSTEM_WORKQUEUE_STACK_SIZE=2048
CONFIG_FLASH=y
CONFIG_FLASH_PAGE_LAYOUT=y
CONFIG_FLASH_MAP=y
CONFIG_FCB=y
CONFIG_SETTINGS=y
CONFIG_SETTINGS_FCB=y


# Settings needed for custom PCB based on RIGADO BMD-340
CONFIG_CLOCK_CONTROL_NRF_K32SRC_XTAL=n
CONFIG_CLOCK_CONTROL_NRF_K32SRC_RC=y
CONFIG_GPIO_AS_PINRESET=n

Towards the last lines of the prj.conf file contains some clock configuration information. If using a development board like the nrf52480-PCA10056 these can be commented out. If you use a custom board, perhaps based on the BMD340 module, these are needed. The difference is that on the latter boards there wont be any external clock source and an internal clock source should be used. The GPIO_AS_PINRESET option is related to how I flash these custom boards using SWD and openocd and don't have access to any on board j-link programmer as on the vendor provided development boards.

To help with building of Zephyr I use some scripts. The first one, below, I call set_fw_build.sh. This is from the lispBM directory. Running the script creates a build directory and runs cmake on the source tree (that contains the CMakeLists.txt file. This run of cmake populates the build directory with (among other things) a Makefile.

#!/bin/bash

if [ -d "repl-zephyr_build" ]; then
    echo "Build directory exists!"
else
    mkdir repl-zephyr_build
    cd repl-zephyr_build
    cmake -G "Eclipse CDT4 - Unix Makefiles" -DBOARD=nrf52840_pca10056 ../repl-zephyr
fi    

Before jumping into the build directory and running make, some additional setup may be needed (depending on how you have configured your system for Zephyr). Zephyr needs to now a bit more about our intentions. For example, we are going to use the arm-none-eabi gcc compiler suite for cross-compilation to out target platform.

There is another file, called zephyr-source-me.sh, that is meant to be sourced (source zephyr-source-me) to include it's settings into the currently running bash shells environment. So, in this file, the ZEPHYR_TOOLCHAIN_VARIANT is set to cross-compile and the CROSS_COMPILE variable is set to the prefix that should be used for all calls to different tools (such as gcc).

It also sources another file of settings from the Zephyr directory tree called zephyr-env.sh

#!/bin/bash

# Tweak this line to reflect your setup
export ZEPHYR_TOOLCHAIN_VARIANT=cross-compile
export CROSS_COMPILE=$HOME/opt/gcc-arm-none-eabi-9-2019-q4-major/bin/arm-none-eabi-

source ../zephyrproject/zephyr/zephyr-env.sh 

After these last bits of setup, it is time to go into the build directory and run make, hopefully successfully.

If the build is successful, the resulting binary can be flashed onto a custom board using an ST-Link and OpenOCD by issuing the following command.

openocd -f interface/stlink.cfg -f target/nrf52.cfg  -c "init" -c "program repl-zephyr_build/zephyr/zephyr.hex verify reset exit"

If instead you are using the nrf52480-PCA10056 I think it should be as easy as running make flash after successfully running make.


HOME

Please contact me with questions, suggestions or feedback at blog (dot) joel (dot) svensson (at) gmail (dot) com or join the google group .

© Copyright 2020 Bo Joel Svensson

This page was generated using Pandoc.