Java Periphery has been merged into Java UIO. I'll leave this up as a place holder for a while.
Java Periphery is a high performance library for GPIO, LED, PWM, SPI, I2C, MMIO and Serial peripheral I/O interface access in userspace Linux. Rather than try to build this from scratch I used c-periphery and HawtJNI to generate the JNI wrappers. This saves a lot of hand coding and allows for easier synchronization with c-periphery changes moving forward.
- Generates JNI source code for c-periphery.
- Generates an autoconf and msbuild source project to build the native library. This gets attached to the Maven project as as the native source zip file.
- Builds the native source tar for the current platform.
- Built native library is stored in a platform specific jar. This gets attached to the Maven project as a platform specific jar file.
- Cross platform MMIO GPIO that doesn't require one off code for each board. Only a simple property file is required to map registers.
- All demo programs require command line arguments to set devices, lines, etc. No need to compile one off examples, change property files or use worthless mocks.
- All wrapper classes support AutoCloseable, so you can use the try-with-resources statement to automatically close and free native resources. This prevents hard to track down native memory leaks.
try (final var spi = new Spi("/dev/spidev1.0", 0, 500000)) {
final var txBuf = new byte[128];
// Change some data at beginning and end.
txBuf[0] = (byte) 0xff;
txBuf[127] = (byte) 0x80;
final var rxBuf = new byte[128];
Spi.spiTransfer(spi.getHandle(), txBuf, rxBuf, txBuf.length);
logger.info(String.format("%02X, %02X", (short) rxBuf[0] & 0xff, (short) rxBuf[127] & 0xff));
}
Behold the FrankenDuo which is used to test all Java Periphery features.
Java Periphery will be targeting Armbian, but the code should work with most Linux distributions. Demo apps are included that illustrate how to leverage the bindings. The idea is to have consistent APIs across C, Python, Lua, Dart and JVM languages without having to use one off board specific drivers, deprecated wiringPi or the deprecated sysfs interface. The possibility of using other JVM based languages such as Groovy, Kotlin, Scala, etc. opens up more language opportunities.
- Why Linux userspace? This is really the only way to get cross platform libraries to work since most SBCs have different chip sets. The trade off is performance compared to native C written to specific chip sets. However, since I'm wrapping C with JNI it guarantees the fastest userspace experience for Java.
- Why Armbian? Because Armbian supports many SBCs and the idea is to be truly SBC cross platform. See downloads.
- Why Java 17? Because Java 17 is the current LTS version of Java. Java 11 the previous LTS release is end of life September 2023. I'm only moving forward with Java. You can always create a fork and make a Java 8 or Java 11 version of Java Periphery.
- Why Zulu OpenJDK? Because it's easy to download without all the crap Oracle puts you through. You can always use another JDK 17 vendor, but you will have to do that manually. Liberica JDK offers an ARM32 version of JDK 17, but it's 50% slower than Zulu JDK 11. At This point in time there are few JDK 17 choices for ARM32. I will update the project once a decent JDK 17 is available for ARM32.
- If you are using Armbian then use
armbian-config
or edit/boot/armbianEnv.txt
to configure various devices. Userspace devices are exposed through /dev or /sys. Verify the device is showing up prior to trying demo apps.sudo apt install armbian-config
- If you are not using Armbian then you will need to know how to configure devices to be exposed to userspace for your Linux distribution and SBC model. Check each log in scripts directory to be sure there were no errors after running install.sh.
- Since linux 4.8 the GPIO sysfs interface is deprecated. Userspace should use the character device instead.
- I have tested 32 bit and 64 bit boards using the latest Armbian release or in the case of the Raspberry Pi Ubuntu Server. The ability to switch seamlessly between boards gives you a wide range of SBC choices. I'm currently testing with Ubuntu 20.04 LTS Focal Fossa using 5.4 to 5.9 kernels.
On the NanoPi Duo the built in button causes it to shutdown by default. You can remove the r_gpio_keys section in the DTB as follows (this may work on other SBCs, but you'll need to know the correct dtb file and section to remove) :
cd /boot/dtb
sudo cp sun8i-h2-plus-nanopi-duo.dtb sun8i-h2-plus-nanopi-duo.dtb.old
sudo dtc -@ -I dtb -O dts -o sun8i-h2-plus-nanopi-duo.dts sun8i-h2-plus-nanopi-duo.dtb
sudo nano sun8i-h2-plus-nanopi-duo.dts
- Remove
r_gpio_keys
section
- Remove
sudo dtc -@ -I dts -O dtb -o sun8i-h2-plus-nanopi-duo.dtb sun8i-h2-plus-nanopi-duo.dts
reboot
If you want to access devices without root do the following (you can try udev rules instead if you wish):
sudo usermod -a -G dialout username
(Use a non-root username)sudo groupadd periphery
sudo usermod -a -G periphery username
(Use a non-root username)sudo nano /etc/rc.local
chown -R root:periphery /dev/gpiochip* #/dev/gpiomem for sandbox
chmod -R ug+rw /dev/gpiochip* #/dev/gpiomem for sandbox
chown -R root:periphery /dev/i2c*
chmod -R ug+rw /dev/i2c*
chown -R root:periphery /dev/spidev*
chmod -R ug+rw /dev/spidev*
chown -R root:periphery /sys/devices/platform/leds/leds
chmod -R ug+rw /sys/devices/platform/leds/leds
- PWM udev rules
- You need kernel 4.16 or greater to use non-root access for PWM.
sudo nano /etc/udev/rules.d/99-pwm.rules
SUBSYSTEM=="pwm*", PROGRAM="/bin/sh -c '\ chown -R root:periphery /sys/class/pwm && chmod -R 770 /sys/class/pwm;\ chown -R root:periphery /sys/devices/platform/soc/*.pwm/pwm/pwmchip* && chmod -R 770 /sys/devices/platform/soc/*.pwm/pwm/pwmchip*\ '"
sudo apt install git
cd ~/
git clone --depth 1 https://github.com/sgjava/java-periphery.git
The install script assumes a clean OS install. If you would like to install on a OS with your own version of Java 17, etc. then you can look at what install.sh does and do it manually. What does the script do?
- Install build dependencies for HawtJNI
- Installs Zulu OpenJDK 17 to /usr/lib/jvm (JDK 11 is used for ARM32)
- Installs Maven to /opt
- Build HawtJNI (using my fork that works with JDK 17)
- Build Java Periphery
The Java Periphery POM uses download-maven-plugin to download c-periphery source
to
src/main/native-package/src
. The files are cached in~/.m2/repository/.cache/download-maven-plugin
, so they are not downloaded again unless they are updated. If you want to build the GPIO C code to use sysfs comment remove-DPERIPHERY_GPIO_CDEV_SUPPORT=1
from<configureArgs>
in thehawtjni-maven-plugin
section of the POM.
Make sure you run script as non-root user and without sudo.
cd ~/java-periphery/scripts
./install.sh
- Check various log files if you have issues running the demo code. Something could have gone wrong during the build/bindings generation processes.
The gcc default include paths usually do not point to the latest gpio.h header. In order to use the latest features in c-periphery you will need to use the correct gpio.h include. After the install.sh script completes:
sudo armbian-config
Software, Headers_installgrep -R -i "GPIOHANDLE_REQUEST_BIAS_DISABLE" /usr/src
mkdir -p $HOME/include/linux
cp /usr/src/linux-headers-5.9.11-sunxi/include/uapi/linux/gpio.h $HOME/include/linux/.
(use actual path)cd ~/java-periphery
mvn clean install "-Dcflags=-I$HOME/include"
- Add
-Dmaven.compiler.source=11 -Dmaven.compiler.target=11
for ARM32
I have created a generic way to achieve fast GPIO for times when performance (bit banging, software based PWM, low CPU latency, etc) is required. I have written a mapper, so you can extract the data register masks without having to do it by hand from the datasheet. Doing this totally by hand is tedious and error prone. The method I use is using a well know interface (GPIO device) to make changes and detecting register deltas. You still need to create a input file with various board specific parameters. Make sure you disable all hardware in armbian-config System, Hardware and remove console=serial from /boot/armbianEnv.txt. You want multi-function pins to act as GPIO pins.
NanoPi Duo (H2+) example:
sudo java -cp $HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT.jar:$HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT-linux32.jar com.codeferm.periphery.mmio.Gen -i duo.properties -o duo-map.properties
sudo java -cp $HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT.jar:$HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT-linux32.jar com.codeferm.periphery.mmio.Perf -i duo-map.properties -d 0 -l 203
NanoPi Neo Plus2 (H5) example:
sudo java -cp $HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT.jar:$HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT-linux64.jar com.codeferm.periphery.mmio.Gen -i neoplus2.properties -o neoplus2-map.properties
sudo java -cp $HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT.jar:$HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT-linux64.jar com.codeferm.periphery.mmio.Perf -i neoplus2-map.properties -d 1 -l 203
As you can see above the same performance test code works on a 32 bit H2+ and a 64 bit H5 CPU. This means almost all boards can be easily supported with the right input file. This is probably the only high performance GPIO code that is truly cross platform. No custom adapters or other one off code is required currently. Also, I use the same pin numbers as the GPIO device, so no goofy wiringPi or BCM pin numbering. Keep in mind that only one core is used, so the CPU will never exceed 25% on a quad core system.
If you want to map your own board you start by getting the data sheet and finding the data registers. I've written a little memory tool MemScan that will allow you to see what bits change for a range of registers using mode, data and pull operations. For example on the ODROID C2 lets look at chip 0 and line 9:
sudo java -cp $HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT.jar:$HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT-linux64.jar com.codeferm.periphery.mmio.MemScan -a 0xc8100024 -w 0x03 -d 0 -l 9
Output:
11:55:39.538 [main] INFO MemScan - Mode difference found at offset 0x00000000 before 0xa0003ef7 after 0xa0003cf7 difference 0x00000200
11:55:39.540 [main] INFO MemScan - Mode difference found at offset 0x00000004 before 0x80003ef7 after 0x80003cf7 difference 0x00000200
11:55:39.543 [main] INFO MemScan - Data difference found at offset 0x00000000 before 0xa0003cf7 after 0xa2003cf7 difference 0x02000000
11:55:39.545 [main] INFO MemScan - Data difference found at offset 0x00000004 before 0x80003cf7 after 0x80003ef7 difference 0x00000200
11:55:39.548 [main] ERROR MemScan - Device 0 line 9 Error Kernel version does not support configuring GPIO line bias
Note the bias error is due to no compiling with latest gpio.h header.
Note that most performance tests focus on writes and not CPU overhead, so it's hard to compare. Technically you will actually be doing something like bit banging to simulate a protocol, so you need extra CPU bandwidth to do that. Please note write frequency is based on square wave (rapid on/off). You can increase clock speed to improve performance on some boards. I used the OS defaults.
SBC | OS | CPU Freq | GPIOD Write KHz | MMIO Write KHz | Average CPU |
---|---|---|---|---|---|
Nano Pi Duo v1.0 | Armbian Focal | 1.0 GHz | 242 | 1880 | 25% |
Nano Pi M1 | Armbian Focal | 1.2 GHz | 320 | 2355 | 25% |
Nano Pi Neo Plus2 | Armbian Focal | 1.0 GHz | 347 | 2356 | 25% |
Odroid C2 | Armbian Focal | 1.5 GHz | 365 | 2346 | 25% |
Odroid XU4 | Armbian Focal | 2.0 GHz | 44 | 300 | 12% |
Raspberry Pi 3 | Ubuntu Focal | 1.2 GHz | 119 | 4541 | 25% |
This is based on testing on a NanoPi Duo. gpiochip0 starts at 0 and gpiochip1 start at 352. Consider the following table:
Name | Chip Name | dev | sysfs |
---|---|---|---|
DEBUG_TX(UART_TXD0)/GPIOA4 | gpiochip0 | 004 | 004 |
DEBUG_RX(UART_RXD0)/GPIOA5/PWM0 | gpiochip0 | 005 | 005 |
I2C0_SCL/GPIOA11 | gpiochip0 | 011 | 011 |
I2C0_SDA/GPIOA12 | gpiochip0 | 012 | 012 |
UART3_TX/SPI1_CS/GPIOA13 | gpiochip0 | 013 | 013 |
UART3_RX/SPI1_CLK/GPIOA14 | gpiochip0 | 014 | 014 |
UART3_RTS/SPI1_MOSI/GPIOA15 | gpiochip0 | 015 | 015 |
UART3_CTS/SPI1_MISO/GPIOA16 | gpiochip0 | 016 | 016 |
UART1_TX/GPIOG6 | gpiochip0 | 198 | 198 |
UART1_RX/GPIOG7 | gpiochip0 | 199 | 199 |
GPIOG11 | gpiochip0 | 203 | 203 |
ON BOARD BUTTON | gpiochip1 | 003 | 355 |
GPIOL11/IR-RX | gpiochip1 | 011 | 363 |
So basically you just need to know the starting number for each chip and realize GPIO character devices always starts at 0 and calculate the offset. Thus gpiochip1 starts at 352 and the on board button is at 355, so 355 - 352 = 3 for GPIO character device.
java -cp $HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT.jar:$HOME/java-periphery/target/java-periphery-1.0.0-SNAPSHOT-linux32.jar com.codeferm.periphery.demo.LedBlink --help
Note that the native library jar has a suffix such as linux32, so depending on your target platform it could be different. To see a list of demos browse code. Just pass in --help to get list of command line arguments.
After bulding Java Periphery simpily add the following artifact:
<groupId>com.codeferm</groupId>
<artifactId>java-periphery</artifactId>
<version>1.0.0-SNAPSHOT</version>
Azul Mission Control allows
you to profile your applications.
Download
zmc and launch on your desktop. To profile your Java Periphery application use:
java -XX:+FlightRecorder -Djava.rmi.server.hostname=your_ip -Dcom.sun.management.jmxremote=true -Dcom.sun.management.jmxremote.port=8888 -Dcom.sun.management.jmxremote.authenticate=false -Dcom.sun.management.jmxremote.ssl=false -cp java-periphery-1.0.0-SNAPSHOT.jar:java-periphery-1.0.0-SNAPSHOT-linux32.jar com.codeferm.periphery.demo.GpioPerf