PYNQ SD Card image

This page will explain how SD card images can be built for PYNQ embedded platforms (Zynq, Zynq Ultrascale+, Zynq RFSoC).

Note: the PYNQ images for supported boards are provided as precompiled downloadable SD card images and do not need rebuilt. The SD card build flow is only required to modify SD cards’ contents or target a new board.

Specifically, The SD card build flow will create the BOOT.bin, the u-boot bootloader, the Linux Device tree blob, the Linux kernel and the PYNQ-Linux root filesystem.

The source files for the PYNQ image flow build can be found here:

<PYNQ repository>/sdbuild

More details on configuring the root filesystem can be found in the README file in the folder above.

Prepare the Building Environment

It is recommended to use a Ubuntu OS to build the image. The currently supported Ubuntu OS are listed below:

Supported OS

Code name

Ubuntu 22.04

Jammy

Use Docker to prepare the build environment

Starting with the PYNQ v3.1 release, Docker is the recommended way to build PYNQ images. Docker simplifies setup by managing dependencies and environment configuration within an isolated container.

If you do not have a supported Ubuntu machine, you can use Docker to create an isolated build environment on your host OS using the following steps:

  1. Install Docker on your host OS by following the official Docker installation instructions.

  2. Install the required AMD tools on your host OS (not inside Docker):

    • Vivado, Vitis, and PetaLinux, version 2024.1

    • Ensure your host OS is supported by the AMD tools (see UG973)

    Note

    AMD tools must be installed on the host system, not inside the Docker container.

  3. Clone the PYNQ repository and in a bash shell, build the Docker image:

    git clone --recursive https://github.com/Xilinx/PYNQ.git PYNQ
cd PYNQ/sdbuild
docker build \
  --build-arg USERNAME=$(whoami) \
  --build-arg USER_UID=$(id -u) \
  --build-arg USER_GID=$(id -g) \
  -t pynqdock:latest .

    If you are in a csh-like shell, switch to Bash by typing: bash.

    The --build-arg values ensure that files created inside the container will be owned by your user on the host system, avoiding permission issues.

  4. Run the Docker container, mounting your AMD tools and PYNQ repository:

    docker run \
  --init \
  --rm \
  -it \
  -v /tools/Xilinx:/tools/Xilinx:ro \
  -v /home/user/petalinux:/home/user/petalinux:ro \
  -v $(pwd):/workspace \
  --name pynq-sdbuild-env \
  --privileged \
  pynqdock:latest \
  /bin/bash

    Replace /tools/Xilinx and /home/user/petalinux with the actual paths to your Xilinx installations on the host system. The :ro option mounts tool directories read-only, and --privileged is required for parts of the build process.

  5. Inside the container, set up the tool environment:

    source /tools/Xilinx/Vivado/2024.1/settings64.sh
source /home/user/petalinux/settings.sh

    Adjust the paths to match your actual Xilinx tool installation paths.

  6. You are now ready to build PYNQ images inside the Docker container. Navigate to the sdbuild directory and follow the building instructions in the next section.

    Note

    Rebuilding the PYNQ source distribution (SDIST) currently does not work inside Docker due to limitations in the Vitis tools. If you need to rebuild the SDIST, consider using a virtual machine with a supported Ubuntu distribution instead.

Use an existing Ubuntu OS

If you’re not able to use Docker, you can still build PYNQ images on a supported Ubuntu OS.

If you already have a Ubuntu OS, and it is listed in the beginning of this section, you can simply do the following:

  1. Install dependencies using the following script. This is necessary if you are not using our vagrant file to prepare the environment.

    <PYNQ repository>/sdbuild/scripts/setup_host.sh

  2. Install correct version of the Xilinx tools, including PetaLinux, Vivado, and Vitis. See the table below for the correct version of each release.

    Starting from image v2.5, SDx is no longer needed.

    Release version

    Xilinx Tool Version

    v1.4

    2015.4

    v2.0

    2016.1

    v2.1

    2017.4

    v2.2

    2017.4

    v2.3

    2018.2

    v2.4

    2018.3

    v2.5

    2019.1

    v2.6

    2020.1

    v2.7

    2020.2

    v3.0

    2022.1

    v3.1

    2024.1

Building the Image From Source

Once you have the build environment ready, you can build an SD card image following the steps below. You don’t have to rerun the setup_host.sh.

  1. Source the appropriate settings for PetaLinux and Vitis. Suppose you are using Xilinx 2024.1 tools:

    source <path-to-vitis>/Vitis/2024.1/settings64.sh
source <path-to-petalinux>/petalinux-2024.1-final/settings.sh

  2. Depending on the overlays being rebuilt, make sure you have the appropriate Vivado licenses to build for your target board, especially the HDMI IP for the ZCU104 or the CMAC IP for the RFSoC4x2.

  3. Collect a prebuilt board-agnostic root filesystem tarball and a prebuilt PYNQ source distribution. Starting in PYNQ v3.0, by default the SD card build flow expects a prebuilt root filesystem and a PYNQ source distribution to speedup and simplify user rebuilds of SD card images. These binaries can be found at the PYNQ boards page and copied into the sdbuild prebuilt folder

    # For rebuilding all SD cards, both arm and aarch64 root filesystems
# may be required depending on boards being targetted.
cp pynq_rootfs.<arm|aarch64>.tar.gz <PYNQ repository>/sdbuild/prebuilt/pynq_rootfs.<arm|aarch64>.tar.gz
cp pynq-<version>.tar.gz            <PYNQ repository>/sdbuild/prebuilt/pynq_sdist.tar.gz

  4. Navigate to the following directory and run make

    cd <PYNQ repository>/sdbuild/
make

The build flow can take several hours and will rebuild SD cards for the Pynq-Z1, Pynq-Z2 and ZCU104 platforms.

Rebuilding the prebuilt board-agnostic image

In order to simplify and speed-up the image building process, users should re-use the prebuilt board-agnostic image appropriate to the architecture - arm for Zynq-7000 and aarch64 for Zynq UltraScale+, downloadable at the boards page of our website. This will allow you to completely skip the board-agnostic stage.

You can force a root filesystem build by setting the REBUILD_PYNQ_ROOTFS variable when invoking make:

cd <PYNQ repository>/sdbuild/
make REBUILD_PYNQ_ROOTFS=True BOARDS=<board>

Rebuilding the PYNQ source distribution tarball

To avoid rebuilding the PYNQ source distribution package, and consequently bypass the need to build bitstreams for the PYNQ-Z1, PYNQ-Z2 and the ZCU104, a prebuilt PYNQ sdist tarball should be reused as described in steps listed above.

You can force a PYNQ source distribution rebuild by setting the REBUILD_PYNQ_SDIST variable when invoking make

cd <PYNQ repository>/sdbuild/
make REBUILD_PYNQ_SDIST=True

Please also refer to the sdbuild readme on our GitHub repository for more info regarding the image-build flow.

Unmount images before building again

Sometimes the SD image building process can error out, leaving mounted images in your host OS. You need to unmount these images before trying the make process again. Starting from image v2.6, users can do the following to unmount the images.

cd <PYNQ repository>/sdbuild/
make delete

The above command not only unmounts all the images, but also deletes the failed images. This makes sure the users do not use the failed images when continuing the SD build process.

To unmount images but not delete them, use the following command instead.

cd <PYNQ repository>/sdbuild/
make unmount

If you want to ignore all the previous staged or cached SD build artifacts and start from scratch again, you can use the following command. This will unmount and delete the failed images, and remove all the previously built images at different stages.

cd <PYNQ repository>/sdbuild/
make clean

Retargeting to a Different Board

Additional boards are supported through external board repositories. A board repository consists of a directory for each board consisting of a spec file and any other files. The board repository is treated the same way as the <PYNQ repository>/boards directory.

Elements of the specification file

The specification file should be name <BOARD>.spec where BOARD is the name of the board directory. A minimal spec file contains the following information

ARCH_${BOARD} := arm
BSP_${BOARD} := mybsp.bsp
BITSTREAM_${BOARD} := mybitstream.bsp
FPGA_MANAGER_${BOARD} := 1

where ${BOARD} is also the name of the board. The ARCH should be arm for Zynq-7000 or aarch64 for Zynq UltraScale+. If no bitstream is provided then the one included in the BSP will be used by default. All paths in this file should be relative to the board directory.

To customise the BSP a petalinux_bsp folder can be included in the board directory the contents of which will be added to the provided BSP before the project is created. See the ZCU104 for an example of this in action. This is designed to allow for additional drivers, kernel or boot-file patches and device tree configuration that are helpful to support elements of PYNQ to be added to a pre-existing BSP.

If a suitable PetaLinux BSP is unavailable for the board then BSP_${BOARD} can be left blank; in this case, users have two options:

  1. Place a <design_name>.xsa file in the petalinux_bsp/hardware_project folder. As part of the build flow, a new BSP will be created from this XSA file.

  2. Place a makefile along with tcl files which can generate the hardware design in the petalinux_bsp/hardware_project folder. As part of the build flow, the hardware design along with the XSA file will be generated, then a new BSP will be created from this XSA file.

Starting from image v2.6, we allow users to disable FPGA manager by setting FPGA_MANAGER_${BOARD} to 0. This may have many use cases. For example, users may want the bitstream to be downloaded at boot to enable some board components as early as possible. Another use case is that users want to enable interrupt for XRT. The side effect of this, is that users may not be able to reload a bitstream after boot.

If FPGA_MANAGER_${BOARD} is set to 1 or FPGA_MANAGER_${BOARD} is not defined at all, FPGA manager will be enabled. In this case, the bitstream will be downloaded later in user applications; and users can only use XRT in polling mode. This is the default behavior of PYNQ since we want users to be able to download any bitstream after boot.

Board-specific packages

A packages directory can be included in board directory with the same layout as the <PYNQ repository>/sdbuild/packages directory. Each subdirectory is a package that can optionally be installed as part of image creation. See <PYNQ repository>/sdbuild/packages/README.md for a description of the format of a PYNQ sdbuild package.

To add a package to the image you must also define a STAGE4_PACKAGE_${BOARD} variable in your spec file. These can either packages in the standard sdbuild library or ones contained within the board package. It is often useful to add the pynq package to this list which will ensure that a customised PYNQ installation is included in your final image.

Leveraging boot.py to modify SD card boot behavior

Starting from the v2.6.0 release, PYNQ SD card images include a boot.py file in the boot partition that runs automatically after the board has been booted. Whatever is inside this file runs during boot and can be modified any time for a custom next-boot behavior (e.g. changing the host name, connecting the board to WiFi, etc.).

This file can be accessed using a SD Card reader on your host machine or from a running PYNQ board - if you are live on the board inside Linux, the file is located in the /boot folder. Note that /boot is the boot partition of the board and no other files should be modified.

If you see some existing code running inside the boot.py file, it probably came from a PYNQ sdbuild package that modified that file. To see an example of an sdbuild package writing the boot.py file see the ZCU104’s boot_leds package which simply flashes the boards LEDs to signify Linux has booted on the board.

Using the PYNQ package

The pynq package will treat your board directory the same as any of the officially supported boards. This means, in particular, that:

  1. A notebooks folder, if it exists, will be copied into the jupyter_notebooks folder in the image. Notebooks here will overwrite any of the default ones.

  2. Any directory containing a bitstream will be treated as an overlay and copied into the overlays folder of the PYNQ installation. Any notebooks will likewise by installed in an overlay-specific subdirectory.

Building from a board repository

To build from a third-party board repository, pass the BOARDDIR variable to the sdbuild makefile.

cd <PYNQ repository>/sdbuild/
make BOARDDIR=${BOARD_REPO}

The board repo should be provided as an absolute path. The BOARDDIR variable can be combined with the BOARDS variable if the repository contains multiple boards and only a subset should be built.