Skip to content

Latest commit

 

History

History
114 lines (85 loc) · 6.19 KB

README.md

File metadata and controls

114 lines (85 loc) · 6.19 KB

'docker' workflow Status

This repository contains containerized open and free development tools for Dynamic Binary Hardware Injection (DBHI). All the images are available at docker.io/aptman/dbhi. Images provided by docker-library/official-images are used as a base, and three host platforms are supported: amd64, arm64v8 and arm32v7. Manifests are also available, in order to allow platform-agnostic development.

  • Any host:

    • aptman/dbhi:bionic*: GHDL (with LLVM backend and --default-pic), and VUnit (Python 3). This is the base image.
    • aptman/dbhi:bionic-dr*: DynamoRIO, based on base image.
    • aptman/dbhi:bionic-cosim*: base image plus GTKWave, Flask, Pillow and numpy.
    • aptman/dbhi:bionic-octave*: base image plus Octave.
  • arm|arm64 only:

    • aptman/dbhi:bionic-mambo-*: MAMBO, based on base image.
  • amd64 only:

NOTE: binaries/artifacts built in aptman/dbhi:bionic* images can be executed on v2.3, v2.4 or v2.5 SDCard images provided at Xilinx/PYNQ, since those are based on Ubuntu 18.04 (bionic). Releases are available for PYNQ, ZCU104 and ZCU111 boards.

Usage

Some images include tools, such as GTKWave or Octave, that provide GUI interfaces. These interfaces require an X server, which is expected to be executed outside of the container. However, docker does not provide built-in options to automatically share the display from the host. Furthermore, non-linux environments do not provide an X server by default. Fortunately, x11docker and runx allow to easily set up custom X servers on either GNU/Linux or Windows.

On GNU/Linux:

x11docker --hostdisplay -i aptman/dbhi:bionic-cosim bash

On Windows, with MSYS2 and VcxSrv:

x11docker --runx --no-auth -i aptman/dbhi:bionic-cosim bash

On Windows, with WSL or Cygwin, and Cygwin/X:

x11docker -i aptman/dbhi:bionic-cosim bash

Apart from these basic options, x11docker provides many features focused on security; and remote access is supported. See JOSS 10.21105/joss.01349.

Continuous integration

Images for the three target platforms are built in a GitHub Actions workflow. dbhi/qus is used to enable execution of images for foreign architectures (arm64v8|arm32v7 on amd64). See push.yml and run.sh for further details.

Building docker images for/on ARM targets

  • SiFive does not provide prebuilt packages for ARM hosts: sifive.com/boards. That's why spinalhdl images for Arm platforms are not provided, although SpinalHDL can be successfully executed.

  • Building image dr hangs the system on RPi3B. We should investigate why. However, DynamoRIO can be succesfully built natively.

  • Building DynamoRIO on an arm32v7 container on a 64 bit host requires to set -DCMAKE_SYSTEM_NAME=Linux -DCMAKE_SYSTEM_PROCESSOR=armv7, as explained at DynamoRIO/dynamorio#3381.

TODO Update either the Dockerfile or the build script according to DynamoRIO/dynamorio#3381. Setting CMAKE_SYSTEM_NAME and CMAKE_SYSTEM_PROCESSOR seems not to be required if the ARM docker containers are executed on x86_64 through qemu-user-static binaries (either arm or AArch64).


Compilation of either GHDL or DynamoRIO fails on AArch64 with kata-runtime (see katacontainers.io). Multiple errors such as the following are shown:

  • Clock skew detected. Your build may be incomplete.
  • Warning: File <file> has modification time 0.008 s in the future
  • cc1: out of memory allocating 65537 bytes after a total of 1466368 bytes

This has been reported at kata-containers/runtime#1293.

We tried to work around some of these issues with --cpuset-cpus, so that make was executed on a single core. However, DynamoRIO requires more than 3GB of RAM, and runtime option -m would fail with kata-runtime (see kata-containers/runtime#1245).

The solution was to set runc as the default runtime. This was done through /etc/systemd/system/docker.service.d/kata-containers.conf and/or /etc/docker/daemon.json. See Install Docker for Kata Containers on Ubuntu.