OpenCores

OR1K:FutureWork

From OR1K

Contents

Future work

We have a list of projects, some of which are currently in progress, and some of which we want others to start work on.

OpenRISC is a very broad project and there is likely something to do for any discipline. That said, below are some projects we have identified and would like help solving.

Why and How to Volunteer

Taking responsibility for maintaining a package is a great way to gain experience of practical program management. If you are a student, or early in your career, it adds important credibility to your resume. And since it is all open source, your expertise can be publicly verified by any potential employer.

To take part, post a message on the mailing lists, or email the Release Manager, Julius Baxter or come and talk to us on IRC.

Project Ideas

OpenRISC System Creator

Brief explanation: OpenRISC project main problem is difficulty in creating a full system design (CPU, switch, peripherals) and visualizing them.

Task is to create a tool to solve this problem.

Several approaches are possible:

- You'll design Mercurial/Git/Npm like tool for OpenRISC system design. By executing 'openrisc init' empty system will get created, by doing 'openrisc create switch' switch will be placed in the system etc.. Example:

 openrisc init
 openrisc create switch sw0
 openrisc create cpu -t or64k cpu0
 openrisc connect sw0.m0 cpu0.s_inst0
 openrisc connect sw0.m1 cpu0.s_data0
 
 etc

- you'll pick and define DSL (Domain Specific Language) for system description

- you'll use YAML/JSON as system description.

Expected results: Make it easier to create OpenRISC systems.

Knowledge Prerequisite: Any scripting language. System design and computer architecture knowledge

Mentor: Wojciech Koszek <wkoszek@freebsd.czest.pl>

OpenRISC Web-based system for system visualization

Brief explanation: OpenRISC needs easy-to-use system for showing how system looks like in a graphical way. Think of connecting boxes together with each other on the schematics. There's no way to do this now in OpenRISC, since there's no software.

Expected results: Goal is to:

- understand HW/SW elements in modern embedded systems.

- figure out in JQuery/Javascript/HTML5 how to write an application with drag&drop capabilities to place CPU (1 box), Switch (1 box) and some other blocks, and wire them all together. Then store this config in memory somehow (JSON) and preserve it on the client side (LocalStorage on HTML5) and submit it to the server.

- server should at least receive it

Knowledge Prerequisite: HTML5, Jquery, Graph/chart javascript library (d3?)

Mentor: Wojciech Koszek <wkoszek@freebsd.czest.pl>


OpenRISC package manager

Brief explanation:

If you look how most software people install packages, you'll see how easy it is.

Hardware people don't have this notion of installation. Installing Wishbone/OpenRISC modules is a whole bunch of pain and nostalgia.

The idea is to solve the problem with packages, so that constructing a system would be easily possible by just typing couple of commands, which would fetch RTL sources and prepare the blocks together for you, so you can easily simulate them. RTL poses new challenges:

- upon fetching, stuff should be simulated and verified to meet the specs

- it should be matched like shared libraries, since some versions of a switch may not work with the CPU

- should be tagged: WishBone stuff can't work with AXI interconnect stuff. It'd be nice to do:

openriscpkg install openrisc
openriscpkg install wb_mux
openriscpkg install wb_uart
openriscpkg sim wb_mux
openriscpkg package wb_mux

Knowledge Prerequisite:

Any scripting language.

System design and computer architecture knowledge

Understanding of Verilog

Willingness to learn of Wishbone architecture

OpenRISC FreeBSD OS port

Brief explanation:

Linux is the only operating system running on OpenRISC. Proposal is to offer alternative software stack for OpenRISC users by porting FreeBSD OS to OpenRISC.

Task would involve:

  • running or1ksim and getting basic Hello world running
  • installing FreeBSD in VM
  • porting FreeBSD to 'accept' new OpenRISC toolchain on FreeBSD
  • making OpenRISC toolchain work
  • porting FreeBSD kernel to OpenRISC

MVP for this product would be minimal kernel that can boot inside of OR1KSIM and print something on UART. We'd expect kernel should reach driver execution stage.


Knowledge Prerequisite:

Advanced ANSI C knowledge

OS design and architecture.

Computer architecture

Compiler understanding

Linux kernel VDSO support

Brief explanation: Syscalls are expensive. VDSO helps speeding this up.

This is probably a small project. For GSoC it might be suitable to combine with other projects.

Since or1k does not contain any atomic operations these are implemented as syscalls. With VDSO these should perform much faster.

Expected results: Faster syscalls, better multi-threaded performance.

Knowledge Prerequisite: C, How to build a Linux kernel

Mentor: Christian 'bluecmd' Svensson

Extend language support for OpenRISC

Brief explanation: OpenRISCs GCC disables languages such as Go, Java, D etc. This project would be to:

  • Find out what kind of work is needed to make these languages available (libffi for one).
  • Prioritize with the community.
  • Start implementation of dependencies and the language support.
  • Set up regression testing.
  • Document usage and caveats.

Expected results: Increase programming languages available on OpenRISC.

Knowledge Prerequisite: C, Linux internals

Mentor: Christian "bluecmd" Svensson

GDB for Linux

Brief explanation: GDB running in userspace for Linux is in a weird state.

As far as bluecmd knows GDB should work on bare metal, but not for userspace Linux. It would be good to clean this up, make it work and upstream it.

The project would be something like this:

  • Find out what works and does not work with the GDB as it is today
  • Fix GDB
  • Set up automatable regression testing
  • Document how to use
  • Upstream the patches

Expected results: GDB running on OpenRISC under Linux

Knowledge Prerequisite: C, Linux internals

Mentor: ? (maybe bluecmd)

Port libffi/gobject-introspection/with friends

Brief explanation: This project would be to take a look on how to port libraries that needs to know stuff about the machine it's running under. Libffi and gobject-introspections are two examples of this.

libffi provides interfaces to C functions for programs like Python. Not having it is a bit of a pain.

GObject-introspection is a C wrapper for GNOME objects. It is rumored that this is a bit tricky to port, but that is only rumor.


This is a smaller project, possibly part of the Extend language support project. Some work has been started by Sebastian Macke: https://github.com/s-macke/libffi

Expected results: Wider support and more software available

Knowledge Prerequisite: C

Mentor: Christian "bluecmd" Svensson

Updated to support OR1K 1.0 versioning system

Brief explanation: The various implementations (mainly OR1200, or1ksim) need to be updated to support the versioning system which is new to OR1K version 1.0.

Specifically, VR2 and ISRs should be implemented and containing meaningful values.

Any software bundled with these implementations, and the libraries such as newlib, should be updated, too (at the very least the spr-defs.h files should be updated - patches for this do exist but need to be committed.)

Once models look to be working OK with the new versioning scheme, things like the Linux kernel port should be updated to support the new system.

Expected results: ?

Knowledge Prerequisite: ?

Mentor: ?

Future toolchain development

Brief explanation: At a technical level, future development of the GNU toolchain is planned to include

  • improved optimization through inclusion of accurate process scheduling models.
  • support for profiling.
  • more efficient thread handling.

There are some structural things we would also really like to have in place:

  • A nice tutorial, showing off the installation and use of the new toolchain.
  • Maintainers for various platforms (Cygwin, Mac OS X)
  • Maintainers for packaging toolchain for Debian (apt) and Fedora (yum) package
  • managers.

Feel free to post any suggestions or offers to assist the project to the mailing lists or talk to us on IRC.

Expected results: Better performance for software running on OR1K

Knowledge Prerequisite: C, Linux internals

Mentor: ?

Probably one of:

Additional board implementations

Brief explanation: Add more supported platforms.

This is a never ending work - we always need more platforms!

Expected results: More platforms to run OpenRISC on

Knowledge Prerequisite: Hardware, HDL, FPGA

Mentor: ?

People doing this kind of work:


Upstreaming toolchains

Brief explanation: Our toolchains are looking better and better. A bigger and bigger issue is having to merge with upstream. It would be very nice if we could come up with a plan and begin merging the work with upstream.

A big part of this is to figure out what code is used and written by whom. We have revision histories spanning back to when CVS was used, over to SVN and finally Git.

Expected results:

  • Easy usage for people, no need to pull patches from our repositories.
  • Less administrative burden for OpenRISC.
  • Needed for stuff like Debian.

Knowledge Prerequisite: Revision Control Systems

Mentor: ?

Match specification to implementation

TODO: FORMAT TO TEMPLATE

The objective is to bring OR1200 specification document into line with what is actually in the RTL. A few sections still need going over (debug interface, for one.)

This task is being handled by Yann Vernier [mailto:yann@orsoc.se yann@orsoc.se].

Add multiple associativity to cache and MMU

TODO: FORMAT TO TEMPLATE

This is being led by Michael Unneback (michael@orsoc.se).


Wishlist

TODO: FORMAT TO TEMPLATE

  • Improve verification of OR1K implementations
    • Unify test code of or1ksim and ORPSoC
    • Extend test code base
    • Develop automated scripts to run GNU GCC regression suite against Verilated ORPSoC design and or1ksim
  • Identify architectural improvements
    • Possible ISA extensions
    • Instruction class restructure
  • Further development of the toolchain

Project Template

Brief explanation:

Expected results:

Knowledge Prerequisite:

Mentor:

When adding an idea, please try to include the following data:

  • if the application is not widely known, a description of what it does and where its code lives
  • a brief explanation
  • the expected results
  • pre-requisites for working on your project
  • if applicable, links to more information or discussions
  • your name and email address for contact (if you're willing to be a mentor)

If you are not a developer but have a good idea for a proposal, get in contact with relevant developers first.

© copyright 1999-2017 OpenCores.org, equivalent to ORSoC AB, all rights reserved. OpenCores®, registered trademark.