OpenPiton Research Platform

OpenPiton is the world's first open source, general purpose, multithreaded manycore processor. It is a tiled manycore framework scalable from one to 1/2 billion cores. It is a 64-bit architecture using SPARC v9 ISA with a distributed directory-based cache coherence protocol across on-chip networks. It is highly configurable in both core and uncore components. OpenPiton has been verified in both ASIC and multiple Xilinx FPGA prototypes running full-stack Debian linux. We have released both the Verilog RTL code as well as synthesis and back-end flow. We believe OpenPiton is a great framework for researchers in computer architecture, OS, compilers, EDA, security and more.

OpenPiton has been published in ASPLOS 2016: Jonathan Balkind, Michael McKeown, Yaosheng Fu, Tri Nguyen, Yanqi Zhou, Alexey Lavrov, Mohammad Shahrad, Adi Fuchs, Samuel Payne, Xiaohua Liang, Matthew Matl, and David Wentzlaff. "OpenPiton: An Open Source Manycore Research Framework." In Proceedings of the 21st International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS '16), April 2016.

Find out more

• More information about OpenPiton is available at www.openpiton.org
• Follow us on Twitter!
• Get help from others by joining our Google Group
• Keep up-to-date with the latest releases at the OpenPiton Blog

If you use OpenPiton in your research please reference our ASPLOS 2016 paper mentioned above and send us a citation of your work.

Documentation

There are several detailed pieces of documentation about OpenPiton in the docs folder listed below:

• OpenPiton Simulation Manual
• OpenPiton Microarchitecture Specification
• OpenPiton FPGA Prototype Manual
• OpenPiton Synthesis and Back-end Manual

We also host GitHub repositories for other parts of the project, including:

• Piton Linux Kernel
• Piton Hypervisor

Environment Setup

• The PITON_ROOT environment variable should point to the root of the OpenPiton repository

• The Synopsys environment for simulation should be setup separately by the user. Besides adding correct paths to your PATH and LD_LIBRARY_PATH (usually accomplished by a script provided by Synopsys), the OpenPiton tools specifically reference the VCS_HOME environment variable which should point to the root of the Synopsys VCS installation.

• Run source $PITON_ROOT/piton/piton_settings.bash to setup the environment

  • A CShell version of this script is provided, but OpenPiton has not been tested for and currently does not support CShell

• Top level directory structure:

  • piton/
    • All OpenPiton design and verification files
  • docs/
    • OpenPiton documentation
  • build/
    • Working directory for simulation and simulation models

Notes on Environment and Dependencies

• Depending on your system setup, Synopsys tools may require the -full64 flag. This can easily be accomplished by adding a bash function as shown in the following example for VCS (also required for URG):

  function vcs() { command vcs -full64 "$@"; }; export -f vcs

• On many systems, an error with goldfinger, or other errors not described below, may indicate that you should run the mktools command once to rebuild a number of the tools before continuing. If you see issues later with building or running simulations, try running mktools if you have not already.

• In some cases, you may need to recompile the PLI libraries we provide. This is done using mkplilib with the argument for the simulator you want to rebuild for. You may need to run mkplilib clean first, then depending on which simulator, you can build with: mkplilib vcs, mkplilib ncverilog, mkplilib icarus, or mkplilib modelsim.

• If you see an error with bw_cpp then you may need to install gcc/g++ (to get cpp), or csh (csh on ubuntu, tcsh on centos)

• If you see an error with goldfinger or g_as then you may need to install 32-bit glibc (libc6-i386 on ubuntu, glibc.i686 on centos)

• If you see an error with goldfinger or m4 then you may need to install libelf (libelf-dev on ubuntu, elfutils-libelf-devel on centos)

• You also need the Perl Bit::Vector package installed on your machine (libbit-vector-perl on ubuntu, perl-Bit-Vector.x86_64 on centos, also installable via CPAN)

==========================

Building a simulation model

1. cd $PITON_ROOT/build
2. sims -sys=manycore -x_tiles=1 -y_tiles=1 -vcs_build builds a single tile OpenPiton simulation model.
3. A directory for the simulation model will be created in $PITON_ROOT/build and the simulation model can now be used to run tests. For more details on building simulation models, please refer to the OpenPiton documentation.

Note: if you would like to decrease the testbench monitor output to a minimum, append -config_rtl=MINIMAL_MONITORING to your build command in step 2. above.

==========================

Running a simulation

1. cd $PITON_ROOT/build
2. sims -sys=manycore -x_tiles=1 -y_tiles=1 -vcs_run princeton-test-test.s runs a simple array summation test given the simulation model is already built.
3. The simulation will run and generate many log files and simulation output to stdout. For more details on running a simulation, provided tests/simulations in the test suite, and understanding the simulation log files and output, please refer to the OpenPiton documentation.

==========================

Running a regression

A regression is a set of simulations/tests which run on the same simulation model.

1. cd $PITON_ROOT/build
2. sims -sim_type=vcs -group=tile1_mini runs the simulations in the tile1_mini regression group.
3. The simuation model will be built and all simulations will be run sequentially. In addition to the simulation model directory, a directory will be created in the form <date>_<id> which contains the simulation results.
4. cd <date>_<id>
5. regreport $PWD > report.log will process the results from each of the regressions and place the aggregated results in the file report.log. For more details on running a regression, the available regression groups, understanding the regression output, and specifying a new regression group, please refer to the OpenPiton documentation.

==========================

Running a continuous integration bundle

Continuous integration bundles are sets of simulations, regression groups, and/or unit tests. The simulations within a bundle are not required to have the same simulation model. The continuous integration tool requires a job queue manager (e.g. SLURM, PBS, etc.) to be present on the system in order parallelize simulations.

1. cd $PITON_ROOT/build
2. contint --bundle=git_push runs the git_push continuous integration bundle which we ran on every commit when developing Piton. It contains a regression group, some assembly tests, and some unit tests.
3. The simulation models will be built and all simulation jobs will be submitted
4. After all simulation jobs complete, the results will be aggregated and printed to the screen. The individual simulation results will be saved in a new directory in the form contint_<bundle name>_<date>_<id> and can be reprocessed later to view the aggregated results again.
5. The exit code of the command in Step 2 indicates whether all tests passed (zero exit code) or at least one failed (non-zero exit code).
6. For more details on running continuous integration bundles, the available bundles, understanding the output, reprocessing completed bundles, and creating new bundles, please refer to the OpenPiton documentation.

==========================

Support for the Ariane RV64IMAC Core

This version of OpenPiton supports the 64bit Ariane RISC-V processor from ETH Zurich. To this end, Ariane has been equipped with a different L1 cache subsystem that follows a write-through protocol and that has support for cache invalidations and atomics. This L1 cache system is designed to connect directly to the L1.5 cache provided by OpenPiton's P-Mesh.

Check out the sections below to see how to run the RISC-V tests or simple bare-metal C programs in simulation.

Environment Setup

In addition to the OpenPiton setup described above, you have to adapt the paths in the ariane_setup.sh script to match with your installation (we support Questasim, VCS and Verilator at the moment). Source this script from the OpenPiton root folder and build the RISC-V tools with ariane_build_tools.sh if you are running this for the first time:

1. cd $PITON_ROOT/
2. source piton/ariane_setup.sh
3. piton/ariane_build_tools.sh

Step 3. will then download and compile the RISC-V toolchain, the assembly tests and Verilator.

Note that the address map is different from the standard OpenPiton configuration. DRAM is mapped to 0x8000_0000, hence the assembly tests and C programs are linked with this offset. Have a look at piton/design/xilinx/genesys2/devices_ariane.xml for a complete address mapping overview.

Also note that we use a slightly adapted version of syscalls.c. Instead of using the RISC-V FESVR, we use the OpenPiton testbench monitors to observe whether a test has passed or not. Hence we added the corresponding pass/fail traps to the exit function in syscalls.c.

For simulation Questasim 10.6b, VCS 2017.03 or Verilator 4.014 is needed (older versions may work, but are untested).

You will need Vivado 2018.2 or newer to build an FPGA bitstream with Ariane.

Running RISC-V Tests and Benchmarks

The RISC-V benchmarks are precompiled in the tool setup step mentioned above. You can run individual benchmarks by first building the simulation model with

1. cd $PITON_ROOT/build
2. sims -sys=manycore -x_tiles=1 -y_tiles=1 -msm_build -ariane

Then, invoke a specific riscv test with the -precompiled switch as follows

sims -sys=manycore -msm_run -x_tiles=1 -y_tiles=1 rv64ui-p-addi.S -ariane -precompiled

This will look for the precompiled ISA test binary named rv64ui-p-addi in the RISC-V tests folder $ARIANE_ROOT/tmp/riscv-tests/build/isa and run it.

In order to run a RISC-V benchmark, do

sims -sys=manycore -msm_run -x_tiles=1 -y_tiles=1 dhrystone.riscv -ariane -precompiled

The printf output will be directed to fake_uart.log in this case (in the build folder).

Note: if you see the Warning: [l15_adapter] return type 004 is not (yet) supported by l15 adapter. warning in the simulation output, do not worry. This is only generated since Ariane does currently not support OpenPiton's packet-based interrupt packets arriving over the memory interface.

Running Custom Programs

You can also run test programs written in C. The following example program just prints 32 times "hello_world" to the fake UART (see fake_uart.log file).

1. cd $PITON_ROOT/build
2. sims -sys=manycore -x_tiles=1 -y_tiles=1 -msm_build -ariane
3. sims -sys=manycore -msm_run -x_tiles=1 -y_tiles=1 hello_world.c -ariane -rtl_timeout 10000000

And a simple hello world program running on multiple tiles can run as follows:

1. cd $PITON_ROOT/build
2. sims -sys=manycore -x_tiles=4 -y_tiles=4 -msm_build -ariane
3. sims -sys=manycore -msm_run -x_tiles=4 -y_tiles=4 hello_world_many.c -ariane -finish_mask 0x1111111111111111 -rtl_timeout 1000000

In the example above, we have a 4x4 Ariane tile configuration, where each core just prints its own hart ID (hardware thread ID) to the fake UART. Synchronization among the harts is achieved using an atomic ADD operation.

Note that we have to adjust the finish mask in this case, since we expect all 16 cores to hit the pass/fail trap.

Regressions

The RISC-V ISA tests, benchmarks and some additonal simple example programs have been added to the regression suite of OpenPiton, and can be invoked as described below.

• RISC-V ISA tests are grouped into the following four batches, where the last two are the regressions for atomic memory operations (AMOs):

sims -group=ariane_tile1_asm_tests_p -sim_type=msm

sims -group=ariane_tile1_asm_tests_v -sim_type=msm

sims -group=ariane_tile1_amo_tests_p -sim_type=msm

sims -group=ariane_tile1_amo_tests_v -sim_type=msm

• RISC-V benchmarks can be run with:

sims -group=ariane_tile1_benchmarks -sim_type=msm

• Simple hello world programs and AMO tests for 1 tile can be invoked with

sims -group=ariane_tile1_simple -sim_type=msm

• And a multicore "hello world" example running on 16 tiles can be run with

sims -group=ariane_tile16_simple -sim_type=msm

If you would like to get an overview of the exit status of a regression batch, step into the regression subfolder and call regreport . -summary.

FPGA Mapping on Genesys2 Board

The bitfile for a 1x1 tile Ariane configuration for the Genesys2 board can be built using the follong command:

protosyn -b genesys2 -d system --core=ariane --uart-dmw ddr

It is recommended to use Vivado 2018.2 or later since earlier versions might not produce a working bitstream.

Once you have loaded the bitstream onto the FPGA using the Vivado Hardware Manager or a USB drive plugged into the Genesys2, you first need to connect the UART/USB port of the Genesys2 board to your computer and flip switch 7 on the board as described in the OpenPiton FPGA Prototype Manual. Then you can use pitonstream to run a list of tests on the FPGA:

pitonstream -b genesys2 -d system -f ./tests.txt --core=ariane

The tests that you would like to run need to be specified in the test.txt file, one test per line (e.g. hello_world.c).

You can also run the precompiled RISCV benchmarks by using the following command

pitonstream -b genesys2 -d system -f ./piton/design/chip/tile/ariane/ci/riscv-benchmarks.list --core=ariane --precompiled

Note the -precompiled switch here, which has the same effect as when used with the sims command.

Debugging via JTAG

OpenPiton+Ariane supports the RISC-V External Debug Draft Spec and hence you can debug (and program) the FPGA using OpenOCD. We provide two example scripts for OpenOCD below.

To get started, connect the micro-USB port that is labeled with JTAG to your machine. This port is attached to the FTDI 2232 USB-to-serial chip on the Genesys 2 board, and is usually used to access the native JTAG interface of the Kintex-7 FPGA (e.g. to program the device using Vivado). However, the FTDI chip also exposes a second serial link that is routed to GPIO pins on the FPGA, and we leverage this to wire up the JTAG from the RISC-V debug module.

If you are on an Ubuntu based system you need to add the following udev rule to /etc/udev/rules.d/99-ftdi.rules

SUBSYSTEM=="usb", ACTION=="add", ATTRS{idProduct}=="6010", ATTRS{idVendor}=="0403", MODE="664", GROUP="plugdev"

Once attached to your system, the FTDI chip should be listed when you type lsusb

Bus 005 Device 019: ID 0403:6010 Future Technology Devices International, Ltd FT2232C/D/H Dual UART/FIFO IC

If this is the case, you can go on and start openocd with the fpga/ariane.cfg configuration file below.

$ openocd -f fpga/ariane.cfg
Open On-Chip Debugger 0.10.0+dev-00195-g933cb87 (2018-09-14-19:32)
Licensed under GNU GPL v2
For bug reports, read
    http://openocd.org/doc/doxygen/bugs.html
adapter speed: 1000 kHz
Info : auto-selecting first available session transport "jtag". To override use 'transport select <transport>'.
Info : clock speed 1000 kHz
Info : TAP riscv.cpu does not have IDCODE
Info : datacount=2 progbufsize=8
Info : Examined RISC-V core; found 1 harts
Info :  hart 0: XLEN=64, misa=0x8000000000141105
Info : Listening on port 3333 for gdb connections
Ready for Remote Connections
Info : Listening on port 6666 for tcl connections
Info : Listening on port 4444 for telnet connections
Info : accepting 'gdb' connection on tcp/3333

Note that this simple OpenOCD script currently only supports one hart to be debugged at a time. Select the hart to debug by changing the core id (look for the -coreid switch in the ariane.cfg file). If you would like to debug multiple harts at once, you can use ariane-multi-hart.cfg.

Then you will be able to either connect through telnet or with gdb:

$ riscv64-unknown-elf-gdb /path/to/elf
(gdb) target remote localhost:3333
(gdb) load
Loading section .text, size 0x6508 lma 0x80000000
Loading section .rodata, size 0x900 lma 0x80006508
(gdb) b putchar
(gdb) c
Continuing.

Program received signal SIGTRAP, Trace/breakpoint trap.
0x0000000080009126 in putchar (s=72) at lib/qprintf.c:69
69    uart_sendchar(s);
(gdb) si
0x000000008000912a  69    uart_sendchar(s);
(gdb) p/x $mepc
$1 = 0xfffffffffffdb5ee

You can read or write device memory by using:

(gdb) x/i 0x1000
    0x1000: lui t0,0x4
(gdb) set {int} 0x1000 = 22
(gdb) set $pc = 0x1000

In order to compile programs that you can load with GDB, use the following command:

sims -sys=manycore -novcs_build -midas_only hello_world.c -ariane -x_tiles=1 -y_tiles=1 -gcc_args="-g"

Note that the tile configuration needs to correspond to your actual platform configuration if your program is a multi-hart program. Otherwise you can omit these switches (the additional cores will not execute the program in that case).

Booting SMP Linux on Genesys2 or VC707

We currently support single core and SMP Linux on the Genesys2, Nexys Video, and VC707 FPGA development boards. For familiarisation and to ensure your hardware is setup correctly first try running with a released bitfile and SD card image.

To prepare the SD card with a Linux image you need to format it with sgdisk then write the image with dd.

1. Download the Ariane Linux OS image from either the ariane-sdk release or the Princeton archive, extract and save the .bin file as bbl.bin in the current directory. If you want to build your own Linux image please see ariane-sdk.
2. $ sudo fdisk -l Search carefully for the corresponding disk label of the SD card, e.g. /dev/sdb
3. $ sudo sgdisk --clear --new=1:2048:67583 --new=2 --typecode=1:3000 --typecode=2:8300 /dev/sdb Create a new GPT partition table and two partitions: 1st partition 32MB (ONIE boot), 2nd partition rest (Linux root).
4. $ sudo dd if=bbl.bin of=/dev/sdb1 oflag=sync bs=1M Write the bbl.bin file to the first partition. E.g. where your disk label is /dev/sdb use /dev/sdb1 (append a 1).
5. Insert the SD card into the FPGA development board. You can leave it there until you want to build your own Linux OS image.

Note that the board specific settings are encoded in the device tree that is automatically generated and compiled into the FPGA bitfile, so no specific configuration of the Linux kernel is needed.

Next up is generating the bitfile which assumes you've setup your PATH by sourcing /opt/xilinx/Vivado/2018.2/settings64.sh and piton/ariane_setup.sh. The default configuration is 1 core for all boards, but you can override this with command line arguments. In order to build an FPGA image for these boards, use one of the following commands representing the maximum configurations:

• protosyn -b nexysVideo -d system --core=ariane --uart-dmw ddr
• protosyn -b genesys2 -d system --core=ariane --uart-dmw ddr --x_tiles=2
• protosyn -b vc707 -d system --core=ariane --uart-dmw ddr --x_tiles=3 --y_tiles=1 (81% LUT utilization)

Vivado version 2017.3 and previous are known to fail for various reasons but may generate an unusable bitfile. Please use Vivado 2018.2.

This command will take a while (1-2 hours is typical for the first run before IP has been generated) to generate a bitfile at build/vc707/system/vc707_system/vc707_system.runs/impl_1/system.bit. To get started you can, alternatively, try a released bitfile from the Princeton archive,

Now that you have a prepared SD card inserted into the dev board, and a bitfile it's time to boot up. The Linux OS provides console access over UART.

1. Connect a mini-USB cable to the port labelled UART and power on the board which allows the interfaces such as /dev/ttyUSB2 to become available.
2. Open a console with 115200/8N1. E.g. something like screen /dev/ttyUSB0 115200 or sudo minicom -D /dev/ttyUSB2 If there are multiple ttyUSB devices just open a console to each of them.
3. Connect a micro-USB cable to the port labelled JTAG and connect from within the Vivado Hardware Manager. This can be done in the GUI by opening the project: vivado build/vc707/system/vc707_system/vc707_system.xpr &
4. Program the device with the generated bitfile, which Vivado should find automatically. Once programming is finished (around 10s) reset will be immediately lifted and you should see the Linux boot process being reported on the UART console.

When the device comes out of reset, the zero-stage bootloader copies the Linux image, including the first stage bootloader, from the SD card into DDR, and executes it. Be patient, copying from SD takes a couple of seconds. When the boot process is finished a login prompt is displayed. The username is root without a password. Now you can test things by running standard unix commands (# cat /proc/cpuinfo), or playing tetris (# /tetris).

There is also preliminary support for the VCU118, but not all features work yet on that board. For the VCU118 board you need the PMOD SD adapter from Digilent to be able to use an SD card (the slot on the VCU118 board is not directly connected to the FPGA). As the PMOD0 port has open-drain level-shifters, you also have to replace the R1-R4 and R7-8 resistors with 470 Ohm 0201 SMD resistors on the Digilent PMOD SD adapter to make sure that signal rise times are short enough.

Running OpenPiton simulations on F1 instances in AWS: step guide

Here is the generic flow to run OpenPiton on F1 instance. We created a public image (agfi-0d87a634f93fe7c83), which you can use to try OpenPiton on F1 without synthesizing it.

1. We assume that you already have F1 instance up and running. If not - steps 1,2,4,5 from this (https://github.com/vegaluisjose/aws-fpga-notes) guide will help you.

2. ssh into your instance, clone OpenPiton repo (https://github.com/PrincetonUniversity/openpiton).

3. cd into repo, run these bash commands:

    export PITON_ROOT=`pwd`
    export AWS_FPGA_REPO_DIR="$PITON_ROOT/piton/design/aws"
    export CL_DIR="$AWS_FPGA_REPO_DIR/hdk/cl/developer_designs/piton_aws"
    source piton/piton_settings.bash
    source piton/ariane_setup.sh

4. Load the fpga image into board:

    fpga-load-local-image -S 0 -I agfi-0d87a634f93fe7c83 

After this step the fpga is programmed, but the reset signal is high, so the system is still not working.

5. Compile software

    cd $CL_DIR/software/src 
    make

This will compile programs "uart" and "dma_os". You should have xdma driver preinstalled (https://github.com/Xilinx/dma_ip_drivers/tree/master/XDMA/linux-kernel).

6. Run "uart" program

    ./uart & 

This will create a pseudo-terminal and tell you the location of the corresponding file (e.g. /dev/pts/3)

7. Write OS image in memory:

    ./dma_os $FILE_LOCATION 

This will put the os image from FILE_LOCATION in the appropriate place in memory. Note, before writing the image you should revert each 8 bytes of it (consequences of strange behavior of xdma driver). You can do this with

    objcopy -I binary -O binary --reverse-bytes=8 $FILE_LOCATION 

8. Reset fpga:

    ./fpga-reset 

After this the processor will start working and printing UART data in your pseudo-terminal. You can connect to the terminal using your favourite terminal program (e.g. screen, tio).

9. Managing the fpga: In the software directory we provide you with some useful programs:

• uart : starts pseudo-terminal, connected to the OpenPiton's uart
• dma_os : copies file from argument into the SD part of the memory
• fpga-reset : resets the fpga
• fpga-poweroff : sets the reset in the fpga to high, basically powering off OpenPiton so that you could write the data into memory without memory corruptions.

Besides, you might find useful some utilities, provided by AWS themselves:

• fpga-clear-local-image : clears the image from FPGA
• fpga-set-virtual-dip-switch : sets the value of 16 virtual dip switches. Note that the last switch (15th) is connected to OpenPiton's reset
• fpga-get-virtual-led : reads the value of 16 virtual leds

Synthesizing OpenPiton image for F1

The flow is very simillar to synthesizing image for any other FPGA we support, but has own nuances. Good news: you can run the flow on your own machine if you have Vivado 2018.2 or higher.

1. Create the S3 credentials and configure your S3 bucket. You can find step guides here (https://github.com/vegaluisjose/aws-fpga-notes).

2. Clone OpenPiton repo (https://github.com/PrincetonUniversity/openpiton).

3. cd into repo, run these bash commands:

    export PITON_ROOT=`pwd`
    export AWS_FPGA_REPO_DIR="$PITON_ROOT/piton/design/aws"
    export CL_DIR="$AWS_FPGA_REPO_DIR/hdk/cl/developer_designs/piton_aws"
    source piton/piton_settings.bash
    source piton/ariane_setup.sh
    source "$AWS_FPGA_REPO_DIR/hdk_setup.sh"

The last command will try to ask for root password to apply patch for Vivado, but you don't have to do it - the flow still works even without patch.

4. Run the synthesis:

    protosyn -b f1 -c ariane ... 

This will create the custom logic tar archive, which we'll later upload on AWS servers. The synthesis itself is run in nohup, but protosyn will tell you the location of the log files, so that you can follow the progress.

5. After the synthesis is complete (takes about 2-3 hours on fast PC), go to results folder:

    cd $PITON_ROOT/build/f1/piton_aws/build/checkpoints/to_aws 

This is the folder, where all the result tars are located.

6. Copy the resulting tar archive in the S3 bucket you created before

7. Send the command for final synthesis:

    aws ec2 create-fpga-image --name NAME_OF_IMAGE --input-storage-location Bucket=YOUR_S3_BUCKET,Key=NAME_OF_YOUR_TAR_ARCHIVE 

The command will tell print the afi and agfi of your image. You can track the synthesis progress with

    aws ec2 describe-fpga-images --fpga-image-ids AFI_OF_YOUR_IMAGE 

8. After the synthesis is done - you can go load it in your F1 instance!

Planned Improvements

The following items are currently under development and will be released soon.

• Thorough validation of cache coherence.
• RISC-V FESVR support in simulation.
• Synthesis flow for large FPGAs.
• Performance enhancements (cache re-parameterization, write-buffer throughput).

Stay tuned!