Posted on

ECP5G Versa board under Linux

The ECP5 platform has caught quite some attention a while ago, being on the rather high end with respect to GigaBit LVDS communication of all sorts. It’s the successor of the ECP3 which has been performing well with HDMI applications in the past.

Lattice Semiconductor had launched a Promo action for this Versa ECP5G board. With great assistance from Future Electronics Switzerland I was able to get hold of a devkit.

Running the Lattice Diamond toolchain on a Linux environment has so far been straightforward, with minor quirks on the GUI side. Let me revisit the important items to get up and running. If you happen to have a Linux OS installed that does not match the Lattice Semi recommendations for a development system, you might want to look at the Docker approach on how to set up the environment.

Programmer preparation

Porting a very simple CPU with UART interface to the platform, the final step is flashing things down into the board using the Lattice Diamond Programmer. Usually, when not having installed any udev rules, these are the steps you have to go through with root powers, in order to get the USB FTDI programmer interface recognized:

Bus 001 Device 013: ID 0403:6010 Future Technology Devices International, Ltd FT2232C Dual USB-UART/FIFO IC

According to this device enumeration, you give access to the device:

chmod a+rw /dev/bus/usb/001/013

and unbind the first ttyUSB0 device from the ftdi_sio driver – unfortunately, the Lattice driver is not able to unbind from within.

echo -n $TTY_ID > /sys/bus/usb/drivers/ftdi_sio/unbind

Replace $TTY_ID by the tty device entry in /sys/bus/usb/drivers/ftdi_sio/ that typically is of the form „1-3.2:1.0“ (the trailing 0 stands for port A where the JTAG is at). If you have plugged in more than one FTDI adapter with UART capabilities, you will see several and need to figure out which is which.

Now you can download code into the programmer.

To automatize the process, you could also use the script below:


allow_io=`lsusb | sed -n 's/^Bus \([0-9]*\) Device \([0-9]*\): ID 0403:6010 .*/\1\/\2/p'`

unbind_tty=`ls /sys/bus/usb/drivers/ftdi_sio/ | sed -n 's/\(.*\:1\.0\).*/\1/p'`

sudo chmod a+rw \/dev\/bus\/usb\/$allow_io
sudo sh -c "echo $unbind_tty > /sys/bus/usb/drivers/ftdi_sio/unbind"

SPI flash download

When downloading your design into the SPI flash, make sure you have MASTER_SPI_PORT=ENABLED set in your *.lpf file. Otherwise the programmer will fail with an error report on CHECK_ID:

ERROR - Verification Error...when Processing function: 'CHECK_ID'

Then you’ll have to use the fast download (into SRAM) with an enabled Master SPI in order to be able to flash again.

Design issues

When starting to port our SoC design to this board, quite a few issues came up:

  • Don’t bother developing with Diamond v3.7. Some very obscure behaviour with wrong I/O mapping cost quite some headache. Seems to be solved in v3.8
  • v3.8 however is somewhat misleading with respect to output and return states from the Synopsys Synthesis engine ‚Synplify‘. Make sure to check your design thoroughly, if Synplify throws an error, Diamond sometimes would not recognize that and map/PAR an old design netlist.
  • Weird random behaviour can occur under Linux with Diamond during PAR, like error messages with respect to path names. This seems to happen especially with long names and underscores. The behaviour has been around in many previous versions of Diamond, Tech Support has refused to accept this as a bug so far. The workaround is to call PAR using the command line (TCL) or use the Run Manager.
    Addendum: It seems that this bug is fixed in v3.9, but there is no release note about it.


Talking through the UART

In theory, you should be able to talk to your design through the UART (if your design supports it) by firing up minicom:

minicom -o -D /dev/ttyUSB1

Now here comes the catch: The default EEPROM from my Versa 5G board did not have a correct descriptor, however it came with the default VID:PID from FTDI, so the ftdi_sio driver would recognize it, but in fact not communicate, neither report an error.

So, in order to properly use this board, you may have to erase the EEPROM of the FTDI adapter on the Versa kit using FTDIs Mprog tool or alike. You might want to save the previous EEPROM content for reference, however it does not seem needed, the Programmer recognizes the Board just fine.

Finally, after downloading our SoC setup into the board, it is talking:

Booting, HW rev: 04 -- Running at 50 MHZ

------------- test shell -------------
-- ZpuSoC for Versa ECP5 --
-- (c) 2017 --
-- type 'h' for help --

Simulation issues

When trying to fully simulate the SoC setup with PLL primitives and some instanced ip cores created by the Clarity module (obviously that’s the IPExpress descendant for ECP5), it turns out that some of the simulation primitives for the VHDL side are missing.

Some of them could be converted using the VHDL conversion trick from Icarus Verilog by the following Makefile rule:

%.vhdl: %.v
    iverilog -tvhdl -o $@ -pdepth=1 $<

However, some of the needed components might not convert without additional tweaking. Hopefully, Lattice Semi will come out with updated VHDL libraries.

IPcore simulation under GHDL

When generating IP cores that depend on library items and running them through GHDL, you might see this error message:

warning: component instance "scuba_vlo_inst" is not bound

However, if you have prepared your FPGA primitive components library, like right, the primitive simulation models should be in there (search for ‚vlo‘ in the *.cf file if in doubt).

The reason why this is happening is that there could be component prototype declarations in the IP core file that shouldn’t be there in order to reference to the components from the library. So: Just remove the „component“ declaration sections and all should link fine. The drawback of this is, that you need two IP core versions, one for simulation and one for synthesis. If you have a better solution, let me know.

Next steps

Now, the fancy stuff on this board to be evaluated is:

  • DDR3 memory
  • Two GigE capable interfaces

Also, the ECP5 on this board has enough resources to run several ZPUng cores simultaneously. For safety reasons, we wouldn’t want the IoT crap run in the same environment as our controlling main loop.

Therefore, a second processor (Core B) is instanced for running the Ethernet Stack (lwip) only, while maintaining a simple DMA channel to the controller Core A for communication. If core B is compromised, Core A will still maintain its „hardened“ control loop and not go haywire.

Implementing DDR3 is tricky. Therefore you might want to use the DDR3 IP core supplied by Lattice. On the demo kit, it will work for a few hours and then pull the global reset.


Of course, I was very curious about the Ethernet ports on this board. It’s armed with two GigE capable Marvell Phys whose data sheets are a little hard to get hold of, but one might also look at various source code around the web or just check the reference design from Lattice. The reference design uses lwip, since I only need and want UDP, I ported a zerocopy-capable UDP stack I developed for the Blackfin EMAC to the ZPUng SoC („cranach“), which is equipped with some DMA capable scratch pad memory for a proper packet queue.

After all, the FPGA is now able to speak netpp, so I can turn on an LED for example:

> netpp UDP:192.168.05:2016 LED.Yellow 1

Resource usage

You might want to know how much logic and RAM is consumed by this solution.

Design Summary
   Number of registers:   2770 out of 44457 (6%)
      PFU registers:         2767 out of 43848 (6%)
      PIO registers:            3 out of   609 (0%)
   Number of SLICEs:      2963 out of 21924 (14%)
      SLICEs as Logic/ROM:   2891 out of 21924 (13%)
      SLICEs as RAM:           72 out of 16443 (0%)
      SLICEs as Carry:        309 out of 21924 (1%)
   Number of LUT4s:        4154 out of 43848 (9%)
   Number of block RAMs:  28 out of 108 (26%)
   Number of DCS:  1 out of 2 (50%)
   Number of PLLs:  1 out of 4 (25%)

As for the actual program code, containing:

  • UDP stack supporting ARP, ICMP ping
  • Minimal shell (UART)
  • System I/O drivers (UART, Timer, MAC, PWM)
  • netpp minimal server with some LED handling

This is what’s effectively downloaded into the target:

(gdb) init
Loading section .fixed_vectors, size 0x400 lma 0x0
Loading section .l1.text, size 0x4af5 lma 0x400
Loading section .rodata, size 0x168 lma 0x4ef8
Loading section .rodata.str1.4, size 0xf60 lma 0x5060
Loading section .data, size 0x2c0 lma 0x5fc0
Start address 0x0, load size 25213

There’s a significant amount of string data for debugging in the .rodata.str1.4 section due to debugging info, plus some netpp descriptors. These again could be ‚overlayed‘ to the SPI flash, as they are not too frequently accessed. To be investigated next…

Posted on

Dockerized FPGA toolchains

For a while, LXC (linux container) technology might have been known for the better chroot. Docker takes this approach even further by letting you mess with changes and undo them easily. You can just install foreign binaries and play with dependencies without compromising your desktop host’s runtime libraries. This article describes how to easily put commercial FPGA toolchains into a docker environment and carry them around on an external hard disk for quick installation on a new developer machine.

Lattice Diamond

The Lattice Diamond toolchain comes as RPM package and is recommended to be run under a Redhat OS. It is possible to convert to a DEB and install and run it on a Debian system likewise, but we try basing it on a minimal RPM compatible environment as an existing CentOS container.

This short howto describes the necessary steps for Diamond v3.8.

First, create a file called Dockerfile in a sandbox work directory, like:

FROM centos

RUN yum update ; \
	yum install -y freetype libSM libXrender fontconfig libXext libXt \
		tcl xorg-x11-fonts-Type1 net-tools libusb-0.1.4 usbutils \

RUN adduser -u 1000 -g 100 diamond; echo "." >> /home/diamond/.bashrc

COPY /home/diamond

ENTRYPOINT ["/bin/bash"]

You could automate things more by copying the RPM into the docker container, but that would just take useless space in the image (and stripping this back down would take little elegant extra action). Therefore we mount the directory where the RPM was downloaded to into the container. Unfortunately, there is no clean way to pull it from an official source.

You may have to sort out permissions first (by adding yourself to the ‚docker‘ group) or prepend ’sudo‘ to each docker call.

Other than that, follow these steps:

  1. Change UID and GID for adduser command in the Dockerfile, if necessary
  2. Run
    docker build -t diamond .
  3. Then you can start the docker container using the following command. Note you have to replace $ETHADDR by the ethernet MAC address you have registered your license.dat to. Also, make $PATH_TO_RPM_INSTALLDIR point to the directory where you installed the RPM from.
    docker run -ti -e DISPLAY=0:0 \
    --mac-address=$ETHADDR \
    -v /dev/bus/usb/:/dev/bus/usb/ \
    -v /tmp/.X11-unix/:/tmp/.X11-unix diamond:latest
  4. Then install diamond by
    rpm -i /mnt/diamond_3_8-base_x64-115-3-x86_64-linux.rpm
  5. You still need to copy your license.dat into your /home/diamond/ directory within the docker container. If it’s supposed to be elsewhere, edit the LM_LICENSE_FILE environment variable in
  6. Then you should be able to start the diamond GUI as user ‚diamond‘:
    su -l diamond
  7. Finally, if you are happy with your changes, you might want to commit everything to a new image:
    docker commit -m "Diamond install" $HASH_OF_YOUR_CONTAINER diamond:v0

A few notes

The -v option takes care about sharing your X11 sockets with the docker sandbox. Note that there also some options inside the to make QT work in this limited environment. Don’t try to run Diamond as root, as the X11 forwarding will not be allowed.

For Diamond v3.9, the libXt package needs to be installed, the Dockerfile listing was updated accordingly.

Xilinx ISE

The same procedure works  for the ISE 14.7 toolchain. The Dockerfile in this case is almost similar, although includes a few X11 extras.

FROM centos

RUN yum update ; \
    yum install -y freetype libSM libXrender fontconfig libXext \
        tcl xorg-x11-fonts-Type1 net-tools libXScrnSaver-1.2.2 \
        libXi libXrandr \
        libusb-0.1.4 usbutils

RUN adduser -u 1000 -g 100 ise; echo "." >> /home/ise/.bashrc

COPY /home/ise

ENTRYPOINT ["/bin/bash"]

Downloading and unpacking ISE

Make sure you have downloaded the following files from the Xilinx website:


Then untar the first one by

> tar xf Xilinx_ISE_DS_14.7_1015_1-1.tar

This directory path will have to be exported to docker under $PATH_TO_UNPACKED_XILINX_TAR below.

The file:



alias ise=$XILINXDIR/ISE/bin/lin64/ise

Likewise, the docker container is run by something like:

docker run -ti -e DISPLAY=0:0 \
--mac-address=$ETHADDR \
-v /dev/bus/usb/:/dev/bus/usb/ \
-v /tmp/.X11-unix/:/tmp/.X11-unix ise:latest

Once you’re inside docker, install as root by running /mnt/xsetup. This may take a long time. Left to do:

  • Install a license file
  • Mess with the USB drivers for Impact. This is left open to the user. I am using xc3sprog from my host system.
Posted on

netpp on programmable logic: IoT for the FPGA

For quite a while I wouldn’t have said, it’s impossible, but wouldn’t put much effort in it either. Why, if you have a spare $1 microprocessor that can run a simple netpp communication stack just fine.
Well, sometimes it’s time to try something else: Running a soft core CPU (ZPU) on small FPGAs has found some interest, due to the limited resource consumption, but full freedom when it comes to specific interfaces, such as

  • Programmable PWM engines (motor control)
  • Many RS485 capable interfaces (that classic uCs don’t have)
  • Safety proof specific state machines

The ZPU will even fit on a $5 FPGA and still leave enough space for the specific interfaces. It is a slow stack machine, even the fastest pipelined implementations don’t really beat the MIPS alike architectures, however this doesn’t bother us when we just have to configure a set of registers, moreover, the ZPU architecture compensates with quite some code density.

Also, for an update, we have implemented a faster, pipelined and configureable version of a stack machine with ZPU compatible front end, called ZPUng.

To keep a long sermon short: A full netpp stack running over a UART interface fits in less than 15kB of memory. And runs on a MachXO2-7000 from Lattice, for example, with less than 50% logic usage.
Who’s still saying that an FPGA is too dumb for the internet?

Ok, there’s one little missing piece: The TCP/IP stack and the ethernet MAC. It may make sense to separate core and communication section for safety reasons and use:

  • esp8266 WLAN chip (not for industrial)
  • Wiznet 5100/5300 TCP-on-chip

This solution was presented on the Embedded World Conference 2016 in Nuremberg. The published documentation can be downloaded for free in the web shop.

Posted on

Virtual ROM on small FPGAs

Readonly data – outsourced

When running display output applications on small FPGAs where printing of strings is required, using the internal block RAM for character sets would be a waste of resources, or even: not sufficient. The 64kB character map for a normal and inverted font using both RGB and BGR subpixel smoothed renderings (more below) would eat up more than available on the Spartan3-250k device, for example.
The Spartan3 on my good old Papilio has a SPI flash attached with less than 50% actual usage for the FPGA bit stream. Tadaa. Plenty of space for character bitmaps. So we can just store the bitmap in an unused sector on the SPI and blit-copy from SPI flash to the LCD display directly.
But what if there’s more read-only data, like second stage program code or coprocessor microcode that is loaded on the fly by an applet? If we don’t need it for the boot process, it should not live in the block ram permanently, but still execute from block RAM. Screams for a cache, doesn’t it?
So the simple solution actually is, to create a small controller entity ’scache‘ inside the system specific peripheral of the SoC. This simply watches access to certain addresses and generates an exception once this address is hit. The exception vector then jumps into a handler routine which does all the SPI flash loading into the physical cache memory area. Then, for the next LOAD instruction, the virtual address internally translates to the physical cache address.
This requires very simple logic, however runs through some program code and needs a bit of time to load from the SPI flash.
Turns out this is barely noticeable for the LCD display.

Under the hood: Linker scripting

Ok, so there is plenty more data from the .rodata section. We could implement some kind of overlay loader and a kind of file system, but why make it complicated, when our data is somewhat static. We just relocate the cached external data using a linker script. Say, our program memory ranges from 0x0000 to 0x2000, the cache is allocated after that. Then we define a memory specification in the linker script as shown below:

        l1ram(rwx): ORIGIN = 0x0000, LENGTH = 0x2000
        l1cache(rwx): ORIGIN = 0x2000, LENGTH = 0x2000
        xdata(r): ORIGIN = 0x10000, LENGTH = 0x8000

For the .rodata section, we can simply use a line like

.rodata         : { *(.rodata .rodata.* .gnu.linkonce.r.*) } > xdata

to allocate the read-only areas into the virtual SPI ROM starting from 0x10000. Eventually, compiling the whole story will result in an ELF file which we only have to separate into the boot ROM area and the second stage binaries that live in the SPI flash.

But we may want to have read-only data that should be present at all times, like library code that is frequently used. For this purpose, we can just define our own segments and use the __attribute__ decorator for the data that we want in BRAM:

__attribute__((section(".l1.rodata"))) char g_fifobuf[32];

I will not dive into the details of linker scripting, but likewise you can also allocate entire library or object files in specific memory segments.

This simple technique allows you to pack quite a bit of code into small FPGA SoCs.

The LCD example

LCD screen
LCD subpixel smoothing example

Here you can see an example of the character table in action. I have mentioned the RGB/BGR subpixel smoothing above which lead to a bit higher ROM usage. The used LCDs have quite a bit of functionality, like setting the orientation of the screen display. You could just not care about when designing a character table and use a simple black/white scheme. However, when trying to achieve a 4×8 pixel character set, the font will look quite unreadable and you’re better off with the subpixel smoothing. This again is sensitive to the pixel order, so for top and bottom display orientation you will already need the font rendered in two variants, plus in color. Now, as we have the space, it is just much simpler to drop the entire bit map into the flash instead of figuring out compression techniques.

Program code

Likewise, not too frequently used program code can be dropped into external SPI flash. This comes in handy when a program becomes bigger during development and the space usage can not be determined a priori. The ZPUng is able to handle a specific exception signal that is risen by the SCACHE controller. The cache handler microcode routine then loads code from SPI flash and resumes execution while translating the virtual PC address into a physical address inside the cache area, where instructions are effectively fetched from. This is the trick used to run the netpp communication library for IoT applications  on FPGAs with limited resources, like the Papilio One.