Skip to content

b5g-ex/nerves_system_kr260

Repository files navigation

Kria KR260

Hex version

This is the base Nerves System configuration for the Kria KR260.

Kria KR260

The above image is obtained from AMD's official website under the Media Library Terms and Conditions.

Feature Description
CPU 1.333 GHz quad-core Cortex-A53
Memory 4 GB DDR4
Storage MicroSD, 512 Mb QSPI, 16 GB eMMC, 64 Kb EEPROM
Linux kernel linux-xlnx
IEx terminal HDMI and USB keyboard (can be changed to UART)
GPIO, I2C, SPI Raspberry Pi HAT I/F - Elixir Circuits
ADC No
PWM No
UART 1 available - ttyPS1
Display Display Port 1.2a
Camera No
Ethernet 2 avaiable PS ethernet and 2 PL
WiFi No
Bluetooth No
Audio No
Pmod 4 available - 12-pin, Untested
USB 4 available

Using

The most common way of using this Nerves System is create a project with mix nerves.new and to export MIX_TARGET=kr260. See the Getting started guide for more information.

If you need custom modifications to this system for your device, clone this repository and update as described in Making custom systems.

Provisioning devices

This system supports storing provisioning information in a small key-value store outside of any filesystem. Provisioning is an optional step and reasonable defaults are provided if this is missing.

Provisioning information can be queried using the Nerves.Runtime KV store's Nerves.Runtime.KV.get/1 function.

Keys used by this system are:

Key Example Value Description
nerves_serial_number "12345678" By default, this string is used to create unique hostnames and Erlang node names. If unset, it defaults to 4-digits, part of MAC address.

The normal procedure would be to set these keys once in manufacturing or before deployment and then leave them alone.

For example, to provision a serial number on a running device, run the following and reboot:

iex> cmd("fw_setenv nerves_serial_number 12345678")

This system supports setting the serial number offline. To do this, set the NERVES_SERIAL_NUMBER environment variable when burning the firmware. If you're programming MicroSD cards using fwup, the commandline is:

sudo NERVES_SERIAL_NUMBER=12345678 fwup path_to_firmware.fw

Serial numbers are stored on the MicroSD card so if the MicroSD card is replaced, the serial number will need to be reprogrammed. The numbers are stored in a U-boot environment block. This is a special region that is separate from the application partition so reformatting the application partition will not lose the serial number or any other data stored in this block.

Additional key value pairs can be provisioned by overriding the default provisioning.conf file location by setting the environment variable NERVES_PROVISIONING=/path/to/provisioning.conf. The default provisioning.conf will set the nerves_serial_number, if you override the location to this file, you will be responsible for setting this yourself.

FPGA

This system provides dfx-mgr, so we can load/unload accelerators by following steps.

  • Add dfx-mgrd under the application supervisor, lib/[project name]/application.ex.
    It's recommended to use MuonTrap.Daemon.
    IMPORTANT: dfx-mgrd must be to run on RW partition. (It writes state.txt on it...
  def children(_target) do
    [
      # Children for all targets except host
      # Starts a worker by calling: Kr260Test.Worker.start_link(arg)
      # {Kr260Test.Worker, arg},
    ] ++ dfx_mgrd()
  end

  defp dfx_mgrd() do
    dfx_mgrd = "/usr/bin/dfx-mgrd"

    if File.exists?(dfx_mgrd) do
      # nerves_system_kr260 provides `/lib/firmware` as tmpfs
      [{MuonTrap.Daemon, [dfx_mgrd, [], [cd: "/lib/firmware"]]}]
    else
      []
    end
  end
  • load/unload accelerators by dfx-mgr-client.
    It's recommended to create wrapper but following example use client directly :).
# list packages
iex()> cmd "dfx-mgr-client -listPackage"
                     Accelerator          Accel_type                            Base           Base_type      #slots(PL+AIE)         Active_slot

                k26-starter-kits            XRT_FLAT                k26-starter-kits            XRT_FLAT               (0+0)                  0
# unload
iex()> cmd "dfx-mgr-client -remove"     
remove from slot 0 returns: 0 (Ok)
# load
iex()> cmd "dfx-mgr-client -load k26-starter-kits"
Loaded to slot 0

About the default firmware

The default firmware can be specified by rootfs_overlay/etc/dfx-mgrd/default_firmware.

User can change it by your Nerves project's rootfs_overlay/etc/dfx-mgrd/default_firmware.

Would like to know more details? Let's read https://github.com/Xilinx/dfx-mgr/tree/xilinx_v2022.1_update2#daemonconf.

NOTE:This system specified k26-starter-kits for default, but doesn't include the firmware.

FAQ

Q: Where should accelerators be located?

A: Put them under your Nerves project's rootfs_overlay/usr/lib/firmware/xilinx.

dfx-mgr details for DEV

  1. dfx-mgrd needs some RW area.
    1. /configfs, literally configfs
      We prepared it as configfs by erlinit, see. rootfs_overlay/etc/erlinit.config
    2. /lib/firmware, accelerator's *.dtbo is copied to here by dfx-mgrd
      We prepared it as tmpfs by erlinit, see. rootfs_overlay/etc/erlinit.config
    3. dfx-mgrd must run on RW area, 'cause it writes some files to run.
      We recommended to use tmpfs(RAM) for it.

Fan Control

In the case of Ubuntu for Kria SOMs, fan control is enabled after boot. So the fan will not run at full speed. Ubuntu uses /usr/sbin/fancontrol, bash script, and it is configured by /etc/fancontrol. /usr/sbin/fancontrol controls the fan via /sys/class/hwmon/hwmonN/pwmN(tempN_input).

On the other hand, this system doesn't control the fan. This system provides /sys/class/hwmon/hwmonN/pwmN(tempN_input), but not uses bash. So we don't use fancontrol.