Time-Sensitive Networking (TSN) is a set of standards under development by the Time-Sensitive Networking task group of the IEEE 802.1 working group.
...
TSN kernel drivers are currently support supported in Xilinx Linux staging area: https://github.com/Xilinx/linux-xlnx/tree/xilinx-v2022.1/drivers/staging/xilinx-tsn
TSN user space utilities and sample configurations are provided to enable TSN functionality. Please refer to the TSN SW user guide and the following sections for more details.
Kernel Configuration
The following config options should be enabled in order to build the TSN Subsystem:
CONFIG_XILINX_TSN
CONFIG_AXIENET_HAS_TADMA
...
The following additional config is required/selected by the TSN subsystem:
CONFIG_NET_SWITCHDEV
CONFIG_STP
CONFIG_NETFILTER
...
Please refer to PL Ethernet and DMA documentation for additional information: Documentation/devicetree/bindings/net/xilinx_axienet.txt
For more details on phy bindings please refer "Documentation/devicetree/bindings/net/phy.txt"
TSN System
- Xilinx’s TSN IP Switch have has three ports Endpoint (Port 0), MAC1 (Port 1) and MAC2 (Port 2)
- Endpoint is connected to MCDMA (Multichannel DMA), each MCDMA channel is a dedicated channel for each type of traffic, i.e. Best Effort, Scheduled Traffic and Reserved. There could be other use cases where there would be separate channels for management traffic
- MAC1 is connected to external world by PHY1
- MAC2 is connected to external world by PHY2
...
c. 802.1ASREV (not all features may be available. See ptp4l/openAvnu documentation).
PTP offload on R5 in zynqmp
...
Running gPTP(802.1AS
...
This enables customers who intend to completely isolate and have a resilient PTP function.
Running gPTP(802.1AS):
...
):
gPTP daemon can be run in two ways. One way is to run it from OpenAvnu, and the other is to run it from ptp4linux. Latter is preferred as it prints rms values at slave to identify sync with master.
...
You would observe that every second 12 ST packets are sent of which 8 packets are sent in the 100us of 10 ms cycle and the rest 4 packets are sent in the next cycle.
As we can see in the above picture, from 4th packet until 11th packet (8 packets) are received sequentially and the next packet starts after 10ms delay (packet no. 12).
Running IPIC:
IP interception translates the transmit packet with the configured source, destination MAC addresses, VLAN ID and PCP values if the packets' IPv4 tuples (source IP, destination IP, DSCP, protocol, source port number and destination port number) match. IPIC module maintains hash entries of IPv4 tuples and if the out-going packets' tuple data match with that of hash entries , translates the IP stream.
...
Make sure iperf3 server is run at the receiver as below and observe the received packets with destination MAC, source MAC addresses and VLAN ID.
- iperf3 -s
Running openlldp:
When the system has preemption capabilities, openlldp is run to detect frame preemption capabilities of the peer and if the peer has the capabilities, to enable the preemption.
...
0th bit of preemption enable register is 1 implying preemption is enabled, and 31st bit of preemption control status register is 1 implying preemption is active.
Running Spanning Tree Protocol
...
Steps to demonstrate Preemption
1. Change MAC addresses (if two xilinx boards are used back to back).
2. Launch lldpad daemon.
# lldpad -d
# lldptool set-lldp -i eth1 adminStatus=rxtx
3. Setup lldpad for additional ethernet capabilities TLV. Make sure preemption is enabled and active.
#lldptool -i eth1 set-tlv -V addEth enableTx=yes
4. Bring up the network interfaces.
5. Establish PTP sync, run qbv on tx side.
6. Add switch CAM entries as necessary, and send traffic, both ST and BE (express and preemptable) from tx side.
7. Run br_prog <interface name> stats, check if the TX MAC MERGE FRAGMENT COUNT(on tx side), RX MAC MERGE FRAGMENT COUNT(On rx side), are increasing.
Running Spanning Tree Protocol
On EP+Switch system running STP using standard linux based tools has been enabled. To enable the HW uses source MAC based filtering mechanism to identify which port a STP frame shall exit from. The TSN driver sets up these filters during driver initialization time. But to make this work, each node of the ring shall have unique mac address set up in device tree during compile time.
...
The TSN drivers and subsystem package(Yocto version) is delivered to our licensed customers via TSN lounge Access on Xilinx.com . Although the TSN drivers are available on xilinx kernel https://github.com/Xilinx/linux-xlnx, they may not be the latest. As we streamline driver delivery with our Yocto release available against Vivado release, its always recommended to use Lounge release. Eventually all TSN drivers will be pushed to Xilinx github kernel also the device tree infrastructure will be pushed to https://github.com/Xilinx/device-tree-xlnx.until 2020.1
From 2022.1 onwards, the TSN kernel drivers are present in the AMD/Xilinx linux-xlnx kernel as specified here 25329749
PTP on slave looses sync with master, when master reboots
...
| Code Block |
|---|
ip link add name br0 type bridge ip link set dev eth1 master br0 ip link set dev eth2 master br0 ifconfig br0 up ip link set dev br0 type bridge stp_state 1 |
→ Under heavy traffic PTP can go out of sync with TIMEOUT errors indicating that PTP process and underlying timestamp fetch did not receive enough time/priority to run. Tuning this depends on the traffic in the system but the typicall recommendations are below:
- Increase tx_timestamp_timeout to 10 in PTP config file. If using a board to board setup, this can be updated on both master and slave config to account for increased TX and RX traffic.
- If using tsn_talker to send traffic, no more than 3 instances of tsn talker can be used at full throughput with packet size of 1472.
→ Dynamic devicetree overlay is not supported with TSN 2022.1 kernel driver.
→ TSN PCP values are mandatory and there are no default values for PCP if not chosen as described in the section "PCP and Traffic Class".
...
→ Under heavy traffic PTP can go out of sync with TIMEOUT errors indicating that PTP process and underlying timestamp fetch did not receive enough time/priority to run. Tuning this depends on the traffic in the system but the typicall recommendations are below:
- Increase tx_timestamp_timeout to 10 in PTP config file. If using a board to board setup, this can be updated on both master and slave config to account for increased TX and RX traffic.
- If using tsn_talker to send traffic, no more than 3 instances of tsn talker can be used at full throughput with packet size of 1472.
→ Dynamic devicetree overlay is not supported with TSN 2022.1 kernel driver.
→ TSN PCP values are mandatory and there are no default values for PCP if not chosen as described in the section "PCP and Traffic Class".
Changelog
2023.2
Summary
- Support for 802.3br configuration via ethtool
- TSN probe order related fixes to handle dependencies with EP, TEMAC and DMAs
- Minor warning on ep and temac MAC address for users of MAC filter
Commits
https://github.com/Xilinx/linux-xlnx/commits/xilinx-v2023.2/drivers/staging/xilinx-tsn
2023.1
- Rebase on 6.1.0 Linux kernel
2022.2
- Bug fixes.
- Removed Hard coded interface names for tsn interfaces.
- Set default PCP to queue mapping, incase its not defined via bootargs.
2022.1
- TSN kernel driver support added to staging area
...