Re: [PATCH net-next v4 1/8] dt-bindings: net: add Xilinx TSN Endpoint Ethernet MAC

From: Neeli, Srinivas

Date: Fri Nov 20 2026 - 01:13:00 EST


Hi Andrew,

On 11/15/2026 3:13 AM, Andrew Lunn wrote:
Is this a synthesis parameter? No other switch that i know of has it
in DT. The driver hard codes the value based on the compatible. Why
does that not work for you?
  Yes, MAX_LEARNING_ENTRIES is a Vivado synthesis parameter of the IP
  (256 to 1920), and no register reports it
I assume that will be fixed in the next version of the IP?
    Exposing synthesis parameters is standard behavior for FPGA IPs, and not all of these are exposed through registers.
    Hence, I am afraid there is no plan to change this in the IP.
    For Xilinx/AMD FPGA IPs, these values can vary depending on the Vivado configuration,
    so the same compatible may have different values for the same property depending on the design.
    We have similar cases in the AXI Ethernet driver, for example, xlnx,rxcsum and xlnx,txcsum, which are based on synthesis parameters.

Are there many other synthesis parameter which are not exported in
registers? I just want to get an idea if this is a one off, or you are
going to add lots of DT parameters in the next set of patches.
  Yes, there are a few additional parameters that we plan to expose through DT properties in the upcoming series.
  The TSN subsystem has several configurable parameters, and I understand that DT should describe the hardware irrespective of its usage by software.
  But given the volume of such parameters, we are considering adding DT properties in an incremental fashion, as and when there are consumers.


  The number of ports is fixed by the IP. The switch always has two
  external MAC ports and one internal endpoint port, and no configuration
  has more external MACs, so the binding never needs to grow another port.
And that is signed off by top management? The next switch you design
will be a fresh design, no reuse of the existing IP and will require a
whole new DSA driver?

  The switch node follows the IP's register map. The switch fabric and the
  two MAC blocks are separate register windows of the switch IP, and each
  MAC window also holds that MAC's MDIO controller and PTP block. The PTP
  timer interrupt is shared by the whole switch, not owned by one port.

  For these reasons I would prefer to keep the per-MAC resources on the
  switch node

+ switch@38000 {
+ compatible = "xlnx,tsn-switch";
+ reg = <0x38000 0x8000>,
+ <0x00000 0x14000>,
+ <0x20000 0x14000>;
+ reg-names = "switch", "mac1", "mac2";
I ask, because it would not be too hard to swap this address map
around, put the switch registers first, and then the port registers,
and you could have as many ports as you want.

Back to what you just said:

each
MAC window also holds that MAC's MDIO controller and PTP block.
So the MDIO controller and the PTP block are MAC properties, so should
be inside the MAC node.

The PTP
timer interrupt is shared by the whole switch
So that should be at the switch level.

Andrew
  Yes, I agree with your suggestion. I will move the MAC and MDIO properties under ethernet-ports in the next series.

Thanks
Srinivas Neeli