On Tue, Jan 21, 2025 at 02:08:40PM +0100, Maxime Chevallier wrote:

On Tue, 21 Jan 2025 15:54:53 +0800
Yijie Yang [off-list ref] wrote:

quoted

Correct the definition of 'phy-mode' to reflect that RX and TX delays are
added by the board, not the MAC, to prevent confusion and ensure accurate
documentation.

That's not entirely correct though. The purpose of the RGMII variants
(TXID, RXID, ID) are mostly to know whether or not the PHY must add
internal delays. That would be when the MAC can't AND there's no PCB
delay traces. Some MAC can insert delays.

There's documentation here as well on that point :

https://elixir.bootlin.com/linux/v6.13-rc3/source/Documentation/networking/phy.rst#L82

This is part of the problem. This describes
PHY_INTERFACE_MODE_RGMII_*, and the value passed to phylib. The
documentation even says:

   The values of phy_interface_t must be understood from the
   perspective of the PHY device itself,

But the value in DT is about the board as a whole, it describes the
hardware. Software gets to decide if the MAC or the PHY adds the
delays, if the board does not provide the delay.

So, MACs may insert delays. A modification for the doc, if needed,
would rather be :

-      # RX and TX delays are added by the MAC when required
+      # RX and TX delays are added by the MAC or PCB traces when required

From the perspective of the board, this is wrong. 'rgmii' means the
board provides the delays.

There is a parallel discussion going on, about how aspeed have also
got there implementation wrong. See:

https://lore.kernel.org/netdev/0ee94fd3-d099-4d82-9ba8-eb1939450cc3@lunn.ch/ (local)

and the test of that thread.

	Andrew