Re: [net-next: PATCH v3 5/6] net: mvpp2: enable using phylink with ACPI

On Wed, Jun 23, 2021 at 11:45:04PM +0200, Marcin Wojtas wrote:

Hi,

śr., 23 cze 2021 o 22:37 Andrew Lunn [off-list ref] napisał(a):

quoted

quoted

+static bool mvpp2_use_acpi_compat_mode(struct fwnode_handle *port_fwnode)
+{
+     if (!is_acpi_node(port_fwnode))
+             return false;
+
+     return (!fwnode_property_present(port_fwnode, "phy-handle") &&
+             !fwnode_property_present(port_fwnode, "managed") &&
+             !fwnode_get_named_child_node(port_fwnode, "fixed-link"));

I'm not too sure about this last one. You only use fixed-link when
connecting to an Ethernet switch. I doubt anybody will try ACPI and a
switch. It has been agreed, ACPI is for simple hardware, and you need
to use DT for advanced hardware configurations.

What is your use case for fixed-link?

Regardless of the "simple hardware" definition or whether DSA + ACPI
feasibility, you can still have e.g. the switch left in "unmanaged"
mode (or whatever the firmware configures), connected via fixed-link
to the MAC. The same effect as booting with DT, but not loading the
DSA/switch driver - the "CPU port" can be used as a normal netdev
interface.

You can do this, but i would not recommend it. Without having STP,
your network is going to be vulnerable to broadcast storms killing
your network.

I'd also prefer to have all 3 major interface types supported in
phylink, explicitly checked in the driver - it has not been supported
yet, but can be in the future, so let's have them covered in the
backward compatibility check.

Maybe i'm not understanding this correctly, but isn't this condition
enforcing there must be a fixed link in order to use the new ACPI
binding? But i expect most boards never need a fixed-link, it is
optional after all.

	 Andrew