[PATCH net-next v5 3/4] net: phy: Allow pre-declaration of MDIO devices

+/**
+ * mdio_register_board_info - register MDIO devices for a given board
+ * @info: array of devices descriptors
+ * @n: number of descriptors provided
+ * Context: can sleep
+ *
+ * The board info passed can be marked with __initdata but be pointers
+ * such as platform_data etc. are copied as-is
+ */
+int mdiobus_register_board_info(const struct mdio_board_info *info,
+				unsigned int n)
+{
+	struct mdio_board_entry *be;
+	unsigned int i;
+
+	be = kcalloc(n, sizeof(*be), GFP_KERNEL);
+	if (!be)
+		return -ENOMEM;
+
+	for (i = 0; i < n; i++, be++, info++) {
+		memcpy(&be->board_info, info, sizeof(*info));
+		mutex_lock(&mdio_board_lock);
+		list_add_tail(&be->list, &mdio_board_list);
+		mutex_unlock(&mdio_board_lock);
+	}
+
+	return 0;

Hi Florian

I've recently been playing with a hot-pluggable SPI bus controller. It
is a USB device, hence can come and go. On the SPI bus i have an
SRAM. On order to instantiate the MTD device, i need SPI board info. I
cannot add the board info until after the SPI bus master appears,
since i need to know its ID to fill in the board info. At the moment,
i have udev script which when the SPI bus master appears loads a
little kernel module which registers the board info.

Such a scheme will not work here. You need to iterate the list of MDIO
devices at the end of mdiobus_register_board_info() to see if the just
registered board info applies to any existing MDIO bus.

I don't think we yet have any hardware which would do this. But there
have been patches to one of the USB-Ethernet dongles to allow it run
without a PHY. My guess is, to allow an SFP module. But it is not too
big a step for somebody to make a USB attached Ethernet switch.

Maybe consider adding this functionality? Also an unregister call?

      Andrew