On Sun, Dec 23, 2018 at 11:43:05AM +0100, Marek Vasut wrote:

On 12/23/18 11:23 AM, Andrew Lunn wrote:

quoted

quoted

quoted

+static int smsc95xx_phy_address(struct usbnet *dev)
+{
+	u32 read_buf;
+	int ret, id1, id2, phyad;
+
+	ret = smsc95xx_read_reg(dev, HW_CFG, &read_buf);
+	if (ret < 0)
+		return ret;
+
+	/* Check if using external PHY, if not, use internal PHY address */
+	if (!(read_buf & HW_CFG_PSEL_))
+		return SMSC95XX_INTERNAL_PHY_ID;
+
+	/*
+	 * Detect external PHY address. Here we probe the MDIO bus from
+	 * the highest address, since some PHYs respond also on address
+	 * zero, which they consider MDIO broadcast address. We really
+	 * want to get their proper address instead though, so we scan
+	 * address zero last.
+	 */
+	for (phyad = 0x1f; phyad >= 0; phyad--) {
+		id1 = smsc95xx_mdio_read(dev->net, phyad, MII_PHYSID1);
+		id2 = smsc95xx_mdio_read(dev->net, phyad, MII_PHYSID2);
+		/* Check for valid response from the PHY */
+		if (id1 > 0 && id2 > 0 && id1 != 0x7fff && id2 != 0xffff)
+			return phyad;
+	}

This would be so much easier if the driver used the core mdio/phy
code. Just set mdio->phy_mask to ~BIT(0) and then use
phy_find_first().

That's in the pipeline, along with PM cleanups, but low prio.

Great. Does using the broadcast address actually cause a problem?  If
not, i would say lets drop this part of the patch until you do the
cleanup.

     Andrew