On Tue, May 05, 2020 at 03:22:03PM +0200, Andrew Lunn wrote:

quoted

quoted

+int ethnl_cable_test_fault_length(struct phy_device *phydev, u8 pair, u32 cm)
+{
+	struct nlattr *nest;
+	int ret = -EMSGSIZE;
+
+	nest = nla_nest_start(phydev->skb,
+			      ETHTOOL_A_CABLE_TEST_NTF_FAULT_LENGTH);
+	if (!nest)
+		return -EMSGSIZE;
+
+	if (nla_put_u8(phydev->skb, ETHTOOL_A_CABLE_FAULT_LENGTH_PAIR, pair))
+		goto err;
+	if (nla_put_u16(phydev->skb, ETHTOOL_A_CABLE_FAULT_LENGTH_CM, cm))
+		goto err;

This should be nla_put_u32().

Yes. I think i messed up a rebase merge conflict somewhere. I'm also
surprised user space is not complaining.

There is no difference on little endian architectures as nla_put_*()
helpers all call __nla_reserve() which fills the padding with zero
bytes. IIRC there was a case where wrong attribute type had been used
for quite long without anyone noticing.

Michal