Re: [PATCH net-next 2/2] cxgb4: Add support for ndo_get_vf_config

On Wednesday, August 08/24/16, 2016 at 07:15:49 +0000, Yuval Mintz wrote:

quoted

+static void fill_vf_station_mac_addr(struct adapter *adap)
+{
+	unsigned int i;
+	u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
+	int err;
+	u8 *na;
+	u16 a, b;
+
+	err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
+	if (!err) {
+		na = adap->params.vpd.na;
+		for (i = 0; i < ETH_ALEN; i++)
+			hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
+				      hex2val(na[2 * i + 1]));
+		a = (hw_addr[0] << 8) | hw_addr[1];
+		b = (hw_addr[1] << 8) | hw_addr[2];
+		a ^= b;
+		a |= 0x0200;    /* locally assigned Ethernet MAC address */
+		a &= ~0x0100;   /* not a multicast Ethernet MAC address */
+		macaddr[0] = a >> 8;
+		macaddr[1] = a & 0xff;
+
+		for (i = 2; i < 5; i++)
+			macaddr[i] = hw_addr[i + 1];
+
+		for (i = 0; i < adap->num_vfs; i++) {
+			macaddr[5] = adap->pf * 16 + i;
+			ether_addr_copy(adap->vfinfo[i].vf_mac_addr,
macaddr);
+		}
+	}
+}

That's some... magical magic? :-)
But I couldn't see anywhere in the patch where this MAC is propagated
to the VF, only to the new NDO(). Care to explain how does the VF learn it?
	

The same logic is used by the firmware to generate the MAC address for the VF.
If MAC address isn't provided through IFLA_VF_MAC. This function is only used
to populate the stationary MAC's for the VF, when user hasn't provided one.

-Hari