Comment # 39 on bug 1191795 from Oliver Neukum

(In reply to Frank Kr���ger from comment #38)
> This is like looking for a needle in a haystack.

Well, this bug report is not connected to an error condition in the kernel.
Finding things that do not happen is always the hardest, in particular if it
happens deep in the hard ware.

Maybe I can shed light on the code in question:

static int rtl8169_runtime_suspend(struct device *device)
{
        struct rtl8169_private *tp = dev_get_drvdata(device);

        if (!tp->TxDescArray) {
                netif_device_detach(tp->dev);
                return 0;
        }

        rtnl_lock();
        __rtl8169_set_wol(tp, WAKE_PHY);
        rtl8169_net_suspend(tp);
        rtnl_unlock();

        return 0;
}

That is the whole logic. The key line is:

__rtl8169_set_wol(tp, WAKE_PHY);

You see that the device is setup to be woken for activity of the PHY.