net: phy: fix too strict check in phy_start_aneg

When adding checks to detect wrong usage of the phylib API we added
a check to phy_start_aneg() which is too strict. If the phylib
state machine is in state PHY_HALTED we should allow reconfiguring
and restarting aneg, and just don't touch the state.

Fixes: 2b3e88e ("net: phy: improve phy state checking")
Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/phy/phy.c

@@ -543,13 +543,6 @@ int phy_start_aneg(struct phy_device *phydev)
 
 	mutex_lock(&phydev->lock);
 
-	if (!__phy_is_started(phydev)) {
-		WARN(1, "called from state %s\n",
-		     phy_state_to_str(phydev->state));
-		err = -EBUSY;
-		goto out_unlock;
-	}
-
 	if (AUTONEG_DISABLE == phydev->autoneg)
 		phy_sanitize_settings(phydev);
 
@@ -560,11 +553,13 @@ int phy_start_aneg(struct phy_device *phydev)
 	if (err < 0)
 		goto out_unlock;
 
-	if (phydev->autoneg == AUTONEG_ENABLE) {
-		err = phy_check_link_status(phydev);
-	} else {
-		phydev->state = PHY_FORCING;
-		phydev->link_timeout = PHY_FORCE_TIMEOUT;
+	if (__phy_is_started(phydev)) {
+		if (phydev->autoneg == AUTONEG_ENABLE) {
+			err = phy_check_link_status(phydev);
+		} else {
+			phydev->state = PHY_FORCING;
+			phydev->link_timeout = PHY_FORCE_TIMEOUT;
+		}
 	}
 
 out_unlock: