On 10/23/19 4:10 PM, Russell King - ARM Linux admin wrote: > On Wed, Oct 23, 2019 at 03:32:20PM +0300, Ioana Ciornei wrote: >> The appropriate assert on the rtnl lock is not present in phylink's >> connect functions which makes unusual calls to them not to be catched. >> Add the appropriate ASSERT_RTNL(). > > As I previously replied, this is not necessary. It is safe to attach > PHYs _prior_ to the netdev being registered without taking the rtnl > lock, just like phylib's phy_connect()/phy_attach() are safe. >
Yeah.. I was trying to catch other errors like in the case of dpaa2-eth where we also connect/disconnect when we are notified by the firmware of a DPMAC disconnect. I now understand why the assert is not always needed in the _connect(). Ioana >> >> Signed-off-by: Ioana Ciornei <[email protected]> >> --- >> drivers/net/phy/phylink.c | 4 ++++ >> 1 file changed, 4 insertions(+) >> >> diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c >> index be7a2c0fa59b..d0aa0c861b2d 100644 >> --- a/drivers/net/phy/phylink.c >> +++ b/drivers/net/phy/phylink.c >> @@ -786,6 +786,8 @@ static int __phylink_connect_phy(struct phylink *pl, >> struct phy_device *phy, >> */ >> int phylink_connect_phy(struct phylink *pl, struct phy_device *phy) >> { >> + ASSERT_RTNL(); >> + >> /* Use PHY device/driver interface */ >> if (pl->link_interface == PHY_INTERFACE_MODE_NA) { >> pl->link_interface = phy->interface; >> @@ -815,6 +817,8 @@ int phylink_of_phy_connect(struct phylink *pl, struct >> device_node *dn, >> struct phy_device *phy_dev; >> int ret; >> >> + ASSERT_RTNL(); >> + >> /* Fixed links and 802.3z are handled without needing a PHY */ >> if (pl->link_an_mode == MLO_AN_FIXED || >> (pl->link_an_mode == MLO_AN_INBAND && >> -- >> 1.9.1 >> >> >
