[PATCH net 0/2] Two phylink pause fixes

[PATCH net 0/2] Two phylink pause fixes

[Russell King - ARM Linux admin]

While testing, I discovered two issues with ethtool -A with phylink.
First, if there is a PHY bound to the network device, we hit a
deadlock when phylib tries to notify us of the link changing as a
result of triggering a renegotiation.

Second, when we are manually forcing the pause settings, and there
is no renegotiation triggered, we do not update the MAC via the new
mac_link_up approach.

These two patches solve both problems, and will need to be backported
to v5.7; they do not apply cleanly there due to the introduction of
PCS in the v5.8 merge window.

 drivers/net/phy/phylink.c | 45 ++++++++++++++++++++++++++++++++-------------
 1 file changed, 32 insertions(+), 13 deletions(-)

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!

[PATCH net 1/2] net: phylink: fix ethtool -A with attached PHYs

[Russell King]

Fix a phylink's ethtool set_pauseparam support deadlock caused by phylib
interacting with phylink: we must not hold the state lock while calling
phylib functions that may call into phylink_phy_change().

Fixes: f904f15ea9b5 ("net: phylink: allow ethtool -A to change flow control advertisement")
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
---
 drivers/net/phy/phylink.c | 20 +++++++++++---------
 1 file changed, 11 insertions(+), 9 deletions(-)

diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c
index 0ab65fb75258..453f9a399bb1 100644
--- a/drivers/net/phy/phylink.c
+++ b/drivers/net/phy/phylink.c
@@ -1502,18 +1502,20 @@ int phylink_ethtool_set_pauseparam(struct phylink *pl,
 	linkmode_set_pause(config->advertising, pause->tx_pause,
 			   pause->rx_pause);
 
-	/* If we have a PHY, phylib will call our link state function if the
-	 * mode has changed, which will trigger a resolve and update the MAC
-	 * configuration.
+	if (!pl->phydev && !test_bit(PHYLINK_DISABLE_STOPPED,
+				     &pl->phylink_disable_state))
+		phylink_pcs_config(pl, true, &pl->link_config);
+
+	mutex_unlock(&pl->state_mutex);
+
+	/* If we have a PHY, a change of the pause frame advertisement will
+	 * cause phylib to renegotiate (if AN is enabled) which will in turn
+	 * call our phylink_phy_change() and trigger a resolve.  Note that
+	 * we can't hold our state mutex while calling phy_set_asym_pause().
 	 */
-	if (pl->phydev) {
+	if (pl->phydev)
 		phy_set_asym_pause(pl->phydev, pause->rx_pause,
 				   pause->tx_pause);
-	} else if (!test_bit(PHYLINK_DISABLE_STOPPED,
-			     &pl->phylink_disable_state)) {
-		phylink_pcs_config(pl, true, &pl->link_config);
-	}
-	mutex_unlock(&pl->state_mutex);
 
 	return 0;
 }
-- 
2.20.1

[PATCH net 2/2] net: phylink: ensure manual pause mode configuration takes effect

[Russell King]

We have been relying on link events and mac_config() when the manual
pause modes are changed.  With recent developments, such as moving
the programming of link state to mac_link_up(), this no longer works.

To ensure that we update the MAC, we must generate a link-down followed
by a link-up event; we can do that by setting mac_link_dropped and
triggering a resolve.

Fixes: 91a208f2185a ("net: phylink: propagate resolved link config via mac_link_up()")
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
---
 drivers/net/phy/phylink.c | 27 ++++++++++++++++++++++-----
 1 file changed, 22 insertions(+), 5 deletions(-)

diff --git a/drivers/net/phy/phylink.c b/drivers/net/phy/phylink.c
index 453f9a399bb1..3b7c70e6c5dd 100644
--- a/drivers/net/phy/phylink.c
+++ b/drivers/net/phy/phylink.c
@@ -1463,6 +1463,8 @@ int phylink_ethtool_set_pauseparam(struct phylink *pl,
 				   struct ethtool_pauseparam *pause)
 {
 	struct phylink_link_state *config = &pl->link_config;
+	bool manual_changed;
+	int pause_state;
 
 	ASSERT_RTNL();
 
@@ -1477,15 +1479,15 @@ int phylink_ethtool_set_pauseparam(struct phylink *pl,
 	    !pause->autoneg && pause->rx_pause != pause->tx_pause)
 		return -EINVAL;
 
-	mutex_lock(&pl->state_mutex);
-	config->pause = 0;
+	pause_state = 0;
 	if (pause->autoneg)
-		config->pause |= MLO_PAUSE_AN;
+		pause_state |= MLO_PAUSE_AN;
 	if (pause->rx_pause)
-		config->pause |= MLO_PAUSE_RX;
+		pause_state |= MLO_PAUSE_RX;
 	if (pause->tx_pause)
-		config->pause |= MLO_PAUSE_TX;
+		pause_state |= MLO_PAUSE_TX;
 
+	mutex_lock(&pl->state_mutex);
 	/*
 	 * See the comments for linkmode_set_pause(), wrt the deficiencies
 	 * with the current implementation.  A solution to this issue would
@@ -1502,6 +1504,12 @@ int phylink_ethtool_set_pauseparam(struct phylink *pl,
 	linkmode_set_pause(config->advertising, pause->tx_pause,
 			   pause->rx_pause);
 
+	manual_changed = (config->pause ^ pause_state) & MLO_PAUSE_AN ||
+			 (!(pause_state & MLO_PAUSE_AN) &&
+			   (config->pause ^ pause_state) & MLO_PAUSE_TXRX_MASK);
+
+	config->pause = pause_state;
+
 	if (!pl->phydev && !test_bit(PHYLINK_DISABLE_STOPPED,
 				     &pl->phylink_disable_state))
 		phylink_pcs_config(pl, true, &pl->link_config);
@@ -1517,6 +1525,15 @@ int phylink_ethtool_set_pauseparam(struct phylink *pl,
 		phy_set_asym_pause(pl->phydev, pause->rx_pause,
 				   pause->tx_pause);
 
+	/* If the manual pause settings changed, make sure we trigger a
+	 * resolve to update their state; we can not guarantee that the
+	 * link will cycle.
+	 */
+	if (manual_changed) {
+		pl->mac_link_dropped = true;
+		phylink_run_resolve(pl);
+	}
+
 	return 0;
 }
 EXPORT_SYMBOL_GPL(phylink_ethtool_set_pauseparam);
-- 
2.20.1

Re: [PATCH net 0/2] Two phylink pause fixes

[David Miller]

From: Russell King - ARM Linux admin <linux@armlinux.org.uk>
Date: Tue, 23 Jun 2020 17:46:42 +0100

> While testing, I discovered two issues with ethtool -A with phylink.
> First, if there is a PHY bound to the network device, we hit a
> deadlock when phylib tries to notify us of the link changing as a
> result of triggering a renegotiation.
> 
> Second, when we are manually forcing the pause settings, and there
> is no renegotiation triggered, we do not update the MAC via the new
> mac_link_up approach.
> 
> These two patches solve both problems, and will need to be backported
> to v5.7; they do not apply cleanly there due to the introduction of
> PCS in the v5.8 merge window.

Series applied, thank you.