It is mandatory for a software to issue a reset upon modifying RGMII
Receive Timing Control and RGMII Transmit Timing Control bit fields of MAC
Specific Control register 2 (page 2, register 21) otherwise the changes
won't be perceived by the PHY (the same is applicable for a lot of other
registers). Not setting the RGMII delays on the platforms that imply it'
being done on the PHY side will consequently cause the traffic loss. We
discovered that the denoted soft-reset is missing in the
m88e1121_config_aneg() method for the case if the RGMII delays are
modified but the MDIx polarity isn't changed or the auto-negotiation is
left enabled, thus causing the traffic loss on our platform with Marvell
Alaska 88E1510 installed. Let's fix that by issuing the soft-reset if the
delays have been actually set in the m88e1121_config_aneg_rgmii_delays()
method.

Fixes: d6ab93364734 ("net: phy: marvell: Avoid unnecessary soft reset")
Signed-off-by: Pavel Parkhomenko <Pavel.Parkhomenko@baikalelectronics.ru>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Reviewed-by: Serge Semin <fancer.lancer@gmail.com>
Cc: stable@vger.kernel.org

---

Link: https://lore.kernel.org/netdev/96759fee7240fd095cb9cc1f6eaf2d9113b57cf0.camel@baikalelectronics.ru/
Changelog v2:
- Add "net" suffix into the PATCH-clause of the subject.
- Cc the patch to the stable tree list.
- Rebase onto the latset netdev/net branch with the top commit 59085208e4a2
("net: mscc: ocelot: fix all IP traffic getting trapped to CPU with PTP over
IP")
---
 drivers/net/phy/marvell.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/net/phy/marvell.c b/drivers/net/phy/marvell.c
index fa71fb7a66b5..e2fd1252be48 100644
--- a/drivers/net/phy/marvell.c
+++ b/drivers/net/phy/marvell.c
@@ -553,9 +553,9 @@ static int m88e1121_config_aneg_rgmii_delays(struct phy_device *phydev)
 	else
 		mscr = 0;
 
-	return phy_modify_paged(phydev, MII_MARVELL_MSCR_PAGE,
-				MII_88E1121_PHY_MSCR_REG,
-				MII_88E1121_PHY_MSCR_DELAY_MASK, mscr);
+	return phy_modify_paged_changed(phydev, MII_MARVELL_MSCR_PAGE,
+					MII_88E1121_PHY_MSCR_REG,
+					MII_88E1121_PHY_MSCR_DELAY_MASK, mscr);
 }
 
 static int m88e1121_config_aneg(struct phy_device *phydev)
@@ -569,11 +569,13 @@ static int m88e1121_config_aneg(struct phy_device *phydev)
 			return err;
 	}
 
+	changed = err;
+
 	err = marvell_set_polarity(phydev, phydev->mdix_ctrl);
 	if (err < 0)
 		return err;
 
-	changed = err;
+	changed |= err;
 
 	err = genphy_config_aneg(phydev);
 	if (err < 0)
-- 
2.32.0

Hello:

This patch was applied to netdev/net.git (master)
by Jakub Kicinski <kuba@kernel.org>:

On Sat, 5 Feb 2022 23:39:32 +0300 you wrote:
> It is mandatory for a software to issue a reset upon modifying RGMII
> Receive Timing Control and RGMII Transmit Timing Control bit fields of MAC
> Specific Control register 2 (page 2, register 21) otherwise the changes
> won't be perceived by the PHY (the same is applicable for a lot of other
> registers). Not setting the RGMII delays on the platforms that imply it'
> being done on the PHY side will consequently cause the traffic loss. We
> discovered that the denoted soft-reset is missing in the
> m88e1121_config_aneg() method for the case if the RGMII delays are
> modified but the MDIx polarity isn't changed or the auto-negotiation is
> left enabled, thus causing the traffic loss on our platform with Marvell
> Alaska 88E1510 installed. Let's fix that by issuing the soft-reset if the
> delays have been actually set in the m88e1121_config_aneg_rgmii_delays()
> method.
> 
> [...]

Here is the summary with links:
  - [net,v2] net: phy: marvell: Fix RGMII Tx/Rx delays setting in 88e1121-compatible PHYs
    https://git.kernel.org/netdev/net/c/fe4f57bf7b58

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html