[PATCH net-next v2 0/2] dpaa2-eth: devlink port number from the DPMAC

[PATCH net-next v2 0/2] dpaa2-eth: devlink port number from the DPMAC

[Vincent Jardin via B4 Relay]

dpaa2-eth registers every DPNI as a physical devlink port without a
port number, so all the DPAA2 interfaces report the same
phys_port_name "p0", including the DPNIs that have no DPMAC behind
them.

Patch 1 uses the DPMAC id as the physical port number. The attributes
of a registered devlink port cannot change, so when the DPNI endpoint
changes at runtime a warning says the number is stale until the DPNI
is rebound.

Patch 2 registers the DPNIs without a DPMAC as virtual ports. They
have no phys_port_name, so "p<dpmac id>" is unique.

With both patches, udev can name an interface using its DPMAC, for
instance:

  SUBSYSTEM=="net", ACTION=="add", DRIVERS=="fsl_dpaa2_eth", \
    ATTR{phys_port_name}=="p3", NAME="dpmac3"

Signed-off-by: Vincent Jardin <vjardin@free.fr>
---
Changes in v2:
- 1/2: shorter warning (Ioana Ciornei)
- 1/2: keep the registered port number in priv under mac_lock, so avoid
  IRQ race with the port registration at probe (Sashiko)
- 1/2: the port check moved to 1/2 from 2/2 (Ioana Ciornei)
- 2/2: new subject, commit log updates (Ioana Ciornei, Sashiko)
- rebased on net-next
- Link to v1: https://lore.kernel.org/r/20260923-for-upstream-dpaa2_ports-v1-0-b59c865edd00@free.fr

---
Vincent Jardin (2):
      dpaa2-eth: use the DPMAC id as the devlink physical port number
      dpaa2-eth: mark DPNIs without a DPMAC as virtual devlink ports

 .../ethernet/freescale/dpaa2/dpaa2-eth-devlink.c   | 45 +++++++++++++++++++++-
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c   |  2 +
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h   |  3 ++
 3 files changed, 48 insertions(+), 2 deletions(-)
---
base-commit: 014d795c73837ea2339a4ea8e8f82c6e959b845d
change-id: 20260923-for-upstream-dpaa2_ports-04087b33bb8f

Best regards,
-- 
Vincent Jardin <vjardin@free.fr>

[PATCH net-next v2 1/2] dpaa2-eth: use the DPMAC id as the devlink physical port number

[Vincent Jardin via B4 Relay]

From: Vincent Jardin <vjardin@free.fr>

dpaa2-eth registers every DPNI as a devlink port of flavour physical
without setting a port number. All DPAA2 interfaces report the same
phys_port_name, p0 (see devlink port show).

The DPNI endpoint can change at runtime while the attributes of a
registered devlink port cannot, so log a warning when it happens.

Another benefit: udev rules can match ATTR{phys_port_name} to name an
interface based on its DPMAC.

Signed-off-by: Vincent Jardin <vjardin@free.fr>
---
 .../ethernet/freescale/dpaa2/dpaa2-eth-devlink.c   | 40 +++++++++++++++++++++-
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c   |  2 ++
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h   |  3 ++
 3 files changed, 44 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c
index 8775c931106b..4c0a04c2d458 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c
@@ -213,20 +213,58 @@ void dpaa2_eth_dl_unregister(struct dpaa2_eth_priv *priv)
 	devlink_unregister(priv->devlink);
 }
 
+static u32 dpaa2_eth_dl_port_number(struct dpaa2_eth_priv *priv)
+{
+	lockdep_assert_held(&priv->mac_lock);
+
+	return dpaa2_eth_has_mac(priv) ? priv->mac->mc_dev->obj_desc.id : 0;
+}
+
 int dpaa2_eth_dl_port_add(struct dpaa2_eth_priv *priv)
 {
 	struct devlink_port *devlink_port = &priv->devlink_port;
 	struct devlink_port_attrs attrs = {};
+	u32 port_number;
+	int err;
+
+	mutex_lock(&priv->mac_lock);
+	port_number = dpaa2_eth_dl_port_number(priv);
+	priv->dl_port_number = port_number;
+	priv->dl_port_valid = true;
+	mutex_unlock(&priv->mac_lock);
 
 	attrs.flavour = DEVLINK_PORT_FLAVOUR_PHYSICAL;
+	attrs.phys.port_number = port_number;
 	devlink_port_attrs_set(devlink_port, &attrs);
-	return devlink_port_register(priv->devlink, devlink_port, 0);
+	err = devlink_port_register(priv->devlink, devlink_port, 0);
+	if (err) {
+		mutex_lock(&priv->mac_lock);
+		priv->dl_port_valid = false;
+		mutex_unlock(&priv->mac_lock);
+	}
+
+	return err;
+}
+
+void dpaa2_eth_dl_port_check(struct dpaa2_eth_priv *priv)
+{
+	mutex_lock(&priv->mac_lock);
+	if (priv->dl_port_valid &&
+	    priv->dl_port_number != dpaa2_eth_dl_port_number(priv))
+		netdev_warn(priv->net_dev,
+			    "devlink port number %u is stale, rebind to update\n",
+			    priv->dl_port_number);
+	mutex_unlock(&priv->mac_lock);
 }
 
 void dpaa2_eth_dl_port_del(struct dpaa2_eth_priv *priv)
 {
 	struct devlink_port *devlink_port = &priv->devlink_port;
 
+	mutex_lock(&priv->mac_lock);
+	priv->dl_port_valid = false;
+	mutex_unlock(&priv->mac_lock);
+
 	devlink_port_unregister(devlink_port);
 }
 
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
index 6f1046c9cc51..def2d5c0e7f7 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c
@@ -4771,6 +4771,8 @@ static irqreturn_t dpni_irq0_handler_thread(int irq_num, void *arg)
 			dpaa2_eth_disconnect_mac(priv);
 		else
 			dpaa2_eth_connect_mac(priv);
+
+		dpaa2_eth_dl_port_check(priv);
 	}
 
 	return IRQ_HANDLED;
diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
index 834cba8c3a41..3e0acd15549e 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.h
@@ -632,6 +632,8 @@ struct dpaa2_eth_priv {
 	struct devlink *devlink;
 	struct dpaa2_eth_trap_data *trap_data;
 	struct devlink_port devlink_port;
+	u32 dl_port_number;
+	bool dl_port_valid;
 
 	u32 rx_copybreak;
 
@@ -802,6 +804,7 @@ void dpaa2_eth_dl_unregister(struct dpaa2_eth_priv *priv);
 
 int dpaa2_eth_dl_port_add(struct dpaa2_eth_priv *priv);
 void dpaa2_eth_dl_port_del(struct dpaa2_eth_priv *priv);
+void dpaa2_eth_dl_port_check(struct dpaa2_eth_priv *priv);
 
 int dpaa2_eth_dl_traps_register(struct dpaa2_eth_priv *priv);
 void dpaa2_eth_dl_traps_unregister(struct dpaa2_eth_priv *priv);

-- 
2.43.0

Re: [PATCH net-next v2 1/2] dpaa2-eth: use the DPMAC id as the devlink physical port number

[Simon Horman]

On Mon, Sep 28, 2026 at 01:22:45PM +0200, Vincent Jardin via B4 Relay wrote:
> From: Vincent Jardin <vjardin@free.fr>
> 
> dpaa2-eth registers every DPNI as a devlink port of flavour physical
> without setting a port number. All DPAA2 interfaces report the same
> phys_port_name, p0 (see devlink port show).
> 
> The DPNI endpoint can change at runtime while the attributes of a
> registered devlink port cannot, so log a warning when it happens.
> 
> Another benefit: udev rules can match ATTR{phys_port_name} to name an
> interface based on its DPMAC.
> 
> Signed-off-by: Vincent Jardin <vjardin@free.fr>

Reviewed-by: Simon Horman <horms@kernel.org>

[PATCH net-next v2 2/2] dpaa2-eth: mark DPNIs without a DPMAC as virtual devlink ports

[Vincent Jardin via B4 Relay]

From: Vincent Jardin <vjardin@free.fr>

A DPNI whose endpoint is not a DPMAC has no physical connector: it
may have no endpoint at all, or be connected to another DPNI or to a
DPSW port. Until now, such a DPNI was still registered as a physical
devlink port with number 0, so userspace saw a physical port that does
not exist.

Register these DPNIs as virtual devlink ports instead. A virtual port
has no phys_port_name, so only the DPNIs that serve a DPMAC at probe
time report one, "p<dpmac id>", and those names do not collide.

Signed-off-by: Vincent Jardin <vjardin@free.fr>
---
 drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c
index 4c0a04c2d458..6defc00199bf 100644
--- a/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c
+++ b/drivers/net/ethernet/freescale/dpaa2/dpaa2-eth-devlink.c
@@ -233,7 +233,10 @@ int dpaa2_eth_dl_port_add(struct dpaa2_eth_priv *priv)
 	priv->dl_port_valid = true;
 	mutex_unlock(&priv->mac_lock);
 
-	attrs.flavour = DEVLINK_PORT_FLAVOUR_PHYSICAL;
+	if (port_number)
+		attrs.flavour = DEVLINK_PORT_FLAVOUR_PHYSICAL;
+	else
+		attrs.flavour = DEVLINK_PORT_FLAVOUR_VIRTUAL;
 	attrs.phys.port_number = port_number;
 	devlink_port_attrs_set(devlink_port, &attrs);
 	err = devlink_port_register(priv->devlink, devlink_port, 0);

-- 
2.43.0

Re: [PATCH net-next v2 2/2] dpaa2-eth: mark DPNIs without a DPMAC as virtual devlink ports

[Simon Horman]

On Mon, Sep 28, 2026 at 01:22:46PM +0200, Vincent Jardin via B4 Relay wrote:
> From: Vincent Jardin <vjardin@free.fr>
> 
> A DPNI whose endpoint is not a DPMAC has no physical connector: it
> may have no endpoint at all, or be connected to another DPNI or to a
> DPSW port. Until now, such a DPNI was still registered as a physical
> devlink port with number 0, so userspace saw a physical port that does
> not exist.
> 
> Register these DPNIs as virtual devlink ports instead. A virtual port
> has no phys_port_name, so only the DPNIs that serve a DPMAC at probe
> time report one, "p<dpmac id>", and those names do not collide.
> 
> Signed-off-by: Vincent Jardin <vjardin@free.fr>

Reviewed-by: Simon Horman <horms@kernel.org>

Re: [PATCH net-next v2 0/2] dpaa2-eth: devlink port number from the DPMAC

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Mon, 28 Sep 2026 13:22:44 +0200 you wrote:
> dpaa2-eth registers every DPNI as a physical devlink port without a
> port number, so all the DPAA2 interfaces report the same
> phys_port_name "p0", including the DPNIs that have no DPMAC behind
> them.
> 
> Patch 1 uses the DPMAC id as the physical port number. The attributes
> of a registered devlink port cannot change, so when the DPNI endpoint
> changes at runtime a warning says the number is stale until the DPNI
> is rebound.
> 
> [...]

Here is the summary with links:
  - [net-next,v2,1/2] dpaa2-eth: use the DPMAC id as the devlink physical port number
    https://git.kernel.org/netdev/net-next/c/526025d552ba
  - [net-next,v2,2/2] dpaa2-eth: mark DPNIs without a DPMAC as virtual devlink ports
    https://git.kernel.org/netdev/net-next/c/aed5a323ec37

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