[PATCH net] net: ixp4xx_eth: handle probe deferral from of_get_mac_address()

[PATCH net] net: ixp4xx_eth: handle probe deferral from of_get_mac_address()

[Rosen Penev]

ixp4xx_of_get_platdata() returns NULL on failure, which the probe converts
to -ENODEV, discarding the real cause of the failure. Return ERR_PTR()
with the appropriate error so callers can distinguish cases such as
missing DT properties (-EINVAL) and, importantly, handle -EPROBE_DEFER
from of_get_mac_address().

Fixes: 95aafe911db6 ("net: ethernet: ixp4xx: Support device tree probing")
Assisted-by: opencode:hy3-free
Signed-off-by: Rosen Penev <rosenp@gmail.com>
---
 drivers/net/ethernet/xscale/ixp4xx_eth.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/xscale/ixp4xx_eth.c b/drivers/net/ethernet/xscale/ixp4xx_eth.c
index b0faa0f1780d..b8889f87ca03 100644
--- a/drivers/net/ethernet/xscale/ixp4xx_eth.c
+++ b/drivers/net/ethernet/xscale/ixp4xx_eth.c
@@ -1444,13 +1444,13 @@ static struct eth_plat_info *ixp4xx_of_get_platdata(struct device *dev)
 
 	plat = devm_kzalloc(dev, sizeof(*plat), GFP_KERNEL);
 	if (!plat)
-		return NULL;
+		return ERR_PTR(-ENOMEM);
 
 	ret = of_parse_phandle_with_fixed_args(np, "intel,npe-handle", 1, 0,
 					       &npe_spec);
 	if (ret) {
 		dev_err(dev, "no NPE engine specified\n");
-		return NULL;
+		return ERR_PTR(-EINVAL);
 	}
 	/* NPE ID 0x00, 0x10, 0x20... */
 	plat->npe = (npe_spec.args[0] << 4);
@@ -1468,7 +1468,7 @@ static struct eth_plat_info *ixp4xx_of_get_platdata(struct device *dev)
 					       &queue_spec);
 	if (ret) {
 		dev_err(dev, "no rx queue phandle\n");
-		return NULL;
+		return ERR_PTR(-EINVAL);
 	}
 	plat->rxq = queue_spec.args[0];
 
@@ -1477,11 +1477,13 @@ static struct eth_plat_info *ixp4xx_of_get_platdata(struct device *dev)
 					       &queue_spec);
 	if (ret) {
 		dev_err(dev, "no txready queue phandle\n");
-		return NULL;
+		return ERR_PTR(-EINVAL);
 	}
 	plat->txreadyq = queue_spec.args[0];
 
 	ret = of_get_mac_address(np, mac);
+	if (ret == -EPROBE_DEFER)
+		return ERR_PTR(ret);
 	if (!ret) {
 		dev_info(dev, "Setting macaddr from DT %pM\n", mac);
 		memcpy(plat->hwaddr, mac, ETH_ALEN);
@@ -1501,8 +1503,8 @@ static int ixp4xx_eth_probe(struct platform_device *pdev)
 	int err;
 
 	plat = ixp4xx_of_get_platdata(dev);
-	if (!plat)
-		return -ENODEV;
+	if (IS_ERR(plat))
+		return PTR_ERR(plat);
 
 	if (!(ndev = devm_alloc_etherdev(dev, sizeof(struct port))))
 		return -ENOMEM;
-- 
2.55.0

Re: [PATCH net] net: ixp4xx_eth: handle probe deferral from of_get_mac_address()

[Linus Walleij]

On Thu, Aug 27, 2026 at 3:36 AM Rosen Penev <rosenp@gmail.com> wrote:

> ixp4xx_of_get_platdata() returns NULL on failure, which the probe converts
> to -ENODEV, discarding the real cause of the failure. Return ERR_PTR()
> with the appropriate error so callers can distinguish cases such as
> missing DT properties (-EINVAL) and, importantly, handle -EPROBE_DEFER
> from of_get_mac_address().
>
> Fixes: 95aafe911db6 ("net: ethernet: ixp4xx: Support device tree probing")
> Assisted-by: opencode:hy3-free
> Signed-off-by: Rosen Penev <rosenp@gmail.com>

Reviewed-by: Linus Walleij <linusw@kernel.org>

Thanks Rosen!
Linus Walleij

Re: [PATCH net] net: ixp4xx_eth: handle probe deferral from of_get_mac_address()

[Jakub Kicinski]

On Wed, 26 Aug 2026 18:36:34 -0700 Rosen Penev wrote:
> ixp4xx_of_get_platdata() returns NULL on failure, which the probe converts
> to -ENODEV, discarding the real cause of the failure. Return ERR_PTR()
> with the appropriate error so callers can distinguish cases such as
> missing DT properties (-EINVAL) and, importantly, handle -EPROBE_DEFER
> from of_get_mac_address().

Doesn't seem like this should be treated as a fix, AI says:

Can a device tree that actually reaches this new code path pass
dtbs_check? -EPROBE_DEFER out of of_get_mac_address() only comes from
the nvmem lookup:

net/core/of_net.c:of_get_mac_address() {
	...
	return of_get_mac_address_nvmem(np, addr);
}

which needs nvmem-cells and nvmem-cell-names = "mac-address" in the node.
Documentation/devicetree/bindings/net/intel,ixp4xx-ethernet.yaml ends with
additionalProperties: false and pulls in ethernet-controller.yaml only via
allOf/$ref, re-listing just phy-mode, phy-handle and fixed-link. It never
lists nvmem-cells, nvmem-cell-names, mac-address or local-mac-address.

Should the binding gain those properties (or switch to
unevaluatedProperties: false) so the configuration this patch handles
can be described?


And if this goes to net-next there's a bunch of pre-existing
bugs in the area which will conflict, so frankly let's just leave 
this be.