* [PATCH net-next v4 1/6] net: enetc: Add pseudo-MAC support for ENETCv4 Ports via a fixed-link node
[not found] <cover.1794581338.git.claudiu.manoil@nxp.com>
@ 2026-11-13 15:02 ` Claudiu Manoil
2026-11-17 15:55 ` netdev-bot+sashiko
2026-11-13 15:02 ` [PATCH net-next v4 2/6] net: enetc: Allocate int_vector array dynamically Claudiu Manoil
` (4 subsequent siblings)
5 siblings, 1 reply; 10+ messages in thread
From: Claudiu Manoil @ 2026-11-13 15:02 UTC (permalink / raw)
To: netdev
Cc: s32, Vladimir Oltean, Wei Fang, Clark Wang, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni,
Rob Herring, Saravana Kannan, Russell King, imx, linux-kernel,
devicetree
ENETCv4 has special internal links when connected to the on-chip NETC
switch via internal switch ports, called pseudo-MAC links. These
pseudo-MACs are proprietary, they don't implement any standard IEEE
interface (like MII), and can be modeled as fixed links with the link
speed determined at boot time by the Port PCR[PSPEED] register
configuration.
We also need to be able to probe the ENETCv4 Ports featuring pseudo-MACs
as pure PCI devices, i.e. without any "ethernet" DT node representation.
The typical use case for this consists in a board with NETC connected
via PCI to another host which probes the ENETC. Note that in this
scenario, the pseudo-MAC ENETCs are connected internally to a NETC
switch that is not owned by Linux.
Since such a port has no "ethernet" DT node, its fixed link has to be
synthesized at probe time. Build the fixed-link subnode directly in C
with an of_changeset: of_changeset_create_node() adds a "fixed-link"
node onto the port's own OF node, and of_changeset_add_prop_*() populate
its "speed" and "full-duplex" properties. The changeset is applied with
of_changeset_apply(), so the fixed-link node is spliced onto the ENETC
netdev's fwnode exactly as if it had come from static DT, and phylink
picks it up through dev_fwnode().
The operating speed written into the node is read from PCR[PSPEED] at
probe time and mapped to a phylink-supported value, so the fixed link
reflects the actual boot-time configuration, as long as it's a phylink
supported value. Otherwise, PCR[PSPEED] will be updated to the closest
rounded down supported value.
The pseudo-MAC path is selected when the port has no OF node, or when it
only has the empty PCI-synthesized node (OF_DYNAMIC), which carries no
fixed-link description. Ports described by static DT take the regular
of_get_phy_mode() path instead.
Building the fixed-link node requires the of_changeset API, which is only
available with CONFIG_OF_DYNAMIC, so select it. As OF_DYNAMIC itself
depends on OF, make NXP_ENETC4 depend on OF as well, rather than force
OF_DYNAMIC on without the core OF code being built.
The changeset is reverted and destroyed on teardown and on the probe
error unwind. The node-less pseudo-MAC path has a runtime dependency on
CONFIG_PCI_DYNAMIC_OF_NODES: that is what makes the PCI core synthesize
the per-function OF node (of_pci_make_dev_node()) that the fixed-link
node is attached onto. This is intentionally not a hard Kconfig
dependency, since the ENETC4 driver also serves regular ports described
by static DT; when PCI_DYNAMIC_OF_NODES is absent the PCI function has no
of_node and the pseudo-MAC probe path fails gracefully with a clear
-ENODEV error.
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
---
v4:
- Made NXP_ENETC4 depend on OF, since OF_DYNAMIC (selected for the
of_changeset API) is only available with OF; added to the commit
message. Reported by the kernel test robot on v2 and in review.
v3:
- Fixed the CONFIG_OF_DYNAMIC dependency.
- Refactored enetc4_add_ppm_fixed_link(), as per Wei's feedback.
- Improved the commit message.
v2:
- Switched from a DT overlay blob (of_overlay_fdt_apply() + a .dtso
compiled to .dtbo) to of_changeset_create_node(), per review
feedback. Drops the embedded overlay externs, the Makefile .dtbo
wrapping, and the OF_OVERLAY Kconfig dependency.
- Baked the operating speed directly into the changeset-created
fixed-link node (read once from PCR[PSPEED] at probe time); the
phylink get_fixed_state callback is no longer needed and was removed.
drivers/net/ethernet/freescale/enetc/Kconfig | 2 +
.../net/ethernet/freescale/enetc/enetc4_pf.c | 109 +++++++++++++++++-
.../net/ethernet/freescale/enetc/enetc_pf.h | 2 +
.../freescale/enetc/enetc_pf_common.c | 41 ++++++-
.../freescale/enetc/enetc_pf_common.h | 1 +
5 files changed, 146 insertions(+), 9 deletions(-)
diff --git a/drivers/net/ethernet/freescale/enetc/Kconfig b/drivers/net/ethernet/freescale/enetc/Kconfig
index f425f82a6213..465c913683c8 100644
--- a/drivers/net/ethernet/freescale/enetc/Kconfig
+++ b/drivers/net/ethernet/freescale/enetc/Kconfig
@@ -49,12 +49,14 @@ config NXP_ENETC4
tristate "ENETC4 PF driver"
depends on PTP_1588_CLOCK_OPTIONAL
depends on PCI_MSI
+ depends on OF
select FSL_ENETC_CORE
select FSL_ENETC_MDIO
select NXP_ENETC_PF_COMMON
select NXP_NTMP
select PHYLINK
select DIMLIB
+ select OF_DYNAMIC
help
This driver supports NXP ENETC devices with major revision 4. ENETC is
as the NIC functionality in NETC, it supports virtualization/isolation
diff --git a/drivers/net/ethernet/freescale/enetc/enetc4_pf.c b/drivers/net/ethernet/freescale/enetc/enetc4_pf.c
index 71c971618388..b7e7201e6a71 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc4_pf.c
+++ b/drivers/net/ethernet/freescale/enetc/enetc4_pf.c
@@ -1,8 +1,9 @@
// SPDX-License-Identifier: (GPL-2.0+ OR BSD-3-Clause)
-/* Copyright 2024 NXP */
+/* Copyright 2024, 2026 NXP */
#include <linux/clk.h>
#include <linux/module.h>
+#include <linux/of.h>
#include <linux/of_net.h>
#include <linux/of_platform.h>
#include <linux/unaligned.h>
@@ -932,6 +933,14 @@ static void enetc4_pl_mac_link_down(struct phylink_config *config,
enetc4_mac_tx_graceful_stop(pf);
}
+static void enetc4_get_pcr_speed(struct enetc_hw *hw, int *speed)
+{
+ u32 val = enetc_port_rd(hw, ENETC4_PCR);
+ int pspeed = FIELD_GET(PCR_PSPEED, val);
+
+ *speed = (pspeed + 1) * 10;
+}
+
static const struct phylink_mac_ops enetc_pl_mac_ops = {
.mac_select_pcs = enetc4_pl_mac_select_pcs,
.mac_config = enetc4_pl_mac_config,
@@ -946,23 +955,108 @@ static void enetc4_pci_remove(void *data)
enetc_pci_remove(pdev);
}
+static void enetc4_put_ppm_fixed_link(struct enetc_pf *pf)
+{
+ if (!pf->ppm_ocs)
+ return;
+
+ of_changeset_revert(pf->ppm_ocs);
+ of_changeset_destroy(pf->ppm_ocs);
+ kfree(pf->ppm_ocs);
+ pf->ppm_ocs = NULL;
+}
+
+/* Pseudo-MAC ports have no real PHY; the link is fixed, with the operating
+ * speed set at probe time from PCR[PSPEED]. Describe this with a fixed-link
+ * subnode synthesized at runtime onto the port's own OF node, so phylink
+ * picks it up through dev_fwnode() exactly as if it came from static DT.
+ */
+static int enetc4_add_ppm_fixed_link(struct enetc_pf *pf,
+ struct device_node *np)
+{
+ struct device_node *fixed_link;
+ struct of_changeset *ocs;
+ int speed, err;
+
+ enetc4_get_pcr_speed(&pf->si->hw, &speed);
+ speed = enetc_phylink_match_pseudo_mac_speed(speed);
+
+ ocs = kzalloc_obj(*ocs);
+ if (!ocs)
+ return -ENOMEM;
+
+ of_changeset_init(ocs);
+
+ fixed_link = of_changeset_create_node(ocs, np, "fixed-link");
+ if (!fixed_link) {
+ err = -ENOMEM;
+ goto err_destroy;
+ }
+
+ err = of_changeset_add_prop_u32(ocs, fixed_link, "speed", speed);
+ if (err)
+ goto err_put_node;
+
+ err = of_changeset_add_prop_bool(ocs, fixed_link, "full-duplex");
+ if (err)
+ goto err_put_node;
+
+ err = of_changeset_apply(ocs);
+ if (err)
+ goto err_put_node;
+
+ of_node_put(fixed_link);
+ pf->ppm_ocs = ocs;
+
+ return 0;
+
+err_put_node:
+ of_node_put(fixed_link);
+err_destroy:
+ of_changeset_destroy(ocs);
+ kfree(ocs);
+
+ return err;
+}
+
static int enetc4_link_init(struct enetc_ndev_priv *priv,
struct device_node *node)
{
+ bool dynamic = node && of_node_check_flag(node, OF_DYNAMIC);
struct enetc_pf *pf = enetc_si_priv(priv->si);
struct device *dev = priv->dev;
int err;
- err = of_get_phy_mode(node, &pf->if_mode);
- if (err) {
- dev_err(dev, "Failed to get PHY mode\n");
- return err;
+ /* Pseudo-MAC ENETCs are described by a runtime-synthesized fixed-link
+ * node rather than static DT. This covers both a missing OF node and a
+ * PCI-synthesized (OF_DYNAMIC) node, which is an empty node created
+ * by the PCI core and thus carries no fixed-link description.
+ */
+ if (enetc_is_pseudo_mac(priv->si) && (!node || dynamic)) {
+ pf->if_mode = PHY_INTERFACE_MODE_INTERNAL;
+
+ if (!node) {
+ dev_err(dev, "Missing of_node for Pseudo-MAC port\n");
+ return -ENODEV;
+ }
+
+ err = enetc4_add_ppm_fixed_link(pf, node);
+ if (err) {
+ dev_err(dev, "Failed to add fixed-link node\n");
+ return err;
+ }
+ } else {
+ err = of_get_phy_mode(node, &pf->if_mode);
+ if (err) {
+ dev_err(dev, "Failed to get PHY mode\n");
+ return err;
+ }
}
err = enetc_mdiobus_create(pf, node);
if (err) {
dev_err(dev, "Failed to create MDIO bus\n");
- return err;
+ goto err_mdiobus_create;
}
err = enetc_phylink_create(priv, node, &enetc_pl_mac_ops);
@@ -975,6 +1069,8 @@ static int enetc4_link_init(struct enetc_ndev_priv *priv,
err_phylink_create:
enetc_mdiobus_destroy(pf);
+err_mdiobus_create:
+ enetc4_put_ppm_fixed_link(pf);
return err;
}
@@ -985,6 +1081,7 @@ static void enetc4_link_deinit(struct enetc_ndev_priv *priv)
enetc_phylink_destroy(priv);
enetc_mdiobus_destroy(pf);
+ enetc4_put_ppm_fixed_link(pf);
}
static void enetc4_pf_link_status_task(struct work_struct *work)
diff --git a/drivers/net/ethernet/freescale/enetc/enetc_pf.h b/drivers/net/ethernet/freescale/enetc/enetc_pf.h
index 1f92d16d8ca8..ab11e40cfdbb 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc_pf.h
+++ b/drivers/net/ethernet/freescale/enetc/enetc_pf.h
@@ -67,6 +67,8 @@ struct enetc_pf {
bool sriov_enabled;
bool link_up;
u16 link_status_ms_mask;
+
+ struct of_changeset *ppm_ocs; /* pseudo-MAC fixed-link changeset */
};
#define phylink_to_enetc_pf(config) \
diff --git a/drivers/net/ethernet/freescale/enetc/enetc_pf_common.c b/drivers/net/ethernet/freescale/enetc/enetc_pf_common.c
index 8facc8bf8693..b032b586f07b 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc_pf_common.c
+++ b/drivers/net/ethernet/freescale/enetc/enetc_pf_common.c
@@ -436,6 +436,43 @@ void enetc_mdiobus_destroy(struct enetc_pf *pf)
}
EXPORT_SYMBOL_GPL(enetc_mdiobus_destroy);
+static struct {
+ unsigned long mac_cap;
+ int speed; /* descending order sorted */
+} enetc_phylink_pseudo_mac_caps[] = {
+ { MAC_25000FD, SPEED_25000 },
+ { MAC_20000FD, SPEED_20000 },
+ { MAC_10000FD, SPEED_10000 },
+ { MAC_5000FD, SPEED_5000 },
+ { MAC_2500FD, SPEED_2500 },
+ { MAC_1000FD, SPEED_1000 },
+ { MAC_100FD, SPEED_100 },
+ { MAC_10FD, SPEED_10 },
+};
+
+static unsigned long enetc_phylink_get_pseudo_mac_caps(void)
+{
+ unsigned long mac_caps = 0;
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(enetc_phylink_pseudo_mac_caps); i++)
+ mac_caps |= enetc_phylink_pseudo_mac_caps[i].mac_cap;
+
+ return mac_caps;
+}
+
+int enetc_phylink_match_pseudo_mac_speed(int speed)
+{
+ int i;
+
+ for (i = 0; i < ARRAY_SIZE(enetc_phylink_pseudo_mac_caps); i++)
+ if (enetc_phylink_pseudo_mac_caps[i].speed <= speed)
+ return enetc_phylink_pseudo_mac_caps[i].speed;
+
+ return SPEED_UNKNOWN;
+}
+EXPORT_SYMBOL_GPL(enetc_phylink_match_pseudo_mac_speed);
+
int enetc_phylink_create(struct enetc_ndev_priv *priv, struct device_node *node,
const struct phylink_mac_ops *ops)
{
@@ -471,9 +508,7 @@ int enetc_phylink_create(struct enetc_ndev_priv *priv, struct device_node *node,
phy_interface_set_rgmii(pf->phylink_config.supported_interfaces);
} else {
- mac_caps |= MAC_10FD | MAC_100FD | MAC_1000FD | MAC_2500FD |
- MAC_5000FD | MAC_10000FD | MAC_20000FD |
- MAC_25000FD;
+ mac_caps |= enetc_phylink_get_pseudo_mac_caps();
}
pf->phylink_config.mac_capabilities = mac_caps;
diff --git a/drivers/net/ethernet/freescale/enetc/enetc_pf_common.h b/drivers/net/ethernet/freescale/enetc/enetc_pf_common.h
index c9eed879d5a3..aede1aad6362 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc_pf_common.h
+++ b/drivers/net/ethernet/freescale/enetc/enetc_pf_common.h
@@ -13,6 +13,7 @@ void enetc_mdiobus_destroy(struct enetc_pf *pf);
int enetc_phylink_create(struct enetc_ndev_priv *priv, struct device_node *node,
const struct phylink_mac_ops *ops);
void enetc_phylink_destroy(struct enetc_ndev_priv *priv);
+int enetc_phylink_match_pseudo_mac_speed(int speed);
void enetc_set_default_rss_key(struct enetc_pf *pf);
int enetc_vlan_rx_add_vid(struct net_device *ndev, __be16 prot, u16 vid);
int enetc_vlan_rx_del_vid(struct net_device *ndev, __be16 prot, u16 vid);
--
2.34.1
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH net-next v4 1/6] net: enetc: Add pseudo-MAC support for ENETCv4 Ports via a fixed-link node
2026-11-13 15:02 ` [PATCH net-next v4 1/6] net: enetc: Add pseudo-MAC support for ENETCv4 Ports via a fixed-link node Claudiu Manoil
@ 2026-11-17 15:55 ` netdev-bot+sashiko
0 siblings, 0 replies; 10+ messages in thread
From: netdev-bot+sashiko @ 2026-11-17 15:55 UTC (permalink / raw)
To: claudiu.manoil; +Cc: netdev, s32, bhelgaas, linux-pci, linux-kernel, kuba
Thank you for your contribution! Sashiko AI review found 3 potential
issue(s) to consider.
Critical: 0 · High: 0 · Medium: 1 · Low: 2
If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).
- [Medium] In enetc4_link_init()
(drivers/net/ethernet/freescale/enetc/enetc4_pf.c:1025-1047), the…
- [Low] In drivers/net/ethernet/freescale/enetc/enetc4_pf.c,
enetc4_link_init() (lines 1035-1047) only builds the pseudo-MAC fixed-…
- [Low] enetc4_add_ppm_fixed_link()
(drivers/net/ethernet/freescale/enetc/enetc4_pf.c:996) writes a…
--
net: enetc: Add pseudo-MAC support for ENETCv4 Ports via a fixed-link node
This lets ENETCv4 pseudo-MAC ports probe without a static DT "ethernet"
node. At probe time an of_changeset adds a "fixed-link" child node whose
speed is read from PCR[PSPEED]. The pseudo-MAC path runs when the port has
no OF node or only an OF_DYNAMIC node, and the changeset is reverted on
teardown and on the probe error unwind.
> diff --git a/drivers/net/ethernet/freescale/enetc/enetc4_pf.c b/drivers/net/ethernet/freescale/enetc/enetc4_pf.c
> index 71c971618388b..b7e7201e6a71a 100644
> --- a/drivers/net/ethernet/freescale/enetc/enetc4_pf.c
> +++ b/drivers/net/ethernet/freescale/enetc/enetc4_pf.c
[ ... ]
> +static int enetc4_add_ppm_fixed_link(struct enetc_pf *pf,
> + struct device_node *np)
> +{
> + struct device_node *fixed_link;
> + struct of_changeset *ocs;
> + int speed, err;
> +
> + enetc4_get_pcr_speed(&pf->si->hw, &speed);
> + speed = enetc_phylink_match_pseudo_mac_speed(speed);
> +
[ ... ]
> + fixed_link = of_changeset_create_node(ocs, np, "fixed-link");
> + if (!fixed_link) {
> + err = -ENOMEM;
> + goto err_destroy;
> + }
> +
> + err = of_changeset_add_prop_u32(ocs, fixed_link, "speed", speed);
[Severity: Low]
This isn't a bug, but can this write a speed value that the fixed-link
binding doesn't allow?
enetc_phylink_pseudo_mac_caps[] in enetc_pf_common.c includes SPEED_25000
and SPEED_20000. When PCR[PSPEED] maps to 20000 Mbps or more,
enetc_phylink_match_pseudo_mac_speed() returns 20000 or 25000.
In Documentation/devicetree/bindings/net/ethernet-controller.yaml, the
fixed-link speed property only allows:
enum: [10, 100, 1000, 2500, 5000, 10000]
phylink still accepts these values. But the node is built at runtime, so
dt-schema never checks it.
Should the binding be extended (as was done for 5000)? Or should the
driver stick to the speeds the binding lists?
> + if (err)
> + goto err_put_node;
[ ... ]
> static int enetc4_link_init(struct enetc_ndev_priv *priv,
> struct device_node *node)
> {
> + bool dynamic = node && of_node_check_flag(node, OF_DYNAMIC);
> struct enetc_pf *pf = enetc_si_priv(priv->si);
> struct device *dev = priv->dev;
> int err;
>
> - err = of_get_phy_mode(node, &pf->if_mode);
> - if (err) {
> - dev_err(dev, "Failed to get PHY mode\n");
> - return err;
> + /* Pseudo-MAC ENETCs are described by a runtime-synthesized fixed-link
> + * node rather than static DT. This covers both a missing OF node and a
> + * PCI-synthesized (OF_DYNAMIC) node, which is an empty node created
> + * by the PCI core and thus carries no fixed-link description.
> + */
[Severity: Low]
Is this comment accurate?
The PCI-synthesized node isn't empty. of_pci_make_dev_node() fills it in
through of_pci_add_properties() (compatible, reg, ...) before it applies
the changeset. The only accurate part is that it has no fixed-link child.
The branch also doesn't really cover a missing OF node. That case just
returns -ENODEV below.
Related to this: at this commit, the PCI core doesn't create a node for an
ENETC endpoint function. pci_bus_add_device() only does it for bridges:
if (pci_is_bridge(dev))
of_pci_make_dev_node(dev);
drivers/pci/quirks.c has no ENETC entries either. So a pseudo-MAC port
without a node always hits the "Missing of_node for Pseudo-MAC port" error
here.
A later patch in the series, "PCI: Create OF node for NXP ENETC pseudo-MAC
functions", fixes that. It adds DECLARE_PCI_FIXUP_FINAL() entries for
PCI_VENDOR_ID_PHILIPS device IDs 0xe101 and 0xe110 that call
of_pci_make_dev_node(). With the whole series applied, only the comment
wording is left.
> + if (enetc_is_pseudo_mac(priv->si) && (!node || dynamic)) {
> + pf->if_mode = PHY_INTERFACE_MODE_INTERNAL;
[Severity: Medium]
Is the OF_DYNAMIC flag enough to show that the node has no link
description?
__of_node_dup() sets OF_DYNAMIC, and DT overlays create their nodes with
it in add_changeset_node(). An overlay can also be applied on top of the
node from of_pci_make_dev_node() to add phy-mode or fixed-link.
In either case, this branch:
- skips of_get_phy_mode()
- forces PHY_INTERFACE_MODE_INTERNAL
- calls enetc4_add_ppm_fixed_link(), which calls
of_changeset_create_node(ocs, np, "fixed-link") without checking for an
existing fixed-link child
__of_attach_node() puts the new node at the head of the child list:
np->sibling = np->parent->child;
np->parent->child = np;
phylink_parse_fixedlink() looks the node up with
fwnode_get_named_child_node(), so it would find the driver-synthesized
fixed-link before the DT one.
Would that leave two fixed-link children and replace the DT-provided speed
with the PCR[PSPEED] value?
The commit message says:
The pseudo-MAC path is selected when the port has no OF node, or when it
only has the empty PCI-synthesized node (OF_DYNAMIC), which carries no
fixed-link description.
Nothing in the code checks that, though. Would a check such as
of_get_child_by_name(node, "fixed-link") or
of_property_present(node, "phy-mode") make the code match the
description?
> +
> + if (!node) {
> + dev_err(dev, "Missing of_node for Pseudo-MAC port\n");
> + return -ENODEV;
> + }
[ ... ]
--
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/cover.1794581338.git.claudiu.manoil%40nxp.com
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH net-next v4 2/6] net: enetc: Allocate int_vector array dynamically
[not found] <cover.1794581338.git.claudiu.manoil@nxp.com>
2026-11-13 15:02 ` [PATCH net-next v4 1/6] net: enetc: Add pseudo-MAC support for ENETCv4 Ports via a fixed-link node Claudiu Manoil
@ 2026-11-13 15:02 ` Claudiu Manoil
2026-11-13 15:02 ` [PATCH net-next v4 3/6] net: enetc: Enable probing of version 4.4 devices Claudiu Manoil
` (3 subsequent siblings)
5 siblings, 0 replies; 10+ messages in thread
From: Claudiu Manoil @ 2026-11-13 15:02 UTC (permalink / raw)
To: netdev
Cc: s32, Vladimir Oltean, Wei Fang, Clark Wang, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni, imx,
linux-kernel
The number of Rx/Tx ring interrupt vectors, priv->bdr_int_num, is set
in enetc_init_si_rings_params() to the lesser of the number of online
CPUs and the number of Rx rings of the station interface. The
int_vector[] array in struct enetc_ndev_priv, however, has a fixed
size of ENETC_MAX_BDR_INT (6), a limit that dates back to the maximum
CPU count of the SoCs integrating ENETC at the time.
That assumption does not hold for S32N79, which has 8 CPUs and gives
the ENETC PF up to 8 Rx rings. There, bdr_int_num can reach 8, and
enetc_alloc_msix() writes priv->int_vector[i] past the end of the
array, corrupting the adjacent fields of the netdev private area.
No SoC currently supported upstream exceeds 6 vectors, so this is not
reachable today; it must be fixed before S32N79 devices can be enabled
by a following patch.
Allocate int_vector[] with devm_kcalloc() in enetc_alloc_msix(), once
bdr_int_num is known. The memory is released together with the PCI
device, so no explicit free is needed on the error or teardown paths.
This is common code, so the PF and VF drivers benefit alike. Drop the
ENETC_MAX_BDR_INT define, which only existed to size this array.
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
---
v4:
- New patch. Prevents an out-of-bounds access to priv->int_vector[]
when bdr_int_num exceeds 6, as is the case on S32N79, as requested
in review.
drivers/net/ethernet/freescale/enetc/enetc.c | 8 ++++++++
drivers/net/ethernet/freescale/enetc/enetc.h | 3 +--
2 files changed, 9 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/freescale/enetc/enetc.c b/drivers/net/ethernet/freescale/enetc/enetc.c
index c6571b5cc842..855b2d3b8718 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc.c
+++ b/drivers/net/ethernet/freescale/enetc/enetc.c
@@ -3641,6 +3641,13 @@ int enetc_alloc_msix(struct enetc_ndev_priv *priv)
if (n != nvec)
return -EPERM;
+ priv->int_vector = devm_kcalloc(&pdev->dev, priv->bdr_int_num,
+ sizeof(*priv->int_vector), GFP_KERNEL);
+ if (!priv->int_vector) {
+ err = -ENOMEM;
+ goto fail_int_vector;
+ }
+
/* # of tx rings per int vector */
v_tx_rings = priv->num_tx_rings / priv->bdr_int_num;
v_remainder = priv->num_tx_rings % priv->bdr_int_num;
@@ -3676,6 +3683,7 @@ int enetc_alloc_msix(struct enetc_ndev_priv *priv)
while (i--)
enetc_int_vector_destroy(priv, i);
+fail_int_vector:
pci_free_irq_vectors(pdev);
return err;
diff --git a/drivers/net/ethernet/freescale/enetc/enetc.h b/drivers/net/ethernet/freescale/enetc/enetc.h
index d9e91832a9c1..689ad82b2949 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc.h
+++ b/drivers/net/ethernet/freescale/enetc/enetc.h
@@ -407,7 +407,6 @@ struct enetc_cls_rule {
int used;
};
-#define ENETC_MAX_BDR_INT 6 /* fixed to max # of available cpus */
struct psfp_cap {
u32 max_streamid;
u32 max_psfp_filter;
@@ -456,7 +455,7 @@ struct enetc_ndev_priv {
struct enetc_si *si;
int bdr_int_num; /* number of Rx/Tx ring interrupts */
- struct enetc_int_vector *int_vector[ENETC_MAX_BDR_INT];
+ struct enetc_int_vector **int_vector;
u16 num_rx_rings, num_tx_rings;
u16 rx_bd_count, tx_bd_count;
--
2.34.1
^ permalink raw reply [flat|nested] 10+ messages in thread* [PATCH net-next v4 3/6] net: enetc: Enable probing of version 4.4 devices
[not found] <cover.1794581338.git.claudiu.manoil@nxp.com>
2026-11-13 15:02 ` [PATCH net-next v4 1/6] net: enetc: Add pseudo-MAC support for ENETCv4 Ports via a fixed-link node Claudiu Manoil
2026-11-13 15:02 ` [PATCH net-next v4 2/6] net: enetc: Allocate int_vector array dynamically Claudiu Manoil
@ 2026-11-13 15:02 ` Claudiu Manoil
2026-11-17 15:55 ` netdev-bot+sashiko
2026-11-13 15:02 ` [PATCH net-next v4 4/6] arm64: dts: s32n79: add NETC IEP18 ECAM node for ENETC PPM ports Claudiu Manoil
` (2 subsequent siblings)
5 siblings, 1 reply; 10+ messages in thread
From: Claudiu Manoil @ 2026-11-13 15:02 UTC (permalink / raw)
To: netdev
Cc: s32, Vladimir Oltean, Wei Fang, Clark Wang, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni, imx,
linux-kernel
This commit adds driver data and platform info entries for ENETC v4.4 PF
and VF devices on S32N79 SoCs with a 500 MHz system clock frequency.
The ENETC v4.4 devices are present on S32N79 SoCs and feature only
Pseudo-MAC ENETC ports, whose PFs use either the 0xe101 or the 0xe110
PCI device ID, and the VFs answer to the usual device ID of 0xef00.
While the ENETC PF instances are usually designated by the 0xe101 ID,
in newer IP versions the 0xe110 ID was introduced to indicate the
power-on reset default NETC switch management interface. These IDs
do not indicate any functional difference between the designated
ENETC instances, and can be interchanged and re-assigned at boot time
(i.e. from firmware).
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
---
v4:
- Added a leading sentence to the commit message, stating that the
patch adds driver data and platform info entries for ENETC v4.4
devices with a 500 MHz system clock frequency.
v3:
- Added sysclk_freq info for the v4.4 ENETC devices.
v2:
- Described why there are two ENETC PF IDs in the commit message, as per
review.
drivers/net/ethernet/freescale/enetc/enetc.c | 26 +++++++++++++++++++
.../net/ethernet/freescale/enetc/enetc_hw.h | 4 ++-
2 files changed, 29 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/freescale/enetc/enetc.c b/drivers/net/ethernet/freescale/enetc/enetc.c
index 855b2d3b8718..e3aa117d323d 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc.c
+++ b/drivers/net/ethernet/freescale/enetc/enetc.c
@@ -3840,6 +3840,20 @@ static const struct enetc_drvdata enetc4_vf_data = {
.eth_ops = &enetc_vf_ethtool_ops,
};
+static const struct enetc_drvdata enetc4_s32n79_ppm_data = {
+ .sysclk_freq = ENETC_CLK_500M,
+ .tx_csum = true,
+ .max_frags = ENETC4_MAX_SKB_FRAGS,
+ .eth_ops = &enetc4_ppm_ethtool_ops,
+};
+
+static const struct enetc_drvdata enetc4_s32n79_vf_data = {
+ .sysclk_freq = ENETC_CLK_500M,
+ .tx_csum = true,
+ .max_frags = ENETC4_MAX_SKB_FRAGS,
+ .eth_ops = &enetc_vf_ethtool_ops,
+};
+
static const struct enetc_platform_info enetc_info[] = {
{ .revision = ENETC_REV_1_0,
.dev_id = ENETC_DEV_ID_PF,
@@ -3870,6 +3884,18 @@ static const struct enetc_platform_info enetc_info[] = {
.dev_id = NXP_ENETC_VF_DEV_ID,
.data = &enetc4_vf_data,
},
+ { .revision = ENETC_REV_4_4,
+ .dev_id = NXP_ENETC_PPM_DEV_ID,
+ .data = &enetc4_s32n79_ppm_data,
+ },
+ { .revision = ENETC_REV_4_4,
+ .dev_id = NXP_ENETC_PF_DEV_ID,
+ .data = &enetc4_s32n79_ppm_data,
+ },
+ { .revision = ENETC_REV_4_4,
+ .dev_id = NXP_ENETC_VF_DEV_ID,
+ .data = &enetc4_s32n79_vf_data,
+ },
};
int enetc_get_driver_data(struct enetc_si *si)
diff --git a/drivers/net/ethernet/freescale/enetc/enetc_hw.h b/drivers/net/ethernet/freescale/enetc/enetc_hw.h
index 2c9d9042eb0b..8e6a9d366d40 100644
--- a/drivers/net/ethernet/freescale/enetc/enetc_hw.h
+++ b/drivers/net/ethernet/freescale/enetc/enetc_hw.h
@@ -414,6 +414,7 @@ enum enetc_bdr_type {TX, RX};
#define ENETC_REV_1_0 0x0100
#define ENETC_REV_4_1 0X0401
#define ENETC_REV_4_3 0x0403
+#define ENETC_REV_4_4 0x0404
#define ENETC_G_EIPBRR1 0x0bfc
#define ENETC_G_EPFBLPR(n) (0xd00 + 4 * (n))
@@ -1048,8 +1049,9 @@ struct enetc_cbd {
u8 status_flags;
};
-#define ENETC_CLK_400M 400000000ULL
#define ENETC_CLK_333M 333000000ULL
+#define ENETC_CLK_400M 400000000ULL
+#define ENETC_CLK_500M 500000000ULL
static inline u32 enetc_cycles_to_usecs(u32 cycles, u64 clk_freq)
{
--
2.34.1
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH net-next v4 3/6] net: enetc: Enable probing of version 4.4 devices
2026-11-13 15:02 ` [PATCH net-next v4 3/6] net: enetc: Enable probing of version 4.4 devices Claudiu Manoil
@ 2026-11-17 15:55 ` netdev-bot+sashiko
0 siblings, 0 replies; 10+ messages in thread
From: netdev-bot+sashiko @ 2026-11-17 15:55 UTC (permalink / raw)
To: claudiu.manoil; +Cc: netdev, s32, bhelgaas, linux-pci, linux-kernel, kuba
Thank you for your contribution! Sashiko AI review found 2 potential
issue(s) to consider.
Critical: 0 · High: 0 · Medium: 2 · Low: 0
If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).
- [Medium] This patch adds ENETC_REV_4_4 entries to enetc_info[] so that
S32N79 PFs (0xe101/0xe110) and VFs (0xef00) now probe.
- [Medium] When an ENETC v4 VF cannot get the IP minor revision from its
PF over the VSI-PSI mailbox, enetc_vf_get_revision()…
--
net: enetc: Enable probing of version 4.4 devices
This commit adds driver data and platform info entries for the ENETC v4.4
PF and VF devices on S32N79 SoCs. These run with a 500 MHz system clock.
The PFs use the 0xe101 or 0xe110 device ID and the VFs use 0xef00.
> diff --git a/drivers/net/ethernet/freescale/enetc/enetc.c b/drivers/net/ethernet/freescale/enetc/enetc.c
> index 855b2d3b87182..e3aa117d323d5 100644
> --- a/drivers/net/ethernet/freescale/enetc/enetc.c
> +++ b/drivers/net/ethernet/freescale/enetc/enetc.c
> @@ -3840,6 +3840,20 @@ static const struct enetc_drvdata enetc4_vf_data = {
[ ... ]
> +static const struct enetc_drvdata enetc4_s32n79_vf_data = {
> + .sysclk_freq = ENETC_CLK_500M,
> + .tx_csum = true,
> + .max_frags = ENETC4_MAX_SKB_FRAGS,
> + .eth_ops = &enetc_vf_ethtool_ops,
> +};
[Severity: Medium]
Can an S32N79 VF end up with the 333 MHz drvdata instead of this one?
The VF gets its minor revision from the PF over the VSI-PSI mailbox. If
that query fails, enetc_vf_get_revision() falls back to 4.1:
drivers/net/ethernet/freescale/enetc/enetc_vf.c:enetc_vf_get_revision() {
...
ip_mn = enetc_vf_get_ip_minor_revision(si);
if (ip_mn >= 0) {
si->revision = (si->pdev->revision << 8) | ip_mn;
return;
}
si->revision = ENETC_REV_4_1;
...
}
The query can fail in several ways:
- -EIO when the mailbox is busy
- -ETIMEDOUT after the 200 ms poll
- an error status returned by the PSI
- -ENOMEM from dma_alloc_coherent()
S32N79 VFs use the same 0xef00 device ID. So in that case
enetc_get_driver_data() matches {ENETC_REV_4_1, NXP_ENETC_VF_DEV_ID} and
selects enetc4_vf_data, which uses ENETC_CLK_333M. enetc_vf_netdev_setup()
then copies that value into priv->sysclk_freq.
Before this patch every v4 platform ran at 333 MHz, so the fallback made no
difference. On a 500 MHz platform, wouldn't enetc_usecs_to_cycles() program
the default 600 us tx_ictt, the DIM rx_ictt and the set_coalesce values at
roughly 0.67x the requested timeout?
enetc_get_coalesce() converts back with the same clock. So ethtool -c would
still show the requested values while the hardware fires about 1.5x sooner.
Should the fallback change now that v4 platforms don't all use the same
clock frequency?
> @@ -3870,6 +3884,18 @@ static const struct enetc_platform_info enetc_info[] = {
> .dev_id = NXP_ENETC_VF_DEV_ID,
> .data = &enetc4_vf_data,
> },
> + { .revision = ENETC_REV_4_4,
> + .dev_id = NXP_ENETC_PPM_DEV_ID,
> + .data = &enetc4_s32n79_ppm_data,
> + },
> + { .revision = ENETC_REV_4_4,
> + .dev_id = NXP_ENETC_PF_DEV_ID,
> + .data = &enetc4_s32n79_ppm_data,
> + },
> + { .revision = ENETC_REV_4_4,
> + .dev_id = NXP_ENETC_VF_DEV_ID,
> + .data = &enetc4_s32n79_vf_data,
> + },
[Severity: Medium]
Does enetc_get_phc_index_by_pdev() in enetc_ethtool.c also need an
ENETC_REV_4_4 case? It has its own per-revision table for the timer devfn:
drivers/net/ethernet/freescale/enetc/enetc_ethtool.c:enetc_get_phc_index_by_pdev() {
...
switch (si->revision) {
case ENETC_REV_1_0:
devfn = PCI_DEVFN(0, 4);
break;
case ENETC_REV_4_1:
devfn = PCI_DEVFN(24, 0);
break;
case ENETC_REV_4_3:
devfn = PCI_DEVFN(0, 1);
break;
default:
return -1;
}
...
}
enetc4_s32n79_vf_data uses enetc_vf_ethtool_ops, which provides
enetc_get_ts_info(). SR-IOV VFs have no OF node, so enetc_get_phc_index()
always takes the per-pdev path. For a v4.4 VF the call chain is:
enetc_get_ts_info()
enetc_get_phc_index()
enetc_get_phc_index_by_pdev()
default: return -1
enetc_get_ts_info() then jumps to timestamp_tx_sw. It reports no PHC and
only SOF_TIMESTAMPING_TX_SOFTWARE.
enetc_hwtstamp_set() and enetc_hwtstamp_get() still accept hardware
timestamping configs, because they only check
enetc_ptp_clock_is_enabled().
Would ptp4l and similar tools then fail to find a PHC on S32N79 VFs?
The v4.4 PF uses enetc4_ppm_ethtool_ops, which has no get_ts_info, so this
seems to affect only the VFs.
The code doesn't show whether a NETC timer is reachable on the VF's PCI
bus on S32N79. If PTP is intentionally left unsupported on v4.4, could the
commit message say so?
--
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/cover.1794581338.git.claudiu.manoil%40nxp.com
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH net-next v4 4/6] arm64: dts: s32n79: add NETC IEP18 ECAM node for ENETC PPM ports
[not found] <cover.1794581338.git.claudiu.manoil@nxp.com>
` (2 preceding siblings ...)
2026-11-13 15:02 ` [PATCH net-next v4 3/6] net: enetc: Enable probing of version 4.4 devices Claudiu Manoil
@ 2026-11-13 15:02 ` Claudiu Manoil
2026-11-13 15:02 ` [PATCH net-next v4 5/6] arm64: dts: s32n79-rdb: enable NETC IEP18 Claudiu Manoil
2026-11-13 15:02 ` [PATCH net-next v4 6/6] PCI: Create OF node for NXP ENETC pseudo-MAC functions Claudiu Manoil
5 siblings, 0 replies; 10+ messages in thread
From: Claudiu Manoil @ 2026-11-13 15:02 UTC (permalink / raw)
To: netdev
Cc: s32, Frank Li, Sascha Hauer, Pengutronix Kernel Team,
Fabio Estevam, Rob Herring, Krzysztof Kozlowski, Conor Dooley,
imx, linux-arm-kernel, devicetree, linux-kernel
Add the netc_iep18 PCIe ECAM host node describing the NETC integrated
endpoint bus allocated to the first Linux cohort on S32N79. The bus is
enumerated via the generic pci-host-ecam-generic driver and hosts the
ENETC v4 Pseudo-MAC (PPM) PF and its VFs, which connect internally to
the on-chip NETC switch (not owned by Linux).
The node is kept disabled at the SoC level; boards that wire up NETC
enable it. MSIs are routed through the GIC-v3 ITS (msi-map) and DMA is
translated by the SMMUv3 (iommu-map).
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
---
v4: none
v3: none
v2:
- Replaced the 'msi-parent' stub property with 'msi-map', as per H/W spec.
arch/arm64/boot/dts/freescale/s32n79.dtsi | 24 +++++++++++++++++++++++
1 file changed, 24 insertions(+)
diff --git a/arch/arm64/boot/dts/freescale/s32n79.dtsi b/arch/arm64/boot/dts/freescale/s32n79.dtsi
index 94ab58783fdc..65cf65e82832 100644
--- a/arch/arm64/boot/dts/freescale/s32n79.dtsi
+++ b/arch/arm64/boot/dts/freescale/s32n79.dtsi
@@ -122,6 +122,30 @@ irqsteer_coss: interrupt-controller@4ed00000 {
};
};
+ nwss-bus {
+ compatible = "simple-bus";
+ #address-cells = <2>;
+ #size-cells = <2>;
+ ranges = <0x0 0x48800000 0x0 0x48800000 0x0 0x4b0000>,
+ <0x2 0x40000000 0x2 0x40000000 0x0 0x20000000>;
+
+ netc_iep18: pcie@241200000 {
+ compatible = "pci-host-ecam-generic";
+ reg = <0x2 0x41200000 0x0 0x100000>;
+ #address-cells = <3>;
+ #size-cells = <2>;
+ device_type = "pci";
+ bus-range = <0x0 0x0>;
+ iommu-map = <0x0 &smmu 0xc0000 0x100>;
+ msi-map = <0x0 &its 0xc0000 0x100>;
+ dma-coherent;
+ status = "disabled";
+
+ /* All NETC PCIe functions */
+ ranges = <0x82000000 0x0 0x48800000 0x0 0x48800000 0x0 0x4b0000>;
+ };
+ };
+
cpus {
#address-cells = <1>;
#size-cells = <0>;
--
2.34.1
^ permalink raw reply [flat|nested] 10+ messages in thread* [PATCH net-next v4 5/6] arm64: dts: s32n79-rdb: enable NETC IEP18
[not found] <cover.1794581338.git.claudiu.manoil@nxp.com>
` (3 preceding siblings ...)
2026-11-13 15:02 ` [PATCH net-next v4 4/6] arm64: dts: s32n79: add NETC IEP18 ECAM node for ENETC PPM ports Claudiu Manoil
@ 2026-11-13 15:02 ` Claudiu Manoil
2026-11-17 15:55 ` netdev-bot+sashiko
2026-11-13 15:02 ` [PATCH net-next v4 6/6] PCI: Create OF node for NXP ENETC pseudo-MAC functions Claudiu Manoil
5 siblings, 1 reply; 10+ messages in thread
From: Claudiu Manoil @ 2026-11-13 15:02 UTC (permalink / raw)
To: netdev
Cc: s32, Frank Li, Sascha Hauer, Pengutronix Kernel Team,
Fabio Estevam, Rob Herring, Krzysztof Kozlowski, Conor Dooley,
imx, linux-arm-kernel, devicetree, linux-kernel
Enable the NETC v4 ECAM bus #18 on the S32N79-RDB board so that the
ENETC v4 Pseudo-MAC port (PPM) PF and VF instances are probed.
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
---
v4: none
v3: none
v2: none
arch/arm64/boot/dts/freescale/s32n79-rdb.dts | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/arch/arm64/boot/dts/freescale/s32n79-rdb.dts b/arch/arm64/boot/dts/freescale/s32n79-rdb.dts
index 1feccd61258e..897ba24ac03a 100644
--- a/arch/arm64/boot/dts/freescale/s32n79-rdb.dts
+++ b/arch/arm64/boot/dts/freescale/s32n79-rdb.dts
@@ -68,3 +68,7 @@ &usdhc0 {
no-sdio;
status = "okay";
};
+
+&netc_iep18 {
+ status = "okay";
+};
--
2.34.1
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH net-next v4 5/6] arm64: dts: s32n79-rdb: enable NETC IEP18
2026-11-13 15:02 ` [PATCH net-next v4 5/6] arm64: dts: s32n79-rdb: enable NETC IEP18 Claudiu Manoil
@ 2026-11-17 15:55 ` netdev-bot+sashiko
0 siblings, 0 replies; 10+ messages in thread
From: netdev-bot+sashiko @ 2026-11-17 15:55 UTC (permalink / raw)
To: claudiu.manoil; +Cc: netdev, s32, bhelgaas, linux-pci, linux-kernel, kuba
Thank you for your contribution! Sashiko AI review found 3 potential
issue(s) to consider.
Critical: 0 · High: 0 · Medium: 2 · Low: 1
If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).
- [Medium] This patch turns on `&netc_iep18` in
arch/arm64/boot/dts/freescale/s32n79-rdb.dts so that, per the commit…
- [Medium] Setting `&netc_iep18` to okay makes the ENETC v4 PPM PF and
VFs live DMA masters.
- [Low] The new `&netc_iep18` override in
arch/arm64/boot/dts/freescale/s32n79-rdb.dts is added after `&usdhc0`.
--
arm64: dts: s32n79-rdb: enable NETC IEP18
This commit enables the NETC v4 ECAM bus #18 on the S32N79-RDB board. The
goal is for the ENETC v4 Pseudo-MAC port (PPM) PF and VF instances to be
probed.
> diff --git a/arch/arm64/boot/dts/freescale/s32n79-rdb.dts b/arch/arm64/boot/dts/freescale/s32n79-rdb.dts
> index 1feccd61258ef..897ba24ac03ae 100644
> --- a/arch/arm64/boot/dts/freescale/s32n79-rdb.dts
> +++ b/arch/arm64/boot/dts/freescale/s32n79-rdb.dts
> @@ -68,3 +68,7 @@ &usdhc0 {
> no-sdio;
> status = "okay";
> };
> +
> +&netc_iep18 {
[Severity: Low]
This isn't a bug, but Documentation/devicetree/bindings/dts-coding-style.rst
says:
When extending nodes in the board DTS via &label, the entries shall be
ordered either alpha-numerically or by keeping the order from DTSI,
The existing overrides in this file are in alphabetical order (irqsteer_coss,
uart0, uart5, uart6, uart7, usdhc0). Should &netc_iep18 go between
&irqsteer_coss and &uart0? Keeping the DTSI order would also put it before
&usdhc0.
> + status = "okay";
> +};
[Severity: Medium]
The commit message says the bus is enabled "so that the ENETC v4 Pseudo-MAC
port (PPM) PF and VF instances are probed". Does the PF actually probe when
the kernel is built from the in-tree arm64 defconfig?
The netc_iep18 host bridge node in s32n79.dtsi has no child nodes, so the
ENETC PF function does not get a static of_node. For a pseudo-MAC port,
enetc4_link_init() in drivers/net/ethernet/freescale/enetc/enetc4_pf.c then
fails here:
if (!node) {
dev_err(dev, "Missing of_node for Pseudo-MAC port\n");
return -ENODEV;
}
The only other way to get an of_node is of_pci_make_dev_node(). The last
patch in the series, "PCI: Create OF node for NXP ENETC pseudo-MAC
functions", adds it as a final fixup for 0xe101/0xe110:
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_PHILIPS, 0xe101, of_pci_make_dev_node);
However, of_pci_make_dev_node() is only built with
CONFIG_PCI_DYNAMIC_OF_NODES. Without it, drivers/pci/pci.h provides an empty
inline stub.
PCI_DYNAMIC_OF_NODES is a plain bool with no default, and NXP_ENETC4 only
selects OF_DYNAMIC. arch/arm64/configs/defconfig sets
CONFIG_PCI_HOST_GENERIC=y, CONFIG_NXP_ENETC4=m and CONFIG_FSL_ENETC_VF=m, but
it does not set CONFIG_PCI_DYNAMIC_OF_NODES.
So on a default arm64 build, won't the only device behind the newly enabled
bus fail with -ENODEV? The PF creates the VFs through SR-IOV, so the VFs
would not appear either. This is still the case at the end of the series.
Would one of these work: giving the PF a static child node in DT, adding
CONFIG_PCI_DYNAMIC_OF_NODES=y to the arm64 defconfig, or having NXP_ENETC4
imply or select it?
[Severity: Medium]
With this bus enabled, the PPM PF and VFs become live DMA masters. The
netc_iep18 node in s32n79.dtsi has:
iommu-map = <0x0 &smmu 0xc0000 0x100>;
msi-map = <0x0 &its 0xc0000 0x100>;
dma-coherent;
The smmu node it points to is disabled at the SoC level:
smmu: iommu@4fc00000 {
compatible = "arm,smmu-v3";
...
status = "disabled";
No board DTS under arch/arm64/boot/dts/freescale/ enables it.
of_iommu_xlate() in drivers/iommu/of_iommu.c then ignores the mapping without
any message:
if (!of_device_is_available(iommu_spec->np))
return -ENODEV;
As a result, the PF and VFs fall back to dma-direct and use raw physical
addresses for the BD rings and buffers.
The SoC dtsi commit in this series says "DMA is translated by the SMMUv3
(iommu-map)". That does not seem to hold on this board.
Doesn't DMA now depend entirely on the SMMU state left by boot firmware?
Suppose firmware left the SMMU enabled with its own stream table, or set it
to abort unmatched streams. NETC DMA would then fault or be blocked, even
though the netdev probes.
The dma-coherent property makes Linux skip cache maintenance. That too relies
on stream attributes configured outside of Linux.
Should &smmu be enabled along with this bus? Alternatively, could the
firmware requirement be documented (SMMU in bypass with coherent attributes
for StreamIDs 0xc0000 and up) and the dtsi commit message corrected?
--
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/cover.1794581338.git.claudiu.manoil%40nxp.com
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH net-next v4 6/6] PCI: Create OF node for NXP ENETC pseudo-MAC functions
[not found] <cover.1794581338.git.claudiu.manoil@nxp.com>
` (4 preceding siblings ...)
2026-11-13 15:02 ` [PATCH net-next v4 5/6] arm64: dts: s32n79-rdb: enable NETC IEP18 Claudiu Manoil
@ 2026-11-13 15:02 ` Claudiu Manoil
2026-11-17 15:55 ` netdev-bot+sashiko
5 siblings, 1 reply; 10+ messages in thread
From: Claudiu Manoil @ 2026-11-13 15:02 UTC (permalink / raw)
To: netdev; +Cc: s32, Bjorn Helgaas, linux-pci, linux-kernel
NXP ENETCv4 pseudo-MAC (PPM) ports can be probed as pure PCI endpoints,
without any "ethernet" node in the static device tree. The ENETC4 driver
describes their fixed link by attaching a fixed-link node to the PCI
function's own OF node, which requires that node to exist.
Request the PCI core to synthesize a per-function OF node for these
devices by registering of_pci_make_dev_node() as a final fixup, following
the existing precedent for the Xilinx, Red Hat and EFAR devices. The
helper is a no-op when the function already has a static OF node, so this
has no effect on platforms that describe ENETC via static DT. It only
takes effect when CONFIG_PCI_DYNAMIC_OF_NODES is enabled.
Both ENETC4 PCI device IDs are covered (0xe101 and 0xe110), since either
can be instantiated as a pseudo-MAC port on S32N79.
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
---
v4: none
v3: none
v2: none (reworded the overlay reference to match patch 1's fixed-link
approach; no functional change)
drivers/pci/quirks.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
index 7f1f47ea2ee4..71f272d8a296 100644
--- a/drivers/pci/quirks.c
+++ b/drivers/pci/quirks.c
@@ -6563,6 +6563,8 @@ DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_EFAR, 0x9660, of_pci_make_dev_node);
DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_TOSHIBA, 0x0220, of_pci_make_dev_node);
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_PHILIPS, 0xe101, of_pci_make_dev_node);
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_PHILIPS, 0xe110, of_pci_make_dev_node);
/*
* Devices known to require a longer delay before first config space access
--
2.34.1
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH net-next v4 6/6] PCI: Create OF node for NXP ENETC pseudo-MAC functions
2026-11-13 15:02 ` [PATCH net-next v4 6/6] PCI: Create OF node for NXP ENETC pseudo-MAC functions Claudiu Manoil
@ 2026-11-17 15:55 ` netdev-bot+sashiko
0 siblings, 0 replies; 10+ messages in thread
From: netdev-bot+sashiko @ 2026-11-17 15:55 UTC (permalink / raw)
To: claudiu.manoil; +Cc: netdev, s32, bhelgaas, linux-pci, linux-kernel, kuba
Thank you for your contribution! Sashiko AI review found 2 potential
issue(s) to consider.
Critical: 0 · High: 0 · Medium: 1 · Low: 1
If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).
- [Medium] The two new fixups for 0xe101 and 0xe110 in
drivers/pci/quirks.c only do something when…
- [Low] The comment above the of_pci_make_dev_node() fixup block in
drivers/pci/quirks.c (lines 6553-6560) gives one reason for the…
--
PCI: Create OF node for NXP ENETC pseudo-MAC functions
This commit registers of_pci_make_dev_node() as a final fixup for the NXP
ENETC4 PCI device IDs 0xe101 and 0xe110. The new per-function OF node lets
the ENETC4 driver attach a fixed-link child to pseudo-MAC ports that have
no static DT node. The fixup only does anything when
CONFIG_PCI_DYNAMIC_OF_NODES is enabled.
> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
> index 7f1f47ea2ee41..71f272d8a2965 100644
> --- a/drivers/pci/quirks.c
> +++ b/drivers/pci/quirks.c
> @@ -6563,6 +6563,8 @@ DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_XILINX, 0x5021, of_pci_make_dev_node);
> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_REDHAT, 0x0005, of_pci_make_dev_node);
> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_EFAR, 0x9660, of_pci_make_dev_node);
> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_TOSHIBA, 0x0220, of_pci_make_dev_node);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_PHILIPS, 0xe101, of_pci_make_dev_node);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_PHILIPS, 0xe110, of_pci_make_dev_node);
[Severity: Medium]
With the default arm64 config, will these fixups do anything on the
S32N79 platform this series targets?
If CONFIG_PCI_DYNAMIC_OF_NODES is not set, of_pci_make_dev_node() is an
empty stub:
drivers/pci/pci.h:
#else
static inline void of_pci_make_dev_node(struct pci_dev *pdev) { }
PCI_DYNAMIC_OF_NODES is a user-visible bool, and nothing in the tree
selects it. NXP_ENETC4 selects OF_DYNAMIC, so there is no build problem.
However, it does not select or imply PCI_DYNAMIC_OF_NODES.
arch/arm64/configs/defconfig sets CONFIG_ARCH_S32=y and
CONFIG_NXP_ENETC4=m, but it does not set PCI_DYNAMIC_OF_NODES.
The s32n79.dtsi change in this series adds netc_iep18 without child nodes
for the PCI functions. So dev->of_node stays NULL for the pseudo-MAC PF,
and enetc4_link_init() fails the probe:
drivers/net/ethernet/freescale/enetc/enetc4_pf.c:enetc4_link_init() {
...
if (!node) {
dev_err(dev, "Missing of_node for Pseudo-MAC port\n");
return -ENODEV;
}
...
}
This error message does not mention the missing config option.
So with the default arm64 config, "arm64: dts: s32n79-rdb: enable NETC
IEP18" does not get the PPM PF/VF instances probed, which is what it sets
out to do.
The commit messages say this dependency is intentional, because the
driver also handles ports described by static DT. That is a reason not to
add a hard "depends on". Would an "imply PCI_DYNAMIC_OF_NODES" on
NXP_ENETC4, or a defconfig update, work instead? The Kconfig help text
notes that turning the option on also creates OF nodes for all PCI
bridges.
[Severity: Low]
This isn't a bug, but the comment above this block of fixups no longer
covers every entry:
/*
* For a PCI device with multiple downstream devices, its driver may use
* a flattened device tree to describe the downstream devices.
* To overlay the flattened device tree, the PCI device and all its ancestor
* devices need to have device tree nodes on system base device tree. Thus,
* before driver probing, it might need to add a device tree node as the final
* fixup.
*/
An ENETC4 pseudo-MAC function has no downstream devices and loads no FDT
overlay. enetc4_add_ppm_fixed_link() only applies an of_changeset that
adds a fixed-link child to the function's own node.
The PCI_DYNAMIC_OF_NODES help text in drivers/pci/Kconfig also describes
only the overlay use case.
Could the comment, and perhaps the help text, be updated to cover this
use?
--
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/cover.1794581338.git.claudiu.manoil%40nxp.com
^ permalink raw reply [flat|nested] 10+ messages in thread