[PATCH v9] phy: Add generic PHY driver used by MACB/GEM on EyeQ5

[PATCH v9] phy: Add generic PHY driver used by MACB/GEM on EyeQ5

[Théo Lebrun]

EyeQ5 SoCs integrate two GEM instances. A system-controller register
region named "OLB" has some control over the Ethernet PHY integration.

Extend the current OLB ecosystem with a new generic PHY driver.
 - OLB is carried by one main platform driver: clk-eyeq.
 - It instantiates auxiliary devices: reset-eyeq & pinctrl-eyeq5.
 - We add a new one: phy-eyeq5-eth.

About related patches:

 - The MACB series [1] has been merged in v6.19-rc1. It makes MACB
   consume a generic PHY from devicetree with the EyeQ5 compatible.

 - clk patches are on the lkml [3]; they make clk-eyeq instantiate this
   new auxiliary device. They also ensure we get a dev->of_node
   assigned. Patches used to be [2] in the same series.

 - MIPS patches are on the lkml [4]; they add MACB/GEM instances in
   devicetree and their associated PHYs. They also update dt-bindings
   to reflect this new feature OLB provides. Patches used to be [2] in
   the same series.

Have a nice day,
Thanks!
Théo

[0]: https://lore.kernel.org/lkml/20250627-macb-v2-15-ff8207d0bb77@bootlin.com/
[1]: https://lore.kernel.org/lkml/20251022-macb-eyeq5-v2-0-7c140abb0581@bootlin.com/
[2]: https://lore.kernel.org/all/20260127-macb-phy-v6-0-cdd840588188@bootlin.com/
[3]: https://lore.kernel.org/all/20260225-macb-phy-v7-0-665bd8619d51@bootlin.com/
[4]: https://lore.kernel.org/all/20260225-macb-phy-v7-0-d3c9842ec931@bootlin.com/

Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
---
Changes in v9:
- Explicitly cast dev_get_platdata() from `void *` to `void __iomem *`,
  to fix "different address spaces" sparse warning.
- Link to v8: https://patch.msgid.link/20260306-macb-phy-v8-0-b5c48ee61402@bootlin.com

Changes in v8:
- Rebase upon linux-phy/next. Drop sorting Kconfig/Makefile patch
  because one exists upstream.
- Instead of storing a phy_interface_t to know which submode to target,
  store a custom enum with two states (RGMII/SGMII). Ignore RGMII
  timings in this generic PHY driver because they are handled by the
  net PHY.
- eq5_phy_set_mode() calls into eq5_phy_validate() and now propagates
  its error return value. Previously it shadowed EINVAL with EOPNOTSUPP.
- Link to v7: https://lore.kernel.org/r/20260225-macb-phy-v7-0-e5211a61db56@bootlin.com

Changes in v7:
- Separate PHY / clk / MIPS patches into three series.
- Implement phy_validate().
- phy_power_on() now supports being called without a prior
  phy_set_mode() because at probe we read the hardware state to
  initialise inst->phy_instance.
- phy_set_mode() now support being called while the PHY is powered on.
- Add sgmii_support bool for each PHY instance to reject SGMII
  configuration on PHY 1 which only supports RGMII.
- Drop dev_dbg() calls.
- Drop readl(gp) in phy_init().
- Replace inst->priv field by inst->dev; that is the only value we need
  from the driver private data. Drop priv->dev field that is unused.
- Call into phy_exit() from phy_init() as the sequence is the same.
  Add comment to explain the reasoning.
- Take Reviewed-by: Luca on "phy: sort Kconfig and Makefile".
- Rebase onto v7.0-rc1 and test on EyeQ5. Only diff to report:
  PHY_COMMON_PROPS and PHY_COMMON_PROPS_TEST are kept at the top in the
  sorting patch; we do not strictly respect an alphabetical ordering.
- Link to v6: https://lore.kernel.org/r/20260127-macb-phy-v6-0-cdd840588188@bootlin.com

Changes in v6:
- Rebase upon v6.19-rc7; nothing to report.
- Add new patch "phy: sort Kconfig and Makefile".
- phy-eyeq5-eth: drop useless explicit __iomem cast to
  dev_get_platdata() return value.
- I did *not* drop the Kconfig `default MACH_EYEQ5` nor driver
  `dev_dbg()`. I think both are useful and should be kept. See
  last revision discussion here:
  https://lore.kernel.org/lkml/DFGSMN8268O0.33TYCQDBVHUHZ@bootlin.com/
- Link to v5: https://lore.kernel.org/r/20251215-macb-phy-v5-0-a9dfea39da34@bootlin.com

Changes in v5:
- phy-eyeq5-eth:
  - fix #includes: add delay, gfp_types, module and drop array_size,
    bug, cleanup, container_of, lockdep, mutex.
  - eq5_phy_xlate(): avoid magic value, use EQ5_PHY_COUNT.
  - use dev_err_probe() in error cases of devm_phy_create() and
    devm_of_phy_provider_register().
- 3x Reviewed-by: Luca Ceresoli.
- Add Neil Armstrong to Cc as new PHY subsystem reviewer.
- Rebase on v6.19-rc1, tested on hardware, no changes.
- Link to v4: https://lore.kernel.org/r/20251124-macb-phy-v4-0-955c625a81a7@bootlin.com

Changes in v4:
- Append my SoB to Jerome's patch:
  [PATCH v4 3/7] clk: eyeq: use the auxiliary device creation helper
- Rebase on net-next & linux-{clk,mips,phy}. Nothing to report.
- Link to v3: https://lore.kernel.org/r/20251119-macb-phy-v3-0-e9a7be186a33@bootlin.com

Changes in v3:
- Take Philipp Zabel's Reviewed-by & Acked-by trailers on reset patch.
- Take Thomas Bogendoerfer's two Acked-by trailers on DT patches.
- Rebase on net-next & test on target. Nothing to report.
- Link to v2: https://lore.kernel.org/r/20251101-macb-phy-v2-0-c1519eef16d3@bootlin.com

Changes in v2:
- Take Acked-by: Conor Dooley on dt-bindings-patch.
- s/%ld/%tu/ for printing ptrdiff_t; warnings on 32-bit archs.
  Reported by NIPA's netdev/build_32bit test.
  https://patchwork.kernel.org/project/netdevbpf/patch/20251021-macb-eyeq5-v1-7-3b0b5a9d2f85@bootlin.com/
  https://netdev.bots.linux.dev/static/nipa/1014126/14277857/build_32bit/stderr
- Link to v1: https://lore.kernel.org/r/20251022-macb-phy-v1-0-f29f28fae721@bootlin.com

Changes since MACB V1:
- Drop the old "mobileye,olb" properties from DT patches; found while
  running dtbs_check and dt_binding_check.
- Drop all patches targeting net-next. That is MACB dt-bindings patch
  and MACB driver code. See there here [1].
- Link to v1: https://lore.kernel.org/lkml/20251021-macb-eyeq5-v1-0-3b0b5a9d2f85@bootlin.com/

Past versions of MACB patches:
 - March 2025: [PATCH net-next 00/13] Support the Cadence MACB/GEM
   instances on Mobileye EyeQ5 SoCs
   https://lore.kernel.org/lkml/20250321-macb-v1-0-537b7e37971d@bootlin.com/
 - June 2025: [PATCH net-next v2 00/18] Support the Cadence MACB/GEM
   instances on Mobileye EyeQ5 SoCs
   https://lore.kernel.org/lkml/20250627-macb-v2-0-ff8207d0bb77@bootlin.com/
 - August 2025: [PATCH net v3 00/16] net: macb: various fixes & cleanup
   https://lore.kernel.org/lkml/20250808-macb-fixes-v3-0-08f1fcb5179f@bootlin.com/

---
Théo Lebrun (1):
      phy: Add driver for EyeQ5 Ethernet PHY wrapper

 MAINTAINERS                 |   1 +
 drivers/phy/Kconfig         |  13 ++
 drivers/phy/Makefile        |   1 +
 drivers/phy/phy-eyeq5-eth.c | 280 ++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 295 insertions(+)
---
base-commit: caf08514bbee0736c31d8d4f406e3415cdf726bb
change-id: 20251022-macb-phy-21bc4e1dfbb7

Best regards,
--  
Théo Lebrun <theo.lebrun@bootlin.com>

[PATCH v9] phy: Add driver for EyeQ5 Ethernet PHY wrapper

[Théo Lebrun]

EyeQ5 embeds a system-controller called OLB. It features many unrelated
registers, and some of those are registers used to configure the
integration of the RGMII/SGMII Cadence PHY used by MACB/GEM instances.

Wrap in a neat generic PHY provider, exposing two PHYs with standard
phy_init() / phy_set_mode() / phy_power_on() operations.

Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
---
 MAINTAINERS                 |   1 +
 drivers/phy/Kconfig         |  13 ++
 drivers/phy/Makefile        |   1 +
 drivers/phy/phy-eyeq5-eth.c | 280 ++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 295 insertions(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index 55af015174a5..6bc2ae3bbd4b 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -17812,6 +17812,7 @@ F:	arch/mips/boot/dts/mobileye/
 F:	arch/mips/configs/eyeq5_defconfig
 F:	arch/mips/mobileye/board-epm5.its.S
 F:	drivers/clk/clk-eyeq.c
+F:	drivers/phy/phy-eyeq5-eth.c
 F:	drivers/pinctrl/pinctrl-eyeq5.c
 F:	drivers/reset/reset-eyeq.c
 F:	include/dt-bindings/clock/mobileye,eyeq5-clk.h
diff --git a/drivers/phy/Kconfig b/drivers/phy/Kconfig
index 3970aa1f300f..cb973a5c0d28 100644
--- a/drivers/phy/Kconfig
+++ b/drivers/phy/Kconfig
@@ -67,6 +67,19 @@ config PHY_CAN_TRANSCEIVER
 	  functional modes using gpios and sets the attribute max link
 	  rate, for CAN drivers.
 
+config PHY_EYEQ5_ETH
+	tristate "Ethernet PHY Driver on EyeQ5"
+	depends on OF
+	depends on MACH_EYEQ5 || COMPILE_TEST
+	select AUXILIARY_BUS
+	select GENERIC_PHY
+	default MACH_EYEQ5
+	help
+	  Enable this to support the Ethernet PHY integrated on EyeQ5.
+	  It supports both RGMII and SGMII. Registers are located in a
+	  shared register region called OLB. If M is selected, the
+	  module will be called phy-eyeq5-eth.
+
 config PHY_GOOGLE_USB
 	tristate "Google Tensor SoC USB PHY driver"
 	select GENERIC_PHY
diff --git a/drivers/phy/Makefile b/drivers/phy/Makefile
index f49d83f00a3d..05be5759cd10 100644
--- a/drivers/phy/Makefile
+++ b/drivers/phy/Makefile
@@ -9,6 +9,7 @@ obj-$(CONFIG_GENERIC_PHY)		+= phy-core.o
 obj-$(CONFIG_GENERIC_PHY_MIPI_DPHY)	+= phy-core-mipi-dphy.o
 obj-$(CONFIG_PHY_AIROHA_PCIE)		+= phy-airoha-pcie.o
 obj-$(CONFIG_PHY_CAN_TRANSCEIVER)	+= phy-can-transceiver.o
+obj-$(CONFIG_PHY_EYEQ5_ETH)		+= phy-eyeq5-eth.o
 obj-$(CONFIG_PHY_GOOGLE_USB)		+= phy-google-usb.o
 obj-$(CONFIG_USB_LGM_PHY)		+= phy-lgm-usb.o
 obj-$(CONFIG_PHY_LPC18XX_USB_OTG)	+= phy-lpc18xx-usb-otg.o
diff --git a/drivers/phy/phy-eyeq5-eth.c b/drivers/phy/phy-eyeq5-eth.c
new file mode 100644
index 000000000000..c03d77c360f7
--- /dev/null
+++ b/drivers/phy/phy-eyeq5-eth.c
@@ -0,0 +1,280 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#include <linux/auxiliary_bus.h>
+#include <linux/bitfield.h>
+#include <linux/bits.h>
+#include <linux/delay.h>
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/errno.h>
+#include <linux/gfp_types.h>
+#include <linux/init.h>
+#include <linux/io.h>
+#include <linux/iopoll.h>
+#include <linux/mod_devicetable.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/phy.h>
+#include <linux/phy/phy.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+
+#define EQ5_PHY_COUNT	2
+
+#define EQ5_PHY0_GP	0x128
+#define EQ5_PHY1_GP	0x12c
+#define EQ5_PHY0_SGMII	0x134
+#define EQ5_PHY1_SGMII	0x138
+
+#define EQ5_GP_TX_SWRST_DIS	BIT(0)		// Tx SW reset
+#define EQ5_GP_TX_M_CLKE	BIT(1)		// Tx M clock enable
+#define EQ5_GP_SYS_SWRST_DIS	BIT(2)		// Sys SW reset
+#define EQ5_GP_SYS_M_CLKE	BIT(3)		// Sys clock enable
+#define EQ5_GP_SGMII_MODE	BIT(4)		// SGMII mode
+#define EQ5_GP_RGMII_DRV	GENMASK(8, 5)	// RGMII drive strength
+
+#define EQ5_SGMII_PWR_EN	BIT(0)
+#define EQ5_SGMII_RST_DIS	BIT(1)
+#define EQ5_SGMII_PLL_EN	BIT(2)
+#define EQ5_SGMII_SIG_DET_SW	BIT(3)
+#define EQ5_SGMII_PWR_STATE	BIT(4)
+#define EQ5_SGMII_PLL_ACK	BIT(18)
+#define EQ5_SGMII_PWR_STATE_ACK	GENMASK(24, 20)
+
+/*
+ * Instead of storing a phy_interface_t, we store this enum.
+ *
+ * We do not deal with RGMII timings in this generic PHY driver,
+ * it is all handled inside the net PHY.
+ */
+enum eq5_phy_submode {
+	EQ5_PHY_SUBMODE_SGMII,
+	EQ5_PHY_SUBMODE_RGMII,
+};
+
+struct eq5_phy_inst {
+	struct device		*dev;
+	struct phy		*phy;
+	void __iomem		*gp, *sgmii;
+	enum eq5_phy_submode	submode;
+	bool			sgmii_support;
+};
+
+struct eq5_phy_private {
+	struct eq5_phy_inst	phys[EQ5_PHY_COUNT];
+};
+
+static int eq5_phy_exit(struct phy *phy)
+{
+	struct eq5_phy_inst *inst = phy_get_drvdata(phy);
+
+	writel(0, inst->gp);
+	writel(0, inst->sgmii);
+	udelay(5); /* settling time */
+	return 0;
+}
+
+static int eq5_phy_init(struct phy *phy)
+{
+	struct eq5_phy_inst *inst = phy_get_drvdata(phy);
+	u32 reg;
+
+	/*
+	 * Hardware stops listening to our instructions once it is started.
+	 * It must be reset to reconfigure it.
+	 */
+	eq5_phy_exit(phy);
+
+	reg = EQ5_GP_TX_SWRST_DIS | EQ5_GP_TX_M_CLKE |
+	      EQ5_GP_SYS_SWRST_DIS | EQ5_GP_SYS_M_CLKE |
+	      FIELD_PREP(EQ5_GP_RGMII_DRV, 0x9);
+	writel(reg, inst->gp);
+
+	return 0;
+}
+
+static int eq5_phy_power_on(struct phy *phy)
+{
+	struct eq5_phy_inst *inst = phy_get_drvdata(phy);
+	u32 reg;
+
+	if (inst->submode == EQ5_PHY_SUBMODE_SGMII) {
+		writel(readl(inst->gp) | EQ5_GP_SGMII_MODE, inst->gp);
+
+		reg = EQ5_SGMII_PWR_EN | EQ5_SGMII_RST_DIS | EQ5_SGMII_PLL_EN;
+		writel(reg, inst->sgmii);
+
+		if (readl_poll_timeout(inst->sgmii, reg,
+				       reg & EQ5_SGMII_PLL_ACK, 1, 100)) {
+			dev_err(inst->dev, "PLL timeout\n");
+			return -ETIMEDOUT;
+		}
+
+		reg = readl(inst->sgmii);
+		reg |= EQ5_SGMII_PWR_STATE | EQ5_SGMII_SIG_DET_SW;
+		writel(reg, inst->sgmii);
+	} else {
+		writel(readl(inst->gp) & ~EQ5_GP_SGMII_MODE, inst->gp);
+		writel(0, inst->sgmii);
+	}
+
+	return 0;
+}
+
+static int eq5_phy_power_off(struct phy *phy)
+{
+	struct eq5_phy_inst *inst = phy_get_drvdata(phy);
+
+	writel(readl(inst->gp) & ~EQ5_GP_SGMII_MODE, inst->gp);
+	writel(0, inst->sgmii);
+
+	return 0;
+}
+
+static int eq5_phy_validate(struct phy *phy, enum phy_mode mode, int submode,
+			    union phy_configure_opts *opts)
+{
+	struct eq5_phy_inst *inst = phy_get_drvdata(phy);
+
+	if (mode != PHY_MODE_ETHERNET)
+		return -EINVAL;
+
+	if (phy_interface_mode_is_rgmii(submode))
+		return 0;
+
+	if (inst->sgmii_support && submode == PHY_INTERFACE_MODE_SGMII)
+		return 0;
+
+	return -EINVAL;
+}
+
+static int eq5_phy_set_mode(struct phy *phy, enum phy_mode mode, int submode)
+{
+	struct eq5_phy_inst *inst = phy_get_drvdata(phy);
+	enum eq5_phy_submode target_submode;
+	int ret;
+
+	ret = eq5_phy_validate(phy, mode, submode, NULL);
+	if (ret)
+		return ret;
+
+	if (submode == PHY_INTERFACE_MODE_SGMII)
+		target_submode = EQ5_PHY_SUBMODE_SGMII;
+	else
+		target_submode = EQ5_PHY_SUBMODE_RGMII;
+
+	if (target_submode == inst->submode)
+		return 0;
+
+	inst->submode = target_submode;
+
+	if (phy->power_count) {
+		eq5_phy_init(phy);
+		return eq5_phy_power_on(phy);
+	}
+
+	return 0;
+}
+
+static const struct phy_ops eq5_phy_ops = {
+	.init		= eq5_phy_init,
+	.exit		= eq5_phy_exit,
+	.power_on	= eq5_phy_power_on,
+	.power_off	= eq5_phy_power_off,
+	.set_mode	= eq5_phy_set_mode,
+	.validate	= eq5_phy_validate,
+};
+
+static struct phy *eq5_phy_xlate(struct device *dev,
+				 const struct of_phandle_args *args)
+{
+	struct eq5_phy_private *priv = dev_get_drvdata(dev);
+
+	if (args->args_count != 1 || args->args[0] >= EQ5_PHY_COUNT)
+		return ERR_PTR(-EINVAL);
+
+	return priv->phys[args->args[0]].phy;
+}
+
+static int eq5_phy_probe_phy(struct device *dev, struct eq5_phy_private *priv,
+			     unsigned int index, void __iomem *base,
+			     unsigned int gp, unsigned int sgmii,
+			     bool sgmii_support)
+{
+	struct eq5_phy_inst *inst = &priv->phys[index];
+	struct phy *phy;
+
+	phy = devm_phy_create(dev, dev->of_node, &eq5_phy_ops);
+	if (IS_ERR(phy))
+		return dev_err_probe(dev, PTR_ERR(phy),
+				     "failed to create PHY %u\n", index);
+
+	inst->dev = dev;
+	inst->phy = phy;
+	inst->gp = base + gp;
+	inst->sgmii = base + sgmii;
+	inst->sgmii_support = sgmii_support;
+	phy_set_drvdata(phy, inst);
+
+	/*
+	 * Init inst->submode based on probe hardware state, allowing
+	 * consumers to power us on without first setting the mode.
+	 */
+	if (sgmii_support && (readl(inst->gp) & EQ5_GP_SGMII_MODE))
+		inst->submode = EQ5_PHY_SUBMODE_SGMII;
+	else
+		inst->submode = EQ5_PHY_SUBMODE_RGMII;
+
+	return 0;
+}
+
+static int eq5_phy_probe(struct auxiliary_device *adev,
+			 const struct auxiliary_device_id *id)
+{
+	struct device *dev = &adev->dev;
+	struct phy_provider *provider;
+	struct eq5_phy_private *priv;
+	void __iomem *base;
+	int ret;
+
+	priv = devm_kzalloc(dev, sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+
+	dev_set_drvdata(dev, priv);
+
+	base = (void __iomem *)dev_get_platdata(dev);
+
+	ret = eq5_phy_probe_phy(dev, priv, 0, base, EQ5_PHY0_GP,
+				EQ5_PHY0_SGMII, true);
+	if (ret)
+		return ret;
+
+	ret = eq5_phy_probe_phy(dev, priv, 1, base, EQ5_PHY1_GP,
+				EQ5_PHY1_SGMII, false);
+	if (ret)
+		return ret;
+
+	provider = devm_of_phy_provider_register(dev, eq5_phy_xlate);
+	if (IS_ERR(provider))
+		return dev_err_probe(dev, PTR_ERR(provider),
+				     "registering provider failed\n");
+
+	return 0;
+}
+
+static const struct auxiliary_device_id eq5_phy_id_table[] = {
+	{ .name = "clk_eyeq.phy" },
+	{}
+};
+MODULE_DEVICE_TABLE(auxiliary, eq5_phy_id_table);
+
+static struct auxiliary_driver eq5_phy_driver = {
+	.probe = eq5_phy_probe,
+	.id_table = eq5_phy_id_table,
+};
+module_auxiliary_driver(eq5_phy_driver);
+
+MODULE_DESCRIPTION("EyeQ5 Ethernet PHY driver");
+MODULE_AUTHOR("Théo Lebrun <theo.lebrun@bootlin.com>");
+MODULE_LICENSE("GPL");

-- 
2.53.0

Re: [PATCH v9] phy: Add driver for EyeQ5 Ethernet PHY wrapper

[Vladimir Oltean]

On Mon, Mar 09, 2026 at 03:37:34PM +0100, Théo Lebrun wrote:
> EyeQ5 embeds a system-controller called OLB. It features many unrelated
> registers, and some of those are registers used to configure the
> integration of the RGMII/SGMII Cadence PHY used by MACB/GEM instances.
> 
> Wrap in a neat generic PHY provider, exposing two PHYs with standard
> phy_init() / phy_set_mode() / phy_power_on() operations.
> 
> Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com>
> ---

Reviewed-by: Vladimir Oltean <olteanv@gmail.com>

Re: [PATCH v9] phy: Add generic PHY driver used by MACB/GEM on EyeQ5

[Vinod Koul]

On Mon, 09 Mar 2026 15:37:33 +0100, Théo Lebrun wrote:
> EyeQ5 SoCs integrate two GEM instances. A system-controller register
> region named "OLB" has some control over the Ethernet PHY integration.
> 
> Extend the current OLB ecosystem with a new generic PHY driver.
>  - OLB is carried by one main platform driver: clk-eyeq.
>  - It instantiates auxiliary devices: reset-eyeq & pinctrl-eyeq5.
>  - We add a new one: phy-eyeq5-eth.
> 
> [...]

Applied, thanks!

[1/1] phy: Add driver for EyeQ5 Ethernet PHY wrapper
      commit: b6e33443876d0ca7e93cf949455e3c1a1a0aae24

Best regards,
-- 
~Vinod