[PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[Hee-Suk Kim]

Giga Lite is a Realtek proprietary mode which allows links of up to
500 Mbps over two-pair cabling. In South Korea, it is widely used to
support 500 Mbps plans by most ISPs. The vendor driver [1] sets PHY
page 0xa42, reg 0x14, bit 9 to allow it on RTL8168H and FP. The driver
does not handle RTL8168EP but I confirmed the same works for it by
setting the PHY register via SIOCSMIIREG on an SK Broadband 500 Mbps
plan [2].

With the bit set, the link still came up at about 470 Mbps even when
only 10/100 was advertised. So set it only when autonegotiating with
1000baseT/Full advertised, and clear it otherwise, to respect a lower
advertised speed. This matches what the vendor driver does for
RTL8168H and FP.

PHYSR reports 0x30 in its speed field to denote a Giga Lite link. This
value is not handled in rtlgen_decode_physr(), so the reported speed
varies between 100 and 1000 Mbps on the same link. Handle it to report
1000, as there is no 500 Mbps uAPI.

[1] https://github.com/mtorromeo/r8168/blob/b3e24e4aa74a3561618d0abf01b708bd55def57a/src/r8168_n.c#L6375-L6390
[2] https://www.tworld.co.kr/web/product/callplan/NI00000567

Assisted-by: LLM
Signed-off-by: Hee-Suk Kim <hskim@dilluti0n.com>
---
Test results on a ThinkPad T14 Gen 5 (RTL8168ep/8111ep, XID 502) with
an SK Broadband 500 Mbps plan ("before"/"after" = without/with this
patch, throughput by fast.com):

  kernel  bit 9          advertised   PHYSR  reported     throughput
  ------  -------------  -----------  -----  -----------  ----------
  before  clear          10/100/1000  0x10   100          94 Mbps
  before  set manually   10/100/1000  0x30   100 or 1000  470 Mbps
  after   set manually   10/100       0x30   1000         470 Mbps
  after   clear          10/100       0x10   100          95 Mbps
  after   set            10/100/1000  0x30   1000         470 Mbps

"set manually" means bit 9 was set via SIOCSMIIREG. Happy to share the
script if useful.

 MAINTAINERS                               |  1 +
 drivers/net/ethernet/realtek/r8169_main.c |  4 ++++
 drivers/net/phy/realtek/realtek_main.c    | 26 +++++++++++++++++++++++
 include/linux/realtek_phy.h               | 10 +++++++++
 4 files changed, 41 insertions(+)
 create mode 100644 include/linux/realtek_phy.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 509b35b5a0e7..a54370ae80a3 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -9789,6 +9789,7 @@ F:	include/linux/phy_fixed.h
 F:	include/linux/phy_link_topology.h
 F:	include/linux/phylib_stubs.h
 F:	include/linux/platform_data/mdio-bcm-unimac.h
+F:	include/linux/realtek_phy.h
 F:	include/net/phy/
 F:	include/trace/events/mdio.h
 F:	include/uapi/linux/mdio.h
diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
index 43c46d483eb8..fe49197c4d7b 100644
--- a/drivers/net/ethernet/realtek/r8169_main.c
+++ b/drivers/net/ethernet/realtek/r8169_main.c
@@ -28,6 +28,7 @@
 #include <linux/bitfield.h>
 #include <linux/phylink.h>
 #include <linux/prefetch.h>
+#include <linux/realtek_phy.h>
 #include <linux/ipv6.h>
 #include <linux/unaligned.h>
 #include <net/ip6_checksum.h>
@@ -5619,6 +5620,9 @@ static int r8169_mdio_register(struct rtl8169_private *tp)
 		return -EUNATCH;
 	}

+	if (tp->mac_version == RTL_GIGA_MAC_VER_51)
+		tp->phydev->dev_flags |= RTL_PHY_F_GIGA_LITE;
+
 	/* Keep EEE off but the version stays EEE-capable so that link-up
 	 * still clears the MAC TX-LPI bits.
 	 */
diff --git a/drivers/net/phy/realtek/realtek_main.c b/drivers/net/phy/realtek/realtek_main.c
index e14960192e79..319122270bca 100644
--- a/drivers/net/phy/realtek/realtek_main.c
+++ b/drivers/net/phy/realtek/realtek_main.c
@@ -14,6 +14,7 @@
 #include <linux/of.h>
 #include <linux/phy.h>
 #include <linux/pm_wakeirq.h>
+#include <linux/realtek_phy.h>
 #include <linux/netdevice.h>
 #include <linux/module.h>
 #include <linux/delay.h>
@@ -1706,6 +1707,7 @@ static void rtlgen_decode_physr(struct phy_device *phydev, int val)
 		phydev->speed = SPEED_100;
 		break;
 	case 0x0020:
+	case 0x0030:	/* Giga Lite */
 		phydev->speed = SPEED_1000;
 		break;
 	case 0x0200:
@@ -1735,6 +1737,29 @@ static void rtlgen_decode_physr(struct phy_device *phydev, int val)
 	}
 }

+static int rtlgen_config_aneg(struct phy_device *phydev)
+{
+	int ret = 0;
+
+	/* With Giga Lite enabled, the PHY can link in Giga Lite even
+	 * when 1000baseT/Full is not advertised.  Only allow it when
+	 * autonegotiating with 1000baseT/Full advertised, so that a
+	 * lower speed set by the user is respected.
+	 */
+	if (phydev->dev_flags & RTL_PHY_F_GIGA_LITE) {
+		bool en = phydev->autoneg == AUTONEG_ENABLE &&
+			  linkmode_test_bit(ETHTOOL_LINK_MODE_1000baseT_Full_BIT,
+					    phydev->advertising);
+
+		ret = phy_modify_paged_changed(phydev, 0xa42, 0x14, BIT(9),
+					       en ? BIT(9) : 0);
+		if (ret < 0)
+			return ret;
+	}
+
+	return __genphy_config_aneg(phydev, ret > 0);
+}
+
 static int rtlgen_read_status(struct phy_device *phydev)
 {
 	int ret, val;
@@ -3248,6 +3273,7 @@ static struct phy_driver realtek_drvs[] = {
 	}, {
 		.name		= "Generic FE-GE Realtek PHY",
 		.match_phy_device = rtlgen_match_phy_device,
+		.config_aneg	= rtlgen_config_aneg,
 		.read_status	= rtlgen_read_status,
 		.suspend	= genphy_suspend,
 		.resume		= rtlgen_resume,
diff --git a/include/linux/realtek_phy.h b/include/linux/realtek_phy.h
new file mode 100644
index 000000000000..7fcde616c070
--- /dev/null
+++ b/include/linux/realtek_phy.h
@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _REALTEK_PHY_H
+#define _REALTEK_PHY_H
+
+#include <linux/bits.h>
+
+/* phy_device::dev_flags */
+#define RTL_PHY_F_GIGA_LITE	BIT(0)
+
+#endif /* _REALTEK_PHY_H */

base-commit: d8674294aefef02266c4d47ad10131f1bffbe534
--
2.55.0

Re: [PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[Heiner Kallweit]

On 10.10.2026 15:53, Hee-Suk Kim wrote:
> Giga Lite is a Realtek proprietary mode which allows links of up to
> 500 Mbps over two-pair cabling. In South Korea, it is widely used to
> support 500 Mbps plans by most ISPs. The vendor driver [1] sets PHY
> page 0xa42, reg 0x14, bit 9 to allow it on RTL8168H and FP. The driver
> does not handle RTL8168EP but I confirmed the same works for it by
> setting the PHY register via SIOCSMIIREG on an SK Broadband 500 Mbps
> plan [2].
> 
> With the bit set, the link still came up at about 470 Mbps even when
> only 10/100 was advertised. So set it only when autonegotiating with
> 1000baseT/Full advertised, and clear it otherwise, to respect a lower
> advertised speed. This matches what the vendor driver does for
> RTL8168H and FP.
> 
> PHYSR reports 0x30 in its speed field to denote a Giga Lite link. This
> value is not handled in rtlgen_decode_physr(), so the reported speed
> varies between 100 and 1000 Mbps on the same link. Handle it to report
> 1000, as there is no 500 Mbps uAPI.
> 
> [1] https://github.com/mtorromeo/r8168/blob/b3e24e4aa74a3561618d0abf01b708bd55def57a/src/r8168_n.c#L6375-L6390
> [2] https://www.tworld.co.kr/web/product/callplan/NI00000567
> 
> Assisted-by: LLM
> Signed-off-by: Hee-Suk Kim <hskim@dilluti0n.com>
> ---
> Test results on a ThinkPad T14 Gen 5 (RTL8168ep/8111ep, XID 502) with
> an SK Broadband 500 Mbps plan ("before"/"after" = without/with this
> patch, throughput by fast.com):
> 
>   kernel  bit 9          advertised   PHYSR  reported     throughput
>   ------  -------------  -----------  -----  -----------  ----------
>   before  clear          10/100/1000  0x10   100          94 Mbps
>   before  set manually   10/100/1000  0x30   100 or 1000  470 Mbps
>   after   set manually   10/100       0x30   1000         470 Mbps
>   after   clear          10/100       0x10   100          95 Mbps
>   after   set            10/100/1000  0x30   1000         470 Mbps
> 
> "set manually" means bit 9 was set via SIOCSMIIREG. Happy to share the
> script if useful.
> 
>  MAINTAINERS                               |  1 +
>  drivers/net/ethernet/realtek/r8169_main.c |  4 ++++
>  drivers/net/phy/realtek/realtek_main.c    | 26 +++++++++++++++++++++++

It would be better to separate the patches for the Realtek PHY driver and
the MAC driver.

>  include/linux/realtek_phy.h               | 10 +++++++++
>  4 files changed, 41 insertions(+)
>  create mode 100644 include/linux/realtek_phy.h
> 
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 509b35b5a0e7..a54370ae80a3 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -9789,6 +9789,7 @@ F:	include/linux/phy_fixed.h
>  F:	include/linux/phy_link_topology.h
>  F:	include/linux/phylib_stubs.h
>  F:	include/linux/platform_data/mdio-bcm-unimac.h
> +F:	include/linux/realtek_phy.h
>  F:	include/net/phy/
>  F:	include/trace/events/mdio.h
>  F:	include/uapi/linux/mdio.h
> diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
> index 43c46d483eb8..fe49197c4d7b 100644
> --- a/drivers/net/ethernet/realtek/r8169_main.c
> +++ b/drivers/net/ethernet/realtek/r8169_main.c
> @@ -28,6 +28,7 @@
>  #include <linux/bitfield.h>
>  #include <linux/phylink.h>
>  #include <linux/prefetch.h>
> +#include <linux/realtek_phy.h>
>  #include <linux/ipv6.h>
>  #include <linux/unaligned.h>
>  #include <net/ip6_checksum.h>
> @@ -5619,6 +5620,9 @@ static int r8169_mdio_register(struct rtl8169_private *tp)
>  		return -EUNATCH;
>  	}
> 
> +	if (tp->mac_version == RTL_GIGA_MAC_VER_51)

Any specific reason you restrict the functionality to this chip version?
In the commit message you describe that vendor driver also supports the
500Mbps mode for RTL8168H and RTL8168FP.
Also vendor drivers r8125, r8126, r8127 include functionality for dealing
with 500mbps mode settings. So more chip versions seem to support it.
Hopefully Realtek people can provide some more insight.

> +		tp->phydev->dev_flags |= RTL_PHY_F_GIGA_LITE;
> +
>  	/* Keep EEE off but the version stays EEE-capable so that link-up
>  	 * still clears the MAC TX-LPI bits.
>  	 */
> diff --git a/drivers/net/phy/realtek/realtek_main.c b/drivers/net/phy/realtek/realtek_main.c
> index e14960192e79..319122270bca 100644
> --- a/drivers/net/phy/realtek/realtek_main.c
> +++ b/drivers/net/phy/realtek/realtek_main.c
> @@ -14,6 +14,7 @@
>  #include <linux/of.h>
>  #include <linux/phy.h>
>  #include <linux/pm_wakeirq.h>
> +#include <linux/realtek_phy.h>
>  #include <linux/netdevice.h>
>  #include <linux/module.h>
>  #include <linux/delay.h>
> @@ -1706,6 +1707,7 @@ static void rtlgen_decode_physr(struct phy_device *phydev, int val)
>  		phydev->speed = SPEED_100;
>  		break;
>  	case 0x0020:
> +	case 0x0030:	/* Giga Lite */
>  		phydev->speed = SPEED_1000;

So for Giga lite (500Mbps) a wrong speed (1Gbps) is reported to the user.
Not sure whether we want to add a SPEED_500 constant. On one hand it's a non-standard mode,
on the other hand it's apparently widely used in South Korea.
@Andrew: How do you see this?

Alternatively we could add a notification to the user if Giga lite is auto-negotiated.

>  		break;
>  	case 0x0200:
> @@ -1735,6 +1737,29 @@ static void rtlgen_decode_physr(struct phy_device *phydev, int val)
>  	}
>  }
> 
> +static int rtlgen_config_aneg(struct phy_device *phydev)
> +{
> +	int ret = 0;
> +
> +	/* With Giga Lite enabled, the PHY can link in Giga Lite even

As Giga lite is a Realtek-proprietary mode, you may want to add some words
about what Giga lite is.

> +	 * when 1000baseT/Full is not advertised.  Only allow it when
> +	 * autonegotiating with 1000baseT/Full advertised, so that a
> +	 * lower speed set by the user is respected.
> +	 */
> +	if (phydev->dev_flags & RTL_PHY_F_GIGA_LITE) {
> +		bool en = phydev->autoneg == AUTONEG_ENABLE &&
> +			  linkmode_test_bit(ETHTOOL_LINK_MODE_1000baseT_Full_BIT,
> +					    phydev->advertising);
> +
> +		ret = phy_modify_paged_changed(phydev, 0xa42, 0x14, BIT(9),
> +					       en ? BIT(9) : 0);
> +		if (ret < 0)
> +			return ret;
> +	}
> +
> +	return __genphy_config_aneg(phydev, ret > 0);
> +}
> +
>  static int rtlgen_read_status(struct phy_device *phydev)
>  {
>  	int ret, val;
> @@ -3248,6 +3273,7 @@ static struct phy_driver realtek_drvs[] = {
>  	}, {
>  		.name		= "Generic FE-GE Realtek PHY",
>  		.match_phy_device = rtlgen_match_phy_device,
> +		.config_aneg	= rtlgen_config_aneg,
>  		.read_status	= rtlgen_read_status,
>  		.suspend	= genphy_suspend,
>  		.resume		= rtlgen_resume,
> diff --git a/include/linux/realtek_phy.h b/include/linux/realtek_phy.h
> new file mode 100644
> index 000000000000..7fcde616c070
> --- /dev/null
> +++ b/include/linux/realtek_phy.h
> @@ -0,0 +1,10 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef _REALTEK_PHY_H
> +#define _REALTEK_PHY_H
> +
> +#include <linux/bits.h>
> +
> +/* phy_device::dev_flags */
> +#define RTL_PHY_F_GIGA_LITE	BIT(0)
> +
> +#endif /* _REALTEK_PHY_H */
> 
> base-commit: d8674294aefef02266c4d47ad10131f1bffbe534
> --
> 2.55.0

Re: [PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[Andrew Lunn]

On Sun, Oct 11, 2026 at 10:59:42PM +0200, Heiner Kallweit wrote:
> On 10.10.2026 15:53, Hee-Suk Kim wrote:
> > Giga Lite is a Realtek proprietary mode which allows links of up to
> > 500 Mbps over two-pair cabling. In South Korea, it is widely used to
> > support 500 Mbps plans by most ISPs. The vendor driver [1] sets PHY
> > page 0xa42, reg 0x14, bit 9 to allow it on RTL8168H and FP. The driver
> > does not handle RTL8168EP but I confirmed the same works for it by
> > setting the PHY register via SIOCSMIIREG on an SK Broadband 500 Mbps
> > plan [2].
> > 
> > With the bit set, the link still came up at about 470 Mbps even when
> > only 10/100 was advertised. So set it only when autonegotiating with
> > 1000baseT/Full advertised, and clear it otherwise, to respect a lower
> > advertised speed. This matches what the vendor driver does for
> > RTL8168H and FP.
> > 
> > PHYSR reports 0x30 in its speed field to denote a Giga Lite link. This
> > value is not handled in rtlgen_decode_physr(), so the reported speed
> > varies between 100 and 1000 Mbps on the same link. Handle it to report
> > 1000, as there is no 500 Mbps uAPI.
> > 
> > [1] https://github.com/mtorromeo/r8168/blob/b3e24e4aa74a3561618d0abf01b708bd55def57a/src/r8168_n.c#L6375-L6390
> > [2] https://www.tworld.co.kr/web/product/callplan/NI00000567

I'm in agreement with Heiner here. If we add this, we do it properly.

Please add ETHTOOL_LINK_MODE_500baseT2_Full. Make the PHY report this
link mode in it features, add the speed, allow it to be negotiated,
reported in ethtool etc.

	Andrew

Re: [PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[Daniel Golle]

On Mon, Oct 12, 2026 at 02:46:27PM +0200, Andrew Lunn wrote:
> On Sun, Oct 11, 2026 at 10:59:42PM +0200, Heiner Kallweit wrote:
> > On 10.10.2026 15:53, Hee-Suk Kim wrote:
> > > Giga Lite is a Realtek proprietary mode which allows links of up to
> > > 500 Mbps over two-pair cabling. In South Korea, it is widely used to
> > > support 500 Mbps plans by most ISPs. The vendor driver [1] sets PHY
> > > page 0xa42, reg 0x14, bit 9 to allow it on RTL8168H and FP. The driver
> > > does not handle RTL8168EP but I confirmed the same works for it by
> > > setting the PHY register via SIOCSMIIREG on an SK Broadband 500 Mbps
> > > plan [2].
> > > 
> > > With the bit set, the link still came up at about 470 Mbps even when
> > > only 10/100 was advertised. So set it only when autonegotiating with
> > > 1000baseT/Full advertised, and clear it otherwise, to respect a lower
> > > advertised speed. This matches what the vendor driver does for
> > > RTL8168H and FP.
> > > 
> > > PHYSR reports 0x30 in its speed field to denote a Giga Lite link. This
> > > value is not handled in rtlgen_decode_physr(), so the reported speed
> > > varies between 100 and 1000 Mbps on the same link. Handle it to report
> > > 1000, as there is no 500 Mbps uAPI.
> > > 
> > > [1] https://github.com/mtorromeo/r8168/blob/b3e24e4aa74a3561618d0abf01b708bd55def57a/src/r8168_n.c#L6375-L6390
> > > [2] https://www.tworld.co.kr/web/product/callplan/NI00000567
> 
> I'm in agreement with Heiner here. If we add this, we do it properly.
> 
> Please add ETHTOOL_LINK_MODE_500baseT2_Full. Make the PHY report this
> link mode in it features, add the speed, allow it to be negotiated,
> reported in ethtool etc.

While at it also consider ETHTOOL_LINK_MODE_1000baseT2_Full which is
supported as "2.5G Lite" by the RTL8221/RTL8226 2.5G PHY family.

Re: [PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[Hee-Suk Kim]

Heiner Kallweit <hkallweit1@gmail.com> writes:

> On 10.10.2026 15:53, Hee-Suk Kim wrote:

>> @@ -5619,6 +5620,9 @@ static int r8169_mdio_register(struct rtl8169_private *tp)
>>  		return -EUNATCH;
>>  	}
>> 
>> +	if (tp->mac_version == RTL_GIGA_MAC_VER_51)
>
> Any specific reason you restrict the functionality to this chip version?
> In the commit message you describe that vendor driver also supports the
> 500Mbps mode for RTL8168H and RTL8168FP.

No, just that RTL8168EP is the only one I could test. I'll include
RTL8168H and RTL8168FP in v2.

> Also vendor drivers r8125, r8126, r8127 include functionality for dealing
> with 500mbps mode settings. So more chip versions seem to support it.
> Hopefully Realtek people can provide some more insight.

I've found all three [1][2][3] do the same as r8168, so I could cover
them as well. It would be good if Realtek people could confirm that
PHYSR reports 0x30 for a Giga Lite link on them, since the vendor
drivers read the speed from the MAC instead.

They also set bits of OCP 0xa5ea when 2.5G/5G/10G is advertised, which
appear to enable the corresponding Lite modes (half the speed over two
pairs). I'd leave those out of this series for now, as I'm not aware of
a use case for them.

[1] https://github.com/awesometic/realtek-r8125-dkms/blob/9a152bf920334aead090d2fb4d91f67a84923123/src/r8125_n.c#L5948-L5980
[2] https://github.com/openwrt/rtl8126/blob/27721fbedc45897cf1f155f5eb44de76962a1ba8/r8126_n.c#L5009-L5033
[3] https://github.com/openwrt/rtl8127/blob/f80bc64922ac76f90618f61245fb29743c018d0a/r8127_n.c#L5330-L5359

> So for Giga lite (500Mbps) a wrong speed (1Gbps) is reported to the user.
> Not sure whether we want to add a SPEED_500 constant. On one hand it's a non-standard mode,
> on the other hand it's apparently widely used in South Korea.
> @Andrew: How do you see this?
>
> Alternatively we could add a notification to the user if Giga lite is auto-negotiated.
>

For reference, the vendor drivers report a Giga Lite link as 1000.

FWIW I'd prefer 500, just because userspace would show it as is.

Thanks,
hsk

Re: [PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[Hee-Suk Kim]

Daniel Golle <daniel@makrotopia.org> writes:

> On Mon, Oct 12, 2026 at 02:46:27PM +0200, Andrew Lunn wrote:
>> Please add ETHTOOL_LINK_MODE_500baseT2_Full. Make the PHY report this
>> link mode in it features, add the speed, allow it to be negotiated,
>> reported in ethtool etc.

Will do in v2.

> While at it also consider ETHTOOL_LINK_MODE_1000baseT2_Full which is
> supported as "2.5G Lite" by the RTL8221/RTL8226 2.5G PHY family.

Happy to add it as well, but I'd need to know what PHYSR reports for a
2.5G Lite link on those PHYs, as I can't test it myself. Do you happen
to know?

Thanks,
hsk

Re: [PATCH net-next] r8169: allow Giga Lite fallback on RTL8168EP

[Heiner Kallweit]

On 12.10.2026 16:53, Hee-Suk Kim wrote:
> Daniel Golle <daniel@makrotopia.org> writes:
> 
>> On Mon, Oct 12, 2026 at 02:46:27PM +0200, Andrew Lunn wrote:
>>> Please add ETHTOOL_LINK_MODE_500baseT2_Full. Make the PHY report this
>>> link mode in it features, add the speed, allow it to be negotiated,
>>> reported in ethtool etc.
> 
> Will do in v2.
> 
>> While at it also consider ETHTOOL_LINK_MODE_1000baseT2_Full which is
>> supported as "2.5G Lite" by the RTL8221/RTL8226 2.5G PHY family.
> 
> Happy to add it as well, but I'd need to know what PHYSR reports for a
> 2.5G Lite link on those PHYs, as I can't test it myself. Do you happen
> to know?
> 
Also the needed register configuration may be different. Would be helpful to get advise
from Javen/Realtek on:
- Which lite modes exist?
- Which chip versions support which lite modes?
- What is the needed configuration per lite mode (according to vendor drivers the config may
  be different per advertised speeds)

I can't tell as I have no access to Realtek documentation.

> Thanks,
> hsk