Merge branch 'ethtool-introduce-phy-mse-diagnostics-uapi-and-drivers'

Oleksij Rempel says:

====================
ethtool: introduce PHY MSE diagnostics UAPI and drivers

This series introduces a generic kernel-userspace API for retrieving PHY
Mean Square Error (MSE) diagnostics, together with netlink integration,
a fast-path reporting hook in LINKSTATE_GET, and initial driver
implementations for the KSZ9477 and DP83TD510E PHYs.

MSE is defined by the OPEN Alliance "Advanced diagnostic features for
100BASE-T1 automotive Ethernet PHYs" specification [1] as a measure of
slicer error rate, typically used internally to derive the Signal
Quality Indicator (SQI). While SQI is useful as a normalized quality
index, it hides raw measurement data, varies in scaling and thresholds
between vendors, and may not indicate certain failure modes - for
example, cases where autonegotiation would fail even though SQI reports
a good link. In practice, such scenarios can only be investigated in
fixed-link mode; here, MSE can provide an empirically estimated value
indicating conditions under which autonegotiation would not succeed.

Example output with current implementation:
root@DistroKit:~ ethtool lan1
Settings for lan1:
...
        Speed: 1000Mb/s
        Duplex: Full
...
        Link detected: yes
        SQI: 5/7
        MSE: 3/127 (channel: worst)

root@DistroKit:~ ethtool --show-mse lan1
MSE diagnostics for lan1:
MSE Configuration:
        Max Average MSE: 127
        Refresh Rate: 2000000 ps
        Symbols per Sample: 250
        Supported capabilities: average channel-a channel-b channel-c
                                channel-d worst

MSE Snapshot (Channel: a):
        Average MSE: 4

MSE Snapshot (Channel: b):
        Average MSE: 3

MSE Snapshot (Channel: c):
        Average MSE: 2

MSE Snapshot (Channel: d):
        Average MSE: 3

[1] https://opensig.org/wp-content/uploads/2024/01/Advanced_PHY_features_for_automotive_Ethernet_V1.0.pdf
====================

Link: https://patch.msgid.link/20251027122801.982364-1-o.rempel@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

10 files changed, 897 insertions, 1 deletions

diff --git a/Documentation/netlink/specs/ethtool.yaml b/Documentation/netlink/specs/ethtool.yaml
index 6a0fb1974513..05d2b6508b59 100644
--- a/Documentation/netlink/specs/ethtool.yaml
+++ b/Documentation/netlink/specs/ethtool.yaml
@@ -1823,6 +1823,73 @@ attribute-sets:
         type: uint
         enum: pse-event
         doc: List of events reported by the PSE controller
+  -
+    name: mse-capabilities
+    doc: MSE capabilities attribute set
+    attr-cnt-name: --ethtool-a-mse-capabilities-cnt
+    attributes:
+      -
+        name: max-average-mse
+        type: uint
+      -
+        name: max-peak-mse
+        type: uint
+      -
+        name: refresh-rate-ps
+        type: uint
+      -
+        name: num-symbols
+        type: uint
+  -
+    name: mse-snapshot
+    doc: MSE snapshot attribute set
+    attr-cnt-name: --ethtool-a-mse-snapshot-cnt
+    attributes:
+      -
+        name: average-mse
+        type: uint
+      -
+        name: peak-mse
+        type: uint
+      -
+        name: worst-peak-mse
+        type: uint
+  -
+    name: mse
+    attr-cnt-name: --ethtool-a-mse-cnt
+    attributes:
+      -
+        name: header
+        type: nest
+        nested-attributes: header
+      -
+        name: capabilities
+        type: nest
+        nested-attributes: mse-capabilities
+      -
+        name: channel-a
+        type: nest
+        nested-attributes: mse-snapshot
+      -
+        name: channel-b
+        type: nest
+        nested-attributes: mse-snapshot
+      -
+        name: channel-c
+        type: nest
+        nested-attributes: mse-snapshot
+      -
+        name: channel-d
+        type: nest
+        nested-attributes: mse-snapshot
+      -
+        name: worst-channel
+        type: nest
+        nested-attributes: mse-snapshot
+      -
+        name: link
+        type: nest
+        nested-attributes: mse-snapshot
 
 operations:
   enum-model: directional
@@ -2756,6 +2823,25 @@ operations:
         attributes:
           - header
           - context
+    -
+      name: mse-get
+      doc: Get PHY MSE measurement data and capabilities.
+      attribute-set: mse
+      do: &mse-get-op
+        request:
+          attributes:
+            - header
+        reply:
+          attributes:
+            - header
+            - capabilities
+            - channel-a
+            - channel-b
+            - channel-c
+            - channel-d
+            - worst-channel
+            - link
+      dump: *mse-get-op
 
 mcast-groups:
   list:
diff --git a/Documentation/networking/ethtool-netlink.rst b/Documentation/networking/ethtool-netlink.rst
index b270886c5f5d..af56c304cef4 100644
--- a/Documentation/networking/ethtool-netlink.rst
+++ b/Documentation/networking/ethtool-netlink.rst
@@ -242,6 +242,7 @@ Userspace to kernel:
   ``ETHTOOL_MSG_RSS_SET``               set RSS settings
   ``ETHTOOL_MSG_RSS_CREATE_ACT``        create an additional RSS context
   ``ETHTOOL_MSG_RSS_DELETE_ACT``        delete an additional RSS context
+  ``ETHTOOL_MSG_MSE_GET``               get MSE diagnostic data
   ===================================== =================================
 
 Kernel to userspace:
@@ -299,6 +300,7 @@ Kernel to userspace:
   ``ETHTOOL_MSG_RSS_CREATE_ACT_REPLY``     create an additional RSS context
   ``ETHTOOL_MSG_RSS_CREATE_NTF``           additional RSS context created
   ``ETHTOOL_MSG_RSS_DELETE_NTF``           additional RSS context deleted
+  ``ETHTOOL_MSG_MSE_GET_REPLY``            MSE diagnostic data
   ======================================== =================================
 
 ``GET`` requests are sent by userspace applications to retrieve device
@@ -2458,6 +2460,68 @@ Kernel response contents:
 
 For a description of each attribute, see ``TSCONFIG_GET``.
 
+MSE_GET
+=======
+
+Retrieves detailed Mean Square Error (MSE) diagnostic information from the PHY.
+
+Request Contents:
+
+  ====================================  ======  ============================
+  ``ETHTOOL_A_MSE_HEADER``              nested  request header
+  ====================================  ======  ============================
+
+Kernel Response Contents:
+
+  ====================================  ======  ================================
+  ``ETHTOOL_A_MSE_HEADER``              nested  reply header
+  ``ETHTOOL_A_MSE_CAPABILITIES``        nested  capability/scale info for MSE
+                                                measurements
+  ``ETHTOOL_A_MSE_CHANNEL_A``           nested  snapshot for Channel A
+  ``ETHTOOL_A_MSE_CHANNEL_B``           nested  snapshot for Channel B
+  ``ETHTOOL_A_MSE_CHANNEL_C``           nested  snapshot for Channel C
+  ``ETHTOOL_A_MSE_CHANNEL_D``           nested  snapshot for Channel D
+  ``ETHTOOL_A_MSE_WORST_CHANNEL``       nested  snapshot for worst channel
+  ``ETHTOOL_A_MSE_LINK``                nested  snapshot for link-wide aggregate
+  ====================================  ======  ================================
+
+MSE Capabilities
+----------------
+
+This nested attribute reports the capability / scaling properties used to
+interpret snapshot values.
+
+  ============================================== ======  =========================
+  ``ETHTOOL_A_MSE_CAPABILITIES_MAX_AVERAGE_MSE`` uint    max avg_mse scale
+  ``ETHTOOL_A_MSE_CAPABILITIES_MAX_PEAK_MSE``    uint    max peak_mse scale
+  ``ETHTOOL_A_MSE_CAPABILITIES_REFRESH_RATE_PS`` uint    sample rate (picoseconds)
+  ``ETHTOOL_A_MSE_CAPABILITIES_NUM_SYMBOLS``     uint    symbols per HW sample
+  ============================================== ======  =========================
+
+The max-average/peak fields are included only if the corresponding metric
+is supported by the PHY. Their absence indicates that the metric is not
+available.
+
+See ``struct phy_mse_capability`` kernel documentation in
+``include/linux/phy.h``.
+
+MSE Snapshot
+------------
+
+Each per-channel nest contains an atomic snapshot of MSE values for that
+selector (channel A/B/C/D, worst channel, or link).
+
+  ==========================================  ======  ===================
+  ``ETHTOOL_A_MSE_SNAPSHOT_AVERAGE_MSE``      uint    average MSE value
+  ``ETHTOOL_A_MSE_SNAPSHOT_PEAK_MSE``         uint    current peak MSE
+  ``ETHTOOL_A_MSE_SNAPSHOT_WORST_PEAK_MSE``   uint    worst-case peak MSE
+  ==========================================  ======  ===================
+
+Within each channel nest, only the metrics supported by the PHY will be present.
+
+See ``struct phy_mse_snapshot`` kernel documentation in
+``include/linux/phy.h``.
+
 Request translation
 ===================
 
diff --git a/drivers/net/phy/dp83td510.c b/drivers/net/phy/dp83td510.c
index 23af1ac194fa..d75dae6071ad 100644
--- a/drivers/net/phy/dp83td510.c
+++ b/drivers/net/phy/dp83td510.c
@@ -61,6 +61,7 @@
 #define DP83TD510E_MASTER_SLAVE_RESOL_FAIL	BIT(15)
 
 #define DP83TD510E_MSE_DETECT			0xa85
+#define DP83TD510E_MSE_MAX			U16_MAX
 
 #define DP83TD510_SQI_MAX	7
 
@@ -249,6 +250,64 @@ struct dp83td510_priv {
 #define DP83TD510E_ALCD_COMPLETE			BIT(15)
 #define DP83TD510E_ALCD_CABLE_LENGTH			GENMASK(10, 0)
 
+static int dp83td510_get_mse_capability(struct phy_device *phydev,
+					struct phy_mse_capability *cap)
+{
+	/* DP83TD510E documents only a single (average) MSE register
+	 * (used to derive SQI); no peak or worst-peak counters are
+	 * described. Advertise only PHY_MSE_CAP_AVG.
+	 */
+	cap->supported_caps = PHY_MSE_CAP_AVG;
+	/* 10BASE-T1L is a single-pair medium, so there are no B/C/D channels.
+	 * We still advertise PHY_MSE_CAP_CHANNEL_A to indicate that the PHY
+	 * can attribute the measurement to a specific pair (the only one),
+	 * rather than exposing it only as a link-aggregate.
+	 *
+	 * Rationale:
+	 *  - Keeps the ethtool MSE_GET selection logic consistent: per-channel
+	 *    (A/B/C/D) is preferred over WORST/LINK, so userspace receives a
+	 *    CHANNEL_A nest instead of LINK.
+	 *  - Signals to tools that "per-pair" data is available (even if there's
+	 *    just one pair), avoiding the impression that only aggregate values
+	 *    are supported.
+	 *  - Remains compatible with multi-pair PHYs and uniform UI handling.
+	 *
+	 * Note: WORST and other channels are not advertised on 10BASE-T1L.
+	 */
+	cap->supported_caps |= PHY_MSE_CHANNEL_A | PHY_MSE_CAP_LINK;
+	cap->max_average_mse = DP83TD510E_MSE_MAX;
+
+	/* The datasheet does not specify the refresh rate or symbol count,
+	 * but based on similar PHYs and standards, we can assume a common
+	 * value. For 10BASE-T1L, the symbol rate is 7.5 MBd. A common
+	 * diagnostic interval is around 1ms.
+	 * 7.5e6 symbols/sec * 0.001 sec = 7500 symbols.
+	 */
+	cap->refresh_rate_ps = 1000000000; /* 1 ms */
+	cap->num_symbols = 7500;
+
+	return 0;
+}
+
+static int dp83td510_get_mse_snapshot(struct phy_device *phydev,
+				      enum phy_mse_channel channel,
+				      struct phy_mse_snapshot *snapshot)
+{
+	int ret;
+
+	if (channel != PHY_MSE_CHANNEL_LINK &&
+	    channel != PHY_MSE_CHANNEL_A)
+		return -EOPNOTSUPP;
+
+	ret = phy_read_mmd(phydev, MDIO_MMD_VEND2, DP83TD510E_MSE_DETECT);
+	if (ret < 0)
+		return ret;
+
+	snapshot->average_mse = ret;
+
+	return 0;
+}
+
 static int dp83td510_led_brightness_set(struct phy_device *phydev, u8 index,
 					enum led_brightness brightness)
 {
@@ -893,6 +952,9 @@ static struct phy_driver dp83td510_driver[] = {
 	.get_phy_stats	= dp83td510_get_phy_stats,
 	.update_stats	= dp83td510_update_stats,
 
+	.get_mse_capability = dp83td510_get_mse_capability,
+	.get_mse_snapshot = dp83td510_get_mse_snapshot,
+
 	.led_brightness_set = dp83td510_led_brightness_set,
 	.led_hw_is_supported = dp83td510_led_hw_is_supported,
 	.led_hw_control_set = dp83td510_led_hw_control_set,
diff --git a/drivers/net/phy/micrel.c b/drivers/net/phy/micrel.c
index edca0024b7c7..06080b1c753d 100644
--- a/drivers/net/phy/micrel.c
+++ b/drivers/net/phy/micrel.c
@@ -2325,6 +2325,106 @@ static int kszphy_get_sqi_max(struct phy_device *phydev)
 	return KSZ9477_SQI_MAX;
 }
 
+static int kszphy_get_mse_capability(struct phy_device *phydev,
+				     struct phy_mse_capability *cap)
+{
+	/* Capabilities depend on link mode:
+	 * - 1000BASE-T: per-pair SQI registers exist => expose A..D
+	 *   and a WORST selector.
+	 * - 100BASE-TX: HW provides a single MSE/SQI reading in the "channel A"
+	 *   register, but with auto MDI-X there is no MDI-X resolution bit,
+	 *   so we cannot map that register to a specific wire pair reliably.
+	 *   To avoid misleading per-channel data, advertise only LINK.
+	 * Other speeds: no MSE exposure via this driver.
+	 *
+	 * Note: WORST is *not* a hardware selector on this family.
+	 * We expose it because the driver computes it in software
+	 * by scanning per-channel readouts (A..D) and picking the
+	 * maximum average MSE.
+	 */
+	if (phydev->speed == SPEED_1000)
+		cap->supported_caps = PHY_MSE_CAP_CHANNEL_A |
+				      PHY_MSE_CAP_CHANNEL_B |
+				      PHY_MSE_CAP_CHANNEL_C |
+				      PHY_MSE_CAP_CHANNEL_D |
+				      PHY_MSE_CAP_WORST_CHANNEL;
+	else if (phydev->speed == SPEED_100)
+		cap->supported_caps = PHY_MSE_CAP_LINK;
+	else
+		return -EOPNOTSUPP;
+
+	cap->max_average_mse = FIELD_MAX(KSZ9477_MMD_SQI_MASK);
+	cap->refresh_rate_ps = 2000000; /* 2 us */
+	/* Estimated from link modulation (125 MBd per channel) and documented
+	 * refresh rate of 2 us
+	 */
+	cap->num_symbols = 250;
+
+	cap->supported_caps |= PHY_MSE_CAP_AVG;
+
+	return 0;
+}
+
+static int kszphy_get_mse_snapshot(struct phy_device *phydev,
+				   enum phy_mse_channel channel,
+				   struct phy_mse_snapshot *snapshot)
+{
+	u8 num_channels;
+	int ret;
+
+	if (phydev->speed == SPEED_1000)
+		num_channels = 4;
+	else if (phydev->speed == SPEED_100)
+		num_channels = 1;
+	else
+		return -EOPNOTSUPP;
+
+	if (channel == PHY_MSE_CHANNEL_WORST) {
+		u32 worst_val = 0;
+		int i;
+
+		/* WORST is implemented in software: select the maximum
+		 * average MSE across the available per-channel registers.
+		 * Only defined when multiple channels exist (1000BASE-T).
+		 */
+		if (num_channels < 2)
+			return -EOPNOTSUPP;
+
+		for (i = 0; i < num_channels; i++) {
+			ret = phy_read_mmd(phydev, MDIO_MMD_PMAPMD,
+					KSZ9477_MMD_SIGNAL_QUALITY_CHAN_A + i);
+			if (ret < 0)
+				return ret;
+
+			ret = FIELD_GET(KSZ9477_MMD_SQI_MASK, ret);
+			if (ret > worst_val)
+				worst_val = ret;
+		}
+		snapshot->average_mse = worst_val;
+	} else if (channel == PHY_MSE_CHANNEL_LINK && num_channels == 1) {
+		ret = phy_read_mmd(phydev, MDIO_MMD_PMAPMD,
+				   KSZ9477_MMD_SIGNAL_QUALITY_CHAN_A);
+		if (ret < 0)
+			return ret;
+		snapshot->average_mse = FIELD_GET(KSZ9477_MMD_SQI_MASK, ret);
+	} else if (channel >= PHY_MSE_CHANNEL_A &&
+		   channel <= PHY_MSE_CHANNEL_D) {
+		/* Per-channel readouts are valid only for 1000BASE-T. */
+		if (phydev->speed != SPEED_1000)
+			return -EOPNOTSUPP;
+
+		ret = phy_read_mmd(phydev, MDIO_MMD_PMAPMD,
+				   KSZ9477_MMD_SIGNAL_QUALITY_CHAN_A + channel);
+		if (ret < 0)
+			return ret;
+		snapshot->average_mse = FIELD_GET(KSZ9477_MMD_SQI_MASK, ret);
+	} else {
+		return -EOPNOTSUPP;
+	}
+
+	return 0;
+}
+
 static void kszphy_enable_clk(struct phy_device *phydev)
 {
 	struct kszphy_priv *priv = phydev->priv;
@@ -6497,6 +6597,8 @@ static struct phy_driver ksphy_driver[] = {
 	.cable_test_get_status	= ksz9x31_cable_test_get_status,
 	.get_sqi	= kszphy_get_sqi,
 	.get_sqi_max	= kszphy_get_sqi_max,
+	.get_mse_capability = kszphy_get_mse_capability,
+	.get_mse_snapshot = kszphy_get_mse_snapshot,
 } };
 
 module_phy_driver(ksphy_driver);
diff --git a/include/linux/phy.h b/include/linux/phy.h
index 358dd6f0ff96..e3474f03cbc1 100644
--- a/include/linux/phy.h
+++ b/include/linux/phy.h
@@ -903,6 +903,165 @@ struct phy_led {
 
 #define to_phy_led(d) container_of(d, struct phy_led, led_cdev)
 
+/*
+ * PHY_MSE_CAP_* - Bitmask flags for Mean Square Error (MSE) capabilities
+ *
+ * These flags describe which MSE metrics and selectors are implemented
+ * by the PHY for the current link mode. They are used in
+ * struct phy_mse_capability.supported_caps.
+ *
+ * Standardization:
+ *   The OPEN Alliance (OA) defines the presence of MSE/SQI/pMSE but not their
+ *   numeric scaling, update intervals, or aggregation windows.  See:
+ *     OA 100BASE-T1 TC1 v1.0, sections 6.1.1-6.1.3
+ *     OA 1000BASE-T1 TC12 v2.2, sections 6.1.1-6.1.2
+ *
+ * Description of flags:
+ *
+ *   PHY_MSE_CAP_CHANNEL_A
+ *     Per-pair diagnostics for Channel A are supported.  Mapping to the
+ *     physical wire pair may depend on MDI/MDI-X polarity.
+ *
+ *   PHY_MSE_CAP_CHANNEL_B, _C, _D
+ *     Same as above for channels B-D.
+ *
+ *   PHY_MSE_CAP_WORST_CHANNEL
+ *     The PHY or driver can identify and report the single worst-performing
+ *     channel without querying each one individually.
+ *
+ *   PHY_MSE_CAP_LINK
+ *     The PHY provides only a link-wide aggregate measurement or cannot map
+ *     results to a specific pair (for example 100BASE-TX with unknown
+ *     MDI/MDI-X).
+ *
+ *   PHY_MSE_CAP_AVG
+ *     Average MSE (mean DCQ metric) is supported.  For 100/1000BASE-T1 the OA
+ *     recommends 2^16 symbols, scaled 0..511, but the exact scaling is
+ *     vendor-specific.
+ *
+ *   PHY_MSE_CAP_PEAK
+ *     Peak MSE (current peak within the measurement window) is supported.
+ *     Defined as pMSE for 100BASE-T1; vendor-specific for others.
+ *
+ *   PHY_MSE_CAP_WORST_PEAK
+ *     Latched worst-case peak MSE since the last read (read-to-clear if
+ *     implemented).  Optional in OA 100BASE-T1 TC1 6.1.3.
+ */
+#define PHY_MSE_CAP_CHANNEL_A BIT(0)
+#define PHY_MSE_CAP_CHANNEL_B BIT(1)
+#define PHY_MSE_CAP_CHANNEL_C BIT(2)
+#define PHY_MSE_CAP_CHANNEL_D BIT(3)
+#define PHY_MSE_CAP_WORST_CHANNEL BIT(4)
+#define PHY_MSE_CAP_LINK BIT(5)
+#define PHY_MSE_CAP_AVG BIT(6)
+#define PHY_MSE_CAP_PEAK BIT(7)
+#define PHY_MSE_CAP_WORST_PEAK BIT(8)
+
+/*
+ * enum phy_mse_channel - Identifiers for selecting MSE measurement channels
+ *
+ * PHY_MSE_CHANNEL_A - PHY_MSE_CHANNEL_D
+ *   Select per-pair measurement for the corresponding channel.
+ *
+ * PHY_MSE_CHANNEL_WORST
+ *   Select the single worst-performing channel reported by hardware.
+ *
+ * PHY_MSE_CHANNEL_LINK
+ *   Select link-wide aggregate data (used when per-pair results are
+ *   unavailable).
+ */
+enum phy_mse_channel {
+	PHY_MSE_CHANNEL_A,
+	PHY_MSE_CHANNEL_B,
+	PHY_MSE_CHANNEL_C,
+	PHY_MSE_CHANNEL_D,
+	PHY_MSE_CHANNEL_WORST,
+	PHY_MSE_CHANNEL_LINK,
+};
+
+/**
+ * struct phy_mse_capability - Capabilities of Mean Square Error (MSE)
+ *                             measurement interface
+ *
+ * Standardization notes:
+ *
+ * - Presence of MSE/SQI/pMSE is defined by OPEN Alliance specs, but numeric
+ *   scaling, refresh/update rate and aggregation windows are not fixed and
+ *   are vendor-/product-specific. (OA 100BASE-T1 TC1 v1.0 6.1.*;
+ *   OA 1000BASE-T1 TC12 v2.2 6.1.*)
+ *
+ * - Typical recommendations: 2^16 symbols and 0..511 scaling for MSE; pMSE only
+ *   defined for 100BASE-T1 (sliding window example), others are vendor
+ *   extensions. Drivers must report actual scale/limits here.
+ *
+ * Describes the MSE measurement capabilities for the current link mode. These
+ * properties are dynamic and may change when link settings are modified.
+ * Callers should re-query this capability after any link state change to
+ * ensure they have the most up-to-date information.
+ *
+ * Callers should only request measurements for channels and types that are
+ * indicated as supported by the @supported_caps bitmask. If @supported_caps
+ * is 0, the device provides no MSE diagnostics, and driver operations should
+ * typically return -EOPNOTSUPP.
+ *
+ * Snapshot values for average and peak MSE can be normalized to a 0..1 ratio
+ * by dividing the raw snapshot by the corresponding @max_average_mse or
+ * @max_peak_mse value.
+ *
+ * @max_average_mse: The maximum value for an average MSE snapshot. This
+ *   defines the scale for the measurement. If the PHY_MSE_CAP_AVG capability is
+ *   supported, this value MUST be greater than 0. (vendor-specific units).
+ * @max_peak_mse: The maximum value for a peak MSE snapshot. If either
+ *   PHY_MSE_CAP_PEAK or PHY_MSE_CAP_WORST_PEAK is supported, this value MUST
+ *   be greater than 0. (vendor-specific units).
+ * @refresh_rate_ps: The typical interval, in picoseconds, between hardware
+ *   updates of the MSE values. This is an estimate, and callers should not
+ *   assume synchronous sampling. (vendor-specific units).
+ * @num_symbols: The number of symbols aggregated per hardware sample to
+ *   calculate the MSE. (vendor-specific units).
+ * @supported_caps: A bitmask of PHY_MSE_CAP_* values indicating which
+ *   measurement types (e.g., average, peak) and channels
+ *   (e.g., per-pair or link-wide) are supported.
+ */
+struct phy_mse_capability {
+	u64 max_average_mse;
+	u64 max_peak_mse;
+	u64 refresh_rate_ps;
+	u64 num_symbols;
+	u32 supported_caps;
+};
+
+/**
+ * struct phy_mse_snapshot - A snapshot of Mean Square Error (MSE) diagnostics
+ *
+ * Holds a set of MSE diagnostic values that were all captured from a single
+ * measurement window.
+ *
+ * Values are raw, device-scaled and not normalized. Use struct
+ * phy_mse_capability to interpret the scale and sampling window.
+ *
+ * @average_mse: The average MSE value over the measurement window.
+ *   OPEN Alliance references MSE as a DCQ metric; recommends 2^16 symbols and
+ *   0..511 scaling. Exact scale and refresh are vendor-specific.
+ *   (100BASE-T1 TC1 v1.0 6.1.1; 1000BASE-T1 TC12 v2.2 6.1.1).
+ *
+ * @peak_mse: The peak MSE value observed within the measurement window.
+ *   For 100BASE-T1, "pMSE" is optional and may be implemented via a sliding
+ *   128-symbol window with periodic capture; not standardized for 1000BASE-T1.
+ *   (100BASE-T1 TC1 v1.0 6.1.3, Table "DCQ.peakMSE").
+ *
+ * @worst_peak_mse: A latched high-water mark of the peak MSE since last read
+ *   (read-to-clear if implemented). OPEN Alliance shows a latched "worst case
+ *   peak MSE" for 100BASE-T1 pMSE; availability/semantics outside that are
+ *   vendor-specific. (100BASE-T1 TC1 v1.0 6.1.3, DCQ.peakMSE high byte;
+ *   1000BASE-T1 TC12 v2.2 treats DCQ details as vendor-specific.)
+ */
+struct phy_mse_snapshot {
+	u64 average_mse;
+	u64 peak_mse;
+	u64 worst_peak_mse;
+};
+
 /**
  * struct phy_driver - Driver structure for a particular PHY type
  *
@@ -1184,6 +1343,53 @@ struct phy_driver {
 	/** @get_sqi_max: Get the maximum signal quality indication */
 	int (*get_sqi_max)(struct phy_device *dev);
 
+	/**
+	 * @get_mse_capability: Get capabilities and scale of MSE measurement
+	 * @dev:    PHY device
+	 * @cap: Output (filled on success)
+	 *
+	 * Fill @cap with the PHY's MSE capability for the current
+	 * link mode: scale limits (max_average_mse, max_peak_mse), update
+	 * interval (refresh_rate_ps), sample length (num_symbols) and the
+	 * capability bitmask (supported_caps).
+	 *
+	 * Implementations may defer capability report until hardware has
+	 * converged; in that case they should return -EAGAIN and allow the
+	 * caller to retry later.
+	 *
+	 * Return: 0 on success. On failure, returns a negative errno code, such
+	 * as -EOPNOTSUPP if MSE measurement is not supported by the PHY or in
+	 * the current link mode, or -EAGAIN if the capability information is
+	 * not yet available.
+	 */
+	int (*get_mse_capability)(struct phy_device *dev,
+				  struct phy_mse_capability *cap);
+
+	/**
+	 * @get_mse_snapshot: Retrieve a snapshot of MSE diagnostic values
+	 * @dev:      PHY device
+	 * @channel:  Channel identifier (PHY_MSE_CHANNEL_*)
+	 * @snapshot: Output (filled on success)
+	 *
+	 * Fill @snapshot with a correlated set of MSE values from the most
+	 * recent measurement window.
+	 *
+	 * Callers must validate @channel against supported_caps returned by
+	 * get_mse_capability(). Drivers must not coerce @channel; if the
+	 * requested selector is not implemented by the device or current link
+	 * mode, the operation must fail.
+	 *
+	 * worst_peak_mse is latched and must be treated as read-to-clear.
+	 *
+	 * Return: 0 on success. On failure, returns a negative errno code, such
+	 * as -EOPNOTSUPP if MSE measurement is not supported by the PHY or in
+	 * the current link mode, or -EAGAIN if measurements are not yet
+	 * available.
+	 */
+	int (*get_mse_snapshot)(struct phy_device *dev,
+				enum phy_mse_channel channel,
+				struct phy_mse_snapshot *snapshot);
+
 	/* PLCA RS interface */
 	/** @get_plca_cfg: Return the current PLCA configuration */
 	int (*get_plca_cfg)(struct phy_device *dev,
diff --git a/include/uapi/linux/ethtool_netlink_generated.h b/include/uapi/linux/ethtool_netlink_generated.h
index 0e8ac0d974e2..b71b175df46d 100644
--- a/include/uapi/linux/ethtool_netlink_generated.h
+++ b/include/uapi/linux/ethtool_netlink_generated.h
@@ -804,6 +804,39 @@ enum {
 };
 
 enum {
+	ETHTOOL_A_MSE_CAPABILITIES_MAX_AVERAGE_MSE = 1,
+	ETHTOOL_A_MSE_CAPABILITIES_MAX_PEAK_MSE,
+	ETHTOOL_A_MSE_CAPABILITIES_REFRESH_RATE_PS,
+	ETHTOOL_A_MSE_CAPABILITIES_NUM_SYMBOLS,
+
+	__ETHTOOL_A_MSE_CAPABILITIES_CNT,
+	ETHTOOL_A_MSE_CAPABILITIES_MAX = (__ETHTOOL_A_MSE_CAPABILITIES_CNT - 1)
+};
+
+enum {
+	ETHTOOL_A_MSE_SNAPSHOT_AVERAGE_MSE = 1,
+	ETHTOOL_A_MSE_SNAPSHOT_PEAK_MSE,
+	ETHTOOL_A_MSE_SNAPSHOT_WORST_PEAK_MSE,
+
+	__ETHTOOL_A_MSE_SNAPSHOT_CNT,
+	ETHTOOL_A_MSE_SNAPSHOT_MAX = (__ETHTOOL_A_MSE_SNAPSHOT_CNT - 1)
+};
+
+enum {
+	ETHTOOL_A_MSE_HEADER = 1,
+	ETHTOOL_A_MSE_CAPABILITIES,
+	ETHTOOL_A_MSE_CHANNEL_A,
+	ETHTOOL_A_MSE_CHANNEL_B,
+	ETHTOOL_A_MSE_CHANNEL_C,
+	ETHTOOL_A_MSE_CHANNEL_D,
+	ETHTOOL_A_MSE_WORST_CHANNEL,
+	ETHTOOL_A_MSE_LINK,
+
+	__ETHTOOL_A_MSE_CNT,
+	ETHTOOL_A_MSE_MAX = (__ETHTOOL_A_MSE_CNT - 1)
+};
+
+enum {
 	ETHTOOL_MSG_USER_NONE = 0,
 	ETHTOOL_MSG_STRSET_GET = 1,
 	ETHTOOL_MSG_LINKINFO_GET,
@@ -855,6 +888,7 @@ enum {
 	ETHTOOL_MSG_RSS_SET,
 	ETHTOOL_MSG_RSS_CREATE_ACT,
 	ETHTOOL_MSG_RSS_DELETE_ACT,
+	ETHTOOL_MSG_MSE_GET,
 
 	__ETHTOOL_MSG_USER_CNT,
 	ETHTOOL_MSG_USER_MAX = (__ETHTOOL_MSG_USER_CNT - 1)
@@ -915,6 +949,7 @@ enum {
 	ETHTOOL_MSG_RSS_CREATE_ACT_REPLY,
 	ETHTOOL_MSG_RSS_CREATE_NTF,
 	ETHTOOL_MSG_RSS_DELETE_NTF,
+	ETHTOOL_MSG_MSE_GET_REPLY,
 
 	__ETHTOOL_MSG_KERNEL_CNT,
 	ETHTOOL_MSG_KERNEL_MAX = (__ETHTOOL_MSG_KERNEL_CNT - 1)
diff --git a/net/ethtool/Makefile b/net/ethtool/Makefile
index 1e493553b977..629c10916670 100644
--- a/net/ethtool/Makefile
+++ b/net/ethtool/Makefile
@@ -9,4 +9,4 @@ ethtool_nl-y	:= netlink.o bitset.o strset.o linkinfo.o linkmodes.o rss.o \
 		   channels.o coalesce.o pause.o eee.o tsinfo.o cabletest.o \
 		   tunnels.o fec.o eeprom.o stats.o phc_vclocks.o mm.o \
 		   module.o cmis_fw_update.o cmis_cdb.o pse-pd.o plca.o \
-		   phy.o tsconfig.o
+		   phy.o tsconfig.o mse.o
diff --git a/net/ethtool/mse.c b/net/ethtool/mse.c
new file mode 100644
index 000000000000..6aac004c3ffc
--- /dev/null
+++ b/net/ethtool/mse.c
@@ -0,0 +1,329 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+#include <linux/ethtool.h>
+#include <linux/phy.h>
+#include <linux/slab.h>
+
+#include "netlink.h"
+#include "common.h"
+
+/* Channels A-D only; WORST and LINK are exclusive alternatives */
+#define PHY_MSE_CHANNEL_COUNT 4
+
+struct mse_req_info {
+	struct ethnl_req_info base;
+};
+
+struct mse_snapshot_entry {
+	struct phy_mse_snapshot snapshot;
+	int channel;
+};
+
+struct mse_reply_data {
+	struct ethnl_reply_data base;
+	struct phy_mse_capability capability;
+	struct mse_snapshot_entry *snapshots;
+	unsigned int num_snapshots;
+};
+
+static struct mse_reply_data *
+mse_repdata(const struct ethnl_reply_data *reply_base)
+{
+	return container_of(reply_base, struct mse_reply_data, base);
+}
+
+const struct nla_policy ethnl_mse_get_policy[] = {
+	[ETHTOOL_A_MSE_HEADER] = NLA_POLICY_NESTED(ethnl_header_policy_phy),
+};
+
+static int get_snapshot_if_supported(struct phy_device *phydev,
+				     struct mse_reply_data *data,
+				     unsigned int *idx, u32 cap_bit,
+				     enum phy_mse_channel channel)
+{
+	int ret;
+
+	if (data->capability.supported_caps & cap_bit) {
+		ret = phydev->drv->get_mse_snapshot(phydev, channel,
+					&data->snapshots[*idx].snapshot);
+		if (ret)
+			return ret;
+		data->snapshots[*idx].channel = channel;
+		(*idx)++;
+	}
+
+	return 0;
+}
+
+static int mse_get_channels(struct phy_device *phydev,
+			    struct mse_reply_data *data)
+{
+	unsigned int i = 0;
+	int ret;
+
+	if (!data->capability.supported_caps)
+		return 0;
+
+	data->snapshots = kcalloc(PHY_MSE_CHANNEL_COUNT,
+				  sizeof(*data->snapshots), GFP_KERNEL);
+	if (!data->snapshots)
+		return -ENOMEM;
+
+	/* Priority 1: Individual channels */
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_A,
+					PHY_MSE_CHANNEL_A);
+	if (ret)
+		return ret;
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_B,
+					PHY_MSE_CHANNEL_B);
+	if (ret)
+		return ret;
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_C,
+					PHY_MSE_CHANNEL_C);
+	if (ret)
+		return ret;
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_CHANNEL_D,
+					PHY_MSE_CHANNEL_D);
+	if (ret)
+		return ret;
+
+	/* If any individual channels were found, we are done. */
+	if (i > 0) {
+		data->num_snapshots = i;
+		return 0;
+	}
+
+	/* Priority 2: Worst channel, if no individual channels supported. */
+	ret = get_snapshot_if_supported(phydev, data, &i,
+					PHY_MSE_CAP_WORST_CHANNEL,
+					PHY_MSE_CHANNEL_WORST);
+	if (ret)
+		return ret;
+
+	/* If worst channel was found, we are done. */
+	if (i > 0) {
+		data->num_snapshots = i;
+		return 0;
+	}
+
+	/* Priority 3: Link-wide, if nothing else is supported. */
+	ret = get_snapshot_if_supported(phydev, data, &i, PHY_MSE_CAP_LINK,
+					PHY_MSE_CHANNEL_LINK);
+	if (ret)
+		return ret;
+
+	data->num_snapshots = i;
+	return 0;
+}
+
+static int mse_prepare_data(const struct ethnl_req_info *req_base,
+			    struct ethnl_reply_data *reply_base,
+			    const struct genl_info *info)
+{
+	struct mse_reply_data *data = mse_repdata(reply_base);
+	struct net_device *dev = reply_base->dev;
+	struct phy_device *phydev;
+	int ret;
+
+	phydev = ethnl_req_get_phydev(req_base, info->attrs,
+				      ETHTOOL_A_MSE_HEADER, info->extack);
+	if (IS_ERR(phydev))
+		return PTR_ERR(phydev);
+	if (!phydev)
+		return -EOPNOTSUPP;
+
+	ret = ethnl_ops_begin(dev);
+	if (ret)
+		return ret;
+
+	mutex_lock(&phydev->lock);
+
+	if (!phydev->drv || !phydev->drv->get_mse_capability ||
+	    !phydev->drv->get_mse_snapshot) {
+		ret = -EOPNOTSUPP;
+		goto out_unlock;
+	}
+	if (!phydev->link) {
+		ret = -ENETDOWN;
+		goto out_unlock;
+	}
+
+	ret = phydev->drv->get_mse_capability(phydev, &data->capability);
+	if (ret)
+		goto out_unlock;
+
+	ret = mse_get_channels(phydev, data);
+
+out_unlock:
+	mutex_unlock(&phydev->lock);
+	ethnl_ops_complete(dev);
+	if (ret)
+		kfree(data->snapshots);
+	return ret;
+}
+
+static void mse_cleanup_data(struct ethnl_reply_data *reply_base)
+{
+	struct mse_reply_data *data = mse_repdata(reply_base);
+
+	kfree(data->snapshots);
+}
+
+static int mse_reply_size(const struct ethnl_req_info *req_base,
+			  const struct ethnl_reply_data *reply_base)
+{
+	const struct mse_reply_data *data = mse_repdata(reply_base);
+	size_t len = 0;
+	unsigned int i;
+
+	/* ETHTOOL_A_MSE_CAPABILITIES */
+	len += nla_total_size(0);
+	if (data->capability.supported_caps & PHY_MSE_CAP_AVG)
+		/* ETHTOOL_A_MSE_CAPABILITIES_MAX_AVERAGE_MSE */
+		len += nla_total_size(sizeof(u64));
+	if (data->capability.supported_caps & (PHY_MSE_CAP_PEAK |
+					       PHY_MSE_CAP_WORST_PEAK))
+		/* ETHTOOL_A_MSE_CAPABILITIES_MAX_PEAK_MSE */
+		len += nla_total_size(sizeof(u64));
+	/* ETHTOOL_A_MSE_CAPABILITIES_REFRESH_RATE_PS */
+	len += nla_total_size(sizeof(u64));
+	/* ETHTOOL_A_MSE_CAPABILITIES_NUM_SYMBOLS */
+	len += nla_total_size(sizeof(u64));
+
+	for (i = 0; i < data->num_snapshots; i++) {
+		size_t snapshot_len = 0;
+
+		/* Per-channel nest (e.g., ETHTOOL_A_MSE_CHANNEL_A / _B / _C /
+		 * _D / _WORST_CHANNEL / _LINK)
+		 */
+		snapshot_len += nla_total_size(0);
+
+		if (data->capability.supported_caps & PHY_MSE_CAP_AVG)
+			snapshot_len += nla_total_size(sizeof(u64));
+		if (data->capability.supported_caps & PHY_MSE_CAP_PEAK)
+			snapshot_len += nla_total_size(sizeof(u64));
+		if (data->capability.supported_caps & PHY_MSE_CAP_WORST_PEAK)
+			snapshot_len += nla_total_size(sizeof(u64));
+
+		len += snapshot_len;
+	}
+
+	return len;
+}
+
+static int mse_channel_to_attr(int ch)
+{
+	switch (ch) {
+	case PHY_MSE_CHANNEL_A:
+		return ETHTOOL_A_MSE_CHANNEL_A;
+	case PHY_MSE_CHANNEL_B:
+		return ETHTOOL_A_MSE_CHANNEL_B;
+	case PHY_MSE_CHANNEL_C:
+		return ETHTOOL_A_MSE_CHANNEL_C;
+	case PHY_MSE_CHANNEL_D:
+		return ETHTOOL_A_MSE_CHANNEL_D;
+	case PHY_MSE_CHANNEL_WORST:
+		return ETHTOOL_A_MSE_WORST_CHANNEL;
+	case PHY_MSE_CHANNEL_LINK:
+		return ETHTOOL_A_MSE_LINK;
+	default:
+		return -EINVAL;
+	}
+}
+
+static int mse_fill_reply(struct sk_buff *skb,
+			  const struct ethnl_req_info *req_base,
+			  const struct ethnl_reply_data *reply_base)
+{
+	const struct mse_reply_data *data = mse_repdata(reply_base);
+	struct nlattr *nest;
+	unsigned int i;
+	int ret;
+
+	nest = nla_nest_start(skb, ETHTOOL_A_MSE_CAPABILITIES);
+	if (!nest)
+		return -EMSGSIZE;
+
+	if (data->capability.supported_caps & PHY_MSE_CAP_AVG) {
+		ret = nla_put_uint(skb,
+				   ETHTOOL_A_MSE_CAPABILITIES_MAX_AVERAGE_MSE,
+				   data->capability.max_average_mse);
+		if (ret < 0)
+			goto nla_put_nest_failure;
+	}
+
+	if (data->capability.supported_caps & (PHY_MSE_CAP_PEAK |
+					       PHY_MSE_CAP_WORST_PEAK)) {
+		ret = nla_put_uint(skb, ETHTOOL_A_MSE_CAPABILITIES_MAX_PEAK_MSE,
+				   data->capability.max_peak_mse);
+		if (ret < 0)
+			goto nla_put_nest_failure;
+	}
+
+	ret = nla_put_uint(skb, ETHTOOL_A_MSE_CAPABILITIES_REFRESH_RATE_PS,
+			   data->capability.refresh_rate_ps);
+	if (ret < 0)
+		goto nla_put_nest_failure;
+
+	ret = nla_put_uint(skb, ETHTOOL_A_MSE_CAPABILITIES_NUM_SYMBOLS,
+			   data->capability.num_symbols);
+	if (ret < 0)
+		goto nla_put_nest_failure;
+
+	nla_nest_end(skb, nest);
+
+	for (i = 0; i < data->num_snapshots; i++) {
+		const struct mse_snapshot_entry *s = &data->snapshots[i];
+		int chan_attr;
+
+		chan_attr = mse_channel_to_attr(s->channel);
+		if (chan_attr < 0)
+			return chan_attr;
+
+		nest = nla_nest_start(skb, chan_attr);
+		if (!nest)
+			return -EMSGSIZE;
+
+		if (data->capability.supported_caps & PHY_MSE_CAP_AVG) {
+			ret = nla_put_uint(skb,
+					   ETHTOOL_A_MSE_SNAPSHOT_AVERAGE_MSE,
+					   s->snapshot.average_mse);
+			if (ret)
+				goto nla_put_nest_failure;
+		}
+		if (data->capability.supported_caps & PHY_MSE_CAP_PEAK) {
+			ret = nla_put_uint(skb, ETHTOOL_A_MSE_SNAPSHOT_PEAK_MSE,
+					   s->snapshot.peak_mse);
+			if (ret)
+				goto nla_put_nest_failure;
+		}
+		if (data->capability.supported_caps & PHY_MSE_CAP_WORST_PEAK) {
+			ret = nla_put_uint(skb,
+					   ETHTOOL_A_MSE_SNAPSHOT_WORST_PEAK_MSE,
+					   s->snapshot.worst_peak_mse);
+			if (ret)
+				goto nla_put_nest_failure;
+		}
+
+		nla_nest_end(skb, nest);
+	}
+
+	return 0;
+
+nla_put_nest_failure:
+	nla_nest_cancel(skb, nest);
+	return ret;
+}
+
+const struct ethnl_request_ops ethnl_mse_request_ops = {
+	.request_cmd = ETHTOOL_MSG_MSE_GET,
+	.reply_cmd = ETHTOOL_MSG_MSE_GET_REPLY,
+	.hdr_attr = ETHTOOL_A_MSE_HEADER,
+	.req_info_size = sizeof(struct mse_req_info),
+	.reply_data_size = sizeof(struct mse_reply_data),
+
+	.prepare_data = mse_prepare_data,
+	.cleanup_data = mse_cleanup_data,
+	.reply_size = mse_reply_size,
+	.fill_reply = mse_fill_reply,
+};
diff --git a/net/ethtool/netlink.c b/net/ethtool/netlink.c
index 2f813f25f07e..6e5f0f4f815a 100644
--- a/net/ethtool/netlink.c
+++ b/net/ethtool/netlink.c
@@ -420,6 +420,7 @@ ethnl_default_requests[__ETHTOOL_MSG_USER_CNT] = {
 	[ETHTOOL_MSG_TSCONFIG_GET]	= &ethnl_tsconfig_request_ops,
 	[ETHTOOL_MSG_TSCONFIG_SET]	= &ethnl_tsconfig_request_ops,
 	[ETHTOOL_MSG_PHY_GET]		= &ethnl_phy_request_ops,
+	[ETHTOOL_MSG_MSE_GET]		= &ethnl_mse_request_ops,
 };
 
 static struct ethnl_dump_ctx *ethnl_dump_context(struct netlink_callback *cb)
@@ -1534,6 +1535,15 @@ static const struct genl_ops ethtool_genl_ops[] = {
 		.policy	= ethnl_rss_delete_policy,
 		.maxattr = ARRAY_SIZE(ethnl_rss_delete_policy) - 1,
 	},
+	{
+		.cmd	= ETHTOOL_MSG_MSE_GET,
+		.doit	= ethnl_default_doit,
+		.start	= ethnl_perphy_start,
+		.dumpit	= ethnl_perphy_dumpit,
+		.done	= ethnl_perphy_done,
+		.policy = ethnl_mse_get_policy,
+		.maxattr = ARRAY_SIZE(ethnl_mse_get_policy) - 1,
+	},
 };
 
 static const struct genl_multicast_group ethtool_nl_mcgrps[] = {
diff --git a/net/ethtool/netlink.h b/net/ethtool/netlink.h
index 1d4f9ecb3d26..89010eaa67df 100644
--- a/net/ethtool/netlink.h
+++ b/net/ethtool/netlink.h
@@ -442,6 +442,7 @@ extern const struct ethnl_request_ops ethnl_plca_status_request_ops;
 extern const struct ethnl_request_ops ethnl_mm_request_ops;
 extern const struct ethnl_request_ops ethnl_phy_request_ops;
 extern const struct ethnl_request_ops ethnl_tsconfig_request_ops;
+extern const struct ethnl_request_ops ethnl_mse_request_ops;
 
 extern const struct nla_policy ethnl_header_policy[ETHTOOL_A_HEADER_FLAGS + 1];
 extern const struct nla_policy ethnl_header_policy_stats[ETHTOOL_A_HEADER_FLAGS + 1];
@@ -497,6 +498,7 @@ extern const struct nla_policy ethnl_module_fw_flash_act_policy[ETHTOOL_A_MODULE
 extern const struct nla_policy ethnl_phy_get_policy[ETHTOOL_A_PHY_HEADER + 1];
 extern const struct nla_policy ethnl_tsconfig_get_policy[ETHTOOL_A_TSCONFIG_HEADER + 1];
 extern const struct nla_policy ethnl_tsconfig_set_policy[ETHTOOL_A_TSCONFIG_MAX + 1];
+extern const struct nla_policy ethnl_mse_get_policy[ETHTOOL_A_MSE_HEADER + 1];
 
 int ethnl_set_features(struct sk_buff *skb, struct genl_info *info);
 int ethnl_act_cable_test(struct sk_buff *skb, struct genl_info *info);