Merge branch 'eth-fbnic-add-timestamping-support'

Vadim Fedorenko says: ==================== eth: fbnic: add timestamping support The series is to add timestamping support for Meta's NIC driver. Changelog: v3 -> v4: - use adjust_by_scaled_ppm() instead of open coding it - adjust cached value of high bits of timestamp to be sure it is older then incoming timestamps v2 -> v3: - rebase on top of net-next - add doc to describe retur value of fbnic_ts40_to_ns() v1 -> v2: - adjust comment about using u64 stats locking primitive - fix typo in the first patch - Cc Richard ==================== Link: https://patch.msgid.link/20241008181436.4120604-1-vadfed@meta.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-05-06 11:26:22 -04:00 · 2024-10-10 12:52:29 +02:00
parent 9e542ff8b7 96f358f75d
commit d9d28b6f6a
11 changed files with 722 additions and 8 deletions
--- a/drivers/net/ethernet/meta/fbnic/Makefile
+++ b/drivers/net/ethernet/meta/fbnic/Makefile
@@ -18,4 +18,5 @@ fbnic-y := fbnic_devlink.o \
 	   fbnic_phylink.o \
 	   fbnic_rpc.o \
 	   fbnic_tlv.o \
-	   fbnic_txrx.o
+	   fbnic_txrx.o \
+	   fbnic_time.o
--- a/drivers/net/ethernet/meta/fbnic/fbnic.h
+++ b/drivers/net/ethernet/meta/fbnic/fbnic.h
@@ -6,6 +6,7 @@

 #include <linux/interrupt.h>
 #include <linux/io.h>
+#include <linux/ptp_clock_kernel.h>
 #include <linux/types.h>
 #include <linux/workqueue.h>

@@ -49,6 +50,16 @@ struct fbnic_dev {
 	/* Number of TCQs/RCQs available on hardware */
 	u16 max_num_queues;

+	/* Lock protecting writes to @time_high, @time_offset of fbnic_netdev,
+	 * and the HW time CSR machinery.
+	 */
+	spinlock_t time_lock;
+	/* Externally accessible PTP clock, may be NULL */
+	struct ptp_clock *ptp;
+	struct ptp_clock_info ptp_info;
+	/* Last @time_high refresh time in jiffies (to catch stalls) */
+	unsigned long last_read;
+
 	/* Local copy of hardware statistics */
 	struct fbnic_hw_stats hw_stats;
 };
--- a/drivers/net/ethernet/meta/fbnic/fbnic_csr.h
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_csr.h
@@ -413,6 +413,44 @@ enum {
 #define FBNIC_TMI_DROP_CTRL		0x04401		/* 0x11004 */
 #define FBNIC_TMI_DROP_CTRL_EN			CSR_BIT(0)
 #define FBNIC_CSR_END_TMI		0x0443f	/* CSR section delimiter */
+
+/* Precision Time Protocol Registers */
+#define FBNIC_CSR_START_PTP		0x04800 /* CSR section delimiter */
+#define FBNIC_PTP_REG_BASE		0x04800		/* 0x12000 */
+
+#define FBNIC_PTP_CTRL			0x04800		/* 0x12000 */
+#define FBNIC_PTP_CTRL_EN			CSR_BIT(0)
+#define FBNIC_PTP_CTRL_MONO_EN			CSR_BIT(4)
+#define FBNIC_PTP_CTRL_TQS_OUT_EN		CSR_BIT(8)
+#define FBNIC_PTP_CTRL_MAC_OUT_IVAL		CSR_GENMASK(16, 12)
+#define FBNIC_PTP_CTRL_TICK_IVAL		CSR_GENMASK(23, 20)
+
+#define FBNIC_PTP_ADJUST		0x04801		/* 0x12004 */
+#define FBNIC_PTP_ADJUST_INIT			CSR_BIT(0)
+#define FBNIC_PTP_ADJUST_SUB_NUDGE		CSR_BIT(8)
+#define FBNIC_PTP_ADJUST_ADD_NUDGE		CSR_BIT(16)
+#define FBNIC_PTP_ADJUST_ADDEND_SET		CSR_BIT(24)
+
+#define FBNIC_PTP_INIT_HI		0x04802		/* 0x12008 */
+#define FBNIC_PTP_INIT_LO		0x04803		/* 0x1200c */
+
+#define FBNIC_PTP_NUDGE_NS		0x04804		/* 0x12010 */
+#define FBNIC_PTP_NUDGE_SUBNS		0x04805		/* 0x12014 */
+
+#define FBNIC_PTP_ADD_VAL_NS		0x04806		/* 0x12018 */
+#define FBNIC_PTP_ADD_VAL_NS_MASK		CSR_GENMASK(15, 0)
+#define FBNIC_PTP_ADD_VAL_SUBNS		0x04807	/* 0x1201c */
+
+#define FBNIC_PTP_CTR_VAL_HI		0x04808		/* 0x12020 */
+#define FBNIC_PTP_CTR_VAL_LO		0x04809		/* 0x12024 */
+
+#define FBNIC_PTP_MONO_PTP_CTR_HI	0x0480a		/* 0x12028 */
+#define FBNIC_PTP_MONO_PTP_CTR_LO	0x0480b		/* 0x1202c */
+
+#define FBNIC_PTP_CDC_FIFO_STATUS	0x0480c		/* 0x12030 */
+#define FBNIC_PTP_SPARE			0x0480d		/* 0x12034 */
+#define FBNIC_CSR_END_PTP		0x0480d /* CSR section delimiter */
+
 /* Rx Buffer Registers */
 #define FBNIC_CSR_START_RXB		0x08000	/* CSR section delimiter */
 enum {
@@ -548,6 +586,7 @@ enum {
 };

 #define FBNIC_RPC_ACT_TBL0_DMA_HINT		CSR_GENMASK(24, 16)
+#define FBNIC_RPC_ACT_TBL0_TS_ENA		CSR_BIT(28)
 #define FBNIC_RPC_ACT_TBL0_RSS_CTXT_ID		CSR_BIT(30)

 #define FBNIC_RPC_ACT_TBL1_DEFAULT	0x0840b		/* 0x2102c */
--- a/drivers/net/ethernet/meta/fbnic/fbnic_ethtool.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_ethtool.c
@@ -6,6 +6,35 @@
 #include "fbnic_netdev.h"
 #include "fbnic_tlv.h"

+static int
+fbnic_get_ts_info(struct net_device *netdev,
+		  struct kernel_ethtool_ts_info *tsinfo)
+{
+	struct fbnic_net *fbn = netdev_priv(netdev);
+
+	tsinfo->phc_index = ptp_clock_index(fbn->fbd->ptp);
+
+	tsinfo->so_timestamping =
+		SOF_TIMESTAMPING_TX_SOFTWARE |
+		SOF_TIMESTAMPING_TX_HARDWARE |
+		SOF_TIMESTAMPING_RX_HARDWARE |
+		SOF_TIMESTAMPING_RAW_HARDWARE;
+
+	tsinfo->tx_types =
+		BIT(HWTSTAMP_TX_OFF) |
+		BIT(HWTSTAMP_TX_ON);
+
+	tsinfo->rx_filters =
+		BIT(HWTSTAMP_FILTER_NONE) |
+		BIT(HWTSTAMP_FILTER_PTP_V1_L4_EVENT) |
+		BIT(HWTSTAMP_FILTER_PTP_V2_L4_EVENT) |
+		BIT(HWTSTAMP_FILTER_PTP_V2_L2_EVENT) |
+		BIT(HWTSTAMP_FILTER_PTP_V2_EVENT) |
+		BIT(HWTSTAMP_FILTER_ALL);
+
+	return 0;
+}
+
 static void
 fbnic_get_drvinfo(struct net_device *netdev, struct ethtool_drvinfo *drvinfo)
 {
@@ -64,8 +93,33 @@ fbnic_get_eth_mac_stats(struct net_device *netdev,
 			  &mac_stats->eth_mac.FrameTooLongErrors);
 }

+static void fbnic_get_ts_stats(struct net_device *netdev,
+			       struct ethtool_ts_stats *ts_stats)
+{
+	struct fbnic_net *fbn = netdev_priv(netdev);
+	u64 ts_packets, ts_lost;
+	struct fbnic_ring *ring;
+	unsigned int start;
+	int i;
+
+	ts_stats->pkts = fbn->tx_stats.ts_packets;
+	ts_stats->lost = fbn->tx_stats.ts_lost;
+	for (i = 0; i < fbn->num_tx_queues; i++) {
+		ring = fbn->tx[i];
+		do {
+			start = u64_stats_fetch_begin(&ring->stats.syncp);
+			ts_packets = ring->stats.ts_packets;
+			ts_lost = ring->stats.ts_lost;
+		} while (u64_stats_fetch_retry(&ring->stats.syncp, start));
+		ts_stats->pkts += ts_packets;
+		ts_stats->lost += ts_lost;
+	}
+}
+
 static const struct ethtool_ops fbnic_ethtool_ops = {
 	.get_drvinfo		= fbnic_get_drvinfo,
+	.get_ts_info		= fbnic_get_ts_info,
+	.get_ts_stats		= fbnic_get_ts_stats,
 	.get_eth_mac_stats	= fbnic_get_eth_mac_stats,
 };

--- a/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_netdev.c
@@ -42,18 +42,24 @@ int __fbnic_open(struct fbnic_net *fbn)
 		goto free_resources;
 	}

-	err = fbnic_fw_init_heartbeat(fbd, false);
+	err = fbnic_time_start(fbn);
 	if (err)
 		goto release_ownership;

+	err = fbnic_fw_init_heartbeat(fbd, false);
+	if (err)
+		goto time_stop;
+
 	err = fbnic_pcs_irq_enable(fbd);
 	if (err)
-		goto release_ownership;
+		goto time_stop;
 	/* Pull the BMC config and initialize the RPC */
 	fbnic_bmc_rpc_init(fbd);
 	fbnic_rss_reinit(fbd, fbn);

 	return 0;
+time_stop:
+	fbnic_time_stop(fbn);
 release_ownership:
 	fbnic_fw_xmit_ownership_msg(fbn->fbd, false);
 free_resources:
@@ -82,6 +88,7 @@ static int fbnic_stop(struct net_device *netdev)
 	fbnic_down(fbn);
 	fbnic_pcs_irq_disable(fbn->fbd);

+	fbnic_time_stop(fbn);
 	fbnic_fw_xmit_ownership_msg(fbn->fbd, false);

 	fbnic_free_resources(fbn);
@@ -317,6 +324,84 @@ void fbnic_clear_rx_mode(struct net_device *netdev)
 	__dev_mc_unsync(netdev, NULL);
 }

+static int fbnic_hwtstamp_get(struct net_device *netdev,
+			      struct kernel_hwtstamp_config *config)
+{
+	struct fbnic_net *fbn = netdev_priv(netdev);
+
+	*config = fbn->hwtstamp_config;
+
+	return 0;
+}
+
+static int fbnic_hwtstamp_set(struct net_device *netdev,
+			      struct kernel_hwtstamp_config *config,
+			      struct netlink_ext_ack *extack)
+{
+	struct fbnic_net *fbn = netdev_priv(netdev);
+	int old_rx_filter;
+
+	if (config->source != HWTSTAMP_SOURCE_NETDEV)
+		return -EOPNOTSUPP;
+
+	if (!kernel_hwtstamp_config_changed(config, &fbn->hwtstamp_config))
+		return 0;
+
+	/* Upscale the filters */
+	switch (config->rx_filter) {
+	case HWTSTAMP_FILTER_NONE:
+	case HWTSTAMP_FILTER_ALL:
+	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_EVENT:
+		break;
+	case HWTSTAMP_FILTER_NTP_ALL:
+		config->rx_filter = HWTSTAMP_FILTER_ALL;
+		break;
+	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
+	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
+		config->rx_filter = HWTSTAMP_FILTER_PTP_V1_L4_EVENT;
+		break;
+	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
+		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_L4_EVENT;
+		break;
+	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
+		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_L2_EVENT;
+		break;
+	case HWTSTAMP_FILTER_PTP_V2_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
+		config->rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
+		break;
+	default:
+		return -ERANGE;
+	}
+
+	/* Configure */
+	old_rx_filter = fbn->hwtstamp_config.rx_filter;
+	memcpy(&fbn->hwtstamp_config, config, sizeof(*config));
+
+	if (old_rx_filter != config->rx_filter && netif_running(fbn->netdev)) {
+		fbnic_rss_reinit(fbn->fbd, fbn);
+		fbnic_write_rules(fbn->fbd);
+	}
+
+	/* Save / report back filter configuration
+	 * Note that our filter configuration is inexact. Instead of
+	 * filtering for a specific UDP port or L2 Ethertype we are
+	 * filtering in all UDP or all non-IP packets for timestamping. So
+	 * if anything other than FILTER_ALL is requested we report
+	 * FILTER_SOME indicating that we will be timestamping a few
+	 * additional packets.
+	 */
+	if (config->rx_filter > HWTSTAMP_FILTER_ALL)
+		config->rx_filter = HWTSTAMP_FILTER_SOME;
+
+	return 0;
+}
+
 static void fbnic_get_stats64(struct net_device *dev,
 			      struct rtnl_link_stats64 *stats64)
 {
@@ -394,6 +479,8 @@ static const struct net_device_ops fbnic_netdev_ops = {
 	.ndo_set_mac_address	= fbnic_set_mac,
 	.ndo_set_rx_mode	= fbnic_set_rx_mode,
 	.ndo_get_stats64	= fbnic_get_stats64,
+	.ndo_hwtstamp_get	= fbnic_hwtstamp_get,
+	.ndo_hwtstamp_set	= fbnic_hwtstamp_set,
 };

 static void fbnic_get_queue_stats_rx(struct net_device *dev, int idx,
--- a/drivers/net/ethernet/meta/fbnic/fbnic_netdev.h
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_netdev.h
@@ -33,6 +33,15 @@ struct fbnic_net {
 	u8 fec;
 	u8 link_mode;

+	/* Cached top bits of the HW time counter for 40b -> 64b conversion */
+	u32 time_high;
+	/* Protect readers of @time_offset, writers take @time_lock. */
+	struct u64_stats_sync time_seq;
+	/* Offset in ns between free running NIC PHC and time set via PTP
+	 * clock callbacks
+	 */
+	s64 time_offset;
+
 	u16 num_tx_queues;
 	u16 num_rx_queues;

@@ -45,6 +54,9 @@ struct fbnic_net {
 	struct fbnic_queue_stats rx_stats;
 	u64 link_down_events;

+	/* Time stampinn filter config */
+	struct kernel_hwtstamp_config hwtstamp_config;
+
 	struct list_head napis;
 };

@@ -60,6 +72,12 @@ void fbnic_reset_queues(struct fbnic_net *fbn,
 			unsigned int tx, unsigned int rx);
 void fbnic_set_ethtool_ops(struct net_device *dev);

+int fbnic_ptp_setup(struct fbnic_dev *fbd);
+void fbnic_ptp_destroy(struct fbnic_dev *fbd);
+void fbnic_time_init(struct fbnic_net *fbn);
+int fbnic_time_start(struct fbnic_net *fbn);
+void fbnic_time_stop(struct fbnic_net *fbn);
+
 void __fbnic_set_rx_mode(struct net_device *netdev);
 void fbnic_clear_rx_mode(struct net_device *netdev);

--- a/drivers/net/ethernet/meta/fbnic/fbnic_pci.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_pci.c
@@ -300,14 +300,20 @@ static int fbnic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 		goto init_failure_mode;
 	}

+	err = fbnic_ptp_setup(fbd);
+	if (err)
+		goto ifm_free_netdev;
+
 	err = fbnic_netdev_register(netdev);
 	if (err) {
 		dev_err(&pdev->dev, "Netdev registration failed: %d\n", err);
-		goto ifm_free_netdev;
+		goto ifm_destroy_ptp;
 	}

 	return 0;

+ifm_destroy_ptp:
+	fbnic_ptp_destroy(fbd);
 ifm_free_netdev:
 	fbnic_netdev_free(fbd);
 init_failure_mode:
@@ -342,6 +348,7 @@ static void fbnic_remove(struct pci_dev *pdev)

 		fbnic_netdev_unregister(netdev);
 		cancel_delayed_work_sync(&fbd->service_task);
+		fbnic_ptp_destroy(fbd);
 		fbnic_netdev_free(fbd);
 	}

--- a/drivers/net/ethernet/meta/fbnic/fbnic_rpc.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_rpc.c
@@ -244,6 +244,12 @@ void fbnic_bmc_rpc_init(struct fbnic_dev *fbd)
 	 ((_ip) ? FBNIC_RPC_TCAM_ACT1_IP_VALID : 0) |	\
 	 ((_v6) ? FBNIC_RPC_TCAM_ACT1_IP_IS_V6 : 0))

+#define FBNIC_TSTAMP_MASK(_all, _udp, _ether)			\
+	(((_all) ? ((1u << FBNIC_NUM_HASH_OPT) - 1) : 0) |	\
+	 ((_udp) ? (1u << FBNIC_UDP6_HASH_OPT) |		\
+		   (1u << FBNIC_UDP4_HASH_OPT) : 0) |		\
+	 ((_ether) ? (1u << FBNIC_ETHER_HASH_OPT) : 0))
+
 void fbnic_rss_reinit(struct fbnic_dev *fbd, struct fbnic_net *fbn)
 {
 	static const u32 act1_value[FBNIC_NUM_HASH_OPT] = {
@@ -255,6 +261,7 @@ void fbnic_rss_reinit(struct fbnic_dev *fbd, struct fbnic_net *fbn)
 		FBNIC_ACT1_INIT(0, 0, 1, 0),	/* IP4 */
 		0				/* Ether */
 	};
+	u32 tstamp_mask = 0;
 	unsigned int i;

 	/* To support scenarios where a BMC is present we must write the
@@ -264,6 +271,28 @@ void fbnic_rss_reinit(struct fbnic_dev *fbd, struct fbnic_net *fbn)
 	BUILD_BUG_ON(FBNIC_RSS_EN_NUM_UNICAST * 2 != FBNIC_RSS_EN_NUM_ENTRIES);
 	BUILD_BUG_ON(ARRAY_SIZE(act1_value) != FBNIC_NUM_HASH_OPT);

+	/* Set timestamp mask with 1b per flow type */
+	if (fbn->hwtstamp_config.rx_filter != HWTSTAMP_FILTER_NONE) {
+		switch (fbn->hwtstamp_config.rx_filter) {
+		case HWTSTAMP_FILTER_ALL:
+			tstamp_mask = FBNIC_TSTAMP_MASK(1, 1, 1);
+			break;
+		case HWTSTAMP_FILTER_PTP_V2_EVENT:
+			tstamp_mask = FBNIC_TSTAMP_MASK(0, 1, 1);
+			break;
+		case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
+		case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
+			tstamp_mask = FBNIC_TSTAMP_MASK(0, 1, 0);
+			break;
+		case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
+			tstamp_mask = FBNIC_TSTAMP_MASK(0, 0, 1);
+			break;
+		default:
+			netdev_warn(fbn->netdev, "Unsupported hwtstamp_rx_filter\n");
+			break;
+		}
+	}
+
 	/* Program RSS hash enable mask for host in action TCAM/table. */
 	for (i = fbnic_bmc_present(fbd) ? 0 : FBNIC_RSS_EN_NUM_UNICAST;
 	     i < FBNIC_RSS_EN_NUM_ENTRIES; i++) {
@@ -287,6 +316,8 @@ void fbnic_rss_reinit(struct fbnic_dev *fbd, struct fbnic_net *fbn)

 		if (!dest)
 			dest = FBNIC_RPC_ACT_TBL0_DROP;
+		else if (tstamp_mask & (1u << flow_type))
+			dest |= FBNIC_RPC_ACT_TBL0_TS_ENA;

 		if (act1_value[flow_type] & FBNIC_RPC_TCAM_ACT1_L4_VALID)
 			dest |= FIELD_PREP(FBNIC_RPC_ACT_TBL0_DMA_HINT,
--- a/drivers/net/ethernet/meta/fbnic/fbnic_time.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_time.c
@@ -0,0 +1,303 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) Meta Platforms, Inc. and affiliates. */
+
+#include <linux/bitfield.h>
+#include <linux/jiffies.h>
+#include <linux/limits.h>
+#include <linux/ptp_clock_kernel.h>
+#include <linux/timer.h>
+
+#include "fbnic.h"
+#include "fbnic_csr.h"
+#include "fbnic_netdev.h"
+
+/* FBNIC timing & PTP implementation
+ * Datapath uses truncated 40b timestamps for scheduling and event reporting.
+ * We need to promote those to full 64b, hence we periodically cache the top
+ * 32bit of the HW time counter. Since this makes our time reporting non-atomic
+ * we leave the HW clock free running and adjust time offsets in SW as needed.
+ * Time offset is 64bit - we need a seq counter for 32bit machines.
+ * Time offset and the cache of top bits are independent so we don't need
+ * a coherent snapshot of both - READ_ONCE()/WRITE_ONCE() + writer side lock
+ * are enough.
+ */
+
+/* Period of refresh of top bits of timestamp, give ourselves a 8x margin.
+ * This should translate to once a minute.
+ * The use of nsecs_to_jiffies() should be safe for a <=40b nsec value.
+ */
+#define FBNIC_TS_HIGH_REFRESH_JIF	nsecs_to_jiffies((1ULL << 40) / 16)
+
+static struct fbnic_dev *fbnic_from_ptp_info(struct ptp_clock_info *ptp)
+{
+	return container_of(ptp, struct fbnic_dev, ptp_info);
+}
+
+/* This function is "slow" because we could try guessing which high part
+ * is correct based on low instead of re-reading, and skip reading @hi
+ * twice altogether if @lo is far enough from 0.
+ */
+static u64 __fbnic_time_get_slow(struct fbnic_dev *fbd)
+{
+	u32 hi, lo;
+
+	lockdep_assert_held(&fbd->time_lock);
+
+	do {
+		hi = fbnic_rd32(fbd, FBNIC_PTP_CTR_VAL_HI);
+		lo = fbnic_rd32(fbd, FBNIC_PTP_CTR_VAL_LO);
+	} while (hi != fbnic_rd32(fbd, FBNIC_PTP_CTR_VAL_HI));
+
+	return (u64)hi << 32 | lo;
+}
+
+static void __fbnic_time_set_addend(struct fbnic_dev *fbd, u64 addend)
+{
+	lockdep_assert_held(&fbd->time_lock);
+
+	fbnic_wr32(fbd, FBNIC_PTP_ADD_VAL_NS,
+		   FIELD_PREP(FBNIC_PTP_ADD_VAL_NS_MASK, addend >> 32));
+	fbnic_wr32(fbd, FBNIC_PTP_ADD_VAL_SUBNS, (u32)addend);
+}
+
+static void fbnic_ptp_fresh_check(struct fbnic_dev *fbd)
+{
+	if (time_is_after_jiffies(fbd->last_read +
+				  FBNIC_TS_HIGH_REFRESH_JIF * 3 / 2))
+		return;
+
+	dev_warn(fbd->dev, "NIC timestamp refresh stall, delayed by %lu sec\n",
+		 (jiffies - fbd->last_read - FBNIC_TS_HIGH_REFRESH_JIF) / HZ);
+}
+
+static void fbnic_ptp_refresh_time(struct fbnic_dev *fbd, struct fbnic_net *fbn)
+{
+	unsigned long flags;
+	u32 hi;
+
+	spin_lock_irqsave(&fbd->time_lock, flags);
+	hi = fbnic_rd32(fbn->fbd, FBNIC_PTP_CTR_VAL_HI);
+	if (!fbnic_present(fbd))
+		goto out; /* Don't bother handling, reset is pending */
+	/* Let's keep high cached value a bit lower to avoid race with
+	 * incoming timestamps. The logic in fbnic_ts40_to_ns() will
+	 * take care of overflow in this case. It will make cached time
+	 * ~1 minute lower and incoming timestamp will always be later
+	 * then cached time.
+	 */
+	WRITE_ONCE(fbn->time_high, hi - 16);
+	fbd->last_read = jiffies;
+ out:
+	spin_unlock_irqrestore(&fbd->time_lock, flags);
+}
+
+static long fbnic_ptp_do_aux_work(struct ptp_clock_info *ptp)
+{
+	struct fbnic_dev *fbd = fbnic_from_ptp_info(ptp);
+	struct fbnic_net *fbn;
+
+	fbn = netdev_priv(fbd->netdev);
+
+	fbnic_ptp_fresh_check(fbd);
+	fbnic_ptp_refresh_time(fbd, fbn);
+
+	return FBNIC_TS_HIGH_REFRESH_JIF;
+}
+
+static int fbnic_ptp_adjfine(struct ptp_clock_info *ptp, long scaled_ppm)
+{
+	struct fbnic_dev *fbd = fbnic_from_ptp_info(ptp);
+	u64 addend, dclk_period;
+	unsigned long flags;
+
+	/* d_clock is 600 MHz; which in Q16.32 fixed point ns is: */
+	dclk_period = (((u64)1000000000) << 32) / FBNIC_CLOCK_FREQ;
+	addend = adjust_by_scaled_ppm(dclk_period, scaled_ppm);
+
+	spin_lock_irqsave(&fbd->time_lock, flags);
+	__fbnic_time_set_addend(fbd, addend);
+	fbnic_wr32(fbd, FBNIC_PTP_ADJUST, FBNIC_PTP_ADJUST_ADDEND_SET);
+
+	/* Flush, make sure FBNIC_PTP_ADD_VAL_* is stable for at least 4 clks */
+	fbnic_rd32(fbd, FBNIC_PTP_SPARE);
+	spin_unlock_irqrestore(&fbd->time_lock, flags);
+
+	return fbnic_present(fbd) ? 0 : -EIO;
+}
+
+static int fbnic_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
+{
+	struct fbnic_dev *fbd = fbnic_from_ptp_info(ptp);
+	struct fbnic_net *fbn;
+	unsigned long flags;
+
+	fbn = netdev_priv(fbd->netdev);
+
+	spin_lock_irqsave(&fbd->time_lock, flags);
+	u64_stats_update_begin(&fbn->time_seq);
+	WRITE_ONCE(fbn->time_offset, READ_ONCE(fbn->time_offset) + delta);
+	u64_stats_update_end(&fbn->time_seq);
+	spin_unlock_irqrestore(&fbd->time_lock, flags);
+
+	return 0;
+}
+
+static int
+fbnic_ptp_gettimex64(struct ptp_clock_info *ptp, struct timespec64 *ts,
+		     struct ptp_system_timestamp *sts)
+{
+	struct fbnic_dev *fbd = fbnic_from_ptp_info(ptp);
+	struct fbnic_net *fbn;
+	unsigned long flags;
+	u64 time_ns;
+	u32 hi, lo;
+
+	fbn = netdev_priv(fbd->netdev);
+
+	spin_lock_irqsave(&fbd->time_lock, flags);
+
+	do {
+		hi = fbnic_rd32(fbd, FBNIC_PTP_CTR_VAL_HI);
+		ptp_read_system_prets(sts);
+		lo = fbnic_rd32(fbd, FBNIC_PTP_CTR_VAL_LO);
+		ptp_read_system_postts(sts);
+		/* Similarly to comment above __fbnic_time_get_slow()
+		 * - this can be optimized if needed.
+		 */
+	} while (hi != fbnic_rd32(fbd, FBNIC_PTP_CTR_VAL_HI));
+
+	time_ns = ((u64)hi << 32 | lo) + fbn->time_offset;
+	spin_unlock_irqrestore(&fbd->time_lock, flags);
+
+	if (!fbnic_present(fbd))
+		return -EIO;
+
+	*ts = ns_to_timespec64(time_ns);
+
+	return 0;
+}
+
+static int
+fbnic_ptp_settime64(struct ptp_clock_info *ptp, const struct timespec64 *ts)
+{
+	struct fbnic_dev *fbd = fbnic_from_ptp_info(ptp);
+	struct fbnic_net *fbn;
+	unsigned long flags;
+	u64 dev_ns, host_ns;
+	int ret;
+
+	fbn = netdev_priv(fbd->netdev);
+
+	host_ns = timespec64_to_ns(ts);
+
+	spin_lock_irqsave(&fbd->time_lock, flags);
+
+	dev_ns = __fbnic_time_get_slow(fbd);
+
+	if (fbnic_present(fbd)) {
+		u64_stats_update_begin(&fbn->time_seq);
+		WRITE_ONCE(fbn->time_offset, host_ns - dev_ns);
+		u64_stats_update_end(&fbn->time_seq);
+		ret = 0;
+	} else {
+		ret = -EIO;
+	}
+	spin_unlock_irqrestore(&fbd->time_lock, flags);
+
+	return ret;
+}
+
+static const struct ptp_clock_info fbnic_ptp_info = {
+	.owner			= THIS_MODULE,
+	/* 1,000,000,000 - 1 PPB to ensure increment is positive
+	 * after max negative adjustment.
+	 */
+	.max_adj		= 999999999,
+	.do_aux_work		= fbnic_ptp_do_aux_work,
+	.adjfine		= fbnic_ptp_adjfine,
+	.adjtime		= fbnic_ptp_adjtime,
+	.gettimex64		= fbnic_ptp_gettimex64,
+	.settime64		= fbnic_ptp_settime64,
+};
+
+static void fbnic_ptp_reset(struct fbnic_dev *fbd)
+{
+	struct fbnic_net *fbn = netdev_priv(fbd->netdev);
+	u64 dclk_period;
+
+	fbnic_wr32(fbd, FBNIC_PTP_CTRL,
+		   FBNIC_PTP_CTRL_EN |
+		   FIELD_PREP(FBNIC_PTP_CTRL_TICK_IVAL, 1));
+
+	/* d_clock is 600 MHz; which in Q16.32 fixed point ns is: */
+	dclk_period = (((u64)1000000000) << 32) / FBNIC_CLOCK_FREQ;
+
+	__fbnic_time_set_addend(fbd, dclk_period);
+
+	fbnic_wr32(fbd, FBNIC_PTP_INIT_HI, 0);
+	fbnic_wr32(fbd, FBNIC_PTP_INIT_LO, 0);
+
+	fbnic_wr32(fbd, FBNIC_PTP_ADJUST, FBNIC_PTP_ADJUST_INIT);
+
+	fbnic_wr32(fbd, FBNIC_PTP_CTRL,
+		   FBNIC_PTP_CTRL_EN |
+		   FBNIC_PTP_CTRL_TQS_OUT_EN |
+		   FIELD_PREP(FBNIC_PTP_CTRL_MAC_OUT_IVAL, 3) |
+		   FIELD_PREP(FBNIC_PTP_CTRL_TICK_IVAL, 1));
+
+	fbnic_rd32(fbd, FBNIC_PTP_SPARE);
+
+	fbn->time_offset = 0;
+	fbn->time_high = 0;
+}
+
+void fbnic_time_init(struct fbnic_net *fbn)
+{
+	/* This is not really a statistic, but the lockng primitive fits
+	 * our usecase perfectly, we need an atomic 8 bytes READ_ONCE() /
+	 * WRITE_ONCE() behavior.
+	 */
+	u64_stats_init(&fbn->time_seq);
+}
+
+int fbnic_time_start(struct fbnic_net *fbn)
+{
+	fbnic_ptp_refresh_time(fbn->fbd, fbn);
+	/* Assume that fbnic_ptp_do_aux_work() will never be called if not
+	 * scheduled here
+	 */
+	return ptp_schedule_worker(fbn->fbd->ptp, FBNIC_TS_HIGH_REFRESH_JIF);
+}
+
+void fbnic_time_stop(struct fbnic_net *fbn)
+{
+	ptp_cancel_worker_sync(fbn->fbd->ptp);
+	fbnic_ptp_fresh_check(fbn->fbd);
+}
+
+int fbnic_ptp_setup(struct fbnic_dev *fbd)
+{
+	struct device *dev = fbd->dev;
+	unsigned long flags;
+
+	spin_lock_init(&fbd->time_lock);
+
+	spin_lock_irqsave(&fbd->time_lock, flags); /* Appease lockdep */
+	fbnic_ptp_reset(fbd);
+	spin_unlock_irqrestore(&fbd->time_lock, flags);
+
+	memcpy(&fbd->ptp_info, &fbnic_ptp_info, sizeof(fbnic_ptp_info));
+
+	fbd->ptp = ptp_clock_register(&fbd->ptp_info, dev);
+	if (IS_ERR(fbd->ptp))
+		dev_err(dev, "Failed to register PTP: %pe\n", fbd->ptp);
+
+	return PTR_ERR_OR_ZERO(fbd->ptp);
+}
+
+void fbnic_ptp_destroy(struct fbnic_dev *fbd)
+{
+	if (!fbd->ptp)
+		return;
+	ptp_clock_unregister(fbd->ptp);
+}
--- a/drivers/net/ethernet/meta/fbnic/fbnic_txrx.c
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_txrx.c
@@ -12,9 +12,14 @@
 #include "fbnic_netdev.h"
 #include "fbnic_txrx.h"

+enum {
+	FBNIC_XMIT_CB_TS	= 0x01,
+};
+
 struct fbnic_xmit_cb {
 	u32 bytecount;
 	u8 desc_count;
+	u8 flags;
 	int hw_head;
 };

@@ -43,6 +48,46 @@ static void fbnic_ring_wr32(struct fbnic_ring *ring, unsigned int csr, u32 val)
 	writel(val, csr_base + csr);
 }

+/**
+ * fbnic_ts40_to_ns() - convert descriptor timestamp to PHC time
+ * @fbn: netdev priv of the FB NIC
+ * @ts40: timestamp read from a descriptor
+ *
+ * Return: u64 value of PHC time in nanoseconds
+ *
+ * Convert truncated 40 bit device timestamp as read from a descriptor
+ * to the full PHC time in nanoseconds.
+ */
+static __maybe_unused u64 fbnic_ts40_to_ns(struct fbnic_net *fbn, u64 ts40)
+{
+	unsigned int s;
+	u64 time_ns;
+	s64 offset;
+	u8 ts_top;
+	u32 high;
+
+	do {
+		s = u64_stats_fetch_begin(&fbn->time_seq);
+		offset = READ_ONCE(fbn->time_offset);
+	} while (u64_stats_fetch_retry(&fbn->time_seq, s));
+
+	high = READ_ONCE(fbn->time_high);
+
+	/* Bits 63..40 from periodic clock reads, 39..0 from ts40 */
+	time_ns = (u64)(high >> 8) << 40 | ts40;
+
+	/* Compare bits 32-39 between periodic reads and ts40,
+	 * see if HW clock may have wrapped since last read. We are sure
+	 * that periodic reads are always at least ~1 minute behind, so
+	 * this logic works perfectly fine.
+	 */
+	ts_top = ts40 >> 32;
+	if (ts_top < (u8)high && (u8)high - ts_top > U8_MAX / 2)
+		time_ns += 1ULL << 40;
+
+	return time_ns + offset;
+}
+
 static unsigned int fbnic_desc_unused(struct fbnic_ring *ring)
 {
 	return (ring->head - ring->tail - 1) & ring->size_mask;
@@ -110,11 +155,32 @@ static void fbnic_unmap_page_twd(struct device *dev, __le64 *twd)
 #define FBNIC_TWD_TYPE(_type) \
 	cpu_to_le64(FIELD_PREP(FBNIC_TWD_TYPE_MASK, FBNIC_TWD_TYPE_##_type))

+static bool fbnic_tx_tstamp(struct sk_buff *skb)
+{
+	struct fbnic_net *fbn;
+
+	if (!unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
+		return false;
+
+	fbn = netdev_priv(skb->dev);
+	if (fbn->hwtstamp_config.tx_type == HWTSTAMP_TX_OFF)
+		return false;
+
+	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+	FBNIC_XMIT_CB(skb)->flags |= FBNIC_XMIT_CB_TS;
+	FBNIC_XMIT_CB(skb)->hw_head = -1;
+
+	return true;
+}
+
 static bool
 fbnic_tx_offloads(struct fbnic_ring *ring, struct sk_buff *skb, __le64 *meta)
 {
 	unsigned int l2len, i3len;

+	if (fbnic_tx_tstamp(skb))
+		*meta |= cpu_to_le64(FBNIC_TWD_FLAG_REQ_TS);
+
 	if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL))
 		return false;

@@ -205,6 +271,9 @@ fbnic_tx_map(struct fbnic_ring *ring, struct sk_buff *skb, __le64 *meta)

 	ring->tail = tail;

+	/* Record SW timestamp */
+	skb_tx_timestamp(skb);
+
 	/* Verify there is room for another packet */
 	fbnic_maybe_stop_tx(skb->dev, ring, FBNIC_MAX_SKB_DESC);

@@ -316,7 +385,7 @@ static void fbnic_clean_twq0(struct fbnic_napi_vector *nv, int napi_budget,
 			     struct fbnic_ring *ring, bool discard,
 			     unsigned int hw_head)
 {
-	u64 total_bytes = 0, total_packets = 0;
+	u64 total_bytes = 0, total_packets = 0, ts_lost = 0;
 	unsigned int head = ring->head;
 	struct netdev_queue *txq;
 	unsigned int clean_desc;
@@ -331,6 +400,13 @@ static void fbnic_clean_twq0(struct fbnic_napi_vector *nv, int napi_budget,
 		if (desc_cnt > clean_desc)
 			break;

+		if (unlikely(FBNIC_XMIT_CB(skb)->flags & FBNIC_XMIT_CB_TS)) {
+			FBNIC_XMIT_CB(skb)->hw_head = hw_head;
+			if (likely(!discard))
+				break;
+			ts_lost++;
+		}
+
 		ring->tx_buf[head] = NULL;

 		clean_desc -= desc_cnt;
@@ -368,6 +444,7 @@ static void fbnic_clean_twq0(struct fbnic_napi_vector *nv, int napi_budget,
 	if (unlikely(discard)) {
 		u64_stats_update_begin(&ring->stats.syncp);
 		ring->stats.dropped += total_packets;
+		ring->stats.ts_lost += ts_lost;
 		u64_stats_update_end(&ring->stats.syncp);

 		netdev_tx_completed_queue(txq, total_packets, total_bytes);
@@ -384,6 +461,56 @@ static void fbnic_clean_twq0(struct fbnic_napi_vector *nv, int napi_budget,
 				 FBNIC_TX_DESC_WAKEUP);
 }

+static void fbnic_clean_tsq(struct fbnic_napi_vector *nv,
+			    struct fbnic_ring *ring,
+			    u64 tcd, int *ts_head, int *head0)
+{
+	struct skb_shared_hwtstamps hwtstamp;
+	struct fbnic_net *fbn;
+	struct sk_buff *skb;
+	int head;
+	u64 ns;
+
+	head = (*ts_head < 0) ? ring->head : *ts_head;
+
+	do {
+		unsigned int desc_cnt;
+
+		if (head == ring->tail) {
+			if (unlikely(net_ratelimit()))
+				netdev_err(nv->napi.dev,
+					   "Tx timestamp without matching packet\n");
+			return;
+		}
+
+		skb = ring->tx_buf[head];
+		desc_cnt = FBNIC_XMIT_CB(skb)->desc_count;
+
+		head += desc_cnt;
+		head &= ring->size_mask;
+	} while (!(FBNIC_XMIT_CB(skb)->flags & FBNIC_XMIT_CB_TS));
+
+	fbn = netdev_priv(nv->napi.dev);
+	ns = fbnic_ts40_to_ns(fbn, FIELD_GET(FBNIC_TCD_TYPE1_TS_MASK, tcd));
+
+	memset(&hwtstamp, 0, sizeof(hwtstamp));
+	hwtstamp.hwtstamp = ns_to_ktime(ns);
+
+	*ts_head = head;
+
+	FBNIC_XMIT_CB(skb)->flags &= ~FBNIC_XMIT_CB_TS;
+	if (*head0 < 0) {
+		head = FBNIC_XMIT_CB(skb)->hw_head;
+		if (head >= 0)
+			*head0 = head;
+	}
+
+	skb_tstamp_tx(skb, &hwtstamp);
+	u64_stats_update_begin(&ring->stats.syncp);
+	ring->stats.ts_packets++;
+	u64_stats_update_end(&ring->stats.syncp);
+}
+
 static void fbnic_page_pool_init(struct fbnic_ring *ring, unsigned int idx,
 				 struct page *page)
 {
@@ -417,10 +544,12 @@ static void fbnic_page_pool_drain(struct fbnic_ring *ring, unsigned int idx,
 }

 static void fbnic_clean_twq(struct fbnic_napi_vector *nv, int napi_budget,
-			    struct fbnic_q_triad *qt, s32 head0)
+			    struct fbnic_q_triad *qt, s32 ts_head, s32 head0)
 {
 	if (head0 >= 0)
 		fbnic_clean_twq0(nv, napi_budget, &qt->sub0, false, head0);
+	else if (ts_head >= 0)
+		fbnic_clean_twq0(nv, napi_budget, &qt->sub0, false, ts_head);
 }

 static void
@@ -428,9 +557,9 @@ fbnic_clean_tcq(struct fbnic_napi_vector *nv, struct fbnic_q_triad *qt,
 		int napi_budget)
 {
 	struct fbnic_ring *cmpl = &qt->cmpl;
+	s32 head0 = -1, ts_head = -1;
 	__le64 *raw_tcd, done;
 	u32 head = cmpl->head;
-	s32 head0 = -1;

 	done = (head & (cmpl->size_mask + 1)) ? 0 : cpu_to_le64(FBNIC_TCD_DONE);
 	raw_tcd = &cmpl->desc[head & cmpl->size_mask];
@@ -453,6 +582,12 @@ fbnic_clean_tcq(struct fbnic_napi_vector *nv, struct fbnic_q_triad *qt,
 			 * they are skipped for now.
 			 */
 			break;
+		case FBNIC_TCD_TYPE_1:
+			if (WARN_ON_ONCE(tcd & FBNIC_TCD_TWQ1))
+				break;
+
+			fbnic_clean_tsq(nv, &qt->sub0, tcd, &ts_head, &head0);
+			break;
 		default:
 			break;
 		}
@@ -472,7 +607,7 @@ fbnic_clean_tcq(struct fbnic_napi_vector *nv, struct fbnic_q_triad *qt,
 	}

 	/* Unmap and free processed buffers */
-	fbnic_clean_twq(nv, napi_budget, qt, head0);
+	fbnic_clean_twq(nv, napi_budget, qt, ts_head, head0);
 }

 static void fbnic_clean_bdq(struct fbnic_napi_vector *nv, int napi_budget,
@@ -707,6 +842,10 @@ static struct sk_buff *fbnic_build_skb(struct fbnic_napi_vector *nv,
 	/* Set MAC header specific fields */
 	skb->protocol = eth_type_trans(skb, nv->napi.dev);

+	/* Add timestamp if present */
+	if (pkt->hwtstamp)
+		skb_hwtstamps(skb)->hwtstamp = pkt->hwtstamp;
+
 	return skb;
 }

@@ -717,6 +856,23 @@ static enum pkt_hash_types fbnic_skb_hash_type(u64 rcd)
 						     PKT_HASH_TYPE_L2;
 }

+static void fbnic_rx_tstamp(struct fbnic_napi_vector *nv, u64 rcd,
+			    struct fbnic_pkt_buff *pkt)
+{
+	struct fbnic_net *fbn;
+	u64 ns, ts;
+
+	if (!FIELD_GET(FBNIC_RCD_OPT_META_TS, rcd))
+		return;
+
+	fbn = netdev_priv(nv->napi.dev);
+	ts = FIELD_GET(FBNIC_RCD_OPT_META_TS_MASK, rcd);
+	ns = fbnic_ts40_to_ns(fbn, ts);
+
+	/* Add timestamp to shared info */
+	pkt->hwtstamp = ns_to_ktime(ns);
+}
+
 static void fbnic_populate_skb_fields(struct fbnic_napi_vector *nv,
 				      u64 rcd, struct sk_buff *skb,
 				      struct fbnic_q_triad *qt)
@@ -781,6 +937,8 @@ static int fbnic_clean_rcq(struct fbnic_napi_vector *nv,
 			if (FIELD_GET(FBNIC_RCD_OPT_META_TYPE_MASK, rcd))
 				break;

+			fbnic_rx_tstamp(nv, rcd, pkt);
+
 			/* We currently ignore the action table index */
 			break;
 		case FBNIC_RCD_TYPE_META:
@@ -907,6 +1065,8 @@ static void fbnic_aggregate_ring_tx_counters(struct fbnic_net *fbn,
 	fbn->tx_stats.bytes += stats->bytes;
 	fbn->tx_stats.packets += stats->packets;
 	fbn->tx_stats.dropped += stats->dropped;
+	fbn->tx_stats.ts_lost += stats->ts_lost;
+	fbn->tx_stats.ts_packets += stats->ts_packets;
 }

 static void fbnic_remove_tx_ring(struct fbnic_net *fbn,
--- a/drivers/net/ethernet/meta/fbnic/fbnic_txrx.h
+++ b/drivers/net/ethernet/meta/fbnic/fbnic_txrx.h
@@ -47,6 +47,7 @@ struct fbnic_net;

 struct fbnic_pkt_buff {
 	struct xdp_buff buff;
+	ktime_t hwtstamp;
 	u32 data_truesize;
 	u16 data_len;
 	u16 nr_frags;
@@ -56,6 +57,8 @@ struct fbnic_queue_stats {
 	u64 packets;
 	u64 bytes;
 	u64 dropped;
+	u64 ts_packets;
+	u64 ts_lost;
 	struct u64_stats_sync syncp;
 };