sfc: Add RX packet timestamping for EF10

The EF10 firmware can optionally insert RX timestamps in the packet
prefix.  These only include the clock minor value.  We must also
enable periodic time sync events on each event queue which provide
the high bits of the clock value.

[bwh: Combined and rebased several changes.
 Added the above description and some sanity checks for inline vs
 separate timestamps.
 Changed efx_rx_skb_attach_timestamp() to read the packet prefix
 from the skb head area.]
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>

7 files changed, 277 insertions, 15 deletions

diff --git a/drivers/net/ethernet/sfc/ef10.c b/drivers/net/ethernet/sfc/ef10.c
index ca922b958a2e..a3d891764030 100644
--- a/drivers/net/ethernet/sfc/ef10.c
+++ b/drivers/net/ethernet/sfc/ef10.c
@@ -1469,8 +1469,9 @@ static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
 		       efx_rx_queue_index(rx_queue));
-	MCDI_POPULATE_DWORD_1(inbuf, INIT_RXQ_IN_FLAGS,
-			      INIT_RXQ_IN_FLAG_PREFIX, 1);
+	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
+			      INIT_RXQ_IN_FLAG_PREFIX, 1,
+			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
 
@@ -3406,6 +3407,119 @@ static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
 }
 
+static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
+					   bool temp)
+{
+	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
+	int rc;
+
+	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
+	    channel->sync_events_state == SYNC_EVENTS_VALID ||
+	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
+		return 0;
+	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
+
+	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
+	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
+	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
+		       channel->channel);
+
+	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
+			  inbuf, sizeof(inbuf), NULL, 0, NULL);
+
+	if (rc != 0)
+		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
+						    SYNC_EVENTS_DISABLED;
+
+	return rc;
+}
+
+static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
+					    bool temp)
+{
+	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
+	int rc;
+
+	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
+	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
+		return 0;
+	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
+		channel->sync_events_state = SYNC_EVENTS_DISABLED;
+		return 0;
+	}
+	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
+					    SYNC_EVENTS_DISABLED;
+
+	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
+	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
+	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
+		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
+	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
+		       channel->channel);
+
+	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
+			  inbuf, sizeof(inbuf), NULL, 0, NULL);
+
+	return rc;
+}
+
+static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
+					   bool temp)
+{
+	int (*set)(struct efx_channel *channel, bool temp);
+	struct efx_channel *channel;
+
+	set = en ?
+	      efx_ef10_rx_enable_timestamping :
+	      efx_ef10_rx_disable_timestamping;
+
+	efx_for_each_channel(channel, efx) {
+		int rc = set(channel, temp);
+		if (en && rc != 0) {
+			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
+			return rc;
+		}
+	}
+
+	return 0;
+}
+
+static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
+				      struct hwtstamp_config *init)
+{
+	int rc;
+
+	switch (init->rx_filter) {
+	case HWTSTAMP_FILTER_NONE:
+		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
+		/* if TX timestamping is still requested then leave PTP on */
+		return efx_ptp_change_mode(efx,
+					   init->tx_type != HWTSTAMP_TX_OFF, 0);
+	case HWTSTAMP_FILTER_ALL:
+	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
+	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
+	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
+	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
+	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
+	case HWTSTAMP_FILTER_PTP_V2_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
+		init->rx_filter = HWTSTAMP_FILTER_ALL;
+		rc = efx_ptp_change_mode(efx, true, 0);
+		if (!rc)
+			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
+		if (rc)
+			efx_ptp_change_mode(efx, false, 0);
+		return rc;
+	default:
+		return -ERANGE;
+	}
+}
+
 const struct efx_nic_type efx_hunt_a0_nic_type = {
 	.mem_map_size = efx_ef10_mem_map_size,
 	.probe = efx_ef10_probe,
@@ -3484,11 +3598,14 @@ const struct efx_nic_type efx_hunt_a0_nic_type = {
 	.mtd_sync = efx_mcdi_mtd_sync,
 #endif
 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
+	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
+	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
 
 	.revision = EFX_REV_HUNT_A0,
 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
+	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
 	.can_rx_scatter = true,
 	.always_rx_scatter = true,
 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
@@ -3497,4 +3614,6 @@ const struct efx_nic_type efx_hunt_a0_nic_type = {
 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
 	.mcdi_max_ver = 2,
 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
+	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
+			    1 << HWTSTAMP_FILTER_ALL,
 };
diff --git a/drivers/net/ethernet/sfc/efx.c b/drivers/net/ethernet/sfc/efx.c
index 893cd78c3a6c..b9929d9ed69d 100644
--- a/drivers/net/ethernet/sfc/efx.c
+++ b/drivers/net/ethernet/sfc/efx.c
@@ -2586,6 +2586,8 @@ static int efx_init_struct(struct efx_nic *efx,
 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
 	efx->rx_packet_hash_offset =
 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
+	efx->rx_packet_ts_offset =
+		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
 	spin_lock_init(&efx->stats_lock);
 	mutex_init(&efx->mac_lock);
 	efx->phy_op = &efx_dummy_phy_operations;
diff --git a/drivers/net/ethernet/sfc/mcdi.c b/drivers/net/ethernet/sfc/mcdi.c
index d608838f6729..540f57915d6f 100644
--- a/drivers/net/ethernet/sfc/mcdi.c
+++ b/drivers/net/ethernet/sfc/mcdi.c
@@ -1018,6 +1018,9 @@ void efx_mcdi_process_event(struct efx_channel *channel,
 	case MCDI_EVENT_CODE_PTP_PPS:
 		efx_ptp_event(efx, event);
 		break;
+	case MCDI_EVENT_CODE_PTP_TIME:
+		efx_time_sync_event(channel, event);
+		break;
 	case MCDI_EVENT_CODE_TX_FLUSH:
 	case MCDI_EVENT_CODE_RX_FLUSH:
 		/* Two flush events will be sent: one to the same event
diff --git a/drivers/net/ethernet/sfc/net_driver.h b/drivers/net/ethernet/sfc/net_driver.h
index 1a3e4972c68f..0d0e8eb88332 100644
--- a/drivers/net/ethernet/sfc/net_driver.h
+++ b/drivers/net/ethernet/sfc/net_driver.h
@@ -369,6 +369,13 @@ enum efx_rx_alloc_method {
 	RX_ALLOC_METHOD_PAGE = 2,
 };
 
+enum efx_sync_events_state {
+	SYNC_EVENTS_DISABLED = 0,
+	SYNC_EVENTS_QUIESCENT,
+	SYNC_EVENTS_REQUESTED,
+	SYNC_EVENTS_VALID,
+};
+
 /**
  * struct efx_channel - An Efx channel
  *
@@ -408,6 +415,9 @@ enum efx_rx_alloc_method {
  *	by __efx_rx_packet(), if @rx_pkt_n_frags != 0
  * @rx_queue: RX queue for this channel
  * @tx_queue: TX queues for this channel
+ * @sync_events_state: Current state of sync events on this channel
+ * @sync_timestamp_major: Major part of the last ptp sync event
+ * @sync_timestamp_minor: Minor part of the last ptp sync event
  */
 struct efx_channel {
 	struct efx_nic *efx;
@@ -446,6 +456,10 @@ struct efx_channel {
 
 	struct efx_rx_queue rx_queue;
 	struct efx_tx_queue tx_queue[EFX_TXQ_TYPES];
+
+	enum efx_sync_events_state sync_events_state;
+	u32 sync_timestamp_major;
+	u32 sync_timestamp_minor;
 };
 
 /**
@@ -686,6 +700,8 @@ struct vfdi_status;
  *	(valid only if @rx_prefix_size != 0; always negative)
  * @rx_packet_len_offset: Offset of RX packet length from start of packet data
  *	(valid only for NICs that set %EFX_RX_PKT_PREFIX_LEN; always negative)
+ * @rx_packet_ts_offset: Offset of timestamp from start of packet data
+ *	(valid only if channel->sync_timestamps_enabled; always negative)
  * @rx_hash_key: Toeplitz hash key for RSS
  * @rx_indir_table: Indirection table for RSS
  * @rx_scatter: Scatter mode enabled for receives
@@ -820,6 +836,7 @@ struct efx_nic {
 	unsigned int rx_prefix_size;
 	int rx_packet_hash_offset;
 	int rx_packet_len_offset;
+	int rx_packet_ts_offset;
 	u8 rx_hash_key[40];
 	u32 rx_indir_table[128];
 	bool rx_scatter;
@@ -1035,6 +1052,8 @@ struct efx_mtd_partition {
  *	also notifies the driver that a writer has finished using this
  *	partition.
  * @ptp_write_host_time: Send host time to MC as part of sync protocol
+ * @ptp_set_ts_sync_events: Enable or disable sync events for inline RX
+ *	timestamping, possibly only temporarily for the purposes of a reset.
  * @ptp_set_ts_config: Set hardware timestamp configuration.  The flags
  *	and tx_type will already have been validated but this operation
  *	must validate and update rx_filter.
@@ -1047,6 +1066,7 @@ struct efx_mtd_partition {
  * @max_dma_mask: Maximum possible DMA mask
  * @rx_prefix_size: Size of RX prefix before packet data
  * @rx_hash_offset: Offset of RX flow hash within prefix
+ * @rx_ts_offset: Offset of timestamp within prefix
  * @rx_buffer_padding: Size of padding at end of RX packet
  * @can_rx_scatter: NIC is able to scatter packets to multiple buffers
  * @always_rx_scatter: NIC will always scatter packets to multiple buffers
@@ -1158,6 +1178,7 @@ struct efx_nic_type {
 	int (*mtd_sync)(struct mtd_info *mtd);
 #endif
 	void (*ptp_write_host_time)(struct efx_nic *efx, u32 host_time);
+	int (*ptp_set_ts_sync_events)(struct efx_nic *efx, bool en, bool temp);
 	int (*ptp_set_ts_config)(struct efx_nic *efx,
 				 struct hwtstamp_config *init);
 
@@ -1170,6 +1191,7 @@ struct efx_nic_type {
 	u64 max_dma_mask;
 	unsigned int rx_prefix_size;
 	unsigned int rx_hash_offset;
+	unsigned int rx_ts_offset;
 	unsigned int rx_buffer_padding;
 	bool can_rx_scatter;
 	bool always_rx_scatter;
diff --git a/drivers/net/ethernet/sfc/nic.h b/drivers/net/ethernet/sfc/nic.h
index 999ef285662d..5d9e2dc121f7 100644
--- a/drivers/net/ethernet/sfc/nic.h
+++ b/drivers/net/ethernet/sfc/nic.h
@@ -566,6 +566,15 @@ int efx_ptp_change_mode(struct efx_nic *efx, bool enable_wanted,
 			unsigned int new_mode);
 int efx_ptp_tx(struct efx_nic *efx, struct sk_buff *skb);
 void efx_ptp_event(struct efx_nic *efx, efx_qword_t *ev);
+void efx_time_sync_event(struct efx_channel *channel, efx_qword_t *ev);
+void __efx_rx_skb_attach_timestamp(struct efx_channel *channel,
+				   struct sk_buff *skb);
+static inline void efx_rx_skb_attach_timestamp(struct efx_channel *channel,
+					       struct sk_buff *skb)
+{
+	if (channel->sync_events_state == SYNC_EVENTS_VALID)
+		__efx_rx_skb_attach_timestamp(channel, skb);
+}
 void efx_ptp_start_datapath(struct efx_nic *efx);
 void efx_ptp_stop_datapath(struct efx_nic *efx);
 
diff --git a/drivers/net/ethernet/sfc/ptp.c b/drivers/net/ethernet/sfc/ptp.c
index 1d1a6f7325da..843e98dfb1b2 100644
--- a/drivers/net/ethernet/sfc/ptp.c
+++ b/drivers/net/ethernet/sfc/ptp.c
@@ -216,6 +216,8 @@ struct efx_ptp_timeset {
  * struct efx_ptp_data - Precision Time Protocol (PTP) state
  * @efx: The NIC context
  * @channel: The PTP channel (Siena only)
+ * @rx_ts_inline: Flag for whether RX timestamps are inline (else they are
+ *	separate events)
  * @rxq: Receive queue (awaiting timestamps)
  * @txq: Transmit queue
  * @evt_list: List of MC receive events awaiting packets
@@ -258,6 +260,7 @@ struct efx_ptp_timeset {
 struct efx_ptp_data {
 	struct efx_nic *efx;
 	struct efx_channel *channel;
+	bool rx_ts_inline;
 	struct sk_buff_head rxq;
 	struct sk_buff_head txq;
 	struct list_head evt_list;
@@ -317,8 +320,8 @@ static void efx_ptp_ns_to_s_ns(s64 ns, u32 *nic_major, u32 *nic_minor)
 	*nic_minor = ts.tv_nsec;
 }
 
-static ktime_t efx_ptp_s_ns_to_ktime(u32 nic_major, u32 nic_minor,
-				     s32 correction)
+static ktime_t efx_ptp_s_ns_to_ktime_correction(u32 nic_major, u32 nic_minor,
+						s32 correction)
 {
 	ktime_t kt = ktime_set(nic_major, nic_minor);
 	if (correction >= 0)
@@ -359,11 +362,16 @@ static void efx_ptp_ns_to_s27(s64 ns, u32 *nic_major, u32 *nic_minor)
 	*nic_minor = min;
 }
 
-static ktime_t efx_ptp_s27_to_ktime(u32 nic_major, u32 nic_minor,
-				    s32 correction)
+static inline ktime_t efx_ptp_s27_to_ktime(u32 nic_major, u32 nic_minor)
 {
-	u32 ns;
+	u32 ns = (u32)(((u64)nic_minor * NSEC_PER_SEC +
+			(1ULL << (S27_TO_NS_SHIFT - 1))) >> S27_TO_NS_SHIFT);
+	return ktime_set(nic_major, ns);
+}
 
+static ktime_t efx_ptp_s27_to_ktime_correction(u32 nic_major, u32 nic_minor,
+					       s32 correction)
+{
 	/* Apply the correction and deal with carry */
 	nic_minor += correction;
 	if ((s32)nic_minor < 0) {
@@ -374,10 +382,7 @@ static ktime_t efx_ptp_s27_to_ktime(u32 nic_major, u32 nic_minor,
 		nic_major++;
 	}
 
-	ns = (u32)(((u64)nic_minor * NSEC_PER_SEC +
-		    (1ULL << (S27_TO_NS_SHIFT - 1))) >> S27_TO_NS_SHIFT);
-
-	return ktime_set(nic_major, ns);
+	return efx_ptp_s27_to_ktime(nic_major, nic_minor);
 }
 
 /* Get PTP attributes and set up time conversions */
@@ -407,10 +412,10 @@ static int efx_ptp_get_attributes(struct efx_nic *efx)
 
 	if (fmt == MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_27FRACTION) {
 		ptp->ns_to_nic_time = efx_ptp_ns_to_s27;
-		ptp->nic_to_kernel_time = efx_ptp_s27_to_ktime;
+		ptp->nic_to_kernel_time = efx_ptp_s27_to_ktime_correction;
 	} else if (fmt == MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_NANOSECONDS) {
 		ptp->ns_to_nic_time = efx_ptp_ns_to_s_ns;
-		ptp->nic_to_kernel_time = efx_ptp_s_ns_to_ktime;
+		ptp->nic_to_kernel_time = efx_ptp_s_ns_to_ktime_correction;
 	} else {
 		return -ERANGE;
 	}
@@ -806,6 +811,9 @@ static void efx_ptp_drop_time_expired_events(struct efx_nic *efx)
 	struct list_head *cursor;
 	struct list_head *next;
 
+	if (ptp->rx_ts_inline)
+		return;
+
 	/* Drop time-expired events */
 	spin_lock_bh(&ptp->evt_lock);
 	if (!list_empty(&ptp->evt_list)) {
@@ -839,6 +847,8 @@ static enum ptp_packet_state efx_ptp_match_rx(struct efx_nic *efx,
 	struct efx_ptp_match *match;
 	enum ptp_packet_state rc = PTP_PACKET_STATE_UNMATCHED;
 
+	WARN_ON_ONCE(ptp->rx_ts_inline);
+
 	spin_lock_bh(&ptp->evt_lock);
 	evts_waiting = !list_empty(&ptp->evt_list);
 	spin_unlock_bh(&ptp->evt_lock);
@@ -1061,8 +1071,6 @@ static void efx_ptp_pps_worker(struct work_struct *work)
 	ptp_clock_event(ptp->phc_clock, &ptp_evt);
 }
 
-/* Process any pending transmissions and timestamp any received packets.
- */
 static void efx_ptp_worker(struct work_struct *work)
 {
 	struct efx_ptp_data *ptp_data =
@@ -1120,6 +1128,7 @@ int efx_ptp_probe(struct efx_nic *efx, struct efx_channel *channel)
 
 	ptp->efx = efx;
 	ptp->channel = channel;
+	ptp->rx_ts_inline = efx_nic_rev(efx) >= EFX_REV_HUNT_A0;
 
 	rc = efx_nic_alloc_buffer(efx, &ptp->start, sizeof(int), GFP_KERNEL);
 	if (rc != 0)
@@ -1491,6 +1500,9 @@ static void ptp_event_rx(struct efx_nic *efx, struct efx_ptp_data *ptp)
 {
 	struct efx_ptp_event_rx *evt = NULL;
 
+	if (WARN_ON_ONCE(ptp->rx_ts_inline))
+		return;
+
 	if (ptp->evt_frag_idx != 3) {
 		ptp_event_failure(efx, 3);
 		return;
@@ -1587,6 +1599,93 @@ void efx_ptp_event(struct efx_nic *efx, efx_qword_t *ev)
 	}
 }
 
+void efx_time_sync_event(struct efx_channel *channel, efx_qword_t *ev)
+{
+	channel->sync_timestamp_major = MCDI_EVENT_FIELD(*ev, PTP_TIME_MAJOR);
+	channel->sync_timestamp_minor =
+		MCDI_EVENT_FIELD(*ev, PTP_TIME_MINOR_26_19) << 19;
+	/* if sync events have been disabled then we want to silently ignore
+	 * this event, so throw away result.
+	 */
+	(void) cmpxchg(&channel->sync_events_state, SYNC_EVENTS_REQUESTED,
+		       SYNC_EVENTS_VALID);
+}
+
+/* make some assumptions about the time representation rather than abstract it,
+ * since we currently only support one type of inline timestamping and only on
+ * EF10.
+ */
+#define MINOR_TICKS_PER_SECOND 0x8000000
+/* Fuzz factor for sync events to be out of order with RX events */
+#define FUZZ (MINOR_TICKS_PER_SECOND / 10)
+#define EXPECTED_SYNC_EVENTS_PER_SECOND 4
+
+static inline u32 efx_rx_buf_timestamp_minor(struct efx_nic *efx, const u8 *eh)
+{
+#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
+	return __le32_to_cpup((const __le32 *)(eh + efx->rx_packet_ts_offset));
+#else
+	const u8 *data = eh + efx->rx_packet_ts_offset;
+	return (u32)data[0]       |
+	       (u32)data[1] << 8  |
+	       (u32)data[2] << 16 |
+	       (u32)data[3] << 24;
+#endif
+}
+
+void __efx_rx_skb_attach_timestamp(struct efx_channel *channel,
+				   struct sk_buff *skb)
+{
+	struct efx_nic *efx = channel->efx;
+	u32 pkt_timestamp_major, pkt_timestamp_minor;
+	u32 diff, carry;
+	struct skb_shared_hwtstamps *timestamps;
+
+	pkt_timestamp_minor = (efx_rx_buf_timestamp_minor(efx,
+							  skb_mac_header(skb)) +
+			       (u32) efx->ptp_data->ts_corrections.rx) &
+			      (MINOR_TICKS_PER_SECOND - 1);
+
+	/* get the difference between the packet and sync timestamps,
+	 * modulo one second
+	 */
+	diff = (pkt_timestamp_minor - channel->sync_timestamp_minor) &
+		(MINOR_TICKS_PER_SECOND - 1);
+	/* do we roll over a second boundary and need to carry the one? */
+	carry = channel->sync_timestamp_minor + diff > MINOR_TICKS_PER_SECOND ?
+		1 : 0;
+
+	if (diff <= MINOR_TICKS_PER_SECOND / EXPECTED_SYNC_EVENTS_PER_SECOND +
+		    FUZZ) {
+		/* packet is ahead of the sync event by a quarter of a second or
+		 * less (allowing for fuzz)
+		 */
+		pkt_timestamp_major = channel->sync_timestamp_major + carry;
+	} else if (diff >= MINOR_TICKS_PER_SECOND - FUZZ) {
+		/* packet is behind the sync event but within the fuzz factor.
+		 * This means the RX packet and sync event crossed as they were
+		 * placed on the event queue, which can sometimes happen.
+		 */
+		pkt_timestamp_major = channel->sync_timestamp_major - 1 + carry;
+	} else {
+		/* it's outside tolerance in both directions. this might be
+		 * indicative of us missing sync events for some reason, so
+		 * we'll call it an error rather than risk giving a bogus
+		 * timestamp.
+		 */
+		netif_vdbg(efx, drv, efx->net_dev,
+			  "packet timestamp %x too far from sync event %x:%x\n",
+			  pkt_timestamp_minor, channel->sync_timestamp_major,
+			  channel->sync_timestamp_minor);
+		return;
+	}
+
+	/* attach the timestamps to the skb */
+	timestamps = skb_hwtstamps(skb);
+	timestamps->hwtstamp =
+		efx_ptp_s27_to_ktime(pkt_timestamp_major, pkt_timestamp_minor);
+}
+
 static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta)
 {
 	struct efx_ptp_data *ptp_data = container_of(ptp,
@@ -1728,9 +1827,15 @@ void efx_ptp_start_datapath(struct efx_nic *efx)
 {
 	if (efx_ptp_restart(efx))
 		netif_err(efx, drv, efx->net_dev, "Failed to restart PTP.\n");
+	/* re-enable timestamping if it was previously enabled */
+	if (efx->type->ptp_set_ts_sync_events)
+		efx->type->ptp_set_ts_sync_events(efx, true, true);
 }
 
 void efx_ptp_stop_datapath(struct efx_nic *efx)
 {
+	/* temporarily disable timestamping */
+	if (efx->type->ptp_set_ts_sync_events)
+		efx->type->ptp_set_ts_sync_events(efx, false, true);
 	efx_ptp_stop(efx);
 }
diff --git a/drivers/net/ethernet/sfc/rx.c b/drivers/net/ethernet/sfc/rx.c
index d12abc543975..1fde9b8ac456 100644
--- a/drivers/net/ethernet/sfc/rx.c
+++ b/drivers/net/ethernet/sfc/rx.c
@@ -624,6 +624,8 @@ static void efx_rx_deliver(struct efx_channel *channel, u8 *eh,
 	if (likely(rx_buf->flags & EFX_RX_PKT_CSUMMED))
 		skb->ip_summed = CHECKSUM_UNNECESSARY;
 
+	efx_rx_skb_attach_timestamp(channel, skb);
+
 	if (channel->type->receive_skb)
 		if (channel->type->receive_skb(channel, skb))
 			return;