Merge branch 'sfc-support-PTP-on-8000-and-X2000-series-NICs'

Edward Cree says:

====================
sfc: support PTP on 8000 and X2000 series NICs

Starting from the 8000-series (Medford 1), SFC NICs can timestamp TX packets
 sent through an ordinary DMA queue, rather than a special control-plane
 operation as in the 7000-series.  Patches 2-8 implement support for this.
The X2000-series (Medford 2) changes the format of timestamps, from seconds+
 (2^27)ths to seconds + quarter nanoseconds, as well as changing the shift
 of the frequency adjustment for increased precision.  Patches 9-12
 implement support for these changes.
Patch #1 is an unrelated fix for NAPI budget handling, needed in order for
 TX completion changes in the later patches to apply cleanly.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/sfc/ef10.c

@@ -322,6 +322,25 @@ static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
 	return 0;
 }
 
+static void efx_ef10_read_licensed_features(struct efx_nic *efx)
+{
+	MCDI_DECLARE_BUF(inbuf, MC_CMD_LICENSING_V3_IN_LEN);
+	MCDI_DECLARE_BUF(outbuf, MC_CMD_LICENSING_V3_OUT_LEN);
+	struct efx_ef10_nic_data *nic_data = efx->nic_data;
+	size_t outlen;
+	int rc;
+
+	MCDI_SET_DWORD(inbuf, LICENSING_V3_IN_OP,
+		       MC_CMD_LICENSING_V3_IN_OP_REPORT_LICENSE);
+	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_LICENSING_V3, inbuf, sizeof(inbuf),
+				outbuf, sizeof(outbuf), &outlen);
+	if (rc || (outlen < MC_CMD_LICENSING_V3_OUT_LEN))
+		return;
+
+	nic_data->licensed_features = MCDI_QWORD(outbuf,
+					 LICENSING_V3_OUT_LICENSED_FEATURES);
+}
+
 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
 {
 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
@@ -722,6 +741,8 @@ static int efx_ef10_probe(struct efx_nic *efx)
 	if (rc < 0)
 		goto fail5;
 
+	efx_ef10_read_licensed_features(efx);
+
 	/* We can have one VI for each vi_stride-byte region.
 	 * However, until we use TX option descriptors we need two TX queues
 	 * per channel.
@@ -760,14 +781,7 @@ static int efx_ef10_probe(struct efx_nic *efx)
 	if (rc && rc != -EPERM)
 		goto fail5;
 
-	rc = efx_ptp_probe(efx, NULL);
-	/* Failure to probe PTP is not fatal.
-	 * In the case of EPERM, efx_ptp_probe will print its own message (in
-	 * efx_ptp_get_attributes()), so we don't need to.
-	 */
-	if (rc && rc != -EPERM)
-		netif_warn(efx, drv, efx->net_dev,
-			   "Failed to probe PTP, rc=%d\n", rc);
+	efx_ptp_defer_probe_with_channel(efx);
 
 #ifdef CONFIG_SFC_SRIOV
 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
@@ -937,6 +951,11 @@ static int efx_ef10_link_piobufs(struct efx_nic *efx)
 
 	/* Link a buffer to each TX queue */
 	efx_for_each_channel(channel, efx) {
+		/* Extra channels, even those with TXQs (PTP), do not require
+		 * PIO resources.
+		 */
+		if (!channel->type->want_pio)
+			continue;
 		efx_for_each_channel_tx_queue(tx_queue, channel) {
 			/* We assign the PIO buffers to queues in
 			 * reverse order to allow for the following
@@ -1284,7 +1303,9 @@ static int efx_ef10_dimension_resources(struct efx_nic *efx)
 	void __iomem *membase;
 	int rc;
 
-	channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
+	channel_vis = max(efx->n_channels,
+			  (efx->n_tx_channels + efx->n_extra_tx_channels) *
+			  EFX_TXQ_TYPES);
 
 #ifdef EFX_USE_PIO
 	/* Try to allocate PIO buffers if wanted and if the full
@@ -2408,12 +2429,25 @@ static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
 	int i;
 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
 
+	/* Only attempt to enable TX timestamping if we have the license for it,
+	 * otherwise TXQ init will fail
+	 */
+	if (!(nic_data->licensed_features &
+	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
+		tx_queue->timestamping = false;
+		/* Disable sync events on this channel. */
+		if (efx->type->ptp_set_ts_sync_events)
+			efx->type->ptp_set_ts_sync_events(efx, false, false);
+	}
+
 	/* TSOv2 is a limited resource that can only be configured on a limited
 	 * number of queues. TSO without checksum offload is not really a thing,
 	 * so we only enable it for those queues.
+	 * TSOv2 cannot be used with Hardware timestamping.
 	 */
 	if (csum_offload && (nic_data->datapath_caps2 &
-			(1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))) {
+			(1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN)) &&
+	    !tx_queue->timestamping) {
 		tso_v2 = true;
 		netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
 				channel->channel);
@@ -2439,14 +2473,16 @@ static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
 
 	do {
-		MCDI_POPULATE_DWORD_3(inbuf, INIT_TXQ_IN_FLAGS,
+		MCDI_POPULATE_DWORD_4(inbuf, INIT_TXQ_IN_FLAGS,
 				/* This flag was removed from mcdi_pcol.h for
 				 * the non-_EXT version of INIT_TXQ.  However,
 				 * firmware still honours it.
 				 */
 				INIT_TXQ_EXT_IN_FLAG_TSOV2_EN, tso_v2,
 				INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
-				INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
+				INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload,
+				INIT_TXQ_EXT_IN_FLAG_TIMESTAMP,
+						tx_queue->timestamping);
 
 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
 					NULL, 0, NULL);
@@ -2472,12 +2508,13 @@ static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
 	tx_queue->insert_count = 1;
 	txd = efx_tx_desc(tx_queue, 0);
-	EFX_POPULATE_QWORD_4(*txd,
+	EFX_POPULATE_QWORD_5(*txd,
 			     ESF_DZ_TX_DESC_IS_OPT, true,
 			     ESF_DZ_TX_OPTION_TYPE,
 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
-			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
+			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload,
+			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
 	tx_queue->write_count = 1;
 
 	if (tso_v2) {
@@ -3572,31 +3609,92 @@ static int efx_ef10_handle_rx_event(struct efx_channel *channel,
 	return n_packets;
 }
 
-static int
+static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
+{
+	u32 tstamp;
+
+	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
+	tstamp <<= 16;
+	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
+
+	return tstamp;
+}
+
+static void
 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
 {
 	struct efx_nic *efx = channel->efx;
 	struct efx_tx_queue *tx_queue;
 	unsigned int tx_ev_desc_ptr;
 	unsigned int tx_ev_q_label;
-	int tx_descs = 0;
+	unsigned int tx_ev_type;
+	u64 ts_part;
 
 	if (unlikely(READ_ONCE(efx->reset_pending)))
-		return 0;
+		return;
 
 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
-		return 0;
+		return;
 
-	/* Transmit completion */
-	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
+	/* Get the transmit queue */
 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
 	tx_queue = efx_channel_get_tx_queue(channel,
 					    tx_ev_q_label % EFX_TXQ_TYPES);
-	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
-		    tx_queue->ptr_mask);
-	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
 
-	return tx_descs;
+	if (!tx_queue->timestamping) {
+		/* Transmit completion */
+		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
+		efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
+		return;
+	}
+
+	/* Transmit timestamps are only available for 8XXX series. They result
+	 * in three events per packet. These occur in order, and are:
+	 *  - the normal completion event
+	 *  - the low part of the timestamp
+	 *  - the high part of the timestamp
+	 *
+	 * Each part of the timestamp is itself split across two 16 bit
+	 * fields in the event.
+	 */
+	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
+
+	switch (tx_ev_type) {
+	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
+		/* In case of Queue flush or FLR, we might have received
+		 * the previous TX completion event but not the Timestamp
+		 * events.
+		 */
+		if (tx_queue->completed_desc_ptr != tx_queue->ptr_mask)
+			efx_xmit_done(tx_queue, tx_queue->completed_desc_ptr);
+
+		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event,
+						 ESF_DZ_TX_DESCR_INDX);
+		tx_queue->completed_desc_ptr =
+					tx_ev_desc_ptr & tx_queue->ptr_mask;
+		break;
+
+	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
+		ts_part = efx_ef10_extract_event_ts(event);
+		tx_queue->completed_timestamp_minor = ts_part;
+		break;
+
+	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
+		ts_part = efx_ef10_extract_event_ts(event);
+		tx_queue->completed_timestamp_major = ts_part;
+
+		efx_xmit_done(tx_queue, tx_queue->completed_desc_ptr);
+		tx_queue->completed_desc_ptr = tx_queue->ptr_mask;
+		break;
+
+	default:
+		netif_err(efx, hw, efx->net_dev,
+			  "channel %d unknown tx event type %d (data "
+			  EFX_QWORD_FMT ")\n",
+			  channel->channel, tx_ev_type,
+			  EFX_QWORD_VAL(*event));
+		break;
+	}
 }
 
 static void
@@ -3658,7 +3756,6 @@ static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
 	efx_qword_t event, *p_event;
 	unsigned int read_ptr;
 	int ev_code;
-	int tx_descs = 0;
 	int spent = 0;
 
 	if (quota <= 0)
@@ -3698,13 +3795,7 @@ static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
 			}
 			break;
 		case ESE_DZ_EV_CODE_TX_EV:
-			tx_descs += efx_ef10_handle_tx_event(channel, &event);
-			if (tx_descs > efx->txq_entries) {
-				spent = quota;
-				goto out;
-			} else if (++spent == quota) {
-				goto out;
-			}
+			efx_ef10_handle_tx_event(channel, &event);
 			break;
 		case ESE_DZ_EV_CODE_DRIVER_EV:
 			efx_ef10_handle_driver_event(channel, &event);
@@ -6179,7 +6270,8 @@ static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
 	      efx_ef10_rx_enable_timestamping :
 	      efx_ef10_rx_disable_timestamping;
 
-	efx_for_each_channel(channel, efx) {
+	channel = efx_ptp_channel(efx);
+	if (channel) {
 		int rc = set(channel, temp);
 		if (en && rc != 0) {
 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);

drivers/net/ethernet/sfc/efx.c

@@ -896,12 +896,20 @@ void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
 	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
 }
 
+bool efx_default_channel_want_txqs(struct efx_channel *channel)
+{
+	return channel->channel - channel->efx->tx_channel_offset <
+		channel->efx->n_tx_channels;
+}
+
 static const struct efx_channel_type efx_default_channel_type = {
 	.pre_probe		= efx_channel_dummy_op_int,
 	.post_remove		= efx_channel_dummy_op_void,
 	.get_name		= efx_get_channel_name,
 	.copy			= efx_copy_channel,
+	.want_txqs		= efx_default_channel_want_txqs,
 	.keep_eventq		= false,
+	.want_pio		= true,
 };
 
 int efx_channel_dummy_op_int(struct efx_channel *channel)
@@ -1501,6 +1509,7 @@ static int efx_probe_interrupts(struct efx_nic *efx)
 	}
 
 	/* Assign extra channels if possible */
+	efx->n_extra_tx_channels = 0;
 	j = efx->n_channels;
 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
 		if (!efx->extra_channel_type[i])
@@ -1512,6 +1521,8 @@ static int efx_probe_interrupts(struct efx_nic *efx)
 			--j;
 			efx_get_channel(efx, j)->type =
 				efx->extra_channel_type[i];
+			if (efx_channel_has_tx_queues(efx_get_channel(efx, j)))
+				efx->n_extra_tx_channels++;
 		}
 	}
 

drivers/net/ethernet/sfc/farch.c

@@ -818,17 +818,16 @@ static void efx_farch_magic_event(struct efx_channel *channel, u32 magic)
  * The NIC batches TX completion events; the message we receive is of
  * the form "complete all TX events up to this index".
  */
-static int
+static void
 efx_farch_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
 {
 	unsigned int tx_ev_desc_ptr;
 	unsigned int tx_ev_q_label;
 	struct efx_tx_queue *tx_queue;
 	struct efx_nic *efx = channel->efx;
-	int tx_packets = 0;
 
 	if (unlikely(READ_ONCE(efx->reset_pending)))
-		return 0;
+		return;
 
 	if (likely(EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_COMP))) {
 		/* Transmit completion */
@@ -836,8 +835,6 @@ efx_farch_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
 		tx_ev_q_label = EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_Q_LABEL);
 		tx_queue = efx_channel_get_tx_queue(
 			channel, tx_ev_q_label % EFX_TXQ_TYPES);
-		tx_packets = ((tx_ev_desc_ptr - tx_queue->read_count) &
-			      tx_queue->ptr_mask);
 		efx_xmit_done(tx_queue, tx_ev_desc_ptr);
 	} else if (EFX_QWORD_FIELD(*event, FSF_AZ_TX_EV_WQ_FF_FULL)) {
 		/* Rewrite the FIFO write pointer */
@@ -856,8 +853,6 @@ efx_farch_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
 			  EFX_QWORD_FMT"\n", channel->channel,
 			  EFX_QWORD_VAL(*event));
 	}
-
-	return tx_packets;
 }
 
 /* Detect errors included in the rx_evt_pkt_ok bit. */
@@ -1090,7 +1085,7 @@ efx_farch_handle_tx_flush_done(struct efx_nic *efx, efx_qword_t *event)
 	int qid;
 
 	qid = EFX_QWORD_FIELD(*event, FSF_AZ_DRIVER_EV_SUBDATA);
-	if (qid < EFX_TXQ_TYPES * efx->n_tx_channels) {
+	if (qid < EFX_TXQ_TYPES * (efx->n_tx_channels + efx->n_extra_tx_channels)) {
 		tx_queue = efx_get_tx_queue(efx, qid / EFX_TXQ_TYPES,
 					    qid % EFX_TXQ_TYPES);
 		if (atomic_cmpxchg(&tx_queue->flush_outstanding, 1, 0)) {
@@ -1270,7 +1265,6 @@ int efx_farch_ev_process(struct efx_channel *channel, int budget)
 	unsigned int read_ptr;
 	efx_qword_t event, *p_event;
 	int ev_code;
-	int tx_packets = 0;
 	int spent = 0;
 
 	if (budget <= 0)
@@ -1304,12 +1298,7 @@ int efx_farch_ev_process(struct efx_channel *channel, int budget)
 				goto out;
 			break;
 		case FSE_AZ_EV_CODE_TX_EV:
-			tx_packets += efx_farch_handle_tx_event(channel,
-								&event);
-			if (tx_packets > efx->txq_entries) {
-				spent = budget;
-				goto out;
-			}
+			efx_farch_handle_tx_event(channel, &event);
 			break;
 		case FSE_AZ_EV_CODE_DRV_GEN_EV:
 			efx_farch_handle_generated_event(channel, &event);
@@ -1680,20 +1669,21 @@ void efx_farch_rx_pull_indir_table(struct efx_nic *efx)
  */
 void efx_farch_dimension_resources(struct efx_nic *efx, unsigned sram_lim_qw)
 {
-	unsigned vi_count, buftbl_min;
+	unsigned vi_count, buftbl_min, total_tx_channels;
 
 #ifdef CONFIG_SFC_SRIOV
 	struct siena_nic_data *nic_data = efx->nic_data;
 #endif
 
+	total_tx_channels = efx->n_tx_channels + efx->n_extra_tx_channels;
 	/* Account for the buffer table entries backing the datapath channels
 	 * and the descriptor caches for those channels.
 	 */
 	buftbl_min = ((efx->n_rx_channels * EFX_MAX_DMAQ_SIZE +
-		       efx->n_tx_channels * EFX_TXQ_TYPES * EFX_MAX_DMAQ_SIZE +
+		       total_tx_channels * EFX_TXQ_TYPES * EFX_MAX_DMAQ_SIZE +
 		       efx->n_channels * EFX_MAX_EVQ_SIZE)
 		      * sizeof(efx_qword_t) / EFX_BUF_SIZE);
-	vi_count = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
+	vi_count = max(efx->n_channels, total_tx_channels * EFX_TXQ_TYPES);
 
 #ifdef CONFIG_SFC_SRIOV
 	if (efx->type->sriov_wanted) {

drivers/net/ethernet/sfc/net_driver.h

@@ -191,6 +191,7 @@ struct efx_tx_buffer {
  *	Size of the region is efx_piobuf_size.
  * @piobuf_offset: Buffer offset to be specified in PIO descriptors
  * @initialised: Has hardware queue been initialised?
+ * @timestamping: Is timestamping enabled for this channel?
  * @handle_tso: TSO xmit preparation handler.  Sets up the TSO metadata and
  *	may also map tx data, depending on the nature of the TSO implementation.
  * @read_count: Current read pointer.
@@ -202,6 +203,10 @@ struct efx_tx_buffer {
  *	avoid cache-line ping-pong between the xmit path and the
  *	completion path.
  * @merge_events: Number of TX merged completion events
+ * @completed_desc_ptr: Most recent completed pointer - only used with
+ *      timestamping.
+ * @completed_timestamp_major: Top part of the most recent tx timestamp.
+ * @completed_timestamp_minor: Low part of the most recent tx timestamp.
  * @insert_count: Current insert pointer
  *	This is the number of buffers that have been added to the
  *	software ring.
@@ -247,6 +252,7 @@ struct efx_tx_queue {
 	void __iomem *piobuf;
 	unsigned int piobuf_offset;
 	bool initialised;
+	bool timestamping;
 
 	/* Function pointers used in the fast path. */
 	int (*handle_tso)(struct efx_tx_queue*, struct sk_buff*, bool *);
@@ -257,6 +263,9 @@ struct efx_tx_queue {
 	unsigned int merge_events;
 	unsigned int bytes_compl;
 	unsigned int pkts_compl;
+	unsigned int completed_desc_ptr;
+	u32 completed_timestamp_major;
+	u32 completed_timestamp_minor;
 
 	/* Members used only on the xmit path */
 	unsigned int insert_count ____cacheline_aligned_in_smp;
@@ -522,8 +531,12 @@ struct efx_msi_context {
  * @copy: Copy the channel state prior to reallocation.  May be %NULL if
  *	reallocation is not supported.
  * @receive_skb: Handle an skb ready to be passed to netif_receive_skb()
+ * @want_txqs: Determine whether this channel should have TX queues
+ *	created.  If %NULL, TX queues are not created.
  * @keep_eventq: Flag for whether event queue should be kept initialised
  *	while the device is stopped
+ * @want_pio: Flag for whether PIO buffers should be linked to this
+ *	channel's TX queues.
  */
 struct efx_channel_type {
 	void (*handle_no_channel)(struct efx_nic *);
@@ -532,7 +545,9 @@ struct efx_channel_type {
 	void (*get_name)(struct efx_channel *, char *buf, size_t len);
 	struct efx_channel *(*copy)(const struct efx_channel *);
 	bool (*receive_skb)(struct efx_channel *, struct sk_buff *);
+	bool (*want_txqs)(struct efx_channel *);
 	bool keep_eventq;
+	bool want_pio;
 };
 
 enum efx_led_mode {
@@ -735,6 +750,7 @@ struct vfdi_status;
  * @n_channels: Number of channels in use
  * @n_rx_channels: Number of channels used for RX (= number of RX queues)
  * @n_tx_channels: Number of channels used for TX
+ * @n_extra_tx_channels: Number of extra channels with TX queues
  * @rx_ip_align: RX DMA address offset to have IP header aligned in
  *	in accordance with NET_IP_ALIGN
  * @rx_dma_len: Current maximum RX DMA length
@@ -881,6 +897,7 @@ struct efx_nic {
 	unsigned rss_spread;
 	unsigned tx_channel_offset;
 	unsigned n_tx_channels;
+	unsigned n_extra_tx_channels;
 	unsigned int rx_ip_align;
 	unsigned int rx_dma_len;
 	unsigned int rx_buffer_order;
@@ -1363,8 +1380,8 @@ efx_get_tx_queue(struct efx_nic *efx, unsigned index, unsigned type)
 
 static inline bool efx_channel_has_tx_queues(struct efx_channel *channel)
 {
-	return channel->channel - channel->efx->tx_channel_offset <
-		channel->efx->n_tx_channels;
+	return channel->type && channel->type->want_txqs &&
+				channel->type->want_txqs(channel);
 }
 
 static inline struct efx_tx_queue *

drivers/net/ethernet/sfc/nic.h

@@ -440,6 +440,7 @@ struct efx_ef10_nic_data {
 	struct efx_udp_tunnel udp_tunnels[16];
 	bool udp_tunnels_dirty;
 	struct mutex udp_tunnels_lock;
+	u64 licensed_features;
 };
 
 int efx_init_sriov(void);
@@ -448,6 +449,7 @@ void efx_fini_sriov(void);
 struct ethtool_ts_info;
 int efx_ptp_probe(struct efx_nic *efx, struct efx_channel *channel);
 void efx_ptp_defer_probe_with_channel(struct efx_nic *efx);
+struct efx_channel *efx_ptp_channel(struct efx_nic *efx);
 void efx_ptp_remove(struct efx_nic *efx);
 int efx_ptp_set_ts_config(struct efx_nic *efx, struct ifreq *ifr);
 int efx_ptp_get_ts_config(struct efx_nic *efx, struct ifreq *ifr);
@@ -471,6 +473,8 @@ static inline void efx_rx_skb_attach_timestamp(struct efx_channel *channel,
 }
 void efx_ptp_start_datapath(struct efx_nic *efx);
 void efx_ptp_stop_datapath(struct efx_nic *efx);
+bool efx_ptp_use_mac_tx_timestamps(struct efx_nic *efx);
+ktime_t efx_ptp_nic_to_kernel_time(struct efx_tx_queue *tx_queue);
 
 extern const struct efx_nic_type falcon_a1_nic_type;
 extern const struct efx_nic_type falcon_b0_nic_type;

drivers/net/ethernet/sfc/ptp.c

@@ -149,18 +149,14 @@ enum ptp_packet_state {
 /* Maximum parts-per-billion adjustment that is acceptable */
 #define MAX_PPB			1000000
 
-/* Number of bits required to hold the above */
-#define	MAX_PPB_BITS		20
-
-/* Number of extra bits allowed when calculating fractional ns.
- * EXTRA_BITS + MC_CMD_PTP_IN_ADJUST_BITS + MAX_PPB_BITS should
- * be less than 63.
- */
-#define	PPB_EXTRA_BITS		2
-
 /* Precalculate scale word to avoid long long division at runtime */
-#define	PPB_SCALE_WORD	((1LL << (PPB_EXTRA_BITS + MC_CMD_PTP_IN_ADJUST_BITS +\
-			MAX_PPB_BITS)) / 1000000000LL)
+/* This is equivalent to 2^66 / 10^9. */
+#define PPB_SCALE_WORD  ((1LL << (57)) / 1953125LL)
+
+/* How much to shift down after scaling to convert to FP40 */
+#define PPB_SHIFT_FP40		26
+/* ... and FP44. */
+#define PPB_SHIFT_FP44		22
 
 #define PTP_SYNC_ATTEMPTS	4
 
@@ -218,8 +214,8 @@ struct efx_ptp_timeset {
  * @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
+ * @rxq: Receive SKB queue (awaiting timestamps)
+ * @txq: Transmit SKB queue
  * @evt_list: List of MC receive events awaiting packets
  * @evt_free_list: List of free events
  * @evt_lock: Lock for manipulating evt_list and evt_free_list
@@ -233,19 +229,36 @@ struct efx_ptp_timeset {
  * @config: Current timestamp configuration
  * @enabled: PTP operation enabled
  * @mode: Mode in which PTP operating (PTP version)
- * @time_format: Time format supported by this NIC
  * @ns_to_nic_time: Function to convert from scalar nanoseconds to NIC time
  * @nic_to_kernel_time: Function to convert from NIC to kernel time
+ * @nic_time.minor_max: Wrap point for NIC minor times
+ * @nic_time.sync_event_diff_min: Minimum acceptable difference between time
+ * in packet prefix and last MCDI time sync event i.e. how much earlier than
+ * the last sync event time a packet timestamp can be.
+ * @nic_time.sync_event_diff_max: Maximum acceptable difference between time
+ * in packet prefix and last MCDI time sync event i.e. how much later than
+ * the last sync event time a packet timestamp can be.
+ * @nic_time.sync_event_minor_shift: Shift required to make minor time from
+ * field in MCDI time sync event.
  * @min_synchronisation_ns: Minimum acceptable corrected sync window
- * @ts_corrections.tx: Required driver correction of transmit timestamps
- * @ts_corrections.rx: Required driver correction of receive timestamps
+ * @capabilities: Capabilities flags from the NIC
+ * @ts_corrections.ptp_tx: Required driver correction of PTP packet transmit
+ *                         timestamps
+ * @ts_corrections.ptp_rx: Required driver correction of PTP packet receive
+ *                         timestamps
  * @ts_corrections.pps_out: PPS output error (information only)
  * @ts_corrections.pps_in: Required driver correction of PPS input timestamps
+ * @ts_corrections.general_tx: Required driver correction of general packet
+ *                             transmit timestamps
+ * @ts_corrections.general_rx: Required driver correction of general packet
+ *                             receive timestamps
  * @evt_frags: Partly assembled PTP events
  * @evt_frag_idx: Current fragment number
  * @evt_code: Last event code
  * @start: Address at which MC indicates ready for synchronisation
  * @host_time_pps: Host time at last PPS
+ * @adjfreq_ppb_shift: Shift required to convert scaled parts-per-billion
+ * frequency adjustment into a fixed point fractional nanosecond format.
  * @current_adjfreq: Current ppb adjustment.
  * @phc_clock: Pointer to registered phc device (if primary function)
  * @phc_clock_info: Registration structure for phc device
@@ -264,6 +277,7 @@ struct efx_ptp_timeset {
  * @oversize_sync_windows: Number of corrected sync windows that are too large
  * @rx_no_timestamp: Number of packets received without a timestamp.
  * @timeset: Last set of synchronisation statistics.
+ * @xmit_skb: Transmit SKB function.
  */
 struct efx_ptp_data {
 	struct efx_nic *efx;
@@ -284,22 +298,31 @@ struct efx_ptp_data {
 	struct hwtstamp_config config;
 	bool enabled;
 	unsigned int mode;
-	unsigned int time_format;
 	void (*ns_to_nic_time)(s64 ns, u32 *nic_major, u32 *nic_minor);
 	ktime_t (*nic_to_kernel_time)(u32 nic_major, u32 nic_minor,
 				      s32 correction);
+	struct {
+		u32 minor_max;
+		u32 sync_event_diff_min;
+		u32 sync_event_diff_max;
+		unsigned int sync_event_minor_shift;
+	} nic_time;
 	unsigned int min_synchronisation_ns;
+	unsigned int capabilities;
 	struct {
-		s32 tx;
-		s32 rx;
+		s32 ptp_tx;
+		s32 ptp_rx;
 		s32 pps_out;
 		s32 pps_in;
+		s32 general_tx;
+		s32 general_rx;
 	} ts_corrections;
 	efx_qword_t evt_frags[MAX_EVENT_FRAGS];
 	int evt_frag_idx;
 	int evt_code;
 	struct efx_buffer start;
 	struct pps_event_time host_time_pps;
+	unsigned int adjfreq_ppb_shift;
 	s64 current_adjfreq;
 	struct ptp_clock *phc_clock;
 	struct ptp_clock_info phc_clock_info;
@@ -319,6 +342,7 @@ struct efx_ptp_data {
 	unsigned int rx_no_timestamp;
 	struct efx_ptp_timeset
 	timeset[MC_CMD_PTP_OUT_SYNCHRONIZE_TIMESET_MAXNUM];
+	void (*xmit_skb)(struct efx_nic *efx, struct sk_buff *skb);
 };
 
 static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta);
@@ -329,6 +353,24 @@ static int efx_phc_settime(struct ptp_clock_info *ptp,
 static int efx_phc_enable(struct ptp_clock_info *ptp,
 			  struct ptp_clock_request *request, int on);
 
+bool efx_ptp_use_mac_tx_timestamps(struct efx_nic *efx)
+{
+	struct efx_ef10_nic_data *nic_data = efx->nic_data;
+
+	return ((efx_nic_rev(efx) >= EFX_REV_HUNT_A0) &&
+		(nic_data->datapath_caps2 &
+		 (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_MAC_TIMESTAMPING_LBN)
+		));
+}
+
+/* PTP 'extra' channel is still a traffic channel, but we only create TX queues
+ * if PTP uses MAC TX timestamps, not if PTP uses the MC directly to transmit.
+ */
+bool efx_ptp_want_txqs(struct efx_channel *channel)
+{
+	return efx_ptp_use_mac_tx_timestamps(channel->efx);
+}
+
 #define PTP_SW_STAT(ext_name, field_name)				\
 	{ #ext_name, 0, offsetof(struct efx_ptp_data, field_name) }
 #define PTP_MC_STAT(ext_name, mcdi_name)				\
@@ -471,6 +513,89 @@ static ktime_t efx_ptp_s27_to_ktime_correction(u32 nic_major, u32 nic_minor,
 	return efx_ptp_s27_to_ktime(nic_major, nic_minor);
 }
 
+/* For Medford2 platforms the time is in seconds and quarter nanoseconds. */
+static void efx_ptp_ns_to_s_qns(s64 ns, u32 *nic_major, u32 *nic_minor)
+{
+	struct timespec64 ts = ns_to_timespec64(ns);
+
+	*nic_major = (u32)ts.tv_sec;
+	*nic_minor = ts.tv_nsec * 4;
+}
+
+static ktime_t efx_ptp_s_qns_to_ktime_correction(u32 nic_major, u32 nic_minor,
+						 s32 correction)
+{
+	ktime_t kt;
+
+	nic_minor = DIV_ROUND_CLOSEST(nic_minor, 4);
+	correction = DIV_ROUND_CLOSEST(correction, 4);
+
+	kt = ktime_set(nic_major, nic_minor);
+
+	if (correction >= 0)
+		kt = ktime_add_ns(kt, (u64)correction);
+	else
+		kt = ktime_sub_ns(kt, (u64)-correction);
+	return kt;
+}
+
+struct efx_channel *efx_ptp_channel(struct efx_nic *efx)
+{
+	return efx->ptp_data ? efx->ptp_data->channel : NULL;
+}
+
+static u32 last_sync_timestamp_major(struct efx_nic *efx)
+{
+	struct efx_channel *channel = efx_ptp_channel(efx);
+	u32 major = 0;
+
+	if (channel)
+		major = channel->sync_timestamp_major;
+	return major;
+}
+
+/* The 8000 series and later can provide the time from the MAC, which is only
+ * 48 bits long and provides meta-information in the top 2 bits.
+ */
+static ktime_t
+efx_ptp_mac_nic_to_ktime_correction(struct efx_nic *efx,
+				    struct efx_ptp_data *ptp,
+				    u32 nic_major, u32 nic_minor,
+				    s32 correction)
+{
+	ktime_t kt = { 0 };
+
+	if (!(nic_major & 0x80000000)) {
+		WARN_ON_ONCE(nic_major >> 16);
+		/* Use the top bits from the latest sync event. */
+		nic_major &= 0xffff;
+		nic_major |= (last_sync_timestamp_major(efx) & 0xffff0000);
+
+		kt = ptp->nic_to_kernel_time(nic_major, nic_minor,
+					     correction);
+	}
+	return kt;
+}
+
+ktime_t efx_ptp_nic_to_kernel_time(struct efx_tx_queue *tx_queue)
+{
+	struct efx_nic *efx = tx_queue->efx;
+	struct efx_ptp_data *ptp = efx->ptp_data;
+	ktime_t kt;
+
+	if (efx_ptp_use_mac_tx_timestamps(efx))
+		kt = efx_ptp_mac_nic_to_ktime_correction(efx, ptp,
+				tx_queue->completed_timestamp_major,
+				tx_queue->completed_timestamp_minor,
+				ptp->ts_corrections.general_tx);
+	else
+		kt = ptp->nic_to_kernel_time(
+				tx_queue->completed_timestamp_major,
+				tx_queue->completed_timestamp_minor,
+				ptp->ts_corrections.general_tx);
+	return kt;
+}
+
 /* Get PTP attributes and set up time conversions */
 static int efx_ptp_get_attributes(struct efx_nic *efx)
 {
@@ -502,31 +627,71 @@ static int efx_ptp_get_attributes(struct efx_nic *efx)
 		return rc;
 	}
 
-	if (fmt == MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_27FRACTION) {
+	switch (fmt) {
+	case 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_correction;
-	} else if (fmt == MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_NANOSECONDS) {
+		ptp->nic_time.minor_max = 1 << 27;
+		ptp->nic_time.sync_event_minor_shift = 19;
+		break;
+	case 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_correction;
-	} else {
+		ptp->nic_time.minor_max = 1000000000;
+		ptp->nic_time.sync_event_minor_shift = 22;
+		break;
+	case MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_QTR_NANOSECONDS:
+		ptp->ns_to_nic_time = efx_ptp_ns_to_s_qns;
+		ptp->nic_to_kernel_time = efx_ptp_s_qns_to_ktime_correction;
+		ptp->nic_time.minor_max = 4000000000;
+		ptp->nic_time.sync_event_minor_shift = 24;
+		break;
+	default:
 		return -ERANGE;
 	}
 
-	ptp->time_format = fmt;
-
-	/* MC_CMD_PTP_OP_GET_ATTRIBUTES is an extended version of an older
-	 * operation MC_CMD_PTP_OP_GET_TIME_FORMAT that also returns a value
-	 * to use for the minimum acceptable corrected synchronization window.
+	/* Precalculate acceptable difference between the minor time in the
+	 * packet prefix and the last MCDI time sync event. We expect the
+	 * packet prefix timestamp to be after of sync event by up to one
+	 * sync event interval (0.25s) but we allow it to exceed this by a
+	 * fuzz factor of (0.1s)
+	 */
+	ptp->nic_time.sync_event_diff_min = ptp->nic_time.minor_max
+		- (ptp->nic_time.minor_max / 10);
+	ptp->nic_time.sync_event_diff_max = (ptp->nic_time.minor_max / 4)
+		+ (ptp->nic_time.minor_max / 10);
+
+	/* MC_CMD_PTP_OP_GET_ATTRIBUTES has been extended twice from an older
+	 * operation MC_CMD_PTP_OP_GET_TIME_FORMAT. The function now may return
+	 * a value to use for the minimum acceptable corrected synchronization
+	 * window and may return further capabilities.
 	 * If we have the extra information store it. For older firmware that
 	 * does not implement the extended command use the default value.
 	 */
-	if (rc == 0 && out_len >= MC_CMD_PTP_OUT_GET_ATTRIBUTES_LEN)
+	if (rc == 0 &&
+	    out_len >= MC_CMD_PTP_OUT_GET_ATTRIBUTES_CAPABILITIES_OFST)
 		ptp->min_synchronisation_ns =
 			MCDI_DWORD(outbuf,
 				   PTP_OUT_GET_ATTRIBUTES_SYNC_WINDOW_MIN);
 	else
 		ptp->min_synchronisation_ns = DEFAULT_MIN_SYNCHRONISATION_NS;
 
+	if (rc == 0 &&
+	    out_len >= MC_CMD_PTP_OUT_GET_ATTRIBUTES_LEN)
+		ptp->capabilities = MCDI_DWORD(outbuf,
+					PTP_OUT_GET_ATTRIBUTES_CAPABILITIES);
+	else
+		ptp->capabilities = 0;
+
+	/* Set up the shift for conversion between frequency
+	 * adjustments in parts-per-billion and the fixed-point
+	 * fractional ns format that the adapter uses.
+	 */
+	if (ptp->capabilities & (1 << MC_CMD_PTP_OUT_GET_ATTRIBUTES_FP44_FREQ_ADJ_LBN))
+		ptp->adjfreq_ppb_shift = PPB_SHIFT_FP44;
+	else
+		ptp->adjfreq_ppb_shift = PPB_SHIFT_FP40;
+
 	return 0;
 }
 
@@ -534,8 +699,9 @@ static int efx_ptp_get_attributes(struct efx_nic *efx)
 static int efx_ptp_get_timestamp_corrections(struct efx_nic *efx)
 {
 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_GET_TIMESTAMP_CORRECTIONS_LEN);
-	MCDI_DECLARE_BUF(outbuf, MC_CMD_PTP_OUT_GET_TIMESTAMP_CORRECTIONS_LEN);
+	MCDI_DECLARE_BUF(outbuf, MC_CMD_PTP_OUT_GET_TIMESTAMP_CORRECTIONS_V2_LEN);
 	int rc;
+	size_t out_len;
 
 	/* Get the timestamp corrections from the NIC. If this operation is
 	 * not supported (older NICs) then no correction is required.
@@ -545,21 +711,37 @@ static int efx_ptp_get_timestamp_corrections(struct efx_nic *efx)
 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
 
 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
-				outbuf, sizeof(outbuf), NULL);
+				outbuf, sizeof(outbuf), &out_len);
 	if (rc == 0) {
-		efx->ptp_data->ts_corrections.tx = MCDI_DWORD(outbuf,
+		efx->ptp_data->ts_corrections.ptp_tx = MCDI_DWORD(outbuf,
 			PTP_OUT_GET_TIMESTAMP_CORRECTIONS_TRANSMIT);
-		efx->ptp_data->ts_corrections.rx = MCDI_DWORD(outbuf,
+		efx->ptp_data->ts_corrections.ptp_rx = MCDI_DWORD(outbuf,
 			PTP_OUT_GET_TIMESTAMP_CORRECTIONS_RECEIVE);
 		efx->ptp_data->ts_corrections.pps_out = MCDI_DWORD(outbuf,
 			PTP_OUT_GET_TIMESTAMP_CORRECTIONS_PPS_OUT);
 		efx->ptp_data->ts_corrections.pps_in = MCDI_DWORD(outbuf,
 			PTP_OUT_GET_TIMESTAMP_CORRECTIONS_PPS_IN);
+
+		if (out_len >= MC_CMD_PTP_OUT_GET_TIMESTAMP_CORRECTIONS_V2_LEN) {
+			efx->ptp_data->ts_corrections.general_tx = MCDI_DWORD(
+				outbuf,
+				PTP_OUT_GET_TIMESTAMP_CORRECTIONS_V2_GENERAL_TX);
+			efx->ptp_data->ts_corrections.general_rx = MCDI_DWORD(
+				outbuf,
+				PTP_OUT_GET_TIMESTAMP_CORRECTIONS_V2_GENERAL_RX);
+		} else {
+			efx->ptp_data->ts_corrections.general_tx =
+				efx->ptp_data->ts_corrections.ptp_tx;
+			efx->ptp_data->ts_corrections.general_rx =
+				efx->ptp_data->ts_corrections.ptp_rx;
+		}
 	} else if (rc == -EINVAL) {
-		efx->ptp_data->ts_corrections.tx = 0;
-		efx->ptp_data->ts_corrections.rx = 0;
+		efx->ptp_data->ts_corrections.ptp_tx = 0;
+		efx->ptp_data->ts_corrections.ptp_rx = 0;
 		efx->ptp_data->ts_corrections.pps_out = 0;
 		efx->ptp_data->ts_corrections.pps_in = 0;
+		efx->ptp_data->ts_corrections.general_tx = 0;
+		efx->ptp_data->ts_corrections.general_rx = 0;
 	} else {
 		efx_mcdi_display_error(efx, MC_CMD_PTP, sizeof(inbuf), outbuf,
 				       sizeof(outbuf), rc);
@@ -873,8 +1055,24 @@ static int efx_ptp_synchronize(struct efx_nic *efx, unsigned int num_readings)
 	return rc;
 }
 
+/* Transmit a PTP packet via the dedicated hardware timestamped queue. */
+static void efx_ptp_xmit_skb_queue(struct efx_nic *efx, struct sk_buff *skb)
+{
+	struct efx_ptp_data *ptp_data = efx->ptp_data;
+	struct efx_tx_queue *tx_queue;
+	u8 type = skb->ip_summed == CHECKSUM_PARTIAL ? EFX_TXQ_TYPE_OFFLOAD : 0;
+
+	tx_queue = &ptp_data->channel->tx_queue[type];
+	if (tx_queue && tx_queue->timestamping) {
+		efx_enqueue_skb(tx_queue, skb);
+	} else {
+		WARN_ONCE(1, "PTP channel has no timestamped tx queue\n");
+		dev_kfree_skb_any(skb);
+	}
+}
+
 /* Transmit a PTP packet, via the MCDI interface, to the wire. */
-static int efx_ptp_xmit_skb(struct efx_nic *efx, struct sk_buff *skb)
+static void efx_ptp_xmit_skb_mc(struct efx_nic *efx, struct sk_buff *skb)
 {
 	struct efx_ptp_data *ptp_data = efx->ptp_data;
 	struct skb_shared_hwtstamps timestamps;
@@ -910,16 +1108,16 @@ static int efx_ptp_xmit_skb(struct efx_nic *efx, struct sk_buff *skb)
 	timestamps.hwtstamp = ptp_data->nic_to_kernel_time(
 		MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_MAJOR),
 		MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_MINOR),
-		ptp_data->ts_corrections.tx);
+		ptp_data->ts_corrections.ptp_tx);
 
 	skb_tstamp_tx(skb, &timestamps);
 
 	rc = 0;
 
 fail:
-	dev_kfree_skb(skb);
+	dev_kfree_skb_any(skb);
 
-	return rc;
+	return;
 }
 
 static void efx_ptp_drop_time_expired_events(struct efx_nic *efx)
@@ -1189,7 +1387,7 @@ static void efx_ptp_worker(struct work_struct *work)
 	efx_ptp_process_events(efx, &tempq);
 
 	while ((skb = skb_dequeue(&ptp_data->txq)))
-		efx_ptp_xmit_skb(efx, skb);
+		ptp_data->xmit_skb(efx, skb);
 
 	while ((skb = __skb_dequeue(&tempq)))
 		efx_ptp_process_rx(efx, skb);
@@ -1239,6 +1437,14 @@ int efx_ptp_probe(struct efx_nic *efx, struct efx_channel *channel)
 		goto fail2;
 	}
 
+	if (efx_ptp_use_mac_tx_timestamps(efx)) {
+		ptp->xmit_skb = efx_ptp_xmit_skb_queue;
+		/* Request sync events on this channel. */
+		channel->sync_events_state = SYNC_EVENTS_QUIESCENT;
+	} else {
+		ptp->xmit_skb = efx_ptp_xmit_skb_mc;
+	}
+
 	INIT_WORK(&ptp->work, efx_ptp_worker);
 	ptp->config.flags = 0;
 	ptp->config.tx_type = HWTSTAMP_TX_OFF;
@@ -1303,11 +1509,21 @@ int efx_ptp_probe(struct efx_nic *efx, struct efx_channel *channel)
 static int efx_ptp_probe_channel(struct efx_channel *channel)
 {
 	struct efx_nic *efx = channel->efx;
+	int rc;
 
 	channel->irq_moderation_us = 0;
 	channel->rx_queue.core_index = 0;
 
-	return efx_ptp_probe(efx, channel);
+	rc = efx_ptp_probe(efx, channel);
+	/* Failure to probe PTP is not fatal; this channel will just not be
+	 * used for anything.
+	 * In the case of EPERM, efx_ptp_probe will print its own message (in
+	 * efx_ptp_get_attributes()), so we don't need to.
+	 */
+	if (rc && rc != -EPERM)
+		netif_warn(efx, drv, efx->net_dev,
+			   "Failed to probe PTP, rc=%d\n", rc);
+	return 0;
 }
 
 void efx_ptp_remove(struct efx_nic *efx)
@@ -1332,6 +1548,7 @@ void efx_ptp_remove(struct efx_nic *efx)
 
 	efx_nic_free_buffer(efx, &efx->ptp_data->start);
 	kfree(efx->ptp_data);
+	efx->ptp_data = NULL;
 }
 
 static void efx_ptp_remove_channel(struct efx_channel *channel)
@@ -1548,6 +1765,17 @@ void efx_ptp_get_ts_info(struct efx_nic *efx, struct ethtool_ts_info *ts_info)
 	ts_info->so_timestamping |= (SOF_TIMESTAMPING_TX_HARDWARE |
 				     SOF_TIMESTAMPING_RX_HARDWARE |
 				     SOF_TIMESTAMPING_RAW_HARDWARE);
+	/* Check licensed features.  If we don't have the license for TX
+	 * timestamps, the NIC will not support them.
+	 */
+	if (efx_ptp_use_mac_tx_timestamps(efx)) {
+		struct efx_ef10_nic_data *nic_data = efx->nic_data;
+
+		if (!(nic_data->licensed_features &
+		      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN)))
+			ts_info->so_timestamping &=
+				~SOF_TIMESTAMPING_TX_HARDWARE;
+	}
 	if (primary && primary->ptp_data && primary->ptp_data->phc_clock)
 		ts_info->phc_index =
 			ptp_clock_index(primary->ptp_data->phc_clock);
@@ -1627,7 +1855,7 @@ static void ptp_event_rx(struct efx_nic *efx, struct efx_ptp_data *ptp)
 		evt->hwtimestamp = efx->ptp_data->nic_to_kernel_time(
 			EFX_QWORD_FIELD(ptp->evt_frags[0], MCDI_EVENT_DATA),
 			EFX_QWORD_FIELD(ptp->evt_frags[1], MCDI_EVENT_DATA),
-			ptp->ts_corrections.rx);
+			ptp->ts_corrections.ptp_rx);
 		evt->expiry = jiffies + msecs_to_jiffies(PKT_EVENT_LIFETIME_MS);
 		list_add_tail(&evt->link, &ptp->evt_list);
 
@@ -1709,9 +1937,20 @@ 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)
 {
+	struct efx_nic *efx = channel->efx;
+	struct efx_ptp_data *ptp = efx->ptp_data;
+
+	/* When extracting the sync timestamp minor value, we should discard
+	 * the least significant two bits. These are not required in order
+	 * to reconstruct full-range timestamps and they are optionally used
+	 * to report status depending on the options supplied when subscribing
+	 * for sync events.
+	 */
 	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;
+		(MCDI_EVENT_FIELD(*ev, PTP_TIME_MINOR_MS_8BITS) & 0xFC)
+			<< ptp->nic_time.sync_event_minor_shift;
+
 	/* if sync events have been disabled then we want to silently ignore
 	 * this event, so throw away result.
 	 */
@@ -1719,15 +1958,6 @@ void efx_time_sync_event(struct efx_channel *channel, efx_qword_t *ev)
 		       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)
@@ -1745,31 +1975,33 @@ void __efx_rx_skb_attach_timestamp(struct efx_channel *channel,
 				   struct sk_buff *skb)
 {
 	struct efx_nic *efx = channel->efx;
+	struct efx_ptp_data *ptp = efx->ptp_data;
 	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);
+	if (channel->sync_events_state != SYNC_EVENTS_VALID)
+		return;
+
+	pkt_timestamp_minor = efx_rx_buf_timestamp_minor(efx, skb_mac_header(skb));
 
 	/* 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);
+	diff = pkt_timestamp_minor - channel->sync_timestamp_minor;
+	if (pkt_timestamp_minor < channel->sync_timestamp_minor)
+		diff += ptp->nic_time.minor_max;
+
 	/* do we roll over a second boundary and need to carry the one? */
-	carry = channel->sync_timestamp_minor + diff > MINOR_TICKS_PER_SECOND ?
+	carry = (channel->sync_timestamp_minor >= ptp->nic_time.minor_max - diff) ?
 		1 : 0;
 
-	if (diff <= MINOR_TICKS_PER_SECOND / EXPECTED_SYNC_EVENTS_PER_SECOND +
-		    FUZZ) {
+	if (diff <= ptp->nic_time.sync_event_diff_max) {
 		/* 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) {
+	} else if (diff >= ptp->nic_time.sync_event_diff_min) {
 		/* 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.
@@ -1791,7 +2023,9 @@ void __efx_rx_skb_attach_timestamp(struct efx_channel *channel,
 	/* attach the timestamps to the skb */
 	timestamps = skb_hwtstamps(skb);
 	timestamps->hwtstamp =
-		efx_ptp_s27_to_ktime(pkt_timestamp_major, pkt_timestamp_minor);
+		ptp->nic_to_kernel_time(pkt_timestamp_major,
+					pkt_timestamp_minor,
+					ptp->ts_corrections.general_rx);
 }
 
 static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta)
@@ -1809,9 +2043,10 @@ static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta)
 	else if (delta < -MAX_PPB)
 		delta = -MAX_PPB;
 
-	/* Convert ppb to fixed point ns. */
-	adjustment_ns = (((s64)delta * PPB_SCALE_WORD) >>
-			 (PPB_EXTRA_BITS + MAX_PPB_BITS));
+	/* Convert ppb to fixed point ns taking care to round correctly. */
+	adjustment_ns = ((s64)delta * PPB_SCALE_WORD +
+			 (1 << (ptp_data->adjfreq_ppb_shift - 1))) >>
+			ptp_data->adjfreq_ppb_shift;
 
 	MCDI_SET_DWORD(inadj, PTP_IN_OP, MC_CMD_PTP_OP_ADJUST);
 	MCDI_SET_DWORD(inadj, PTP_IN_PERIPH_ID, 0);
@@ -1911,13 +2146,14 @@ static int efx_phc_enable(struct ptp_clock_info *ptp,
 	return 0;
 }
 
-static const struct efx_channel_type efx_ptp_channel_type = {
+const struct efx_channel_type efx_ptp_channel_type = {
 	.handle_no_channel	= efx_ptp_handle_no_channel,
 	.pre_probe		= efx_ptp_probe_channel,
 	.post_remove		= efx_ptp_remove_channel,
 	.get_name		= efx_ptp_get_channel_name,
 	/* no copy operation; there is no need to reallocate this channel */
 	.receive_skb		= efx_ptp_rx,
+	.want_txqs		= efx_ptp_want_txqs,
 	.keep_eventq		= false,
 };
 

drivers/net/ethernet/sfc/tx.c

@@ -77,9 +77,23 @@ static void efx_dequeue_buffer(struct efx_tx_queue *tx_queue,
 	}
 
 	if (buffer->flags & EFX_TX_BUF_SKB) {
+		struct sk_buff *skb = (struct sk_buff *)buffer->skb;
+
 		EFX_WARN_ON_PARANOID(!pkts_compl || !bytes_compl);
 		(*pkts_compl)++;
-		(*bytes_compl) += buffer->skb->len;
+		(*bytes_compl) += skb->len;
+		if (tx_queue->timestamping &&
+		    (tx_queue->completed_timestamp_major ||
+		     tx_queue->completed_timestamp_minor)) {
+			struct skb_shared_hwtstamps hwtstamp;
+
+			hwtstamp.hwtstamp =
+				efx_ptp_nic_to_kernel_time(tx_queue);
+			skb_tstamp_tx(skb, &hwtstamp);
+
+			tx_queue->completed_timestamp_major = 0;
+			tx_queue->completed_timestamp_minor = 0;
+		}
 		dev_consume_skb_any((struct sk_buff *)buffer->skb);
 		netif_vdbg(tx_queue->efx, tx_done, tx_queue->efx->net_dev,
 			   "TX queue %d transmission id %x complete\n",
@@ -828,6 +842,11 @@ void efx_init_tx_queue(struct efx_tx_queue *tx_queue)
 	tx_queue->old_read_count = 0;
 	tx_queue->empty_read_count = 0 | EFX_EMPTY_COUNT_VALID;
 	tx_queue->xmit_more_available = false;
+	tx_queue->timestamping = (efx_ptp_use_mac_tx_timestamps(efx) &&
+				  tx_queue->channel == efx_ptp_channel(efx));
+	tx_queue->completed_desc_ptr = tx_queue->ptr_mask;
+	tx_queue->completed_timestamp_major = 0;
+	tx_queue->completed_timestamp_minor = 0;
 
 	/* Set up default function pointers. These may get replaced by
 	 * efx_nic_init_tx() based off NIC/queue capabilities.