bnxt_en: Transmit and retrieve packet timestamps

Setup the TXBD to enable TX timestamp if requested.  At TX packet DMA
completion, if we requested TX timestamp on that packet, we defer to
.do_aux_work() to obtain the TX timestamp from the firmware before we
free the TX SKB.

v2: Use .do_aux_work() to get the TX timestamp from firmware.
Reviewed-by: Edwin Peer <edwin.peer@broadcom.com>
Signed-off-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/broadcom/bnxt/bnxt.c

@@ -421,12 +421,25 @@ static netdev_tx_t bnxt_start_xmit(struct sk_buff *skb, struct net_device *dev)
 			vlan_tag_flags |= 1 << TX_BD_CFA_META_TPID_SHIFT;
 	}
 
-	if (unlikely(skb->no_fcs)) {
-		lflags |= cpu_to_le32(TX_BD_FLAGS_NO_CRC);
-		goto normal_tx;
+	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
+		struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
+
+		if (ptp && ptp->tx_tstamp_en && !skb_is_gso(skb) &&
+		    atomic_dec_if_positive(&ptp->tx_avail) >= 0) {
+			if (!bnxt_ptp_parse(skb, &ptp->tx_seqid)) {
+				lflags |= cpu_to_le32(TX_BD_FLAGS_STAMP);
+				skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+			} else {
+				atomic_inc(&bp->ptp_cfg->tx_avail);
+			}
+		}
 	}
 
-	if (free_size == bp->tx_ring_size && length <= bp->tx_push_thresh) {
+	if (unlikely(skb->no_fcs))
+		lflags |= cpu_to_le32(TX_BD_FLAGS_NO_CRC);
+
+	if (free_size == bp->tx_ring_size && length <= bp->tx_push_thresh &&
+	    !lflags) {
 		struct tx_push_buffer *tx_push_buf = txr->tx_push;
 		struct tx_push_bd *tx_push = &tx_push_buf->push_bd;
 		struct tx_bd_ext *tx_push1 = &tx_push->txbd2;
@@ -593,6 +606,8 @@ static netdev_tx_t bnxt_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 	netdev_tx_sent_queue(txq, skb->len);
 
+	skb_tx_timestamp(skb);
+
 	/* Sync BD data before updating doorbell */
 	wmb();
 
@@ -622,6 +637,9 @@ static netdev_tx_t bnxt_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	return NETDEV_TX_OK;
 
 tx_dma_error:
+	if (BNXT_TX_PTP_IS_SET(lflags))
+		atomic_inc(&bp->ptp_cfg->tx_avail);
+
 	last_frag = i;
 
 	/* start back at beginning and unmap skb */
@@ -656,6 +674,7 @@ static void bnxt_tx_int(struct bnxt *bp, struct bnxt_napi *bnapi, int nr_pkts)
 
 	for (i = 0; i < nr_pkts; i++) {
 		struct bnxt_sw_tx_bd *tx_buf;
+		bool compl_deferred = false;
 		struct sk_buff *skb;
 		int j, last;
 
@@ -682,12 +701,21 @@ static void bnxt_tx_int(struct bnxt *bp, struct bnxt_napi *bnapi, int nr_pkts)
 				skb_frag_size(&skb_shinfo(skb)->frags[j]),
 				PCI_DMA_TODEVICE);
 		}
+		if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) {
+			if (bp->flags & BNXT_FLAG_CHIP_P5) {
+				if (!bnxt_get_tx_ts_p5(bp, skb))
+					compl_deferred = true;
+				else
+					atomic_inc(&bp->ptp_cfg->tx_avail);
+			}
+		}
 
 next_tx_int:
 		cons = NEXT_TX(cons);
 
 		tx_bytes += skb->len;
-		dev_kfree_skb_any(skb);
+		if (!compl_deferred)
+			dev_kfree_skb_any(skb);
 	}
 
 	netdev_tx_completed_queue(txq, nr_pkts, tx_bytes);

drivers/net/ethernet/broadcom/bnxt/bnxt.h

@@ -89,6 +89,8 @@ struct tx_bd_ext {
 	#define TX_BD_CFA_META_KEY_VLAN                         (1 << 28)
 };
 
+#define BNXT_TX_PTP_IS_SET(lflags) ((lflags) & cpu_to_le32(TX_BD_FLAGS_STAMP))
+
 struct rx_bd {
 	__le32 rx_bd_len_flags_type;
 	#define RX_BD_TYPE					(0x3f << 0)

drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.c

@@ -15,10 +15,32 @@
 #include <linux/net_tstamp.h>
 #include <linux/timecounter.h>
 #include <linux/timekeeping.h>
+#include <linux/ptp_classify.h>
 #include "bnxt_hsi.h"
 #include "bnxt.h"
 #include "bnxt_ptp.h"
 
+int bnxt_ptp_parse(struct sk_buff *skb, u16 *seq_id)
+{
+	unsigned int ptp_class;
+	struct ptp_header *hdr;
+
+	ptp_class = ptp_classify_raw(skb);
+
+	switch (ptp_class & PTP_CLASS_VMASK) {
+	case PTP_CLASS_V1:
+	case PTP_CLASS_V2:
+		hdr = ptp_parse_header(skb, ptp_class);
+		if (!hdr)
+			return -EINVAL;
+
+		*seq_id	 = ntohs(hdr->sequence_id);
+		return 0;
+	default:
+		return -ERANGE;
+	}
+}
+
 static int bnxt_ptp_settime(struct ptp_clock_info *ptp_info,
 			    const struct timespec64 *ts)
 {
@@ -57,6 +79,28 @@ static void bnxt_ptp_get_current_time(struct bnxt *bp)
 	spin_unlock_bh(&ptp->ptp_lock);
 }
 
+static int bnxt_hwrm_port_ts_query(struct bnxt *bp, u32 flags, u64 *ts)
+{
+	struct hwrm_port_ts_query_output *resp = bp->hwrm_cmd_resp_addr;
+	struct hwrm_port_ts_query_input req = {0};
+	int rc;
+
+	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_PORT_TS_QUERY, -1, -1);
+	req.flags = cpu_to_le32(flags);
+	if ((flags & PORT_TS_QUERY_REQ_FLAGS_PATH) ==
+	    PORT_TS_QUERY_REQ_FLAGS_PATH_TX) {
+		req.enables = cpu_to_le16(BNXT_PTP_QTS_TX_ENABLES);
+		req.ptp_seq_id = cpu_to_le32(bp->ptp_cfg->tx_seqid);
+		req.ts_req_timeout = cpu_to_le16(BNXT_PTP_QTS_TIMEOUT);
+	}
+	mutex_lock(&bp->hwrm_cmd_lock);
+	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
+	if (!rc)
+		*ts = le64_to_cpu(resp->ptp_msg_ts);
+	mutex_unlock(&bp->hwrm_cmd_lock);
+	return rc;
+}
+
 static int bnxt_ptp_gettimex(struct ptp_clock_info *ptp_info,
 			     struct timespec64 *ts,
 			     struct ptp_system_timestamp *sts)
@@ -269,16 +313,62 @@ static u64 bnxt_cc_read(const struct cyclecounter *cc)
 	return bnxt_refclk_read(ptp->bp, NULL);
 }
 
+static void bnxt_stamp_tx_skb(struct bnxt *bp, struct sk_buff *skb)
+{
+	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
+	struct skb_shared_hwtstamps timestamp;
+	u64 ts = 0, ns = 0;
+	int rc;
+
+	rc = bnxt_hwrm_port_ts_query(bp, PORT_TS_QUERY_REQ_FLAGS_PATH_TX, &ts);
+	if (!rc) {
+		memset(&timestamp, 0, sizeof(timestamp));
+		spin_lock_bh(&ptp->ptp_lock);
+		ns = timecounter_cyc2time(&ptp->tc, ts);
+		spin_unlock_bh(&ptp->ptp_lock);
+		timestamp.hwtstamp = ns_to_ktime(ns);
+		skb_tstamp_tx(ptp->tx_skb, &timestamp);
+	} else {
+		netdev_err(bp->dev, "TS query for TX timer failed rc = %x\n",
+			   rc);
+	}
+
+	dev_kfree_skb_any(ptp->tx_skb);
+	ptp->tx_skb = NULL;
+	atomic_inc(&ptp->tx_avail);
+}
+
 static long bnxt_ptp_ts_aux_work(struct ptp_clock_info *ptp_info)
 {
 	struct bnxt_ptp_cfg *ptp = container_of(ptp_info, struct bnxt_ptp_cfg,
 						ptp_info);
+	unsigned long now = jiffies;
 	struct bnxt *bp = ptp->bp;
 
+	if (ptp->tx_skb)
+		bnxt_stamp_tx_skb(bp, ptp->tx_skb);
+
+	if (!time_after_eq(now, ptp->next_period))
+		return ptp->next_period - now;
+
 	bnxt_ptp_get_current_time(bp);
+	ptp->next_period = now + HZ;
 	return HZ;
 }
 
+int bnxt_get_tx_ts_p5(struct bnxt *bp, struct sk_buff *skb)
+{
+	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
+
+	if (ptp->tx_skb) {
+		netdev_err(bp->dev, "deferring skb:one SKB is still outstanding\n");
+		return -EBUSY;
+	}
+	ptp->tx_skb = skb;
+	ptp_schedule_worker(ptp->ptp_clock, 0);
+	return 0;
+}
+
 int bnxt_get_rx_ts_p5(struct bnxt *bp, u64 *ts, u32 pkt_ts)
 {
 	struct bnxt_ptp_cfg *ptp = bp->ptp_cfg;
@@ -375,5 +465,9 @@ void bnxt_ptp_clear(struct bnxt *bp)
 		ptp_clock_unregister(ptp->ptp_clock);
 
 	ptp->ptp_clock = NULL;
+	if (ptp->tx_skb) {
+		dev_kfree_skb_any(ptp->tx_skb);
+		ptp->tx_skb = NULL;
+	}
 	bnxt_unmap_ptp_regs(bp);
 }

drivers/net/ethernet/broadcom/bnxt/bnxt_ptp.h

@@ -70,8 +70,10 @@ do {						\
 	((dst) = READ_ONCE(src))
 #endif
 
+int bnxt_ptp_parse(struct sk_buff *skb, u16 *seq_id);
 int bnxt_hwtstamp_set(struct net_device *dev, struct ifreq *ifr);
 int bnxt_hwtstamp_get(struct net_device *dev, struct ifreq *ifr);
+int bnxt_get_tx_ts_p5(struct bnxt *bp, struct sk_buff *skb);
 int bnxt_get_rx_ts_p5(struct bnxt *bp, u64 *ts, u32 pkt_ts);
 void bnxt_ptp_start(struct bnxt *bp);
 int bnxt_ptp_init(struct bnxt *bp);