net: aquantia: implement data PTP datapath

Here we do alloc/free IRQs for PTP rings.
We also implement processing of PTP packets on TX and RX sides.
Signed-off-by: Egor Pomozov <epomozov@marvell.com>
Co-developed-by: Sergey Samoilenko <sergey.samoilenko@aquantia.com>
Signed-off-by: Sergey Samoilenko <sergey.samoilenko@aquantia.com>
Co-developed-by: Dmitry Bezrukov <dmitry.bezrukov@aquantia.com>
Signed-off-by: Dmitry Bezrukov <dmitry.bezrukov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/aquantia/atlantic/aq_cfg.h

 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
  * aQuantia Corporation Network Driver
- * Copyright (C) 2014-2017 aQuantia Corporation. All rights reserved
+ * Copyright (C) 2014-2019 aQuantia Corporation. All rights reserved
  */
 
 /* File aq_cfg.h: Definition of configuration parameters and constants. */
@@ -27,7 +27,7 @@
 
 #define AQ_CFG_INTERRUPT_MODERATION_USEC_MAX (0x1FF * 2)
 
-#define AQ_CFG_IRQ_MASK                      0x1FFU
+#define AQ_CFG_IRQ_MASK                      0x3FFU
 
 #define AQ_CFG_VECS_MAX   8U
 #define AQ_CFG_TCS_MAX    8U

drivers/net/ethernet/aquantia/atlantic/aq_hw.h

@@ -244,6 +244,12 @@ struct aq_hw_ops {
 
 	int (*hw_rx_tc_mode_get)(struct aq_hw_s *self, u32 *tc_mode);
 
+	int (*hw_ring_hwts_rx_fill)(struct aq_hw_s *self,
+				    struct aq_ring_s *aq_ring);
+
+	int (*hw_ring_hwts_rx_receive)(struct aq_hw_s *self,
+				       struct aq_ring_s *ring);
+
 	void (*hw_get_ptp_ts)(struct aq_hw_s *self, u64 *stamp);
 
 	int (*hw_adj_clock_freq)(struct aq_hw_s *self, s32 delta);
@@ -252,6 +258,12 @@ struct aq_hw_ops {
 
 	int (*hw_set_sys_clock)(struct aq_hw_s *self, u64 time, u64 ts);
 
+	u16 (*rx_extract_ts)(struct aq_hw_s *self, u8 *p, unsigned int len,
+			     u64 *timestamp);
+
+	int (*extract_hwts)(struct aq_hw_s *self, u8 *p, unsigned int len,
+			    u64 *timestamp);
+
 	int (*hw_set_fc)(struct aq_hw_s *self, u32 fc, u32 tc);
 };
 

drivers/net/ethernet/aquantia/atlantic/aq_main.c

 // SPDX-License-Identifier: GPL-2.0-only
 /*
  * aQuantia Corporation Network Driver
- * Copyright (C) 2014-2017 aQuantia Corporation. All rights reserved
+ * Copyright (C) 2014-2019 aQuantia Corporation. All rights reserved
  */
 
 /* File aq_main.c: Main file for aQuantia Linux driver. */
@@ -10,10 +10,13 @@
 #include "aq_nic.h"
 #include "aq_pci_func.h"
 #include "aq_ethtool.h"
+#include "aq_ptp.h"
 #include "aq_filters.h"
 
 #include <linux/netdevice.h>
 #include <linux/module.h>
+#include <linux/ip.h>
+#include <linux/udp.h>
 
 MODULE_LICENSE("GPL v2");
 MODULE_VERSION(AQ_CFG_DRV_VERSION);
@@ -93,6 +96,24 @@ static int aq_ndev_start_xmit(struct sk_buff *skb, struct net_device *ndev)
 {
 	struct aq_nic_s *aq_nic = netdev_priv(ndev);
 
+	if (unlikely(aq_utils_obj_test(&aq_nic->flags, AQ_NIC_PTP_DPATH_UP))) {
+		/* Hardware adds the Timestamp for PTPv2 802.AS1
+		 * and PTPv2 IPv4 UDP.
+		 * We have to push even general 320 port messages to the ptp
+		 * queue explicitly. This is a limitation of current firmware
+		 * and hardware PTP design of the chip. Otherwise ptp stream
+		 * will fail to sync
+		 */
+		if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) ||
+		    unlikely((ip_hdr(skb)->version == 4) &&
+			     (ip_hdr(skb)->protocol == IPPROTO_UDP) &&
+			     ((udp_hdr(skb)->dest == htons(319)) ||
+			      (udp_hdr(skb)->dest == htons(320)))) ||
+		    unlikely(eth_hdr(skb)->h_proto == htons(ETH_P_1588)))
+			return aq_ptp_xmit(aq_nic, skb);
+	}
+
+	skb_tx_timestamp(skb);
 	return aq_nic_xmit(aq_nic, skb);
 }
 

drivers/net/ethernet/aquantia/atlantic/aq_nic.c

@@ -146,8 +146,11 @@ static int aq_nic_update_link_status(struct aq_nic_s *self)
 			self->aq_hw->aq_link_status.mbps);
 		aq_nic_update_interrupt_moderation_settings(self);
 
-		if (self->aq_ptp)
+		if (self->aq_ptp) {
 			aq_ptp_clock_init(self);
+			aq_ptp_tm_offset_set(self,
+					     self->aq_hw->aq_link_status.mbps);
+		}
 
 		/* Driver has to update flow control settings on RX block
 		 * on any link event.
@@ -196,6 +199,8 @@ static void aq_nic_service_task(struct work_struct *work)
 					     service_task);
 	int err;
 
+	aq_ptp_service_task(self);
+
 	if (aq_utils_obj_test(&self->flags, AQ_NIC_FLAGS_IS_NOT_READY))
 		return;
 
@@ -408,6 +413,10 @@ int aq_nic_start(struct aq_nic_s *self)
 				goto err_exit;
 		}
 
+		err = aq_ptp_irq_alloc(self);
+		if (err < 0)
+			goto err_exit;
+
 		if (self->aq_nic_cfg.link_irq_vec) {
 			int irqvec = pci_irq_vector(self->pdev,
 						   self->aq_nic_cfg.link_irq_vec);
@@ -440,9 +449,8 @@ int aq_nic_start(struct aq_nic_s *self)
 	return err;
 }
 
-static unsigned int aq_nic_map_skb(struct aq_nic_s *self,
-				   struct sk_buff *skb,
-				   struct aq_ring_s *ring)
+unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
+			    struct aq_ring_s *ring)
 {
 	unsigned int ret = 0U;
 	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
@@ -973,6 +981,8 @@ int aq_nic_stop(struct aq_nic_s *self)
 	else
 		aq_pci_func_free_irqs(self);
 
+	aq_ptp_irq_free(self);
+
 	for (i = 0U, aq_vec = self->aq_vec[0];
 		self->aq_vecs > i; ++i, aq_vec = self->aq_vec[i])
 		aq_vec_stop(aq_vec);

drivers/net/ethernet/aquantia/atlantic/aq_nic.h

@@ -54,6 +54,7 @@ struct aq_nic_cfg_s {
 #define AQ_NIC_FLAG_STOPPING    0x00000008U
 #define AQ_NIC_FLAG_RESETTING   0x00000010U
 #define AQ_NIC_FLAG_CLOSING     0x00000020U
+#define AQ_NIC_PTP_DPATH_UP     0x02000000U
 #define AQ_NIC_LINK_DOWN        0x04000000U
 #define AQ_NIC_FLAG_ERR_UNPLUG  0x40000000U
 #define AQ_NIC_FLAG_ERR_HW      0x80000000U
@@ -129,6 +130,8 @@ void aq_nic_cfg_start(struct aq_nic_s *self);
 int aq_nic_ndev_register(struct aq_nic_s *self);
 void aq_nic_ndev_free(struct aq_nic_s *self);
 int aq_nic_start(struct aq_nic_s *self);
+unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
+			    struct aq_ring_s *ring);
 int aq_nic_xmit(struct aq_nic_s *self, struct sk_buff *skb);
 int aq_nic_get_regs(struct aq_nic_s *self, struct ethtool_regs *regs, void *p);
 int aq_nic_get_regs_count(struct aq_nic_s *self);

drivers/net/ethernet/aquantia/atlantic/aq_pci_func.c

 // SPDX-License-Identifier: GPL-2.0-only
 /*
  * aQuantia Corporation Network Driver
- * Copyright (C) 2014-2017 aQuantia Corporation. All rights reserved
+ * Copyright (C) 2014-2019 aQuantia Corporation. All rights reserved
  */
 
 /* File aq_pci_func.c: Definition of PCI functions. */
@@ -269,6 +269,9 @@ static int aq_pci_probe(struct pci_dev *pdev,
 	numvecs = min((u8)AQ_CFG_VECS_DEF,
 		      aq_nic_get_cfg(self)->aq_hw_caps->msix_irqs);
 	numvecs = min(numvecs, num_online_cpus());
+	/* Request IRQ vector for PTP */
+	numvecs += 1;
+
 	numvecs += AQ_HW_SERVICE_IRQS;
 	/*enable interrupts */
 #if !AQ_CFG_FORCE_LEGACY_INT

drivers/net/ethernet/aquantia/atlantic/aq_ptp.c

@@ -8,12 +8,24 @@
  */
 
 #include <linux/ptp_clock_kernel.h>
+#include <linux/interrupt.h>
 #include <linux/clocksource.h>
 
 #include "aq_nic.h"
 #include "aq_ptp.h"
 #include "aq_ring.h"
 
+#define AQ_PTP_TX_TIMEOUT        (HZ *  10)
+
+enum ptp_speed_offsets {
+	ptp_offset_idx_10 = 0,
+	ptp_offset_idx_100,
+	ptp_offset_idx_1000,
+	ptp_offset_idx_2500,
+	ptp_offset_idx_5000,
+	ptp_offset_idx_10000,
+};
+
 struct ptp_skb_ring {
 	struct sk_buff **buff;
 	spinlock_t lock;
@@ -22,6 +34,12 @@ struct ptp_skb_ring {
 	unsigned int tail;
 };
 
+struct ptp_tx_timeout {
+	spinlock_t lock;
+	bool active;
+	unsigned long tx_start;
+};
+
 struct aq_ptp_s {
 	struct aq_nic_s *aq_nic;
 	spinlock_t ptp_lock;
@@ -29,8 +47,16 @@ struct aq_ptp_s {
 	struct ptp_clock *ptp_clock;
 	struct ptp_clock_info ptp_info;
 
+	atomic_t offset_egress;
+	atomic_t offset_ingress;
+
 	struct aq_ring_param_s ptp_ring_param;
 
+	struct ptp_tx_timeout ptp_tx_timeout;
+
+	unsigned int idx_vector;
+	struct napi_struct napi;
+
 	struct aq_ring_s ptp_tx;
 	struct aq_ring_s ptp_rx;
 	struct aq_ring_s hwts_rx;
@@ -38,6 +64,101 @@ struct aq_ptp_s {
 	struct ptp_skb_ring skb_ring;
 };
 
+struct ptp_tm_offset {
+	unsigned int mbps;
+	int egress;
+	int ingress;
+};
+
+static struct ptp_tm_offset ptp_offset[6];
+
+void aq_ptp_tm_offset_set(struct aq_nic_s *aq_nic, unsigned int mbps)
+{
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+	int i, egress, ingress;
+
+	if (!aq_ptp)
+		return;
+
+	egress = 0;
+	ingress = 0;
+
+	for (i = 0; i < ARRAY_SIZE(ptp_offset); i++) {
+		if (mbps == ptp_offset[i].mbps) {
+			egress = ptp_offset[i].egress;
+			ingress = ptp_offset[i].ingress;
+			break;
+		}
+	}
+
+	atomic_set(&aq_ptp->offset_egress, egress);
+	atomic_set(&aq_ptp->offset_ingress, ingress);
+}
+
+static int __aq_ptp_skb_put(struct ptp_skb_ring *ring, struct sk_buff *skb)
+{
+	unsigned int next_head = (ring->head + 1) % ring->size;
+
+	if (next_head == ring->tail)
+		return -ENOMEM;
+
+	ring->buff[ring->head] = skb_get(skb);
+	ring->head = next_head;
+
+	return 0;
+}
+
+static int aq_ptp_skb_put(struct ptp_skb_ring *ring, struct sk_buff *skb)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&ring->lock, flags);
+	ret = __aq_ptp_skb_put(ring, skb);
+	spin_unlock_irqrestore(&ring->lock, flags);
+
+	return ret;
+}
+
+static struct sk_buff *__aq_ptp_skb_get(struct ptp_skb_ring *ring)
+{
+	struct sk_buff *skb;
+
+	if (ring->tail == ring->head)
+		return NULL;
+
+	skb = ring->buff[ring->tail];
+	ring->tail = (ring->tail + 1) % ring->size;
+
+	return skb;
+}
+
+static struct sk_buff *aq_ptp_skb_get(struct ptp_skb_ring *ring)
+{
+	unsigned long flags;
+	struct sk_buff *skb;
+
+	spin_lock_irqsave(&ring->lock, flags);
+	skb = __aq_ptp_skb_get(ring);
+	spin_unlock_irqrestore(&ring->lock, flags);
+
+	return skb;
+}
+
+static unsigned int aq_ptp_skb_buf_len(struct ptp_skb_ring *ring)
+{
+	unsigned long flags;
+	unsigned int len;
+
+	spin_lock_irqsave(&ring->lock, flags);
+	len = (ring->head >= ring->tail) ?
+	ring->head - ring->tail :
+	ring->size - ring->tail + ring->head;
+	spin_unlock_irqrestore(&ring->lock, flags);
+
+	return len;
+}
+
 static int aq_ptp_skb_ring_init(struct ptp_skb_ring *ring, unsigned int size)
 {
 	struct sk_buff **buff = kmalloc(sizeof(*buff) * size, GFP_KERNEL);
@@ -55,10 +176,75 @@ static int aq_ptp_skb_ring_init(struct ptp_skb_ring *ring, unsigned int size)
 	return 0;
 }
 
+static void aq_ptp_skb_ring_clean(struct ptp_skb_ring *ring)
+{
+	struct sk_buff *skb;
+
+	while ((skb = aq_ptp_skb_get(ring)) != NULL)
+		dev_kfree_skb_any(skb);
+}
+
 static void aq_ptp_skb_ring_release(struct ptp_skb_ring *ring)
 {
-	kfree(ring->buff);
-	ring->buff = NULL;
+	if (ring->buff) {
+		aq_ptp_skb_ring_clean(ring);
+		kfree(ring->buff);
+		ring->buff = NULL;
+	}
+}
+
+static void aq_ptp_tx_timeout_init(struct ptp_tx_timeout *timeout)
+{
+	spin_lock_init(&timeout->lock);
+	timeout->active = false;
+}
+
+static void aq_ptp_tx_timeout_start(struct aq_ptp_s *aq_ptp)
+{
+	struct ptp_tx_timeout *timeout = &aq_ptp->ptp_tx_timeout;
+	unsigned long flags;
+
+	spin_lock_irqsave(&timeout->lock, flags);
+	timeout->active = true;
+	timeout->tx_start = jiffies;
+	spin_unlock_irqrestore(&timeout->lock, flags);
+}
+
+static void aq_ptp_tx_timeout_update(struct aq_ptp_s *aq_ptp)
+{
+	if (!aq_ptp_skb_buf_len(&aq_ptp->skb_ring)) {
+		struct ptp_tx_timeout *timeout = &aq_ptp->ptp_tx_timeout;
+		unsigned long flags;
+
+		spin_lock_irqsave(&timeout->lock, flags);
+		timeout->active = false;
+		spin_unlock_irqrestore(&timeout->lock, flags);
+	}
+}
+
+static void aq_ptp_tx_timeout_check(struct aq_ptp_s *aq_ptp)
+{
+	struct ptp_tx_timeout *timeout = &aq_ptp->ptp_tx_timeout;
+	unsigned long flags;
+	bool timeout_flag;
+
+	timeout_flag = false;
+
+	spin_lock_irqsave(&timeout->lock, flags);
+	if (timeout->active) {
+		timeout_flag = time_is_before_jiffies(timeout->tx_start +
+						      AQ_PTP_TX_TIMEOUT);
+		/* reset active flag if timeout detected */
+		if (timeout_flag)
+			timeout->active = false;
+	}
+	spin_unlock_irqrestore(&timeout->lock, flags);
+
+	if (timeout_flag) {
+		aq_ptp_skb_ring_clean(&aq_ptp->skb_ring);
+		netdev_err(aq_ptp->aq_nic->ndev,
+			   "PTP Timeout. Clearing Tx Timestamp SKBs\n");
+	}
 }
 
 /* aq_ptp_adjfine
@@ -148,6 +334,263 @@ static int aq_ptp_settime(struct ptp_clock_info *ptp,
 	return 0;
 }
 
+static void aq_ptp_convert_to_hwtstamp(struct aq_ptp_s *aq_ptp,
+				       struct skb_shared_hwtstamps *hwtstamp,
+				       u64 timestamp)
+{
+	memset(hwtstamp, 0, sizeof(*hwtstamp));
+	hwtstamp->hwtstamp = ns_to_ktime(timestamp);
+}
+
+/* aq_ptp_tx_hwtstamp - utility function which checks for TX time stamp
+ * @adapter: the private adapter struct
+ *
+ * if the timestamp is valid, we convert it into the timecounter ns
+ * value, then store that result into the hwtstamps structure which
+ * is passed up the network stack
+ */
+void aq_ptp_tx_hwtstamp(struct aq_nic_s *aq_nic, u64 timestamp)
+{
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+	struct sk_buff *skb = aq_ptp_skb_get(&aq_ptp->skb_ring);
+	struct skb_shared_hwtstamps hwtstamp;
+
+	if (!skb) {
+		netdev_err(aq_nic->ndev, "have timestamp but tx_queus empty\n");
+		return;
+	}
+
+	timestamp += atomic_read(&aq_ptp->offset_egress);
+	aq_ptp_convert_to_hwtstamp(aq_ptp, &hwtstamp, timestamp);
+	skb_tstamp_tx(skb, &hwtstamp);
+	dev_kfree_skb_any(skb);
+
+	aq_ptp_tx_timeout_update(aq_ptp);
+}
+
+/* aq_ptp_rx_hwtstamp - utility function which checks for RX time stamp
+ * @adapter: pointer to adapter struct
+ * @skb: particular skb to send timestamp with
+ *
+ * if the timestamp is valid, we convert it into the timecounter ns
+ * value, then store that result into the hwtstamps structure which
+ * is passed up the network stack
+ */
+static void aq_ptp_rx_hwtstamp(struct aq_ptp_s *aq_ptp, struct sk_buff *skb,
+			       u64 timestamp)
+{
+	timestamp -= atomic_read(&aq_ptp->offset_ingress);
+	aq_ptp_convert_to_hwtstamp(aq_ptp, skb_hwtstamps(skb), timestamp);
+}
+
+bool aq_ptp_ring(struct aq_nic_s *aq_nic, struct aq_ring_s *ring)
+{
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+
+	if (!aq_ptp)
+		return false;
+
+	return &aq_ptp->ptp_tx == ring ||
+	       &aq_ptp->ptp_rx == ring || &aq_ptp->hwts_rx == ring;
+}
+
+u16 aq_ptp_extract_ts(struct aq_nic_s *aq_nic, struct sk_buff *skb, u8 *p,
+		      unsigned int len)
+{
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+	u64 timestamp = 0;
+	u16 ret = aq_nic->aq_hw_ops->rx_extract_ts(aq_nic->aq_hw,
+						   p, len, &timestamp);
+
+	if (ret > 0)
+		aq_ptp_rx_hwtstamp(aq_ptp, skb, timestamp);
+
+	return ret;
+}
+
+static int aq_ptp_poll(struct napi_struct *napi, int budget)
+{
+	struct aq_ptp_s *aq_ptp = container_of(napi, struct aq_ptp_s, napi);
+	struct aq_nic_s *aq_nic = aq_ptp->aq_nic;
+	bool was_cleaned = false;
+	int work_done = 0;
+	int err;
+
+	/* Processing PTP TX traffic */
+	err = aq_nic->aq_hw_ops->hw_ring_tx_head_update(aq_nic->aq_hw,
+							&aq_ptp->ptp_tx);
+	if (err < 0)
+		goto err_exit;
+
+	if (aq_ptp->ptp_tx.sw_head != aq_ptp->ptp_tx.hw_head) {
+		aq_ring_tx_clean(&aq_ptp->ptp_tx);
+
+		was_cleaned = true;
+	}
+
+	/* Processing HW_TIMESTAMP RX traffic */
+	err = aq_nic->aq_hw_ops->hw_ring_hwts_rx_receive(aq_nic->aq_hw,
+							 &aq_ptp->hwts_rx);
+	if (err < 0)
+		goto err_exit;
+
+	if (aq_ptp->hwts_rx.sw_head != aq_ptp->hwts_rx.hw_head) {
+		aq_ring_hwts_rx_clean(&aq_ptp->hwts_rx, aq_nic);
+
+		err = aq_nic->aq_hw_ops->hw_ring_hwts_rx_fill(aq_nic->aq_hw,
+							      &aq_ptp->hwts_rx);
+
+		was_cleaned = true;
+	}
+
+	/* Processing PTP RX traffic */
+	err = aq_nic->aq_hw_ops->hw_ring_rx_receive(aq_nic->aq_hw,
+						    &aq_ptp->ptp_rx);
+	if (err < 0)
+		goto err_exit;
+
+	if (aq_ptp->ptp_rx.sw_head != aq_ptp->ptp_rx.hw_head) {
+		unsigned int sw_tail_old;
+
+		err = aq_ring_rx_clean(&aq_ptp->ptp_rx, napi, &work_done, budget);
+		if (err < 0)
+			goto err_exit;
+
+		sw_tail_old = aq_ptp->ptp_rx.sw_tail;
+		err = aq_ring_rx_fill(&aq_ptp->ptp_rx);
+		if (err < 0)
+			goto err_exit;
+
+		err = aq_nic->aq_hw_ops->hw_ring_rx_fill(aq_nic->aq_hw,
+							 &aq_ptp->ptp_rx,
+							 sw_tail_old);
+		if (err < 0)
+			goto err_exit;
+	}
+
+	if (was_cleaned)
+		work_done = budget;
+
+	if (work_done < budget) {
+		napi_complete_done(napi, work_done);
+		aq_nic->aq_hw_ops->hw_irq_enable(aq_nic->aq_hw,
+					1 << aq_ptp->ptp_ring_param.vec_idx);
+	}
+
+err_exit:
+	return work_done;
+}
+
+static irqreturn_t aq_ptp_isr(int irq, void *private)
+{
+	struct aq_ptp_s *aq_ptp = private;
+	int err = 0;
+
+	if (!aq_ptp) {
+		err = -EINVAL;
+		goto err_exit;
+	}
+	napi_schedule(&aq_ptp->napi);
+
+err_exit:
+	return err >= 0 ? IRQ_HANDLED : IRQ_NONE;
+}
+
+int aq_ptp_xmit(struct aq_nic_s *aq_nic, struct sk_buff *skb)
+{
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+	struct aq_ring_s *ring = &aq_ptp->ptp_tx;
+	unsigned long irq_flags;
+	int err = NETDEV_TX_OK;
+	unsigned int frags;
+
+	if (skb->len <= 0) {
+		dev_kfree_skb_any(skb);
+		goto err_exit;
+	}
+
+	frags = skb_shinfo(skb)->nr_frags + 1;
+	/* Frags cannot be bigger 16KB
+	 * because PTP usually works
+	 * without Jumbo even in a background
+	 */
+	if (frags > AQ_CFG_SKB_FRAGS_MAX || frags > aq_ring_avail_dx(ring)) {
+		/* Drop packet because it doesn't make sence to delay it */
+		dev_kfree_skb_any(skb);
+		goto err_exit;
+	}
+
+	err = aq_ptp_skb_put(&aq_ptp->skb_ring, skb);
+	if (err) {
+		netdev_err(aq_nic->ndev, "SKB Ring is overflow (%u)!\n",
+			   ring->size);
+		return NETDEV_TX_BUSY;
+	}
+	skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+	aq_ptp_tx_timeout_start(aq_ptp);
+	skb_tx_timestamp(skb);
+
+	spin_lock_irqsave(&aq_nic->aq_ptp->ptp_ring_lock, irq_flags);
+	frags = aq_nic_map_skb(aq_nic, skb, ring);
+
+	if (likely(frags)) {
+		err = aq_nic->aq_hw_ops->hw_ring_tx_xmit(aq_nic->aq_hw,
+						       ring, frags);
+		if (err >= 0) {
+			++ring->stats.tx.packets;
+			ring->stats.tx.bytes += skb->len;
+		}
+	} else {
+		err = NETDEV_TX_BUSY;
+	}
+	spin_unlock_irqrestore(&aq_nic->aq_ptp->ptp_ring_lock, irq_flags);
+
+err_exit:
+	return err;
+}
+
+void aq_ptp_service_task(struct aq_nic_s *aq_nic)
+{
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+
+	if (!aq_ptp)
+		return;
+
+	aq_ptp_tx_timeout_check(aq_ptp);
+}
+
+int aq_ptp_irq_alloc(struct aq_nic_s *aq_nic)
+{
+	struct pci_dev *pdev = aq_nic->pdev;
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+	int err = 0;
+
+	if (!aq_ptp)
+		return 0;
+
+	if (pdev->msix_enabled || pdev->msi_enabled) {
+		err = request_irq(pci_irq_vector(pdev, aq_ptp->idx_vector),
+				  aq_ptp_isr, 0, aq_nic->ndev->name, aq_ptp);
+	} else {
+		err = -EINVAL;
+		goto err_exit;
+	}
+
+err_exit:
+	return err;
+}
+
+void aq_ptp_irq_free(struct aq_nic_s *aq_nic)
+{
+	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
+	struct pci_dev *pdev = aq_nic->pdev;
+
+	if (!aq_ptp)
+		return;
+
+	free_irq(pci_irq_vector(pdev, aq_ptp->idx_vector), aq_ptp);
+}
+
 int aq_ptp_ring_init(struct aq_nic_s *aq_nic)
 {
 	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
@@ -189,6 +632,12 @@ int aq_ptp_ring_init(struct aq_nic_s *aq_nic)
 	err = aq_nic->aq_hw_ops->hw_ring_rx_init(aq_nic->aq_hw,
 						 &aq_ptp->hwts_rx,
 						 &aq_ptp->ptp_ring_param);
+	if (err < 0)
+		goto err_exit;
+	err = aq_nic->aq_hw_ops->hw_ring_hwts_rx_fill(aq_nic->aq_hw,
+						      &aq_ptp->hwts_rx);
+	if (err < 0)
+		goto err_exit;
 
 	return err;
 
@@ -219,6 +668,8 @@ int aq_ptp_ring_start(struct aq_nic_s *aq_nic)
 	if (err < 0)
 		goto err_exit;
 
+	napi_enable(&aq_ptp->napi);
+
 err_exit:
 	return err;
 }
@@ -234,6 +685,8 @@ void aq_ptp_ring_stop(struct aq_nic_s *aq_nic)
 	aq_nic->aq_hw_ops->hw_ring_rx_stop(aq_nic->aq_hw, &aq_ptp->ptp_rx);
 
 	aq_nic->aq_hw_ops->hw_ring_rx_stop(aq_nic->aq_hw, &aq_ptp->hwts_rx);
+
+	napi_disable(&aq_ptp->napi);
 }
 
 void aq_ptp_ring_deinit(struct aq_nic_s *aq_nic)
@@ -306,6 +759,12 @@ int aq_ptp_ring_alloc(struct aq_nic_s *aq_nic)
 		goto err_exit_hwts_rx;
 	}
 
+	aq_ptp->ptp_ring_param.vec_idx = aq_ptp->idx_vector;
+	aq_ptp->ptp_ring_param.cpu = aq_ptp->ptp_ring_param.vec_idx +
+			aq_nic_get_cfg(aq_nic)->aq_rss.base_cpu_number;
+	cpumask_set_cpu(aq_ptp->ptp_ring_param.cpu,
+			&aq_ptp->ptp_ring_param.affinity_mask);
+
 	return 0;
 
 err_exit_hwts_rx:
@@ -347,6 +806,60 @@ static struct ptp_clock_info aq_ptp_clock = {
 	.pin_config	= NULL,
 };
 
+#define ptp_offset_init(__idx, __mbps, __egress, __ingress)   do { \
+		ptp_offset[__idx].mbps = (__mbps); \
+		ptp_offset[__idx].egress = (__egress); \
+		ptp_offset[__idx].ingress = (__ingress); } \
+		while (0)
+
+static void aq_ptp_offset_init_from_fw(const struct hw_aq_ptp_offset *offsets)
+{
+	int i;
+
+	/* Load offsets for PTP */
+	for (i = 0; i < ARRAY_SIZE(ptp_offset); i++) {
+		switch (i) {
+		/* 100M */
+		case ptp_offset_idx_100:
+			ptp_offset_init(i, 100,
+					offsets->egress_100,
+					offsets->ingress_100);
+			break;
+		/* 1G */
+		case ptp_offset_idx_1000:
+			ptp_offset_init(i, 1000,
+					offsets->egress_1000,
+					offsets->ingress_1000);
+			break;
+		/* 2.5G */
+		case ptp_offset_idx_2500:
+			ptp_offset_init(i, 2500,
+					offsets->egress_2500,
+					offsets->ingress_2500);
+			break;
+		/* 5G */
+		case ptp_offset_idx_5000:
+			ptp_offset_init(i, 5000,
+					offsets->egress_5000,
+					offsets->ingress_5000);
+			break;
+		/* 10G */
+		case ptp_offset_idx_10000:
+			ptp_offset_init(i, 10000,
+					offsets->egress_10000,
+					offsets->ingress_10000);
+			break;
+		}
+	}
+}
+
+static void aq_ptp_offset_init(const struct hw_aq_ptp_offset *offsets)
+{
+	memset(ptp_offset, 0, sizeof(ptp_offset));
+
+	aq_ptp_offset_init_from_fw(offsets);
+}
+
 void aq_ptp_clock_init(struct aq_nic_s *aq_nic)
 {
 	struct aq_ptp_s *aq_ptp = aq_nic->aq_ptp;
@@ -380,6 +893,8 @@ int aq_ptp_init(struct aq_nic_s *aq_nic, unsigned int idx_vec)
 		return 0;
 	}
 
+	aq_ptp_offset_init(&mbox.info.ptp_offset);
+
 	aq_ptp = kzalloc(sizeof(*aq_ptp), GFP_KERNEL);
 	if (!aq_ptp) {
 		err = -ENOMEM;
@@ -399,6 +914,15 @@ int aq_ptp_init(struct aq_nic_s *aq_nic, unsigned int idx_vec)
 		goto err_exit;
 	}
 	aq_ptp->ptp_clock = clock;
+	aq_ptp_tx_timeout_init(&aq_ptp->ptp_tx_timeout);
+
+	atomic_set(&aq_ptp->offset_egress, 0);
+	atomic_set(&aq_ptp->offset_ingress, 0);
+
+	netif_napi_add(aq_nic_get_ndev(aq_nic), &aq_ptp->napi,
+		       aq_ptp_poll, AQ_CFG_NAPI_WEIGHT);
+
+	aq_ptp->idx_vector = idx_vec;
 
 	aq_nic->aq_ptp = aq_ptp;
 
@@ -439,6 +963,12 @@ void aq_ptp_free(struct aq_nic_s *aq_nic)
 	aq_nic->aq_fw_ops->enable_ptp(aq_nic->aq_hw, 0);
 	mutex_unlock(&aq_nic->fwreq_mutex);
 
+	netif_napi_del(&aq_ptp->napi);
 	kfree(aq_ptp);
 	aq_nic->aq_ptp = NULL;
 }
+
+struct ptp_clock *aq_ptp_get_ptp_clock(struct aq_ptp_s *aq_ptp)
+{
+	return aq_ptp->ptp_clock;
+}

drivers/net/ethernet/aquantia/atlantic/aq_ptp.h

@@ -17,6 +17,9 @@ int aq_ptp_init(struct aq_nic_s *aq_nic, unsigned int idx_vec);
 void aq_ptp_unregister(struct aq_nic_s *aq_nic);
 void aq_ptp_free(struct aq_nic_s *aq_nic);
 
+int aq_ptp_irq_alloc(struct aq_nic_s *aq_nic);
+void aq_ptp_irq_free(struct aq_nic_s *aq_nic);
+
 int aq_ptp_ring_alloc(struct aq_nic_s *aq_nic);
 void aq_ptp_ring_free(struct aq_nic_s *aq_nic);
 
@@ -25,6 +28,22 @@ int aq_ptp_ring_start(struct aq_nic_s *aq_nic);
 void aq_ptp_ring_stop(struct aq_nic_s *aq_nic);
 void aq_ptp_ring_deinit(struct aq_nic_s *aq_nic);
 
+void aq_ptp_service_task(struct aq_nic_s *aq_nic);
+
+void aq_ptp_tm_offset_set(struct aq_nic_s *aq_nic, unsigned int mbps);
+
 void aq_ptp_clock_init(struct aq_nic_s *aq_nic);
 
+/* Traffic processing functions */
+int aq_ptp_xmit(struct aq_nic_s *aq_nic, struct sk_buff *skb);
+void aq_ptp_tx_hwtstamp(struct aq_nic_s *aq_nic, u64 timestamp);
+
+/* Return either ring is belong to PTP or not*/
+bool aq_ptp_ring(struct aq_nic_s *aq_nic, struct aq_ring_s *ring);
+
+u16 aq_ptp_extract_ts(struct aq_nic_s *aq_nic, struct sk_buff *skb, u8 *p,
+		      unsigned int len);
+
+struct ptp_clock *aq_ptp_get_ptp_clock(struct aq_ptp_s *aq_ptp);
+
 #endif /* AQ_PTP_H */

drivers/net/ethernet/aquantia/atlantic/aq_ring.c

@@ -10,6 +10,7 @@
 #include "aq_nic.h"
 #include "aq_hw.h"
 #include "aq_hw_utils.h"
+#include "aq_ptp.h"
 
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
@@ -314,6 +315,7 @@ int aq_ring_rx_clean(struct aq_ring_s *self,
 		self->sw_head = aq_ring_next_dx(self, self->sw_head),
 		--budget, ++(*work_done)) {
 		struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head];
+		bool is_ptp_ring = aq_ptp_ring(self->aq_nic, self);
 		struct aq_ring_buff_s *buff_ = NULL;
 		struct sk_buff *skb = NULL;
 		unsigned int next_ = 0U;
@@ -378,6 +380,11 @@ int aq_ring_rx_clean(struct aq_ring_s *self,
 				err = -ENOMEM;
 				goto err_exit;
 			}
+			if (is_ptp_ring)
+				buff->len -=
+					aq_ptp_extract_ts(self->aq_nic, skb,
+						aq_buf_vaddr(&buff->rxdata),
+						buff->len);
 			skb_put(skb, buff->len);
 			page_ref_inc(buff->rxdata.page);
 		} else {
@@ -386,6 +393,11 @@ int aq_ring_rx_clean(struct aq_ring_s *self,
 				err = -ENOMEM;
 				goto err_exit;
 			}
+			if (is_ptp_ring)
+				buff->len -=
+					aq_ptp_extract_ts(self->aq_nic, skb,
+						aq_buf_vaddr(&buff->rxdata),
+						buff->len);
 
 			hdr_len = buff->len;
 			if (hdr_len > AQ_CFG_RX_HDR_SIZE)
@@ -445,8 +457,8 @@ int aq_ring_rx_clean(struct aq_ring_s *self,
 		skb_set_hash(skb, buff->rss_hash,
 			     buff->is_hash_l4 ? PKT_HASH_TYPE_L4 :
 			     PKT_HASH_TYPE_NONE);
-
-		skb_record_rx_queue(skb, self->idx);
+		/* Send all PTP traffic to 0 queue */
+		skb_record_rx_queue(skb, is_ptp_ring ? 0 : self->idx);
 
 		++self->stats.rx.packets;
 		self->stats.rx.bytes += skb->len;
@@ -458,6 +470,21 @@ int aq_ring_rx_clean(struct aq_ring_s *self,
 	return err;
 }
 
+void aq_ring_hwts_rx_clean(struct aq_ring_s *self, struct aq_nic_s *aq_nic)
+{
+	while (self->sw_head != self->hw_head) {
+		u64 ns;
+
+		aq_nic->aq_hw_ops->extract_hwts(aq_nic->aq_hw,
+						self->dx_ring +
+						(self->sw_head * self->dx_size),
+						self->dx_size, &ns);
+		aq_ptp_tx_hwtstamp(aq_nic, ns);
+
+		self->sw_head = aq_ring_next_dx(self, self->sw_head);
+	}
+}
+
 int aq_ring_rx_fill(struct aq_ring_s *self)
 {
 	unsigned int page_order = self->page_order;

drivers/net/ethernet/aquantia/atlantic/aq_ring.h

@@ -177,5 +177,6 @@ int aq_ring_rx_fill(struct aq_ring_s *self);
 struct aq_ring_s *aq_ring_hwts_rx_alloc(struct aq_ring_s *self,
 		struct aq_nic_s *aq_nic, unsigned int idx,
 		unsigned int size, unsigned int dx_size);
+void aq_ring_hwts_rx_clean(struct aq_ring_s *self, struct aq_nic_s *aq_nic);
 
 #endif /* AQ_RING_H */

drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_b0.c

@@ -682,6 +682,46 @@ static int hw_atl_b0_hw_ring_rx_fill(struct aq_hw_s *self,
 	return aq_hw_err_from_flags(self);
 }
 
+static int hw_atl_b0_hw_ring_hwts_rx_fill(struct aq_hw_s *self,
+					  struct aq_ring_s *ring)
+{
+	unsigned int i;
+
+	for (i = aq_ring_avail_dx(ring); i--;
+			ring->sw_tail = aq_ring_next_dx(ring, ring->sw_tail)) {
+		struct hw_atl_rxd_s *rxd =
+			(struct hw_atl_rxd_s *)
+			&ring->dx_ring[ring->sw_tail * HW_ATL_B0_RXD_SIZE];
+
+		rxd->buf_addr = ring->dx_ring_pa + ring->size * ring->dx_size;
+		rxd->hdr_addr = 0U;
+	}
+	/* Make sure descriptors are updated before bump tail*/
+	wmb();
+
+	hw_atl_reg_rx_dma_desc_tail_ptr_set(self, ring->sw_tail, ring->idx);
+
+	return aq_hw_err_from_flags(self);
+}
+
+static int hw_atl_b0_hw_ring_hwts_rx_receive(struct aq_hw_s *self,
+					     struct aq_ring_s *ring)
+{
+	while (ring->hw_head != ring->sw_tail) {
+		struct hw_atl_rxd_hwts_wb_s *hwts_wb =
+			(struct hw_atl_rxd_hwts_wb_s *)
+			(ring->dx_ring + (ring->hw_head * HW_ATL_B0_RXD_SIZE));
+
+		/* RxD is not done */
+		if (!(hwts_wb->sec_lw0 & 0x1U))
+			break;
+
+		ring->hw_head = aq_ring_next_dx(ring, ring->hw_head);
+	}
+
+	return aq_hw_err_from_flags(self);
+}
+
 static int hw_atl_b0_hw_ring_tx_head_update(struct aq_hw_s *self,
 					    struct aq_ring_s *ring)
 {
@@ -1133,6 +1173,61 @@ static int hw_atl_b0_adj_clock_freq(struct aq_hw_s *self, s32 ppb)
 	return self->aq_fw_ops->send_fw_request(self, &fwreq, size);
 }
 
+static u16 hw_atl_b0_rx_extract_ts(struct aq_hw_s *self, u8 *p,
+				   unsigned int len, u64 *timestamp)
+{
+	unsigned int offset = 14;
+	struct ethhdr *eth;
+	u64 sec;
+	u8 *ptr;
+	u32 ns;
+
+	if (len <= offset || !timestamp)
+		return 0;
+
+	/* The TIMESTAMP in the end of package has following format:
+	 * (big-endian)
+	 *   struct {
+	 *     uint64_t sec;
+	 *     uint32_t ns;
+	 *     uint16_t stream_id;
+	 *   };
+	 */
+	ptr = p + (len - offset);
+	memcpy(&sec, ptr, sizeof(sec));
+	ptr += sizeof(sec);
+	memcpy(&ns, ptr, sizeof(ns));
+
+	sec = be64_to_cpu(sec) & 0xffffffffffffllu;
+	ns = be32_to_cpu(ns);
+	*timestamp = sec * NSEC_PER_SEC + ns + self->ptp_clk_offset;
+
+	eth = (struct ethhdr *)p;
+
+	return (eth->h_proto == htons(ETH_P_1588)) ? 12 : 14;
+}
+
+static int hw_atl_b0_extract_hwts(struct aq_hw_s *self, u8 *p, unsigned int len,
+				  u64 *timestamp)
+{
+	struct hw_atl_rxd_hwts_wb_s *hwts_wb = (struct hw_atl_rxd_hwts_wb_s *)p;
+	u64 tmp, sec, ns;
+
+	sec = 0;
+	tmp = (hwts_wb->sec_lw0 >> 2) & 0x3ff;
+	sec += tmp;
+	tmp = (u64)((hwts_wb->sec_lw1 >> 16) & 0xffff) << 10;
+	sec += tmp;
+	tmp = (u64)(hwts_wb->sec_hw & 0xfff) << 26;
+	sec += tmp;
+	tmp = (u64)((hwts_wb->sec_hw >> 22) & 0x3ff) << 38;
+	sec += tmp;
+	ns = sec * NSEC_PER_SEC + hwts_wb->ns;
+	if (timestamp)
+		*timestamp = ns + self->ptp_clk_offset;
+	return 0;
+}
+
 static int hw_atl_b0_hw_fl3l4_clear(struct aq_hw_s *self,
 				    struct aq_rx_filter_l3l4 *data)
 {
@@ -1309,11 +1404,16 @@ const struct aq_hw_ops hw_atl_ops_b0 = {
 	.hw_tx_tc_mode_get       = hw_atl_b0_tx_tc_mode_get,
 	.hw_rx_tc_mode_get       = hw_atl_b0_rx_tc_mode_get,
 
+	.hw_ring_hwts_rx_fill        = hw_atl_b0_hw_ring_hwts_rx_fill,
+	.hw_ring_hwts_rx_receive     = hw_atl_b0_hw_ring_hwts_rx_receive,
+
 	.hw_get_ptp_ts           = hw_atl_b0_get_ptp_ts,
 	.hw_adj_sys_clock        = hw_atl_b0_adj_sys_clock,
 	.hw_set_sys_clock        = hw_atl_b0_set_sys_clock,
 	.hw_adj_clock_freq       = hw_atl_b0_adj_clock_freq,
 
+	.rx_extract_ts           = hw_atl_b0_rx_extract_ts,
+	.extract_hwts            = hw_atl_b0_extract_hwts,
 	.hw_set_offload          = hw_atl_b0_hw_offload_set,
 	.hw_set_fc                   = hw_atl_b0_set_fc,
 };