[PATCH net-next v2 0/2] net: micrel: Change how to read TX timestamp

[PATCH net-next v2 0/2] net: micrel: Change how to read TX timestamp

[Horatiu Vultur]

Change how to read the timestamp of the received frames. Currently it is
required to read over the MDIO bus each of the timestamps but it is
possible to change to receive the timestamp (the nanosecond part and the
least significant two bits of the second) in the frame.
The first patch makes this change, while the second patch optimized the
read of second part. Because it is not required to read the second part
of the timestmap for each received frame but it is OK to read the second
part twice per second and then cache it.

v1->v2:
- create a patch series instead of single patch
- add optimization for reading the second part of the received timestamp,
  read the second twice per second instead of for each frame

Horatiu Vultur (2):
  net: micrel: Change to receive timestamp in the frame for lan8841
  net: micrel: Schedule work to read seconds for lan8841

 drivers/net/phy/micrel.c | 272 ++++++++++++++++++++++++++-------------
 1 file changed, 180 insertions(+), 92 deletions(-)

-- 
2.38.0

[PATCH net-next v2 1/2] net: micrel: Change to receive timestamp in the frame for lan8841

[Horatiu Vultur]

Currently for each timestamp frame, the SW needs to go and read the
received timestamp over the MDIO bus. But the HW has the capability
to store the received nanoseconds part and the least significant two
bits of the seconds in the reserved field of the PTP header. In this
way we could save few mdio transactions (actually a little more
transactions because the access to the PTP registers are indirect)
for each received frame. But is still required to read the seconds
part of each received frame.
Doing these changes to start to get the received timestamp in the
reserved field of the header, will give a great CPU usage performance.
Running ptp4l with logSyncInterval of -9 will give a ~50% CPU
improvment.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
---
 drivers/net/phy/micrel.c | 231 +++++++++++++++++++++++----------------
 1 file changed, 139 insertions(+), 92 deletions(-)

diff --git a/drivers/net/phy/micrel.c b/drivers/net/phy/micrel.c
index 0ff4fd9f1a183..28365006b2067 100644
--- a/drivers/net/phy/micrel.c
+++ b/drivers/net/phy/micrel.c
@@ -3502,6 +3502,9 @@ static int lan8814_probe(struct phy_device *phydev)
 #define LAN8841_PTP_CMD_CTL_PTP_RESET		BIT(0)
 #define LAN8841_PTP_RX_PARSE_CONFIG		368
 #define LAN8841_PTP_TX_PARSE_CONFIG		432
+#define LAN8841_PTP_RX_MODE			381
+#define LAN8841_PTP_INSERT_TS_EN		BIT(0)
+#define LAN8841_PTP_INSERT_TS_32BIT		BIT(1)
 
 static int lan8841_config_init(struct phy_device *phydev)
 {
@@ -3650,68 +3653,18 @@ static void lan8841_ptp_process_tx_ts(struct kszphy_ptp_priv *ptp_priv)
 		lan8814_match_tx_skb(ptp_priv, sec, nsec, seq);
 }
 
-#define LAN8841_PTP_RX_INGRESS_SEC_LO			389
-#define LAN8841_PTP_RX_INGRESS_SEC_HI			388
-#define LAN8841_PTP_RX_INGRESS_NS_LO			387
-#define LAN8841_PTP_RX_INGRESS_NS_HI			386
-#define LAN8841_PTP_RX_INGRESS_NSEC_HI_VALID		BIT(15)
-#define LAN8841_PTP_RX_MSG_HEADER2			391
-
-static struct lan8814_ptp_rx_ts *lan8841_ptp_get_rx_ts(struct kszphy_ptp_priv *ptp_priv)
-{
-	struct phy_device *phydev = ptp_priv->phydev;
-	struct lan8814_ptp_rx_ts *rx_ts;
-	u32 sec, nsec;
-	u16 seq;
-
-	nsec = phy_read_mmd(phydev, 2, LAN8841_PTP_RX_INGRESS_NS_HI);
-	if (!(nsec & LAN8841_PTP_RX_INGRESS_NSEC_HI_VALID))
-		return NULL;
-
-	nsec = ((nsec & 0x3fff) << 16);
-	nsec = nsec | phy_read_mmd(phydev, 2, LAN8841_PTP_RX_INGRESS_NS_LO);
-
-	sec = phy_read_mmd(phydev, 2, LAN8841_PTP_RX_INGRESS_SEC_HI);
-	sec = sec << 16;
-	sec = sec | phy_read_mmd(phydev, 2, LAN8841_PTP_RX_INGRESS_SEC_LO);
-
-	seq = phy_read_mmd(phydev, 2, LAN8841_PTP_RX_MSG_HEADER2);
-
-	rx_ts = kzalloc(sizeof(*rx_ts), GFP_KERNEL);
-	if (!rx_ts)
-		return NULL;
-
-	rx_ts->seconds = sec;
-	rx_ts->nsec = nsec;
-	rx_ts->seq_id = seq;
-
-	return rx_ts;
-}
-
-static void lan8841_ptp_process_rx_ts(struct kszphy_ptp_priv *ptp_priv)
-{
-	struct lan8814_ptp_rx_ts *rx_ts;
-
-	while ((rx_ts = lan8841_ptp_get_rx_ts(ptp_priv)) != NULL)
-		lan8814_match_rx_ts(ptp_priv, rx_ts);
-}
-
 #define LAN8841_PTP_INT_STS			259
 #define LAN8841_PTP_INT_STS_PTP_TX_TS_OVRFL_INT	BIT(13)
 #define LAN8841_PTP_INT_STS_PTP_TX_TS_INT	BIT(12)
-#define LAN8841_PTP_INT_STS_PTP_RX_TS_OVRFL_INT	BIT(9)
-#define LAN8841_PTP_INT_STS_PTP_RX_TS_INT	BIT(8)
 #define LAN8841_PTP_INT_STS_PTP_GPIO_CAP_INT	BIT(2)
 
-static void lan8841_ptp_flush_fifo(struct kszphy_ptp_priv *ptp_priv, bool egress)
+static void lan8841_ptp_flush_fifo(struct kszphy_ptp_priv *ptp_priv)
 {
 	struct phy_device *phydev = ptp_priv->phydev;
 	int i;
 
 	for (i = 0; i < FIFO_SIZE; ++i)
-		phy_read_mmd(phydev, 2,
-			     egress ? LAN8841_PTP_TX_MSG_HEADER2 :
-				      LAN8841_PTP_RX_MSG_HEADER2);
+		phy_read_mmd(phydev, 2, LAN8841_PTP_TX_MSG_HEADER2);
 
 	phy_read_mmd(phydev, 2, LAN8841_PTP_INT_STS);
 }
@@ -3789,23 +3742,17 @@ static void lan8841_handle_ptp_interrupt(struct phy_device *phydev)
 		if (status & LAN8841_PTP_INT_STS_PTP_TX_TS_INT)
 			lan8841_ptp_process_tx_ts(ptp_priv);
 
-		if (status & LAN8841_PTP_INT_STS_PTP_RX_TS_INT)
-			lan8841_ptp_process_rx_ts(ptp_priv);
-
 		if (status & LAN8841_PTP_INT_STS_PTP_GPIO_CAP_INT)
 			lan8841_gpio_process_cap(ptp_priv);
 
 		if (status & LAN8841_PTP_INT_STS_PTP_TX_TS_OVRFL_INT) {
-			lan8841_ptp_flush_fifo(ptp_priv, true);
+			lan8841_ptp_flush_fifo(ptp_priv);
 			skb_queue_purge(&ptp_priv->tx_queue);
 		}
 
-		if (status & LAN8841_PTP_INT_STS_PTP_RX_TS_OVRFL_INT) {
-			lan8841_ptp_flush_fifo(ptp_priv, false);
-			skb_queue_purge(&ptp_priv->rx_queue);
-		}
-
-	} while (status);
+	} while (status & (LAN8841_PTP_INT_STS_PTP_TX_TS_INT |
+			   LAN8841_PTP_INT_STS_PTP_GPIO_CAP_INT |
+			   LAN8841_PTP_INT_STS_PTP_TX_TS_OVRFL_INT));
 }
 
 #define LAN8841_INTS_PTP		BIT(9)
@@ -3869,32 +3816,44 @@ static int lan8841_ts_info(struct mii_timestamper *mii_ts,
 #define LAN8841_PTP_INT_EN			260
 #define LAN8841_PTP_INT_EN_PTP_TX_TS_OVRFL_EN	BIT(13)
 #define LAN8841_PTP_INT_EN_PTP_TX_TS_EN		BIT(12)
-#define LAN8841_PTP_INT_EN_PTP_RX_TS_OVRFL_EN	BIT(9)
-#define LAN8841_PTP_INT_EN_PTP_RX_TS_EN		BIT(8)
 
-static void lan8841_ptp_enable_int(struct kszphy_ptp_priv *ptp_priv,
-				   bool enable)
+static void lan8841_ptp_enable_processing(struct kszphy_ptp_priv *ptp_priv,
+					  bool enable)
 {
 	struct phy_device *phydev = ptp_priv->phydev;
 
-	if (enable)
-		/* Enable interrupts */
+	if (enable) {
+		/* Enable interrupts on the TX side */
 		phy_modify_mmd(phydev, 2, LAN8841_PTP_INT_EN,
 			       LAN8841_PTP_INT_EN_PTP_TX_TS_OVRFL_EN |
-			       LAN8841_PTP_INT_EN_PTP_RX_TS_OVRFL_EN |
-			       LAN8841_PTP_INT_EN_PTP_TX_TS_EN |
-			       LAN8841_PTP_INT_EN_PTP_RX_TS_EN,
+			       LAN8841_PTP_INT_EN_PTP_TX_TS_EN,
 			       LAN8841_PTP_INT_EN_PTP_TX_TS_OVRFL_EN |
-			       LAN8841_PTP_INT_EN_PTP_RX_TS_OVRFL_EN |
-			       LAN8841_PTP_INT_EN_PTP_TX_TS_EN |
-			       LAN8841_PTP_INT_EN_PTP_RX_TS_EN);
-	else
-		/* Disable interrupts */
+			       LAN8841_PTP_INT_EN_PTP_TX_TS_EN);
+
+		/* Enable the modification of the frame on RX side,
+		 * this will add the ns and 2 bits of sec in the reserved field
+		 * of the PTP header
+		 */
+		phy_modify_mmd(phydev, KSZ9131RN_MMD_COMMON_CTRL_REG,
+			       LAN8841_PTP_RX_MODE,
+			       LAN8841_PTP_INSERT_TS_EN |
+			       LAN8841_PTP_INSERT_TS_32BIT,
+			       LAN8841_PTP_INSERT_TS_EN |
+			       LAN8841_PTP_INSERT_TS_32BIT);
+	} else {
+		/* Disable interrupts on the TX side */
 		phy_modify_mmd(phydev, 2, LAN8841_PTP_INT_EN,
 			       LAN8841_PTP_INT_EN_PTP_TX_TS_OVRFL_EN |
-			       LAN8841_PTP_INT_EN_PTP_RX_TS_OVRFL_EN |
-			       LAN8841_PTP_INT_EN_PTP_TX_TS_EN |
-			       LAN8841_PTP_INT_EN_PTP_RX_TS_EN, 0);
+			       LAN8841_PTP_INT_EN_PTP_TX_TS_EN, 0);
+
+		/* Disable modification of the RX frames */
+		phy_modify_mmd(phydev, KSZ9131RN_MMD_COMMON_CTRL_REG,
+			       LAN8841_PTP_RX_MODE,
+			       LAN8841_PTP_INSERT_TS_EN |
+			       LAN8841_PTP_INSERT_TS_32BIT, 0);
+
+		ptp_cancel_worker_sync(ptp_priv->ptp_clock);
+	}
 }
 
 #define LAN8841_PTP_RX_TIMESTAMP_EN		379
@@ -3905,7 +3864,6 @@ static int lan8841_hwtstamp(struct mii_timestamper *mii_ts, struct ifreq *ifr)
 {
 	struct kszphy_ptp_priv *ptp_priv = container_of(mii_ts, struct kszphy_ptp_priv, mii_ts);
 	struct phy_device *phydev = ptp_priv->phydev;
-	struct lan8814_ptp_rx_ts *rx_ts, *tmp;
 	struct hwtstamp_config config;
 	int txcfg = 0, rxcfg = 0;
 	int pkt_ts_enable;
@@ -3969,24 +3927,47 @@ static int lan8841_hwtstamp(struct mii_timestamper *mii_ts, struct ifreq *ifr)
 				PTP_TX_MOD_TX_PTP_SYNC_TS_INSERT_ : 0);
 
 	/* Now enable/disable the timestamping */
-	lan8841_ptp_enable_int(ptp_priv,
-			       config.rx_filter != HWTSTAMP_FILTER_NONE);
-
-	/* In case of multiple starts and stops, these needs to be cleared */
-	list_for_each_entry_safe(rx_ts, tmp, &ptp_priv->rx_ts_list, list) {
-		list_del(&rx_ts->list);
-		kfree(rx_ts);
-	}
+	lan8841_ptp_enable_processing(ptp_priv,
+				      config.rx_filter != HWTSTAMP_FILTER_NONE);
 
 	skb_queue_purge(&ptp_priv->rx_queue);
 	skb_queue_purge(&ptp_priv->tx_queue);
 
-	lan8841_ptp_flush_fifo(ptp_priv, false);
-	lan8841_ptp_flush_fifo(ptp_priv, true);
+	lan8841_ptp_flush_fifo(ptp_priv);
 
 	return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ? -EFAULT : 0;
 }
 
+#define LAN8841_SKB_CB(skb)	((struct lan8841_skb_cb *)(skb)->cb)
+
+struct lan8841_skb_cb {
+	struct ptp_header *header;
+};
+
+static bool lan8841_rxtstamp(struct mii_timestamper *mii_ts,
+			     struct sk_buff *skb, int type)
+{
+	struct kszphy_ptp_priv *ptp_priv =
+			container_of(mii_ts, struct kszphy_ptp_priv, mii_ts);
+	struct ptp_header *header = ptp_parse_header(skb, type);
+
+	if (!header)
+		return false;
+
+	if (ptp_priv->rx_filter == HWTSTAMP_FILTER_NONE ||
+	    type == PTP_CLASS_NONE)
+		return false;
+
+	if ((type & ptp_priv->version) == 0 || (type & ptp_priv->layer) == 0)
+		return false;
+
+	LAN8841_SKB_CB(skb)->header = header;
+	skb_queue_tail(&ptp_priv->rx_queue, skb);
+	ptp_schedule_worker(ptp_priv->ptp_clock, 0);
+
+	return true;
+}
+
 #define LAN8841_EVENT_A		0
 #define LAN8841_EVENT_B		1
 #define LAN8841_PTP_LTC_TARGET_SEC_HI(event)	((event) == LAN8841_EVENT_A ? 278 : 288)
@@ -4127,6 +4108,30 @@ static int lan8841_ptp_gettime64(struct ptp_clock_info *ptp,
 	return 0;
 }
 
+static void lan8841_ptp_getseconds(struct ptp_clock_info *ptp,
+				   struct timespec64 *ts)
+{
+	struct kszphy_ptp_priv *ptp_priv = container_of(ptp, struct kszphy_ptp_priv,
+							ptp_clock_info);
+	struct phy_device *phydev = ptp_priv->phydev;
+	time64_t s;
+
+	mutex_lock(&ptp_priv->ptp_lock);
+	/* Issue the command to read the LTC */
+	phy_write_mmd(phydev, 2, LAN8841_PTP_CMD_CTL,
+		      LAN8841_PTP_CMD_CTL_PTP_LTC_READ);
+
+	/* Read the LTC */
+	s = phy_read_mmd(phydev, 2, LAN8841_PTP_LTC_RD_SEC_HI);
+	s <<= 16;
+	s |= phy_read_mmd(phydev, 2, LAN8841_PTP_LTC_RD_SEC_MID);
+	s <<= 16;
+	s |= phy_read_mmd(phydev, 2, LAN8841_PTP_LTC_RD_SEC_LO);
+	mutex_unlock(&ptp_priv->ptp_lock);
+
+	set_normalized_timespec64(ts, s, 0);
+}
+
 #define LAN8841_PTP_LTC_STEP_ADJ_LO			276
 #define LAN8841_PTP_LTC_STEP_ADJ_HI			275
 #define LAN8841_PTP_LTC_STEP_ADJ_DIR			BIT(15)
@@ -4629,6 +4634,37 @@ static int lan8841_ptp_enable(struct ptp_clock_info *ptp,
 	return 0;
 }
 
+static long lan8841_ptp_do_aux_work(struct ptp_clock_info *ptp)
+{
+	struct kszphy_ptp_priv *ptp_priv = container_of(ptp, struct kszphy_ptp_priv,
+							ptp_clock_info);
+	struct skb_shared_hwtstamps *shhwtstamps;
+	struct timespec64 ts;
+	struct sk_buff *skb;
+	u32 ts_header;
+
+	while ((skb = skb_dequeue(&ptp_priv->rx_queue)) != NULL) {
+		lan8841_ptp_getseconds(ptp, &ts);
+		ts_header = __be32_to_cpu(LAN8841_SKB_CB(skb)->header->reserved2);
+
+		shhwtstamps = skb_hwtstamps(skb);
+		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
+
+		/* Check for any wrap arounds for the second part */
+		if ((ts.tv_sec & GENMASK(1, 0)) < ts_header >> 30)
+			ts.tv_sec -= GENMASK(1, 0) + 1;
+
+		shhwtstamps->hwtstamp =
+			ktime_set((ts.tv_sec & ~(GENMASK(1, 0))) | ts_header >> 30,
+				  ts_header & GENMASK(29, 0));
+		LAN8841_SKB_CB(skb)->header->reserved2 = 0;
+
+		netif_rx(skb);
+	}
+
+	return -1;
+}
+
 static struct ptp_clock_info lan8841_ptp_clock_info = {
 	.owner		= THIS_MODULE,
 	.name		= "lan8841 ptp",
@@ -4639,6 +4675,7 @@ static struct ptp_clock_info lan8841_ptp_clock_info = {
 	.adjfine	= lan8841_ptp_adjfine,
 	.verify         = lan8841_ptp_verify,
 	.enable         = lan8841_ptp_enable,
+	.do_aux_work	= lan8841_ptp_do_aux_work,
 	.n_per_out      = LAN8841_PTP_GPIO_NUM,
 	.n_ext_ts       = LAN8841_PTP_GPIO_NUM,
 	.n_pins         = LAN8841_PTP_GPIO_NUM,
@@ -4705,7 +4742,7 @@ static int lan8841_probe(struct phy_device *phydev)
 	ptp_priv->phydev = phydev;
 	mutex_init(&ptp_priv->ptp_lock);
 
-	ptp_priv->mii_ts.rxtstamp = lan8814_rxtstamp;
+	ptp_priv->mii_ts.rxtstamp = lan8841_rxtstamp;
 	ptp_priv->mii_ts.txtstamp = lan8814_txtstamp;
 	ptp_priv->mii_ts.hwtstamp = lan8841_hwtstamp;
 	ptp_priv->mii_ts.ts_info = lan8841_ts_info;
@@ -4715,6 +4752,16 @@ static int lan8841_probe(struct phy_device *phydev)
 	return 0;
 }
 
+static int lan8841_suspend(struct phy_device *phydev)
+{
+	struct kszphy_priv *priv = phydev->priv;
+	struct kszphy_ptp_priv *ptp_priv = &priv->ptp_priv;
+
+	ptp_cancel_worker_sync(ptp_priv->ptp_clock);
+
+	return genphy_suspend(phydev);
+}
+
 static struct phy_driver ksphy_driver[] = {
 {
 	.phy_id		= PHY_ID_KS8737,
@@ -4938,7 +4985,7 @@ static struct phy_driver ksphy_driver[] = {
 	.get_sset_count = kszphy_get_sset_count,
 	.get_strings	= kszphy_get_strings,
 	.get_stats	= kszphy_get_stats,
-	.suspend	= genphy_suspend,
+	.suspend	= lan8841_suspend,
 	.resume		= genphy_resume,
 	.cable_test_start	= lan8814_cable_test_start,
 	.cable_test_get_status	= ksz886x_cable_test_get_status,
-- 
2.38.0

[PATCH net-next v2 2/2] net: micrel: Schedule work to read seconds for lan8841

[Horatiu Vultur]

Instead of reading the seconds part of the received frame for each of
the frames, schedule a workqueue to read the seconds part every 500ms
and then for each of the received frames use this information instead of
reading again the seconds part. Because if for example running with 512
frames per second, there is no point to read 512 times the second part.
Of course care needs to be taken in case of the less two significant
bits are 0 or 3, to make sure there are no seconds wraparound.
This will improve the CPU usage by ~20% and also it is possible to receive
1024 Sync frames per second.

Signed-off-by: Horatiu Vultur <horatiu.vultur@microchip.com>
---
 drivers/net/phy/micrel.c | 49 ++++++++++++++++++++++++++++++++++++----
 1 file changed, 45 insertions(+), 4 deletions(-)

diff --git a/drivers/net/phy/micrel.c b/drivers/net/phy/micrel.c
index 28365006b2067..9832eea404377 100644
--- a/drivers/net/phy/micrel.c
+++ b/drivers/net/phy/micrel.c
@@ -33,6 +33,7 @@
 #include <linux/ptp_classify.h>
 #include <linux/net_tstamp.h>
 #include <linux/gpio/consumer.h>
+#include <linux/workqueue.h>
 
 /* Operation Mode Strap Override */
 #define MII_KSZPHY_OMSO				0x16
@@ -254,6 +255,9 @@
 #define PS_TO_REG				200
 #define FIFO_SIZE				8
 
+/* Delay used to get the second part from the LTC */
+#define LAN8841_GET_SEC_LTC_DELAY		(500 * NSEC_PER_MSEC)
+
 struct kszphy_hw_stat {
 	const char *string;
 	u8 reg;
@@ -321,6 +325,9 @@ struct kszphy_ptp_priv {
 	/* Lock for ptp_clock */
 	struct mutex ptp_lock;
 	struct ptp_pin_desc *pin_config;
+
+	s64 seconds;
+	struct delayed_work seconds_work;
 };
 
 struct kszphy_priv {
@@ -3840,6 +3847,12 @@ static void lan8841_ptp_enable_processing(struct kszphy_ptp_priv *ptp_priv,
 			       LAN8841_PTP_INSERT_TS_32BIT,
 			       LAN8841_PTP_INSERT_TS_EN |
 			       LAN8841_PTP_INSERT_TS_32BIT);
+
+		/* Schedule the work to read the seconds, which will be used in
+		 * the received timestamp
+		 */
+		schedule_delayed_work(&ptp_priv->seconds_work,
+				      nsecs_to_jiffies(LAN8841_GET_SEC_LTC_DELAY));
 	} else {
 		/* Disable interrupts on the TX side */
 		phy_modify_mmd(phydev, 2, LAN8841_PTP_INT_EN,
@@ -3853,6 +3866,11 @@ static void lan8841_ptp_enable_processing(struct kszphy_ptp_priv *ptp_priv,
 			       LAN8841_PTP_INSERT_TS_32BIT, 0);
 
 		ptp_cancel_worker_sync(ptp_priv->ptp_clock);
+
+		/* Stop the work, as there is no reason to continue to read the
+		 * seconds if there is no timestamping enabled
+		 */
+		cancel_delayed_work_sync(&ptp_priv->seconds_work);
 	}
 }
 
@@ -4062,6 +4080,8 @@ static int lan8841_ptp_settime64(struct ptp_clock_info *ptp,
 	phy_write_mmd(phydev, 2, LAN8841_PTP_LTC_SET_NS_LO, lower_16_bits(ts->tv_nsec));
 	phy_write_mmd(phydev, 2, LAN8841_PTP_LTC_SET_NS_HI, upper_16_bits(ts->tv_nsec) & 0x3fff);
 
+	ptp_priv->seconds = ts->tv_sec;
+
 	/* Set the command to load the LTC */
 	phy_write_mmd(phydev, 2, LAN8841_PTP_CMD_CTL,
 		      LAN8841_PTP_CMD_CTL_PTP_LTC_LOAD);
@@ -4116,7 +4136,6 @@ static void lan8841_ptp_getseconds(struct ptp_clock_info *ptp,
 	struct phy_device *phydev = ptp_priv->phydev;
 	time64_t s;
 
-	mutex_lock(&ptp_priv->ptp_lock);
 	/* Issue the command to read the LTC */
 	phy_write_mmd(phydev, 2, LAN8841_PTP_CMD_CTL,
 		      LAN8841_PTP_CMD_CTL_PTP_LTC_READ);
@@ -4127,7 +4146,6 @@ static void lan8841_ptp_getseconds(struct ptp_clock_info *ptp,
 	s |= phy_read_mmd(phydev, 2, LAN8841_PTP_LTC_RD_SEC_MID);
 	s <<= 16;
 	s |= phy_read_mmd(phydev, 2, LAN8841_PTP_LTC_RD_SEC_LO);
-	mutex_unlock(&ptp_priv->ptp_lock);
 
 	set_normalized_timespec64(ts, s, 0);
 }
@@ -4644,15 +4662,20 @@ static long lan8841_ptp_do_aux_work(struct ptp_clock_info *ptp)
 	u32 ts_header;
 
 	while ((skb = skb_dequeue(&ptp_priv->rx_queue)) != NULL) {
-		lan8841_ptp_getseconds(ptp, &ts);
+		mutex_lock(&ptp_priv->ptp_lock);
+		ts.tv_sec = ptp_priv->seconds;
+		mutex_unlock(&ptp_priv->ptp_lock);
+
 		ts_header = __be32_to_cpu(LAN8841_SKB_CB(skb)->header->reserved2);
 
 		shhwtstamps = skb_hwtstamps(skb);
 		memset(shhwtstamps, 0, sizeof(*shhwtstamps));
 
 		/* Check for any wrap arounds for the second part */
-		if ((ts.tv_sec & GENMASK(1, 0)) < ts_header >> 30)
+		if ((ts.tv_sec & GENMASK(1, 0)) == 0 && (ts_header >> 30) == 3)
 			ts.tv_sec -= GENMASK(1, 0) + 1;
+		else if ((ts.tv_sec & GENMASK(1, 0)) == 3 && (ts_header >> 30) == 0)
+			ts.tv_sec += 1;
 
 		shhwtstamps->hwtstamp =
 			ktime_set((ts.tv_sec & ~(GENMASK(1, 0))) | ts_header >> 30,
@@ -4665,6 +4688,21 @@ static long lan8841_ptp_do_aux_work(struct ptp_clock_info *ptp)
 	return -1;
 }
 
+static void lan8841_ptp_seconds_work(struct work_struct *seconds_work)
+{
+	struct kszphy_ptp_priv *ptp_priv =
+		container_of(seconds_work, struct kszphy_ptp_priv, seconds_work.work);
+	struct timespec64 ts;
+
+	mutex_lock(&ptp_priv->ptp_lock);
+	lan8841_ptp_getseconds(&ptp_priv->ptp_clock_info, &ts);
+	ptp_priv->seconds = ts.tv_sec;
+	mutex_unlock(&ptp_priv->ptp_lock);
+
+	schedule_delayed_work(&ptp_priv->seconds_work,
+			      nsecs_to_jiffies(LAN8841_GET_SEC_LTC_DELAY));
+}
+
 static struct ptp_clock_info lan8841_ptp_clock_info = {
 	.owner		= THIS_MODULE,
 	.name		= "lan8841 ptp",
@@ -4749,6 +4787,8 @@ static int lan8841_probe(struct phy_device *phydev)
 
 	phydev->mii_ts = &ptp_priv->mii_ts;
 
+	INIT_DELAYED_WORK(&ptp_priv->seconds_work, lan8841_ptp_seconds_work);
+
 	return 0;
 }
 
@@ -4758,6 +4798,7 @@ static int lan8841_suspend(struct phy_device *phydev)
 	struct kszphy_ptp_priv *ptp_priv = &priv->ptp_priv;
 
 	ptp_cancel_worker_sync(ptp_priv->ptp_clock);
+	cancel_delayed_work_sync(&ptp_priv->seconds_work);
 
 	return genphy_suspend(phydev);
 }
-- 
2.38.0

Re: [PATCH net-next v2 0/2] net: micrel: Change how to read TX timestamp

[Horatiu Vultur]

The 06/13/2023 11:45, Horatiu Vultur wrote:

Argh... I forgot add Richard to this thread. Therefore CC him now.

> Change how to read the timestamp of the received frames. Currently it is
> required to read over the MDIO bus each of the timestamps but it is
> possible to change to receive the timestamp (the nanosecond part and the
> least significant two bits of the second) in the frame.
> The first patch makes this change, while the second patch optimized the
> read of second part. Because it is not required to read the second part
> of the timestmap for each received frame but it is OK to read the second
> part twice per second and then cache it.
> 
> v1->v2:
> - create a patch series instead of single patch
> - add optimization for reading the second part of the received timestamp,
>   read the second twice per second instead of for each frame
> 
> Horatiu Vultur (2):
>   net: micrel: Change to receive timestamp in the frame for lan8841
>   net: micrel: Schedule work to read seconds for lan8841
> 
>  drivers/net/phy/micrel.c | 272 ++++++++++++++++++++++++++-------------
>  1 file changed, 180 insertions(+), 92 deletions(-)
> 
> -- 
> 2.38.0
> 

-- 
/Horatiu

Re: [PATCH net-next v2 2/2] net: micrel: Schedule work to read seconds for lan8841

[Richard Cochran]

On Tue, Jun 13, 2023 at 11:45:26AM +0200, Horatiu Vultur wrote:
> @@ -3840,6 +3847,12 @@ static void lan8841_ptp_enable_processing(struct kszphy_ptp_priv *ptp_priv,
>  			       LAN8841_PTP_INSERT_TS_32BIT,
>  			       LAN8841_PTP_INSERT_TS_EN |
>  			       LAN8841_PTP_INSERT_TS_32BIT);
> +
> +		/* Schedule the work to read the seconds, which will be used in
> +		 * the received timestamp
> +		 */
> +		schedule_delayed_work(&ptp_priv->seconds_work,
> +				      nsecs_to_jiffies(LAN8841_GET_SEC_LTC_DELAY));

Why not do this in the PTP kworker thread?

The thread's scheduling can be easily tuned with chrt to give it
appropriate priority, but work can't.

Also, If you have seconds thread, then you don't have to defer the
received frames.

Thanks,
Richard

Re: [PATCH net-next v2 1/2] net: micrel: Change to receive timestamp in the frame for lan8841

[Richard Cochran]

On Tue, Jun 13, 2023 at 11:45:25AM +0200, Horatiu Vultur wrote:
> Doing these changes to start to get the received timestamp in the
> reserved field of the header, will give a great CPU usage performance.
> Running ptp4l with logSyncInterval of -9 will give a ~50% CPU
> improvment.

One Sync every 1.95 milliseconds seems a bit excessive.
Just saying.

Thanks,
Richard

Re: [PATCH net-next v2 2/2] net: micrel: Schedule work to read seconds for lan8841

[Horatiu Vultur]

The 06/13/2023 21:49, Richard Cochran wrote:

Hi Richard,

> 
> On Tue, Jun 13, 2023 at 11:45:26AM +0200, Horatiu Vultur wrote:
> > @@ -3840,6 +3847,12 @@ static void lan8841_ptp_enable_processing(struct kszphy_ptp_priv *ptp_priv,
> >                              LAN8841_PTP_INSERT_TS_32BIT,
> >                              LAN8841_PTP_INSERT_TS_EN |
> >                              LAN8841_PTP_INSERT_TS_32BIT);
> > +
> > +             /* Schedule the work to read the seconds, which will be used in
> > +              * the received timestamp
> > +              */
> > +             schedule_delayed_work(&ptp_priv->seconds_work,
> > +                                   nsecs_to_jiffies(LAN8841_GET_SEC_LTC_DELAY));
> 
> Why not do this in the PTP kworker thread?

I presume you mean the work of reading the second part to be done in the
PTP kworker thread and not scheduling the seconds_work.
Because then it make sense to me and I think is a great idea.
> 
> The thread's scheduling can be easily tuned with chrt to give it
> appropriate priority, but work can't.

Nice, I didn't know about this.

> 
> Also, If you have seconds thread, then you don't have to defer the
> received frames.

Exactly, the PTP kworker thread will cache the seconds value while
lan8841_rxtstamp will read this value, so no need to defer these frames.

In this way I can get rid of seconds_work.

> 
> Thanks,
> Richard
> 

-- 
/Horatiu