netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH net-next v5 0/2] Broadcom PTP PHY support
@ 2022-05-18 22:39 Jonathan Lemon
  2022-05-18 22:39 ` [PATCH net-next v5 1/2] net: phy: broadcom: Add Broadcom PTP hooks to bcm-phy-lib Jonathan Lemon
                   ` (2 more replies)
  0 siblings, 3 replies; 13+ messages in thread
From: Jonathan Lemon @ 2022-05-18 22:39 UTC (permalink / raw)
  To: netdev
  Cc: f.fainelli, andrew, hkallweit1, linux, bcm-kernel-feedback-list,
	kernel-team, davem, kuba, pabeni, edumazet

This adds PTP support for the Broadcom PHY BCM54210E (and the
specific variant BCM54213PE that the rpi-5.15 branch uses).

This has only been tested on the RPI CM4, which has one port.

There are other Broadcom chips which may benefit from using the
same framework here, although with different register sets.

v4->v5:
 Reorder so bcm-phy-lib.h shows up first, fixing dependency issues.
 Use set_normalized_timespec for adjusting ns in adjtime
 Return upscaled config setting from hwtstamp

v3->v4:
 Squash bcm-phy-lib.h and broadcom.c changes into one patch
 Reorder so the main patch shows up first.

v2->v3:
 Rearrange patches so they apply in order
 Use ERR_CAST

v1->v2:
 Squash Kconfig into main patch
 Move config checks into bcm-phy-lib.h
 Fix delta_ns calculations in adjtime
 Uppercase mode selector macros
 Only use NSE_INIT when necessary
 Remove the inserted Broadcom RX timestamp from the PTP packet
 Add perout (chip generated) and fsync out (timer generated)
 Remove PHY_ID_BCM54213PE special casing (needed for rpi tree)

Thanks to Lasse Johnsen <lasse@timebeat.app> for pointing out
that the chip's periodic output generation isn't sync'd to any
time base.

Jonathan Lemon (2):
  net: phy: broadcom: Add Broadcom PTP hooks to bcm-phy-lib
  net: phy: broadcom: Add PTP support for some Broadcom PHYs.

 drivers/net/phy/Kconfig       |  10 +
 drivers/net/phy/Makefile      |   1 +
 drivers/net/phy/bcm-phy-lib.h |  14 +
 drivers/net/phy/bcm-phy-ptp.c | 868 ++++++++++++++++++++++++++++++++++
 drivers/net/phy/broadcom.c    |  23 +-
 5 files changed, 912 insertions(+), 4 deletions(-)
 create mode 100644 drivers/net/phy/bcm-phy-ptp.c

-- 
2.34.3

^ permalink raw reply	[flat|nested] 13+ messages in thread

* [PATCH net-next v5 1/2] net: phy: broadcom: Add Broadcom PTP hooks to bcm-phy-lib
  2022-05-18 22:39 [PATCH net-next v5 0/2] Broadcom PTP PHY support Jonathan Lemon
@ 2022-05-18 22:39 ` Jonathan Lemon
  2022-05-18 22:39 ` [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs Jonathan Lemon
  2022-05-28  3:27 ` [PATCH net-next v5 0/2] Broadcom PTP PHY support Richard Cochran
  2 siblings, 0 replies; 13+ messages in thread
From: Jonathan Lemon @ 2022-05-18 22:39 UTC (permalink / raw)
  To: netdev
  Cc: f.fainelli, andrew, hkallweit1, linux, bcm-kernel-feedback-list,
	kernel-team, davem, kuba, pabeni, edumazet

Add 'struct bcm_ptp_private' to bcm54xx_phy_priv which points to
an optional PTP structure attached to the PHY.  This is allocated
on probe if PHY PTP support is configured, and if the driver supports
PTP for the specified PHY.

Add the bcm_ptp_probe() and bcm_ptp_config_init() API functions
to the bcm-phy library.

Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
---
 drivers/net/phy/bcm-phy-lib.h | 14 ++++++++++++++
 drivers/net/phy/broadcom.c    | 23 +++++++++++++++++++----
 2 files changed, 33 insertions(+), 4 deletions(-)

diff --git a/drivers/net/phy/bcm-phy-lib.h b/drivers/net/phy/bcm-phy-lib.h
index c3842f87c33b..c2cc3c2766a1 100644
--- a/drivers/net/phy/bcm-phy-lib.h
+++ b/drivers/net/phy/bcm-phy-lib.h
@@ -87,4 +87,18 @@ int bcm_phy_cable_test_start_rdb(struct phy_device *phydev);
 int bcm_phy_cable_test_start(struct phy_device *phydev);
 int bcm_phy_cable_test_get_status(struct phy_device *phydev, bool *finished);
 
+#if IS_ENABLED(CONFIG_BCM_NET_PHYPTP)
+struct bcm_ptp_private *bcm_ptp_probe(struct phy_device *phydev);
+void bcm_ptp_config_init(struct phy_device *phydev);
+#else
+static inline struct bcm_ptp_private *bcm_ptp_probe(struct phy_device *phydev)
+{
+	return NULL;
+}
+
+static inline void bcm_ptp_config_init(struct phy_device *phydev)
+{
+}
+#endif
+
 #endif /* _LINUX_BCM_PHY_LIB_H */
diff --git a/drivers/net/phy/broadcom.c b/drivers/net/phy/broadcom.c
index e36809aa6d30..a7722599b5f9 100644
--- a/drivers/net/phy/broadcom.c
+++ b/drivers/net/phy/broadcom.c
@@ -27,6 +27,11 @@ MODULE_DESCRIPTION("Broadcom PHY driver");
 MODULE_AUTHOR("Maciej W. Rozycki");
 MODULE_LICENSE("GPL");
 
+struct bcm54xx_phy_priv {
+	u64	*stats;
+	struct bcm_ptp_private *ptp;
+};
+
 static int bcm54xx_config_clock_delay(struct phy_device *phydev)
 {
 	int rc, val;
@@ -313,6 +318,14 @@ static void bcm54xx_adjust_rxrefclk(struct phy_device *phydev)
 		bcm_phy_write_shadow(phydev, BCM54XX_SHD_APD, val);
 }
 
+static void bcm54xx_ptp_config_init(struct phy_device *phydev)
+{
+	struct bcm54xx_phy_priv *priv = phydev->priv;
+
+	if (priv->ptp)
+		bcm_ptp_config_init(phydev);
+}
+
 static int bcm54xx_config_init(struct phy_device *phydev)
 {
 	int reg, err, val;
@@ -390,6 +403,8 @@ static int bcm54xx_config_init(struct phy_device *phydev)
 		bcm_phy_write_exp(phydev, BCM_EXP_MULTICOLOR, val);
 	}
 
+	bcm54xx_ptp_config_init(phydev);
+
 	return 0;
 }
 
@@ -741,10 +756,6 @@ static irqreturn_t brcm_fet_handle_interrupt(struct phy_device *phydev)
 	return IRQ_HANDLED;
 }
 
-struct bcm54xx_phy_priv {
-	u64	*stats;
-};
-
 static int bcm54xx_phy_probe(struct phy_device *phydev)
 {
 	struct bcm54xx_phy_priv *priv;
@@ -761,6 +772,10 @@ static int bcm54xx_phy_probe(struct phy_device *phydev)
 	if (!priv->stats)
 		return -ENOMEM;
 
+	priv->ptp = bcm_ptp_probe(phydev);
+	if (IS_ERR(priv->ptp))
+		return PTR_ERR(priv->ptp);
+
 	return 0;
 }
 
-- 
2.34.3


^ permalink raw reply related	[flat|nested] 13+ messages in thread

* [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-18 22:39 [PATCH net-next v5 0/2] Broadcom PTP PHY support Jonathan Lemon
  2022-05-18 22:39 ` [PATCH net-next v5 1/2] net: phy: broadcom: Add Broadcom PTP hooks to bcm-phy-lib Jonathan Lemon
@ 2022-05-18 22:39 ` Jonathan Lemon
  2022-05-20 17:24   ` Florian Fainelli
  2022-05-29  0:34   ` Richard Cochran
  2022-05-28  3:27 ` [PATCH net-next v5 0/2] Broadcom PTP PHY support Richard Cochran
  2 siblings, 2 replies; 13+ messages in thread
From: Jonathan Lemon @ 2022-05-18 22:39 UTC (permalink / raw)
  To: netdev
  Cc: f.fainelli, andrew, hkallweit1, linux, bcm-kernel-feedback-list,
	kernel-team, davem, kuba, pabeni, edumazet

This adds PTP support for BCM54210E Broadcom PHYs, in particular,
the BCM54213PE, as used in the Rasperry PI CM4.  It has only been
tested on that hardware.

Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
---
 drivers/net/phy/Kconfig       |  10 +
 drivers/net/phy/Makefile      |   1 +
 drivers/net/phy/bcm-phy-ptp.c | 868 ++++++++++++++++++++++++++++++++++
 3 files changed, 879 insertions(+)
 create mode 100644 drivers/net/phy/bcm-phy-ptp.c

diff --git a/drivers/net/phy/Kconfig b/drivers/net/phy/Kconfig
index 9fee639ee5c8..68d3fc6beaae 100644
--- a/drivers/net/phy/Kconfig
+++ b/drivers/net/phy/Kconfig
@@ -108,6 +108,16 @@ config BROADCOM_PHY
 	  Currently supports the BCM5411, BCM5421, BCM5461, BCM54616S, BCM5464,
 	  BCM5481, BCM54810 and BCM5482 PHYs.
 
+config BCM_NET_PHYPTP
+	tristate "Broadcom PHY PTP support"
+	depends on NETWORK_PHY_TIMESTAMPING
+	depends on PHYLIB
+	depends on PTP_1588_CLOCK
+	depends on BROADCOM_PHY
+	depends on NET_PTP_CLASSIFY
+	help
+	  Supports PTP timestamping for certain Broadcom PHYs.
+
 config BCM54140_PHY
 	tristate "Broadcom BCM54140 PHY"
 	depends on PHYLIB
diff --git a/drivers/net/phy/Makefile b/drivers/net/phy/Makefile
index b12b1d86fc99..f7138d3c896b 100644
--- a/drivers/net/phy/Makefile
+++ b/drivers/net/phy/Makefile
@@ -47,6 +47,7 @@ obj-$(CONFIG_BCM84881_PHY)	+= bcm84881.o
 obj-$(CONFIG_BCM87XX_PHY)	+= bcm87xx.o
 obj-$(CONFIG_BCM_CYGNUS_PHY)	+= bcm-cygnus.o
 obj-$(CONFIG_BCM_NET_PHYLIB)	+= bcm-phy-lib.o
+obj-$(CONFIG_BCM_NET_PHYPTP)	+= bcm-phy-ptp.o
 obj-$(CONFIG_BROADCOM_PHY)	+= broadcom.o
 obj-$(CONFIG_CICADA_PHY)	+= cicada.o
 obj-$(CONFIG_CORTINA_PHY)	+= cortina.o
diff --git a/drivers/net/phy/bcm-phy-ptp.c b/drivers/net/phy/bcm-phy-ptp.c
new file mode 100644
index 000000000000..62d48c58370a
--- /dev/null
+++ b/drivers/net/phy/bcm-phy-ptp.c
@@ -0,0 +1,868 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2022 Meta Platforms Inc.
+ * Copyright (C) 2022 Jonathan Lemon <jonathan.lemon@gmail.com>
+ */
+
+#include <asm/unaligned.h>
+#include <linux/mii.h>
+#include <linux/phy.h>
+#include <linux/ptp_classify.h>
+#include <linux/ptp_clock_kernel.h>
+#include <linux/net_tstamp.h>
+#include <linux/netdevice.h>
+#include <linux/workqueue.h>
+
+#include "bcm-phy-lib.h"
+
+/* IEEE 1588 Expansion registers */
+#define SLICE_CTRL		0x0810
+#define  SLICE_TX_EN			BIT(0)
+#define  SLICE_RX_EN			BIT(8)
+#define TX_EVENT_MODE		0x0811
+#define  MODE_TX_UPDATE_CF		BIT(0)
+#define  MODE_TX_REPLACE_TS_CF		BIT(1)
+#define  MODE_TX_REPLACE_TS		GENMASK(1, 0)
+#define RX_EVENT_MODE		0x0819
+#define  MODE_RX_UPDATE_CF		BIT(0)
+#define  MODE_RX_INSERT_TS_48		BIT(1)
+#define  MODE_RX_INSERT_TS_64		GENMASK(1, 0)
+
+#define MODE_EVT_SHIFT_SYNC		0
+#define MODE_EVT_SHIFT_DELAY_REQ	2
+#define MODE_EVT_SHIFT_PDELAY_REQ	4
+#define MODE_EVT_SHIFT_PDELAY_RESP	6
+
+#define MODE_SEL_SHIFT_PORT		0
+#define MODE_SEL_SHIFT_CPU		8
+
+#define RX_MODE_SEL(sel, evt, act) \
+	(((MODE_RX_##act) << (MODE_EVT_SHIFT_##evt)) << (MODE_SEL_SHIFT_##sel))
+
+#define TX_MODE_SEL(sel, evt, act) \
+	(((MODE_TX_##act) << (MODE_EVT_SHIFT_##evt)) << (MODE_SEL_SHIFT_##sel))
+
+/* needs global TS capture first */
+#define TX_TS_CAPTURE		0x0821
+#define  TX_TS_CAP_EN			BIT(0)
+#define RX_TS_CAPTURE		0x0822
+#define  RX_TS_CAP_EN			BIT(0)
+
+#define TIME_CODE_0		0x0854
+#define TIME_CODE_1		0x0855
+#define TIME_CODE_2		0x0856
+#define TIME_CODE_3		0x0857
+#define TIME_CODE_4		0x0858
+
+#define DPLL_SELECT		0x085b
+#define  DPLL_HB_MODE2			BIT(6)
+
+#define SHADOW_CTRL		0x085c
+#define SHADOW_LOAD		0x085d
+#define  TIME_CODE_LOAD			BIT(10)
+#define  SYNC_OUT_LOAD			BIT(9)
+#define  NCO_TIME_LOAD			BIT(7)
+#define  FREQ_LOAD			BIT(6)
+#define INTR_MASK		0x085e
+#define INTR_STATUS		0x085f
+#define  INTC_FSYNC			BIT(0)
+#define  INTC_SOP			BIT(1)
+
+#define NCO_FREQ_LSB		0x0873
+#define NCO_FREQ_MSB		0x0874
+
+#define NCO_TIME_0		0x0875
+#define NCO_TIME_1		0x0876
+#define NCO_TIME_2_CTRL		0x0877
+#define  FREQ_MDIO_SEL			BIT(14)
+
+#define SYNC_OUT_0		0x0878
+#define SYNC_OUT_1		0x0879
+#define SYNC_OUT_2		0x087a
+
+#define SYNOUT_TS_0		0x087c
+#define SYNOUT_TS_1		0x087d
+#define SYNOUT_TS_2		0x087e
+
+#define NSE_CTRL		0x087f
+#define  NSE_GMODE_EN			GENMASK(15, 14)
+#define  NSE_CAPTURE_EN			BIT(13)
+#define  NSE_INIT			BIT(12)
+#define  NSE_CPU_FRAMESYNC		BIT(5)
+#define  NSE_FRAMESYNC_MASK		GENMASK(5, 2)
+#define  NSE_PEROUT_EN			BIT(1)
+#define  NSE_ONESHOT_EN			BIT(0)
+#define  NSE_SYNC_OUT_MASK		GENMASK(1, 0)
+
+#define TS_READ_CTRL		0x0885
+#define  TS_READ_START			BIT(0)
+#define  TS_READ_END			BIT(1)
+
+#define HB_REG_0		0x0886
+#define HB_REG_1		0x0887
+#define HB_REG_2		0x0888
+#define HB_REG_3		0x08ec
+#define HB_REG_4		0x08ed
+#define HB_STAT_CTRL		0x088e
+#define  HB_READ_START			BIT(10)
+#define  HB_READ_END			BIT(11)
+#define  HB_READ_MASK			GENMASK(11, 10)
+
+#define TS_REG_0		0x0889
+#define TS_REG_1		0x088a
+#define TS_REG_2		0x088b
+#define TS_REG_3		0x08c4
+
+#define TS_INFO_0		0x088c
+#define TS_INFO_1		0x088d
+
+#define TIMECODE_CTRL		0x08c3
+#define  TX_TIMECODE_SEL		GENMASK(7, 0)
+#define  RX_TIMECODE_SEL		GENMASK(15, 8)
+
+#define TIME_SYNC		0x0ff5
+#define  TIME_SYNC_EN			BIT(0)
+
+struct bcm_ptp_private {
+	struct phy_device *phydev;
+	struct mii_timestamper mii_ts;
+	struct ptp_clock *ptp_clock;
+	struct ptp_clock_info ptp_info;
+	struct mutex mutex;
+	struct sk_buff_head tx_queue;
+	int tx_type;
+	bool hwts_rx;
+	u16 nse_ctrl;
+
+	struct delayed_work out_work;
+	bool fsync_out;
+};
+
+struct bcm_ptp_skb_cb {
+	unsigned long timeout;
+	u16 seq_id;
+	u8 msgtype;
+	bool discard;
+};
+
+struct bcm_ptp_capture {
+	ktime_t	hwtstamp;
+	u16 seq_id;
+	u8 msgtype;
+	bool tx_dir;
+};
+
+#define BCM_SKB_CB(skb)		((struct bcm_ptp_skb_cb *)(skb)->cb)
+#define SKB_TS_TIMEOUT		10			/* jiffies */
+
+#define BCM_MAX_PULSE_8NS	((1U << 9) - 1)
+#define BCM_MAX_PERIOD_8NS	((1U << 30) - 1)
+
+#define BRCM_PHY_MODEL(phydev) \
+	((phydev)->drv->phy_id & (phydev)->drv->phy_id_mask)
+
+static struct bcm_ptp_private *mii2priv(struct mii_timestamper *mii_ts)
+{
+	return container_of(mii_ts, struct bcm_ptp_private, mii_ts);
+}
+
+static struct bcm_ptp_private *ptp2priv(struct ptp_clock_info *info)
+{
+	return container_of(info, struct bcm_ptp_private, ptp_info);
+}
+
+static int bcm_ptp_framesync(struct phy_device *phydev,
+			     struct ptp_system_timestamp *sts,
+			     u16 ctrl)
+{
+	u16 reg;
+	int i;
+
+	/* prep for framesync */
+	bcm_phy_write_exp(phydev, NSE_CTRL, ctrl);
+
+	ptp_read_system_prets(sts);
+
+	/* trigger framesync */
+	bcm_phy_write_exp(phydev, NSE_CTRL, ctrl | NSE_CPU_FRAMESYNC);
+
+	ptp_read_system_postts(sts);
+
+	if ((ctrl & NSE_CAPTURE_EN) == 0)
+		return 0;
+
+	/* poll for FSYNC interrupt from TS capture */
+	for (i = 0; i < 10; i++) {
+		reg = bcm_phy_read_exp(phydev, INTR_STATUS);
+		if (reg & INTC_FSYNC)
+			break;
+	}
+
+	return reg & INTC_FSYNC ? 0 : -ETIMEDOUT;
+}
+
+static int bcm_ptp_gettime_locked(struct bcm_ptp_private *priv,
+				  struct timespec64 *ts,
+				  struct ptp_system_timestamp *sts)
+{
+	struct phy_device *phydev = priv->phydev;
+	u16 hb[4];
+	int err;
+
+	err = bcm_ptp_framesync(phydev, sts, priv->nse_ctrl | NSE_CAPTURE_EN);
+	if (err)
+		return err;
+
+	bcm_phy_write_exp(phydev, HB_STAT_CTRL, HB_READ_START);
+
+	hb[0] = bcm_phy_read_exp(phydev, HB_REG_0);
+	hb[1] = bcm_phy_read_exp(phydev, HB_REG_1);
+	hb[2] = bcm_phy_read_exp(phydev, HB_REG_2);
+	hb[3] = bcm_phy_read_exp(phydev, HB_REG_3);
+
+	bcm_phy_write_exp(phydev, HB_STAT_CTRL, HB_READ_END);
+	bcm_phy_write_exp(phydev, HB_STAT_CTRL, 0);
+
+	ts->tv_sec = (hb[3] << 16) | hb[2];
+	ts->tv_nsec = (hb[1] << 16) | hb[0];
+
+	return 0;
+}
+
+static int bcm_ptp_gettimex(struct ptp_clock_info *info,
+			    struct timespec64 *ts,
+			    struct ptp_system_timestamp *sts)
+{
+	struct bcm_ptp_private *priv = ptp2priv(info);
+	int err;
+
+	mutex_lock(&priv->mutex);
+	err = bcm_ptp_gettime_locked(priv, ts, sts);
+	mutex_unlock(&priv->mutex);
+
+	return err;
+}
+
+static int bcm_ptp_settime_locked(struct bcm_ptp_private *priv,
+				  const struct timespec64 *ts)
+{
+	struct phy_device *phydev = priv->phydev;
+	u64 ns;
+
+	/* set up time code */
+	bcm_phy_write_exp(phydev, TIME_CODE_0, ts->tv_nsec);
+	bcm_phy_write_exp(phydev, TIME_CODE_1, ts->tv_nsec >> 16);
+	bcm_phy_write_exp(phydev, TIME_CODE_2, ts->tv_sec);
+	bcm_phy_write_exp(phydev, TIME_CODE_3, ts->tv_sec >> 16);
+	bcm_phy_write_exp(phydev, TIME_CODE_4, ts->tv_sec >> 32);
+
+	/* set NCO counter to match */
+	ns = timespec64_to_ns(ts);
+	bcm_phy_write_exp(phydev, NCO_TIME_0, ns >> 4);
+	bcm_phy_write_exp(phydev, NCO_TIME_1, ns >> 20);
+	bcm_phy_write_exp(phydev, NCO_TIME_2_CTRL, (ns >> 36) & 0xfff);
+
+	/* set up load on next frame sync (auto-clears due to NSE_INIT) */
+	bcm_phy_write_exp(phydev, SHADOW_LOAD, TIME_CODE_LOAD | NCO_TIME_LOAD);
+
+	/* must have NSE_INIT in order to write time code */
+	return bcm_ptp_framesync(phydev, NULL, priv->nse_ctrl | NSE_INIT);
+}
+
+static int bcm_ptp_settime(struct ptp_clock_info *info,
+			   const struct timespec64 *ts)
+{
+	struct bcm_ptp_private *priv = ptp2priv(info);
+	int err;
+
+	mutex_lock(&priv->mutex);
+	err = bcm_ptp_settime_locked(priv, ts);
+	mutex_unlock(&priv->mutex);
+
+	return err;
+}
+
+static int bcm_ptp_adjtime_locked(struct bcm_ptp_private *priv,
+				  s64 delta_ns)
+{
+	struct timespec64 ts;
+	int err;
+
+	err = bcm_ptp_gettime_locked(priv, &ts, NULL);
+	if (!err) {
+		set_normalized_timespec64(&ts, ts.tv_sec,
+					  ts.tv_nsec + delta_ns);
+		err = bcm_ptp_settime_locked(priv, &ts);
+	}
+	return err;
+}
+
+static int bcm_ptp_adjtime(struct ptp_clock_info *info, s64 delta_ns)
+{
+	struct bcm_ptp_private *priv = ptp2priv(info);
+	int err;
+
+	mutex_lock(&priv->mutex);
+	err = bcm_ptp_adjtime_locked(priv, delta_ns);
+	mutex_unlock(&priv->mutex);
+
+	return err;
+}
+
+/* A 125Mhz clock should adjust 8ns per pulse.
+ * The frequency adjustment base is 0x8000 0000, or 8*2^28.
+ *
+ * Frequency adjustment is
+ * adj = scaled_ppm * 8*2^28 / (10^6 * 2^16)
+ *   which simplifies to:
+ * adj = scaled_ppm * 2^9 / 5^6
+ */
+static int bcm_ptp_adjfine(struct ptp_clock_info *info, long scaled_ppm)
+{
+	struct bcm_ptp_private *priv = ptp2priv(info);
+	int neg_adj = 0;
+	u32 diff, freq;
+	u64 adj;
+
+	if (scaled_ppm < 0) {
+		neg_adj = 1;
+		scaled_ppm = -scaled_ppm;
+	}
+
+	adj = scaled_ppm << 9;
+	diff = div_u64(adj, 15625);
+	freq = (8 << 28) + (neg_adj ? -diff : diff);
+
+	mutex_lock(&priv->mutex);
+
+	bcm_phy_write_exp(priv->phydev, NCO_FREQ_LSB, freq);
+	bcm_phy_write_exp(priv->phydev, NCO_FREQ_MSB, freq >> 16);
+
+	bcm_phy_write_exp(priv->phydev, NCO_TIME_2_CTRL, FREQ_MDIO_SEL);
+
+	/* load on next framesync */
+	bcm_phy_write_exp(priv->phydev, SHADOW_LOAD, FREQ_LOAD);
+
+	bcm_ptp_framesync(priv->phydev, NULL, priv->nse_ctrl);
+
+	/* clear load */
+	bcm_phy_write_exp(priv->phydev, SHADOW_LOAD, 0);
+
+	mutex_unlock(&priv->mutex);
+
+	return 0;
+}
+
+/* The output pulse and period are stable, but the signal is not
+ * synchronized to anything.
+ */
+static int bcm_ptp_perout_locked(struct bcm_ptp_private *priv,
+				 struct ptp_perout_request *req, int on)
+{
+	struct phy_device *phydev = priv->phydev;
+	u64 period, pulse;
+	u16 val;
+
+	if (!on) {
+		priv->nse_ctrl &= ~NSE_SYNC_OUT_MASK;
+		bcm_phy_write_exp(phydev, NSE_CTRL, priv->nse_ctrl);
+		return 0;
+	}
+
+	if (req->flags & PTP_PEROUT_PHASE)
+		return -EOPNOTSUPP;
+
+	period = ktime_to_ns(ktime_set(req->period.sec, req->period.nsec));
+	if (req->flags & PTP_PEROUT_DUTY_CYCLE)
+		pulse = ktime_to_ns(ktime_set(req->on.sec, req->on.nsec));
+	else
+		pulse = min(period / 2, (u64)BCM_MAX_PULSE_8NS << 3);
+
+	/* convert to 8ns units */
+	pulse >>= 3;
+	period >>= 3;
+
+	if (!pulse || !period)
+		return -EINVAL;
+
+	if (pulse > period)
+		return -EINVAL;
+
+	if (pulse > BCM_MAX_PULSE_8NS || period > BCM_MAX_PERIOD_8NS)
+		return -EINVAL;
+
+	bcm_phy_write_exp(phydev, SYNC_OUT_0, period);
+
+	val = ((pulse & 0x3) << 14) | ((period >> 16) & 0x3fff);
+	bcm_phy_write_exp(phydev, SYNC_OUT_1, val);
+
+	val = ((pulse >> 2) & 0x7f) | (pulse << 7);
+	bcm_phy_write_exp(phydev, SYNC_OUT_2, val);
+
+	/* load values on next framesync */
+	bcm_phy_write_exp(phydev, SHADOW_LOAD, SYNC_OUT_LOAD);
+
+	priv->nse_ctrl |= NSE_PEROUT_EN;
+	return bcm_ptp_framesync(phydev, NULL, priv->nse_ctrl | NSE_INIT);
+}
+
+static void bcm_ptp_fsync_work(struct work_struct *out_work)
+{
+	struct bcm_ptp_private *priv =
+		container_of(out_work, struct bcm_ptp_private, out_work.work);
+	struct phy_device *phydev = priv->phydev;
+	struct timespec64 ts;
+	u64 ns, next;
+
+	mutex_lock(&priv->mutex);
+
+	/* no longer running */
+	if (!priv->fsync_out) {
+		mutex_unlock(&priv->mutex);
+		return;
+	}
+
+	bcm_ptp_gettime_locked(priv, &ts, NULL);
+
+	/* this is 1PPS only */
+	next = NSEC_PER_SEC - ts.tv_nsec;
+	ts.tv_sec += next < NSEC_PER_MSEC ? 2 : 1;
+	ts.tv_nsec = 0;
+
+	ns = timespec64_to_ns(&ts);
+
+	bcm_phy_write_exp(phydev, SYNOUT_TS_0, ns & 0xfff0);
+	bcm_phy_write_exp(phydev, SYNOUT_TS_1, ns >> 16);
+	bcm_phy_write_exp(phydev, SYNOUT_TS_2, ns >> 32);
+
+	/* load values on next framesync */
+	bcm_phy_write_exp(phydev, SHADOW_LOAD, SYNC_OUT_LOAD);
+
+	/* disable existing pulse - oneshot needs toggle. */
+	if (priv->nse_ctrl & NSE_SYNC_OUT_MASK) {
+		priv->nse_ctrl &= ~NSE_SYNC_OUT_MASK;
+		bcm_phy_write_exp(phydev, NSE_CTRL, priv->nse_ctrl);
+	}
+
+	priv->nse_ctrl |= NSE_ONESHOT_EN;
+	bcm_ptp_framesync(phydev, NULL, priv->nse_ctrl | NSE_INIT);
+
+	mutex_unlock(&priv->mutex);
+
+	next = next + NSEC_PER_MSEC;
+	schedule_delayed_work(&priv->out_work, nsecs_to_jiffies(next));
+}
+
+static int bcm_ptp_fsync_locked(struct bcm_ptp_private *priv,
+				struct ptp_perout_request *req, int on)
+{
+	struct phy_device *phydev = priv->phydev;
+	u64 period, pulse;
+	u16 val;
+
+	if (!on) {
+		priv->fsync_out = false;
+		priv->nse_ctrl &= ~NSE_SYNC_OUT_MASK;
+		bcm_phy_write_exp(phydev, NSE_CTRL, priv->nse_ctrl);
+		return 0;
+	}
+
+	/* 1PPS for now */
+	if (req->period.sec != 1 || req->period.nsec != 0)
+		return -EINVAL;
+
+	if (req->flags & PTP_PEROUT_PHASE)
+		return -EOPNOTSUPP;
+
+	period = ktime_to_ns(ktime_set(req->period.sec, req->period.nsec));
+	if (req->flags & PTP_PEROUT_DUTY_CYCLE)
+		pulse = ktime_to_ns(ktime_set(req->on.sec, req->on.nsec));
+	else
+		pulse = min(period / 2, (u64)BCM_MAX_PULSE_8NS << 3);
+
+	/* convert to 8ns units */
+	pulse >>= 3;
+	period >>= 3;
+
+	if (!pulse || !period)
+		return -EINVAL;
+
+	if (pulse > period)
+		return -EINVAL;
+
+	if (pulse > BCM_MAX_PULSE_8NS || period > BCM_MAX_PERIOD_8NS)
+		return -EINVAL;
+
+	bcm_phy_write_exp(phydev, SYNC_OUT_0, period);
+
+	val = ((pulse & 0x3) << 14) | ((period >> 16) & 0x3fff);
+	bcm_phy_write_exp(phydev, SYNC_OUT_1, val);
+
+	val = ((pulse >> 2) & 0x7f) | (pulse << 7);
+	bcm_phy_write_exp(phydev, SYNC_OUT_2, val);
+
+	priv->fsync_out = true;
+	schedule_delayed_work(&priv->out_work, 0);
+
+	return 0;
+}
+
+static bool bcm_ptp_rxtstamp(struct mii_timestamper *mii_ts,
+			     struct sk_buff *skb, int type)
+{
+	struct bcm_ptp_private *priv = mii2priv(mii_ts);
+	struct skb_shared_hwtstamps *hwts;
+	struct ptp_header *header;
+	u32 sec, nsec;
+	u8 *data;
+	int off;
+
+	if (!priv->hwts_rx)
+		return false;
+
+	header = ptp_parse_header(skb, type);
+	if (!header)
+		return false;
+
+	data = (u8 *)(header + 1);
+	sec = get_unaligned_be32(data);
+	nsec = get_unaligned_be32(data + 4);
+
+	hwts = skb_hwtstamps(skb);
+	hwts->hwtstamp = ktime_set(sec, nsec);
+
+	off = data - skb->data + 8;
+	if (off < skb->len) {
+		memmove(data, data + 8, skb->len - off);
+		__pskb_trim(skb, skb->len - 8);
+	}
+
+	return false;
+}
+
+static bool bcm_ptp_get_tstamp(struct bcm_ptp_private *priv,
+			       struct bcm_ptp_capture *capts)
+{
+	struct phy_device *phydev = priv->phydev;
+	u16 ts[4], reg;
+	u32 sec, nsec;
+
+	mutex_lock(&priv->mutex);
+
+	reg = bcm_phy_read_exp(phydev, INTR_STATUS);
+	if ((reg & INTC_SOP) == 0) {
+		mutex_unlock(&priv->mutex);
+		return false;
+	}
+
+	bcm_phy_write_exp(phydev, TS_READ_CTRL, TS_READ_START);
+
+	ts[0] = bcm_phy_read_exp(phydev, TS_REG_0);
+	ts[1] = bcm_phy_read_exp(phydev, TS_REG_1);
+	ts[2] = bcm_phy_read_exp(phydev, TS_REG_2);
+	ts[3] = bcm_phy_read_exp(phydev, TS_REG_3);
+
+	/* not in be32 format for some reason */
+	capts->seq_id = bcm_phy_read_exp(priv->phydev, TS_INFO_0);
+
+	reg = bcm_phy_read_exp(phydev, TS_INFO_1);
+	capts->msgtype = reg >> 12;
+	capts->tx_dir = !!(reg & BIT(11));
+
+	bcm_phy_write_exp(phydev, TS_READ_CTRL, TS_READ_END);
+	bcm_phy_write_exp(phydev, TS_READ_CTRL, 0);
+
+	mutex_unlock(&priv->mutex);
+
+	sec = (ts[3] << 16) | ts[2];
+	nsec = (ts[1] << 16) | ts[0];
+	capts->hwtstamp = ktime_set(sec, nsec);
+
+	return true;
+}
+
+static void bcm_ptp_match_tstamp(struct bcm_ptp_private *priv,
+				 struct bcm_ptp_capture *capts)
+{
+	struct skb_shared_hwtstamps hwts;
+	struct sk_buff *skb, *ts_skb;
+	unsigned long flags;
+	bool first = false;
+
+	ts_skb = NULL;
+	spin_lock_irqsave(&priv->tx_queue.lock, flags);
+	skb_queue_walk(&priv->tx_queue, skb) {
+		if (BCM_SKB_CB(skb)->seq_id == capts->seq_id &&
+		    BCM_SKB_CB(skb)->msgtype == capts->msgtype) {
+			first = skb_queue_is_first(&priv->tx_queue, skb);
+			__skb_unlink(skb, &priv->tx_queue);
+			ts_skb = skb;
+			break;
+		}
+	}
+	spin_unlock_irqrestore(&priv->tx_queue.lock, flags);
+
+	/* TX captures one-step packets, discard them if needed. */
+	if (ts_skb) {
+		if (BCM_SKB_CB(ts_skb)->discard) {
+			kfree_skb(ts_skb);
+		} else {
+			memset(&hwts, 0, sizeof(hwts));
+			hwts.hwtstamp = capts->hwtstamp;
+			skb_complete_tx_timestamp(ts_skb, &hwts);
+		}
+	}
+
+	/* not first match, try and expire entries */
+	if (!first) {
+		while ((skb = skb_dequeue(&priv->tx_queue))) {
+			if (!time_after(jiffies, BCM_SKB_CB(skb)->timeout)) {
+				skb_queue_head(&priv->tx_queue, skb);
+				break;
+			}
+			kfree_skb(skb);
+		}
+	}
+}
+
+static long bcm_ptp_do_aux_work(struct ptp_clock_info *info)
+{
+	struct bcm_ptp_private *priv = ptp2priv(info);
+	struct bcm_ptp_capture capts;
+	bool reschedule = false;
+
+	while (!skb_queue_empty_lockless(&priv->tx_queue)) {
+		if (!bcm_ptp_get_tstamp(priv, &capts)) {
+			reschedule = true;
+			break;
+		}
+		bcm_ptp_match_tstamp(priv, &capts);
+	}
+
+	return reschedule ? 1 : -1;
+}
+
+static int bcm_ptp_enable(struct ptp_clock_info *info,
+			  struct ptp_clock_request *rq, int on)
+{
+	struct bcm_ptp_private *priv = ptp2priv(info);
+	int err = 0;
+
+	switch (rq->type) {
+	case PTP_CLK_REQ_PEROUT:
+		mutex_lock(&priv->mutex);
+		if (rq->perout.index == 0)
+			err = bcm_ptp_perout_locked(priv, &rq->perout, on);
+		else
+			err = bcm_ptp_fsync_locked(priv, &rq->perout, on);
+		mutex_unlock(&priv->mutex);
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+	return err;
+}
+
+static const struct ptp_clock_info bcm_ptp_clock_info = {
+	.owner		= THIS_MODULE,
+	.name		= KBUILD_MODNAME,
+	.max_adj	= 100000000,
+	.gettimex64	= bcm_ptp_gettimex,
+	.settime64	= bcm_ptp_settime,
+	.adjtime	= bcm_ptp_adjtime,
+	.adjfine	= bcm_ptp_adjfine,
+	.enable		= bcm_ptp_enable,
+	.do_aux_work	= bcm_ptp_do_aux_work,
+	.n_per_out	= 2,
+};
+
+static void bcm_ptp_txtstamp(struct mii_timestamper *mii_ts,
+			     struct sk_buff *skb, int type)
+{
+	struct bcm_ptp_private *priv = mii2priv(mii_ts);
+	struct ptp_header *hdr;
+	bool discard = false;
+	int msgtype;
+
+	hdr = ptp_parse_header(skb, type);
+	if (!hdr)
+		goto out;
+	msgtype = ptp_get_msgtype(hdr, type);
+
+	switch (priv->tx_type) {
+	case HWTSTAMP_TX_ONESTEP_P2P:
+		if (msgtype == PTP_MSGTYPE_PDELAY_RESP)
+			discard = true;
+		fallthrough;
+	case HWTSTAMP_TX_ONESTEP_SYNC:
+		if (msgtype == PTP_MSGTYPE_SYNC)
+			discard = true;
+		fallthrough;
+	case HWTSTAMP_TX_ON:
+		BCM_SKB_CB(skb)->timeout = jiffies + SKB_TS_TIMEOUT;
+		BCM_SKB_CB(skb)->seq_id = be16_to_cpu(hdr->sequence_id);
+		BCM_SKB_CB(skb)->msgtype = msgtype;
+		BCM_SKB_CB(skb)->discard = discard;
+		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
+		skb_queue_tail(&priv->tx_queue, skb);
+		ptp_schedule_worker(priv->ptp_clock, 0);
+		return;
+	default:
+		break;
+	}
+
+out:
+	kfree_skb(skb);
+}
+
+static int bcm_ptp_hwtstamp(struct mii_timestamper *mii_ts,
+			    struct ifreq *ifr)
+{
+	struct bcm_ptp_private *priv = mii2priv(mii_ts);
+	struct hwtstamp_config cfg;
+	u16 mode, ctrl;
+
+	if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
+		return -EFAULT;
+
+	switch (cfg.rx_filter) {
+	case HWTSTAMP_FILTER_NONE:
+		priv->hwts_rx = false;
+		break;
+	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
+	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
+	case HWTSTAMP_FILTER_PTP_V2_EVENT:
+	case HWTSTAMP_FILTER_PTP_V2_SYNC:
+	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
+		cfg.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
+		priv->hwts_rx = true;
+		break;
+	default:
+		return -ERANGE;
+	}
+
+	priv->tx_type = cfg.tx_type;
+
+	ctrl  = priv->hwts_rx ? SLICE_RX_EN : 0;
+	ctrl |= priv->tx_type != HWTSTAMP_TX_OFF ? SLICE_TX_EN : 0;
+
+	mode = TX_MODE_SEL(PORT, SYNC, REPLACE_TS) |
+	       TX_MODE_SEL(PORT, DELAY_REQ, REPLACE_TS) |
+	       TX_MODE_SEL(PORT, PDELAY_REQ, REPLACE_TS) |
+	       TX_MODE_SEL(PORT, PDELAY_RESP, REPLACE_TS);
+
+	bcm_phy_write_exp(priv->phydev, TX_EVENT_MODE, mode);
+
+	mode = RX_MODE_SEL(PORT, SYNC, INSERT_TS_64) |
+	       RX_MODE_SEL(PORT, DELAY_REQ, INSERT_TS_64) |
+	       RX_MODE_SEL(PORT, PDELAY_REQ, INSERT_TS_64) |
+	       RX_MODE_SEL(PORT, PDELAY_RESP, INSERT_TS_64);
+
+	bcm_phy_write_exp(priv->phydev, RX_EVENT_MODE, mode);
+
+	bcm_phy_write_exp(priv->phydev, SLICE_CTRL, ctrl);
+
+	if (ctrl & SLICE_TX_EN)
+		bcm_phy_write_exp(priv->phydev, TX_TS_CAPTURE, TX_TS_CAP_EN);
+	else
+		ptp_cancel_worker_sync(priv->ptp_clock);
+
+	/* purge existing data */
+	skb_queue_purge(&priv->tx_queue);
+
+	return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ? -EFAULT : 0;
+}
+
+static int bcm_ptp_ts_info(struct mii_timestamper *mii_ts,
+			   struct ethtool_ts_info *ts_info)
+{
+	struct bcm_ptp_private *priv = mii2priv(mii_ts);
+
+	ts_info->phc_index = ptp_clock_index(priv->ptp_clock);
+	ts_info->so_timestamping =
+		SOF_TIMESTAMPING_TX_HARDWARE |
+		SOF_TIMESTAMPING_RX_HARDWARE |
+		SOF_TIMESTAMPING_RAW_HARDWARE;
+	ts_info->tx_types =
+		BIT(HWTSTAMP_TX_ON) |
+		BIT(HWTSTAMP_TX_OFF) |
+		BIT(HWTSTAMP_TX_ONESTEP_SYNC) |
+		BIT(HWTSTAMP_TX_ONESTEP_P2P);
+	ts_info->rx_filters =
+		BIT(HWTSTAMP_FILTER_NONE) |
+		BIT(HWTSTAMP_FILTER_PTP_V2_EVENT);
+
+	return 0;
+}
+
+void bcm_ptp_config_init(struct phy_device *phydev)
+{
+	/* init network sync engine */
+	bcm_phy_write_exp(phydev, NSE_CTRL, NSE_GMODE_EN | NSE_INIT);
+
+	/* enable time sync (TX/RX SOP capture) */
+	bcm_phy_write_exp(phydev, TIME_SYNC, TIME_SYNC_EN);
+
+	/* use sec.nsec heartbeat capture */
+	bcm_phy_write_exp(phydev, DPLL_SELECT, DPLL_HB_MODE2);
+
+	/* use 64 bit timecode for TX */
+	bcm_phy_write_exp(phydev, TIMECODE_CTRL, TX_TIMECODE_SEL);
+
+	/* always allow FREQ_LOAD on framesync */
+	bcm_phy_write_exp(phydev, SHADOW_CTRL, FREQ_LOAD);
+}
+EXPORT_SYMBOL_GPL(bcm_ptp_config_init);
+
+static void bcm_ptp_init(struct bcm_ptp_private *priv)
+{
+	priv->nse_ctrl = NSE_GMODE_EN;
+
+	mutex_init(&priv->mutex);
+	skb_queue_head_init(&priv->tx_queue);
+
+	priv->mii_ts.rxtstamp = bcm_ptp_rxtstamp;
+	priv->mii_ts.txtstamp = bcm_ptp_txtstamp;
+	priv->mii_ts.hwtstamp = bcm_ptp_hwtstamp;
+	priv->mii_ts.ts_info = bcm_ptp_ts_info;
+
+	priv->phydev->mii_ts = &priv->mii_ts;
+
+	INIT_DELAYED_WORK(&priv->out_work, bcm_ptp_fsync_work);
+}
+
+struct bcm_ptp_private *bcm_ptp_probe(struct phy_device *phydev)
+{
+	struct bcm_ptp_private *priv;
+	struct ptp_clock *clock;
+
+	switch (BRCM_PHY_MODEL(phydev)) {
+	case PHY_ID_BCM54210E:
+		break;
+	default:
+		return NULL;
+	}
+
+	priv = devm_kzalloc(&phydev->mdio.dev, sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return ERR_PTR(-ENOMEM);
+
+	priv->ptp_info = bcm_ptp_clock_info;
+
+	clock = ptp_clock_register(&priv->ptp_info, &phydev->mdio.dev);
+	if (IS_ERR(clock))
+		return ERR_CAST(clock);
+	priv->ptp_clock = clock;
+
+	priv->phydev = phydev;
+	bcm_ptp_init(priv);
+
+	return priv;
+}
+EXPORT_SYMBOL_GPL(bcm_ptp_probe);
+
+MODULE_LICENSE("GPL");
-- 
2.34.3


^ permalink raw reply related	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-18 22:39 ` [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs Jonathan Lemon
@ 2022-05-20 17:24   ` Florian Fainelli
  2022-05-21  2:04     ` Jonathan Lemon
  2022-05-29  0:34   ` Richard Cochran
  1 sibling, 1 reply; 13+ messages in thread
From: Florian Fainelli @ 2022-05-20 17:24 UTC (permalink / raw)
  To: Jonathan Lemon, netdev
  Cc: f.fainelli, andrew, hkallweit1, linux, bcm-kernel-feedback-list,
	kernel-team, davem, kuba, pabeni, edumazet



On 5/18/2022 3:39 PM, Jonathan Lemon wrote:
> This adds PTP support for BCM54210E Broadcom PHYs, in particular,
> the BCM54213PE, as used in the Rasperry PI CM4.  It has only been
> tested on that hardware.
> 
> Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
> ---
[snip]

Looks good to me, just one question below:

> +static void bcm_ptp_init(struct bcm_ptp_private *priv)
> +{
> +	priv->nse_ctrl = NSE_GMODE_EN;
> +
> +	mutex_init(&priv->mutex);
> +	skb_queue_head_init(&priv->tx_queue);
> +
> +	priv->mii_ts.rxtstamp = bcm_ptp_rxtstamp;
> +	priv->mii_ts.txtstamp = bcm_ptp_txtstamp;
> +	priv->mii_ts.hwtstamp = bcm_ptp_hwtstamp;
> +	priv->mii_ts.ts_info = bcm_ptp_ts_info;
> +
> +	priv->phydev->mii_ts = &priv->mii_ts;
> +
> +	INIT_DELAYED_WORK(&priv->out_work, bcm_ptp_fsync_work);

Do we need to make sure that we cancel the workqueue in an 
bcm_ptp_exit() function?

I would imagine that the Ethernet MAC attached to that PHY device having 
stopped its receiver and transmitter should ensure no more packets 
coming in or out, however since this is a delayed/asynchronous work, do 
not we need to protect against use after free?
-- 
Florian

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-20 17:24   ` Florian Fainelli
@ 2022-05-21  2:04     ` Jonathan Lemon
  2022-05-23  2:49       ` Florian Fainelli
  0 siblings, 1 reply; 13+ messages in thread
From: Jonathan Lemon @ 2022-05-21  2:04 UTC (permalink / raw)
  To: Florian Fainelli
  Cc: netdev, andrew, hkallweit1, linux, bcm-kernel-feedback-list,
	kernel-team, davem, kuba, pabeni, edumazet

On Fri, May 20, 2022 at 10:24:25AM -0700, Florian Fainelli wrote:
> 
> 
> On 5/18/2022 3:39 PM, Jonathan Lemon wrote:
> > This adds PTP support for BCM54210E Broadcom PHYs, in particular,
> > the BCM54213PE, as used in the Rasperry PI CM4.  It has only been
> > tested on that hardware.
> > 
> > Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
> > ---
> [snip]
> 
> Looks good to me, just one question below:
> 
> > +static void bcm_ptp_init(struct bcm_ptp_private *priv)
> > +{
> > +	priv->nse_ctrl = NSE_GMODE_EN;
> > +
> > +	mutex_init(&priv->mutex);
> > +	skb_queue_head_init(&priv->tx_queue);
> > +
> > +	priv->mii_ts.rxtstamp = bcm_ptp_rxtstamp;
> > +	priv->mii_ts.txtstamp = bcm_ptp_txtstamp;
> > +	priv->mii_ts.hwtstamp = bcm_ptp_hwtstamp;
> > +	priv->mii_ts.ts_info = bcm_ptp_ts_info;
> > +
> > +	priv->phydev->mii_ts = &priv->mii_ts;
> > +
> > +	INIT_DELAYED_WORK(&priv->out_work, bcm_ptp_fsync_work);
> 
> Do we need to make sure that we cancel the workqueue in an bcm_ptp_exit()
> function?
> 
> I would imagine that the Ethernet MAC attached to that PHY device having
> stopped its receiver and transmitter should ensure no more packets coming in
> or out, however since this is a delayed/asynchronous work, do not we need to
> protect against use after free?

The workqueue is just mamually creatimg a 1PPS pulse on the SYNC_OUT
pin, no packet activity.  Arguably, the .suspend hook could stop all work,
but that seems out of scope here? (and this phy does not suspend/resume)
-- 
Jonathan   (one handed typist for a few weeks)

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-21  2:04     ` Jonathan Lemon
@ 2022-05-23  2:49       ` Florian Fainelli
  2022-05-30 22:18         ` Jonathan Lemon
  0 siblings, 1 reply; 13+ messages in thread
From: Florian Fainelli @ 2022-05-23  2:49 UTC (permalink / raw)
  To: Jonathan Lemon
  Cc: netdev, andrew, hkallweit1, linux, bcm-kernel-feedback-list,
	kernel-team, davem, kuba, pabeni, edumazet



On 5/20/2022 7:04 PM, Jonathan Lemon wrote:
> On Fri, May 20, 2022 at 10:24:25AM -0700, Florian Fainelli wrote:
>>
>>
>> On 5/18/2022 3:39 PM, Jonathan Lemon wrote:
>>> This adds PTP support for BCM54210E Broadcom PHYs, in particular,
>>> the BCM54213PE, as used in the Rasperry PI CM4.  It has only been
>>> tested on that hardware.
>>>
>>> Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
>>> ---
>> [snip]
>>
>> Looks good to me, just one question below:
>>
>>> +static void bcm_ptp_init(struct bcm_ptp_private *priv)
>>> +{
>>> +	priv->nse_ctrl = NSE_GMODE_EN;
>>> +
>>> +	mutex_init(&priv->mutex);
>>> +	skb_queue_head_init(&priv->tx_queue);
>>> +
>>> +	priv->mii_ts.rxtstamp = bcm_ptp_rxtstamp;
>>> +	priv->mii_ts.txtstamp = bcm_ptp_txtstamp;
>>> +	priv->mii_ts.hwtstamp = bcm_ptp_hwtstamp;
>>> +	priv->mii_ts.ts_info = bcm_ptp_ts_info;
>>> +
>>> +	priv->phydev->mii_ts = &priv->mii_ts;
>>> +
>>> +	INIT_DELAYED_WORK(&priv->out_work, bcm_ptp_fsync_work);
>>
>> Do we need to make sure that we cancel the workqueue in an bcm_ptp_exit()
>> function?
>>
>> I would imagine that the Ethernet MAC attached to that PHY device having
>> stopped its receiver and transmitter should ensure no more packets coming in
>> or out, however since this is a delayed/asynchronous work, do not we need to
>> protect against use after free?
> 
> The workqueue is just mamually creatimg a 1PPS pulse on the SYNC_OUT
> pin, no packet activity.  Arguably, the .suspend hook could stop all work,
> but that seems out of scope here? (and this phy does not suspend/resume)

The BCM54210E entry does have a suspend/resume entry so it seems to me 
that we do need to cancel the workqueue as the PHY library will not do 
that on our behalf. What I imagine could happen is that this workqueue 
generates spurious MDIO accesses *after* both the PHY and the bus have 
been suspended (and their driver's clock possibly gated already).
-- 
Florian

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 0/2] Broadcom PTP PHY support
  2022-05-18 22:39 [PATCH net-next v5 0/2] Broadcom PTP PHY support Jonathan Lemon
  2022-05-18 22:39 ` [PATCH net-next v5 1/2] net: phy: broadcom: Add Broadcom PTP hooks to bcm-phy-lib Jonathan Lemon
  2022-05-18 22:39 ` [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs Jonathan Lemon
@ 2022-05-28  3:27 ` Richard Cochran
  2022-05-30 16:48   ` Jonathan Lemon
  2 siblings, 1 reply; 13+ messages in thread
From: Richard Cochran @ 2022-05-28  3:27 UTC (permalink / raw)
  To: Jonathan Lemon
  Cc: netdev, f.fainelli, andrew, hkallweit1, linux,
	bcm-kernel-feedback-list, kernel-team, davem, kuba, pabeni,
	edumazet

On Wed, May 18, 2022 at 03:39:33PM -0700, Jonathan Lemon wrote:
> This adds PTP support for the Broadcom PHY BCM54210E (and the
> specific variant BCM54213PE that the rpi-5.15 branch uses).

I'm interested in this series.  Can you please include me on CC?

Thanks,
Richard

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-18 22:39 ` [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs Jonathan Lemon
  2022-05-20 17:24   ` Florian Fainelli
@ 2022-05-29  0:34   ` Richard Cochran
  2022-05-30 17:07     ` Jonathan Lemon
  1 sibling, 1 reply; 13+ messages in thread
From: Richard Cochran @ 2022-05-29  0:34 UTC (permalink / raw)
  To: Jonathan Lemon
  Cc: netdev, f.fainelli, andrew, hkallweit1, linux,
	bcm-kernel-feedback-list, kernel-team, davem, kuba, pabeni,
	edumazet

On Wed, May 18, 2022 at 03:39:35PM -0700, Jonathan Lemon wrote:

> +static int bcm_ptp_adjtime_locked(struct bcm_ptp_private *priv,
> +				  s64 delta_ns)
> +{
> +	struct timespec64 ts;
> +	int err;
> +
> +	err = bcm_ptp_gettime_locked(priv, &ts, NULL);
> +	if (!err) {
> +		set_normalized_timespec64(&ts, ts.tv_sec,
> +					  ts.tv_nsec + delta_ns);

This also takes a LONG time when delta is large...

> +		err = bcm_ptp_settime_locked(priv, &ts);
> +	}
> +	return err;
> +}

Try this instead:

	s64 ns;

	err = bcm_ptp_gettime_locked(priv, &ts, NULL);
	if (err) {
		return err;
	}
	ns = timespec64_to_ns(&ts);		
	ns += delta_ns;
	ts = ns_to_timespec64(ns);
	err = bcm_ptp_settime_locked(priv, &ts);

	return err;

Thanks,
Richard


^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 0/2] Broadcom PTP PHY support
  2022-05-28  3:27 ` [PATCH net-next v5 0/2] Broadcom PTP PHY support Richard Cochran
@ 2022-05-30 16:48   ` Jonathan Lemon
  0 siblings, 0 replies; 13+ messages in thread
From: Jonathan Lemon @ 2022-05-30 16:48 UTC (permalink / raw)
  To: Richard Cochran
  Cc: netdev, f.fainelli, andrew, hkallweit1, linux,
	bcm-kernel-feedback-list, kernel-team, davem, kuba, pabeni,
	edumazet

On Fri, May 27, 2022 at 08:27:12PM -0700, Richard Cochran wrote:
> On Wed, May 18, 2022 at 03:39:33PM -0700, Jonathan Lemon wrote:
> > This adds PTP support for the Broadcom PHY BCM54210E (and the
> > specific variant BCM54213PE that the rpi-5.15 branch uses).
> 
> I'm interested in this series.  Can you please include me on CC?

Sure thing!  (I thought I had already, if not, it was an unintentional
oversight.)
-- 
Jonathan

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-29  0:34   ` Richard Cochran
@ 2022-05-30 17:07     ` Jonathan Lemon
  2022-05-30 23:04       ` Richard Cochran
  0 siblings, 1 reply; 13+ messages in thread
From: Jonathan Lemon @ 2022-05-30 17:07 UTC (permalink / raw)
  To: Richard Cochran
  Cc: netdev, f.fainelli, andrew, hkallweit1, linux,
	bcm-kernel-feedback-list, kernel-team, davem, kuba, pabeni,
	edumazet

On Sat, May 28, 2022 at 05:34:47PM -0700, Richard Cochran wrote:
> On Wed, May 18, 2022 at 03:39:35PM -0700, Jonathan Lemon wrote:
> 
> > +static int bcm_ptp_adjtime_locked(struct bcm_ptp_private *priv,
> > +				  s64 delta_ns)
> > +{
> > +	struct timespec64 ts;
> > +	int err;
> > +
> > +	err = bcm_ptp_gettime_locked(priv, &ts, NULL);
> > +	if (!err) {
> > +		set_normalized_timespec64(&ts, ts.tv_sec,
> > +					  ts.tv_nsec + delta_ns);
> 
> This also takes a LONG time when delta is large...

Didn't we just go through this?  What constitutes a "large" offset here?
The current version seems acceptable to me:

root@rpi:~/src/rpi # time phc_ctl /dev/ptp0 -- adj 86400
phc_ctl[766492.486]: adjusted clock by 86400.000000 seconds

real    0m0.009s
user    0m0.002s
sys 0m0.007s

root@rpi:~/src/rpi # time phc_ctl /dev/ptp0 -- adj -86400
phc_ctl[766494.647]: adjusted clock by -86400.000000 seconds

real    0m0.009s
user    0m0.009s
sys 0m0.000s

-- 
Jonathan

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-23  2:49       ` Florian Fainelli
@ 2022-05-30 22:18         ` Jonathan Lemon
  0 siblings, 0 replies; 13+ messages in thread
From: Jonathan Lemon @ 2022-05-30 22:18 UTC (permalink / raw)
  To: Florian Fainelli
  Cc: netdev, andrew, hkallweit1, linux, bcm-kernel-feedback-list,
	kernel-team, davem, kuba, pabeni, edumazet, richardcochran

On Sun, May 22, 2022 at 07:49:20PM -0700, Florian Fainelli wrote:
> 
> 
> On 5/20/2022 7:04 PM, Jonathan Lemon wrote:
> > On Fri, May 20, 2022 at 10:24:25AM -0700, Florian Fainelli wrote:
> > > 
> > > 
> > > On 5/18/2022 3:39 PM, Jonathan Lemon wrote:
> > > > This adds PTP support for BCM54210E Broadcom PHYs, in particular,
> > > > the BCM54213PE, as used in the Rasperry PI CM4.  It has only been
> > > > tested on that hardware.
> > > > 
> > > > Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
> > > > ---
> > > [snip]
> > > 
> > > Looks good to me, just one question below:
> > > 
> > > > +static void bcm_ptp_init(struct bcm_ptp_private *priv)
> > > > +{
> > > > +	priv->nse_ctrl = NSE_GMODE_EN;
> > > > +
> > > > +	mutex_init(&priv->mutex);
> > > > +	skb_queue_head_init(&priv->tx_queue);
> > > > +
> > > > +	priv->mii_ts.rxtstamp = bcm_ptp_rxtstamp;
> > > > +	priv->mii_ts.txtstamp = bcm_ptp_txtstamp;
> > > > +	priv->mii_ts.hwtstamp = bcm_ptp_hwtstamp;
> > > > +	priv->mii_ts.ts_info = bcm_ptp_ts_info;
> > > > +
> > > > +	priv->phydev->mii_ts = &priv->mii_ts;
> > > > +
> > > > +	INIT_DELAYED_WORK(&priv->out_work, bcm_ptp_fsync_work);
> > > 
> > > Do we need to make sure that we cancel the workqueue in an bcm_ptp_exit()
> > > function?
> > > 
> > > I would imagine that the Ethernet MAC attached to that PHY device having
> > > stopped its receiver and transmitter should ensure no more packets coming in
> > > or out, however since this is a delayed/asynchronous work, do not we need to
> > > protect against use after free?
> > 
> > The workqueue is just mamually creatimg a 1PPS pulse on the SYNC_OUT
> > pin, no packet activity.  Arguably, the .suspend hook could stop all work,
> > but that seems out of scope here? (and this phy does not suspend/resume)
> 
> The BCM54210E entry does have a suspend/resume entry so it seems to me that
> we do need to cancel the workqueue as the PHY library will not do that on
> our behalf. What I imagine could happen is that this workqueue generates
> spurious MDIO accesses *after* both the PHY and the bus have been suspended
> (and their driver's clock possibly gated already).

Yes, you're right.  I was looking at the rpi-5.15.y tree, which doesn't
have these hooks yet.  I'll add a call to stop the workqueue.

Actually, in the next series, I'll break out the extts/perout into
separate patch.

Thanks for pointing this out!
-- 
Jonathan

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-30 17:07     ` Jonathan Lemon
@ 2022-05-30 23:04       ` Richard Cochran
  2022-05-31  1:15         ` Flugsvamp
  0 siblings, 1 reply; 13+ messages in thread
From: Richard Cochran @ 2022-05-30 23:04 UTC (permalink / raw)
  To: Jonathan Lemon
  Cc: netdev, f.fainelli, andrew, hkallweit1, linux,
	bcm-kernel-feedback-list, kernel-team, davem, kuba, pabeni,
	edumazet

On Mon, May 30, 2022 at 10:07:44AM -0700, Jonathan Lemon wrote:
> On Sat, May 28, 2022 at 05:34:47PM -0700, Richard Cochran wrote:
> > On Wed, May 18, 2022 at 03:39:35PM -0700, Jonathan Lemon wrote:
> > 
> > > +static int bcm_ptp_adjtime_locked(struct bcm_ptp_private *priv,
> > > +				  s64 delta_ns)
> > > +{
> > > +	struct timespec64 ts;
> > > +	int err;
> > > +
> > > +	err = bcm_ptp_gettime_locked(priv, &ts, NULL);
> > > +	if (!err) {
> > > +		set_normalized_timespec64(&ts, ts.tv_sec,
> > > +					  ts.tv_nsec + delta_ns);
> > 
> > This also takes a LONG time when delta is large...
> 
> Didn't we just go through this?  What constitutes a "large" offset here?
> The current version seems acceptable to me:

When the PHY boots, it starts from time zero.

Then as a client it needs to adjust to today, something like:

1653951762.413809006 or Mon May 30 16:02:42 2022

(that means adding 1,653,951,762,413,809,006 nanoseconds)

Try that and see how long it takes to apply the adjustment.

Thanks,
Richard

^ permalink raw reply	[flat|nested] 13+ messages in thread

* Re: [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs.
  2022-05-30 23:04       ` Richard Cochran
@ 2022-05-31  1:15         ` Flugsvamp
  0 siblings, 0 replies; 13+ messages in thread
From: Flugsvamp @ 2022-05-31  1:15 UTC (permalink / raw)
  To: Richard Cochran
  Cc: Jonathan Lemon, netdev, f.fainelli, andrew, hkallweit1, linux,
	bcm-kernel-feedback-list, kernel-team, davem, kuba, pabeni,
	edumazet



> On May 30, 2022, at 4:04 PM, Richard Cochran <richardcochran@gmail.com> wrote:
> 
> On Mon, May 30, 2022 at 10:07:44AM -0700, Jonathan Lemon wrote:
>>> On Sat, May 28, 2022 at 05:34:47PM -0700, Richard Cochran wrote:
>>>> On Wed, May 18, 2022 at 03:39:35PM -0700, Jonathan Lemon wrote:
>>> 
>>>> +static int bcm_ptp_adjtime_locked(struct bcm_ptp_private *priv,
>>>> +                  s64 delta_ns)
>>>> +{
>>>> +    struct timespec64 ts;
>>>> +    int err;
>>>> +
>>>> +    err = bcm_ptp_gettime_locked(priv, &ts, NULL);
>>>> +    if (!err) {
>>>> +        set_normalized_timespec64(&ts, ts.tv_sec,
>>>> +                      ts.tv_nsec + delta_ns);
>>> 
>>> This also takes a LONG time when delta is large...
>> 
>> Didn't we just go through this?  What constitutes a "large" offset here?
>> The current version seems acceptable to me:
> 
> When the PHY boots, it starts from time zero.
> 
> Then as a client it needs to adjust to today, something like:
> 
> 1653951762.413809006 or Mon May 30 16:02:42 2022
> 
> (that means adding 1,653,951,762,413,809,006 nanoseconds)
> 
> Try that and see how long it takes to apply the adjustment.

That’s certainly a big adjustment. I was expecting something along the lines of a settime() call, followed up by adjtime().

Will fix. Seems like there should be a better call sequence for this, though. 

Sent from my iPhone

^ permalink raw reply	[flat|nested] 13+ messages in thread

end of thread, other threads:[~2022-05-31  1:15 UTC | newest]

Thread overview: 13+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-05-18 22:39 [PATCH net-next v5 0/2] Broadcom PTP PHY support Jonathan Lemon
2022-05-18 22:39 ` [PATCH net-next v5 1/2] net: phy: broadcom: Add Broadcom PTP hooks to bcm-phy-lib Jonathan Lemon
2022-05-18 22:39 ` [PATCH net-next v5 2/2] net: phy: broadcom: Add PTP support for some Broadcom PHYs Jonathan Lemon
2022-05-20 17:24   ` Florian Fainelli
2022-05-21  2:04     ` Jonathan Lemon
2022-05-23  2:49       ` Florian Fainelli
2022-05-30 22:18         ` Jonathan Lemon
2022-05-29  0:34   ` Richard Cochran
2022-05-30 17:07     ` Jonathan Lemon
2022-05-30 23:04       ` Richard Cochran
2022-05-31  1:15         ` Flugsvamp
2022-05-28  3:27 ` [PATCH net-next v5 0/2] Broadcom PTP PHY support Richard Cochran
2022-05-30 16:48   ` Jonathan Lemon

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).