spectrum_ptp.c

// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
/* Copyright (c) 2019 Mellanox Technologies. All rights reserved */

#include <linux/ptp_clock_kernel.h>
#include <linux/clocksource.h>
#include <linux/timecounter.h>
#include <linux/spinlock.h>
#include <linux/device.h>
#include <linux/rhashtable.h>
#include <linux/ptp_classify.h>
#include <linux/if_ether.h>
#include <linux/if_vlan.h>

#include "spectrum.h"
#include "spectrum_ptp.h"
#include "core.h"

#define MLXSW_SP1_PTP_CLOCK_CYCLES_SHIFT	29
#define MLXSW_SP1_PTP_CLOCK_FREQ_KHZ		156257 /* 6.4nSec */
#define MLXSW_SP1_PTP_CLOCK_MASK		64

#define MLXSW_SP1_PTP_HT_GC_INTERVAL		500 /* ms */

/* How long, approximately, should the unmatched entries stay in the hash table
 * before they are collected. Should be evenly divisible by the GC interval.
 */
#define MLXSW_SP1_PTP_HT_GC_TIMEOUT		1000 /* ms */

struct mlxsw_sp_ptp_state {
	struct mlxsw_sp *mlxsw_sp;
	struct rhashtable unmatched_ht;
	spinlock_t unmatched_lock; /* protects the HT */
	struct delayed_work ht_gc_dw;
	u32 gc_cycle;
};

struct mlxsw_sp1_ptp_key {
	u8 local_port;
	u8 message_type;
	u16 sequence_id;
	u8 domain_number;
	bool ingress;
};

struct mlxsw_sp1_ptp_unmatched {
	struct mlxsw_sp1_ptp_key key;
	struct rhash_head ht_node;
	struct rcu_head rcu;
	struct sk_buff *skb;
	u64 timestamp;
	u32 gc_cycle;
};

static const struct rhashtable_params mlxsw_sp1_ptp_unmatched_ht_params = {
	.key_len = sizeof_field(struct mlxsw_sp1_ptp_unmatched, key),
	.key_offset = offsetof(struct mlxsw_sp1_ptp_unmatched, key),
	.head_offset = offsetof(struct mlxsw_sp1_ptp_unmatched, ht_node),
};

struct mlxsw_sp_ptp_clock {
	struct mlxsw_core *core;
	spinlock_t lock; /* protect this structure */
	struct cyclecounter cycles;
	struct timecounter tc;
	u32 nominal_c_mult;
	struct ptp_clock *ptp;
	struct ptp_clock_info ptp_info;
	unsigned long overflow_period;
	struct delayed_work overflow_work;
};

static u64 __mlxsw_sp1_ptp_read_frc(struct mlxsw_sp_ptp_clock *clock,
				    struct ptp_system_timestamp *sts)
{
	struct mlxsw_core *mlxsw_core = clock->core;
	u32 frc_h1, frc_h2, frc_l;

	frc_h1 = mlxsw_core_read_frc_h(mlxsw_core);
	ptp_read_system_prets(sts);
	frc_l = mlxsw_core_read_frc_l(mlxsw_core);
	ptp_read_system_postts(sts);
	frc_h2 = mlxsw_core_read_frc_h(mlxsw_core);

	if (frc_h1 != frc_h2) {
		/* wrap around */
		ptp_read_system_prets(sts);
		frc_l = mlxsw_core_read_frc_l(mlxsw_core);
		ptp_read_system_postts(sts);
	}

	return (u64) frc_l | (u64) frc_h2 << 32;
}

static u64 mlxsw_sp1_ptp_read_frc(const struct cyclecounter *cc)
{
	struct mlxsw_sp_ptp_clock *clock =
		container_of(cc, struct mlxsw_sp_ptp_clock, cycles);

	return __mlxsw_sp1_ptp_read_frc(clock, NULL) & cc->mask;
}

static int
mlxsw_sp1_ptp_phc_adjfreq(struct mlxsw_sp_ptp_clock *clock, int freq_adj)
{
	struct mlxsw_core *mlxsw_core = clock->core;
	char mtutc_pl[MLXSW_REG_MTUTC_LEN];

	mlxsw_reg_mtutc_pack(mtutc_pl, MLXSW_REG_MTUTC_OPERATION_ADJUST_FREQ,
			     freq_adj, 0);
	return mlxsw_reg_write(mlxsw_core, MLXSW_REG(mtutc), mtutc_pl);
}

static u64 mlxsw_sp1_ptp_ns2cycles(const struct timecounter *tc, u64 nsec)
{
	u64 cycles = (u64) nsec;

	cycles <<= tc->cc->shift;
	cycles = div_u64(cycles, tc->cc->mult);

	return cycles;
}

static int
mlxsw_sp1_ptp_phc_settime(struct mlxsw_sp_ptp_clock *clock, u64 nsec)
{
	struct mlxsw_core *mlxsw_core = clock->core;
	u64 next_sec, next_sec_in_nsec, cycles;
	char mtutc_pl[MLXSW_REG_MTUTC_LEN];
	char mtpps_pl[MLXSW_REG_MTPPS_LEN];
	int err;

	next_sec = div_u64(nsec, NSEC_PER_SEC) + 1;
	next_sec_in_nsec = next_sec * NSEC_PER_SEC;

	spin_lock_bh(&clock->lock);
	cycles = mlxsw_sp1_ptp_ns2cycles(&clock->tc, next_sec_in_nsec);
	spin_unlock_bh(&clock->lock);

	mlxsw_reg_mtpps_vpin_pack(mtpps_pl, cycles);
	err = mlxsw_reg_write(mlxsw_core, MLXSW_REG(mtpps), mtpps_pl);
	if (err)
		return err;

	mlxsw_reg_mtutc_pack(mtutc_pl,
			     MLXSW_REG_MTUTC_OPERATION_SET_TIME_AT_NEXT_SEC,
			     0, next_sec);
	return mlxsw_reg_write(mlxsw_core, MLXSW_REG(mtutc), mtutc_pl);
}

static int mlxsw_sp1_ptp_adjfine(struct ptp_clock_info *ptp, long scaled_ppm)
{
	struct mlxsw_sp_ptp_clock *clock =
		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
	int neg_adj = 0;
	u32 diff;
	u64 adj;
	s32 ppb;

	ppb = scaled_ppm_to_ppb(scaled_ppm);

	if (ppb < 0) {
		neg_adj = 1;
		ppb = -ppb;
	}

	adj = clock->nominal_c_mult;
	adj *= ppb;
	diff = div_u64(adj, NSEC_PER_SEC);

	spin_lock_bh(&clock->lock);
	timecounter_read(&clock->tc);
	clock->cycles.mult = neg_adj ? clock->nominal_c_mult - diff :
				       clock->nominal_c_mult + diff;
	spin_unlock_bh(&clock->lock);

	return mlxsw_sp1_ptp_phc_adjfreq(clock, neg_adj ? -ppb : ppb);
}

static int mlxsw_sp1_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
{
	struct mlxsw_sp_ptp_clock *clock =
		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
	u64 nsec;

	spin_lock_bh(&clock->lock);
	timecounter_adjtime(&clock->tc, delta);
	nsec = timecounter_read(&clock->tc);
	spin_unlock_bh(&clock->lock);

	return mlxsw_sp1_ptp_phc_settime(clock, nsec);
}

static int mlxsw_sp1_ptp_gettimex(struct ptp_clock_info *ptp,
				  struct timespec64 *ts,
				  struct ptp_system_timestamp *sts)
{
	struct mlxsw_sp_ptp_clock *clock =
		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
	u64 cycles, nsec;

	spin_lock_bh(&clock->lock);
	cycles = __mlxsw_sp1_ptp_read_frc(clock, sts);
	nsec = timecounter_cyc2time(&clock->tc, cycles);
	spin_unlock_bh(&clock->lock);

	*ts = ns_to_timespec64(nsec);

	return 0;
}

static int mlxsw_sp1_ptp_settime(struct ptp_clock_info *ptp,
				 const struct timespec64 *ts)
{
	struct mlxsw_sp_ptp_clock *clock =
		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
	u64 nsec = timespec64_to_ns(ts);

	spin_lock_bh(&clock->lock);
	timecounter_init(&clock->tc, &clock->cycles, nsec);
	nsec = timecounter_read(&clock->tc);
	spin_unlock_bh(&clock->lock);

	return mlxsw_sp1_ptp_phc_settime(clock, nsec);
}

static const struct ptp_clock_info mlxsw_sp1_ptp_clock_info = {
	.owner		= THIS_MODULE,
	.name		= "mlxsw_sp_clock",
	.max_adj	= 100000000,
	.adjfine	= mlxsw_sp1_ptp_adjfine,
	.adjtime	= mlxsw_sp1_ptp_adjtime,
	.gettimex64	= mlxsw_sp1_ptp_gettimex,
	.settime64	= mlxsw_sp1_ptp_settime,
};

static void mlxsw_sp1_ptp_clock_overflow(struct work_struct *work)
{
	struct delayed_work *dwork = to_delayed_work(work);
	struct mlxsw_sp_ptp_clock *clock;

	clock = container_of(dwork, struct mlxsw_sp_ptp_clock, overflow_work);

	spin_lock_bh(&clock->lock);
	timecounter_read(&clock->tc);
	spin_unlock_bh(&clock->lock);
	mlxsw_core_schedule_dw(&clock->overflow_work, clock->overflow_period);
}

struct mlxsw_sp_ptp_clock *
mlxsw_sp1_ptp_clock_init(struct mlxsw_sp *mlxsw_sp, struct device *dev)
{
	u64 overflow_cycles, nsec, frac = 0;
	struct mlxsw_sp_ptp_clock *clock;
	int err;

	clock = kzalloc(sizeof(*clock), GFP_KERNEL);
	if (!clock)
		return ERR_PTR(-ENOMEM);

	spin_lock_init(&clock->lock);
	clock->cycles.read = mlxsw_sp1_ptp_read_frc;
	clock->cycles.shift = MLXSW_SP1_PTP_CLOCK_CYCLES_SHIFT;
	clock->cycles.mult = clocksource_khz2mult(MLXSW_SP1_PTP_CLOCK_FREQ_KHZ,
						  clock->cycles.shift);
	clock->nominal_c_mult = clock->cycles.mult;
	clock->cycles.mask = CLOCKSOURCE_MASK(MLXSW_SP1_PTP_CLOCK_MASK);
	clock->core = mlxsw_sp->core;

	timecounter_init(&clock->tc, &clock->cycles,
			 ktime_to_ns(ktime_get_real()));

	/* Calculate period in seconds to call the overflow watchdog - to make
	 * sure counter is checked at least twice every wrap around.
	 * The period is calculated as the minimum between max HW cycles count
	 * (The clock source mask) and max amount of cycles that can be
	 * multiplied by clock multiplier where the result doesn't exceed
	 * 64bits.
	 */
	overflow_cycles = div64_u64(~0ULL >> 1, clock->cycles.mult);
	overflow_cycles = min(overflow_cycles, div_u64(clock->cycles.mask, 3));

	nsec = cyclecounter_cyc2ns(&clock->cycles, overflow_cycles, 0, &frac);
	clock->overflow_period = nsecs_to_jiffies(nsec);

	INIT_DELAYED_WORK(&clock->overflow_work, mlxsw_sp1_ptp_clock_overflow);
	mlxsw_core_schedule_dw(&clock->overflow_work, 0);

	clock->ptp_info = mlxsw_sp1_ptp_clock_info;
	clock->ptp = ptp_clock_register(&clock->ptp_info, dev);
	if (IS_ERR(clock->ptp)) {
		err = PTR_ERR(clock->ptp);
		dev_err(dev, "ptp_clock_register failed %d\n", err);
		goto err_ptp_clock_register;
	}

	return clock;

err_ptp_clock_register:
	cancel_delayed_work_sync(&clock->overflow_work);
	kfree(clock);
	return ERR_PTR(err);
}

void mlxsw_sp1_ptp_clock_fini(struct mlxsw_sp_ptp_clock *clock)
{
	ptp_clock_unregister(clock->ptp);
	cancel_delayed_work_sync(&clock->overflow_work);
	kfree(clock);
}

static int mlxsw_sp_ptp_parse(struct sk_buff *skb,
			      u8 *p_domain_number,
			      u8 *p_message_type,
			      u16 *p_sequence_id)
{
	unsigned int offset = 0;
	unsigned int ptp_class;
	u8 *data;

	data = skb_mac_header(skb);
	ptp_class = ptp_classify_raw(skb);

	switch (ptp_class & PTP_CLASS_VMASK) {
	case PTP_CLASS_V1:
	case PTP_CLASS_V2:
		break;
	default:
		return -ERANGE;
	}

	if (ptp_class & PTP_CLASS_VLAN)
		offset += VLAN_HLEN;

	switch (ptp_class & PTP_CLASS_PMASK) {
	case PTP_CLASS_IPV4:
		offset += ETH_HLEN + IPV4_HLEN(data + offset) + UDP_HLEN;
		break;
	case PTP_CLASS_IPV6:
		offset += ETH_HLEN + IP6_HLEN + UDP_HLEN;
		break;
	case PTP_CLASS_L2:
		offset += ETH_HLEN;
		break;
	default:
		return -ERANGE;
	}

	/* PTP header is 34 bytes. */
	if (skb->len < offset + 34)
		return -EINVAL;

	*p_message_type = data[offset] & 0x0f;
	*p_domain_number = data[offset + 4];
	*p_sequence_id = (u16)(data[offset + 30]) << 8 | data[offset + 31];
	return 0;
}

/* Returns NULL on successful insertion, a pointer on conflict, or an ERR_PTR on
 * error.
 */
static struct mlxsw_sp1_ptp_unmatched *
mlxsw_sp1_ptp_unmatched_save(struct mlxsw_sp *mlxsw_sp,
			     struct mlxsw_sp1_ptp_key key,
			     struct sk_buff *skb,
			     u64 timestamp)
{
	int cycles = MLXSW_SP1_PTP_HT_GC_TIMEOUT / MLXSW_SP1_PTP_HT_GC_INTERVAL;
	struct mlxsw_sp_ptp_state *ptp_state = mlxsw_sp->ptp_state;
	struct mlxsw_sp1_ptp_unmatched *unmatched;
	struct mlxsw_sp1_ptp_unmatched *conflict;

	unmatched = kzalloc(sizeof(*unmatched), GFP_ATOMIC);
	if (!unmatched)
		return ERR_PTR(-ENOMEM);

	unmatched->key = key;
	unmatched->skb = skb;
	unmatched->timestamp = timestamp;
	unmatched->gc_cycle = mlxsw_sp->ptp_state->gc_cycle + cycles;

	conflict = rhashtable_lookup_get_insert_fast(&ptp_state->unmatched_ht,
					    &unmatched->ht_node,
					    mlxsw_sp1_ptp_unmatched_ht_params);
	if (conflict)
		kfree(unmatched);

	return conflict;
}

static struct mlxsw_sp1_ptp_unmatched *
mlxsw_sp1_ptp_unmatched_lookup(struct mlxsw_sp *mlxsw_sp,
			       struct mlxsw_sp1_ptp_key key)
{
	return rhashtable_lookup(&mlxsw_sp->ptp_state->unmatched_ht, &key,
				 mlxsw_sp1_ptp_unmatched_ht_params);
}

static int
mlxsw_sp1_ptp_unmatched_remove(struct mlxsw_sp *mlxsw_sp,
			       struct mlxsw_sp1_ptp_unmatched *unmatched)
{
	return rhashtable_remove_fast(&mlxsw_sp->ptp_state->unmatched_ht,
				      &unmatched->ht_node,
				      mlxsw_sp1_ptp_unmatched_ht_params);
}

/* This function is called in the following scenarios:
 *
 * 1) When a packet is matched with its timestamp.
 * 2) In several situation when it is necessary to immediately pass on
 *    an SKB without a timestamp.
 * 3) From GC indirectly through mlxsw_sp1_ptp_unmatched_finish().
 *    This case is similar to 2) above.
 */
static void mlxsw_sp1_ptp_packet_finish(struct mlxsw_sp *mlxsw_sp,
					struct sk_buff *skb, u8 local_port,
					bool ingress,
					struct skb_shared_hwtstamps *hwtstamps)
{
	struct mlxsw_sp_port *mlxsw_sp_port;

	/* Between capturing the packet and finishing it, there is a window of
	 * opportunity for the originating port to go away (e.g. due to a
	 * split). Also make sure the SKB device reference is still valid.
	 */
	mlxsw_sp_port = mlxsw_sp->ports[local_port];
	if (!mlxsw_sp_port && (!skb->dev || skb->dev == mlxsw_sp_port->dev)) {
		dev_kfree_skb_any(skb);
		return;
	}

	if (ingress) {
		if (hwtstamps)
			*skb_hwtstamps(skb) = *hwtstamps;
		mlxsw_sp_rx_listener_no_mark_func(skb, local_port, mlxsw_sp);
	} else {
		/* skb_tstamp_tx() allows hwtstamps to be NULL. */
		skb_tstamp_tx(skb, hwtstamps);
		dev_kfree_skb_any(skb);
	}
}

static void mlxsw_sp1_packet_timestamp(struct mlxsw_sp *mlxsw_sp,
				       struct mlxsw_sp1_ptp_key key,
				       struct sk_buff *skb,
				       u64 timestamp)
{
	struct skb_shared_hwtstamps hwtstamps;
	u64 nsec;

	spin_lock_bh(&mlxsw_sp->clock->lock);
	nsec = timecounter_cyc2time(&mlxsw_sp->clock->tc, timestamp);
	spin_unlock_bh(&mlxsw_sp->clock->lock);

	hwtstamps.hwtstamp = ns_to_ktime(nsec);
	mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb,
				    key.local_port, key.ingress, &hwtstamps);
}

static void
mlxsw_sp1_ptp_unmatched_finish(struct mlxsw_sp *mlxsw_sp,
			       struct mlxsw_sp1_ptp_unmatched *unmatched)
{
	if (unmatched->skb && unmatched->timestamp)
		mlxsw_sp1_packet_timestamp(mlxsw_sp, unmatched->key,
					   unmatched->skb,
					   unmatched->timestamp);
	else if (unmatched->skb)
		mlxsw_sp1_ptp_packet_finish(mlxsw_sp, unmatched->skb,
					    unmatched->key.local_port,
					    unmatched->key.ingress, NULL);
	kfree_rcu(unmatched, rcu);
}

static void mlxsw_sp1_ptp_unmatched_free_fn(void *ptr, void *arg)
{
	struct mlxsw_sp1_ptp_unmatched *unmatched = ptr;

	/* This is invoked at a point where the ports are gone already. Nothing
	 * to do with whatever is left in the HT but to free it.
	 */
	if (unmatched->skb)
		dev_kfree_skb_any(unmatched->skb);
	kfree_rcu(unmatched, rcu);
}

static void mlxsw_sp1_ptp_got_piece(struct mlxsw_sp *mlxsw_sp,
				    struct mlxsw_sp1_ptp_key key,
				    struct sk_buff *skb, u64 timestamp)
{
	struct mlxsw_sp1_ptp_unmatched *unmatched, *conflict;
	int err;

	rcu_read_lock();

	unmatched = mlxsw_sp1_ptp_unmatched_lookup(mlxsw_sp, key);

	spin_lock(&mlxsw_sp->ptp_state->unmatched_lock);

	if (unmatched) {
		/* There was an unmatched entry when we looked, but it may have
		 * been removed before we took the lock.
		 */
		err = mlxsw_sp1_ptp_unmatched_remove(mlxsw_sp, unmatched);
		if (err)
			unmatched = NULL;
	}

	if (!unmatched) {
		/* We have no unmatched entry, but one may have been added after
		 * we looked, but before we took the lock.
		 */
		unmatched = mlxsw_sp1_ptp_unmatched_save(mlxsw_sp, key,
							 skb, timestamp);
		if (IS_ERR(unmatched)) {
			if (skb)
				mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb,
							    key.local_port,
							    key.ingress, NULL);
			unmatched = NULL;
		} else if (unmatched) {
			/* Save just told us, under lock, that the entry is
			 * there, so this has to work.
			 */
			err = mlxsw_sp1_ptp_unmatched_remove(mlxsw_sp,
							     unmatched);
			WARN_ON_ONCE(err);
		}
	}

	/* If unmatched is non-NULL here, it comes either from the lookup, or
	 * from the save attempt above. In either case the entry was removed
	 * from the hash table. If unmatched is NULL, a new unmatched entry was
	 * added to the hash table, and there was no conflict.
	 */

	if (skb && unmatched && unmatched->timestamp) {
		unmatched->skb = skb;
	} else if (timestamp && unmatched && unmatched->skb) {
		unmatched->timestamp = timestamp;
	} else if (unmatched) {
		/* unmatched holds an older entry of the same type: either an
		 * skb if we are handling skb, or a timestamp if we are handling
		 * timestamp. We can't match that up, so save what we have.
		 */
		conflict = mlxsw_sp1_ptp_unmatched_save(mlxsw_sp, key,
							skb, timestamp);
		if (IS_ERR(conflict)) {
			if (skb)
				mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb,
							    key.local_port,
							    key.ingress, NULL);
		} else {
			/* Above, we removed an object with this key from the
			 * hash table, under lock, so conflict can not be a
			 * valid pointer.
			 */
			WARN_ON_ONCE(conflict);
		}
	}

	spin_unlock(&mlxsw_sp->ptp_state->unmatched_lock);

	if (unmatched)
		mlxsw_sp1_ptp_unmatched_finish(mlxsw_sp, unmatched);

	rcu_read_unlock();
}

static void mlxsw_sp1_ptp_got_packet(struct mlxsw_sp *mlxsw_sp,
				     struct sk_buff *skb, u8 local_port,
				     bool ingress)
{
	struct mlxsw_sp_port *mlxsw_sp_port;
	struct mlxsw_sp1_ptp_key key;
	u8 types;
	int err;

	mlxsw_sp_port = mlxsw_sp->ports[local_port];
	if (!mlxsw_sp_port)
		goto immediate;

	types = ingress ? mlxsw_sp_port->ptp.ing_types :
			  mlxsw_sp_port->ptp.egr_types;
	if (!types)
		goto immediate;

	memset(&key, 0, sizeof(key));
	key.local_port = local_port;
	key.ingress = ingress;

	err = mlxsw_sp_ptp_parse(skb, &key.domain_number, &key.message_type,
				 &key.sequence_id);
	if (err)
		goto immediate;

	/* For packets whose timestamping was not enabled on this port, don't
	 * bother trying to match the timestamp.
	 */
	if (!((1 << key.message_type) & types))
		goto immediate;

	mlxsw_sp1_ptp_got_piece(mlxsw_sp, key, skb, 0);
	return;

immediate:
	mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb, local_port, ingress, NULL);
}

void mlxsw_sp1_ptp_got_timestamp(struct mlxsw_sp *mlxsw_sp, bool ingress,
				 u8 local_port, u8 message_type,
				 u8 domain_number, u16 sequence_id,
				 u64 timestamp)
{
	struct mlxsw_sp_port *mlxsw_sp_port;
	struct mlxsw_sp1_ptp_key key;
	u8 types;

	mlxsw_sp_port = mlxsw_sp->ports[local_port];
	if (!mlxsw_sp_port)
		return;

	types = ingress ? mlxsw_sp_port->ptp.ing_types :
			  mlxsw_sp_port->ptp.egr_types;

	/* For message types whose timestamping was not enabled on this port,
	 * don't bother with the timestamp.
	 */
	if (!((1 << message_type) & types))
		return;

	memset(&key, 0, sizeof(key));
	key.local_port = local_port;
	key.domain_number = domain_number;
	key.message_type = message_type;
	key.sequence_id = sequence_id;
	key.ingress = ingress;

	mlxsw_sp1_ptp_got_piece(mlxsw_sp, key, NULL, timestamp);
}

void mlxsw_sp1_ptp_receive(struct mlxsw_sp *mlxsw_sp, struct sk_buff *skb,
			   u8 local_port)
{
	skb_reset_mac_header(skb);
	mlxsw_sp1_ptp_got_packet(mlxsw_sp, skb, local_port, true);
}

void mlxsw_sp1_ptp_transmitted(struct mlxsw_sp *mlxsw_sp,
			       struct sk_buff *skb, u8 local_port)
{
	mlxsw_sp1_ptp_got_packet(mlxsw_sp, skb, local_port, false);
}

static void
mlxsw_sp1_ptp_ht_gc_collect(struct mlxsw_sp_ptp_state *ptp_state,
			    struct mlxsw_sp1_ptp_unmatched *unmatched)
{
	int err;

	/* If an unmatched entry has an SKB, it has to be handed over to the
	 * networking stack. This is usually done from a trap handler, which is
	 * invoked in a softirq context. Here we are going to do it in process
	 * context. If that were to be interrupted by a softirq, it could cause
	 * a deadlock when an attempt is made to take an already-taken lock
	 * somewhere along the sending path. Disable softirqs to prevent this.
	 */
	local_bh_disable();

	spin_lock(&ptp_state->unmatched_lock);
	err = rhashtable_remove_fast(&ptp_state->unmatched_ht,
				     &unmatched->ht_node,
				     mlxsw_sp1_ptp_unmatched_ht_params);
	spin_unlock(&ptp_state->unmatched_lock);

	if (err)
		/* The packet was matched with timestamp during the walk. */
		goto out;

	/* mlxsw_sp1_ptp_unmatched_finish() invokes netif_receive_skb(). While
	 * the comment at that function states that it can only be called in
	 * soft IRQ context, this pattern of local_bh_disable() +
	 * netif_receive_skb(), in process context, is seen elsewhere in the
	 * kernel, notably in pktgen.
	 */
	mlxsw_sp1_ptp_unmatched_finish(ptp_state->mlxsw_sp, unmatched);

out:
	local_bh_enable();
}

static void mlxsw_sp1_ptp_ht_gc(struct work_struct *work)
{
	struct delayed_work *dwork = to_delayed_work(work);
	struct mlxsw_sp1_ptp_unmatched *unmatched;
	struct mlxsw_sp_ptp_state *ptp_state;
	struct rhashtable_iter iter;
	u32 gc_cycle;
	void *obj;

	ptp_state = container_of(dwork, struct mlxsw_sp_ptp_state, ht_gc_dw);
	gc_cycle = ptp_state->gc_cycle++;

	rhashtable_walk_enter(&ptp_state->unmatched_ht, &iter);
	rhashtable_walk_start(&iter);
	while ((obj = rhashtable_walk_next(&iter))) {
		if (IS_ERR(obj))
			continue;

		unmatched = obj;
		if (unmatched->gc_cycle <= gc_cycle)
			mlxsw_sp1_ptp_ht_gc_collect(ptp_state, unmatched);
	}
	rhashtable_walk_stop(&iter);
	rhashtable_walk_exit(&iter);

	mlxsw_core_schedule_dw(&ptp_state->ht_gc_dw,
			       MLXSW_SP1_PTP_HT_GC_INTERVAL);
}

struct mlxsw_sp_ptp_state *mlxsw_sp1_ptp_init(struct mlxsw_sp *mlxsw_sp)
{
	struct mlxsw_sp_ptp_state *ptp_state;
	int err;

	ptp_state = kzalloc(sizeof(*ptp_state), GFP_KERNEL);
	if (!ptp_state)
		return ERR_PTR(-ENOMEM);
	ptp_state->mlxsw_sp = mlxsw_sp;

	spin_lock_init(&ptp_state->unmatched_lock);

	err = rhashtable_init(&ptp_state->unmatched_ht,
			      &mlxsw_sp1_ptp_unmatched_ht_params);
	if (err)
		goto err_hashtable_init;

	INIT_DELAYED_WORK(&ptp_state->ht_gc_dw, mlxsw_sp1_ptp_ht_gc);
	mlxsw_core_schedule_dw(&ptp_state->ht_gc_dw,
			       MLXSW_SP1_PTP_HT_GC_INTERVAL);
	return ptp_state;

err_hashtable_init:
	kfree(ptp_state);
	return ERR_PTR(err);
}

void mlxsw_sp1_ptp_fini(struct mlxsw_sp_ptp_state *ptp_state)
{
	cancel_delayed_work_sync(&ptp_state->ht_gc_dw);
	rhashtable_free_and_destroy(&ptp_state->unmatched_ht,
				    &mlxsw_sp1_ptp_unmatched_free_fn, NULL);
	kfree(ptp_state);
}