xref: /linux/drivers/net/ethernet/mellanox/mlxsw/spectrum_ptp.c (revision ba95c7452439756d4f6dceb5a188b7c31dbbe5b6)
1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2019 Mellanox Technologies. All rights reserved */
3 
4 #include <linux/ptp_clock_kernel.h>
5 #include <linux/clocksource.h>
6 #include <linux/timecounter.h>
7 #include <linux/spinlock.h>
8 #include <linux/device.h>
9 #include <linux/rhashtable.h>
10 #include <linux/ptp_classify.h>
11 #include <linux/if_ether.h>
12 #include <linux/if_vlan.h>
13 #include <linux/net_tstamp.h>
14 
15 #include "spectrum.h"
16 #include "spectrum_ptp.h"
17 #include "core.h"
18 
19 #define MLXSW_SP1_PTP_CLOCK_CYCLES_SHIFT	29
20 #define MLXSW_SP1_PTP_CLOCK_FREQ_KHZ		156257 /* 6.4nSec */
21 #define MLXSW_SP1_PTP_CLOCK_MASK		64
22 
23 #define MLXSW_SP1_PTP_HT_GC_INTERVAL		500 /* ms */
24 
25 /* How long, approximately, should the unmatched entries stay in the hash table
26  * before they are collected. Should be evenly divisible by the GC interval.
27  */
28 #define MLXSW_SP1_PTP_HT_GC_TIMEOUT		1000 /* ms */
29 
30 struct mlxsw_sp_ptp_state {
31 	struct mlxsw_sp *mlxsw_sp;
32 	struct rhashtable unmatched_ht;
33 	spinlock_t unmatched_lock; /* protects the HT */
34 	struct delayed_work ht_gc_dw;
35 	u32 gc_cycle;
36 };
37 
38 struct mlxsw_sp1_ptp_key {
39 	u8 local_port;
40 	u8 message_type;
41 	u16 sequence_id;
42 	u8 domain_number;
43 	bool ingress;
44 };
45 
46 struct mlxsw_sp1_ptp_unmatched {
47 	struct mlxsw_sp1_ptp_key key;
48 	struct rhash_head ht_node;
49 	struct rcu_head rcu;
50 	struct sk_buff *skb;
51 	u64 timestamp;
52 	u32 gc_cycle;
53 };
54 
55 static const struct rhashtable_params mlxsw_sp1_ptp_unmatched_ht_params = {
56 	.key_len = sizeof_field(struct mlxsw_sp1_ptp_unmatched, key),
57 	.key_offset = offsetof(struct mlxsw_sp1_ptp_unmatched, key),
58 	.head_offset = offsetof(struct mlxsw_sp1_ptp_unmatched, ht_node),
59 };
60 
61 struct mlxsw_sp_ptp_clock {
62 	struct mlxsw_core *core;
63 	spinlock_t lock; /* protect this structure */
64 	struct cyclecounter cycles;
65 	struct timecounter tc;
66 	u32 nominal_c_mult;
67 	struct ptp_clock *ptp;
68 	struct ptp_clock_info ptp_info;
69 	unsigned long overflow_period;
70 	struct delayed_work overflow_work;
71 };
72 
73 static u64 __mlxsw_sp1_ptp_read_frc(struct mlxsw_sp_ptp_clock *clock,
74 				    struct ptp_system_timestamp *sts)
75 {
76 	struct mlxsw_core *mlxsw_core = clock->core;
77 	u32 frc_h1, frc_h2, frc_l;
78 
79 	frc_h1 = mlxsw_core_read_frc_h(mlxsw_core);
80 	ptp_read_system_prets(sts);
81 	frc_l = mlxsw_core_read_frc_l(mlxsw_core);
82 	ptp_read_system_postts(sts);
83 	frc_h2 = mlxsw_core_read_frc_h(mlxsw_core);
84 
85 	if (frc_h1 != frc_h2) {
86 		/* wrap around */
87 		ptp_read_system_prets(sts);
88 		frc_l = mlxsw_core_read_frc_l(mlxsw_core);
89 		ptp_read_system_postts(sts);
90 	}
91 
92 	return (u64) frc_l | (u64) frc_h2 << 32;
93 }
94 
95 static u64 mlxsw_sp1_ptp_read_frc(const struct cyclecounter *cc)
96 {
97 	struct mlxsw_sp_ptp_clock *clock =
98 		container_of(cc, struct mlxsw_sp_ptp_clock, cycles);
99 
100 	return __mlxsw_sp1_ptp_read_frc(clock, NULL) & cc->mask;
101 }
102 
103 static int
104 mlxsw_sp1_ptp_phc_adjfreq(struct mlxsw_sp_ptp_clock *clock, int freq_adj)
105 {
106 	struct mlxsw_core *mlxsw_core = clock->core;
107 	char mtutc_pl[MLXSW_REG_MTUTC_LEN];
108 
109 	mlxsw_reg_mtutc_pack(mtutc_pl, MLXSW_REG_MTUTC_OPERATION_ADJUST_FREQ,
110 			     freq_adj, 0);
111 	return mlxsw_reg_write(mlxsw_core, MLXSW_REG(mtutc), mtutc_pl);
112 }
113 
114 static u64 mlxsw_sp1_ptp_ns2cycles(const struct timecounter *tc, u64 nsec)
115 {
116 	u64 cycles = (u64) nsec;
117 
118 	cycles <<= tc->cc->shift;
119 	cycles = div_u64(cycles, tc->cc->mult);
120 
121 	return cycles;
122 }
123 
124 static int
125 mlxsw_sp1_ptp_phc_settime(struct mlxsw_sp_ptp_clock *clock, u64 nsec)
126 {
127 	struct mlxsw_core *mlxsw_core = clock->core;
128 	u64 next_sec, next_sec_in_nsec, cycles;
129 	char mtutc_pl[MLXSW_REG_MTUTC_LEN];
130 	char mtpps_pl[MLXSW_REG_MTPPS_LEN];
131 	int err;
132 
133 	next_sec = div_u64(nsec, NSEC_PER_SEC) + 1;
134 	next_sec_in_nsec = next_sec * NSEC_PER_SEC;
135 
136 	spin_lock_bh(&clock->lock);
137 	cycles = mlxsw_sp1_ptp_ns2cycles(&clock->tc, next_sec_in_nsec);
138 	spin_unlock_bh(&clock->lock);
139 
140 	mlxsw_reg_mtpps_vpin_pack(mtpps_pl, cycles);
141 	err = mlxsw_reg_write(mlxsw_core, MLXSW_REG(mtpps), mtpps_pl);
142 	if (err)
143 		return err;
144 
145 	mlxsw_reg_mtutc_pack(mtutc_pl,
146 			     MLXSW_REG_MTUTC_OPERATION_SET_TIME_AT_NEXT_SEC,
147 			     0, next_sec);
148 	return mlxsw_reg_write(mlxsw_core, MLXSW_REG(mtutc), mtutc_pl);
149 }
150 
151 static int mlxsw_sp1_ptp_adjfine(struct ptp_clock_info *ptp, long scaled_ppm)
152 {
153 	struct mlxsw_sp_ptp_clock *clock =
154 		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
155 	int neg_adj = 0;
156 	u32 diff;
157 	u64 adj;
158 	s32 ppb;
159 
160 	ppb = scaled_ppm_to_ppb(scaled_ppm);
161 
162 	if (ppb < 0) {
163 		neg_adj = 1;
164 		ppb = -ppb;
165 	}
166 
167 	adj = clock->nominal_c_mult;
168 	adj *= ppb;
169 	diff = div_u64(adj, NSEC_PER_SEC);
170 
171 	spin_lock_bh(&clock->lock);
172 	timecounter_read(&clock->tc);
173 	clock->cycles.mult = neg_adj ? clock->nominal_c_mult - diff :
174 				       clock->nominal_c_mult + diff;
175 	spin_unlock_bh(&clock->lock);
176 
177 	return mlxsw_sp1_ptp_phc_adjfreq(clock, neg_adj ? -ppb : ppb);
178 }
179 
180 static int mlxsw_sp1_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
181 {
182 	struct mlxsw_sp_ptp_clock *clock =
183 		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
184 	u64 nsec;
185 
186 	spin_lock_bh(&clock->lock);
187 	timecounter_adjtime(&clock->tc, delta);
188 	nsec = timecounter_read(&clock->tc);
189 	spin_unlock_bh(&clock->lock);
190 
191 	return mlxsw_sp1_ptp_phc_settime(clock, nsec);
192 }
193 
194 static int mlxsw_sp1_ptp_gettimex(struct ptp_clock_info *ptp,
195 				  struct timespec64 *ts,
196 				  struct ptp_system_timestamp *sts)
197 {
198 	struct mlxsw_sp_ptp_clock *clock =
199 		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
200 	u64 cycles, nsec;
201 
202 	spin_lock_bh(&clock->lock);
203 	cycles = __mlxsw_sp1_ptp_read_frc(clock, sts);
204 	nsec = timecounter_cyc2time(&clock->tc, cycles);
205 	spin_unlock_bh(&clock->lock);
206 
207 	*ts = ns_to_timespec64(nsec);
208 
209 	return 0;
210 }
211 
212 static int mlxsw_sp1_ptp_settime(struct ptp_clock_info *ptp,
213 				 const struct timespec64 *ts)
214 {
215 	struct mlxsw_sp_ptp_clock *clock =
216 		container_of(ptp, struct mlxsw_sp_ptp_clock, ptp_info);
217 	u64 nsec = timespec64_to_ns(ts);
218 
219 	spin_lock_bh(&clock->lock);
220 	timecounter_init(&clock->tc, &clock->cycles, nsec);
221 	nsec = timecounter_read(&clock->tc);
222 	spin_unlock_bh(&clock->lock);
223 
224 	return mlxsw_sp1_ptp_phc_settime(clock, nsec);
225 }
226 
227 static const struct ptp_clock_info mlxsw_sp1_ptp_clock_info = {
228 	.owner		= THIS_MODULE,
229 	.name		= "mlxsw_sp_clock",
230 	.max_adj	= 100000000,
231 	.adjfine	= mlxsw_sp1_ptp_adjfine,
232 	.adjtime	= mlxsw_sp1_ptp_adjtime,
233 	.gettimex64	= mlxsw_sp1_ptp_gettimex,
234 	.settime64	= mlxsw_sp1_ptp_settime,
235 };
236 
237 static void mlxsw_sp1_ptp_clock_overflow(struct work_struct *work)
238 {
239 	struct delayed_work *dwork = to_delayed_work(work);
240 	struct mlxsw_sp_ptp_clock *clock;
241 
242 	clock = container_of(dwork, struct mlxsw_sp_ptp_clock, overflow_work);
243 
244 	spin_lock_bh(&clock->lock);
245 	timecounter_read(&clock->tc);
246 	spin_unlock_bh(&clock->lock);
247 	mlxsw_core_schedule_dw(&clock->overflow_work, clock->overflow_period);
248 }
249 
250 struct mlxsw_sp_ptp_clock *
251 mlxsw_sp1_ptp_clock_init(struct mlxsw_sp *mlxsw_sp, struct device *dev)
252 {
253 	u64 overflow_cycles, nsec, frac = 0;
254 	struct mlxsw_sp_ptp_clock *clock;
255 	int err;
256 
257 	clock = kzalloc(sizeof(*clock), GFP_KERNEL);
258 	if (!clock)
259 		return ERR_PTR(-ENOMEM);
260 
261 	spin_lock_init(&clock->lock);
262 	clock->cycles.read = mlxsw_sp1_ptp_read_frc;
263 	clock->cycles.shift = MLXSW_SP1_PTP_CLOCK_CYCLES_SHIFT;
264 	clock->cycles.mult = clocksource_khz2mult(MLXSW_SP1_PTP_CLOCK_FREQ_KHZ,
265 						  clock->cycles.shift);
266 	clock->nominal_c_mult = clock->cycles.mult;
267 	clock->cycles.mask = CLOCKSOURCE_MASK(MLXSW_SP1_PTP_CLOCK_MASK);
268 	clock->core = mlxsw_sp->core;
269 
270 	timecounter_init(&clock->tc, &clock->cycles,
271 			 ktime_to_ns(ktime_get_real()));
272 
273 	/* Calculate period in seconds to call the overflow watchdog - to make
274 	 * sure counter is checked at least twice every wrap around.
275 	 * The period is calculated as the minimum between max HW cycles count
276 	 * (The clock source mask) and max amount of cycles that can be
277 	 * multiplied by clock multiplier where the result doesn't exceed
278 	 * 64bits.
279 	 */
280 	overflow_cycles = div64_u64(~0ULL >> 1, clock->cycles.mult);
281 	overflow_cycles = min(overflow_cycles, div_u64(clock->cycles.mask, 3));
282 
283 	nsec = cyclecounter_cyc2ns(&clock->cycles, overflow_cycles, 0, &frac);
284 	clock->overflow_period = nsecs_to_jiffies(nsec);
285 
286 	INIT_DELAYED_WORK(&clock->overflow_work, mlxsw_sp1_ptp_clock_overflow);
287 	mlxsw_core_schedule_dw(&clock->overflow_work, 0);
288 
289 	clock->ptp_info = mlxsw_sp1_ptp_clock_info;
290 	clock->ptp = ptp_clock_register(&clock->ptp_info, dev);
291 	if (IS_ERR(clock->ptp)) {
292 		err = PTR_ERR(clock->ptp);
293 		dev_err(dev, "ptp_clock_register failed %d\n", err);
294 		goto err_ptp_clock_register;
295 	}
296 
297 	return clock;
298 
299 err_ptp_clock_register:
300 	cancel_delayed_work_sync(&clock->overflow_work);
301 	kfree(clock);
302 	return ERR_PTR(err);
303 }
304 
305 void mlxsw_sp1_ptp_clock_fini(struct mlxsw_sp_ptp_clock *clock)
306 {
307 	ptp_clock_unregister(clock->ptp);
308 	cancel_delayed_work_sync(&clock->overflow_work);
309 	kfree(clock);
310 }
311 
312 static int mlxsw_sp_ptp_parse(struct sk_buff *skb,
313 			      u8 *p_domain_number,
314 			      u8 *p_message_type,
315 			      u16 *p_sequence_id)
316 {
317 	unsigned int offset = 0;
318 	unsigned int ptp_class;
319 	u8 *data;
320 
321 	data = skb_mac_header(skb);
322 	ptp_class = ptp_classify_raw(skb);
323 
324 	switch (ptp_class & PTP_CLASS_VMASK) {
325 	case PTP_CLASS_V1:
326 	case PTP_CLASS_V2:
327 		break;
328 	default:
329 		return -ERANGE;
330 	}
331 
332 	if (ptp_class & PTP_CLASS_VLAN)
333 		offset += VLAN_HLEN;
334 
335 	switch (ptp_class & PTP_CLASS_PMASK) {
336 	case PTP_CLASS_IPV4:
337 		offset += ETH_HLEN + IPV4_HLEN(data + offset) + UDP_HLEN;
338 		break;
339 	case PTP_CLASS_IPV6:
340 		offset += ETH_HLEN + IP6_HLEN + UDP_HLEN;
341 		break;
342 	case PTP_CLASS_L2:
343 		offset += ETH_HLEN;
344 		break;
345 	default:
346 		return -ERANGE;
347 	}
348 
349 	/* PTP header is 34 bytes. */
350 	if (skb->len < offset + 34)
351 		return -EINVAL;
352 
353 	*p_message_type = data[offset] & 0x0f;
354 	*p_domain_number = data[offset + 4];
355 	*p_sequence_id = (u16)(data[offset + 30]) << 8 | data[offset + 31];
356 	return 0;
357 }
358 
359 /* Returns NULL on successful insertion, a pointer on conflict, or an ERR_PTR on
360  * error.
361  */
362 static struct mlxsw_sp1_ptp_unmatched *
363 mlxsw_sp1_ptp_unmatched_save(struct mlxsw_sp *mlxsw_sp,
364 			     struct mlxsw_sp1_ptp_key key,
365 			     struct sk_buff *skb,
366 			     u64 timestamp)
367 {
368 	int cycles = MLXSW_SP1_PTP_HT_GC_TIMEOUT / MLXSW_SP1_PTP_HT_GC_INTERVAL;
369 	struct mlxsw_sp_ptp_state *ptp_state = mlxsw_sp->ptp_state;
370 	struct mlxsw_sp1_ptp_unmatched *unmatched;
371 	struct mlxsw_sp1_ptp_unmatched *conflict;
372 
373 	unmatched = kzalloc(sizeof(*unmatched), GFP_ATOMIC);
374 	if (!unmatched)
375 		return ERR_PTR(-ENOMEM);
376 
377 	unmatched->key = key;
378 	unmatched->skb = skb;
379 	unmatched->timestamp = timestamp;
380 	unmatched->gc_cycle = mlxsw_sp->ptp_state->gc_cycle + cycles;
381 
382 	conflict = rhashtable_lookup_get_insert_fast(&ptp_state->unmatched_ht,
383 					    &unmatched->ht_node,
384 					    mlxsw_sp1_ptp_unmatched_ht_params);
385 	if (conflict)
386 		kfree(unmatched);
387 
388 	return conflict;
389 }
390 
391 static struct mlxsw_sp1_ptp_unmatched *
392 mlxsw_sp1_ptp_unmatched_lookup(struct mlxsw_sp *mlxsw_sp,
393 			       struct mlxsw_sp1_ptp_key key)
394 {
395 	return rhashtable_lookup(&mlxsw_sp->ptp_state->unmatched_ht, &key,
396 				 mlxsw_sp1_ptp_unmatched_ht_params);
397 }
398 
399 static int
400 mlxsw_sp1_ptp_unmatched_remove(struct mlxsw_sp *mlxsw_sp,
401 			       struct mlxsw_sp1_ptp_unmatched *unmatched)
402 {
403 	return rhashtable_remove_fast(&mlxsw_sp->ptp_state->unmatched_ht,
404 				      &unmatched->ht_node,
405 				      mlxsw_sp1_ptp_unmatched_ht_params);
406 }
407 
408 /* This function is called in the following scenarios:
409  *
410  * 1) When a packet is matched with its timestamp.
411  * 2) In several situation when it is necessary to immediately pass on
412  *    an SKB without a timestamp.
413  * 3) From GC indirectly through mlxsw_sp1_ptp_unmatched_finish().
414  *    This case is similar to 2) above.
415  */
416 static void mlxsw_sp1_ptp_packet_finish(struct mlxsw_sp *mlxsw_sp,
417 					struct sk_buff *skb, u8 local_port,
418 					bool ingress,
419 					struct skb_shared_hwtstamps *hwtstamps)
420 {
421 	struct mlxsw_sp_port *mlxsw_sp_port;
422 
423 	/* Between capturing the packet and finishing it, there is a window of
424 	 * opportunity for the originating port to go away (e.g. due to a
425 	 * split). Also make sure the SKB device reference is still valid.
426 	 */
427 	mlxsw_sp_port = mlxsw_sp->ports[local_port];
428 	if (!(mlxsw_sp_port && (!skb->dev || skb->dev == mlxsw_sp_port->dev))) {
429 		dev_kfree_skb_any(skb);
430 		return;
431 	}
432 
433 	if (ingress) {
434 		if (hwtstamps)
435 			*skb_hwtstamps(skb) = *hwtstamps;
436 		mlxsw_sp_rx_listener_no_mark_func(skb, local_port, mlxsw_sp);
437 	} else {
438 		/* skb_tstamp_tx() allows hwtstamps to be NULL. */
439 		skb_tstamp_tx(skb, hwtstamps);
440 		dev_kfree_skb_any(skb);
441 	}
442 }
443 
444 static void mlxsw_sp1_packet_timestamp(struct mlxsw_sp *mlxsw_sp,
445 				       struct mlxsw_sp1_ptp_key key,
446 				       struct sk_buff *skb,
447 				       u64 timestamp)
448 {
449 	struct skb_shared_hwtstamps hwtstamps;
450 	u64 nsec;
451 
452 	spin_lock_bh(&mlxsw_sp->clock->lock);
453 	nsec = timecounter_cyc2time(&mlxsw_sp->clock->tc, timestamp);
454 	spin_unlock_bh(&mlxsw_sp->clock->lock);
455 
456 	hwtstamps.hwtstamp = ns_to_ktime(nsec);
457 	mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb,
458 				    key.local_port, key.ingress, &hwtstamps);
459 }
460 
461 static void
462 mlxsw_sp1_ptp_unmatched_finish(struct mlxsw_sp *mlxsw_sp,
463 			       struct mlxsw_sp1_ptp_unmatched *unmatched)
464 {
465 	if (unmatched->skb && unmatched->timestamp)
466 		mlxsw_sp1_packet_timestamp(mlxsw_sp, unmatched->key,
467 					   unmatched->skb,
468 					   unmatched->timestamp);
469 	else if (unmatched->skb)
470 		mlxsw_sp1_ptp_packet_finish(mlxsw_sp, unmatched->skb,
471 					    unmatched->key.local_port,
472 					    unmatched->key.ingress, NULL);
473 	kfree_rcu(unmatched, rcu);
474 }
475 
476 static void mlxsw_sp1_ptp_unmatched_free_fn(void *ptr, void *arg)
477 {
478 	struct mlxsw_sp1_ptp_unmatched *unmatched = ptr;
479 
480 	/* This is invoked at a point where the ports are gone already. Nothing
481 	 * to do with whatever is left in the HT but to free it.
482 	 */
483 	if (unmatched->skb)
484 		dev_kfree_skb_any(unmatched->skb);
485 	kfree_rcu(unmatched, rcu);
486 }
487 
488 static void mlxsw_sp1_ptp_got_piece(struct mlxsw_sp *mlxsw_sp,
489 				    struct mlxsw_sp1_ptp_key key,
490 				    struct sk_buff *skb, u64 timestamp)
491 {
492 	struct mlxsw_sp1_ptp_unmatched *unmatched, *conflict;
493 	int err;
494 
495 	rcu_read_lock();
496 
497 	unmatched = mlxsw_sp1_ptp_unmatched_lookup(mlxsw_sp, key);
498 
499 	spin_lock(&mlxsw_sp->ptp_state->unmatched_lock);
500 
501 	if (unmatched) {
502 		/* There was an unmatched entry when we looked, but it may have
503 		 * been removed before we took the lock.
504 		 */
505 		err = mlxsw_sp1_ptp_unmatched_remove(mlxsw_sp, unmatched);
506 		if (err)
507 			unmatched = NULL;
508 	}
509 
510 	if (!unmatched) {
511 		/* We have no unmatched entry, but one may have been added after
512 		 * we looked, but before we took the lock.
513 		 */
514 		unmatched = mlxsw_sp1_ptp_unmatched_save(mlxsw_sp, key,
515 							 skb, timestamp);
516 		if (IS_ERR(unmatched)) {
517 			if (skb)
518 				mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb,
519 							    key.local_port,
520 							    key.ingress, NULL);
521 			unmatched = NULL;
522 		} else if (unmatched) {
523 			/* Save just told us, under lock, that the entry is
524 			 * there, so this has to work.
525 			 */
526 			err = mlxsw_sp1_ptp_unmatched_remove(mlxsw_sp,
527 							     unmatched);
528 			WARN_ON_ONCE(err);
529 		}
530 	}
531 
532 	/* If unmatched is non-NULL here, it comes either from the lookup, or
533 	 * from the save attempt above. In either case the entry was removed
534 	 * from the hash table. If unmatched is NULL, a new unmatched entry was
535 	 * added to the hash table, and there was no conflict.
536 	 */
537 
538 	if (skb && unmatched && unmatched->timestamp) {
539 		unmatched->skb = skb;
540 	} else if (timestamp && unmatched && unmatched->skb) {
541 		unmatched->timestamp = timestamp;
542 	} else if (unmatched) {
543 		/* unmatched holds an older entry of the same type: either an
544 		 * skb if we are handling skb, or a timestamp if we are handling
545 		 * timestamp. We can't match that up, so save what we have.
546 		 */
547 		conflict = mlxsw_sp1_ptp_unmatched_save(mlxsw_sp, key,
548 							skb, timestamp);
549 		if (IS_ERR(conflict)) {
550 			if (skb)
551 				mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb,
552 							    key.local_port,
553 							    key.ingress, NULL);
554 		} else {
555 			/* Above, we removed an object with this key from the
556 			 * hash table, under lock, so conflict can not be a
557 			 * valid pointer.
558 			 */
559 			WARN_ON_ONCE(conflict);
560 		}
561 	}
562 
563 	spin_unlock(&mlxsw_sp->ptp_state->unmatched_lock);
564 
565 	if (unmatched)
566 		mlxsw_sp1_ptp_unmatched_finish(mlxsw_sp, unmatched);
567 
568 	rcu_read_unlock();
569 }
570 
571 static void mlxsw_sp1_ptp_got_packet(struct mlxsw_sp *mlxsw_sp,
572 				     struct sk_buff *skb, u8 local_port,
573 				     bool ingress)
574 {
575 	struct mlxsw_sp_port *mlxsw_sp_port;
576 	struct mlxsw_sp1_ptp_key key;
577 	u8 types;
578 	int err;
579 
580 	mlxsw_sp_port = mlxsw_sp->ports[local_port];
581 	if (!mlxsw_sp_port)
582 		goto immediate;
583 
584 	types = ingress ? mlxsw_sp_port->ptp.ing_types :
585 			  mlxsw_sp_port->ptp.egr_types;
586 	if (!types)
587 		goto immediate;
588 
589 	memset(&key, 0, sizeof(key));
590 	key.local_port = local_port;
591 	key.ingress = ingress;
592 
593 	err = mlxsw_sp_ptp_parse(skb, &key.domain_number, &key.message_type,
594 				 &key.sequence_id);
595 	if (err)
596 		goto immediate;
597 
598 	/* For packets whose timestamping was not enabled on this port, don't
599 	 * bother trying to match the timestamp.
600 	 */
601 	if (!((1 << key.message_type) & types))
602 		goto immediate;
603 
604 	mlxsw_sp1_ptp_got_piece(mlxsw_sp, key, skb, 0);
605 	return;
606 
607 immediate:
608 	mlxsw_sp1_ptp_packet_finish(mlxsw_sp, skb, local_port, ingress, NULL);
609 }
610 
611 void mlxsw_sp1_ptp_got_timestamp(struct mlxsw_sp *mlxsw_sp, bool ingress,
612 				 u8 local_port, u8 message_type,
613 				 u8 domain_number, u16 sequence_id,
614 				 u64 timestamp)
615 {
616 	struct mlxsw_sp_port *mlxsw_sp_port;
617 	struct mlxsw_sp1_ptp_key key;
618 	u8 types;
619 
620 	mlxsw_sp_port = mlxsw_sp->ports[local_port];
621 	if (!mlxsw_sp_port)
622 		return;
623 
624 	types = ingress ? mlxsw_sp_port->ptp.ing_types :
625 			  mlxsw_sp_port->ptp.egr_types;
626 
627 	/* For message types whose timestamping was not enabled on this port,
628 	 * don't bother with the timestamp.
629 	 */
630 	if (!((1 << message_type) & types))
631 		return;
632 
633 	memset(&key, 0, sizeof(key));
634 	key.local_port = local_port;
635 	key.domain_number = domain_number;
636 	key.message_type = message_type;
637 	key.sequence_id = sequence_id;
638 	key.ingress = ingress;
639 
640 	mlxsw_sp1_ptp_got_piece(mlxsw_sp, key, NULL, timestamp);
641 }
642 
643 void mlxsw_sp1_ptp_receive(struct mlxsw_sp *mlxsw_sp, struct sk_buff *skb,
644 			   u8 local_port)
645 {
646 	skb_reset_mac_header(skb);
647 	mlxsw_sp1_ptp_got_packet(mlxsw_sp, skb, local_port, true);
648 }
649 
650 void mlxsw_sp1_ptp_transmitted(struct mlxsw_sp *mlxsw_sp,
651 			       struct sk_buff *skb, u8 local_port)
652 {
653 	mlxsw_sp1_ptp_got_packet(mlxsw_sp, skb, local_port, false);
654 }
655 
656 static void
657 mlxsw_sp1_ptp_ht_gc_collect(struct mlxsw_sp_ptp_state *ptp_state,
658 			    struct mlxsw_sp1_ptp_unmatched *unmatched)
659 {
660 	int err;
661 
662 	/* If an unmatched entry has an SKB, it has to be handed over to the
663 	 * networking stack. This is usually done from a trap handler, which is
664 	 * invoked in a softirq context. Here we are going to do it in process
665 	 * context. If that were to be interrupted by a softirq, it could cause
666 	 * a deadlock when an attempt is made to take an already-taken lock
667 	 * somewhere along the sending path. Disable softirqs to prevent this.
668 	 */
669 	local_bh_disable();
670 
671 	spin_lock(&ptp_state->unmatched_lock);
672 	err = rhashtable_remove_fast(&ptp_state->unmatched_ht,
673 				     &unmatched->ht_node,
674 				     mlxsw_sp1_ptp_unmatched_ht_params);
675 	spin_unlock(&ptp_state->unmatched_lock);
676 
677 	if (err)
678 		/* The packet was matched with timestamp during the walk. */
679 		goto out;
680 
681 	/* mlxsw_sp1_ptp_unmatched_finish() invokes netif_receive_skb(). While
682 	 * the comment at that function states that it can only be called in
683 	 * soft IRQ context, this pattern of local_bh_disable() +
684 	 * netif_receive_skb(), in process context, is seen elsewhere in the
685 	 * kernel, notably in pktgen.
686 	 */
687 	mlxsw_sp1_ptp_unmatched_finish(ptp_state->mlxsw_sp, unmatched);
688 
689 out:
690 	local_bh_enable();
691 }
692 
693 static void mlxsw_sp1_ptp_ht_gc(struct work_struct *work)
694 {
695 	struct delayed_work *dwork = to_delayed_work(work);
696 	struct mlxsw_sp1_ptp_unmatched *unmatched;
697 	struct mlxsw_sp_ptp_state *ptp_state;
698 	struct rhashtable_iter iter;
699 	u32 gc_cycle;
700 	void *obj;
701 
702 	ptp_state = container_of(dwork, struct mlxsw_sp_ptp_state, ht_gc_dw);
703 	gc_cycle = ptp_state->gc_cycle++;
704 
705 	rhashtable_walk_enter(&ptp_state->unmatched_ht, &iter);
706 	rhashtable_walk_start(&iter);
707 	while ((obj = rhashtable_walk_next(&iter))) {
708 		if (IS_ERR(obj))
709 			continue;
710 
711 		unmatched = obj;
712 		if (unmatched->gc_cycle <= gc_cycle)
713 			mlxsw_sp1_ptp_ht_gc_collect(ptp_state, unmatched);
714 	}
715 	rhashtable_walk_stop(&iter);
716 	rhashtable_walk_exit(&iter);
717 
718 	mlxsw_core_schedule_dw(&ptp_state->ht_gc_dw,
719 			       MLXSW_SP1_PTP_HT_GC_INTERVAL);
720 }
721 
722 static int mlxsw_sp_ptp_mtptpt_set(struct mlxsw_sp *mlxsw_sp,
723 				   enum mlxsw_reg_mtptpt_trap_id trap_id,
724 				   u16 message_type)
725 {
726 	char mtptpt_pl[MLXSW_REG_MTPTPT_LEN];
727 
728 	mlxsw_reg_mtptptp_pack(mtptpt_pl, trap_id, message_type);
729 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(mtptpt), mtptpt_pl);
730 }
731 
732 static int mlxsw_sp1_ptp_set_fifo_clr_on_trap(struct mlxsw_sp *mlxsw_sp,
733 					      bool clr)
734 {
735 	char mogcr_pl[MLXSW_REG_MOGCR_LEN] = {0};
736 	int err;
737 
738 	err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(mogcr), mogcr_pl);
739 	if (err)
740 		return err;
741 
742 	mlxsw_reg_mogcr_ptp_iftc_set(mogcr_pl, clr);
743 	mlxsw_reg_mogcr_ptp_eftc_set(mogcr_pl, clr);
744 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(mogcr), mogcr_pl);
745 }
746 
747 static int mlxsw_sp1_ptp_mtpppc_set(struct mlxsw_sp *mlxsw_sp,
748 				    u16 ing_types, u16 egr_types)
749 {
750 	char mtpppc_pl[MLXSW_REG_MTPPPC_LEN];
751 
752 	mlxsw_reg_mtpppc_pack(mtpppc_pl, ing_types, egr_types);
753 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(mtpppc), mtpppc_pl);
754 }
755 
756 struct mlxsw_sp_ptp_state *mlxsw_sp1_ptp_init(struct mlxsw_sp *mlxsw_sp)
757 {
758 	struct mlxsw_sp_ptp_state *ptp_state;
759 	u16 message_type;
760 	int err;
761 
762 	ptp_state = kzalloc(sizeof(*ptp_state), GFP_KERNEL);
763 	if (!ptp_state)
764 		return ERR_PTR(-ENOMEM);
765 	ptp_state->mlxsw_sp = mlxsw_sp;
766 
767 	spin_lock_init(&ptp_state->unmatched_lock);
768 
769 	err = rhashtable_init(&ptp_state->unmatched_ht,
770 			      &mlxsw_sp1_ptp_unmatched_ht_params);
771 	if (err)
772 		goto err_hashtable_init;
773 
774 	/* Delive these message types as PTP0. */
775 	message_type = BIT(MLXSW_SP_PTP_MESSAGE_TYPE_SYNC) |
776 		       BIT(MLXSW_SP_PTP_MESSAGE_TYPE_DELAY_REQ) |
777 		       BIT(MLXSW_SP_PTP_MESSAGE_TYPE_PDELAY_REQ) |
778 		       BIT(MLXSW_SP_PTP_MESSAGE_TYPE_PDELAY_RESP);
779 	err = mlxsw_sp_ptp_mtptpt_set(mlxsw_sp, MLXSW_REG_MTPTPT_TRAP_ID_PTP0,
780 				      message_type);
781 	if (err)
782 		goto err_mtptpt_set;
783 
784 	/* Everything else is PTP1. */
785 	message_type = ~message_type;
786 	err = mlxsw_sp_ptp_mtptpt_set(mlxsw_sp, MLXSW_REG_MTPTPT_TRAP_ID_PTP1,
787 				      message_type);
788 	if (err)
789 		goto err_mtptpt1_set;
790 
791 	err = mlxsw_sp1_ptp_set_fifo_clr_on_trap(mlxsw_sp, true);
792 	if (err)
793 		goto err_fifo_clr;
794 
795 	INIT_DELAYED_WORK(&ptp_state->ht_gc_dw, mlxsw_sp1_ptp_ht_gc);
796 	mlxsw_core_schedule_dw(&ptp_state->ht_gc_dw,
797 			       MLXSW_SP1_PTP_HT_GC_INTERVAL);
798 	return ptp_state;
799 
800 err_fifo_clr:
801 	mlxsw_sp_ptp_mtptpt_set(mlxsw_sp, MLXSW_REG_MTPTPT_TRAP_ID_PTP1, 0);
802 err_mtptpt1_set:
803 	mlxsw_sp_ptp_mtptpt_set(mlxsw_sp, MLXSW_REG_MTPTPT_TRAP_ID_PTP0, 0);
804 err_mtptpt_set:
805 	rhashtable_destroy(&ptp_state->unmatched_ht);
806 err_hashtable_init:
807 	kfree(ptp_state);
808 	return ERR_PTR(err);
809 }
810 
811 void mlxsw_sp1_ptp_fini(struct mlxsw_sp_ptp_state *ptp_state)
812 {
813 	struct mlxsw_sp *mlxsw_sp = ptp_state->mlxsw_sp;
814 
815 	cancel_delayed_work_sync(&ptp_state->ht_gc_dw);
816 	mlxsw_sp1_ptp_mtpppc_set(mlxsw_sp, 0, 0);
817 	mlxsw_sp1_ptp_set_fifo_clr_on_trap(mlxsw_sp, false);
818 	mlxsw_sp_ptp_mtptpt_set(mlxsw_sp, MLXSW_REG_MTPTPT_TRAP_ID_PTP1, 0);
819 	mlxsw_sp_ptp_mtptpt_set(mlxsw_sp, MLXSW_REG_MTPTPT_TRAP_ID_PTP0, 0);
820 	rhashtable_free_and_destroy(&ptp_state->unmatched_ht,
821 				    &mlxsw_sp1_ptp_unmatched_free_fn, NULL);
822 	kfree(ptp_state);
823 }
824 
825 int mlxsw_sp1_ptp_hwtstamp_get(struct mlxsw_sp_port *mlxsw_sp_port,
826 			       struct hwtstamp_config *config)
827 {
828 	*config = mlxsw_sp_port->ptp.hwtstamp_config;
829 	return 0;
830 }
831 
832 static int mlxsw_sp_ptp_get_message_types(const struct hwtstamp_config *config,
833 					  u16 *p_ing_types, u16 *p_egr_types,
834 					  enum hwtstamp_rx_filters *p_rx_filter)
835 {
836 	enum hwtstamp_rx_filters rx_filter = config->rx_filter;
837 	enum hwtstamp_tx_types tx_type = config->tx_type;
838 	u16 ing_types = 0x00;
839 	u16 egr_types = 0x00;
840 
841 	switch (tx_type) {
842 	case HWTSTAMP_TX_OFF:
843 		egr_types = 0x00;
844 		break;
845 	case HWTSTAMP_TX_ON:
846 		egr_types = 0xff;
847 		break;
848 	case HWTSTAMP_TX_ONESTEP_SYNC:
849 		return -ERANGE;
850 	}
851 
852 	switch (rx_filter) {
853 	case HWTSTAMP_FILTER_NONE:
854 		ing_types = 0x00;
855 		break;
856 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
857 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
858 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
859 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
860 		ing_types = 0x01;
861 		break;
862 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
863 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
864 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
865 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
866 		ing_types = 0x02;
867 		break;
868 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
869 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
870 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
871 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
872 		ing_types = 0x0f;
873 		break;
874 	case HWTSTAMP_FILTER_ALL:
875 		ing_types = 0xff;
876 		break;
877 	case HWTSTAMP_FILTER_SOME:
878 	case HWTSTAMP_FILTER_NTP_ALL:
879 		return -ERANGE;
880 	}
881 
882 	*p_ing_types = ing_types;
883 	*p_egr_types = egr_types;
884 	*p_rx_filter = rx_filter;
885 	return 0;
886 }
887 
888 static int mlxsw_sp1_ptp_mtpppc_update(struct mlxsw_sp_port *mlxsw_sp_port,
889 				       u16 ing_types, u16 egr_types)
890 {
891 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
892 	struct mlxsw_sp_port *tmp;
893 	int i;
894 
895 	/* MTPPPC configures timestamping globally, not per port. Find the
896 	 * configuration that contains all configured timestamping requests.
897 	 */
898 	for (i = 1; i < mlxsw_core_max_ports(mlxsw_sp->core); i++) {
899 		tmp = mlxsw_sp->ports[i];
900 		if (tmp && tmp != mlxsw_sp_port) {
901 			ing_types |= tmp->ptp.ing_types;
902 			egr_types |= tmp->ptp.egr_types;
903 		}
904 	}
905 
906 	return mlxsw_sp1_ptp_mtpppc_set(mlxsw_sp_port->mlxsw_sp,
907 				       ing_types, egr_types);
908 }
909 
910 int mlxsw_sp1_ptp_hwtstamp_set(struct mlxsw_sp_port *mlxsw_sp_port,
911 			       struct hwtstamp_config *config)
912 {
913 	enum hwtstamp_rx_filters rx_filter;
914 	u16 ing_types;
915 	u16 egr_types;
916 	int err;
917 
918 	err = mlxsw_sp_ptp_get_message_types(config, &ing_types, &egr_types,
919 					     &rx_filter);
920 	if (err)
921 		return err;
922 
923 	err = mlxsw_sp1_ptp_mtpppc_update(mlxsw_sp_port, ing_types, egr_types);
924 	if (err)
925 		return err;
926 
927 	mlxsw_sp_port->ptp.hwtstamp_config = *config;
928 	mlxsw_sp_port->ptp.ing_types = ing_types;
929 	mlxsw_sp_port->ptp.egr_types = egr_types;
930 
931 	/* Notify the ioctl caller what we are actually timestamping. */
932 	config->rx_filter = rx_filter;
933 
934 	return 0;
935 }
936 
937 int mlxsw_sp1_ptp_get_ts_info(struct mlxsw_sp *mlxsw_sp,
938 			      struct ethtool_ts_info *info)
939 {
940 	info->phc_index = ptp_clock_index(mlxsw_sp->clock->ptp);
941 
942 	info->so_timestamping = SOF_TIMESTAMPING_TX_HARDWARE |
943 				SOF_TIMESTAMPING_RX_HARDWARE |
944 				SOF_TIMESTAMPING_RAW_HARDWARE;
945 
946 	info->tx_types = BIT(HWTSTAMP_TX_OFF) |
947 			 BIT(HWTSTAMP_TX_ON);
948 
949 	info->rx_filters = BIT(HWTSTAMP_FILTER_NONE) |
950 			   BIT(HWTSTAMP_FILTER_ALL);
951 
952 	return 0;
953 }
954