xref: /openbmc/linux/drivers/net/ethernet/sfc/ptp.c (revision 77d84ff8)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2011-2013 Solarflare Communications Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation, incorporated herein by reference.
8  */
9 
10 /* Theory of operation:
11  *
12  * PTP support is assisted by firmware running on the MC, which provides
13  * the hardware timestamping capabilities.  Both transmitted and received
14  * PTP event packets are queued onto internal queues for subsequent processing;
15  * this is because the MC operations are relatively long and would block
16  * block NAPI/interrupt operation.
17  *
18  * Receive event processing:
19  *	The event contains the packet's UUID and sequence number, together
20  *	with the hardware timestamp.  The PTP receive packet queue is searched
21  *	for this UUID/sequence number and, if found, put on a pending queue.
22  *	Packets not matching are delivered without timestamps (MCDI events will
23  *	always arrive after the actual packet).
24  *	It is important for the operation of the PTP protocol that the ordering
25  *	of packets between the event and general port is maintained.
26  *
27  * Work queue processing:
28  *	If work waiting, synchronise host/hardware time
29  *
30  *	Transmit: send packet through MC, which returns the transmission time
31  *	that is converted to an appropriate timestamp.
32  *
33  *	Receive: the packet's reception time is converted to an appropriate
34  *	timestamp.
35  */
36 #include <linux/ip.h>
37 #include <linux/udp.h>
38 #include <linux/time.h>
39 #include <linux/ktime.h>
40 #include <linux/module.h>
41 #include <linux/net_tstamp.h>
42 #include <linux/pps_kernel.h>
43 #include <linux/ptp_clock_kernel.h>
44 #include "net_driver.h"
45 #include "efx.h"
46 #include "mcdi.h"
47 #include "mcdi_pcol.h"
48 #include "io.h"
49 #include "farch_regs.h"
50 #include "nic.h"
51 
52 /* Maximum number of events expected to make up a PTP event */
53 #define	MAX_EVENT_FRAGS			3
54 
55 /* Maximum delay, ms, to begin synchronisation */
56 #define	MAX_SYNCHRONISE_WAIT_MS		2
57 
58 /* How long, at most, to spend synchronising */
59 #define	SYNCHRONISE_PERIOD_NS		250000
60 
61 /* How often to update the shared memory time */
62 #define	SYNCHRONISATION_GRANULARITY_NS	200
63 
64 /* Minimum permitted length of a (corrected) synchronisation time */
65 #define	MIN_SYNCHRONISATION_NS		120
66 
67 /* Maximum permitted length of a (corrected) synchronisation time */
68 #define	MAX_SYNCHRONISATION_NS		1000
69 
70 /* How many (MC) receive events that can be queued */
71 #define	MAX_RECEIVE_EVENTS		8
72 
73 /* Length of (modified) moving average. */
74 #define	AVERAGE_LENGTH			16
75 
76 /* How long an unmatched event or packet can be held */
77 #define PKT_EVENT_LIFETIME_MS		10
78 
79 /* Offsets into PTP packet for identification.  These offsets are from the
80  * start of the IP header, not the MAC header.  Note that neither PTP V1 nor
81  * PTP V2 permit the use of IPV4 options.
82  */
83 #define PTP_DPORT_OFFSET	22
84 
85 #define PTP_V1_VERSION_LENGTH	2
86 #define PTP_V1_VERSION_OFFSET	28
87 
88 #define PTP_V1_UUID_LENGTH	6
89 #define PTP_V1_UUID_OFFSET	50
90 
91 #define PTP_V1_SEQUENCE_LENGTH	2
92 #define PTP_V1_SEQUENCE_OFFSET	58
93 
94 /* The minimum length of a PTP V1 packet for offsets, etc. to be valid:
95  * includes IP header.
96  */
97 #define	PTP_V1_MIN_LENGTH	64
98 
99 #define PTP_V2_VERSION_LENGTH	1
100 #define PTP_V2_VERSION_OFFSET	29
101 
102 #define PTP_V2_UUID_LENGTH	8
103 #define PTP_V2_UUID_OFFSET	48
104 
105 /* Although PTP V2 UUIDs are comprised a ClockIdentity (8) and PortNumber (2),
106  * the MC only captures the last six bytes of the clock identity. These values
107  * reflect those, not the ones used in the standard.  The standard permits
108  * mapping of V1 UUIDs to V2 UUIDs with these same values.
109  */
110 #define PTP_V2_MC_UUID_LENGTH	6
111 #define PTP_V2_MC_UUID_OFFSET	50
112 
113 #define PTP_V2_SEQUENCE_LENGTH	2
114 #define PTP_V2_SEQUENCE_OFFSET	58
115 
116 /* The minimum length of a PTP V2 packet for offsets, etc. to be valid:
117  * includes IP header.
118  */
119 #define	PTP_V2_MIN_LENGTH	63
120 
121 #define	PTP_MIN_LENGTH		63
122 
123 #define PTP_ADDRESS		0xe0000181	/* 224.0.1.129 */
124 #define PTP_EVENT_PORT		319
125 #define PTP_GENERAL_PORT	320
126 
127 /* Annoyingly the format of the version numbers are different between
128  * versions 1 and 2 so it isn't possible to simply look for 1 or 2.
129  */
130 #define	PTP_VERSION_V1		1
131 
132 #define	PTP_VERSION_V2		2
133 #define	PTP_VERSION_V2_MASK	0x0f
134 
135 enum ptp_packet_state {
136 	PTP_PACKET_STATE_UNMATCHED = 0,
137 	PTP_PACKET_STATE_MATCHED,
138 	PTP_PACKET_STATE_TIMED_OUT,
139 	PTP_PACKET_STATE_MATCH_UNWANTED
140 };
141 
142 /* NIC synchronised with single word of time only comprising
143  * partial seconds and full nanoseconds: 10^9 ~ 2^30 so 2 bits for seconds.
144  */
145 #define	MC_NANOSECOND_BITS	30
146 #define	MC_NANOSECOND_MASK	((1 << MC_NANOSECOND_BITS) - 1)
147 #define	MC_SECOND_MASK		((1 << (32 - MC_NANOSECOND_BITS)) - 1)
148 
149 /* Maximum parts-per-billion adjustment that is acceptable */
150 #define MAX_PPB			1000000
151 
152 /* Number of bits required to hold the above */
153 #define	MAX_PPB_BITS		20
154 
155 /* Number of extra bits allowed when calculating fractional ns.
156  * EXTRA_BITS + MC_CMD_PTP_IN_ADJUST_BITS + MAX_PPB_BITS should
157  * be less than 63.
158  */
159 #define	PPB_EXTRA_BITS		2
160 
161 /* Precalculate scale word to avoid long long division at runtime */
162 #define	PPB_SCALE_WORD	((1LL << (PPB_EXTRA_BITS + MC_CMD_PTP_IN_ADJUST_BITS +\
163 			MAX_PPB_BITS)) / 1000000000LL)
164 
165 #define PTP_SYNC_ATTEMPTS	4
166 
167 /**
168  * struct efx_ptp_match - Matching structure, stored in sk_buff's cb area.
169  * @words: UUID and (partial) sequence number
170  * @expiry: Time after which the packet should be delivered irrespective of
171  *            event arrival.
172  * @state: The state of the packet - whether it is ready for processing or
173  *         whether that is of no interest.
174  */
175 struct efx_ptp_match {
176 	u32 words[DIV_ROUND_UP(PTP_V1_UUID_LENGTH, 4)];
177 	unsigned long expiry;
178 	enum ptp_packet_state state;
179 };
180 
181 /**
182  * struct efx_ptp_event_rx - A PTP receive event (from MC)
183  * @seq0: First part of (PTP) UUID
184  * @seq1: Second part of (PTP) UUID and sequence number
185  * @hwtimestamp: Event timestamp
186  */
187 struct efx_ptp_event_rx {
188 	struct list_head link;
189 	u32 seq0;
190 	u32 seq1;
191 	ktime_t hwtimestamp;
192 	unsigned long expiry;
193 };
194 
195 /**
196  * struct efx_ptp_timeset - Synchronisation between host and MC
197  * @host_start: Host time immediately before hardware timestamp taken
198  * @seconds: Hardware timestamp, seconds
199  * @nanoseconds: Hardware timestamp, nanoseconds
200  * @host_end: Host time immediately after hardware timestamp taken
201  * @waitns: Number of nanoseconds between hardware timestamp being read and
202  *          host end time being seen
203  * @window: Difference of host_end and host_start
204  * @valid: Whether this timeset is valid
205  */
206 struct efx_ptp_timeset {
207 	u32 host_start;
208 	u32 seconds;
209 	u32 nanoseconds;
210 	u32 host_end;
211 	u32 waitns;
212 	u32 window;	/* Derived: end - start, allowing for wrap */
213 };
214 
215 /**
216  * struct efx_ptp_data - Precision Time Protocol (PTP) state
217  * @channel: The PTP channel
218  * @rxq: Receive queue (awaiting timestamps)
219  * @txq: Transmit queue
220  * @evt_list: List of MC receive events awaiting packets
221  * @evt_free_list: List of free events
222  * @evt_lock: Lock for manipulating evt_list and evt_free_list
223  * @evt_overflow: Boolean indicating that event list has overflowed
224  * @rx_evts: Instantiated events (on evt_list and evt_free_list)
225  * @workwq: Work queue for processing pending PTP operations
226  * @work: Work task
227  * @reset_required: A serious error has occurred and the PTP task needs to be
228  *                  reset (disable, enable).
229  * @rxfilter_event: Receive filter when operating
230  * @rxfilter_general: Receive filter when operating
231  * @config: Current timestamp configuration
232  * @enabled: PTP operation enabled
233  * @mode: Mode in which PTP operating (PTP version)
234  * @evt_frags: Partly assembled PTP events
235  * @evt_frag_idx: Current fragment number
236  * @evt_code: Last event code
237  * @start: Address at which MC indicates ready for synchronisation
238  * @host_time_pps: Host time at last PPS
239  * @last_sync_ns: Last number of nanoseconds between readings when synchronising
240  * @base_sync_ns: Number of nanoseconds for last synchronisation.
241  * @base_sync_valid: Whether base_sync_time is valid.
242  * @current_adjfreq: Current ppb adjustment.
243  * @phc_clock: Pointer to registered phc device
244  * @phc_clock_info: Registration structure for phc device
245  * @pps_work: pps work task for handling pps events
246  * @pps_workwq: pps work queue
247  * @nic_ts_enabled: Flag indicating if NIC generated TS events are handled
248  * @txbuf: Buffer for use when transmitting (PTP) packets to MC (avoids
249  *         allocations in main data path).
250  * @debug_ptp_dir: PTP debugfs directory
251  * @missed_rx_sync: Number of packets received without syncrhonisation.
252  * @good_syncs: Number of successful synchronisations.
253  * @no_time_syncs: Number of synchronisations with no good times.
254  * @bad_sync_durations: Number of synchronisations with bad durations.
255  * @bad_syncs: Number of failed synchronisations.
256  * @last_sync_time: Number of nanoseconds for last synchronisation.
257  * @sync_timeouts: Number of synchronisation timeouts
258  * @fast_syncs: Number of synchronisations requiring short delay
259  * @min_sync_delta: Minimum time between event and synchronisation
260  * @max_sync_delta: Maximum time between event and synchronisation
261  * @average_sync_delta: Average time between event and synchronisation.
262  *                      Modified moving average.
263  * @last_sync_delta: Last time between event and synchronisation
264  * @mc_stats: Context value for MC statistics
265  * @timeset: Last set of synchronisation statistics.
266  */
267 struct efx_ptp_data {
268 	struct efx_channel *channel;
269 	struct sk_buff_head rxq;
270 	struct sk_buff_head txq;
271 	struct list_head evt_list;
272 	struct list_head evt_free_list;
273 	spinlock_t evt_lock;
274 	bool evt_overflow;
275 	struct efx_ptp_event_rx rx_evts[MAX_RECEIVE_EVENTS];
276 	struct workqueue_struct *workwq;
277 	struct work_struct work;
278 	bool reset_required;
279 	u32 rxfilter_event;
280 	u32 rxfilter_general;
281 	bool rxfilter_installed;
282 	struct hwtstamp_config config;
283 	bool enabled;
284 	unsigned int mode;
285 	efx_qword_t evt_frags[MAX_EVENT_FRAGS];
286 	int evt_frag_idx;
287 	int evt_code;
288 	struct efx_buffer start;
289 	struct pps_event_time host_time_pps;
290 	unsigned last_sync_ns;
291 	unsigned base_sync_ns;
292 	bool base_sync_valid;
293 	s64 current_adjfreq;
294 	struct ptp_clock *phc_clock;
295 	struct ptp_clock_info phc_clock_info;
296 	struct work_struct pps_work;
297 	struct workqueue_struct *pps_workwq;
298 	bool nic_ts_enabled;
299 	MCDI_DECLARE_BUF(txbuf, MC_CMD_PTP_IN_TRANSMIT_LENMAX);
300 	struct efx_ptp_timeset
301 	timeset[MC_CMD_PTP_OUT_SYNCHRONIZE_TIMESET_MAXNUM];
302 };
303 
304 static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta);
305 static int efx_phc_adjtime(struct ptp_clock_info *ptp, s64 delta);
306 static int efx_phc_gettime(struct ptp_clock_info *ptp, struct timespec *ts);
307 static int efx_phc_settime(struct ptp_clock_info *ptp,
308 			   const struct timespec *e_ts);
309 static int efx_phc_enable(struct ptp_clock_info *ptp,
310 			  struct ptp_clock_request *request, int on);
311 
312 /* Enable MCDI PTP support. */
313 static int efx_ptp_enable(struct efx_nic *efx)
314 {
315 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_ENABLE_LEN);
316 
317 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_ENABLE);
318 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
319 	MCDI_SET_DWORD(inbuf, PTP_IN_ENABLE_QUEUE,
320 		       efx->ptp_data->channel->channel);
321 	MCDI_SET_DWORD(inbuf, PTP_IN_ENABLE_MODE, efx->ptp_data->mode);
322 
323 	return efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
324 			    NULL, 0, NULL);
325 }
326 
327 /* Disable MCDI PTP support.
328  *
329  * Note that this function should never rely on the presence of ptp_data -
330  * may be called before that exists.
331  */
332 static int efx_ptp_disable(struct efx_nic *efx)
333 {
334 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_DISABLE_LEN);
335 
336 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_DISABLE);
337 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
338 	return efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
339 			    NULL, 0, NULL);
340 }
341 
342 static void efx_ptp_deliver_rx_queue(struct sk_buff_head *q)
343 {
344 	struct sk_buff *skb;
345 
346 	while ((skb = skb_dequeue(q))) {
347 		local_bh_disable();
348 		netif_receive_skb(skb);
349 		local_bh_enable();
350 	}
351 }
352 
353 static void efx_ptp_handle_no_channel(struct efx_nic *efx)
354 {
355 	netif_err(efx, drv, efx->net_dev,
356 		  "ERROR: PTP requires MSI-X and 1 additional interrupt"
357 		  "vector. PTP disabled\n");
358 }
359 
360 /* Repeatedly send the host time to the MC which will capture the hardware
361  * time.
362  */
363 static void efx_ptp_send_times(struct efx_nic *efx,
364 			       struct pps_event_time *last_time)
365 {
366 	struct pps_event_time now;
367 	struct timespec limit;
368 	struct efx_ptp_data *ptp = efx->ptp_data;
369 	struct timespec start;
370 	int *mc_running = ptp->start.addr;
371 
372 	pps_get_ts(&now);
373 	start = now.ts_real;
374 	limit = now.ts_real;
375 	timespec_add_ns(&limit, SYNCHRONISE_PERIOD_NS);
376 
377 	/* Write host time for specified period or until MC is done */
378 	while ((timespec_compare(&now.ts_real, &limit) < 0) &&
379 	       ACCESS_ONCE(*mc_running)) {
380 		struct timespec update_time;
381 		unsigned int host_time;
382 
383 		/* Don't update continuously to avoid saturating the PCIe bus */
384 		update_time = now.ts_real;
385 		timespec_add_ns(&update_time, SYNCHRONISATION_GRANULARITY_NS);
386 		do {
387 			pps_get_ts(&now);
388 		} while ((timespec_compare(&now.ts_real, &update_time) < 0) &&
389 			 ACCESS_ONCE(*mc_running));
390 
391 		/* Synchronise NIC with single word of time only */
392 		host_time = (now.ts_real.tv_sec << MC_NANOSECOND_BITS |
393 			     now.ts_real.tv_nsec);
394 		/* Update host time in NIC memory */
395 		efx->type->ptp_write_host_time(efx, host_time);
396 	}
397 	*last_time = now;
398 }
399 
400 /* Read a timeset from the MC's results and partial process. */
401 static void efx_ptp_read_timeset(MCDI_DECLARE_STRUCT_PTR(data),
402 				 struct efx_ptp_timeset *timeset)
403 {
404 	unsigned start_ns, end_ns;
405 
406 	timeset->host_start = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_HOSTSTART);
407 	timeset->seconds = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_SECONDS);
408 	timeset->nanoseconds = MCDI_DWORD(data,
409 					 PTP_OUT_SYNCHRONIZE_NANOSECONDS);
410 	timeset->host_end = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_HOSTEND),
411 	timeset->waitns = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_WAITNS);
412 
413 	/* Ignore seconds */
414 	start_ns = timeset->host_start & MC_NANOSECOND_MASK;
415 	end_ns = timeset->host_end & MC_NANOSECOND_MASK;
416 	/* Allow for rollover */
417 	if (end_ns < start_ns)
418 		end_ns += NSEC_PER_SEC;
419 	/* Determine duration of operation */
420 	timeset->window = end_ns - start_ns;
421 }
422 
423 /* Process times received from MC.
424  *
425  * Extract times from returned results, and establish the minimum value
426  * seen.  The minimum value represents the "best" possible time and events
427  * too much greater than this are rejected - the machine is, perhaps, too
428  * busy. A number of readings are taken so that, hopefully, at least one good
429  * synchronisation will be seen in the results.
430  */
431 static int
432 efx_ptp_process_times(struct efx_nic *efx, MCDI_DECLARE_STRUCT_PTR(synch_buf),
433 		      size_t response_length,
434 		      const struct pps_event_time *last_time)
435 {
436 	unsigned number_readings =
437 		MCDI_VAR_ARRAY_LEN(response_length,
438 				   PTP_OUT_SYNCHRONIZE_TIMESET);
439 	unsigned i;
440 	unsigned total;
441 	unsigned ngood = 0;
442 	unsigned last_good = 0;
443 	struct efx_ptp_data *ptp = efx->ptp_data;
444 	u32 last_sec;
445 	u32 start_sec;
446 	struct timespec delta;
447 
448 	if (number_readings == 0)
449 		return -EAGAIN;
450 
451 	/* Read the set of results and increment stats for any results that
452 	 * appera to be erroneous.
453 	 */
454 	for (i = 0; i < number_readings; i++) {
455 		efx_ptp_read_timeset(
456 			MCDI_ARRAY_STRUCT_PTR(synch_buf,
457 					      PTP_OUT_SYNCHRONIZE_TIMESET, i),
458 			&ptp->timeset[i]);
459 	}
460 
461 	/* Find the last good host-MC synchronization result. The MC times
462 	 * when it finishes reading the host time so the corrected window time
463 	 * should be fairly constant for a given platform.
464 	 */
465 	total = 0;
466 	for (i = 0; i < number_readings; i++)
467 		if (ptp->timeset[i].window > ptp->timeset[i].waitns) {
468 			unsigned win;
469 
470 			win = ptp->timeset[i].window - ptp->timeset[i].waitns;
471 			if (win >= MIN_SYNCHRONISATION_NS &&
472 			    win < MAX_SYNCHRONISATION_NS) {
473 				total += ptp->timeset[i].window;
474 				ngood++;
475 				last_good = i;
476 			}
477 		}
478 
479 	if (ngood == 0) {
480 		netif_warn(efx, drv, efx->net_dev,
481 			   "PTP no suitable synchronisations %dns\n",
482 			   ptp->base_sync_ns);
483 		return -EAGAIN;
484 	}
485 
486 	/* Average minimum this synchronisation */
487 	ptp->last_sync_ns = DIV_ROUND_UP(total, ngood);
488 	if (!ptp->base_sync_valid || (ptp->last_sync_ns < ptp->base_sync_ns)) {
489 		ptp->base_sync_valid = true;
490 		ptp->base_sync_ns = ptp->last_sync_ns;
491 	}
492 
493 	/* Calculate delay from actual PPS to last_time */
494 	delta.tv_nsec =
495 		ptp->timeset[last_good].nanoseconds +
496 		last_time->ts_real.tv_nsec -
497 		(ptp->timeset[last_good].host_start & MC_NANOSECOND_MASK);
498 
499 	/* It is possible that the seconds rolled over between taking
500 	 * the start reading and the last value written by the host.  The
501 	 * timescales are such that a gap of more than one second is never
502 	 * expected.
503 	 */
504 	start_sec = ptp->timeset[last_good].host_start >> MC_NANOSECOND_BITS;
505 	last_sec = last_time->ts_real.tv_sec & MC_SECOND_MASK;
506 	if (start_sec != last_sec) {
507 		if (((start_sec + 1) & MC_SECOND_MASK) != last_sec) {
508 			netif_warn(efx, hw, efx->net_dev,
509 				   "PTP bad synchronisation seconds\n");
510 			return -EAGAIN;
511 		} else {
512 			delta.tv_sec = 1;
513 		}
514 	} else {
515 		delta.tv_sec = 0;
516 	}
517 
518 	ptp->host_time_pps = *last_time;
519 	pps_sub_ts(&ptp->host_time_pps, delta);
520 
521 	return 0;
522 }
523 
524 /* Synchronize times between the host and the MC */
525 static int efx_ptp_synchronize(struct efx_nic *efx, unsigned int num_readings)
526 {
527 	struct efx_ptp_data *ptp = efx->ptp_data;
528 	MCDI_DECLARE_BUF(synch_buf, MC_CMD_PTP_OUT_SYNCHRONIZE_LENMAX);
529 	size_t response_length;
530 	int rc;
531 	unsigned long timeout;
532 	struct pps_event_time last_time = {};
533 	unsigned int loops = 0;
534 	int *start = ptp->start.addr;
535 
536 	MCDI_SET_DWORD(synch_buf, PTP_IN_OP, MC_CMD_PTP_OP_SYNCHRONIZE);
537 	MCDI_SET_DWORD(synch_buf, PTP_IN_PERIPH_ID, 0);
538 	MCDI_SET_DWORD(synch_buf, PTP_IN_SYNCHRONIZE_NUMTIMESETS,
539 		       num_readings);
540 	MCDI_SET_QWORD(synch_buf, PTP_IN_SYNCHRONIZE_START_ADDR,
541 		       ptp->start.dma_addr);
542 
543 	/* Clear flag that signals MC ready */
544 	ACCESS_ONCE(*start) = 0;
545 	rc = efx_mcdi_rpc_start(efx, MC_CMD_PTP, synch_buf,
546 				MC_CMD_PTP_IN_SYNCHRONIZE_LEN);
547 	EFX_BUG_ON_PARANOID(rc);
548 
549 	/* Wait for start from MCDI (or timeout) */
550 	timeout = jiffies + msecs_to_jiffies(MAX_SYNCHRONISE_WAIT_MS);
551 	while (!ACCESS_ONCE(*start) && (time_before(jiffies, timeout))) {
552 		udelay(20);	/* Usually start MCDI execution quickly */
553 		loops++;
554 	}
555 
556 	if (ACCESS_ONCE(*start))
557 		efx_ptp_send_times(efx, &last_time);
558 
559 	/* Collect results */
560 	rc = efx_mcdi_rpc_finish(efx, MC_CMD_PTP,
561 				 MC_CMD_PTP_IN_SYNCHRONIZE_LEN,
562 				 synch_buf, sizeof(synch_buf),
563 				 &response_length);
564 	if (rc == 0)
565 		rc = efx_ptp_process_times(efx, synch_buf, response_length,
566 					   &last_time);
567 
568 	return rc;
569 }
570 
571 /* Transmit a PTP packet, via the MCDI interface, to the wire. */
572 static int efx_ptp_xmit_skb(struct efx_nic *efx, struct sk_buff *skb)
573 {
574 	struct efx_ptp_data *ptp_data = efx->ptp_data;
575 	struct skb_shared_hwtstamps timestamps;
576 	int rc = -EIO;
577 	MCDI_DECLARE_BUF(txtime, MC_CMD_PTP_OUT_TRANSMIT_LEN);
578 	size_t len;
579 
580 	MCDI_SET_DWORD(ptp_data->txbuf, PTP_IN_OP, MC_CMD_PTP_OP_TRANSMIT);
581 	MCDI_SET_DWORD(ptp_data->txbuf, PTP_IN_PERIPH_ID, 0);
582 	MCDI_SET_DWORD(ptp_data->txbuf, PTP_IN_TRANSMIT_LENGTH, skb->len);
583 	if (skb_shinfo(skb)->nr_frags != 0) {
584 		rc = skb_linearize(skb);
585 		if (rc != 0)
586 			goto fail;
587 	}
588 
589 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
590 		rc = skb_checksum_help(skb);
591 		if (rc != 0)
592 			goto fail;
593 	}
594 	skb_copy_from_linear_data(skb,
595 				  MCDI_PTR(ptp_data->txbuf,
596 					   PTP_IN_TRANSMIT_PACKET),
597 				  skb->len);
598 	rc = efx_mcdi_rpc(efx, MC_CMD_PTP,
599 			  ptp_data->txbuf, MC_CMD_PTP_IN_TRANSMIT_LEN(skb->len),
600 			  txtime, sizeof(txtime), &len);
601 	if (rc != 0)
602 		goto fail;
603 
604 	memset(&timestamps, 0, sizeof(timestamps));
605 	timestamps.hwtstamp = ktime_set(
606 		MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_SECONDS),
607 		MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_NANOSECONDS));
608 
609 	skb_tstamp_tx(skb, &timestamps);
610 
611 	rc = 0;
612 
613 fail:
614 	dev_kfree_skb(skb);
615 
616 	return rc;
617 }
618 
619 static void efx_ptp_drop_time_expired_events(struct efx_nic *efx)
620 {
621 	struct efx_ptp_data *ptp = efx->ptp_data;
622 	struct list_head *cursor;
623 	struct list_head *next;
624 
625 	/* Drop time-expired events */
626 	spin_lock_bh(&ptp->evt_lock);
627 	if (!list_empty(&ptp->evt_list)) {
628 		list_for_each_safe(cursor, next, &ptp->evt_list) {
629 			struct efx_ptp_event_rx *evt;
630 
631 			evt = list_entry(cursor, struct efx_ptp_event_rx,
632 					 link);
633 			if (time_after(jiffies, evt->expiry)) {
634 				list_move(&evt->link, &ptp->evt_free_list);
635 				netif_warn(efx, hw, efx->net_dev,
636 					   "PTP rx event dropped\n");
637 			}
638 		}
639 	}
640 	/* If the event overflow flag is set and the event list is now empty
641 	 * clear the flag to re-enable the overflow warning message.
642 	 */
643 	if (ptp->evt_overflow && list_empty(&ptp->evt_list))
644 		ptp->evt_overflow = false;
645 	spin_unlock_bh(&ptp->evt_lock);
646 }
647 
648 static enum ptp_packet_state efx_ptp_match_rx(struct efx_nic *efx,
649 					      struct sk_buff *skb)
650 {
651 	struct efx_ptp_data *ptp = efx->ptp_data;
652 	bool evts_waiting;
653 	struct list_head *cursor;
654 	struct list_head *next;
655 	struct efx_ptp_match *match;
656 	enum ptp_packet_state rc = PTP_PACKET_STATE_UNMATCHED;
657 
658 	spin_lock_bh(&ptp->evt_lock);
659 	evts_waiting = !list_empty(&ptp->evt_list);
660 	spin_unlock_bh(&ptp->evt_lock);
661 
662 	if (!evts_waiting)
663 		return PTP_PACKET_STATE_UNMATCHED;
664 
665 	match = (struct efx_ptp_match *)skb->cb;
666 	/* Look for a matching timestamp in the event queue */
667 	spin_lock_bh(&ptp->evt_lock);
668 	list_for_each_safe(cursor, next, &ptp->evt_list) {
669 		struct efx_ptp_event_rx *evt;
670 
671 		evt = list_entry(cursor, struct efx_ptp_event_rx, link);
672 		if ((evt->seq0 == match->words[0]) &&
673 		    (evt->seq1 == match->words[1])) {
674 			struct skb_shared_hwtstamps *timestamps;
675 
676 			/* Match - add in hardware timestamp */
677 			timestamps = skb_hwtstamps(skb);
678 			timestamps->hwtstamp = evt->hwtimestamp;
679 
680 			match->state = PTP_PACKET_STATE_MATCHED;
681 			rc = PTP_PACKET_STATE_MATCHED;
682 			list_move(&evt->link, &ptp->evt_free_list);
683 			break;
684 		}
685 	}
686 	/* If the event overflow flag is set and the event list is now empty
687 	 * clear the flag to re-enable the overflow warning message.
688 	 */
689 	if (ptp->evt_overflow && list_empty(&ptp->evt_list))
690 		ptp->evt_overflow = false;
691 	spin_unlock_bh(&ptp->evt_lock);
692 
693 	return rc;
694 }
695 
696 /* Process any queued receive events and corresponding packets
697  *
698  * q is returned with all the packets that are ready for delivery.
699  * true is returned if at least one of those packets requires
700  * synchronisation.
701  */
702 static bool efx_ptp_process_events(struct efx_nic *efx, struct sk_buff_head *q)
703 {
704 	struct efx_ptp_data *ptp = efx->ptp_data;
705 	bool rc = false;
706 	struct sk_buff *skb;
707 
708 	while ((skb = skb_dequeue(&ptp->rxq))) {
709 		struct efx_ptp_match *match;
710 
711 		match = (struct efx_ptp_match *)skb->cb;
712 		if (match->state == PTP_PACKET_STATE_MATCH_UNWANTED) {
713 			__skb_queue_tail(q, skb);
714 		} else if (efx_ptp_match_rx(efx, skb) ==
715 			   PTP_PACKET_STATE_MATCHED) {
716 			rc = true;
717 			__skb_queue_tail(q, skb);
718 		} else if (time_after(jiffies, match->expiry)) {
719 			match->state = PTP_PACKET_STATE_TIMED_OUT;
720 			if (net_ratelimit())
721 				netif_warn(efx, rx_err, efx->net_dev,
722 					   "PTP packet - no timestamp seen\n");
723 			__skb_queue_tail(q, skb);
724 		} else {
725 			/* Replace unprocessed entry and stop */
726 			skb_queue_head(&ptp->rxq, skb);
727 			break;
728 		}
729 	}
730 
731 	return rc;
732 }
733 
734 /* Complete processing of a received packet */
735 static inline void efx_ptp_process_rx(struct efx_nic *efx, struct sk_buff *skb)
736 {
737 	local_bh_disable();
738 	netif_receive_skb(skb);
739 	local_bh_enable();
740 }
741 
742 static int efx_ptp_start(struct efx_nic *efx)
743 {
744 	struct efx_ptp_data *ptp = efx->ptp_data;
745 	struct efx_filter_spec rxfilter;
746 	int rc;
747 
748 	ptp->reset_required = false;
749 
750 	/* Must filter on both event and general ports to ensure
751 	 * that there is no packet re-ordering.
752 	 */
753 	efx_filter_init_rx(&rxfilter, EFX_FILTER_PRI_REQUIRED, 0,
754 			   efx_rx_queue_index(
755 				   efx_channel_get_rx_queue(ptp->channel)));
756 	rc = efx_filter_set_ipv4_local(&rxfilter, IPPROTO_UDP,
757 				       htonl(PTP_ADDRESS),
758 				       htons(PTP_EVENT_PORT));
759 	if (rc != 0)
760 		return rc;
761 
762 	rc = efx_filter_insert_filter(efx, &rxfilter, true);
763 	if (rc < 0)
764 		return rc;
765 	ptp->rxfilter_event = rc;
766 
767 	efx_filter_init_rx(&rxfilter, EFX_FILTER_PRI_REQUIRED, 0,
768 			   efx_rx_queue_index(
769 				   efx_channel_get_rx_queue(ptp->channel)));
770 	rc = efx_filter_set_ipv4_local(&rxfilter, IPPROTO_UDP,
771 				       htonl(PTP_ADDRESS),
772 				       htons(PTP_GENERAL_PORT));
773 	if (rc != 0)
774 		goto fail;
775 
776 	rc = efx_filter_insert_filter(efx, &rxfilter, true);
777 	if (rc < 0)
778 		goto fail;
779 	ptp->rxfilter_general = rc;
780 
781 	rc = efx_ptp_enable(efx);
782 	if (rc != 0)
783 		goto fail2;
784 
785 	ptp->evt_frag_idx = 0;
786 	ptp->current_adjfreq = 0;
787 	ptp->rxfilter_installed = true;
788 
789 	return 0;
790 
791 fail2:
792 	efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
793 				  ptp->rxfilter_general);
794 fail:
795 	efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
796 				  ptp->rxfilter_event);
797 
798 	return rc;
799 }
800 
801 static int efx_ptp_stop(struct efx_nic *efx)
802 {
803 	struct efx_ptp_data *ptp = efx->ptp_data;
804 	struct list_head *cursor;
805 	struct list_head *next;
806 	int rc;
807 
808 	if (ptp == NULL)
809 		return 0;
810 
811 	rc = efx_ptp_disable(efx);
812 
813 	if (ptp->rxfilter_installed) {
814 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
815 					  ptp->rxfilter_general);
816 		efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
817 					  ptp->rxfilter_event);
818 		ptp->rxfilter_installed = false;
819 	}
820 
821 	/* Make sure RX packets are really delivered */
822 	efx_ptp_deliver_rx_queue(&efx->ptp_data->rxq);
823 	skb_queue_purge(&efx->ptp_data->txq);
824 
825 	/* Drop any pending receive events */
826 	spin_lock_bh(&efx->ptp_data->evt_lock);
827 	list_for_each_safe(cursor, next, &efx->ptp_data->evt_list) {
828 		list_move(cursor, &efx->ptp_data->evt_free_list);
829 	}
830 	ptp->evt_overflow = false;
831 	spin_unlock_bh(&efx->ptp_data->evt_lock);
832 
833 	return rc;
834 }
835 
836 static int efx_ptp_restart(struct efx_nic *efx)
837 {
838 	if (efx->ptp_data && efx->ptp_data->enabled)
839 		return efx_ptp_start(efx);
840 	return 0;
841 }
842 
843 static void efx_ptp_pps_worker(struct work_struct *work)
844 {
845 	struct efx_ptp_data *ptp =
846 		container_of(work, struct efx_ptp_data, pps_work);
847 	struct efx_nic *efx = ptp->channel->efx;
848 	struct ptp_clock_event ptp_evt;
849 
850 	if (efx_ptp_synchronize(efx, PTP_SYNC_ATTEMPTS))
851 		return;
852 
853 	ptp_evt.type = PTP_CLOCK_PPSUSR;
854 	ptp_evt.pps_times = ptp->host_time_pps;
855 	ptp_clock_event(ptp->phc_clock, &ptp_evt);
856 }
857 
858 /* Process any pending transmissions and timestamp any received packets.
859  */
860 static void efx_ptp_worker(struct work_struct *work)
861 {
862 	struct efx_ptp_data *ptp_data =
863 		container_of(work, struct efx_ptp_data, work);
864 	struct efx_nic *efx = ptp_data->channel->efx;
865 	struct sk_buff *skb;
866 	struct sk_buff_head tempq;
867 
868 	if (ptp_data->reset_required) {
869 		efx_ptp_stop(efx);
870 		efx_ptp_start(efx);
871 		return;
872 	}
873 
874 	efx_ptp_drop_time_expired_events(efx);
875 
876 	__skb_queue_head_init(&tempq);
877 	if (efx_ptp_process_events(efx, &tempq) ||
878 	    !skb_queue_empty(&ptp_data->txq)) {
879 
880 		while ((skb = skb_dequeue(&ptp_data->txq)))
881 			efx_ptp_xmit_skb(efx, skb);
882 	}
883 
884 	while ((skb = __skb_dequeue(&tempq)))
885 		efx_ptp_process_rx(efx, skb);
886 }
887 
888 /* Initialise PTP channel and state.
889  *
890  * Setting core_index to zero causes the queue to be initialised and doesn't
891  * overlap with 'rxq0' because ptp.c doesn't use skb_record_rx_queue.
892  */
893 static int efx_ptp_probe_channel(struct efx_channel *channel)
894 {
895 	struct efx_nic *efx = channel->efx;
896 	struct efx_ptp_data *ptp;
897 	int rc = 0;
898 	unsigned int pos;
899 
900 	channel->irq_moderation = 0;
901 	channel->rx_queue.core_index = 0;
902 
903 	ptp = kzalloc(sizeof(struct efx_ptp_data), GFP_KERNEL);
904 	efx->ptp_data = ptp;
905 	if (!efx->ptp_data)
906 		return -ENOMEM;
907 
908 	rc = efx_nic_alloc_buffer(efx, &ptp->start, sizeof(int), GFP_KERNEL);
909 	if (rc != 0)
910 		goto fail1;
911 
912 	ptp->channel = channel;
913 	skb_queue_head_init(&ptp->rxq);
914 	skb_queue_head_init(&ptp->txq);
915 	ptp->workwq = create_singlethread_workqueue("sfc_ptp");
916 	if (!ptp->workwq) {
917 		rc = -ENOMEM;
918 		goto fail2;
919 	}
920 
921 	INIT_WORK(&ptp->work, efx_ptp_worker);
922 	ptp->config.flags = 0;
923 	ptp->config.tx_type = HWTSTAMP_TX_OFF;
924 	ptp->config.rx_filter = HWTSTAMP_FILTER_NONE;
925 	INIT_LIST_HEAD(&ptp->evt_list);
926 	INIT_LIST_HEAD(&ptp->evt_free_list);
927 	spin_lock_init(&ptp->evt_lock);
928 	for (pos = 0; pos < MAX_RECEIVE_EVENTS; pos++)
929 		list_add(&ptp->rx_evts[pos].link, &ptp->evt_free_list);
930 	ptp->evt_overflow = false;
931 
932 	ptp->phc_clock_info.owner = THIS_MODULE;
933 	snprintf(ptp->phc_clock_info.name,
934 		 sizeof(ptp->phc_clock_info.name),
935 		 "%pm", efx->net_dev->perm_addr);
936 	ptp->phc_clock_info.max_adj = MAX_PPB;
937 	ptp->phc_clock_info.n_alarm = 0;
938 	ptp->phc_clock_info.n_ext_ts = 0;
939 	ptp->phc_clock_info.n_per_out = 0;
940 	ptp->phc_clock_info.pps = 1;
941 	ptp->phc_clock_info.adjfreq = efx_phc_adjfreq;
942 	ptp->phc_clock_info.adjtime = efx_phc_adjtime;
943 	ptp->phc_clock_info.gettime = efx_phc_gettime;
944 	ptp->phc_clock_info.settime = efx_phc_settime;
945 	ptp->phc_clock_info.enable = efx_phc_enable;
946 
947 	ptp->phc_clock = ptp_clock_register(&ptp->phc_clock_info,
948 					    &efx->pci_dev->dev);
949 	if (IS_ERR(ptp->phc_clock)) {
950 		rc = PTR_ERR(ptp->phc_clock);
951 		goto fail3;
952 	}
953 
954 	INIT_WORK(&ptp->pps_work, efx_ptp_pps_worker);
955 	ptp->pps_workwq = create_singlethread_workqueue("sfc_pps");
956 	if (!ptp->pps_workwq) {
957 		rc = -ENOMEM;
958 		goto fail4;
959 	}
960 	ptp->nic_ts_enabled = false;
961 
962 	return 0;
963 fail4:
964 	ptp_clock_unregister(efx->ptp_data->phc_clock);
965 
966 fail3:
967 	destroy_workqueue(efx->ptp_data->workwq);
968 
969 fail2:
970 	efx_nic_free_buffer(efx, &ptp->start);
971 
972 fail1:
973 	kfree(efx->ptp_data);
974 	efx->ptp_data = NULL;
975 
976 	return rc;
977 }
978 
979 static void efx_ptp_remove_channel(struct efx_channel *channel)
980 {
981 	struct efx_nic *efx = channel->efx;
982 
983 	if (!efx->ptp_data)
984 		return;
985 
986 	(void)efx_ptp_disable(channel->efx);
987 
988 	cancel_work_sync(&efx->ptp_data->work);
989 	cancel_work_sync(&efx->ptp_data->pps_work);
990 
991 	skb_queue_purge(&efx->ptp_data->rxq);
992 	skb_queue_purge(&efx->ptp_data->txq);
993 
994 	ptp_clock_unregister(efx->ptp_data->phc_clock);
995 
996 	destroy_workqueue(efx->ptp_data->workwq);
997 	destroy_workqueue(efx->ptp_data->pps_workwq);
998 
999 	efx_nic_free_buffer(efx, &efx->ptp_data->start);
1000 	kfree(efx->ptp_data);
1001 }
1002 
1003 static void efx_ptp_get_channel_name(struct efx_channel *channel,
1004 				     char *buf, size_t len)
1005 {
1006 	snprintf(buf, len, "%s-ptp", channel->efx->name);
1007 }
1008 
1009 /* Determine whether this packet should be processed by the PTP module
1010  * or transmitted conventionally.
1011  */
1012 bool efx_ptp_is_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
1013 {
1014 	return efx->ptp_data &&
1015 		efx->ptp_data->enabled &&
1016 		skb->len >= PTP_MIN_LENGTH &&
1017 		skb->len <= MC_CMD_PTP_IN_TRANSMIT_PACKET_MAXNUM &&
1018 		likely(skb->protocol == htons(ETH_P_IP)) &&
1019 		skb_transport_header_was_set(skb) &&
1020 		skb_network_header_len(skb) >= sizeof(struct iphdr) &&
1021 		ip_hdr(skb)->protocol == IPPROTO_UDP &&
1022 		skb_headlen(skb) >=
1023 		skb_transport_offset(skb) + sizeof(struct udphdr) &&
1024 		udp_hdr(skb)->dest == htons(PTP_EVENT_PORT);
1025 }
1026 
1027 /* Receive a PTP packet.  Packets are queued until the arrival of
1028  * the receive timestamp from the MC - this will probably occur after the
1029  * packet arrival because of the processing in the MC.
1030  */
1031 static bool efx_ptp_rx(struct efx_channel *channel, struct sk_buff *skb)
1032 {
1033 	struct efx_nic *efx = channel->efx;
1034 	struct efx_ptp_data *ptp = efx->ptp_data;
1035 	struct efx_ptp_match *match = (struct efx_ptp_match *)skb->cb;
1036 	u8 *match_data_012, *match_data_345;
1037 	unsigned int version;
1038 
1039 	match->expiry = jiffies + msecs_to_jiffies(PKT_EVENT_LIFETIME_MS);
1040 
1041 	/* Correct version? */
1042 	if (ptp->mode == MC_CMD_PTP_MODE_V1) {
1043 		if (!pskb_may_pull(skb, PTP_V1_MIN_LENGTH)) {
1044 			return false;
1045 		}
1046 		version = ntohs(*(__be16 *)&skb->data[PTP_V1_VERSION_OFFSET]);
1047 		if (version != PTP_VERSION_V1) {
1048 			return false;
1049 		}
1050 
1051 		/* PTP V1 uses all six bytes of the UUID to match the packet
1052 		 * to the timestamp
1053 		 */
1054 		match_data_012 = skb->data + PTP_V1_UUID_OFFSET;
1055 		match_data_345 = skb->data + PTP_V1_UUID_OFFSET + 3;
1056 	} else {
1057 		if (!pskb_may_pull(skb, PTP_V2_MIN_LENGTH)) {
1058 			return false;
1059 		}
1060 		version = skb->data[PTP_V2_VERSION_OFFSET];
1061 		if ((version & PTP_VERSION_V2_MASK) != PTP_VERSION_V2) {
1062 			return false;
1063 		}
1064 
1065 		/* The original V2 implementation uses bytes 2-7 of
1066 		 * the UUID to match the packet to the timestamp. This
1067 		 * discards two of the bytes of the MAC address used
1068 		 * to create the UUID (SF bug 33070).  The PTP V2
1069 		 * enhanced mode fixes this issue and uses bytes 0-2
1070 		 * and byte 5-7 of the UUID.
1071 		 */
1072 		match_data_345 = skb->data + PTP_V2_UUID_OFFSET + 5;
1073 		if (ptp->mode == MC_CMD_PTP_MODE_V2) {
1074 			match_data_012 = skb->data + PTP_V2_UUID_OFFSET + 2;
1075 		} else {
1076 			match_data_012 = skb->data + PTP_V2_UUID_OFFSET + 0;
1077 			BUG_ON(ptp->mode != MC_CMD_PTP_MODE_V2_ENHANCED);
1078 		}
1079 	}
1080 
1081 	/* Does this packet require timestamping? */
1082 	if (ntohs(*(__be16 *)&skb->data[PTP_DPORT_OFFSET]) == PTP_EVENT_PORT) {
1083 		struct skb_shared_hwtstamps *timestamps;
1084 
1085 		match->state = PTP_PACKET_STATE_UNMATCHED;
1086 
1087 		/* Clear all timestamps held: filled in later */
1088 		timestamps = skb_hwtstamps(skb);
1089 		memset(timestamps, 0, sizeof(*timestamps));
1090 
1091 		/* We expect the sequence number to be in the same position in
1092 		 * the packet for PTP V1 and V2
1093 		 */
1094 		BUILD_BUG_ON(PTP_V1_SEQUENCE_OFFSET != PTP_V2_SEQUENCE_OFFSET);
1095 		BUILD_BUG_ON(PTP_V1_SEQUENCE_LENGTH != PTP_V2_SEQUENCE_LENGTH);
1096 
1097 		/* Extract UUID/Sequence information */
1098 		match->words[0] = (match_data_012[0]         |
1099 				   (match_data_012[1] << 8)  |
1100 				   (match_data_012[2] << 16) |
1101 				   (match_data_345[0] << 24));
1102 		match->words[1] = (match_data_345[1]         |
1103 				   (match_data_345[2] << 8)  |
1104 				   (skb->data[PTP_V1_SEQUENCE_OFFSET +
1105 					      PTP_V1_SEQUENCE_LENGTH - 1] <<
1106 				    16));
1107 	} else {
1108 		match->state = PTP_PACKET_STATE_MATCH_UNWANTED;
1109 	}
1110 
1111 	skb_queue_tail(&ptp->rxq, skb);
1112 	queue_work(ptp->workwq, &ptp->work);
1113 
1114 	return true;
1115 }
1116 
1117 /* Transmit a PTP packet.  This has to be transmitted by the MC
1118  * itself, through an MCDI call.  MCDI calls aren't permitted
1119  * in the transmit path so defer the actual transmission to a suitable worker.
1120  */
1121 int efx_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
1122 {
1123 	struct efx_ptp_data *ptp = efx->ptp_data;
1124 
1125 	skb_queue_tail(&ptp->txq, skb);
1126 
1127 	if ((udp_hdr(skb)->dest == htons(PTP_EVENT_PORT)) &&
1128 	    (skb->len <= MC_CMD_PTP_IN_TRANSMIT_PACKET_MAXNUM))
1129 		efx_xmit_hwtstamp_pending(skb);
1130 	queue_work(ptp->workwq, &ptp->work);
1131 
1132 	return NETDEV_TX_OK;
1133 }
1134 
1135 static int efx_ptp_change_mode(struct efx_nic *efx, bool enable_wanted,
1136 			       unsigned int new_mode)
1137 {
1138 	if ((enable_wanted != efx->ptp_data->enabled) ||
1139 	    (enable_wanted && (efx->ptp_data->mode != new_mode))) {
1140 		int rc = 0;
1141 
1142 		if (enable_wanted) {
1143 			/* Change of mode requires disable */
1144 			if (efx->ptp_data->enabled &&
1145 			    (efx->ptp_data->mode != new_mode)) {
1146 				efx->ptp_data->enabled = false;
1147 				rc = efx_ptp_stop(efx);
1148 				if (rc != 0)
1149 					return rc;
1150 			}
1151 
1152 			/* Set new operating mode and establish
1153 			 * baseline synchronisation, which must
1154 			 * succeed.
1155 			 */
1156 			efx->ptp_data->mode = new_mode;
1157 			if (netif_running(efx->net_dev))
1158 				rc = efx_ptp_start(efx);
1159 			if (rc == 0) {
1160 				rc = efx_ptp_synchronize(efx,
1161 							 PTP_SYNC_ATTEMPTS * 2);
1162 				if (rc != 0)
1163 					efx_ptp_stop(efx);
1164 			}
1165 		} else {
1166 			rc = efx_ptp_stop(efx);
1167 		}
1168 
1169 		if (rc != 0)
1170 			return rc;
1171 
1172 		efx->ptp_data->enabled = enable_wanted;
1173 	}
1174 
1175 	return 0;
1176 }
1177 
1178 static int efx_ptp_ts_init(struct efx_nic *efx, struct hwtstamp_config *init)
1179 {
1180 	bool enable_wanted = false;
1181 	unsigned int new_mode;
1182 	int rc;
1183 
1184 	if (init->flags)
1185 		return -EINVAL;
1186 
1187 	if ((init->tx_type != HWTSTAMP_TX_OFF) &&
1188 	    (init->tx_type != HWTSTAMP_TX_ON))
1189 		return -ERANGE;
1190 
1191 	new_mode = efx->ptp_data->mode;
1192 	/* Determine whether any PTP HW operations are required */
1193 	switch (init->rx_filter) {
1194 	case HWTSTAMP_FILTER_NONE:
1195 		break;
1196 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1197 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1198 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1199 		init->rx_filter = HWTSTAMP_FILTER_PTP_V1_L4_EVENT;
1200 		new_mode = MC_CMD_PTP_MODE_V1;
1201 		enable_wanted = true;
1202 		break;
1203 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1204 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1205 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1206 	/* Although these three are accepted only IPV4 packets will be
1207 	 * timestamped
1208 	 */
1209 		init->rx_filter = HWTSTAMP_FILTER_PTP_V2_L4_EVENT;
1210 		new_mode = MC_CMD_PTP_MODE_V2_ENHANCED;
1211 		enable_wanted = true;
1212 		break;
1213 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
1214 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
1215 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1216 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1217 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1218 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1219 		/* Non-IP + IPv6 timestamping not supported */
1220 		return -ERANGE;
1221 		break;
1222 	default:
1223 		return -ERANGE;
1224 	}
1225 
1226 	if (init->tx_type != HWTSTAMP_TX_OFF)
1227 		enable_wanted = true;
1228 
1229 	/* Old versions of the firmware do not support the improved
1230 	 * UUID filtering option (SF bug 33070).  If the firmware does
1231 	 * not accept the enhanced mode, fall back to the standard PTP
1232 	 * v2 UUID filtering.
1233 	 */
1234 	rc = efx_ptp_change_mode(efx, enable_wanted, new_mode);
1235 	if ((rc != 0) && (new_mode == MC_CMD_PTP_MODE_V2_ENHANCED))
1236 		rc = efx_ptp_change_mode(efx, enable_wanted, MC_CMD_PTP_MODE_V2);
1237 	if (rc != 0)
1238 		return rc;
1239 
1240 	efx->ptp_data->config = *init;
1241 
1242 	return 0;
1243 }
1244 
1245 void efx_ptp_get_ts_info(struct efx_nic *efx, struct ethtool_ts_info *ts_info)
1246 {
1247 	struct efx_ptp_data *ptp = efx->ptp_data;
1248 
1249 	if (!ptp)
1250 		return;
1251 
1252 	ts_info->so_timestamping |= (SOF_TIMESTAMPING_TX_HARDWARE |
1253 				     SOF_TIMESTAMPING_RX_HARDWARE |
1254 				     SOF_TIMESTAMPING_RAW_HARDWARE);
1255 	ts_info->phc_index = ptp_clock_index(ptp->phc_clock);
1256 	ts_info->tx_types = 1 << HWTSTAMP_TX_OFF | 1 << HWTSTAMP_TX_ON;
1257 	ts_info->rx_filters = (1 << HWTSTAMP_FILTER_NONE |
1258 			       1 << HWTSTAMP_FILTER_PTP_V1_L4_EVENT |
1259 			       1 << HWTSTAMP_FILTER_PTP_V1_L4_SYNC |
1260 			       1 << HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ |
1261 			       1 << HWTSTAMP_FILTER_PTP_V2_L4_EVENT |
1262 			       1 << HWTSTAMP_FILTER_PTP_V2_L4_SYNC |
1263 			       1 << HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ);
1264 }
1265 
1266 int efx_ptp_ioctl(struct efx_nic *efx, struct ifreq *ifr, int cmd)
1267 {
1268 	struct hwtstamp_config config;
1269 	int rc;
1270 
1271 	/* Not a PTP enabled port */
1272 	if (!efx->ptp_data)
1273 		return -EOPNOTSUPP;
1274 
1275 	if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1276 		return -EFAULT;
1277 
1278 	rc = efx_ptp_ts_init(efx, &config);
1279 	if (rc != 0)
1280 		return rc;
1281 
1282 	return copy_to_user(ifr->ifr_data, &config, sizeof(config))
1283 		? -EFAULT : 0;
1284 }
1285 
1286 static void ptp_event_failure(struct efx_nic *efx, int expected_frag_len)
1287 {
1288 	struct efx_ptp_data *ptp = efx->ptp_data;
1289 
1290 	netif_err(efx, hw, efx->net_dev,
1291 		"PTP unexpected event length: got %d expected %d\n",
1292 		ptp->evt_frag_idx, expected_frag_len);
1293 	ptp->reset_required = true;
1294 	queue_work(ptp->workwq, &ptp->work);
1295 }
1296 
1297 /* Process a completed receive event.  Put it on the event queue and
1298  * start worker thread.  This is required because event and their
1299  * correspoding packets may come in either order.
1300  */
1301 static void ptp_event_rx(struct efx_nic *efx, struct efx_ptp_data *ptp)
1302 {
1303 	struct efx_ptp_event_rx *evt = NULL;
1304 
1305 	if (ptp->evt_frag_idx != 3) {
1306 		ptp_event_failure(efx, 3);
1307 		return;
1308 	}
1309 
1310 	spin_lock_bh(&ptp->evt_lock);
1311 	if (!list_empty(&ptp->evt_free_list)) {
1312 		evt = list_first_entry(&ptp->evt_free_list,
1313 				       struct efx_ptp_event_rx, link);
1314 		list_del(&evt->link);
1315 
1316 		evt->seq0 = EFX_QWORD_FIELD(ptp->evt_frags[2], MCDI_EVENT_DATA);
1317 		evt->seq1 = (EFX_QWORD_FIELD(ptp->evt_frags[2],
1318 					     MCDI_EVENT_SRC)        |
1319 			     (EFX_QWORD_FIELD(ptp->evt_frags[1],
1320 					      MCDI_EVENT_SRC) << 8) |
1321 			     (EFX_QWORD_FIELD(ptp->evt_frags[0],
1322 					      MCDI_EVENT_SRC) << 16));
1323 		evt->hwtimestamp = ktime_set(
1324 			EFX_QWORD_FIELD(ptp->evt_frags[0], MCDI_EVENT_DATA),
1325 			EFX_QWORD_FIELD(ptp->evt_frags[1], MCDI_EVENT_DATA));
1326 		evt->expiry = jiffies + msecs_to_jiffies(PKT_EVENT_LIFETIME_MS);
1327 		list_add_tail(&evt->link, &ptp->evt_list);
1328 
1329 		queue_work(ptp->workwq, &ptp->work);
1330 	} else if (!ptp->evt_overflow) {
1331 		/* Log a warning message and set the event overflow flag.
1332 		 * The message won't be logged again until the event queue
1333 		 * becomes empty.
1334 		 */
1335 		netif_err(efx, rx_err, efx->net_dev, "PTP event queue overflow\n");
1336 		ptp->evt_overflow = true;
1337 	}
1338 	spin_unlock_bh(&ptp->evt_lock);
1339 }
1340 
1341 static void ptp_event_fault(struct efx_nic *efx, struct efx_ptp_data *ptp)
1342 {
1343 	int code = EFX_QWORD_FIELD(ptp->evt_frags[0], MCDI_EVENT_DATA);
1344 	if (ptp->evt_frag_idx != 1) {
1345 		ptp_event_failure(efx, 1);
1346 		return;
1347 	}
1348 
1349 	netif_err(efx, hw, efx->net_dev, "PTP error %d\n", code);
1350 }
1351 
1352 static void ptp_event_pps(struct efx_nic *efx, struct efx_ptp_data *ptp)
1353 {
1354 	if (ptp->nic_ts_enabled)
1355 		queue_work(ptp->pps_workwq, &ptp->pps_work);
1356 }
1357 
1358 void efx_ptp_event(struct efx_nic *efx, efx_qword_t *ev)
1359 {
1360 	struct efx_ptp_data *ptp = efx->ptp_data;
1361 	int code = EFX_QWORD_FIELD(*ev, MCDI_EVENT_CODE);
1362 
1363 	if (!ptp->enabled)
1364 		return;
1365 
1366 	if (ptp->evt_frag_idx == 0) {
1367 		ptp->evt_code = code;
1368 	} else if (ptp->evt_code != code) {
1369 		netif_err(efx, hw, efx->net_dev,
1370 			  "PTP out of sequence event %d\n", code);
1371 		ptp->evt_frag_idx = 0;
1372 	}
1373 
1374 	ptp->evt_frags[ptp->evt_frag_idx++] = *ev;
1375 	if (!MCDI_EVENT_FIELD(*ev, CONT)) {
1376 		/* Process resulting event */
1377 		switch (code) {
1378 		case MCDI_EVENT_CODE_PTP_RX:
1379 			ptp_event_rx(efx, ptp);
1380 			break;
1381 		case MCDI_EVENT_CODE_PTP_FAULT:
1382 			ptp_event_fault(efx, ptp);
1383 			break;
1384 		case MCDI_EVENT_CODE_PTP_PPS:
1385 			ptp_event_pps(efx, ptp);
1386 			break;
1387 		default:
1388 			netif_err(efx, hw, efx->net_dev,
1389 				  "PTP unknown event %d\n", code);
1390 			break;
1391 		}
1392 		ptp->evt_frag_idx = 0;
1393 	} else if (MAX_EVENT_FRAGS == ptp->evt_frag_idx) {
1394 		netif_err(efx, hw, efx->net_dev,
1395 			  "PTP too many event fragments\n");
1396 		ptp->evt_frag_idx = 0;
1397 	}
1398 }
1399 
1400 static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta)
1401 {
1402 	struct efx_ptp_data *ptp_data = container_of(ptp,
1403 						     struct efx_ptp_data,
1404 						     phc_clock_info);
1405 	struct efx_nic *efx = ptp_data->channel->efx;
1406 	MCDI_DECLARE_BUF(inadj, MC_CMD_PTP_IN_ADJUST_LEN);
1407 	s64 adjustment_ns;
1408 	int rc;
1409 
1410 	if (delta > MAX_PPB)
1411 		delta = MAX_PPB;
1412 	else if (delta < -MAX_PPB)
1413 		delta = -MAX_PPB;
1414 
1415 	/* Convert ppb to fixed point ns. */
1416 	adjustment_ns = (((s64)delta * PPB_SCALE_WORD) >>
1417 			 (PPB_EXTRA_BITS + MAX_PPB_BITS));
1418 
1419 	MCDI_SET_DWORD(inadj, PTP_IN_OP, MC_CMD_PTP_OP_ADJUST);
1420 	MCDI_SET_DWORD(inadj, PTP_IN_PERIPH_ID, 0);
1421 	MCDI_SET_QWORD(inadj, PTP_IN_ADJUST_FREQ, adjustment_ns);
1422 	MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_SECONDS, 0);
1423 	MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_NANOSECONDS, 0);
1424 	rc = efx_mcdi_rpc(efx, MC_CMD_PTP, inadj, sizeof(inadj),
1425 			  NULL, 0, NULL);
1426 	if (rc != 0)
1427 		return rc;
1428 
1429 	ptp_data->current_adjfreq = adjustment_ns;
1430 	return 0;
1431 }
1432 
1433 static int efx_phc_adjtime(struct ptp_clock_info *ptp, s64 delta)
1434 {
1435 	struct efx_ptp_data *ptp_data = container_of(ptp,
1436 						     struct efx_ptp_data,
1437 						     phc_clock_info);
1438 	struct efx_nic *efx = ptp_data->channel->efx;
1439 	struct timespec delta_ts = ns_to_timespec(delta);
1440 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_ADJUST_LEN);
1441 
1442 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_ADJUST);
1443 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
1444 	MCDI_SET_QWORD(inbuf, PTP_IN_ADJUST_FREQ, ptp_data->current_adjfreq);
1445 	MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_SECONDS, (u32)delta_ts.tv_sec);
1446 	MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_NANOSECONDS, (u32)delta_ts.tv_nsec);
1447 	return efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
1448 			    NULL, 0, NULL);
1449 }
1450 
1451 static int efx_phc_gettime(struct ptp_clock_info *ptp, struct timespec *ts)
1452 {
1453 	struct efx_ptp_data *ptp_data = container_of(ptp,
1454 						     struct efx_ptp_data,
1455 						     phc_clock_info);
1456 	struct efx_nic *efx = ptp_data->channel->efx;
1457 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_READ_NIC_TIME_LEN);
1458 	MCDI_DECLARE_BUF(outbuf, MC_CMD_PTP_OUT_READ_NIC_TIME_LEN);
1459 	int rc;
1460 
1461 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_READ_NIC_TIME);
1462 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
1463 
1464 	rc = efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
1465 			  outbuf, sizeof(outbuf), NULL);
1466 	if (rc != 0)
1467 		return rc;
1468 
1469 	ts->tv_sec = MCDI_DWORD(outbuf, PTP_OUT_READ_NIC_TIME_SECONDS);
1470 	ts->tv_nsec = MCDI_DWORD(outbuf, PTP_OUT_READ_NIC_TIME_NANOSECONDS);
1471 	return 0;
1472 }
1473 
1474 static int efx_phc_settime(struct ptp_clock_info *ptp,
1475 			   const struct timespec *e_ts)
1476 {
1477 	/* Get the current NIC time, efx_phc_gettime.
1478 	 * Subtract from the desired time to get the offset
1479 	 * call efx_phc_adjtime with the offset
1480 	 */
1481 	int rc;
1482 	struct timespec time_now;
1483 	struct timespec delta;
1484 
1485 	rc = efx_phc_gettime(ptp, &time_now);
1486 	if (rc != 0)
1487 		return rc;
1488 
1489 	delta = timespec_sub(*e_ts, time_now);
1490 
1491 	rc = efx_phc_adjtime(ptp, timespec_to_ns(&delta));
1492 	if (rc != 0)
1493 		return rc;
1494 
1495 	return 0;
1496 }
1497 
1498 static int efx_phc_enable(struct ptp_clock_info *ptp,
1499 			  struct ptp_clock_request *request,
1500 			  int enable)
1501 {
1502 	struct efx_ptp_data *ptp_data = container_of(ptp,
1503 						     struct efx_ptp_data,
1504 						     phc_clock_info);
1505 	if (request->type != PTP_CLK_REQ_PPS)
1506 		return -EOPNOTSUPP;
1507 
1508 	ptp_data->nic_ts_enabled = !!enable;
1509 	return 0;
1510 }
1511 
1512 static const struct efx_channel_type efx_ptp_channel_type = {
1513 	.handle_no_channel	= efx_ptp_handle_no_channel,
1514 	.pre_probe		= efx_ptp_probe_channel,
1515 	.post_remove		= efx_ptp_remove_channel,
1516 	.get_name		= efx_ptp_get_channel_name,
1517 	/* no copy operation; there is no need to reallocate this channel */
1518 	.receive_skb		= efx_ptp_rx,
1519 	.keep_eventq		= false,
1520 };
1521 
1522 void efx_ptp_probe(struct efx_nic *efx)
1523 {
1524 	/* Check whether PTP is implemented on this NIC.  The DISABLE
1525 	 * operation will succeed if and only if it is implemented.
1526 	 */
1527 	if (efx_ptp_disable(efx) == 0)
1528 		efx->extra_channel_type[EFX_EXTRA_CHANNEL_PTP] =
1529 			&efx_ptp_channel_type;
1530 }
1531 
1532 void efx_ptp_start_datapath(struct efx_nic *efx)
1533 {
1534 	if (efx_ptp_restart(efx))
1535 		netif_err(efx, drv, efx->net_dev, "Failed to restart PTP.\n");
1536 }
1537 
1538 void efx_ptp_stop_datapath(struct efx_nic *efx)
1539 {
1540 	efx_ptp_stop(efx);
1541 }
1542