projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
sfc: only advertise TX timestamping if we have the license for it
[linux-2.6-block.git] / drivers / net / ethernet / sfc / ptp.c
CommitLineData
7c236c43 1/****************************************************************************
f7a6d2c4
BH		 2 * Driver for Solarflare network controllers and boards
3 * Copyright 2011-2013 Solarflare Communications Inc.
7c236c43
SH		 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"
8b8a95a1 49#include "farch_regs.h"
7c236c43
SH		 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 */
a6f73460 65#define DEFAULT_MIN_SYNCHRONISATION_NS 120
7c236c43
SH		 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
c939a316
LE		 102#define PTP_V2_UUID_LENGTH 8
103#define PTP_V2_UUID_OFFSET 48
104
7c236c43
SH		 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
135enum 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 */
175struct 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 */
187struct 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
a6f73460
LE		 198 * @major: Hardware timestamp, major
199 * @minor: Hardware timestamp, minor
7c236c43 200 * @host_end: Host time immediately after hardware timestamp taken
a6f73460 201 * @wait: Number of NIC clock ticks between hardware timestamp being read and
7c236c43
SH		 202 * host end time being seen
203 * @window: Difference of host_end and host_start
204 * @valid: Whether this timeset is valid
205 */
206struct efx_ptp_timeset {
207 u32 host_start;
a6f73460
LE		 208 u32 major;
209 u32 minor;
7c236c43 210 u32 host_end;
a6f73460 211 u32 wait;
7c236c43
SH		 212 u32 window; /* Derived: end - start, allowing for wrap */
213};
214
215/**
216 * struct efx_ptp_data - Precision Time Protocol (PTP) state
ac36baf8
BH		 217 * @efx: The NIC context
218 * @channel: The PTP channel (Siena only)
bd9a265d
JC		 219 * @rx_ts_inline: Flag for whether RX timestamps are inline (else they are
220 * separate events)
23418dc1
MH		 221 * @rxq: Receive SKB queue (awaiting timestamps)
222 * @txq: Transmit SKB queue
7c236c43
SH		 223 * @evt_list: List of MC receive events awaiting packets
224 * @evt_free_list: List of free events
225 * @evt_lock: Lock for manipulating evt_list and evt_free_list
226 * @rx_evts: Instantiated events (on evt_list and evt_free_list)
227 * @workwq: Work queue for processing pending PTP operations
228 * @work: Work task
229 * @reset_required: A serious error has occurred and the PTP task needs to be
230 * reset (disable, enable).
231 * @rxfilter_event: Receive filter when operating
232 * @rxfilter_general: Receive filter when operating
233 * @config: Current timestamp configuration
234 * @enabled: PTP operation enabled
235 * @mode: Mode in which PTP operating (PTP version)
a6f73460
LE		 236 * @time_format: Time format supported by this NIC
237 * @ns_to_nic_time: Function to convert from scalar nanoseconds to NIC time
238 * @nic_to_kernel_time: Function to convert from NIC to kernel time
239 * @min_synchronisation_ns: Minimum acceptable corrected sync window
240 * @ts_corrections.tx: Required driver correction of transmit timestamps
241 * @ts_corrections.rx: Required driver correction of receive timestamps
242 * @ts_corrections.pps_out: PPS output error (information only)
243 * @ts_corrections.pps_in: Required driver correction of PPS input timestamps
7c236c43
SH		 244 * @evt_frags: Partly assembled PTP events
245 * @evt_frag_idx: Current fragment number
246 * @evt_code: Last event code
247 * @start: Address at which MC indicates ready for synchronisation
248 * @host_time_pps: Host time at last PPS
7c236c43 249 * @current_adjfreq: Current ppb adjustment.
9aecda95 250 * @phc_clock: Pointer to registered phc device (if primary function)
7c236c43
SH		 251 * @phc_clock_info: Registration structure for phc device
252 * @pps_work: pps work task for handling pps events
253 * @pps_workwq: pps work queue
254 * @nic_ts_enabled: Flag indicating if NIC generated TS events are handled
255 * @txbuf: Buffer for use when transmitting (PTP) packets to MC (avoids
256 * allocations in main data path).
99691c4a
BH		 257 * @good_syncs: Number of successful synchronisations.
258 * @fast_syncs: Number of synchronisations requiring short delay
259 * @bad_syncs: Number of failed synchronisations.
260 * @sync_timeouts: Number of synchronisation timeouts
261 * @no_time_syncs: Number of synchronisations with no good times.
262 * @invalid_sync_windows: Number of sync windows with bad durations.
263 * @undersize_sync_windows: Number of corrected sync windows that are too small
264 * @oversize_sync_windows: Number of corrected sync windows that are too large
265 * @rx_no_timestamp: Number of packets received without a timestamp.
7c236c43 266 * @timeset: Last set of synchronisation statistics.
23418dc1 267 * @xmit_skb: Transmit SKB function.
7c236c43
SH		 268 */
269struct efx_ptp_data {
ac36baf8 270 struct efx_nic *efx;
7c236c43 271 struct efx_channel *channel;
bd9a265d 272 bool rx_ts_inline;
7c236c43
SH		 273 struct sk_buff_head rxq;
274 struct sk_buff_head txq;
275 struct list_head evt_list;
276 struct list_head evt_free_list;
277 spinlock_t evt_lock;
278 struct efx_ptp_event_rx rx_evts[MAX_RECEIVE_EVENTS];
279 struct workqueue_struct *workwq;
280 struct work_struct work;
281 bool reset_required;
282 u32 rxfilter_event;
283 u32 rxfilter_general;
284 bool rxfilter_installed;
285 struct hwtstamp_config config;
286 bool enabled;
287 unsigned int mode;
a6f73460
LE		 288 unsigned int time_format;
289 void (*ns_to_nic_time)(s64 ns, u32 *nic_major, u32 *nic_minor);
290 ktime_t (*nic_to_kernel_time)(u32 nic_major, u32 nic_minor,
291 s32 correction);
292 unsigned int min_synchronisation_ns;
293 struct {
294 s32 tx;
295 s32 rx;
296 s32 pps_out;
297 s32 pps_in;
298 } ts_corrections;
7c236c43
SH		 299 efx_qword_t evt_frags[MAX_EVENT_FRAGS];
300 int evt_frag_idx;
301 int evt_code;
302 struct efx_buffer start;
303 struct pps_event_time host_time_pps;
7c236c43
SH		 304 s64 current_adjfreq;
305 struct ptp_clock *phc_clock;
306 struct ptp_clock_info phc_clock_info;
307 struct work_struct pps_work;
308 struct workqueue_struct *pps_workwq;
309 bool nic_ts_enabled;
aa09a3da 310 _MCDI_DECLARE_BUF(txbuf, MC_CMD_PTP_IN_TRANSMIT_LENMAX);
99691c4a
BH		 311
312 unsigned int good_syncs;
313 unsigned int fast_syncs;
314 unsigned int bad_syncs;
315 unsigned int sync_timeouts;
316 unsigned int no_time_syncs;
317 unsigned int invalid_sync_windows;
318 unsigned int undersize_sync_windows;
319 unsigned int oversize_sync_windows;
320 unsigned int rx_no_timestamp;
7c236c43
SH		 321 struct efx_ptp_timeset
322 timeset[MC_CMD_PTP_OUT_SYNCHRONIZE_TIMESET_MAXNUM];
23418dc1 323 void (*xmit_skb)(struct efx_nic *efx, struct sk_buff *skb);
7c236c43
SH		 324};
325
326static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta);
327static int efx_phc_adjtime(struct ptp_clock_info *ptp, s64 delta);
0fcb5c76 328static int efx_phc_gettime(struct ptp_clock_info *ptp, struct timespec64 *ts);
7c236c43 329static int efx_phc_settime(struct ptp_clock_info *ptp,
0fcb5c76 330 const struct timespec64 *e_ts);
7c236c43
SH		 331static int efx_phc_enable(struct ptp_clock_info *ptp,
332 struct ptp_clock_request *request, int on);
333
9c3afb33
MH		 334bool efx_ptp_use_mac_tx_timestamps(struct efx_nic *efx)
335{
336 struct efx_ef10_nic_data *nic_data = efx->nic_data;
337
338 return ((efx_nic_rev(efx) >= EFX_REV_HUNT_A0) &&
339 (nic_data->datapath_caps2 &
340 (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_MAC_TIMESTAMPING_LBN)
341 ));
342}
343
2935e3c3
EC		 344/* PTP 'extra' channel is still a traffic channel, but we only create TX queues
345 * if PTP uses MAC TX timestamps, not if PTP uses the MC directly to transmit.
346 */
347bool efx_ptp_want_txqs(struct efx_channel *channel)
348{
349 return efx_ptp_use_mac_tx_timestamps(channel->efx);
350}
351
99691c4a
BH		 352#define PTP_SW_STAT(ext_name, field_name) \
353 { #ext_name, 0, offsetof(struct efx_ptp_data, field_name) }
354#define PTP_MC_STAT(ext_name, mcdi_name) \
355 { #ext_name, 32, MC_CMD_PTP_OUT_STATUS_STATS_ ## mcdi_name ## _OFST }
356static const struct efx_hw_stat_desc efx_ptp_stat_desc[] = {
357 PTP_SW_STAT(ptp_good_syncs, good_syncs),
358 PTP_SW_STAT(ptp_fast_syncs, fast_syncs),
359 PTP_SW_STAT(ptp_bad_syncs, bad_syncs),
360 PTP_SW_STAT(ptp_sync_timeouts, sync_timeouts),
361 PTP_SW_STAT(ptp_no_time_syncs, no_time_syncs),
362 PTP_SW_STAT(ptp_invalid_sync_windows, invalid_sync_windows),
363 PTP_SW_STAT(ptp_undersize_sync_windows, undersize_sync_windows),
364 PTP_SW_STAT(ptp_oversize_sync_windows, oversize_sync_windows),
365 PTP_SW_STAT(ptp_rx_no_timestamp, rx_no_timestamp),
366 PTP_MC_STAT(ptp_tx_timestamp_packets, TX),
367 PTP_MC_STAT(ptp_rx_timestamp_packets, RX),
368 PTP_MC_STAT(ptp_timestamp_packets, TS),
369 PTP_MC_STAT(ptp_filter_matches, FM),
370 PTP_MC_STAT(ptp_non_filter_matches, NFM),
371};
372#define PTP_STAT_COUNT ARRAY_SIZE(efx_ptp_stat_desc)
373static const unsigned long efx_ptp_stat_mask[] = {
374 [0 ... BITS_TO_LONGS(PTP_STAT_COUNT) - 1] = ~0UL,
375};
376
377size_t efx_ptp_describe_stats(struct efx_nic *efx, u8 *strings)
378{
379 if (!efx->ptp_data)
380 return 0;
381
382 return efx_nic_describe_stats(efx_ptp_stat_desc, PTP_STAT_COUNT,
383 efx_ptp_stat_mask, strings);
384}
385
386size_t efx_ptp_update_stats(struct efx_nic *efx, u64 *stats)
387{
388 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_STATUS_LEN);
389 MCDI_DECLARE_BUF(outbuf, MC_CMD_PTP_OUT_STATUS_LEN);
390 size_t i;
391 int rc;
392
393 if (!efx->ptp_data)
394 return 0;
395
396 /* Copy software statistics */
397 for (i = 0; i < PTP_STAT_COUNT; i++) {
398 if (efx_ptp_stat_desc[i].dma_width)
399 continue;
400 stats[i] = *(unsigned int *)((char *)efx->ptp_data +
401 efx_ptp_stat_desc[i].offset);
402 }
403
404 /* Fetch MC statistics. We *must* fill in all statistics or
405 * risk leaking kernel memory to userland, so if the MCDI
406 * request fails we pretend we got zeroes.
407 */
408 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_STATUS);
409 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
410 rc = efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
411 outbuf, sizeof(outbuf), NULL);
3de1b513 412 if (rc)
99691c4a 413 memset(outbuf, 0, sizeof(outbuf));
99691c4a
BH		 414 efx_nic_update_stats(efx_ptp_stat_desc, PTP_STAT_COUNT,
415 efx_ptp_stat_mask,
416 stats, _MCDI_PTR(outbuf, 0), false);
417
418 return PTP_STAT_COUNT;
419}
420
a6f73460
LE		 421/* For Siena platforms NIC time is s and ns */
422static void efx_ptp_ns_to_s_ns(s64 ns, u32 *nic_major, u32 *nic_minor)
423{
090e2edb
AB		 424 struct timespec64 ts = ns_to_timespec64(ns);
425 *nic_major = (u32)ts.tv_sec;
a6f73460
LE		 426 *nic_minor = ts.tv_nsec;
427}
428
bd9a265d
JC		 429static ktime_t efx_ptp_s_ns_to_ktime_correction(u32 nic_major, u32 nic_minor,
430 s32 correction)
a6f73460
LE		 431{
432 ktime_t kt = ktime_set(nic_major, nic_minor);
433 if (correction >= 0)
434 kt = ktime_add_ns(kt, (u64)correction);
435 else
436 kt = ktime_sub_ns(kt, (u64)-correction);
437 return kt;
438}
439
440/* To convert from s27 format to ns we multiply then divide by a power of 2.
441 * For the conversion from ns to s27, the operation is also converted to a
442 * multiply and shift.
443 */
444#define S27_TO_NS_SHIFT (27)
445#define NS_TO_S27_MULT (((1ULL << 63) + NSEC_PER_SEC / 2) / NSEC_PER_SEC)
446#define NS_TO_S27_SHIFT (63 - S27_TO_NS_SHIFT)
447#define S27_MINOR_MAX (1 << S27_TO_NS_SHIFT)
448
449/* For Huntington platforms NIC time is in seconds and fractions of a second
450 * where the minor register only uses 27 bits in units of 2^-27s.
451 */
452static void efx_ptp_ns_to_s27(s64 ns, u32 *nic_major, u32 *nic_minor)
453{
090e2edb
AB		 454 struct timespec64 ts = ns_to_timespec64(ns);
455 u32 maj = (u32)ts.tv_sec;
a6f73460
LE		 456 u32 min = (u32)(((u64)ts.tv_nsec * NS_TO_S27_MULT +
457 (1ULL << (NS_TO_S27_SHIFT - 1))) >> NS_TO_S27_SHIFT);
458
459 /* The conversion can result in the minor value exceeding the maximum.
460 * In this case, round up to the next second.
461 */
462 if (min >= S27_MINOR_MAX) {
463 min -= S27_MINOR_MAX;
464 maj++;
465 }
466
467 *nic_major = maj;
468 *nic_minor = min;
469}
470
bd9a265d 471static inline ktime_t efx_ptp_s27_to_ktime(u32 nic_major, u32 nic_minor)
a6f73460 472{
bd9a265d
JC		 473 u32 ns = (u32)(((u64)nic_minor * NSEC_PER_SEC +
474 (1ULL << (S27_TO_NS_SHIFT - 1))) >> S27_TO_NS_SHIFT);
475 return ktime_set(nic_major, ns);
476}
a6f73460 477
bd9a265d
JC		 478static ktime_t efx_ptp_s27_to_ktime_correction(u32 nic_major, u32 nic_minor,
479 s32 correction)
480{
a6f73460
LE		 481 /* Apply the correction and deal with carry */
482 nic_minor += correction;
483 if ((s32)nic_minor < 0) {
484 nic_minor += S27_MINOR_MAX;
485 nic_major--;
486 } else if (nic_minor >= S27_MINOR_MAX) {
487 nic_minor -= S27_MINOR_MAX;
488 nic_major++;
489 }
490
bd9a265d 491 return efx_ptp_s27_to_ktime(nic_major, nic_minor);
a6f73460
LE		 492}
493
c1d0d339
MH		 494struct efx_channel *efx_ptp_channel(struct efx_nic *efx)
495{
496 return efx->ptp_data ? efx->ptp_data->channel : NULL;
497}
498
499static u32 last_sync_timestamp_major(struct efx_nic *efx)
500{
501 struct efx_channel *channel = efx_ptp_channel(efx);
502 u32 major = 0;
503
504 if (channel)
505 major = channel->sync_timestamp_major;
506 return major;
507}
508
509/* The 8000 series and later can provide the time from the MAC, which is only
510 * 48 bits long and provides meta-information in the top 2 bits.
511 */
512static ktime_t
513efx_ptp_mac_s27_to_ktime_correction(struct efx_nic *efx,
514 u32 nic_major, u32 nic_minor,
515 s32 correction)
516{
517 ktime_t kt = { 0 };
518
519 if (!(nic_major & 0x80000000)) {
520 WARN_ON_ONCE(nic_major >> 16);
521 /* Use the top bits from the latest sync event. */
522 nic_major &= 0xffff;
523 nic_major |= (last_sync_timestamp_major(efx) & 0xffff0000);
524
525 kt = efx_ptp_s27_to_ktime_correction(nic_major, nic_minor,
526 correction);
527 }
528 return kt;
529}
530
b9b603d4
MH		 531ktime_t efx_ptp_nic_to_kernel_time(struct efx_tx_queue *tx_queue)
532{
533 struct efx_nic *efx = tx_queue->efx;
534 struct efx_ptp_data *ptp = efx->ptp_data;
535 ktime_t kt;
536
c1d0d339
MH		 537 if (efx_ptp_use_mac_tx_timestamps(efx))
538 kt = efx_ptp_mac_s27_to_ktime_correction(efx,
539 tx_queue->completed_timestamp_major,
540 tx_queue->completed_timestamp_minor, 0);
541 else
542 kt = ptp->nic_to_kernel_time(
543 tx_queue->completed_timestamp_major,
544 tx_queue->completed_timestamp_minor, 0);
b9b603d4
MH		 545 return kt;
546}
547
a6f73460
LE		 548/* Get PTP attributes and set up time conversions */
549static int efx_ptp_get_attributes(struct efx_nic *efx)
550{
551 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_GET_ATTRIBUTES_LEN);
552 MCDI_DECLARE_BUF(outbuf, MC_CMD_PTP_OUT_GET_ATTRIBUTES_LEN);
553 struct efx_ptp_data *ptp = efx->ptp_data;
554 int rc;
555 u32 fmt;
556 size_t out_len;
557
558 /* Get the PTP attributes. If the NIC doesn't support the operation we
559 * use the default format for compatibility with older NICs i.e.
560 * seconds and nanoseconds.
561 */
562 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_GET_ATTRIBUTES);
563 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3de1b513
EC		 564 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
565 outbuf, sizeof(outbuf), &out_len);
566 if (rc == 0) {
a6f73460 567 fmt = MCDI_DWORD(outbuf, PTP_OUT_GET_ATTRIBUTES_TIME_FORMAT);
3de1b513 568 } else if (rc == -EINVAL) {
a6f73460 569 fmt = MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_NANOSECONDS;
3de1b513
EC		 570 } else if (rc == -EPERM) {
571 netif_info(efx, probe, efx->net_dev, "no PTP support\n");
a6f73460 572 return rc;
3de1b513
EC		 573 } else {
574 efx_mcdi_display_error(efx, MC_CMD_PTP, sizeof(inbuf),
575 outbuf, sizeof(outbuf), rc);
576 return rc;
577 }
a6f73460
LE		 578
579 if (fmt == MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_27FRACTION) {
580 ptp->ns_to_nic_time = efx_ptp_ns_to_s27;
bd9a265d 581 ptp->nic_to_kernel_time = efx_ptp_s27_to_ktime_correction;
a6f73460
LE		 582 } else if (fmt == MC_CMD_PTP_OUT_GET_ATTRIBUTES_SECONDS_NANOSECONDS) {
583 ptp->ns_to_nic_time = efx_ptp_ns_to_s_ns;
bd9a265d 584 ptp->nic_to_kernel_time = efx_ptp_s_ns_to_ktime_correction;
a6f73460
LE		 585 } else {
586 return -ERANGE;
587 }
588
589 ptp->time_format = fmt;
590
591 /* MC_CMD_PTP_OP_GET_ATTRIBUTES is an extended version of an older
592 * operation MC_CMD_PTP_OP_GET_TIME_FORMAT that also returns a value
593 * to use for the minimum acceptable corrected synchronization window.
594 * If we have the extra information store it. For older firmware that
595 * does not implement the extended command use the default value.
596 */
597 if (rc == 0 && out_len >= MC_CMD_PTP_OUT_GET_ATTRIBUTES_LEN)
598 ptp->min_synchronisation_ns =
599 MCDI_DWORD(outbuf,
600 PTP_OUT_GET_ATTRIBUTES_SYNC_WINDOW_MIN);
601 else
602 ptp->min_synchronisation_ns = DEFAULT_MIN_SYNCHRONISATION_NS;
603
604 return 0;
605}
606
607/* Get PTP timestamp corrections */
608static int efx_ptp_get_timestamp_corrections(struct efx_nic *efx)
609{
610 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_GET_TIMESTAMP_CORRECTIONS_LEN);
611 MCDI_DECLARE_BUF(outbuf, MC_CMD_PTP_OUT_GET_TIMESTAMP_CORRECTIONS_LEN);
612 int rc;
613
614 /* Get the timestamp corrections from the NIC. If this operation is
615 * not supported (older NICs) then no correction is required.
616 */
617 MCDI_SET_DWORD(inbuf, PTP_IN_OP,
618 MC_CMD_PTP_OP_GET_TIMESTAMP_CORRECTIONS);
619 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
620
3de1b513
EC		 621 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
622 outbuf, sizeof(outbuf), NULL);
a6f73460
LE		 623 if (rc == 0) {
624 efx->ptp_data->ts_corrections.tx = MCDI_DWORD(outbuf,
625 PTP_OUT_GET_TIMESTAMP_CORRECTIONS_TRANSMIT);
626 efx->ptp_data->ts_corrections.rx = MCDI_DWORD(outbuf,
627 PTP_OUT_GET_TIMESTAMP_CORRECTIONS_RECEIVE);
628 efx->ptp_data->ts_corrections.pps_out = MCDI_DWORD(outbuf,
629 PTP_OUT_GET_TIMESTAMP_CORRECTIONS_PPS_OUT);
630 efx->ptp_data->ts_corrections.pps_in = MCDI_DWORD(outbuf,
631 PTP_OUT_GET_TIMESTAMP_CORRECTIONS_PPS_IN);
632 } else if (rc == -EINVAL) {
633 efx->ptp_data->ts_corrections.tx = 0;
634 efx->ptp_data->ts_corrections.rx = 0;
635 efx->ptp_data->ts_corrections.pps_out = 0;
636 efx->ptp_data->ts_corrections.pps_in = 0;
637 } else {
3de1b513
EC		 638 efx_mcdi_display_error(efx, MC_CMD_PTP, sizeof(inbuf), outbuf,
639 sizeof(outbuf), rc);
a6f73460
LE		 640 return rc;
641 }
642
643 return 0;
644}
645
7c236c43
SH		 646/* Enable MCDI PTP support. */
647static int efx_ptp_enable(struct efx_nic *efx)
648{
59cfc479 649 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_ENABLE_LEN);
aa09a3da 650 MCDI_DECLARE_BUF_ERR(outbuf);
1e0b8120 651 int rc;
7c236c43
SH		 652
653 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_ENABLE);
c1d828bd 654 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
7c236c43 655 MCDI_SET_DWORD(inbuf, PTP_IN_ENABLE_QUEUE,
ac36baf8
BH		 656 efx->ptp_data->channel ?
657 efx->ptp_data->channel->channel : 0);
7c236c43
SH		 658 MCDI_SET_DWORD(inbuf, PTP_IN_ENABLE_MODE, efx->ptp_data->mode);
659
1e0b8120
EC		 660 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
661 outbuf, sizeof(outbuf), NULL);
662 rc = (rc == -EALREADY) ? 0 : rc;
663 if (rc)
664 efx_mcdi_display_error(efx, MC_CMD_PTP,
665 MC_CMD_PTP_IN_ENABLE_LEN,
666 outbuf, sizeof(outbuf), rc);
667 return rc;
7c236c43
SH		 668}
669
670/* Disable MCDI PTP support.
671 *
672 * Note that this function should never rely on the presence of ptp_data -
673 * may be called before that exists.
674 */
675static int efx_ptp_disable(struct efx_nic *efx)
676{
59cfc479 677 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_DISABLE_LEN);
aa09a3da 678 MCDI_DECLARE_BUF_ERR(outbuf);
1e0b8120 679 int rc;
7c236c43
SH		 680
681 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_DISABLE);
c1d828bd 682 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
1e0b8120
EC		 683 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
684 outbuf, sizeof(outbuf), NULL);
685 rc = (rc == -EALREADY) ? 0 : rc;
b1336389
EC		 686 /* If we get ENOSYS, the NIC doesn't support PTP, and thus this function
687 * should only have been called during probe.
688 */
689 if (rc == -ENOSYS || rc == -EPERM)
690 netif_info(efx, probe, efx->net_dev, "no PTP support\n");
691 else if (rc)
1e0b8120
EC		 692 efx_mcdi_display_error(efx, MC_CMD_PTP,
693 MC_CMD_PTP_IN_DISABLE_LEN,
694 outbuf, sizeof(outbuf), rc);
695 return rc;
7c236c43
SH		 696}
697
698static void efx_ptp_deliver_rx_queue(struct sk_buff_head *q)
699{
700 struct sk_buff *skb;
701
702 while ((skb = skb_dequeue(q))) {
703 local_bh_disable();
704 netif_receive_skb(skb);
705 local_bh_enable();
706 }
707}
708
709static void efx_ptp_handle_no_channel(struct efx_nic *efx)
710{
711 netif_err(efx, drv, efx->net_dev,
712 "ERROR: PTP requires MSI-X and 1 additional interrupt"
713 "vector. PTP disabled\n");
714}
715
716/* Repeatedly send the host time to the MC which will capture the hardware
717 * time.
718 */
719static void efx_ptp_send_times(struct efx_nic *efx,
720 struct pps_event_time *last_time)
721{
722 struct pps_event_time now;
ade1bdff 723 struct timespec64 limit;
7c236c43 724 struct efx_ptp_data *ptp = efx->ptp_data;
7c236c43
SH		 725 int *mc_running = ptp->start.addr;
726
727 pps_get_ts(&now);
7c236c43 728 limit = now.ts_real;
ade1bdff 729 timespec64_add_ns(&limit, SYNCHRONISE_PERIOD_NS);
7c236c43
SH		 730
731 /* Write host time for specified period or until MC is done */
ade1bdff 732 while ((timespec64_compare(&now.ts_real, &limit) < 0) &&
6aa7de05 733 READ_ONCE(*mc_running)) {
ade1bdff 734 struct timespec64 update_time;
7c236c43
SH		 735 unsigned int host_time;
736
737 /* Don't update continuously to avoid saturating the PCIe bus */
738 update_time = now.ts_real;
ade1bdff 739 timespec64_add_ns(&update_time, SYNCHRONISATION_GRANULARITY_NS);
7c236c43
SH		 740 do {
741 pps_get_ts(&now);
ade1bdff 742 } while ((timespec64_compare(&now.ts_real, &update_time) < 0) &&
6aa7de05 743 READ_ONCE(*mc_running));
7c236c43
SH		 744
745 /* Synchronise NIC with single word of time only */
746 host_time = (now.ts_real.tv_sec << MC_NANOSECOND_BITS |
747 now.ts_real.tv_nsec);
748 /* Update host time in NIC memory */
977a5d5d 749 efx->type->ptp_write_host_time(efx, host_time);
7c236c43
SH		 750 }
751 *last_time = now;
752}
753
754/* Read a timeset from the MC's results and partial process. */
c5bb0e98
BH		 755static void efx_ptp_read_timeset(MCDI_DECLARE_STRUCT_PTR(data),
756 struct efx_ptp_timeset *timeset)
7c236c43
SH		 757{
758 unsigned start_ns, end_ns;
759
760 timeset->host_start = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_HOSTSTART);
a6f73460
LE		 761 timeset->major = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_MAJOR);
762 timeset->minor = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_MINOR);
7c236c43 763 timeset->host_end = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_HOSTEND),
a6f73460 764 timeset->wait = MCDI_DWORD(data, PTP_OUT_SYNCHRONIZE_WAITNS);
7c236c43
SH		 765
766 /* Ignore seconds */
767 start_ns = timeset->host_start & MC_NANOSECOND_MASK;
768 end_ns = timeset->host_end & MC_NANOSECOND_MASK;
769 /* Allow for rollover */
770 if (end_ns < start_ns)
771 end_ns += NSEC_PER_SEC;
772 /* Determine duration of operation */
773 timeset->window = end_ns - start_ns;
774}
775
776/* Process times received from MC.
777 *
778 * Extract times from returned results, and establish the minimum value
779 * seen. The minimum value represents the "best" possible time and events
780 * too much greater than this are rejected - the machine is, perhaps, too
781 * busy. A number of readings are taken so that, hopefully, at least one good
782 * synchronisation will be seen in the results.
783 */
c5bb0e98
BH		 784static int
785efx_ptp_process_times(struct efx_nic *efx, MCDI_DECLARE_STRUCT_PTR(synch_buf),
786 size_t response_length,
787 const struct pps_event_time *last_time)
7c236c43 788{
c5bb0e98
BH		 789 unsigned number_readings =
790 MCDI_VAR_ARRAY_LEN(response_length,
791 PTP_OUT_SYNCHRONIZE_TIMESET);
7c236c43 792 unsigned i;
7c236c43
SH		 793 unsigned ngood = 0;
794 unsigned last_good = 0;
795 struct efx_ptp_data *ptp = efx->ptp_data;
7c236c43
SH		 796 u32 last_sec;
797 u32 start_sec;
ade1bdff 798 struct timespec64 delta;
a6f73460 799 ktime_t mc_time;
7c236c43
SH		 800
801 if (number_readings == 0)
802 return -EAGAIN;
803
dfd8d581
LE		 804 /* Read the set of results and find the last good host-MC
805 * synchronization result. The MC times when it finishes reading the
806 * host time so the corrected window time should be fairly constant
99691c4a
BH		 807 * for a given platform. Increment stats for any results that appear
808 * to be erroneous.
7c236c43
SH		 809 */
810 for (i = 0; i < number_readings; i++) {
dfd8d581 811 s32 window, corrected;
090e2edb 812 struct timespec64 wait;
dfd8d581 813
c5bb0e98
BH		 814 efx_ptp_read_timeset(
815 MCDI_ARRAY_STRUCT_PTR(synch_buf,
816 PTP_OUT_SYNCHRONIZE_TIMESET, i),
817 &ptp->timeset[i]);
7c236c43 818
090e2edb 819 wait = ktime_to_timespec64(
a6f73460 820 ptp->nic_to_kernel_time(0, ptp->timeset[i].wait, 0));
dfd8d581 821 window = ptp->timeset[i].window;
a6f73460 822 corrected = window - wait.tv_nsec;
dfd8d581
LE		 823
824 /* We expect the uncorrected synchronization window to be at
825 * least as large as the interval between host start and end
826 * times. If it is smaller than this then this is mostly likely
827 * to be a consequence of the host's time being adjusted.
828 * Check that the corrected sync window is in a reasonable
829 * range. If it is out of range it is likely to be because an
830 * interrupt or other delay occurred between reading the system
831 * time and writing it to MC memory.
832 */
99691c4a
BH		 833 if (window < SYNCHRONISATION_GRANULARITY_NS) {
834 ++ptp->invalid_sync_windows;
835 } else if (corrected >= MAX_SYNCHRONISATION_NS) {
99691c4a 836 ++ptp->oversize_sync_windows;
13c92e82
BH		 837 } else if (corrected < ptp->min_synchronisation_ns) {
838 ++ptp->undersize_sync_windows;
99691c4a 839 } else {
dfd8d581
LE		 840 ngood++;
841 last_good = i;
7c236c43 842 }
dfd8d581 843 }
7c236c43
SH		 844
845 if (ngood == 0) {
846 netif_warn(efx, drv, efx->net_dev,
94cd60d0 847 "PTP no suitable synchronisations\n");
7c236c43
SH		 848 return -EAGAIN;
849 }
850
92d8f766
BH		 851 /* Calculate delay from last good sync (host time) to last_time.
852 * It is possible that the seconds rolled over between taking
7c236c43
SH		 853 * the start reading and the last value written by the host. The
854 * timescales are such that a gap of more than one second is never
92d8f766 855 * expected. delta is *not* normalised.
7c236c43
SH		 856 */
857 start_sec = ptp->timeset[last_good].host_start >> MC_NANOSECOND_BITS;
858 last_sec = last_time->ts_real.tv_sec & MC_SECOND_MASK;
92d8f766
BH		 859 if (start_sec != last_sec &&
860 ((start_sec + 1) & MC_SECOND_MASK) != last_sec) {
861 netif_warn(efx, hw, efx->net_dev,
862 "PTP bad synchronisation seconds\n");
863 return -EAGAIN;
7c236c43 864 }
92d8f766
BH		 865 delta.tv_sec = (last_sec - start_sec) & 1;
866 delta.tv_nsec =
867 last_time->ts_real.tv_nsec -
868 (ptp->timeset[last_good].host_start & MC_NANOSECOND_MASK);
869
870 /* Convert the NIC time at last good sync into kernel time.
871 * No correction is required - this time is the output of a
872 * firmware process.
873 */
874 mc_time = ptp->nic_to_kernel_time(ptp->timeset[last_good].major,
875 ptp->timeset[last_good].minor, 0);
876
877 /* Calculate delay from NIC top of second to last_time */
090e2edb 878 delta.tv_nsec += ktime_to_timespec64(mc_time).tv_nsec;
7c236c43 879
92d8f766 880 /* Set PPS timestamp to match NIC top of second */
7c236c43
SH		 881 ptp->host_time_pps = *last_time;
882 pps_sub_ts(&ptp->host_time_pps, delta);
883
884 return 0;
885}
886
887/* Synchronize times between the host and the MC */
888static int efx_ptp_synchronize(struct efx_nic *efx, unsigned int num_readings)
889{
890 struct efx_ptp_data *ptp = efx->ptp_data;
59cfc479 891 MCDI_DECLARE_BUF(synch_buf, MC_CMD_PTP_OUT_SYNCHRONIZE_LENMAX);
7c236c43
SH		 892 size_t response_length;
893 int rc;
894 unsigned long timeout;
895 struct pps_event_time last_time = {};
896 unsigned int loops = 0;
897 int *start = ptp->start.addr;
898
899 MCDI_SET_DWORD(synch_buf, PTP_IN_OP, MC_CMD_PTP_OP_SYNCHRONIZE);
c1d828bd 900 MCDI_SET_DWORD(synch_buf, PTP_IN_PERIPH_ID, 0);
7c236c43
SH		 901 MCDI_SET_DWORD(synch_buf, PTP_IN_SYNCHRONIZE_NUMTIMESETS,
902 num_readings);
338f74df
BH		 903 MCDI_SET_QWORD(synch_buf, PTP_IN_SYNCHRONIZE_START_ADDR,
904 ptp->start.dma_addr);
7c236c43
SH		 905
906 /* Clear flag that signals MC ready */
6aa7de05 907 WRITE_ONCE(*start, 0);
df2cd8af
BH		 908 rc = efx_mcdi_rpc_start(efx, MC_CMD_PTP, synch_buf,
909 MC_CMD_PTP_IN_SYNCHRONIZE_LEN);
e01b16a7 910 EFX_WARN_ON_ONCE_PARANOID(rc);
7c236c43
SH		 911
912 /* Wait for start from MCDI (or timeout) */
913 timeout = jiffies + msecs_to_jiffies(MAX_SYNCHRONISE_WAIT_MS);
6aa7de05 914 while (!READ_ONCE(*start) && (time_before(jiffies, timeout))) {
7c236c43
SH		 915 udelay(20); /* Usually start MCDI execution quickly */
916 loops++;
917 }
918
99691c4a
BH		 919 if (loops <= 1)
920 ++ptp->fast_syncs;
921 if (!time_before(jiffies, timeout))
922 ++ptp->sync_timeouts;
923
6aa7de05 924 if (READ_ONCE(*start))
7c236c43
SH		 925 efx_ptp_send_times(efx, &last_time);
926
927 /* Collect results */
928 rc = efx_mcdi_rpc_finish(efx, MC_CMD_PTP,
929 MC_CMD_PTP_IN_SYNCHRONIZE_LEN,
930 synch_buf, sizeof(synch_buf),
931 &response_length);
99691c4a 932 if (rc == 0) {
7c236c43
SH		 933 rc = efx_ptp_process_times(efx, synch_buf, response_length,
934 &last_time);
99691c4a
BH		 935 if (rc == 0)
936 ++ptp->good_syncs;
937 else
938 ++ptp->no_time_syncs;
939 }
940
941 /* Increment the bad syncs counter if the synchronize fails, whatever
942 * the reason.
943 */
944 if (rc != 0)
945 ++ptp->bad_syncs;
7c236c43
SH		 946
947 return rc;
948}
949
23418dc1
MH		 950/* Transmit a PTP packet via the dedicated hardware timestamped queue. */
951static void efx_ptp_xmit_skb_queue(struct efx_nic *efx, struct sk_buff *skb)
952{
953 struct efx_ptp_data *ptp_data = efx->ptp_data;
954 struct efx_tx_queue *tx_queue;
955 u8 type = skb->ip_summed == CHECKSUM_PARTIAL ? EFX_TXQ_TYPE_OFFLOAD : 0;
956
957 tx_queue = &ptp_data->channel->tx_queue[type];
958 if (tx_queue && tx_queue->timestamping) {
959 efx_enqueue_skb(tx_queue, skb);
960 } else {
961 WARN_ONCE(1, "PTP channel has no timestamped tx queue\n");
962 dev_kfree_skb_any(skb);
963 }
964}
965
7c236c43 966/* Transmit a PTP packet, via the MCDI interface, to the wire. */
23418dc1 967static void efx_ptp_xmit_skb_mc(struct efx_nic *efx, struct sk_buff *skb)
7c236c43 968{
c5bb0e98 969 struct efx_ptp_data *ptp_data = efx->ptp_data;
7c236c43
SH		 970 struct skb_shared_hwtstamps timestamps;
971 int rc = -EIO;
59cfc479 972 MCDI_DECLARE_BUF(txtime, MC_CMD_PTP_OUT_TRANSMIT_LEN);
9528b921 973 size_t len;
7c236c43 974
c5bb0e98 975 MCDI_SET_DWORD(ptp_data->txbuf, PTP_IN_OP, MC_CMD_PTP_OP_TRANSMIT);
c1d828bd 976 MCDI_SET_DWORD(ptp_data->txbuf, PTP_IN_PERIPH_ID, 0);
c5bb0e98 977 MCDI_SET_DWORD(ptp_data->txbuf, PTP_IN_TRANSMIT_LENGTH, skb->len);
7c236c43
SH		 978 if (skb_shinfo(skb)->nr_frags != 0) {
979 rc = skb_linearize(skb);
980 if (rc != 0)
981 goto fail;
982 }
983
984 if (skb->ip_summed == CHECKSUM_PARTIAL) {
985 rc = skb_checksum_help(skb);
986 if (rc != 0)
987 goto fail;
988 }
989 skb_copy_from_linear_data(skb,
c5bb0e98
BH		 990 MCDI_PTR(ptp_data->txbuf,
991 PTP_IN_TRANSMIT_PACKET),
9528b921
BH		 992 skb->len);
993 rc = efx_mcdi_rpc(efx, MC_CMD_PTP,
994 ptp_data->txbuf, MC_CMD_PTP_IN_TRANSMIT_LEN(skb->len),
995 txtime, sizeof(txtime), &len);
7c236c43
SH		 996 if (rc != 0)
997 goto fail;
998
999 memset(&timestamps, 0, sizeof(timestamps));
a6f73460
LE		 1000 timestamps.hwtstamp = ptp_data->nic_to_kernel_time(
1001 MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_MAJOR),
1002 MCDI_DWORD(txtime, PTP_OUT_TRANSMIT_MINOR),
1003 ptp_data->ts_corrections.tx);
7c236c43
SH		 1004
1005 skb_tstamp_tx(skb, &timestamps);
1006
1007 rc = 0;
1008
1009fail:
23418dc1 1010 dev_kfree_skb_any(skb);
7c236c43 1011
23418dc1 1012 return;
7c236c43
SH		 1013}
1014
1015static void efx_ptp_drop_time_expired_events(struct efx_nic *efx)
1016{
1017 struct efx_ptp_data *ptp = efx->ptp_data;
1018 struct list_head *cursor;
1019 struct list_head *next;
1020
bd9a265d
JC		 1021 if (ptp->rx_ts_inline)
1022 return;
1023
7c236c43
SH		 1024 /* Drop time-expired events */
1025 spin_lock_bh(&ptp->evt_lock);
1026 if (!list_empty(&ptp->evt_list)) {
1027 list_for_each_safe(cursor, next, &ptp->evt_list) {
1028 struct efx_ptp_event_rx *evt;
1029
1030 evt = list_entry(cursor, struct efx_ptp_event_rx,
1031 link);
1032 if (time_after(jiffies, evt->expiry)) {
9545f4e2 1033 list_move(&evt->link, &ptp->evt_free_list);
7c236c43
SH		 1034 netif_warn(efx, hw, efx->net_dev,
1035 "PTP rx event dropped\n");
1036 }
1037 }
1038 }
1039 spin_unlock_bh(&ptp->evt_lock);
1040}
1041
1042static enum ptp_packet_state efx_ptp_match_rx(struct efx_nic *efx,
1043 struct sk_buff *skb)
1044{
1045 struct efx_ptp_data *ptp = efx->ptp_data;
1046 bool evts_waiting;
1047 struct list_head *cursor;
1048 struct list_head *next;
1049 struct efx_ptp_match *match;
1050 enum ptp_packet_state rc = PTP_PACKET_STATE_UNMATCHED;
1051
bd9a265d
JC		 1052 WARN_ON_ONCE(ptp->rx_ts_inline);
1053
7c236c43
SH		 1054 spin_lock_bh(&ptp->evt_lock);
1055 evts_waiting = !list_empty(&ptp->evt_list);
1056 spin_unlock_bh(&ptp->evt_lock);
1057
1058 if (!evts_waiting)
1059 return PTP_PACKET_STATE_UNMATCHED;
1060
1061 match = (struct efx_ptp_match *)skb->cb;
1062 /* Look for a matching timestamp in the event queue */
1063 spin_lock_bh(&ptp->evt_lock);
1064 list_for_each_safe(cursor, next, &ptp->evt_list) {
1065 struct efx_ptp_event_rx *evt;
1066
1067 evt = list_entry(cursor, struct efx_ptp_event_rx, link);
1068 if ((evt->seq0 == match->words[0]) &&
1069 (evt->seq1 == match->words[1])) {
1070 struct skb_shared_hwtstamps *timestamps;
1071
1072 /* Match - add in hardware timestamp */
1073 timestamps = skb_hwtstamps(skb);
1074 timestamps->hwtstamp = evt->hwtimestamp;
1075
1076 match->state = PTP_PACKET_STATE_MATCHED;
1077 rc = PTP_PACKET_STATE_MATCHED;
9545f4e2 1078 list_move(&evt->link, &ptp->evt_free_list);
7c236c43
SH		 1079 break;
1080 }
1081 }
1082 spin_unlock_bh(&ptp->evt_lock);
1083
1084 return rc;
1085}
1086
1087/* Process any queued receive events and corresponding packets
1088 *
1089 * q is returned with all the packets that are ready for delivery.
7c236c43 1090 */
bbbe7149 1091static void efx_ptp_process_events(struct efx_nic *efx, struct sk_buff_head *q)
7c236c43
SH		 1092{
1093 struct efx_ptp_data *ptp = efx->ptp_data;
7c236c43
SH		 1094 struct sk_buff *skb;
1095
1096 while ((skb = skb_dequeue(&ptp->rxq))) {
1097 struct efx_ptp_match *match;
1098
1099 match = (struct efx_ptp_match *)skb->cb;
1100 if (match->state == PTP_PACKET_STATE_MATCH_UNWANTED) {
1101 __skb_queue_tail(q, skb);
1102 } else if (efx_ptp_match_rx(efx, skb) ==
1103 PTP_PACKET_STATE_MATCHED) {
7c236c43
SH		 1104 __skb_queue_tail(q, skb);
1105 } else if (time_after(jiffies, match->expiry)) {
1106 match->state = PTP_PACKET_STATE_TIMED_OUT;
99691c4a 1107 ++ptp->rx_no_timestamp;
7c236c43
SH		 1108 __skb_queue_tail(q, skb);
1109 } else {
1110 /* Replace unprocessed entry and stop */
1111 skb_queue_head(&ptp->rxq, skb);
1112 break;
1113 }
1114 }
7c236c43
SH		 1115}
1116
1117/* Complete processing of a received packet */
1118static inline void efx_ptp_process_rx(struct efx_nic *efx, struct sk_buff *skb)
1119{
1120 local_bh_disable();
1121 netif_receive_skb(skb);
1122 local_bh_enable();
1123}
1124
62a1c703
BH		 1125static void efx_ptp_remove_multicast_filters(struct efx_nic *efx)
1126{
1127 struct efx_ptp_data *ptp = efx->ptp_data;
1128
1129 if (ptp->rxfilter_installed) {
1130 efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
1131 ptp->rxfilter_general);
1132 efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
1133 ptp->rxfilter_event);
1134 ptp->rxfilter_installed = false;
1135 }
1136}
1137
1138static int efx_ptp_insert_multicast_filters(struct efx_nic *efx)
7c236c43
SH		 1139{
1140 struct efx_ptp_data *ptp = efx->ptp_data;
1141 struct efx_filter_spec rxfilter;
1142 int rc;
1143
ac36baf8 1144 if (!ptp->channel || ptp->rxfilter_installed)
62a1c703 1145 return 0;
7c236c43
SH		 1146
1147 /* Must filter on both event and general ports to ensure
1148 * that there is no packet re-ordering.
1149 */
1150 efx_filter_init_rx(&rxfilter, EFX_FILTER_PRI_REQUIRED, 0,
1151 efx_rx_queue_index(
1152 efx_channel_get_rx_queue(ptp->channel)));
1153 rc = efx_filter_set_ipv4_local(&rxfilter, IPPROTO_UDP,
1154 htonl(PTP_ADDRESS),
1155 htons(PTP_EVENT_PORT));
1156 if (rc != 0)
1157 return rc;
1158
1159 rc = efx_filter_insert_filter(efx, &rxfilter, true);
1160 if (rc < 0)
1161 return rc;
1162 ptp->rxfilter_event = rc;
1163
1164 efx_filter_init_rx(&rxfilter, EFX_FILTER_PRI_REQUIRED, 0,
1165 efx_rx_queue_index(
1166 efx_channel_get_rx_queue(ptp->channel)));
1167 rc = efx_filter_set_ipv4_local(&rxfilter, IPPROTO_UDP,
1168 htonl(PTP_ADDRESS),
1169 htons(PTP_GENERAL_PORT));
1170 if (rc != 0)
1171 goto fail;
1172
1173 rc = efx_filter_insert_filter(efx, &rxfilter, true);
1174 if (rc < 0)
1175 goto fail;
1176 ptp->rxfilter_general = rc;
1177
62a1c703
BH		 1178 ptp->rxfilter_installed = true;
1179 return 0;
1180
1181fail:
1182 efx_filter_remove_id_safe(efx, EFX_FILTER_PRI_REQUIRED,
1183 ptp->rxfilter_event);
1184 return rc;
1185}
1186
1187static int efx_ptp_start(struct efx_nic *efx)
1188{
1189 struct efx_ptp_data *ptp = efx->ptp_data;
1190 int rc;
1191
1192 ptp->reset_required = false;
1193
1194 rc = efx_ptp_insert_multicast_filters(efx);
1195 if (rc)
1196 return rc;
1197
7c236c43
SH		 1198 rc = efx_ptp_enable(efx);
1199 if (rc != 0)
62a1c703 1200 goto fail;
7c236c43
SH		 1201
1202 ptp->evt_frag_idx = 0;
1203 ptp->current_adjfreq = 0;
7c236c43
SH		 1204
1205 return 0;
1206
7c236c43 1207fail:
62a1c703 1208 efx_ptp_remove_multicast_filters(efx);
7c236c43
SH		 1209 return rc;
1210}
1211
1212static int efx_ptp_stop(struct efx_nic *efx)
1213{
1214 struct efx_ptp_data *ptp = efx->ptp_data;
7c236c43
SH		 1215 struct list_head *cursor;
1216 struct list_head *next;
2ea4dc28
AR		 1217 int rc;
1218
1219 if (ptp == NULL)
1220 return 0;
1221
1222 rc = efx_ptp_disable(efx);
7c236c43 1223
62a1c703 1224 efx_ptp_remove_multicast_filters(efx);
7c236c43
SH		 1225
1226 /* Make sure RX packets are really delivered */
1227 efx_ptp_deliver_rx_queue(&efx->ptp_data->rxq);
1228 skb_queue_purge(&efx->ptp_data->txq);
1229
1230 /* Drop any pending receive events */
1231 spin_lock_bh(&efx->ptp_data->evt_lock);
1232 list_for_each_safe(cursor, next, &efx->ptp_data->evt_list) {
9545f4e2 1233 list_move(cursor, &efx->ptp_data->evt_free_list);
7c236c43
SH		 1234 }
1235 spin_unlock_bh(&efx->ptp_data->evt_lock);
1236
1237 return rc;
1238}
1239
2ea4dc28
AR		 1240static int efx_ptp_restart(struct efx_nic *efx)
1241{
1242 if (efx->ptp_data && efx->ptp_data->enabled)
1243 return efx_ptp_start(efx);
1244 return 0;
1245}
1246
7c236c43
SH		 1247static void efx_ptp_pps_worker(struct work_struct *work)
1248{
1249 struct efx_ptp_data *ptp =
1250 container_of(work, struct efx_ptp_data, pps_work);
ac36baf8 1251 struct efx_nic *efx = ptp->efx;
7c236c43
SH		 1252 struct ptp_clock_event ptp_evt;
1253
1254 if (efx_ptp_synchronize(efx, PTP_SYNC_ATTEMPTS))
1255 return;
1256
1257 ptp_evt.type = PTP_CLOCK_PPSUSR;
1258 ptp_evt.pps_times = ptp->host_time_pps;
1259 ptp_clock_event(ptp->phc_clock, &ptp_evt);
1260}
1261
7c236c43
SH		 1262static void efx_ptp_worker(struct work_struct *work)
1263{
1264 struct efx_ptp_data *ptp_data =
1265 container_of(work, struct efx_ptp_data, work);
ac36baf8 1266 struct efx_nic *efx = ptp_data->efx;
7c236c43
SH		 1267 struct sk_buff *skb;
1268 struct sk_buff_head tempq;
1269
1270 if (ptp_data->reset_required) {
1271 efx_ptp_stop(efx);
1272 efx_ptp_start(efx);
1273 return;
1274 }
1275
1276 efx_ptp_drop_time_expired_events(efx);
1277
1278 __skb_queue_head_init(&tempq);
bbbe7149 1279 efx_ptp_process_events(efx, &tempq);
7c236c43 1280
bbbe7149 1281 while ((skb = skb_dequeue(&ptp_data->txq)))
23418dc1 1282 ptp_data->xmit_skb(efx, skb);
7c236c43
SH		 1283
1284 while ((skb = __skb_dequeue(&tempq)))
1285 efx_ptp_process_rx(efx, skb);
1286}
1287
5d0dab01
BH		 1288static const struct ptp_clock_info efx_phc_clock_info = {
1289 .owner = THIS_MODULE,
1290 .name = "sfc",
1291 .max_adj = MAX_PPB,
1292 .n_alarm = 0,
1293 .n_ext_ts = 0,
1294 .n_per_out = 0,
4986b4f0 1295 .n_pins = 0,
5d0dab01
BH		 1296 .pps = 1,
1297 .adjfreq = efx_phc_adjfreq,
1298 .adjtime = efx_phc_adjtime,
0fcb5c76
RC		 1299 .gettime64 = efx_phc_gettime,
1300 .settime64 = efx_phc_settime,
5d0dab01
BH		 1301 .enable = efx_phc_enable,
1302};
1303
ac36baf8
BH		 1304/* Initialise PTP state. */
1305int efx_ptp_probe(struct efx_nic *efx, struct efx_channel *channel)
7c236c43 1306{
7c236c43
SH		 1307 struct efx_ptp_data *ptp;
1308 int rc = 0;
1309 unsigned int pos;
1310
7c236c43
SH		 1311 ptp = kzalloc(sizeof(struct efx_ptp_data), GFP_KERNEL);
1312 efx->ptp_data = ptp;
1313 if (!efx->ptp_data)
1314 return -ENOMEM;
1315
ac36baf8
BH		 1316 ptp->efx = efx;
1317 ptp->channel = channel;
bd9a265d 1318 ptp->rx_ts_inline = efx_nic_rev(efx) >= EFX_REV_HUNT_A0;
ac36baf8 1319
0d19a540 1320 rc = efx_nic_alloc_buffer(efx, &ptp->start, sizeof(int), GFP_KERNEL);
7c236c43
SH		 1321 if (rc != 0)
1322 goto fail1;
1323
7c236c43
SH		 1324 skb_queue_head_init(&ptp->rxq);
1325 skb_queue_head_init(&ptp->txq);
1326 ptp->workwq = create_singlethread_workqueue("sfc_ptp");
1327 if (!ptp->workwq) {
1328 rc = -ENOMEM;
1329 goto fail2;
1330 }
1331
2935e3c3 1332 if (efx_ptp_use_mac_tx_timestamps(efx)) {
23418dc1 1333 ptp->xmit_skb = efx_ptp_xmit_skb_queue;
2935e3c3
EC		 1334 /* Request sync events on this channel. */
1335 channel->sync_events_state = SYNC_EVENTS_QUIESCENT;
1336 } else {
1337 ptp->xmit_skb = efx_ptp_xmit_skb_mc;
1338 }
23418dc1 1339
7c236c43
SH		 1340 INIT_WORK(&ptp->work, efx_ptp_worker);
1341 ptp->config.flags = 0;
1342 ptp->config.tx_type = HWTSTAMP_TX_OFF;
1343 ptp->config.rx_filter = HWTSTAMP_FILTER_NONE;
1344 INIT_LIST_HEAD(&ptp->evt_list);
1345 INIT_LIST_HEAD(&ptp->evt_free_list);
1346 spin_lock_init(&ptp->evt_lock);
1347 for (pos = 0; pos < MAX_RECEIVE_EVENTS; pos++)
1348 list_add(&ptp->rx_evts[pos].link, &ptp->evt_free_list);
1349
a6f73460
LE		 1350 /* Get the NIC PTP attributes and set up time conversions */
1351 rc = efx_ptp_get_attributes(efx);
1352 if (rc < 0)
1353 goto fail3;
1354
1355 /* Get the timestamp corrections */
1356 rc = efx_ptp_get_timestamp_corrections(efx);
1357 if (rc < 0)
1358 goto fail3;
1359
9aecda95
BH		 1360 if (efx->mcdi->fn_flags &
1361 (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY)) {
1362 ptp->phc_clock_info = efx_phc_clock_info;
1363 ptp->phc_clock = ptp_clock_register(&ptp->phc_clock_info,
1364 &efx->pci_dev->dev);
1365 if (IS_ERR(ptp->phc_clock)) {
1366 rc = PTR_ERR(ptp->phc_clock);
1367 goto fail3;
efee95f4
NP		 1368 } else if (ptp->phc_clock) {
1369 INIT_WORK(&ptp->pps_work, efx_ptp_pps_worker);
1370 ptp->pps_workwq = create_singlethread_workqueue("sfc_pps");
1371 if (!ptp->pps_workwq) {
1372 rc = -ENOMEM;
1373 goto fail4;
1374 }
9aecda95 1375 }
7c236c43
SH		 1376 }
1377 ptp->nic_ts_enabled = false;
1378
1379 return 0;
1380fail4:
1381 ptp_clock_unregister(efx->ptp_data->phc_clock);
1382
1383fail3:
1384 destroy_workqueue(efx->ptp_data->workwq);
1385
1386fail2:
1387 efx_nic_free_buffer(efx, &ptp->start);
1388
1389fail1:
1390 kfree(efx->ptp_data);
1391 efx->ptp_data = NULL;
1392
1393 return rc;
1394}
1395
ac36baf8
BH		 1396/* Initialise PTP channel.
1397 *
1398 * Setting core_index to zero causes the queue to be initialised and doesn't
1399 * overlap with 'rxq0' because ptp.c doesn't use skb_record_rx_queue.
1400 */
1401static int efx_ptp_probe_channel(struct efx_channel *channel)
7c236c43
SH		 1402{
1403 struct efx_nic *efx = channel->efx;
23418dc1 1404 int rc;
7c236c43 1405
539de7c5 1406 channel->irq_moderation_us = 0;
ac36baf8
BH		 1407 channel->rx_queue.core_index = 0;
1408
23418dc1
MH		 1409 rc = efx_ptp_probe(efx, channel);
1410 /* Failure to probe PTP is not fatal; this channel will just not be
1411 * used for anything.
1412 * In the case of EPERM, efx_ptp_probe will print its own message (in
1413 * efx_ptp_get_attributes()), so we don't need to.
1414 */
1415 if (rc && rc != -EPERM)
1416 netif_warn(efx, drv, efx->net_dev,
1417 "Failed to probe PTP, rc=%d\n", rc);
1418 return 0;
ac36baf8
BH		 1419}
1420
1421void efx_ptp_remove(struct efx_nic *efx)
1422{
7c236c43
SH		 1423 if (!efx->ptp_data)
1424 return;
1425
ac36baf8 1426 (void)efx_ptp_disable(efx);
7c236c43
SH		 1427
1428 cancel_work_sync(&efx->ptp_data->work);
1429 cancel_work_sync(&efx->ptp_data->pps_work);
1430
1431 skb_queue_purge(&efx->ptp_data->rxq);
1432 skb_queue_purge(&efx->ptp_data->txq);
1433
9aecda95
BH		 1434 if (efx->ptp_data->phc_clock) {
1435 destroy_workqueue(efx->ptp_data->pps_workwq);
1436 ptp_clock_unregister(efx->ptp_data->phc_clock);
1437 }
7c236c43
SH		 1438
1439 destroy_workqueue(efx->ptp_data->workwq);
7c236c43
SH		 1440
1441 efx_nic_free_buffer(efx, &efx->ptp_data->start);
1442 kfree(efx->ptp_data);
23418dc1 1443 efx->ptp_data = NULL;
7c236c43
SH		 1444}
1445
ac36baf8
BH		 1446static void efx_ptp_remove_channel(struct efx_channel *channel)
1447{
1448 efx_ptp_remove(channel->efx);
1449}
1450
7c236c43
SH		 1451static void efx_ptp_get_channel_name(struct efx_channel *channel,
1452 char *buf, size_t len)
1453{
1454 snprintf(buf, len, "%s-ptp", channel->efx->name);
1455}
1456
1457/* Determine whether this packet should be processed by the PTP module
1458 * or transmitted conventionally.
1459 */
1460bool efx_ptp_is_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
1461{
1462 return efx->ptp_data &&
1463 efx->ptp_data->enabled &&
1464 skb->len >= PTP_MIN_LENGTH &&
1465 skb->len <= MC_CMD_PTP_IN_TRANSMIT_PACKET_MAXNUM &&
1466 likely(skb->protocol == htons(ETH_P_IP)) &&
e5a498e9
BH		 1467 skb_transport_header_was_set(skb) &&
1468 skb_network_header_len(skb) >= sizeof(struct iphdr) &&
7c236c43 1469 ip_hdr(skb)->protocol == IPPROTO_UDP &&
e5a498e9
BH		 1470 skb_headlen(skb) >=
1471 skb_transport_offset(skb) + sizeof(struct udphdr) &&
7c236c43
SH		 1472 udp_hdr(skb)->dest == htons(PTP_EVENT_PORT);
1473}
1474
1475/* Receive a PTP packet. Packets are queued until the arrival of
1476 * the receive timestamp from the MC - this will probably occur after the
1477 * packet arrival because of the processing in the MC.
1478 */
4a74dc65 1479static bool efx_ptp_rx(struct efx_channel *channel, struct sk_buff *skb)
7c236c43
SH		 1480{
1481 struct efx_nic *efx = channel->efx;
1482 struct efx_ptp_data *ptp = efx->ptp_data;
1483 struct efx_ptp_match *match = (struct efx_ptp_match *)skb->cb;
c939a316 1484 u8 *match_data_012, *match_data_345;
7c236c43 1485 unsigned int version;
ce320f44 1486 u8 *data;
7c236c43
SH		 1487
1488 match->expiry = jiffies + msecs_to_jiffies(PKT_EVENT_LIFETIME_MS);
1489
1490 /* Correct version? */
1491 if (ptp->mode == MC_CMD_PTP_MODE_V1) {
97d48a10 1492 if (!pskb_may_pull(skb, PTP_V1_MIN_LENGTH)) {
4a74dc65 1493 return false;
7c236c43 1494 }
ce320f44
BH		 1495 data = skb->data;
1496 version = ntohs(*(__be16 *)&data[PTP_V1_VERSION_OFFSET]);
7c236c43 1497 if (version != PTP_VERSION_V1) {
4a74dc65 1498 return false;
7c236c43 1499 }
c939a316
LE		 1500
1501 /* PTP V1 uses all six bytes of the UUID to match the packet
1502 * to the timestamp
1503 */
ce320f44
BH		 1504 match_data_012 = data + PTP_V1_UUID_OFFSET;
1505 match_data_345 = data + PTP_V1_UUID_OFFSET + 3;
7c236c43 1506 } else {
97d48a10 1507 if (!pskb_may_pull(skb, PTP_V2_MIN_LENGTH)) {
4a74dc65 1508 return false;
7c236c43 1509 }
ce320f44
BH		 1510 data = skb->data;
1511 version = data[PTP_V2_VERSION_OFFSET];
7c236c43 1512 if ((version & PTP_VERSION_V2_MASK) != PTP_VERSION_V2) {
4a74dc65 1513 return false;
7c236c43 1514 }
c939a316
LE		 1515
1516 /* The original V2 implementation uses bytes 2-7 of
1517 * the UUID to match the packet to the timestamp. This
1518 * discards two of the bytes of the MAC address used
1519 * to create the UUID (SF bug 33070). The PTP V2
1520 * enhanced mode fixes this issue and uses bytes 0-2
1521 * and byte 5-7 of the UUID.
1522 */
ce320f44 1523 match_data_345 = data + PTP_V2_UUID_OFFSET + 5;
c939a316 1524 if (ptp->mode == MC_CMD_PTP_MODE_V2) {
ce320f44 1525 match_data_012 = data + PTP_V2_UUID_OFFSET + 2;
c939a316 1526 } else {
ce320f44 1527 match_data_012 = data + PTP_V2_UUID_OFFSET + 0;
c939a316
LE		 1528 BUG_ON(ptp->mode != MC_CMD_PTP_MODE_V2_ENHANCED);
1529 }
7c236c43
SH		 1530 }
1531
1532 /* Does this packet require timestamping? */
ce320f44 1533 if (ntohs(*(__be16 *)&data[PTP_DPORT_OFFSET]) == PTP_EVENT_PORT) {
7c236c43
SH		 1534 match->state = PTP_PACKET_STATE_UNMATCHED;
1535
c939a316
LE		 1536 /* We expect the sequence number to be in the same position in
1537 * the packet for PTP V1 and V2
1538 */
1539 BUILD_BUG_ON(PTP_V1_SEQUENCE_OFFSET != PTP_V2_SEQUENCE_OFFSET);
1540 BUILD_BUG_ON(PTP_V1_SEQUENCE_LENGTH != PTP_V2_SEQUENCE_LENGTH);
1541
7c236c43 1542 /* Extract UUID/Sequence information */
c939a316
LE		 1543 match->words[0] = (match_data_012[0] |
1544 (match_data_012[1] << 8) |
1545 (match_data_012[2] << 16) |
1546 (match_data_345[0] << 24));
1547 match->words[1] = (match_data_345[1] |
1548 (match_data_345[2] << 8) |
ce320f44
BH		 1549 (data[PTP_V1_SEQUENCE_OFFSET +
1550 PTP_V1_SEQUENCE_LENGTH - 1] <<
7c236c43
SH		 1551 16));
1552 } else {
1553 match->state = PTP_PACKET_STATE_MATCH_UNWANTED;
1554 }
1555
1556 skb_queue_tail(&ptp->rxq, skb);
1557 queue_work(ptp->workwq, &ptp->work);
4a74dc65
BH		 1558
1559 return true;
7c236c43
SH		 1560}
1561
1562/* Transmit a PTP packet. This has to be transmitted by the MC
1563 * itself, through an MCDI call. MCDI calls aren't permitted
1564 * in the transmit path so defer the actual transmission to a suitable worker.
1565 */
1566int efx_ptp_tx(struct efx_nic *efx, struct sk_buff *skb)
1567{
1568 struct efx_ptp_data *ptp = efx->ptp_data;
1569
1570 skb_queue_tail(&ptp->txq, skb);
1571
1572 if ((udp_hdr(skb)->dest == htons(PTP_EVENT_PORT)) &&
1573 (skb->len <= MC_CMD_PTP_IN_TRANSMIT_PACKET_MAXNUM))
1574 efx_xmit_hwtstamp_pending(skb);
1575 queue_work(ptp->workwq, &ptp->work);
1576
1577 return NETDEV_TX_OK;
1578}
1579
9ec06595
DP		 1580int efx_ptp_get_mode(struct efx_nic *efx)
1581{
1582 return efx->ptp_data->mode;
1583}
1584
1585int efx_ptp_change_mode(struct efx_nic *efx, bool enable_wanted,
1586 unsigned int new_mode)
7c236c43
SH		 1587{
1588 if ((enable_wanted != efx->ptp_data->enabled) ||
1589 (enable_wanted && (efx->ptp_data->mode != new_mode))) {
2ea4dc28 1590 int rc = 0;
7c236c43
SH		 1591
1592 if (enable_wanted) {
1593 /* Change of mode requires disable */
1594 if (efx->ptp_data->enabled &&
1595 (efx->ptp_data->mode != new_mode)) {
1596 efx->ptp_data->enabled = false;
1597 rc = efx_ptp_stop(efx);
1598 if (rc != 0)
1599 return rc;
1600 }
1601
1602 /* Set new operating mode and establish
1603 * baseline synchronisation, which must
1604 * succeed.
1605 */
1606 efx->ptp_data->mode = new_mode;
2ea4dc28
AR		 1607 if (netif_running(efx->net_dev))
1608 rc = efx_ptp_start(efx);
7c236c43
SH		 1609 if (rc == 0) {
1610 rc = efx_ptp_synchronize(efx,
1611 PTP_SYNC_ATTEMPTS * 2);
1612 if (rc != 0)
1613 efx_ptp_stop(efx);
1614 }
1615 } else {
1616 rc = efx_ptp_stop(efx);
1617 }
1618
1619 if (rc != 0)
1620 return rc;
1621
1622 efx->ptp_data->enabled = enable_wanted;
1623 }
1624
1625 return 0;
1626}
1627
1628static int efx_ptp_ts_init(struct efx_nic *efx, struct hwtstamp_config *init)
1629{
7c236c43
SH		 1630 int rc;
1631
1632 if (init->flags)
1633 return -EINVAL;
1634
1635 if ((init->tx_type != HWTSTAMP_TX_OFF) &&
1636 (init->tx_type != HWTSTAMP_TX_ON))
1637 return -ERANGE;
1638
9ec06595
DP		 1639 rc = efx->type->ptp_set_ts_config(efx, init);
1640 if (rc)
7c236c43
SH		 1641 return rc;
1642
1643 efx->ptp_data->config = *init;
7c236c43
SH		 1644 return 0;
1645}
1646
62ebac92 1647void efx_ptp_get_ts_info(struct efx_nic *efx, struct ethtool_ts_info *ts_info)
7c236c43 1648{
7c236c43 1649 struct efx_ptp_data *ptp = efx->ptp_data;
9aecda95
BH		 1650 struct efx_nic *primary = efx->primary;
1651
1652 ASSERT_RTNL();
7c236c43
SH		 1653
1654 if (!ptp)
62ebac92 1655 return;
7c236c43 1656
62ebac92
BH		 1657 ts_info->so_timestamping |= (SOF_TIMESTAMPING_TX_HARDWARE |
1658 SOF_TIMESTAMPING_RX_HARDWARE |
1659 SOF_TIMESTAMPING_RAW_HARDWARE);
6aa47c87
MH		 1660 /* Check licensed features. If we don't have the license for TX
1661 * timestamps, the NIC will not support them.
1662 */
1663 if (efx_ptp_use_mac_tx_timestamps(efx)) {
1664 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1665
1666 if (!(nic_data->licensed_features &
1667 (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN)))
1668 ts_info->so_timestamping &=
1669 ~SOF_TIMESTAMPING_TX_HARDWARE;
1670 }
9aecda95
BH		 1671 if (primary && primary->ptp_data && primary->ptp_data->phc_clock)
1672 ts_info->phc_index =
1673 ptp_clock_index(primary->ptp_data->phc_clock);
7c236c43 1674 ts_info->tx_types = 1 << HWTSTAMP_TX_OFF | 1 << HWTSTAMP_TX_ON;
9ec06595 1675 ts_info->rx_filters = ptp->efx->type->hwtstamp_filters;
7c236c43
SH		 1676}
1677
433dc9b3 1678int efx_ptp_set_ts_config(struct efx_nic *efx, struct ifreq *ifr)
7c236c43
SH		 1679{
1680 struct hwtstamp_config config;
1681 int rc;
1682
1683 /* Not a PTP enabled port */
1684 if (!efx->ptp_data)
1685 return -EOPNOTSUPP;
1686
1687 if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
1688 return -EFAULT;
1689
1690 rc = efx_ptp_ts_init(efx, &config);
1691 if (rc != 0)
1692 return rc;
1693
1694 return copy_to_user(ifr->ifr_data, &config, sizeof(config))
1695 ? -EFAULT : 0;
1696}
1697
433dc9b3
BH		 1698int efx_ptp_get_ts_config(struct efx_nic *efx, struct ifreq *ifr)
1699{
1700 if (!efx->ptp_data)
1701 return -EOPNOTSUPP;
1702
1703 return copy_to_user(ifr->ifr_data, &efx->ptp_data->config,
1704 sizeof(efx->ptp_data->config)) ? -EFAULT : 0;
1705}
1706
7c236c43
SH		 1707static void ptp_event_failure(struct efx_nic *efx, int expected_frag_len)
1708{
1709 struct efx_ptp_data *ptp = efx->ptp_data;
1710
1711 netif_err(efx, hw, efx->net_dev,
1712 "PTP unexpected event length: got %d expected %d\n",
1713 ptp->evt_frag_idx, expected_frag_len);
1714 ptp->reset_required = true;
1715 queue_work(ptp->workwq, &ptp->work);
1716}
1717
1718/* Process a completed receive event. Put it on the event queue and
1719 * start worker thread. This is required because event and their
1720 * correspoding packets may come in either order.
1721 */
1722static void ptp_event_rx(struct efx_nic *efx, struct efx_ptp_data *ptp)
1723{
1724 struct efx_ptp_event_rx *evt = NULL;
1725
bd9a265d
JC		 1726 if (WARN_ON_ONCE(ptp->rx_ts_inline))
1727 return;
1728
7c236c43
SH		 1729 if (ptp->evt_frag_idx != 3) {
1730 ptp_event_failure(efx, 3);
1731 return;
1732 }
1733
1734 spin_lock_bh(&ptp->evt_lock);
1735 if (!list_empty(&ptp->evt_free_list)) {
1736 evt = list_first_entry(&ptp->evt_free_list,
1737 struct efx_ptp_event_rx, link);
1738 list_del(&evt->link);
1739
1740 evt->seq0 = EFX_QWORD_FIELD(ptp->evt_frags[2], MCDI_EVENT_DATA);
1741 evt->seq1 = (EFX_QWORD_FIELD(ptp->evt_frags[2],
1742 MCDI_EVENT_SRC) |
1743 (EFX_QWORD_FIELD(ptp->evt_frags[1],
1744 MCDI_EVENT_SRC) << 8) |
1745 (EFX_QWORD_FIELD(ptp->evt_frags[0],
1746 MCDI_EVENT_SRC) << 16));
a6f73460 1747 evt->hwtimestamp = efx->ptp_data->nic_to_kernel_time(
7c236c43 1748 EFX_QWORD_FIELD(ptp->evt_frags[0], MCDI_EVENT_DATA),
a6f73460
LE		 1749 EFX_QWORD_FIELD(ptp->evt_frags[1], MCDI_EVENT_DATA),
1750 ptp->ts_corrections.rx);
7c236c43
SH		 1751 evt->expiry = jiffies + msecs_to_jiffies(PKT_EVENT_LIFETIME_MS);
1752 list_add_tail(&evt->link, &ptp->evt_list);
1753
1754 queue_work(ptp->workwq, &ptp->work);
f9fd7ec7
LE		 1755 } else if (net_ratelimit()) {
1756 /* Log a rate-limited warning message. */
f3211600 1757 netif_err(efx, rx_err, efx->net_dev, "PTP event queue overflow\n");
7c236c43
SH		 1758 }
1759 spin_unlock_bh(&ptp->evt_lock);
1760}
1761
1762static void ptp_event_fault(struct efx_nic *efx, struct efx_ptp_data *ptp)
1763{
1764 int code = EFX_QWORD_FIELD(ptp->evt_frags[0], MCDI_EVENT_DATA);
1765 if (ptp->evt_frag_idx != 1) {
1766 ptp_event_failure(efx, 1);
1767 return;
1768 }
1769
1770 netif_err(efx, hw, efx->net_dev, "PTP error %d\n", code);
1771}
1772
1773static void ptp_event_pps(struct efx_nic *efx, struct efx_ptp_data *ptp)
1774{
1775 if (ptp->nic_ts_enabled)
1776 queue_work(ptp->pps_workwq, &ptp->pps_work);
1777}
1778
1779void efx_ptp_event(struct efx_nic *efx, efx_qword_t *ev)
1780{
1781 struct efx_ptp_data *ptp = efx->ptp_data;
1782 int code = EFX_QWORD_FIELD(*ev, MCDI_EVENT_CODE);
1783
8f355e5c 1784 if (!ptp) {
acaef3c1 1785 if (!efx->ptp_warned) {
8f355e5c
EC		 1786 netif_warn(efx, drv, efx->net_dev,
1787 "Received PTP event but PTP not set up\n");
acaef3c1
EC		 1788 efx->ptp_warned = true;
1789 }
8f355e5c
EC		 1790 return;
1791 }
1792
7c236c43
SH		 1793 if (!ptp->enabled)
1794 return;
1795
1796 if (ptp->evt_frag_idx == 0) {
1797 ptp->evt_code = code;
1798 } else if (ptp->evt_code != code) {
1799 netif_err(efx, hw, efx->net_dev,
1800 "PTP out of sequence event %d\n", code);
1801 ptp->evt_frag_idx = 0;
1802 }
1803
1804 ptp->evt_frags[ptp->evt_frag_idx++] = *ev;
1805 if (!MCDI_EVENT_FIELD(*ev, CONT)) {
1806 /* Process resulting event */
1807 switch (code) {
1808 case MCDI_EVENT_CODE_PTP_RX:
1809 ptp_event_rx(efx, ptp);
1810 break;
1811 case MCDI_EVENT_CODE_PTP_FAULT:
1812 ptp_event_fault(efx, ptp);
1813 break;
1814 case MCDI_EVENT_CODE_PTP_PPS:
1815 ptp_event_pps(efx, ptp);
1816 break;
1817 default:
1818 netif_err(efx, hw, efx->net_dev,
1819 "PTP unknown event %d\n", code);
1820 break;
1821 }
1822 ptp->evt_frag_idx = 0;
1823 } else if (MAX_EVENT_FRAGS == ptp->evt_frag_idx) {
1824 netif_err(efx, hw, efx->net_dev,
1825 "PTP too many event fragments\n");
1826 ptp->evt_frag_idx = 0;
1827 }
1828}
1829
bd9a265d
JC		 1830void efx_time_sync_event(struct efx_channel *channel, efx_qword_t *ev)
1831{
1832 channel->sync_timestamp_major = MCDI_EVENT_FIELD(*ev, PTP_TIME_MAJOR);
1833 channel->sync_timestamp_minor =
1834 MCDI_EVENT_FIELD(*ev, PTP_TIME_MINOR_26_19) << 19;
1835 /* if sync events have been disabled then we want to silently ignore
1836 * this event, so throw away result.
1837 */
1838 (void) cmpxchg(&channel->sync_events_state, SYNC_EVENTS_REQUESTED,
1839 SYNC_EVENTS_VALID);
1840}
1841
1842/* make some assumptions about the time representation rather than abstract it,
1843 * since we currently only support one type of inline timestamping and only on
1844 * EF10.
1845 */
1846#define MINOR_TICKS_PER_SECOND 0x8000000
1847/* Fuzz factor for sync events to be out of order with RX events */
1848#define FUZZ (MINOR_TICKS_PER_SECOND / 10)
1849#define EXPECTED_SYNC_EVENTS_PER_SECOND 4
1850
1851static inline u32 efx_rx_buf_timestamp_minor(struct efx_nic *efx, const u8 *eh)
1852{
1853#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
1854 return __le32_to_cpup((const __le32 *)(eh + efx->rx_packet_ts_offset));
1855#else
1856 const u8 *data = eh + efx->rx_packet_ts_offset;
1857 return (u32)data[0] |
1858 (u32)data[1] << 8 |
1859 (u32)data[2] << 16 |
1860 (u32)data[3] << 24;
1861#endif
1862}
1863
1864void __efx_rx_skb_attach_timestamp(struct efx_channel *channel,
1865 struct sk_buff *skb)
1866{
1867 struct efx_nic *efx = channel->efx;
1868 u32 pkt_timestamp_major, pkt_timestamp_minor;
1869 u32 diff, carry;
1870 struct skb_shared_hwtstamps *timestamps;
1871
1872 pkt_timestamp_minor = (efx_rx_buf_timestamp_minor(efx,
1873 skb_mac_header(skb)) +
1874 (u32) efx->ptp_data->ts_corrections.rx) &
1875 (MINOR_TICKS_PER_SECOND - 1);
1876
1877 /* get the difference between the packet and sync timestamps,
1878 * modulo one second
1879 */
1880 diff = (pkt_timestamp_minor - channel->sync_timestamp_minor) &
1881 (MINOR_TICKS_PER_SECOND - 1);
1882 /* do we roll over a second boundary and need to carry the one? */
1883 carry = channel->sync_timestamp_minor + diff > MINOR_TICKS_PER_SECOND ?
1884 1 : 0;
1885
1886 if (diff <= MINOR_TICKS_PER_SECOND / EXPECTED_SYNC_EVENTS_PER_SECOND +
1887 FUZZ) {
1888 /* packet is ahead of the sync event by a quarter of a second or
1889 * less (allowing for fuzz)
1890 */
1891 pkt_timestamp_major = channel->sync_timestamp_major + carry;
1892 } else if (diff >= MINOR_TICKS_PER_SECOND - FUZZ) {
1893 /* packet is behind the sync event but within the fuzz factor.
1894 * This means the RX packet and sync event crossed as they were
1895 * placed on the event queue, which can sometimes happen.
1896 */
1897 pkt_timestamp_major = channel->sync_timestamp_major - 1 + carry;
1898 } else {
1899 /* it's outside tolerance in both directions. this might be
1900 * indicative of us missing sync events for some reason, so
1901 * we'll call it an error rather than risk giving a bogus
1902 * timestamp.
1903 */
1904 netif_vdbg(efx, drv, efx->net_dev,
1905 "packet timestamp %x too far from sync event %x:%x\n",
1906 pkt_timestamp_minor, channel->sync_timestamp_major,
1907 channel->sync_timestamp_minor);
1908 return;
1909 }
1910
1911 /* attach the timestamps to the skb */
1912 timestamps = skb_hwtstamps(skb);
1913 timestamps->hwtstamp =
1914 efx_ptp_s27_to_ktime(pkt_timestamp_major, pkt_timestamp_minor);
1915}
1916
7c236c43
SH		 1917static int efx_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta)
1918{
1919 struct efx_ptp_data *ptp_data = container_of(ptp,
1920 struct efx_ptp_data,
1921 phc_clock_info);
ac36baf8 1922 struct efx_nic *efx = ptp_data->efx;
59cfc479 1923 MCDI_DECLARE_BUF(inadj, MC_CMD_PTP_IN_ADJUST_LEN);
7c236c43
SH		 1924 s64 adjustment_ns;
1925 int rc;
1926
1927 if (delta > MAX_PPB)
1928 delta = MAX_PPB;
1929 else if (delta < -MAX_PPB)
1930 delta = -MAX_PPB;
1931
1932 /* Convert ppb to fixed point ns. */
1933 adjustment_ns = (((s64)delta * PPB_SCALE_WORD) >>
1934 (PPB_EXTRA_BITS + MAX_PPB_BITS));
1935
1936 MCDI_SET_DWORD(inadj, PTP_IN_OP, MC_CMD_PTP_OP_ADJUST);
c1d828bd 1937 MCDI_SET_DWORD(inadj, PTP_IN_PERIPH_ID, 0);
338f74df 1938 MCDI_SET_QWORD(inadj, PTP_IN_ADJUST_FREQ, adjustment_ns);
7c236c43
SH		 1939 MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_SECONDS, 0);
1940 MCDI_SET_DWORD(inadj, PTP_IN_ADJUST_NANOSECONDS, 0);
1941 rc = efx_mcdi_rpc(efx, MC_CMD_PTP, inadj, sizeof(inadj),
1942 NULL, 0, NULL);
1943 if (rc != 0)
1944 return rc;
1945
cd6fe65e 1946 ptp_data->current_adjfreq = adjustment_ns;
7c236c43
SH		 1947 return 0;
1948}
1949
1950static int efx_phc_adjtime(struct ptp_clock_info *ptp, s64 delta)
1951{
a6f73460 1952 u32 nic_major, nic_minor;
7c236c43
SH		 1953 struct efx_ptp_data *ptp_data = container_of(ptp,
1954 struct efx_ptp_data,
1955 phc_clock_info);
ac36baf8 1956 struct efx_nic *efx = ptp_data->efx;
59cfc479 1957 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_ADJUST_LEN);
7c236c43 1958
a6f73460
LE		 1959 efx->ptp_data->ns_to_nic_time(delta, &nic_major, &nic_minor);
1960
7c236c43 1961 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_ADJUST);
c1d828bd 1962 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
cd6fe65e 1963 MCDI_SET_QWORD(inbuf, PTP_IN_ADJUST_FREQ, ptp_data->current_adjfreq);
a6f73460
LE		 1964 MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_MAJOR, nic_major);
1965 MCDI_SET_DWORD(inbuf, PTP_IN_ADJUST_MINOR, nic_minor);
7c236c43
SH		 1966 return efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
1967 NULL, 0, NULL);
1968}
1969
0fcb5c76 1970static int efx_phc_gettime(struct ptp_clock_info *ptp, struct timespec64 *ts)
7c236c43
SH		 1971{
1972 struct efx_ptp_data *ptp_data = container_of(ptp,
1973 struct efx_ptp_data,
1974 phc_clock_info);
ac36baf8 1975 struct efx_nic *efx = ptp_data->efx;
59cfc479
BH		 1976 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_READ_NIC_TIME_LEN);
1977 MCDI_DECLARE_BUF(outbuf, MC_CMD_PTP_OUT_READ_NIC_TIME_LEN);
7c236c43 1978 int rc;
a6f73460 1979 ktime_t kt;
7c236c43
SH		 1980
1981 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_READ_NIC_TIME);
c1d828bd 1982 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
7c236c43
SH		 1983
1984 rc = efx_mcdi_rpc(efx, MC_CMD_PTP, inbuf, sizeof(inbuf),
1985 outbuf, sizeof(outbuf), NULL);
1986 if (rc != 0)
1987 return rc;
1988
a6f73460
LE		 1989 kt = ptp_data->nic_to_kernel_time(
1990 MCDI_DWORD(outbuf, PTP_OUT_READ_NIC_TIME_MAJOR),
1991 MCDI_DWORD(outbuf, PTP_OUT_READ_NIC_TIME_MINOR), 0);
0fcb5c76 1992 *ts = ktime_to_timespec64(kt);
7c236c43
SH		 1993 return 0;
1994}
1995
1996static int efx_phc_settime(struct ptp_clock_info *ptp,
0fcb5c76 1997 const struct timespec64 *e_ts)
7c236c43
SH		 1998{
1999 /* Get the current NIC time, efx_phc_gettime.
2000 * Subtract from the desired time to get the offset
2001 * call efx_phc_adjtime with the offset
2002 */
2003 int rc;
0fcb5c76
RC		 2004 struct timespec64 time_now;
2005 struct timespec64 delta;
7c236c43
SH		 2006
2007 rc = efx_phc_gettime(ptp, &time_now);
2008 if (rc != 0)
2009 return rc;
2010
0fcb5c76 2011 delta = timespec64_sub(*e_ts, time_now);
7c236c43 2012
0fcb5c76 2013 rc = efx_phc_adjtime(ptp, timespec64_to_ns(&delta));
7c236c43
SH		 2014 if (rc != 0)
2015 return rc;
2016
2017 return 0;
2018}
2019
2020static int efx_phc_enable(struct ptp_clock_info *ptp,
2021 struct ptp_clock_request *request,
2022 int enable)
2023{
2024 struct efx_ptp_data *ptp_data = container_of(ptp,
2025 struct efx_ptp_data,
2026 phc_clock_info);
2027 if (request->type != PTP_CLK_REQ_PPS)
2028 return -EOPNOTSUPP;
2029
2030 ptp_data->nic_ts_enabled = !!enable;
2031 return 0;
2032}
2033
2935e3c3 2034const struct efx_channel_type efx_ptp_channel_type = {
7c236c43
SH		 2035 .handle_no_channel = efx_ptp_handle_no_channel,
2036 .pre_probe = efx_ptp_probe_channel,
2037 .post_remove = efx_ptp_remove_channel,
2038 .get_name = efx_ptp_get_channel_name,
2039 /* no copy operation; there is no need to reallocate this channel */
2040 .receive_skb = efx_ptp_rx,
2935e3c3 2041 .want_txqs = efx_ptp_want_txqs,
7c236c43
SH		 2042 .keep_eventq = false,
2043};
2044
ac36baf8 2045void efx_ptp_defer_probe_with_channel(struct efx_nic *efx)
7c236c43
SH		 2046{
2047 /* Check whether PTP is implemented on this NIC. The DISABLE
2048 * operation will succeed if and only if it is implemented.
2049 */
2050 if (efx_ptp_disable(efx) == 0)
2051 efx->extra_channel_type[EFX_EXTRA_CHANNEL_PTP] =
2052 &efx_ptp_channel_type;
2053}
2ea4dc28
AR		 2054
2055void efx_ptp_start_datapath(struct efx_nic *efx)
2056{
2057 if (efx_ptp_restart(efx))
2058 netif_err(efx, drv, efx->net_dev, "Failed to restart PTP.\n");
bd9a265d
JC		 2059 /* re-enable timestamping if it was previously enabled */
2060 if (efx->type->ptp_set_ts_sync_events)
2061 efx->type->ptp_set_ts_sync_events(efx, true, true);
2ea4dc28
AR		 2062}
2063
2064void efx_ptp_stop_datapath(struct efx_nic *efx)
2065{
bd9a265d
JC		 2066 /* temporarily disable timestamping */
2067 if (efx->type->ptp_set_ts_sync_events)
2068 efx->type->ptp_set_ts_sync_events(efx, false, true);
2ea4dc28
AR		 2069 efx_ptp_stop(efx);
2070}
Linux 6.x block layer and io_uring tree(s)