?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.7.7
[linux/fpc-iii.git] / net / sched / sch_taprio.c
blobb1eb12d33b9a6ca9d608d03684d884adc257f0ec
1 // SPDX-License-Identifier: GPL-2.0
2
3 /* net/sched/sch_taprio.c Time Aware Priority Scheduler
4 *
5 * Authors: Vinicius Costa Gomes <vinicius.gomes@intel.com>
6 *
7 */
8
9 #include <linux/types.h>
10 #include <linux/slab.h>
11 #include <linux/kernel.h>
12 #include <linux/string.h>
13 #include <linux/list.h>
14 #include <linux/errno.h>
15 #include <linux/skbuff.h>
16 #include <linux/math64.h>
17 #include <linux/module.h>
18 #include <linux/spinlock.h>
19 #include <linux/rcupdate.h>
20 #include <net/netlink.h>
21 #include <net/pkt_sched.h>
22 #include <net/pkt_cls.h>
23 #include <net/sch_generic.h>
24 #include <net/sock.h>
25 #include <net/tcp.h>
26
27 static LIST_HEAD(taprio_list);
28 static DEFINE_SPINLOCK(taprio_list_lock);
29
30 #define TAPRIO_ALL_GATES_OPEN -1
31
32 #define TXTIME_ASSIST_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST)
33 #define FULL_OFFLOAD_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD)
34 #define TAPRIO_FLAGS_INVALID U32_MAX
35
36 struct sched_entry {
37 struct list_head list;
38
39 /* The instant that this entry "closes" and the next one
40 * should open, the qdisc will make some effort so that no
41 * packet leaves after this time.
42 */
43 ktime_t close_time;
44 ktime_t next_txtime;
45 atomic_t budget;
46 int index;
47 u32 gate_mask;
48 u32 interval;
49 u8 command;
50 };
51
52 struct sched_gate_list {
53 struct rcu_head rcu;
54 struct list_head entries;
55 size_t num_entries;
56 ktime_t cycle_close_time;
57 s64 cycle_time;
58 s64 cycle_time_extension;
59 s64 base_time;
60 };
61
62 struct taprio_sched {
63 struct Qdisc **qdiscs;
64 struct Qdisc *root;
65 u32 flags;
66 enum tk_offsets tk_offset;
67 int clockid;
68 atomic64_t picos_per_byte; /* Using picoseconds because for 10Gbps+
69 * speeds it's sub-nanoseconds per byte
70 */
71
72 /* Protects the update side of the RCU protected current_entry */
73 spinlock_t current_entry_lock;
74 struct sched_entry __rcu *current_entry;
75 struct sched_gate_list __rcu *oper_sched;
76 struct sched_gate_list __rcu *admin_sched;
77 struct hrtimer advance_timer;
78 struct list_head taprio_list;
79 struct sk_buff *(*dequeue)(struct Qdisc *sch);
80 struct sk_buff *(*peek)(struct Qdisc *sch);
81 u32 txtime_delay;
82 };
83
84 struct __tc_taprio_qopt_offload {
85 refcount_t users;
86 struct tc_taprio_qopt_offload offload;
87 };
88
89 static ktime_t sched_base_time(const struct sched_gate_list *sched)
90 {
91 if (!sched)
92 return KTIME_MAX;
93
94 return ns_to_ktime(sched->base_time);
95 }
96
97 static ktime_t taprio_get_time(struct taprio_sched *q)
98 {
99 ktime_t mono = ktime_get();
100
101 switch (q->tk_offset) {
102 case TK_OFFS_MAX:
103 return mono;
104 default:
105 return ktime_mono_to_any(mono, q->tk_offset);
106 }
107
108 return KTIME_MAX;
109 }
110
111 static void taprio_free_sched_cb(struct rcu_head *head)
112 {
113 struct sched_gate_list *sched = container_of(head, struct sched_gate_list, rcu);
114 struct sched_entry *entry, *n;
115
116 if (!sched)
117 return;
118
119 list_for_each_entry_safe(entry, n, &sched->entries, list) {
120 list_del(&entry->list);
121 kfree(entry);
122 }
123
124 kfree(sched);
125 }
126
127 static void switch_schedules(struct taprio_sched *q,
128 struct sched_gate_list **admin,
129 struct sched_gate_list **oper)
130 {
131 rcu_assign_pointer(q->oper_sched, *admin);
132 rcu_assign_pointer(q->admin_sched, NULL);
133
134 if (*oper)
135 call_rcu(&(*oper)->rcu, taprio_free_sched_cb);
136
137 *oper = *admin;
138 *admin = NULL;
139 }
140
141 /* Get how much time has been already elapsed in the current cycle. */
142 static s32 get_cycle_time_elapsed(struct sched_gate_list *sched, ktime_t time)
143 {
144 ktime_t time_since_sched_start;
145 s32 time_elapsed;
146
147 time_since_sched_start = ktime_sub(time, sched->base_time);
148 div_s64_rem(time_since_sched_start, sched->cycle_time, &time_elapsed);
149
150 return time_elapsed;
151 }
152
153 static ktime_t get_interval_end_time(struct sched_gate_list *sched,
154 struct sched_gate_list *admin,
155 struct sched_entry *entry,
156 ktime_t intv_start)
157 {
158 s32 cycle_elapsed = get_cycle_time_elapsed(sched, intv_start);
159 ktime_t intv_end, cycle_ext_end, cycle_end;
160
161 cycle_end = ktime_add_ns(intv_start, sched->cycle_time - cycle_elapsed);
162 intv_end = ktime_add_ns(intv_start, entry->interval);
163 cycle_ext_end = ktime_add(cycle_end, sched->cycle_time_extension);
164
165 if (ktime_before(intv_end, cycle_end))
166 return intv_end;
167 else if (admin && admin != sched &&
168 ktime_after(admin->base_time, cycle_end) &&
169 ktime_before(admin->base_time, cycle_ext_end))
170 return admin->base_time;
171 else
172 return cycle_end;
173 }
174
175 static int length_to_duration(struct taprio_sched *q, int len)
176 {
177 return div_u64(len * atomic64_read(&q->picos_per_byte), 1000);
178 }
179
180 /* Returns the entry corresponding to next available interval. If
181 * validate_interval is set, it only validates whether the timestamp occurs
182 * when the gate corresponding to the skb's traffic class is open.
183 */
184 static struct sched_entry *find_entry_to_transmit(struct sk_buff *skb,
185 struct Qdisc *sch,
186 struct sched_gate_list *sched,
187 struct sched_gate_list *admin,
188 ktime_t time,
189 ktime_t *interval_start,
190 ktime_t *interval_end,
191 bool validate_interval)
192 {
193 ktime_t curr_intv_start, curr_intv_end, cycle_end, packet_transmit_time;
194 ktime_t earliest_txtime = KTIME_MAX, txtime, cycle, transmit_end_time;
195 struct sched_entry *entry = NULL, *entry_found = NULL;
196 struct taprio_sched *q = qdisc_priv(sch);
197 struct net_device *dev = qdisc_dev(sch);
198 bool entry_available = false;
199 s32 cycle_elapsed;
200 int tc, n;
201
202 tc = netdev_get_prio_tc_map(dev, skb->priority);
203 packet_transmit_time = length_to_duration(q, qdisc_pkt_len(skb));
204
205 *interval_start = 0;
206 *interval_end = 0;
207
208 if (!sched)
209 return NULL;
210
211 cycle = sched->cycle_time;
212 cycle_elapsed = get_cycle_time_elapsed(sched, time);
213 curr_intv_end = ktime_sub_ns(time, cycle_elapsed);
214 cycle_end = ktime_add_ns(curr_intv_end, cycle);
215
216 list_for_each_entry(entry, &sched->entries, list) {
217 curr_intv_start = curr_intv_end;
218 curr_intv_end = get_interval_end_time(sched, admin, entry,
219 curr_intv_start);
220
221 if (ktime_after(curr_intv_start, cycle_end))
222 break;
223
224 if (!(entry->gate_mask & BIT(tc)) ||
225 packet_transmit_time > entry->interval)
226 continue;
227
228 txtime = entry->next_txtime;
229
230 if (ktime_before(txtime, time) || validate_interval) {
231 transmit_end_time = ktime_add_ns(time, packet_transmit_time);
232 if ((ktime_before(curr_intv_start, time) &&
233 ktime_before(transmit_end_time, curr_intv_end)) ||
234 (ktime_after(curr_intv_start, time) && !validate_interval)) {
235 entry_found = entry;
236 *interval_start = curr_intv_start;
237 *interval_end = curr_intv_end;
238 break;
239 } else if (!entry_available && !validate_interval) {
240 /* Here, we are just trying to find out the
241 * first available interval in the next cycle.
242 */
243 entry_available = 1;
244 entry_found = entry;
245 *interval_start = ktime_add_ns(curr_intv_start, cycle);
246 *interval_end = ktime_add_ns(curr_intv_end, cycle);
247 }
248 } else if (ktime_before(txtime, earliest_txtime) &&
249 !entry_available) {
250 earliest_txtime = txtime;
251 entry_found = entry;
252 n = div_s64(ktime_sub(txtime, curr_intv_start), cycle);
253 *interval_start = ktime_add(curr_intv_start, n * cycle);
254 *interval_end = ktime_add(curr_intv_end, n * cycle);
255 }
256 }
257
258 return entry_found;
259 }
260
261 static bool is_valid_interval(struct sk_buff *skb, struct Qdisc *sch)
262 {
263 struct taprio_sched *q = qdisc_priv(sch);
264 struct sched_gate_list *sched, *admin;
265 ktime_t interval_start, interval_end;
266 struct sched_entry *entry;
267
268 rcu_read_lock();
269 sched = rcu_dereference(q->oper_sched);
270 admin = rcu_dereference(q->admin_sched);
271
272 entry = find_entry_to_transmit(skb, sch, sched, admin, skb->tstamp,
273 &interval_start, &interval_end, true);
274 rcu_read_unlock();
275
276 return entry;
277 }
278
279 static bool taprio_flags_valid(u32 flags)
280 {
281 /* Make sure no other flag bits are set. */
282 if (flags & ~(TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST |
283 TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
284 return false;
285 /* txtime-assist and full offload are mutually exclusive */
286 if ((flags & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST) &&
287 (flags & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
288 return false;
289 return true;
290 }
291
292 /* This returns the tstamp value set by TCP in terms of the set clock. */
293 static ktime_t get_tcp_tstamp(struct taprio_sched *q, struct sk_buff *skb)
294 {
295 unsigned int offset = skb_network_offset(skb);
296 const struct ipv6hdr *ipv6h;
297 const struct iphdr *iph;
298 struct ipv6hdr _ipv6h;
299
300 ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h);
301 if (!ipv6h)
302 return 0;
303
304 if (ipv6h->version == 4) {
305 iph = (struct iphdr *)ipv6h;
306 offset += iph->ihl * 4;
307
308 /* special-case 6in4 tunnelling, as that is a common way to get
309 * v6 connectivity in the home
310 */
311 if (iph->protocol == IPPROTO_IPV6) {
312 ipv6h = skb_header_pointer(skb, offset,
313 sizeof(_ipv6h), &_ipv6h);
314
315 if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP)
316 return 0;
317 } else if (iph->protocol != IPPROTO_TCP) {
318 return 0;
319 }
320 } else if (ipv6h->version == 6 && ipv6h->nexthdr != IPPROTO_TCP) {
321 return 0;
322 }
323
324 return ktime_mono_to_any(skb->skb_mstamp_ns, q->tk_offset);
325 }
326
327 /* There are a few scenarios where we will have to modify the txtime from
328 * what is read from next_txtime in sched_entry. They are:
329 * 1. If txtime is in the past,
330 * a. The gate for the traffic class is currently open and packet can be
331 * transmitted before it closes, schedule the packet right away.
332 * b. If the gate corresponding to the traffic class is going to open later
333 * in the cycle, set the txtime of packet to the interval start.
334 * 2. If txtime is in the future, there are packets corresponding to the
335 * current traffic class waiting to be transmitted. So, the following
336 * possibilities exist:
337 * a. We can transmit the packet before the window containing the txtime
338 * closes.
339 * b. The window might close before the transmission can be completed
340 * successfully. So, schedule the packet in the next open window.
341 */
342 static long get_packet_txtime(struct sk_buff *skb, struct Qdisc *sch)
343 {
344 ktime_t transmit_end_time, interval_end, interval_start, tcp_tstamp;
345 struct taprio_sched *q = qdisc_priv(sch);
346 struct sched_gate_list *sched, *admin;
347 ktime_t minimum_time, now, txtime;
348 int len, packet_transmit_time;
349 struct sched_entry *entry;
350 bool sched_changed;
351
352 now = taprio_get_time(q);
353 minimum_time = ktime_add_ns(now, q->txtime_delay);
354
355 tcp_tstamp = get_tcp_tstamp(q, skb);
356 minimum_time = max_t(ktime_t, minimum_time, tcp_tstamp);
357
358 rcu_read_lock();
359 admin = rcu_dereference(q->admin_sched);
360 sched = rcu_dereference(q->oper_sched);
361 if (admin && ktime_after(minimum_time, admin->base_time))
362 switch_schedules(q, &admin, &sched);
363
364 /* Until the schedule starts, all the queues are open */
365 if (!sched || ktime_before(minimum_time, sched->base_time)) {
366 txtime = minimum_time;
367 goto done;
368 }
369
370 len = qdisc_pkt_len(skb);
371 packet_transmit_time = length_to_duration(q, len);
372
373 do {
374 sched_changed = 0;
375
376 entry = find_entry_to_transmit(skb, sch, sched, admin,
377 minimum_time,
378 &interval_start, &interval_end,
379 false);
380 if (!entry) {
381 txtime = 0;
382 goto done;
383 }
384
385 txtime = entry->next_txtime;
386 txtime = max_t(ktime_t, txtime, minimum_time);
387 txtime = max_t(ktime_t, txtime, interval_start);
388
389 if (admin && admin != sched &&
390 ktime_after(txtime, admin->base_time)) {
391 sched = admin;
392 sched_changed = 1;
393 continue;
394 }
395
396 transmit_end_time = ktime_add(txtime, packet_transmit_time);
397 minimum_time = transmit_end_time;
398
399 /* Update the txtime of current entry to the next time it's
400 * interval starts.
401 */
402 if (ktime_after(transmit_end_time, interval_end))
403 entry->next_txtime = ktime_add(interval_start, sched->cycle_time);
404 } while (sched_changed || ktime_after(transmit_end_time, interval_end));
405
406 entry->next_txtime = transmit_end_time;
407
408 done:
409 rcu_read_unlock();
410 return txtime;
411 }
412
413 static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
414 struct sk_buff **to_free)
415 {
416 struct taprio_sched *q = qdisc_priv(sch);
417 struct Qdisc *child;
418 int queue;
419
420 queue = skb_get_queue_mapping(skb);
421
422 child = q->qdiscs[queue];
423 if (unlikely(!child))
424 return qdisc_drop(skb, sch, to_free);
425
426 if (skb->sk && sock_flag(skb->sk, SOCK_TXTIME)) {
427 if (!is_valid_interval(skb, sch))
428 return qdisc_drop(skb, sch, to_free);
429 } else if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
430 skb->tstamp = get_packet_txtime(skb, sch);
431 if (!skb->tstamp)
432 return qdisc_drop(skb, sch, to_free);
433 }
434
435 qdisc_qstats_backlog_inc(sch, skb);
436 sch->q.qlen++;
437
438 return qdisc_enqueue(skb, child, to_free);
439 }
440
441 static struct sk_buff *taprio_peek_soft(struct Qdisc *sch)
442 {
443 struct taprio_sched *q = qdisc_priv(sch);
444 struct net_device *dev = qdisc_dev(sch);
445 struct sched_entry *entry;
446 struct sk_buff *skb;
447 u32 gate_mask;
448 int i;
449
450 rcu_read_lock();
451 entry = rcu_dereference(q->current_entry);
452 gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
453 rcu_read_unlock();
454
455 if (!gate_mask)
456 return NULL;
457
458 for (i = 0; i < dev->num_tx_queues; i++) {
459 struct Qdisc *child = q->qdiscs[i];
460 int prio;
461 u8 tc;
462
463 if (unlikely(!child))
464 continue;
465
466 skb = child->ops->peek(child);
467 if (!skb)
468 continue;
469
470 if (TXTIME_ASSIST_IS_ENABLED(q->flags))
471 return skb;
472
473 prio = skb->priority;
474 tc = netdev_get_prio_tc_map(dev, prio);
475
476 if (!(gate_mask & BIT(tc)))
477 continue;
478
479 return skb;
480 }
481
482 return NULL;
483 }
484
485 static struct sk_buff *taprio_peek_offload(struct Qdisc *sch)
486 {
487 struct taprio_sched *q = qdisc_priv(sch);
488 struct net_device *dev = qdisc_dev(sch);
489 struct sk_buff *skb;
490 int i;
491
492 for (i = 0; i < dev->num_tx_queues; i++) {
493 struct Qdisc *child = q->qdiscs[i];
494
495 if (unlikely(!child))
496 continue;
497
498 skb = child->ops->peek(child);
499 if (!skb)
500 continue;
501
502 return skb;
503 }
504
505 return NULL;
506 }
507
508 static struct sk_buff *taprio_peek(struct Qdisc *sch)
509 {
510 struct taprio_sched *q = qdisc_priv(sch);
511
512 return q->peek(sch);
513 }
514
515 static void taprio_set_budget(struct taprio_sched *q, struct sched_entry *entry)
516 {
517 atomic_set(&entry->budget,
518 div64_u64((u64)entry->interval * 1000,
519 atomic64_read(&q->picos_per_byte)));
520 }
521
522 static struct sk_buff *taprio_dequeue_soft(struct Qdisc *sch)
523 {
524 struct taprio_sched *q = qdisc_priv(sch);
525 struct net_device *dev = qdisc_dev(sch);
526 struct sk_buff *skb = NULL;
527 struct sched_entry *entry;
528 u32 gate_mask;
529 int i;
530
531 rcu_read_lock();
532 entry = rcu_dereference(q->current_entry);
533 /* if there's no entry, it means that the schedule didn't
534 * start yet, so force all gates to be open, this is in
535 * accordance to IEEE 802.1Qbv-2015 Section 8.6.9.4.5
536 * "AdminGateSates"
537 */
538 gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
539
540 if (!gate_mask)
541 goto done;
542
543 for (i = 0; i < dev->num_tx_queues; i++) {
544 struct Qdisc *child = q->qdiscs[i];
545 ktime_t guard;
546 int prio;
547 int len;
548 u8 tc;
549
550 if (unlikely(!child))
551 continue;
552
553 if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
554 skb = child->ops->dequeue(child);
555 if (!skb)
556 continue;
557 goto skb_found;
558 }
559
560 skb = child->ops->peek(child);
561 if (!skb)
562 continue;
563
564 prio = skb->priority;
565 tc = netdev_get_prio_tc_map(dev, prio);
566
567 if (!(gate_mask & BIT(tc))) {
568 skb = NULL;
569 continue;
570 }
571
572 len = qdisc_pkt_len(skb);
573 guard = ktime_add_ns(taprio_get_time(q),
574 length_to_duration(q, len));
575
576 /* In the case that there's no gate entry, there's no
577 * guard band ...
578 */
579 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
580 ktime_after(guard, entry->close_time)) {
581 skb = NULL;
582 continue;
583 }
584
585 /* ... and no budget. */
586 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
587 atomic_sub_return(len, &entry->budget) < 0) {
588 skb = NULL;
589 continue;
590 }
591
592 skb = child->ops->dequeue(child);
593 if (unlikely(!skb))
594 goto done;
595
596 skb_found:
597 qdisc_bstats_update(sch, skb);
598 qdisc_qstats_backlog_dec(sch, skb);
599 sch->q.qlen--;
600
601 goto done;
602 }
603
604 done:
605 rcu_read_unlock();
606
607 return skb;
608 }
609
610 static struct sk_buff *taprio_dequeue_offload(struct Qdisc *sch)
611 {
612 struct taprio_sched *q = qdisc_priv(sch);
613 struct net_device *dev = qdisc_dev(sch);
614 struct sk_buff *skb;
615 int i;
616
617 for (i = 0; i < dev->num_tx_queues; i++) {
618 struct Qdisc *child = q->qdiscs[i];
619
620 if (unlikely(!child))
621 continue;
622
623 skb = child->ops->dequeue(child);
624 if (unlikely(!skb))
625 continue;
626
627 qdisc_bstats_update(sch, skb);
628 qdisc_qstats_backlog_dec(sch, skb);
629 sch->q.qlen--;
630
631 return skb;
632 }
633
634 return NULL;
635 }
636
637 static struct sk_buff *taprio_dequeue(struct Qdisc *sch)
638 {
639 struct taprio_sched *q = qdisc_priv(sch);
640
641 return q->dequeue(sch);
642 }
643
644 static bool should_restart_cycle(const struct sched_gate_list *oper,
645 const struct sched_entry *entry)
646 {
647 if (list_is_last(&entry->list, &oper->entries))
648 return true;
649
650 if (ktime_compare(entry->close_time, oper->cycle_close_time) == 0)
651 return true;
652
653 return false;
654 }
655
656 static bool should_change_schedules(const struct sched_gate_list *admin,
657 const struct sched_gate_list *oper,
658 ktime_t close_time)
659 {
660 ktime_t next_base_time, extension_time;
661
662 if (!admin)
663 return false;
664
665 next_base_time = sched_base_time(admin);
666
667 /* This is the simple case, the close_time would fall after
668 * the next schedule base_time.
669 */
670 if (ktime_compare(next_base_time, close_time) <= 0)
671 return true;
672
673 /* This is the cycle_time_extension case, if the close_time
674 * plus the amount that can be extended would fall after the
675 * next schedule base_time, we can extend the current schedule
676 * for that amount.
677 */
678 extension_time = ktime_add_ns(close_time, oper->cycle_time_extension);
679
680 /* FIXME: the IEEE 802.1Q-2018 Specification isn't clear about
681 * how precisely the extension should be made. So after
682 * conformance testing, this logic may change.
683 */
684 if (ktime_compare(next_base_time, extension_time) <= 0)
685 return true;
686
687 return false;
688 }
689
690 static enum hrtimer_restart advance_sched(struct hrtimer *timer)
691 {
692 struct taprio_sched *q = container_of(timer, struct taprio_sched,
693 advance_timer);
694 struct sched_gate_list *oper, *admin;
695 struct sched_entry *entry, *next;
696 struct Qdisc *sch = q->root;
697 ktime_t close_time;
698
699 spin_lock(&q->current_entry_lock);
700 entry = rcu_dereference_protected(q->current_entry,
701 lockdep_is_held(&q->current_entry_lock));
702 oper = rcu_dereference_protected(q->oper_sched,
703 lockdep_is_held(&q->current_entry_lock));
704 admin = rcu_dereference_protected(q->admin_sched,
705 lockdep_is_held(&q->current_entry_lock));
706
707 if (!oper)
708 switch_schedules(q, &admin, &oper);
709
710 /* This can happen in two cases: 1. this is the very first run
711 * of this function (i.e. we weren't running any schedule
712 * previously); 2. The previous schedule just ended. The first
713 * entry of all schedules are pre-calculated during the
714 * schedule initialization.
715 */
716 if (unlikely(!entry || entry->close_time == oper->base_time)) {
717 next = list_first_entry(&oper->entries, struct sched_entry,
718 list);
719 close_time = next->close_time;
720 goto first_run;
721 }
722
723 if (should_restart_cycle(oper, entry)) {
724 next = list_first_entry(&oper->entries, struct sched_entry,
725 list);
726 oper->cycle_close_time = ktime_add_ns(oper->cycle_close_time,
727 oper->cycle_time);
728 } else {
729 next = list_next_entry(entry, list);
730 }
731
732 close_time = ktime_add_ns(entry->close_time, next->interval);
733 close_time = min_t(ktime_t, close_time, oper->cycle_close_time);
734
735 if (should_change_schedules(admin, oper, close_time)) {
736 /* Set things so the next time this runs, the new
737 * schedule runs.
738 */
739 close_time = sched_base_time(admin);
740 switch_schedules(q, &admin, &oper);
741 }
742
743 next->close_time = close_time;
744 taprio_set_budget(q, next);
745
746 first_run:
747 rcu_assign_pointer(q->current_entry, next);
748 spin_unlock(&q->current_entry_lock);
749
750 hrtimer_set_expires(&q->advance_timer, close_time);
751
752 rcu_read_lock();
753 __netif_schedule(sch);
754 rcu_read_unlock();
755
756 return HRTIMER_RESTART;
757 }
758
759 static const struct nla_policy entry_policy[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = {
760 [TCA_TAPRIO_SCHED_ENTRY_INDEX] = { .type = NLA_U32 },
761 [TCA_TAPRIO_SCHED_ENTRY_CMD] = { .type = NLA_U8 },
762 [TCA_TAPRIO_SCHED_ENTRY_GATE_MASK] = { .type = NLA_U32 },
763 [TCA_TAPRIO_SCHED_ENTRY_INTERVAL] = { .type = NLA_U32 },
764 };
765
766 static const struct nla_policy taprio_policy[TCA_TAPRIO_ATTR_MAX + 1] = {
767 [TCA_TAPRIO_ATTR_PRIOMAP] = {
768 .len = sizeof(struct tc_mqprio_qopt)
769 },
770 [TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST] = { .type = NLA_NESTED },
771 [TCA_TAPRIO_ATTR_SCHED_BASE_TIME] = { .type = NLA_S64 },
772 [TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY] = { .type = NLA_NESTED },
773 [TCA_TAPRIO_ATTR_SCHED_CLOCKID] = { .type = NLA_S32 },
774 [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME] = { .type = NLA_S64 },
775 [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION] = { .type = NLA_S64 },
776 [TCA_TAPRIO_ATTR_FLAGS] = { .type = NLA_U32 },
777 [TCA_TAPRIO_ATTR_TXTIME_DELAY] = { .type = NLA_U32 },
778 };
779
780 static int fill_sched_entry(struct nlattr **tb, struct sched_entry *entry,
781 struct netlink_ext_ack *extack)
782 {
783 u32 interval = 0;
784
785 if (tb[TCA_TAPRIO_SCHED_ENTRY_CMD])
786 entry->command = nla_get_u8(
787 tb[TCA_TAPRIO_SCHED_ENTRY_CMD]);
788
789 if (tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK])
790 entry->gate_mask = nla_get_u32(
791 tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK]);
792
793 if (tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL])
794 interval = nla_get_u32(
795 tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL]);
796
797 if (interval == 0) {
798 NL_SET_ERR_MSG(extack, "Invalid interval for schedule entry");
799 return -EINVAL;
800 }
801
802 entry->interval = interval;
803
804 return 0;
805 }
806
807 static int parse_sched_entry(struct nlattr *n, struct sched_entry *entry,
808 int index, struct netlink_ext_ack *extack)
809 {
810 struct nlattr *tb[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = { };
811 int err;
812
813 err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_SCHED_ENTRY_MAX, n,
814 entry_policy, NULL);
815 if (err < 0) {
816 NL_SET_ERR_MSG(extack, "Could not parse nested entry");
817 return -EINVAL;
818 }
819
820 entry->index = index;
821
822 return fill_sched_entry(tb, entry, extack);
823 }
824
825 static int parse_sched_list(struct nlattr *list,
826 struct sched_gate_list *sched,
827 struct netlink_ext_ack *extack)
828 {
829 struct nlattr *n;
830 int err, rem;
831 int i = 0;
832
833 if (!list)
834 return -EINVAL;
835
836 nla_for_each_nested(n, list, rem) {
837 struct sched_entry *entry;
838
839 if (nla_type(n) != TCA_TAPRIO_SCHED_ENTRY) {
840 NL_SET_ERR_MSG(extack, "Attribute is not of type 'entry'");
841 continue;
842 }
843
844 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
845 if (!entry) {
846 NL_SET_ERR_MSG(extack, "Not enough memory for entry");
847 return -ENOMEM;
848 }
849
850 err = parse_sched_entry(n, entry, i, extack);
851 if (err < 0) {
852 kfree(entry);
853 return err;
854 }
855
856 list_add_tail(&entry->list, &sched->entries);
857 i++;
858 }
859
860 sched->num_entries = i;
861
862 return i;
863 }
864
865 static int parse_taprio_schedule(struct nlattr **tb,
866 struct sched_gate_list *new,
867 struct netlink_ext_ack *extack)
868 {
869 int err = 0;
870
871 if (tb[TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]) {
872 NL_SET_ERR_MSG(extack, "Adding a single entry is not supported");
873 return -ENOTSUPP;
874 }
875
876 if (tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME])
877 new->base_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME]);
878
879 if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION])
880 new->cycle_time_extension = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION]);
881
882 if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME])
883 new->cycle_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]);
884
885 if (tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST])
886 err = parse_sched_list(
887 tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST], new, extack);
888 if (err < 0)
889 return err;
890
891 if (!new->cycle_time) {
892 struct sched_entry *entry;
893 ktime_t cycle = 0;
894
895 list_for_each_entry(entry, &new->entries, list)
896 cycle = ktime_add_ns(cycle, entry->interval);
897 new->cycle_time = cycle;
898 }
899
900 return 0;
901 }
902
903 static int taprio_parse_mqprio_opt(struct net_device *dev,
904 struct tc_mqprio_qopt *qopt,
905 struct netlink_ext_ack *extack,
906 u32 taprio_flags)
907 {
908 int i, j;
909
910 if (!qopt && !dev->num_tc) {
911 NL_SET_ERR_MSG(extack, "'mqprio' configuration is necessary");
912 return -EINVAL;
913 }
914
915 /* If num_tc is already set, it means that the user already
916 * configured the mqprio part
917 */
918 if (dev->num_tc)
919 return 0;
920
921 /* Verify num_tc is not out of max range */
922 if (qopt->num_tc > TC_MAX_QUEUE) {
923 NL_SET_ERR_MSG(extack, "Number of traffic classes is outside valid range");
924 return -EINVAL;
925 }
926
927 /* taprio imposes that traffic classes map 1:n to tx queues */
928 if (qopt->num_tc > dev->num_tx_queues) {
929 NL_SET_ERR_MSG(extack, "Number of traffic classes is greater than number of HW queues");
930 return -EINVAL;
931 }
932
933 /* Verify priority mapping uses valid tcs */
934 for (i = 0; i <= TC_BITMASK; i++) {
935 if (qopt->prio_tc_map[i] >= qopt->num_tc) {
936 NL_SET_ERR_MSG(extack, "Invalid traffic class in priority to traffic class mapping");
937 return -EINVAL;
938 }
939 }
940
941 for (i = 0; i < qopt->num_tc; i++) {
942 unsigned int last = qopt->offset[i] + qopt->count[i];
943
944 /* Verify the queue count is in tx range being equal to the
945 * real_num_tx_queues indicates the last queue is in use.
946 */
947 if (qopt->offset[i] >= dev->num_tx_queues ||
948 !qopt->count[i] ||
949 last > dev->real_num_tx_queues) {
950 NL_SET_ERR_MSG(extack, "Invalid queue in traffic class to queue mapping");
951 return -EINVAL;
952 }
953
954 if (TXTIME_ASSIST_IS_ENABLED(taprio_flags))
955 continue;
956
957 /* Verify that the offset and counts do not overlap */
958 for (j = i + 1; j < qopt->num_tc; j++) {
959 if (last > qopt->offset[j]) {
960 NL_SET_ERR_MSG(extack, "Detected overlap in the traffic class to queue mapping");
961 return -EINVAL;
962 }
963 }
964 }
965
966 return 0;
967 }
968
969 static int taprio_get_start_time(struct Qdisc *sch,
970 struct sched_gate_list *sched,
971 ktime_t *start)
972 {
973 struct taprio_sched *q = qdisc_priv(sch);
974 ktime_t now, base, cycle;
975 s64 n;
976
977 base = sched_base_time(sched);
978 now = taprio_get_time(q);
979
980 if (ktime_after(base, now)) {
981 *start = base;
982 return 0;
983 }
984
985 cycle = sched->cycle_time;
986
987 /* The qdisc is expected to have at least one sched_entry. Moreover,
988 * any entry must have 'interval' > 0. Thus if the cycle time is zero,
989 * something went really wrong. In that case, we should warn about this
990 * inconsistent state and return error.
991 */
992 if (WARN_ON(!cycle))
993 return -EFAULT;
994
995 /* Schedule the start time for the beginning of the next
996 * cycle.
997 */
998 n = div64_s64(ktime_sub_ns(now, base), cycle);
999 *start = ktime_add_ns(base, (n + 1) * cycle);
1000 return 0;
1001 }
1002
1003 static void setup_first_close_time(struct taprio_sched *q,
1004 struct sched_gate_list *sched, ktime_t base)
1005 {
1006 struct sched_entry *first;
1007 ktime_t cycle;
1008
1009 first = list_first_entry(&sched->entries,
1010 struct sched_entry, list);
1011
1012 cycle = sched->cycle_time;
1013
1014 /* FIXME: find a better place to do this */
1015 sched->cycle_close_time = ktime_add_ns(base, cycle);
1016
1017 first->close_time = ktime_add_ns(base, first->interval);
1018 taprio_set_budget(q, first);
1019 rcu_assign_pointer(q->current_entry, NULL);
1020 }
1021
1022 static void taprio_start_sched(struct Qdisc *sch,
1023 ktime_t start, struct sched_gate_list *new)
1024 {
1025 struct taprio_sched *q = qdisc_priv(sch);
1026 ktime_t expires;
1027
1028 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1029 return;
1030
1031 expires = hrtimer_get_expires(&q->advance_timer);
1032 if (expires == 0)
1033 expires = KTIME_MAX;
1034
1035 /* If the new schedule starts before the next expiration, we
1036 * reprogram it to the earliest one, so we change the admin
1037 * schedule to the operational one at the right time.
1038 */
1039 start = min_t(ktime_t, start, expires);
1040
1041 hrtimer_start(&q->advance_timer, start, HRTIMER_MODE_ABS);
1042 }
1043
1044 static void taprio_set_picos_per_byte(struct net_device *dev,
1045 struct taprio_sched *q)
1046 {
1047 struct ethtool_link_ksettings ecmd;
1048 int speed = SPEED_10;
1049 int picos_per_byte;
1050 int err;
1051
1052 err = __ethtool_get_link_ksettings(dev, &ecmd);
1053 if (err < 0)
1054 goto skip;
1055
1056 if (ecmd.base.speed && ecmd.base.speed != SPEED_UNKNOWN)
1057 speed = ecmd.base.speed;
1058
1059 skip:
1060 picos_per_byte = (USEC_PER_SEC * 8) / speed;
1061
1062 atomic64_set(&q->picos_per_byte, picos_per_byte);
1063 netdev_dbg(dev, "taprio: set %s's picos_per_byte to: %lld, linkspeed: %d\n",
1064 dev->name, (long long)atomic64_read(&q->picos_per_byte),
1065 ecmd.base.speed);
1066 }
1067
1068 static int taprio_dev_notifier(struct notifier_block *nb, unsigned long event,
1069 void *ptr)
1070 {
1071 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1072 struct net_device *qdev;
1073 struct taprio_sched *q;
1074 bool found = false;
1075
1076 ASSERT_RTNL();
1077
1078 if (event != NETDEV_UP && event != NETDEV_CHANGE)
1079 return NOTIFY_DONE;
1080
1081 spin_lock(&taprio_list_lock);
1082 list_for_each_entry(q, &taprio_list, taprio_list) {
1083 qdev = qdisc_dev(q->root);
1084 if (qdev == dev) {
1085 found = true;
1086 break;
1087 }
1088 }
1089 spin_unlock(&taprio_list_lock);
1090
1091 if (found)
1092 taprio_set_picos_per_byte(dev, q);
1093
1094 return NOTIFY_DONE;
1095 }
1096
1097 static void setup_txtime(struct taprio_sched *q,
1098 struct sched_gate_list *sched, ktime_t base)
1099 {
1100 struct sched_entry *entry;
1101 u32 interval = 0;
1102
1103 list_for_each_entry(entry, &sched->entries, list) {
1104 entry->next_txtime = ktime_add_ns(base, interval);
1105 interval += entry->interval;
1106 }
1107 }
1108
1109 static struct tc_taprio_qopt_offload *taprio_offload_alloc(int num_entries)
1110 {
1111 size_t size = sizeof(struct tc_taprio_sched_entry) * num_entries +
1112 sizeof(struct __tc_taprio_qopt_offload);
1113 struct __tc_taprio_qopt_offload *__offload;
1114
1115 __offload = kzalloc(size, GFP_KERNEL);
1116 if (!__offload)
1117 return NULL;
1118
1119 refcount_set(&__offload->users, 1);
1120
1121 return &__offload->offload;
1122 }
1123
1124 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
1125 *offload)
1126 {
1127 struct __tc_taprio_qopt_offload *__offload;
1128
1129 __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1130 offload);
1131
1132 refcount_inc(&__offload->users);
1133
1134 return offload;
1135 }
1136 EXPORT_SYMBOL_GPL(taprio_offload_get);
1137
1138 void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
1139 {
1140 struct __tc_taprio_qopt_offload *__offload;
1141
1142 __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1143 offload);
1144
1145 if (!refcount_dec_and_test(&__offload->users))
1146 return;
1147
1148 kfree(__offload);
1149 }
1150 EXPORT_SYMBOL_GPL(taprio_offload_free);
1151
1152 /* The function will only serve to keep the pointers to the "oper" and "admin"
1153 * schedules valid in relation to their base times, so when calling dump() the
1154 * users looks at the right schedules.
1155 * When using full offload, the admin configuration is promoted to oper at the
1156 * base_time in the PHC time domain. But because the system time is not
1157 * necessarily in sync with that, we can't just trigger a hrtimer to call
1158 * switch_schedules at the right hardware time.
1159 * At the moment we call this by hand right away from taprio, but in the future
1160 * it will be useful to create a mechanism for drivers to notify taprio of the
1161 * offload state (PENDING, ACTIVE, INACTIVE) so it can be visible in dump().
1162 * This is left as TODO.
1163 */
1164 static void taprio_offload_config_changed(struct taprio_sched *q)
1165 {
1166 struct sched_gate_list *oper, *admin;
1167
1168 spin_lock(&q->current_entry_lock);
1169
1170 oper = rcu_dereference_protected(q->oper_sched,
1171 lockdep_is_held(&q->current_entry_lock));
1172 admin = rcu_dereference_protected(q->admin_sched,
1173 lockdep_is_held(&q->current_entry_lock));
1174
1175 switch_schedules(q, &admin, &oper);
1176
1177 spin_unlock(&q->current_entry_lock);
1178 }
1179
1180 static void taprio_sched_to_offload(struct taprio_sched *q,
1181 struct sched_gate_list *sched,
1182 const struct tc_mqprio_qopt *mqprio,
1183 struct tc_taprio_qopt_offload *offload)
1184 {
1185 struct sched_entry *entry;
1186 int i = 0;
1187
1188 offload->base_time = sched->base_time;
1189 offload->cycle_time = sched->cycle_time;
1190 offload->cycle_time_extension = sched->cycle_time_extension;
1191
1192 list_for_each_entry(entry, &sched->entries, list) {
1193 struct tc_taprio_sched_entry *e = &offload->entries[i];
1194
1195 e->command = entry->command;
1196 e->interval = entry->interval;
1197 e->gate_mask = entry->gate_mask;
1198 i++;
1199 }
1200
1201 offload->num_entries = i;
1202 }
1203
1204 static int taprio_enable_offload(struct net_device *dev,
1205 struct tc_mqprio_qopt *mqprio,
1206 struct taprio_sched *q,
1207 struct sched_gate_list *sched,
1208 struct netlink_ext_ack *extack)
1209 {
1210 const struct net_device_ops *ops = dev->netdev_ops;
1211 struct tc_taprio_qopt_offload *offload;
1212 int err = 0;
1213
1214 if (!ops->ndo_setup_tc) {
1215 NL_SET_ERR_MSG(extack,
1216 "Device does not support taprio offload");
1217 return -EOPNOTSUPP;
1218 }
1219
1220 offload = taprio_offload_alloc(sched->num_entries);
1221 if (!offload) {
1222 NL_SET_ERR_MSG(extack,
1223 "Not enough memory for enabling offload mode");
1224 return -ENOMEM;
1225 }
1226 offload->enable = 1;
1227 taprio_sched_to_offload(q, sched, mqprio, offload);
1228
1229 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1230 if (err < 0) {
1231 NL_SET_ERR_MSG(extack,
1232 "Device failed to setup taprio offload");
1233 goto done;
1234 }
1235
1236 done:
1237 taprio_offload_free(offload);
1238
1239 return err;
1240 }
1241
1242 static int taprio_disable_offload(struct net_device *dev,
1243 struct taprio_sched *q,
1244 struct netlink_ext_ack *extack)
1245 {
1246 const struct net_device_ops *ops = dev->netdev_ops;
1247 struct tc_taprio_qopt_offload *offload;
1248 int err;
1249
1250 if (!FULL_OFFLOAD_IS_ENABLED(q->flags))
1251 return 0;
1252
1253 if (!ops->ndo_setup_tc)
1254 return -EOPNOTSUPP;
1255
1256 offload = taprio_offload_alloc(0);
1257 if (!offload) {
1258 NL_SET_ERR_MSG(extack,
1259 "Not enough memory to disable offload mode");
1260 return -ENOMEM;
1261 }
1262 offload->enable = 0;
1263
1264 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1265 if (err < 0) {
1266 NL_SET_ERR_MSG(extack,
1267 "Device failed to disable offload");
1268 goto out;
1269 }
1270
1271 out:
1272 taprio_offload_free(offload);
1273
1274 return err;
1275 }
1276
1277 /* If full offload is enabled, the only possible clockid is the net device's
1278 * PHC. For that reason, specifying a clockid through netlink is incorrect.
1279 * For txtime-assist, it is implicitly assumed that the device's PHC is kept
1280 * in sync with the specified clockid via a user space daemon such as phc2sys.
1281 * For both software taprio and txtime-assist, the clockid is used for the
1282 * hrtimer that advances the schedule and hence mandatory.
1283 */
1284 static int taprio_parse_clockid(struct Qdisc *sch, struct nlattr **tb,
1285 struct netlink_ext_ack *extack)
1286 {
1287 struct taprio_sched *q = qdisc_priv(sch);
1288 struct net_device *dev = qdisc_dev(sch);
1289 int err = -EINVAL;
1290
1291 if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1292 const struct ethtool_ops *ops = dev->ethtool_ops;
1293 struct ethtool_ts_info info = {
1294 .cmd = ETHTOOL_GET_TS_INFO,
1295 .phc_index = -1,
1296 };
1297
1298 if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1299 NL_SET_ERR_MSG(extack,
1300 "The 'clockid' cannot be specified for full offload");
1301 goto out;
1302 }
1303
1304 if (ops && ops->get_ts_info)
1305 err = ops->get_ts_info(dev, &info);
1306
1307 if (err || info.phc_index < 0) {
1308 NL_SET_ERR_MSG(extack,
1309 "Device does not have a PTP clock");
1310 err = -ENOTSUPP;
1311 goto out;
1312 }
1313 } else if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1314 int clockid = nla_get_s32(tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]);
1315
1316 /* We only support static clockids and we don't allow
1317 * for it to be modified after the first init.
1318 */
1319 if (clockid < 0 ||
1320 (q->clockid != -1 && q->clockid != clockid)) {
1321 NL_SET_ERR_MSG(extack,
1322 "Changing the 'clockid' of a running schedule is not supported");
1323 err = -ENOTSUPP;
1324 goto out;
1325 }
1326
1327 switch (clockid) {
1328 case CLOCK_REALTIME:
1329 q->tk_offset = TK_OFFS_REAL;
1330 break;
1331 case CLOCK_MONOTONIC:
1332 q->tk_offset = TK_OFFS_MAX;
1333 break;
1334 case CLOCK_BOOTTIME:
1335 q->tk_offset = TK_OFFS_BOOT;
1336 break;
1337 case CLOCK_TAI:
1338 q->tk_offset = TK_OFFS_TAI;
1339 break;
1340 default:
1341 NL_SET_ERR_MSG(extack, "Invalid 'clockid'");
1342 err = -EINVAL;
1343 goto out;
1344 }
1345
1346 q->clockid = clockid;
1347 } else {
1348 NL_SET_ERR_MSG(extack, "Specifying a 'clockid' is mandatory");
1349 goto out;
1350 }
1351
1352 /* Everything went ok, return success. */
1353 err = 0;
1354
1355 out:
1356 return err;
1357 }
1358
1359 static int taprio_mqprio_cmp(const struct net_device *dev,
1360 const struct tc_mqprio_qopt *mqprio)
1361 {
1362 int i;
1363
1364 if (!mqprio || mqprio->num_tc != dev->num_tc)
1365 return -1;
1366
1367 for (i = 0; i < mqprio->num_tc; i++)
1368 if (dev->tc_to_txq[i].count != mqprio->count[i] ||
1369 dev->tc_to_txq[i].offset != mqprio->offset[i])
1370 return -1;
1371
1372 for (i = 0; i <= TC_BITMASK; i++)
1373 if (dev->prio_tc_map[i] != mqprio->prio_tc_map[i])
1374 return -1;
1375
1376 return 0;
1377 }
1378
1379 /* The semantics of the 'flags' argument in relation to 'change()'
1380 * requests, are interpreted following two rules (which are applied in
1381 * this order): (1) an omitted 'flags' argument is interpreted as
1382 * zero; (2) the 'flags' of a "running" taprio instance cannot be
1383 * changed.
1384 */
1385 static int taprio_new_flags(const struct nlattr *attr, u32 old,
1386 struct netlink_ext_ack *extack)
1387 {
1388 u32 new = 0;
1389
1390 if (attr)
1391 new = nla_get_u32(attr);
1392
1393 if (old != TAPRIO_FLAGS_INVALID && old != new) {
1394 NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
1395 return -EOPNOTSUPP;
1396 }
1397
1398 if (!taprio_flags_valid(new)) {
1399 NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
1400 return -EINVAL;
1401 }
1402
1403 return new;
1404 }
1405
1406 static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
1407 struct netlink_ext_ack *extack)
1408 {
1409 struct nlattr *tb[TCA_TAPRIO_ATTR_MAX + 1] = { };
1410 struct sched_gate_list *oper, *admin, *new_admin;
1411 struct taprio_sched *q = qdisc_priv(sch);
1412 struct net_device *dev = qdisc_dev(sch);
1413 struct tc_mqprio_qopt *mqprio = NULL;
1414 unsigned long flags;
1415 ktime_t start;
1416 int i, err;
1417
1418 err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_ATTR_MAX, opt,
1419 taprio_policy, extack);
1420 if (err < 0)
1421 return err;
1422
1423 if (tb[TCA_TAPRIO_ATTR_PRIOMAP])
1424 mqprio = nla_data(tb[TCA_TAPRIO_ATTR_PRIOMAP]);
1425
1426 err = taprio_new_flags(tb[TCA_TAPRIO_ATTR_FLAGS],
1427 q->flags, extack);
1428 if (err < 0)
1429 return err;
1430
1431 q->flags = err;
1432
1433 err = taprio_parse_mqprio_opt(dev, mqprio, extack, q->flags);
1434 if (err < 0)
1435 return err;
1436
1437 new_admin = kzalloc(sizeof(*new_admin), GFP_KERNEL);
1438 if (!new_admin) {
1439 NL_SET_ERR_MSG(extack, "Not enough memory for a new schedule");
1440 return -ENOMEM;
1441 }
1442 INIT_LIST_HEAD(&new_admin->entries);
1443
1444 rcu_read_lock();
1445 oper = rcu_dereference(q->oper_sched);
1446 admin = rcu_dereference(q->admin_sched);
1447 rcu_read_unlock();
1448
1449 /* no changes - no new mqprio settings */
1450 if (!taprio_mqprio_cmp(dev, mqprio))
1451 mqprio = NULL;
1452
1453 if (mqprio && (oper || admin)) {
1454 NL_SET_ERR_MSG(extack, "Changing the traffic mapping of a running schedule is not supported");
1455 err = -ENOTSUPP;
1456 goto free_sched;
1457 }
1458
1459 err = parse_taprio_schedule(tb, new_admin, extack);
1460 if (err < 0)
1461 goto free_sched;
1462
1463 if (new_admin->num_entries == 0) {
1464 NL_SET_ERR_MSG(extack, "There should be at least one entry in the schedule");
1465 err = -EINVAL;
1466 goto free_sched;
1467 }
1468
1469 err = taprio_parse_clockid(sch, tb, extack);
1470 if (err < 0)
1471 goto free_sched;
1472
1473 taprio_set_picos_per_byte(dev, q);
1474
1475 if (mqprio) {
1476 netdev_set_num_tc(dev, mqprio->num_tc);
1477 for (i = 0; i < mqprio->num_tc; i++)
1478 netdev_set_tc_queue(dev, i,
1479 mqprio->count[i],
1480 mqprio->offset[i]);
1481
1482 /* Always use supplied priority mappings */
1483 for (i = 0; i <= TC_BITMASK; i++)
1484 netdev_set_prio_tc_map(dev, i,
1485 mqprio->prio_tc_map[i]);
1486 }
1487
1488 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1489 err = taprio_enable_offload(dev, mqprio, q, new_admin, extack);
1490 else
1491 err = taprio_disable_offload(dev, q, extack);
1492 if (err)
1493 goto free_sched;
1494
1495 /* Protects against enqueue()/dequeue() */
1496 spin_lock_bh(qdisc_lock(sch));
1497
1498 if (tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]) {
1499 if (!TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1500 NL_SET_ERR_MSG_MOD(extack, "txtime-delay can only be set when txtime-assist mode is enabled");
1501 err = -EINVAL;
1502 goto unlock;
1503 }
1504
1505 q->txtime_delay = nla_get_u32(tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]);
1506 }
1507
1508 if (!TXTIME_ASSIST_IS_ENABLED(q->flags) &&
1509 !FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1510 !hrtimer_active(&q->advance_timer)) {
1511 hrtimer_init(&q->advance_timer, q->clockid, HRTIMER_MODE_ABS);
1512 q->advance_timer.function = advance_sched;
1513 }
1514
1515 if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1516 q->dequeue = taprio_dequeue_offload;
1517 q->peek = taprio_peek_offload;
1518 } else {
1519 /* Be sure to always keep the function pointers
1520 * in a consistent state.
1521 */
1522 q->dequeue = taprio_dequeue_soft;
1523 q->peek = taprio_peek_soft;
1524 }
1525
1526 err = taprio_get_start_time(sch, new_admin, &start);
1527 if (err < 0) {
1528 NL_SET_ERR_MSG(extack, "Internal error: failed get start time");
1529 goto unlock;
1530 }
1531
1532 setup_txtime(q, new_admin, start);
1533
1534 if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1535 if (!oper) {
1536 rcu_assign_pointer(q->oper_sched, new_admin);
1537 err = 0;
1538 new_admin = NULL;
1539 goto unlock;
1540 }
1541
1542 rcu_assign_pointer(q->admin_sched, new_admin);
1543 if (admin)
1544 call_rcu(&admin->rcu, taprio_free_sched_cb);
1545 } else {
1546 setup_first_close_time(q, new_admin, start);
1547
1548 /* Protects against advance_sched() */
1549 spin_lock_irqsave(&q->current_entry_lock, flags);
1550
1551 taprio_start_sched(sch, start, new_admin);
1552
1553 rcu_assign_pointer(q->admin_sched, new_admin);
1554 if (admin)
1555 call_rcu(&admin->rcu, taprio_free_sched_cb);
1556
1557 spin_unlock_irqrestore(&q->current_entry_lock, flags);
1558
1559 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1560 taprio_offload_config_changed(q);
1561 }
1562
1563 new_admin = NULL;
1564 err = 0;
1565
1566 unlock:
1567 spin_unlock_bh(qdisc_lock(sch));
1568
1569 free_sched:
1570 if (new_admin)
1571 call_rcu(&new_admin->rcu, taprio_free_sched_cb);
1572
1573 return err;
1574 }
1575
1576 static void taprio_destroy(struct Qdisc *sch)
1577 {
1578 struct taprio_sched *q = qdisc_priv(sch);
1579 struct net_device *dev = qdisc_dev(sch);
1580 unsigned int i;
1581
1582 spin_lock(&taprio_list_lock);
1583 list_del(&q->taprio_list);
1584 spin_unlock(&taprio_list_lock);
1585
1586 hrtimer_cancel(&q->advance_timer);
1587
1588 taprio_disable_offload(dev, q, NULL);
1589
1590 if (q->qdiscs) {
1591 for (i = 0; i < dev->num_tx_queues && q->qdiscs[i]; i++)
1592 qdisc_put(q->qdiscs[i]);
1593
1594 kfree(q->qdiscs);
1595 }
1596 q->qdiscs = NULL;
1597
1598 netdev_reset_tc(dev);
1599
1600 if (q->oper_sched)
1601 call_rcu(&q->oper_sched->rcu, taprio_free_sched_cb);
1602
1603 if (q->admin_sched)
1604 call_rcu(&q->admin_sched->rcu, taprio_free_sched_cb);
1605 }
1606
1607 static int taprio_init(struct Qdisc *sch, struct nlattr *opt,
1608 struct netlink_ext_ack *extack)
1609 {
1610 struct taprio_sched *q = qdisc_priv(sch);
1611 struct net_device *dev = qdisc_dev(sch);
1612 int i;
1613
1614 spin_lock_init(&q->current_entry_lock);
1615
1616 hrtimer_init(&q->advance_timer, CLOCK_TAI, HRTIMER_MODE_ABS);
1617 q->advance_timer.function = advance_sched;
1618
1619 q->dequeue = taprio_dequeue_soft;
1620 q->peek = taprio_peek_soft;
1621
1622 q->root = sch;
1623
1624 /* We only support static clockids. Use an invalid value as default
1625 * and get the valid one on taprio_change().
1626 */
1627 q->clockid = -1;
1628 q->flags = TAPRIO_FLAGS_INVALID;
1629
1630 spin_lock(&taprio_list_lock);
1631 list_add(&q->taprio_list, &taprio_list);
1632 spin_unlock(&taprio_list_lock);
1633
1634 if (sch->parent != TC_H_ROOT)
1635 return -EOPNOTSUPP;
1636
1637 if (!netif_is_multiqueue(dev))
1638 return -EOPNOTSUPP;
1639
1640 /* pre-allocate qdisc, attachment can't fail */
1641 q->qdiscs = kcalloc(dev->num_tx_queues,
1642 sizeof(q->qdiscs[0]),
1643 GFP_KERNEL);
1644
1645 if (!q->qdiscs)
1646 return -ENOMEM;
1647
1648 if (!opt)
1649 return -EINVAL;
1650
1651 for (i = 0; i < dev->num_tx_queues; i++) {
1652 struct netdev_queue *dev_queue;
1653 struct Qdisc *qdisc;
1654
1655 dev_queue = netdev_get_tx_queue(dev, i);
1656 qdisc = qdisc_create_dflt(dev_queue,
1657 &pfifo_qdisc_ops,
1658 TC_H_MAKE(TC_H_MAJ(sch->handle),
1659 TC_H_MIN(i + 1)),
1660 extack);
1661 if (!qdisc)
1662 return -ENOMEM;
1663
1664 if (i < dev->real_num_tx_queues)
1665 qdisc_hash_add(qdisc, false);
1666
1667 q->qdiscs[i] = qdisc;
1668 }
1669
1670 return taprio_change(sch, opt, extack);
1671 }
1672
1673 static struct netdev_queue *taprio_queue_get(struct Qdisc *sch,
1674 unsigned long cl)
1675 {
1676 struct net_device *dev = qdisc_dev(sch);
1677 unsigned long ntx = cl - 1;
1678
1679 if (ntx >= dev->num_tx_queues)
1680 return NULL;
1681
1682 return netdev_get_tx_queue(dev, ntx);
1683 }
1684
1685 static int taprio_graft(struct Qdisc *sch, unsigned long cl,
1686 struct Qdisc *new, struct Qdisc **old,
1687 struct netlink_ext_ack *extack)
1688 {
1689 struct taprio_sched *q = qdisc_priv(sch);
1690 struct net_device *dev = qdisc_dev(sch);
1691 struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1692
1693 if (!dev_queue)
1694 return -EINVAL;
1695
1696 if (dev->flags & IFF_UP)
1697 dev_deactivate(dev);
1698
1699 *old = q->qdiscs[cl - 1];
1700 q->qdiscs[cl - 1] = new;
1701
1702 if (new)
1703 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1704
1705 if (dev->flags & IFF_UP)
1706 dev_activate(dev);
1707
1708 return 0;
1709 }
1710
1711 static int dump_entry(struct sk_buff *msg,
1712 const struct sched_entry *entry)
1713 {
1714 struct nlattr *item;
1715
1716 item = nla_nest_start_noflag(msg, TCA_TAPRIO_SCHED_ENTRY);
1717 if (!item)
1718 return -ENOSPC;
1719
1720 if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INDEX, entry->index))
1721 goto nla_put_failure;
1722
1723 if (nla_put_u8(msg, TCA_TAPRIO_SCHED_ENTRY_CMD, entry->command))
1724 goto nla_put_failure;
1725
1726 if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_GATE_MASK,
1727 entry->gate_mask))
1728 goto nla_put_failure;
1729
1730 if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INTERVAL,
1731 entry->interval))
1732 goto nla_put_failure;
1733
1734 return nla_nest_end(msg, item);
1735
1736 nla_put_failure:
1737 nla_nest_cancel(msg, item);
1738 return -1;
1739 }
1740
1741 static int dump_schedule(struct sk_buff *msg,
1742 const struct sched_gate_list *root)
1743 {
1744 struct nlattr *entry_list;
1745 struct sched_entry *entry;
1746
1747 if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_BASE_TIME,
1748 root->base_time, TCA_TAPRIO_PAD))
1749 return -1;
1750
1751 if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME,
1752 root->cycle_time, TCA_TAPRIO_PAD))
1753 return -1;
1754
1755 if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION,
1756 root->cycle_time_extension, TCA_TAPRIO_PAD))
1757 return -1;
1758
1759 entry_list = nla_nest_start_noflag(msg,
1760 TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST);
1761 if (!entry_list)
1762 goto error_nest;
1763
1764 list_for_each_entry(entry, &root->entries, list) {
1765 if (dump_entry(msg, entry) < 0)
1766 goto error_nest;
1767 }
1768
1769 nla_nest_end(msg, entry_list);
1770 return 0;
1771
1772 error_nest:
1773 nla_nest_cancel(msg, entry_list);
1774 return -1;
1775 }
1776
1777 static int taprio_dump(struct Qdisc *sch, struct sk_buff *skb)
1778 {
1779 struct taprio_sched *q = qdisc_priv(sch);
1780 struct net_device *dev = qdisc_dev(sch);
1781 struct sched_gate_list *oper, *admin;
1782 struct tc_mqprio_qopt opt = { 0 };
1783 struct nlattr *nest, *sched_nest;
1784 unsigned int i;
1785
1786 rcu_read_lock();
1787 oper = rcu_dereference(q->oper_sched);
1788 admin = rcu_dereference(q->admin_sched);
1789
1790 opt.num_tc = netdev_get_num_tc(dev);
1791 memcpy(opt.prio_tc_map, dev->prio_tc_map, sizeof(opt.prio_tc_map));
1792
1793 for (i = 0; i < netdev_get_num_tc(dev); i++) {
1794 opt.count[i] = dev->tc_to_txq[i].count;
1795 opt.offset[i] = dev->tc_to_txq[i].offset;
1796 }
1797
1798 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1799 if (!nest)
1800 goto start_error;
1801
1802 if (nla_put(skb, TCA_TAPRIO_ATTR_PRIOMAP, sizeof(opt), &opt))
1803 goto options_error;
1804
1805 if (!FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1806 nla_put_s32(skb, TCA_TAPRIO_ATTR_SCHED_CLOCKID, q->clockid))
1807 goto options_error;
1808
1809 if (q->flags && nla_put_u32(skb, TCA_TAPRIO_ATTR_FLAGS, q->flags))
1810 goto options_error;
1811
1812 if (q->txtime_delay &&
1813 nla_put_u32(skb, TCA_TAPRIO_ATTR_TXTIME_DELAY, q->txtime_delay))
1814 goto options_error;
1815
1816 if (oper && dump_schedule(skb, oper))
1817 goto options_error;
1818
1819 if (!admin)
1820 goto done;
1821
1822 sched_nest = nla_nest_start_noflag(skb, TCA_TAPRIO_ATTR_ADMIN_SCHED);
1823 if (!sched_nest)
1824 goto options_error;
1825
1826 if (dump_schedule(skb, admin))
1827 goto admin_error;
1828
1829 nla_nest_end(skb, sched_nest);
1830
1831 done:
1832 rcu_read_unlock();
1833
1834 return nla_nest_end(skb, nest);
1835
1836 admin_error:
1837 nla_nest_cancel(skb, sched_nest);
1838
1839 options_error:
1840 nla_nest_cancel(skb, nest);
1841
1842 start_error:
1843 rcu_read_unlock();
1844 return -ENOSPC;
1845 }
1846
1847 static struct Qdisc *taprio_leaf(struct Qdisc *sch, unsigned long cl)
1848 {
1849 struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1850
1851 if (!dev_queue)
1852 return NULL;
1853
1854 return dev_queue->qdisc_sleeping;
1855 }
1856
1857 static unsigned long taprio_find(struct Qdisc *sch, u32 classid)
1858 {
1859 unsigned int ntx = TC_H_MIN(classid);
1860
1861 if (!taprio_queue_get(sch, ntx))
1862 return 0;
1863 return ntx;
1864 }
1865
1866 static int taprio_dump_class(struct Qdisc *sch, unsigned long cl,
1867 struct sk_buff *skb, struct tcmsg *tcm)
1868 {
1869 struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1870
1871 tcm->tcm_parent = TC_H_ROOT;
1872 tcm->tcm_handle |= TC_H_MIN(cl);
1873 tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
1874
1875 return 0;
1876 }
1877
1878 static int taprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
1879 struct gnet_dump *d)
1880 __releases(d->lock)
1881 __acquires(d->lock)
1882 {
1883 struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1884
1885 sch = dev_queue->qdisc_sleeping;
1886 if (gnet_stats_copy_basic(&sch->running, d, NULL, &sch->bstats) < 0 ||
1887 qdisc_qstats_copy(d, sch) < 0)
1888 return -1;
1889 return 0;
1890 }
1891
1892 static void taprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1893 {
1894 struct net_device *dev = qdisc_dev(sch);
1895 unsigned long ntx;
1896
1897 if (arg->stop)
1898 return;
1899
1900 arg->count = arg->skip;
1901 for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
1902 if (arg->fn(sch, ntx + 1, arg) < 0) {
1903 arg->stop = 1;
1904 break;
1905 }
1906 arg->count++;
1907 }
1908 }
1909
1910 static struct netdev_queue *taprio_select_queue(struct Qdisc *sch,
1911 struct tcmsg *tcm)
1912 {
1913 return taprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
1914 }
1915
1916 static const struct Qdisc_class_ops taprio_class_ops = {
1917 .graft = taprio_graft,
1918 .leaf = taprio_leaf,
1919 .find = taprio_find,
1920 .walk = taprio_walk,
1921 .dump = taprio_dump_class,
1922 .dump_stats = taprio_dump_class_stats,
1923 .select_queue = taprio_select_queue,
1924 };
1925
1926 static struct Qdisc_ops taprio_qdisc_ops __read_mostly = {
1927 .cl_ops = &taprio_class_ops,
1928 .id = "taprio",
1929 .priv_size = sizeof(struct taprio_sched),
1930 .init = taprio_init,
1931 .change = taprio_change,
1932 .destroy = taprio_destroy,
1933 .peek = taprio_peek,
1934 .dequeue = taprio_dequeue,
1935 .enqueue = taprio_enqueue,
1936 .dump = taprio_dump,
1937 .owner = THIS_MODULE,
1938 };
1939
1940 static struct notifier_block taprio_device_notifier = {
1941 .notifier_call = taprio_dev_notifier,
1942 };
1943
1944 static int __init taprio_module_init(void)
1945 {
1946 int err = register_netdevice_notifier(&taprio_device_notifier);
1947
1948 if (err)
1949 return err;
1950
1951 return register_qdisc(&taprio_qdisc_ops);
1952 }
1953
1954 static void __exit taprio_module_exit(void)
1955 {
1956 unregister_qdisc(&taprio_qdisc_ops);
1957 unregister_netdevice_notifier(&taprio_device_notifier);
1958 }
1959
1960 module_init(taprio_module_init);
1961 module_exit(taprio_module_exit);
1962 MODULE_LICENSE("GPL");
Linux with added FPC-III support