1 // SPDX-License-Identifier: GPL-2.0
3 /* net/sched/sch_taprio.c Time Aware Priority Scheduler
5 * Authors: Vinicius Costa Gomes <vinicius.gomes@intel.com>
9 #include <linux/ethtool.h>
10 #include <linux/ethtool_netlink.h>
11 #include <linux/types.h>
12 #include <linux/slab.h>
13 #include <linux/kernel.h>
14 #include <linux/string.h>
15 #include <linux/list.h>
16 #include <linux/errno.h>
17 #include <linux/skbuff.h>
18 #include <linux/math64.h>
19 #include <linux/module.h>
20 #include <linux/spinlock.h>
21 #include <linux/rcupdate.h>
22 #include <linux/time.h>
24 #include <net/netlink.h>
25 #include <net/pkt_sched.h>
26 #include <net/pkt_cls.h>
27 #include <net/sch_generic.h>
31 #define TAPRIO_STAT_NOT_SET (~0ULL)
33 #include "sch_mqprio_lib.h"
35 static LIST_HEAD(taprio_list);
36 static struct static_key_false taprio_have_broken_mqprio;
37 static struct static_key_false taprio_have_working_mqprio;
39 #define TAPRIO_ALL_GATES_OPEN -1
41 #define TXTIME_ASSIST_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST)
42 #define FULL_OFFLOAD_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD)
43 #define TAPRIO_FLAGS_INVALID U32_MAX
46 /* Durations between this GCL entry and the GCL entry where the
47 * respective traffic class gate closes
49 u64 gate_duration[TC_MAX_QUEUE];
50 atomic_t budget[TC_MAX_QUEUE];
51 /* The qdisc makes some effort so that no packet leaves
54 ktime_t gate_close_time[TC_MAX_QUEUE];
55 struct list_head list;
56 /* Used to calculate when to advance the schedule */
65 struct sched_gate_list {
66 /* Longest non-zero contiguous gate durations per traffic class,
67 * or 0 if a traffic class gate never opens during the schedule.
69 u64 max_open_gate_duration[TC_MAX_QUEUE];
70 u32 max_frm_len[TC_MAX_QUEUE]; /* for the fast path */
71 u32 max_sdu[TC_MAX_QUEUE]; /* for dump */
73 struct list_head entries;
75 ktime_t cycle_end_time;
77 s64 cycle_time_extension;
82 struct Qdisc **qdiscs;
85 enum tk_offsets tk_offset;
90 atomic64_t picos_per_byte; /* Using picoseconds because for 10Gbps+
91 * speeds it's sub-nanoseconds per byte
94 /* Protects the update side of the RCU protected current_entry */
95 spinlock_t current_entry_lock;
96 struct sched_entry __rcu *current_entry;
97 struct sched_gate_list __rcu *oper_sched;
98 struct sched_gate_list __rcu *admin_sched;
99 struct hrtimer advance_timer;
100 struct list_head taprio_list;
101 int cur_txq[TC_MAX_QUEUE];
102 u32 max_sdu[TC_MAX_QUEUE]; /* save info from the user */
103 u32 fp[TC_QOPT_MAX_QUEUE]; /* only for dump and offloading */
107 struct __tc_taprio_qopt_offload {
109 struct tc_taprio_qopt_offload offload;
112 static void taprio_calculate_gate_durations(struct taprio_sched *q,
113 struct sched_gate_list *sched)
115 struct net_device *dev = qdisc_dev(q->root);
116 int num_tc = netdev_get_num_tc(dev);
117 struct sched_entry *entry, *cur;
120 list_for_each_entry(entry, &sched->entries, list) {
121 u32 gates_still_open = entry->gate_mask;
123 /* For each traffic class, calculate each open gate duration,
124 * starting at this schedule entry and ending at the schedule
125 * entry containing a gate close event for that TC.
130 if (!gates_still_open)
133 for (tc = 0; tc < num_tc; tc++) {
134 if (!(gates_still_open & BIT(tc)))
137 if (cur->gate_mask & BIT(tc))
138 entry->gate_duration[tc] += cur->interval;
140 gates_still_open &= ~BIT(tc);
143 cur = list_next_entry_circular(cur, &sched->entries, list);
144 } while (cur != entry);
146 /* Keep track of the maximum gate duration for each traffic
147 * class, taking care to not confuse a traffic class which is
148 * temporarily closed with one that is always closed.
150 for (tc = 0; tc < num_tc; tc++)
151 if (entry->gate_duration[tc] &&
152 sched->max_open_gate_duration[tc] < entry->gate_duration[tc])
153 sched->max_open_gate_duration[tc] = entry->gate_duration[tc];
157 static bool taprio_entry_allows_tx(ktime_t skb_end_time,
158 struct sched_entry *entry, int tc)
160 return ktime_before(skb_end_time, entry->gate_close_time[tc]);
163 static ktime_t sched_base_time(const struct sched_gate_list *sched)
168 return ns_to_ktime(sched->base_time);
171 static ktime_t taprio_mono_to_any(const struct taprio_sched *q, ktime_t mono)
173 /* This pairs with WRITE_ONCE() in taprio_parse_clockid() */
174 enum tk_offsets tk_offset = READ_ONCE(q->tk_offset);
180 return ktime_mono_to_any(mono, tk_offset);
184 static ktime_t taprio_get_time(const struct taprio_sched *q)
186 return taprio_mono_to_any(q, ktime_get());
189 static void taprio_free_sched_cb(struct rcu_head *head)
191 struct sched_gate_list *sched = container_of(head, struct sched_gate_list, rcu);
192 struct sched_entry *entry, *n;
194 list_for_each_entry_safe(entry, n, &sched->entries, list) {
195 list_del(&entry->list);
202 static void switch_schedules(struct taprio_sched *q,
203 struct sched_gate_list **admin,
204 struct sched_gate_list **oper)
206 rcu_assign_pointer(q->oper_sched, *admin);
207 rcu_assign_pointer(q->admin_sched, NULL);
210 call_rcu(&(*oper)->rcu, taprio_free_sched_cb);
216 /* Get how much time has been already elapsed in the current cycle. */
217 static s32 get_cycle_time_elapsed(struct sched_gate_list *sched, ktime_t time)
219 ktime_t time_since_sched_start;
222 time_since_sched_start = ktime_sub(time, sched->base_time);
223 div_s64_rem(time_since_sched_start, sched->cycle_time, &time_elapsed);
228 static ktime_t get_interval_end_time(struct sched_gate_list *sched,
229 struct sched_gate_list *admin,
230 struct sched_entry *entry,
233 s32 cycle_elapsed = get_cycle_time_elapsed(sched, intv_start);
234 ktime_t intv_end, cycle_ext_end, cycle_end;
236 cycle_end = ktime_add_ns(intv_start, sched->cycle_time - cycle_elapsed);
237 intv_end = ktime_add_ns(intv_start, entry->interval);
238 cycle_ext_end = ktime_add(cycle_end, sched->cycle_time_extension);
240 if (ktime_before(intv_end, cycle_end))
242 else if (admin && admin != sched &&
243 ktime_after(admin->base_time, cycle_end) &&
244 ktime_before(admin->base_time, cycle_ext_end))
245 return admin->base_time;
250 static int length_to_duration(struct taprio_sched *q, int len)
252 return div_u64(len * atomic64_read(&q->picos_per_byte), PSEC_PER_NSEC);
255 static int duration_to_length(struct taprio_sched *q, u64 duration)
257 return div_u64(duration * PSEC_PER_NSEC, atomic64_read(&q->picos_per_byte));
260 /* Sets sched->max_sdu[] and sched->max_frm_len[] to the minimum between the
261 * q->max_sdu[] requested by the user and the max_sdu dynamically determined by
262 * the maximum open gate durations at the given link speed.
264 static void taprio_update_queue_max_sdu(struct taprio_sched *q,
265 struct sched_gate_list *sched,
266 struct qdisc_size_table *stab)
268 struct net_device *dev = qdisc_dev(q->root);
269 int num_tc = netdev_get_num_tc(dev);
270 u32 max_sdu_from_user;
275 for (tc = 0; tc < num_tc; tc++) {
276 max_sdu_from_user = q->max_sdu[tc] ?: U32_MAX;
278 /* TC gate never closes => keep the queueMaxSDU
279 * selected by the user
281 if (sched->max_open_gate_duration[tc] == sched->cycle_time) {
282 max_sdu_dynamic = U32_MAX;
286 max_frm_len = duration_to_length(q, sched->max_open_gate_duration[tc]);
287 /* Compensate for L1 overhead from size table,
288 * but don't let the frame size go negative
291 max_frm_len -= stab->szopts.overhead;
292 max_frm_len = max_t(int, max_frm_len,
293 dev->hard_header_len + 1);
295 max_sdu_dynamic = max_frm_len - dev->hard_header_len;
296 if (max_sdu_dynamic > dev->max_mtu)
297 max_sdu_dynamic = U32_MAX;
300 max_sdu = min(max_sdu_dynamic, max_sdu_from_user);
302 if (max_sdu != U32_MAX) {
303 sched->max_frm_len[tc] = max_sdu + dev->hard_header_len;
304 sched->max_sdu[tc] = max_sdu;
306 sched->max_frm_len[tc] = U32_MAX; /* never oversized */
307 sched->max_sdu[tc] = 0;
312 /* Returns the entry corresponding to next available interval. If
313 * validate_interval is set, it only validates whether the timestamp occurs
314 * when the gate corresponding to the skb's traffic class is open.
316 static struct sched_entry *find_entry_to_transmit(struct sk_buff *skb,
318 struct sched_gate_list *sched,
319 struct sched_gate_list *admin,
321 ktime_t *interval_start,
322 ktime_t *interval_end,
323 bool validate_interval)
325 ktime_t curr_intv_start, curr_intv_end, cycle_end, packet_transmit_time;
326 ktime_t earliest_txtime = KTIME_MAX, txtime, cycle, transmit_end_time;
327 struct sched_entry *entry = NULL, *entry_found = NULL;
328 struct taprio_sched *q = qdisc_priv(sch);
329 struct net_device *dev = qdisc_dev(sch);
330 bool entry_available = false;
334 tc = netdev_get_prio_tc_map(dev, skb->priority);
335 packet_transmit_time = length_to_duration(q, qdisc_pkt_len(skb));
343 cycle = sched->cycle_time;
344 cycle_elapsed = get_cycle_time_elapsed(sched, time);
345 curr_intv_end = ktime_sub_ns(time, cycle_elapsed);
346 cycle_end = ktime_add_ns(curr_intv_end, cycle);
348 list_for_each_entry(entry, &sched->entries, list) {
349 curr_intv_start = curr_intv_end;
350 curr_intv_end = get_interval_end_time(sched, admin, entry,
353 if (ktime_after(curr_intv_start, cycle_end))
356 if (!(entry->gate_mask & BIT(tc)) ||
357 packet_transmit_time > entry->interval)
360 txtime = entry->next_txtime;
362 if (ktime_before(txtime, time) || validate_interval) {
363 transmit_end_time = ktime_add_ns(time, packet_transmit_time);
364 if ((ktime_before(curr_intv_start, time) &&
365 ktime_before(transmit_end_time, curr_intv_end)) ||
366 (ktime_after(curr_intv_start, time) && !validate_interval)) {
368 *interval_start = curr_intv_start;
369 *interval_end = curr_intv_end;
371 } else if (!entry_available && !validate_interval) {
372 /* Here, we are just trying to find out the
373 * first available interval in the next cycle.
375 entry_available = true;
377 *interval_start = ktime_add_ns(curr_intv_start, cycle);
378 *interval_end = ktime_add_ns(curr_intv_end, cycle);
380 } else if (ktime_before(txtime, earliest_txtime) &&
382 earliest_txtime = txtime;
384 n = div_s64(ktime_sub(txtime, curr_intv_start), cycle);
385 *interval_start = ktime_add(curr_intv_start, n * cycle);
386 *interval_end = ktime_add(curr_intv_end, n * cycle);
393 static bool is_valid_interval(struct sk_buff *skb, struct Qdisc *sch)
395 struct taprio_sched *q = qdisc_priv(sch);
396 struct sched_gate_list *sched, *admin;
397 ktime_t interval_start, interval_end;
398 struct sched_entry *entry;
401 sched = rcu_dereference(q->oper_sched);
402 admin = rcu_dereference(q->admin_sched);
404 entry = find_entry_to_transmit(skb, sch, sched, admin, skb->tstamp,
405 &interval_start, &interval_end, true);
411 static bool taprio_flags_valid(u32 flags)
413 /* Make sure no other flag bits are set. */
414 if (flags & ~(TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST |
415 TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
417 /* txtime-assist and full offload are mutually exclusive */
418 if ((flags & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST) &&
419 (flags & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
424 /* This returns the tstamp value set by TCP in terms of the set clock. */
425 static ktime_t get_tcp_tstamp(struct taprio_sched *q, struct sk_buff *skb)
427 unsigned int offset = skb_network_offset(skb);
428 const struct ipv6hdr *ipv6h;
429 const struct iphdr *iph;
430 struct ipv6hdr _ipv6h;
432 ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h);
436 if (ipv6h->version == 4) {
437 iph = (struct iphdr *)ipv6h;
438 offset += iph->ihl * 4;
440 /* special-case 6in4 tunnelling, as that is a common way to get
441 * v6 connectivity in the home
443 if (iph->protocol == IPPROTO_IPV6) {
444 ipv6h = skb_header_pointer(skb, offset,
445 sizeof(_ipv6h), &_ipv6h);
447 if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP)
449 } else if (iph->protocol != IPPROTO_TCP) {
452 } else if (ipv6h->version == 6 && ipv6h->nexthdr != IPPROTO_TCP) {
456 return taprio_mono_to_any(q, skb->skb_mstamp_ns);
459 /* There are a few scenarios where we will have to modify the txtime from
460 * what is read from next_txtime in sched_entry. They are:
461 * 1. If txtime is in the past,
462 * a. The gate for the traffic class is currently open and packet can be
463 * transmitted before it closes, schedule the packet right away.
464 * b. If the gate corresponding to the traffic class is going to open later
465 * in the cycle, set the txtime of packet to the interval start.
466 * 2. If txtime is in the future, there are packets corresponding to the
467 * current traffic class waiting to be transmitted. So, the following
468 * possibilities exist:
469 * a. We can transmit the packet before the window containing the txtime
471 * b. The window might close before the transmission can be completed
472 * successfully. So, schedule the packet in the next open window.
474 static long get_packet_txtime(struct sk_buff *skb, struct Qdisc *sch)
476 ktime_t transmit_end_time, interval_end, interval_start, tcp_tstamp;
477 struct taprio_sched *q = qdisc_priv(sch);
478 struct sched_gate_list *sched, *admin;
479 ktime_t minimum_time, now, txtime;
480 int len, packet_transmit_time;
481 struct sched_entry *entry;
484 now = taprio_get_time(q);
485 minimum_time = ktime_add_ns(now, q->txtime_delay);
487 tcp_tstamp = get_tcp_tstamp(q, skb);
488 minimum_time = max_t(ktime_t, minimum_time, tcp_tstamp);
491 admin = rcu_dereference(q->admin_sched);
492 sched = rcu_dereference(q->oper_sched);
493 if (admin && ktime_after(minimum_time, admin->base_time))
494 switch_schedules(q, &admin, &sched);
496 /* Until the schedule starts, all the queues are open */
497 if (!sched || ktime_before(minimum_time, sched->base_time)) {
498 txtime = minimum_time;
502 len = qdisc_pkt_len(skb);
503 packet_transmit_time = length_to_duration(q, len);
506 sched_changed = false;
508 entry = find_entry_to_transmit(skb, sch, sched, admin,
510 &interval_start, &interval_end,
517 txtime = entry->next_txtime;
518 txtime = max_t(ktime_t, txtime, minimum_time);
519 txtime = max_t(ktime_t, txtime, interval_start);
521 if (admin && admin != sched &&
522 ktime_after(txtime, admin->base_time)) {
524 sched_changed = true;
528 transmit_end_time = ktime_add(txtime, packet_transmit_time);
529 minimum_time = transmit_end_time;
531 /* Update the txtime of current entry to the next time it's
534 if (ktime_after(transmit_end_time, interval_end))
535 entry->next_txtime = ktime_add(interval_start, sched->cycle_time);
536 } while (sched_changed || ktime_after(transmit_end_time, interval_end));
538 entry->next_txtime = transmit_end_time;
545 /* Devices with full offload are expected to honor this in hardware */
546 static bool taprio_skb_exceeds_queue_max_sdu(struct Qdisc *sch,
549 struct taprio_sched *q = qdisc_priv(sch);
550 struct net_device *dev = qdisc_dev(sch);
551 struct sched_gate_list *sched;
552 int prio = skb->priority;
553 bool exceeds = false;
556 tc = netdev_get_prio_tc_map(dev, prio);
559 sched = rcu_dereference(q->oper_sched);
560 if (sched && skb->len > sched->max_frm_len[tc])
567 static int taprio_enqueue_one(struct sk_buff *skb, struct Qdisc *sch,
568 struct Qdisc *child, struct sk_buff **to_free)
570 struct taprio_sched *q = qdisc_priv(sch);
572 /* sk_flags are only safe to use on full sockets. */
573 if (skb->sk && sk_fullsock(skb->sk) && sock_flag(skb->sk, SOCK_TXTIME)) {
574 if (!is_valid_interval(skb, sch))
575 return qdisc_drop(skb, sch, to_free);
576 } else if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
577 skb->tstamp = get_packet_txtime(skb, sch);
579 return qdisc_drop(skb, sch, to_free);
582 qdisc_qstats_backlog_inc(sch, skb);
585 return qdisc_enqueue(skb, child, to_free);
588 static int taprio_enqueue_segmented(struct sk_buff *skb, struct Qdisc *sch,
590 struct sk_buff **to_free)
592 unsigned int slen = 0, numsegs = 0, len = qdisc_pkt_len(skb);
593 netdev_features_t features = netif_skb_features(skb);
594 struct sk_buff *segs, *nskb;
597 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
598 if (IS_ERR_OR_NULL(segs))
599 return qdisc_drop(skb, sch, to_free);
601 skb_list_walk_safe(segs, segs, nskb) {
602 skb_mark_not_on_list(segs);
603 qdisc_skb_cb(segs)->pkt_len = segs->len;
606 /* FIXME: we should be segmenting to a smaller size
607 * rather than dropping these
609 if (taprio_skb_exceeds_queue_max_sdu(sch, segs))
610 ret = qdisc_drop(segs, sch, to_free);
612 ret = taprio_enqueue_one(segs, sch, child, to_free);
614 if (ret != NET_XMIT_SUCCESS) {
615 if (net_xmit_drop_count(ret))
616 qdisc_qstats_drop(sch);
623 qdisc_tree_reduce_backlog(sch, 1 - numsegs, len - slen);
626 return numsegs > 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP;
629 /* Will not be called in the full offload case, since the TX queues are
630 * attached to the Qdisc created using qdisc_create_dflt()
632 static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
633 struct sk_buff **to_free)
635 struct taprio_sched *q = qdisc_priv(sch);
639 queue = skb_get_queue_mapping(skb);
641 child = q->qdiscs[queue];
642 if (unlikely(!child))
643 return qdisc_drop(skb, sch, to_free);
645 if (taprio_skb_exceeds_queue_max_sdu(sch, skb)) {
646 /* Large packets might not be transmitted when the transmission
647 * duration exceeds any configured interval. Therefore, segment
648 * the skb into smaller chunks. Drivers with full offload are
649 * expected to handle this in hardware.
652 return taprio_enqueue_segmented(skb, sch, child,
655 return qdisc_drop(skb, sch, to_free);
658 return taprio_enqueue_one(skb, sch, child, to_free);
661 static struct sk_buff *taprio_peek(struct Qdisc *sch)
663 WARN_ONCE(1, "taprio only supports operating as root qdisc, peek() not implemented");
667 static void taprio_set_budgets(struct taprio_sched *q,
668 struct sched_gate_list *sched,
669 struct sched_entry *entry)
671 struct net_device *dev = qdisc_dev(q->root);
672 int num_tc = netdev_get_num_tc(dev);
675 for (tc = 0; tc < num_tc; tc++) {
676 /* Traffic classes which never close have infinite budget */
677 if (entry->gate_duration[tc] == sched->cycle_time)
680 budget = div64_u64((u64)entry->gate_duration[tc] * PSEC_PER_NSEC,
681 atomic64_read(&q->picos_per_byte));
683 atomic_set(&entry->budget[tc], budget);
687 /* When an skb is sent, it consumes from the budget of all traffic classes */
688 static int taprio_update_budgets(struct sched_entry *entry, size_t len,
689 int tc_consumed, int num_tc)
691 int tc, budget, new_budget = 0;
693 for (tc = 0; tc < num_tc; tc++) {
694 budget = atomic_read(&entry->budget[tc]);
695 /* Don't consume from infinite budget */
696 if (budget == INT_MAX) {
697 if (tc == tc_consumed)
702 if (tc == tc_consumed)
703 new_budget = atomic_sub_return(len, &entry->budget[tc]);
705 atomic_sub(len, &entry->budget[tc]);
711 static struct sk_buff *taprio_dequeue_from_txq(struct Qdisc *sch, int txq,
712 struct sched_entry *entry,
715 struct taprio_sched *q = qdisc_priv(sch);
716 struct net_device *dev = qdisc_dev(sch);
717 struct Qdisc *child = q->qdiscs[txq];
718 int num_tc = netdev_get_num_tc(dev);
725 if (unlikely(!child))
728 if (TXTIME_ASSIST_IS_ENABLED(q->flags))
729 goto skip_peek_checks;
731 skb = child->ops->peek(child);
735 prio = skb->priority;
736 tc = netdev_get_prio_tc_map(dev, prio);
738 if (!(gate_mask & BIT(tc)))
741 len = qdisc_pkt_len(skb);
742 guard = ktime_add_ns(taprio_get_time(q), length_to_duration(q, len));
744 /* In the case that there's no gate entry, there's no
747 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
748 !taprio_entry_allows_tx(guard, entry, tc))
751 /* ... and no budget. */
752 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
753 taprio_update_budgets(entry, len, tc, num_tc) < 0)
757 skb = child->ops->dequeue(child);
761 qdisc_bstats_update(sch, skb);
762 qdisc_qstats_backlog_dec(sch, skb);
768 static void taprio_next_tc_txq(struct net_device *dev, int tc, int *txq)
770 int offset = dev->tc_to_txq[tc].offset;
771 int count = dev->tc_to_txq[tc].count;
774 if (*txq == offset + count)
778 /* Prioritize higher traffic classes, and select among TXQs belonging to the
779 * same TC using round robin
781 static struct sk_buff *taprio_dequeue_tc_priority(struct Qdisc *sch,
782 struct sched_entry *entry,
785 struct taprio_sched *q = qdisc_priv(sch);
786 struct net_device *dev = qdisc_dev(sch);
787 int num_tc = netdev_get_num_tc(dev);
791 for (tc = num_tc - 1; tc >= 0; tc--) {
792 int first_txq = q->cur_txq[tc];
794 if (!(gate_mask & BIT(tc)))
798 skb = taprio_dequeue_from_txq(sch, q->cur_txq[tc],
801 taprio_next_tc_txq(dev, tc, &q->cur_txq[tc]);
803 if (q->cur_txq[tc] >= dev->num_tx_queues)
804 q->cur_txq[tc] = first_txq;
808 } while (q->cur_txq[tc] != first_txq);
814 /* Broken way of prioritizing smaller TXQ indices and ignoring the traffic
815 * class other than to determine whether the gate is open or not
817 static struct sk_buff *taprio_dequeue_txq_priority(struct Qdisc *sch,
818 struct sched_entry *entry,
821 struct net_device *dev = qdisc_dev(sch);
825 for (i = 0; i < dev->num_tx_queues; i++) {
826 skb = taprio_dequeue_from_txq(sch, i, entry, gate_mask);
834 /* Will not be called in the full offload case, since the TX queues are
835 * attached to the Qdisc created using qdisc_create_dflt()
837 static struct sk_buff *taprio_dequeue(struct Qdisc *sch)
839 struct taprio_sched *q = qdisc_priv(sch);
840 struct sk_buff *skb = NULL;
841 struct sched_entry *entry;
845 entry = rcu_dereference(q->current_entry);
846 /* if there's no entry, it means that the schedule didn't
847 * start yet, so force all gates to be open, this is in
848 * accordance to IEEE 802.1Qbv-2015 Section 8.6.9.4.5
851 gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
855 if (static_branch_unlikely(&taprio_have_broken_mqprio) &&
856 !static_branch_likely(&taprio_have_working_mqprio)) {
857 /* Single NIC kind which is broken */
858 skb = taprio_dequeue_txq_priority(sch, entry, gate_mask);
859 } else if (static_branch_likely(&taprio_have_working_mqprio) &&
860 !static_branch_unlikely(&taprio_have_broken_mqprio)) {
861 /* Single NIC kind which prioritizes properly */
862 skb = taprio_dequeue_tc_priority(sch, entry, gate_mask);
864 /* Mixed NIC kinds present in system, need dynamic testing */
865 if (q->broken_mqprio)
866 skb = taprio_dequeue_txq_priority(sch, entry, gate_mask);
868 skb = taprio_dequeue_tc_priority(sch, entry, gate_mask);
877 static bool should_restart_cycle(const struct sched_gate_list *oper,
878 const struct sched_entry *entry)
880 if (list_is_last(&entry->list, &oper->entries))
883 if (ktime_compare(entry->end_time, oper->cycle_end_time) == 0)
889 static bool should_change_schedules(const struct sched_gate_list *admin,
890 const struct sched_gate_list *oper,
893 ktime_t next_base_time, extension_time;
898 next_base_time = sched_base_time(admin);
900 /* This is the simple case, the end_time would fall after
901 * the next schedule base_time.
903 if (ktime_compare(next_base_time, end_time) <= 0)
906 /* This is the cycle_time_extension case, if the end_time
907 * plus the amount that can be extended would fall after the
908 * next schedule base_time, we can extend the current schedule
911 extension_time = ktime_add_ns(end_time, oper->cycle_time_extension);
913 /* FIXME: the IEEE 802.1Q-2018 Specification isn't clear about
914 * how precisely the extension should be made. So after
915 * conformance testing, this logic may change.
917 if (ktime_compare(next_base_time, extension_time) <= 0)
923 static enum hrtimer_restart advance_sched(struct hrtimer *timer)
925 struct taprio_sched *q = container_of(timer, struct taprio_sched,
927 struct net_device *dev = qdisc_dev(q->root);
928 struct sched_gate_list *oper, *admin;
929 int num_tc = netdev_get_num_tc(dev);
930 struct sched_entry *entry, *next;
931 struct Qdisc *sch = q->root;
935 spin_lock(&q->current_entry_lock);
936 entry = rcu_dereference_protected(q->current_entry,
937 lockdep_is_held(&q->current_entry_lock));
938 oper = rcu_dereference_protected(q->oper_sched,
939 lockdep_is_held(&q->current_entry_lock));
940 admin = rcu_dereference_protected(q->admin_sched,
941 lockdep_is_held(&q->current_entry_lock));
944 switch_schedules(q, &admin, &oper);
946 /* This can happen in two cases: 1. this is the very first run
947 * of this function (i.e. we weren't running any schedule
948 * previously); 2. The previous schedule just ended. The first
949 * entry of all schedules are pre-calculated during the
950 * schedule initialization.
952 if (unlikely(!entry || entry->end_time == oper->base_time)) {
953 next = list_first_entry(&oper->entries, struct sched_entry,
955 end_time = next->end_time;
959 if (should_restart_cycle(oper, entry)) {
960 next = list_first_entry(&oper->entries, struct sched_entry,
962 oper->cycle_end_time = ktime_add_ns(oper->cycle_end_time,
965 next = list_next_entry(entry, list);
968 end_time = ktime_add_ns(entry->end_time, next->interval);
969 end_time = min_t(ktime_t, end_time, oper->cycle_end_time);
971 for (tc = 0; tc < num_tc; tc++) {
972 if (next->gate_duration[tc] == oper->cycle_time)
973 next->gate_close_time[tc] = KTIME_MAX;
975 next->gate_close_time[tc] = ktime_add_ns(entry->end_time,
976 next->gate_duration[tc]);
979 if (should_change_schedules(admin, oper, end_time)) {
980 /* Set things so the next time this runs, the new
983 end_time = sched_base_time(admin);
984 switch_schedules(q, &admin, &oper);
987 next->end_time = end_time;
988 taprio_set_budgets(q, oper, next);
991 rcu_assign_pointer(q->current_entry, next);
992 spin_unlock(&q->current_entry_lock);
994 hrtimer_set_expires(&q->advance_timer, end_time);
997 __netif_schedule(sch);
1000 return HRTIMER_RESTART;
1003 static const struct nla_policy entry_policy[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = {
1004 [TCA_TAPRIO_SCHED_ENTRY_INDEX] = { .type = NLA_U32 },
1005 [TCA_TAPRIO_SCHED_ENTRY_CMD] = { .type = NLA_U8 },
1006 [TCA_TAPRIO_SCHED_ENTRY_GATE_MASK] = { .type = NLA_U32 },
1007 [TCA_TAPRIO_SCHED_ENTRY_INTERVAL] = { .type = NLA_U32 },
1010 static const struct nla_policy taprio_tc_policy[TCA_TAPRIO_TC_ENTRY_MAX + 1] = {
1011 [TCA_TAPRIO_TC_ENTRY_INDEX] = NLA_POLICY_MAX(NLA_U32,
1013 [TCA_TAPRIO_TC_ENTRY_MAX_SDU] = { .type = NLA_U32 },
1014 [TCA_TAPRIO_TC_ENTRY_FP] = NLA_POLICY_RANGE(NLA_U32,
1019 static const struct netlink_range_validation_signed taprio_cycle_time_range = {
1024 static const struct nla_policy taprio_policy[TCA_TAPRIO_ATTR_MAX + 1] = {
1025 [TCA_TAPRIO_ATTR_PRIOMAP] = {
1026 .len = sizeof(struct tc_mqprio_qopt)
1028 [TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST] = { .type = NLA_NESTED },
1029 [TCA_TAPRIO_ATTR_SCHED_BASE_TIME] = { .type = NLA_S64 },
1030 [TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY] = { .type = NLA_NESTED },
1031 [TCA_TAPRIO_ATTR_SCHED_CLOCKID] = { .type = NLA_S32 },
1032 [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME] =
1033 NLA_POLICY_FULL_RANGE_SIGNED(NLA_S64, &taprio_cycle_time_range),
1034 [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION] = { .type = NLA_S64 },
1035 [TCA_TAPRIO_ATTR_FLAGS] = { .type = NLA_U32 },
1036 [TCA_TAPRIO_ATTR_TXTIME_DELAY] = { .type = NLA_U32 },
1037 [TCA_TAPRIO_ATTR_TC_ENTRY] = { .type = NLA_NESTED },
1040 static int fill_sched_entry(struct taprio_sched *q, struct nlattr **tb,
1041 struct sched_entry *entry,
1042 struct netlink_ext_ack *extack)
1044 int min_duration = length_to_duration(q, ETH_ZLEN);
1047 if (tb[TCA_TAPRIO_SCHED_ENTRY_CMD])
1048 entry->command = nla_get_u8(
1049 tb[TCA_TAPRIO_SCHED_ENTRY_CMD]);
1051 if (tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK])
1052 entry->gate_mask = nla_get_u32(
1053 tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK]);
1055 if (tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL])
1056 interval = nla_get_u32(
1057 tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL]);
1059 /* The interval should allow at least the minimum ethernet
1062 if (interval < min_duration) {
1063 NL_SET_ERR_MSG(extack, "Invalid interval for schedule entry");
1067 entry->interval = interval;
1072 static int parse_sched_entry(struct taprio_sched *q, struct nlattr *n,
1073 struct sched_entry *entry, int index,
1074 struct netlink_ext_ack *extack)
1076 struct nlattr *tb[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = { };
1079 err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_SCHED_ENTRY_MAX, n,
1080 entry_policy, NULL);
1082 NL_SET_ERR_MSG(extack, "Could not parse nested entry");
1086 entry->index = index;
1088 return fill_sched_entry(q, tb, entry, extack);
1091 static int parse_sched_list(struct taprio_sched *q, struct nlattr *list,
1092 struct sched_gate_list *sched,
1093 struct netlink_ext_ack *extack)
1102 nla_for_each_nested(n, list, rem) {
1103 struct sched_entry *entry;
1105 if (nla_type(n) != TCA_TAPRIO_SCHED_ENTRY) {
1106 NL_SET_ERR_MSG(extack, "Attribute is not of type 'entry'");
1110 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1112 NL_SET_ERR_MSG(extack, "Not enough memory for entry");
1116 err = parse_sched_entry(q, n, entry, i, extack);
1122 list_add_tail(&entry->list, &sched->entries);
1126 sched->num_entries = i;
1131 static int parse_taprio_schedule(struct taprio_sched *q, struct nlattr **tb,
1132 struct sched_gate_list *new,
1133 struct netlink_ext_ack *extack)
1137 if (tb[TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]) {
1138 NL_SET_ERR_MSG(extack, "Adding a single entry is not supported");
1142 if (tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME])
1143 new->base_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME]);
1145 if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION])
1146 new->cycle_time_extension = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION]);
1148 if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME])
1149 new->cycle_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]);
1151 if (tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST])
1152 err = parse_sched_list(q, tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST],
1157 if (!new->cycle_time) {
1158 struct sched_entry *entry;
1161 list_for_each_entry(entry, &new->entries, list)
1162 cycle = ktime_add_ns(cycle, entry->interval);
1165 NL_SET_ERR_MSG(extack, "'cycle_time' can never be 0");
1169 if (cycle < 0 || cycle > INT_MAX) {
1170 NL_SET_ERR_MSG(extack, "'cycle_time' is too big");
1174 new->cycle_time = cycle;
1177 taprio_calculate_gate_durations(q, new);
1182 static int taprio_parse_mqprio_opt(struct net_device *dev,
1183 struct tc_mqprio_qopt *qopt,
1184 struct netlink_ext_ack *extack,
1187 bool allow_overlapping_txqs = TXTIME_ASSIST_IS_ENABLED(taprio_flags);
1189 if (!qopt && !dev->num_tc) {
1190 NL_SET_ERR_MSG(extack, "'mqprio' configuration is necessary");
1194 /* If num_tc is already set, it means that the user already
1195 * configured the mqprio part
1200 /* taprio imposes that traffic classes map 1:n to tx queues */
1201 if (qopt->num_tc > dev->num_tx_queues) {
1202 NL_SET_ERR_MSG(extack, "Number of traffic classes is greater than number of HW queues");
1206 /* For some reason, in txtime-assist mode, we allow TXQ ranges for
1207 * different TCs to overlap, and just validate the TXQ ranges.
1209 return mqprio_validate_qopt(dev, qopt, true, allow_overlapping_txqs,
1213 static int taprio_get_start_time(struct Qdisc *sch,
1214 struct sched_gate_list *sched,
1217 struct taprio_sched *q = qdisc_priv(sch);
1218 ktime_t now, base, cycle;
1221 base = sched_base_time(sched);
1222 now = taprio_get_time(q);
1224 if (ktime_after(base, now)) {
1229 cycle = sched->cycle_time;
1231 /* The qdisc is expected to have at least one sched_entry. Moreover,
1232 * any entry must have 'interval' > 0. Thus if the cycle time is zero,
1233 * something went really wrong. In that case, we should warn about this
1234 * inconsistent state and return error.
1236 if (WARN_ON(!cycle))
1239 /* Schedule the start time for the beginning of the next
1242 n = div64_s64(ktime_sub_ns(now, base), cycle);
1243 *start = ktime_add_ns(base, (n + 1) * cycle);
1247 static void setup_first_end_time(struct taprio_sched *q,
1248 struct sched_gate_list *sched, ktime_t base)
1250 struct net_device *dev = qdisc_dev(q->root);
1251 int num_tc = netdev_get_num_tc(dev);
1252 struct sched_entry *first;
1256 first = list_first_entry(&sched->entries,
1257 struct sched_entry, list);
1259 cycle = sched->cycle_time;
1261 /* FIXME: find a better place to do this */
1262 sched->cycle_end_time = ktime_add_ns(base, cycle);
1264 first->end_time = ktime_add_ns(base, first->interval);
1265 taprio_set_budgets(q, sched, first);
1267 for (tc = 0; tc < num_tc; tc++) {
1268 if (first->gate_duration[tc] == sched->cycle_time)
1269 first->gate_close_time[tc] = KTIME_MAX;
1271 first->gate_close_time[tc] = ktime_add_ns(base, first->gate_duration[tc]);
1274 rcu_assign_pointer(q->current_entry, NULL);
1277 static void taprio_start_sched(struct Qdisc *sch,
1278 ktime_t start, struct sched_gate_list *new)
1280 struct taprio_sched *q = qdisc_priv(sch);
1283 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1286 expires = hrtimer_get_expires(&q->advance_timer);
1288 expires = KTIME_MAX;
1290 /* If the new schedule starts before the next expiration, we
1291 * reprogram it to the earliest one, so we change the admin
1292 * schedule to the operational one at the right time.
1294 start = min_t(ktime_t, start, expires);
1296 hrtimer_start(&q->advance_timer, start, HRTIMER_MODE_ABS);
1299 static void taprio_set_picos_per_byte(struct net_device *dev,
1300 struct taprio_sched *q)
1302 struct ethtool_link_ksettings ecmd;
1303 int speed = SPEED_10;
1307 err = __ethtool_get_link_ksettings(dev, &ecmd);
1311 if (ecmd.base.speed && ecmd.base.speed != SPEED_UNKNOWN)
1312 speed = ecmd.base.speed;
1315 picos_per_byte = (USEC_PER_SEC * 8) / speed;
1317 atomic64_set(&q->picos_per_byte, picos_per_byte);
1318 netdev_dbg(dev, "taprio: set %s's picos_per_byte to: %lld, linkspeed: %d\n",
1319 dev->name, (long long)atomic64_read(&q->picos_per_byte),
1323 static int taprio_dev_notifier(struct notifier_block *nb, unsigned long event,
1326 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1327 struct sched_gate_list *oper, *admin;
1328 struct qdisc_size_table *stab;
1329 struct taprio_sched *q;
1333 if (event != NETDEV_UP && event != NETDEV_CHANGE)
1336 list_for_each_entry(q, &taprio_list, taprio_list) {
1337 if (dev != qdisc_dev(q->root))
1340 taprio_set_picos_per_byte(dev, q);
1342 stab = rtnl_dereference(q->root->stab);
1344 oper = rtnl_dereference(q->oper_sched);
1346 taprio_update_queue_max_sdu(q, oper, stab);
1348 admin = rtnl_dereference(q->admin_sched);
1350 taprio_update_queue_max_sdu(q, admin, stab);
1358 static void setup_txtime(struct taprio_sched *q,
1359 struct sched_gate_list *sched, ktime_t base)
1361 struct sched_entry *entry;
1364 list_for_each_entry(entry, &sched->entries, list) {
1365 entry->next_txtime = ktime_add_ns(base, interval);
1366 interval += entry->interval;
1370 static struct tc_taprio_qopt_offload *taprio_offload_alloc(int num_entries)
1372 struct __tc_taprio_qopt_offload *__offload;
1374 __offload = kzalloc(struct_size(__offload, offload.entries, num_entries),
1379 refcount_set(&__offload->users, 1);
1381 return &__offload->offload;
1384 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
1387 struct __tc_taprio_qopt_offload *__offload;
1389 __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1392 refcount_inc(&__offload->users);
1396 EXPORT_SYMBOL_GPL(taprio_offload_get);
1398 void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
1400 struct __tc_taprio_qopt_offload *__offload;
1402 __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1405 if (!refcount_dec_and_test(&__offload->users))
1410 EXPORT_SYMBOL_GPL(taprio_offload_free);
1412 /* The function will only serve to keep the pointers to the "oper" and "admin"
1413 * schedules valid in relation to their base times, so when calling dump() the
1414 * users looks at the right schedules.
1415 * When using full offload, the admin configuration is promoted to oper at the
1416 * base_time in the PHC time domain. But because the system time is not
1417 * necessarily in sync with that, we can't just trigger a hrtimer to call
1418 * switch_schedules at the right hardware time.
1419 * At the moment we call this by hand right away from taprio, but in the future
1420 * it will be useful to create a mechanism for drivers to notify taprio of the
1421 * offload state (PENDING, ACTIVE, INACTIVE) so it can be visible in dump().
1422 * This is left as TODO.
1424 static void taprio_offload_config_changed(struct taprio_sched *q)
1426 struct sched_gate_list *oper, *admin;
1428 oper = rtnl_dereference(q->oper_sched);
1429 admin = rtnl_dereference(q->admin_sched);
1431 switch_schedules(q, &admin, &oper);
1434 static u32 tc_map_to_queue_mask(struct net_device *dev, u32 tc_mask)
1436 u32 i, queue_mask = 0;
1438 for (i = 0; i < dev->num_tc; i++) {
1441 if (!(tc_mask & BIT(i)))
1444 offset = dev->tc_to_txq[i].offset;
1445 count = dev->tc_to_txq[i].count;
1447 queue_mask |= GENMASK(offset + count - 1, offset);
1453 static void taprio_sched_to_offload(struct net_device *dev,
1454 struct sched_gate_list *sched,
1455 struct tc_taprio_qopt_offload *offload,
1456 const struct tc_taprio_caps *caps)
1458 struct sched_entry *entry;
1461 offload->base_time = sched->base_time;
1462 offload->cycle_time = sched->cycle_time;
1463 offload->cycle_time_extension = sched->cycle_time_extension;
1465 list_for_each_entry(entry, &sched->entries, list) {
1466 struct tc_taprio_sched_entry *e = &offload->entries[i];
1468 e->command = entry->command;
1469 e->interval = entry->interval;
1470 if (caps->gate_mask_per_txq)
1471 e->gate_mask = tc_map_to_queue_mask(dev,
1474 e->gate_mask = entry->gate_mask;
1479 offload->num_entries = i;
1482 static void taprio_detect_broken_mqprio(struct taprio_sched *q)
1484 struct net_device *dev = qdisc_dev(q->root);
1485 struct tc_taprio_caps caps;
1487 qdisc_offload_query_caps(dev, TC_SETUP_QDISC_TAPRIO,
1488 &caps, sizeof(caps));
1490 q->broken_mqprio = caps.broken_mqprio;
1491 if (q->broken_mqprio)
1492 static_branch_inc(&taprio_have_broken_mqprio);
1494 static_branch_inc(&taprio_have_working_mqprio);
1496 q->detected_mqprio = true;
1499 static void taprio_cleanup_broken_mqprio(struct taprio_sched *q)
1501 if (!q->detected_mqprio)
1504 if (q->broken_mqprio)
1505 static_branch_dec(&taprio_have_broken_mqprio);
1507 static_branch_dec(&taprio_have_working_mqprio);
1510 static int taprio_enable_offload(struct net_device *dev,
1511 struct taprio_sched *q,
1512 struct sched_gate_list *sched,
1513 struct netlink_ext_ack *extack)
1515 const struct net_device_ops *ops = dev->netdev_ops;
1516 struct tc_taprio_qopt_offload *offload;
1517 struct tc_taprio_caps caps;
1520 if (!ops->ndo_setup_tc) {
1521 NL_SET_ERR_MSG(extack,
1522 "Device does not support taprio offload");
1526 qdisc_offload_query_caps(dev, TC_SETUP_QDISC_TAPRIO,
1527 &caps, sizeof(caps));
1529 if (!caps.supports_queue_max_sdu) {
1530 for (tc = 0; tc < TC_MAX_QUEUE; tc++) {
1531 if (q->max_sdu[tc]) {
1532 NL_SET_ERR_MSG_MOD(extack,
1533 "Device does not handle queueMaxSDU");
1539 offload = taprio_offload_alloc(sched->num_entries);
1541 NL_SET_ERR_MSG(extack,
1542 "Not enough memory for enabling offload mode");
1545 offload->cmd = TAPRIO_CMD_REPLACE;
1546 offload->extack = extack;
1547 mqprio_qopt_reconstruct(dev, &offload->mqprio.qopt);
1548 offload->mqprio.extack = extack;
1549 taprio_sched_to_offload(dev, sched, offload, &caps);
1550 mqprio_fp_to_offload(q->fp, &offload->mqprio);
1552 for (tc = 0; tc < TC_MAX_QUEUE; tc++)
1553 offload->max_sdu[tc] = q->max_sdu[tc];
1555 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1557 NL_SET_ERR_MSG_WEAK(extack,
1558 "Device failed to setup taprio offload");
1562 q->offloaded = true;
1565 /* The offload structure may linger around via a reference taken by the
1566 * device driver, so clear up the netlink extack pointer so that the
1567 * driver isn't tempted to dereference data which stopped being valid
1569 offload->extack = NULL;
1570 offload->mqprio.extack = NULL;
1571 taprio_offload_free(offload);
1576 static int taprio_disable_offload(struct net_device *dev,
1577 struct taprio_sched *q,
1578 struct netlink_ext_ack *extack)
1580 const struct net_device_ops *ops = dev->netdev_ops;
1581 struct tc_taprio_qopt_offload *offload;
1587 offload = taprio_offload_alloc(0);
1589 NL_SET_ERR_MSG(extack,
1590 "Not enough memory to disable offload mode");
1593 offload->cmd = TAPRIO_CMD_DESTROY;
1595 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1597 NL_SET_ERR_MSG(extack,
1598 "Device failed to disable offload");
1602 q->offloaded = false;
1605 taprio_offload_free(offload);
1610 /* If full offload is enabled, the only possible clockid is the net device's
1611 * PHC. For that reason, specifying a clockid through netlink is incorrect.
1612 * For txtime-assist, it is implicitly assumed that the device's PHC is kept
1613 * in sync with the specified clockid via a user space daemon such as phc2sys.
1614 * For both software taprio and txtime-assist, the clockid is used for the
1615 * hrtimer that advances the schedule and hence mandatory.
1617 static int taprio_parse_clockid(struct Qdisc *sch, struct nlattr **tb,
1618 struct netlink_ext_ack *extack)
1620 struct taprio_sched *q = qdisc_priv(sch);
1621 struct net_device *dev = qdisc_dev(sch);
1624 if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1625 const struct ethtool_ops *ops = dev->ethtool_ops;
1626 struct ethtool_ts_info info = {
1627 .cmd = ETHTOOL_GET_TS_INFO,
1631 if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1632 NL_SET_ERR_MSG(extack,
1633 "The 'clockid' cannot be specified for full offload");
1637 if (ops && ops->get_ts_info)
1638 err = ops->get_ts_info(dev, &info);
1640 if (err || info.phc_index < 0) {
1641 NL_SET_ERR_MSG(extack,
1642 "Device does not have a PTP clock");
1646 } else if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1647 int clockid = nla_get_s32(tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]);
1648 enum tk_offsets tk_offset;
1650 /* We only support static clockids and we don't allow
1651 * for it to be modified after the first init.
1654 (q->clockid != -1 && q->clockid != clockid)) {
1655 NL_SET_ERR_MSG(extack,
1656 "Changing the 'clockid' of a running schedule is not supported");
1662 case CLOCK_REALTIME:
1663 tk_offset = TK_OFFS_REAL;
1665 case CLOCK_MONOTONIC:
1666 tk_offset = TK_OFFS_MAX;
1668 case CLOCK_BOOTTIME:
1669 tk_offset = TK_OFFS_BOOT;
1672 tk_offset = TK_OFFS_TAI;
1675 NL_SET_ERR_MSG(extack, "Invalid 'clockid'");
1679 /* This pairs with READ_ONCE() in taprio_mono_to_any */
1680 WRITE_ONCE(q->tk_offset, tk_offset);
1682 q->clockid = clockid;
1684 NL_SET_ERR_MSG(extack, "Specifying a 'clockid' is mandatory");
1688 /* Everything went ok, return success. */
1695 static int taprio_parse_tc_entry(struct Qdisc *sch,
1697 u32 max_sdu[TC_QOPT_MAX_QUEUE],
1698 u32 fp[TC_QOPT_MAX_QUEUE],
1699 unsigned long *seen_tcs,
1700 struct netlink_ext_ack *extack)
1702 struct nlattr *tb[TCA_TAPRIO_TC_ENTRY_MAX + 1] = { };
1703 struct net_device *dev = qdisc_dev(sch);
1707 err = nla_parse_nested(tb, TCA_TAPRIO_TC_ENTRY_MAX, opt,
1708 taprio_tc_policy, extack);
1712 if (!tb[TCA_TAPRIO_TC_ENTRY_INDEX]) {
1713 NL_SET_ERR_MSG_MOD(extack, "TC entry index missing");
1717 tc = nla_get_u32(tb[TCA_TAPRIO_TC_ENTRY_INDEX]);
1718 if (tc >= TC_QOPT_MAX_QUEUE) {
1719 NL_SET_ERR_MSG_MOD(extack, "TC entry index out of range");
1723 if (*seen_tcs & BIT(tc)) {
1724 NL_SET_ERR_MSG_MOD(extack, "Duplicate TC entry");
1728 *seen_tcs |= BIT(tc);
1730 if (tb[TCA_TAPRIO_TC_ENTRY_MAX_SDU]) {
1731 val = nla_get_u32(tb[TCA_TAPRIO_TC_ENTRY_MAX_SDU]);
1732 if (val > dev->max_mtu) {
1733 NL_SET_ERR_MSG_MOD(extack, "TC max SDU exceeds device max MTU");
1740 if (tb[TCA_TAPRIO_TC_ENTRY_FP])
1741 fp[tc] = nla_get_u32(tb[TCA_TAPRIO_TC_ENTRY_FP]);
1746 static int taprio_parse_tc_entries(struct Qdisc *sch,
1748 struct netlink_ext_ack *extack)
1750 struct taprio_sched *q = qdisc_priv(sch);
1751 struct net_device *dev = qdisc_dev(sch);
1752 u32 max_sdu[TC_QOPT_MAX_QUEUE];
1753 bool have_preemption = false;
1754 unsigned long seen_tcs = 0;
1755 u32 fp[TC_QOPT_MAX_QUEUE];
1760 for (tc = 0; tc < TC_QOPT_MAX_QUEUE; tc++) {
1761 max_sdu[tc] = q->max_sdu[tc];
1765 nla_for_each_nested(n, opt, rem) {
1766 if (nla_type(n) != TCA_TAPRIO_ATTR_TC_ENTRY)
1769 err = taprio_parse_tc_entry(sch, n, max_sdu, fp, &seen_tcs,
1775 for (tc = 0; tc < TC_QOPT_MAX_QUEUE; tc++) {
1776 q->max_sdu[tc] = max_sdu[tc];
1778 if (fp[tc] != TC_FP_EXPRESS)
1779 have_preemption = true;
1782 if (have_preemption) {
1783 if (!FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1784 NL_SET_ERR_MSG(extack,
1785 "Preemption only supported with full offload");
1789 if (!ethtool_dev_mm_supported(dev)) {
1790 NL_SET_ERR_MSG(extack,
1791 "Device does not support preemption");
1799 static int taprio_mqprio_cmp(const struct net_device *dev,
1800 const struct tc_mqprio_qopt *mqprio)
1804 if (!mqprio || mqprio->num_tc != dev->num_tc)
1807 for (i = 0; i < mqprio->num_tc; i++)
1808 if (dev->tc_to_txq[i].count != mqprio->count[i] ||
1809 dev->tc_to_txq[i].offset != mqprio->offset[i])
1812 for (i = 0; i <= TC_BITMASK; i++)
1813 if (dev->prio_tc_map[i] != mqprio->prio_tc_map[i])
1819 /* The semantics of the 'flags' argument in relation to 'change()'
1820 * requests, are interpreted following two rules (which are applied in
1821 * this order): (1) an omitted 'flags' argument is interpreted as
1822 * zero; (2) the 'flags' of a "running" taprio instance cannot be
1825 static int taprio_new_flags(const struct nlattr *attr, u32 old,
1826 struct netlink_ext_ack *extack)
1831 new = nla_get_u32(attr);
1833 if (old != TAPRIO_FLAGS_INVALID && old != new) {
1834 NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
1838 if (!taprio_flags_valid(new)) {
1839 NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
1846 static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
1847 struct netlink_ext_ack *extack)
1849 struct qdisc_size_table *stab = rtnl_dereference(sch->stab);
1850 struct nlattr *tb[TCA_TAPRIO_ATTR_MAX + 1] = { };
1851 struct sched_gate_list *oper, *admin, *new_admin;
1852 struct taprio_sched *q = qdisc_priv(sch);
1853 struct net_device *dev = qdisc_dev(sch);
1854 struct tc_mqprio_qopt *mqprio = NULL;
1855 unsigned long flags;
1859 err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_ATTR_MAX, opt,
1860 taprio_policy, extack);
1864 if (tb[TCA_TAPRIO_ATTR_PRIOMAP])
1865 mqprio = nla_data(tb[TCA_TAPRIO_ATTR_PRIOMAP]);
1867 err = taprio_new_flags(tb[TCA_TAPRIO_ATTR_FLAGS],
1874 err = taprio_parse_mqprio_opt(dev, mqprio, extack, q->flags);
1878 err = taprio_parse_tc_entries(sch, opt, extack);
1882 new_admin = kzalloc(sizeof(*new_admin), GFP_KERNEL);
1884 NL_SET_ERR_MSG(extack, "Not enough memory for a new schedule");
1887 INIT_LIST_HEAD(&new_admin->entries);
1889 oper = rtnl_dereference(q->oper_sched);
1890 admin = rtnl_dereference(q->admin_sched);
1892 /* no changes - no new mqprio settings */
1893 if (!taprio_mqprio_cmp(dev, mqprio))
1896 if (mqprio && (oper || admin)) {
1897 NL_SET_ERR_MSG(extack, "Changing the traffic mapping of a running schedule is not supported");
1903 err = netdev_set_num_tc(dev, mqprio->num_tc);
1906 for (i = 0; i < mqprio->num_tc; i++) {
1907 netdev_set_tc_queue(dev, i,
1910 q->cur_txq[i] = mqprio->offset[i];
1913 /* Always use supplied priority mappings */
1914 for (i = 0; i <= TC_BITMASK; i++)
1915 netdev_set_prio_tc_map(dev, i,
1916 mqprio->prio_tc_map[i]);
1919 err = parse_taprio_schedule(q, tb, new_admin, extack);
1923 if (new_admin->num_entries == 0) {
1924 NL_SET_ERR_MSG(extack, "There should be at least one entry in the schedule");
1929 err = taprio_parse_clockid(sch, tb, extack);
1933 taprio_set_picos_per_byte(dev, q);
1934 taprio_update_queue_max_sdu(q, new_admin, stab);
1936 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1937 err = taprio_enable_offload(dev, q, new_admin, extack);
1939 err = taprio_disable_offload(dev, q, extack);
1943 /* Protects against enqueue()/dequeue() */
1944 spin_lock_bh(qdisc_lock(sch));
1946 if (tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]) {
1947 if (!TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1948 NL_SET_ERR_MSG_MOD(extack, "txtime-delay can only be set when txtime-assist mode is enabled");
1953 q->txtime_delay = nla_get_u32(tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]);
1956 if (!TXTIME_ASSIST_IS_ENABLED(q->flags) &&
1957 !FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1958 !hrtimer_active(&q->advance_timer)) {
1959 hrtimer_init(&q->advance_timer, q->clockid, HRTIMER_MODE_ABS);
1960 q->advance_timer.function = advance_sched;
1963 err = taprio_get_start_time(sch, new_admin, &start);
1965 NL_SET_ERR_MSG(extack, "Internal error: failed get start time");
1969 setup_txtime(q, new_admin, start);
1971 if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1973 rcu_assign_pointer(q->oper_sched, new_admin);
1979 rcu_assign_pointer(q->admin_sched, new_admin);
1981 call_rcu(&admin->rcu, taprio_free_sched_cb);
1983 setup_first_end_time(q, new_admin, start);
1985 /* Protects against advance_sched() */
1986 spin_lock_irqsave(&q->current_entry_lock, flags);
1988 taprio_start_sched(sch, start, new_admin);
1990 rcu_assign_pointer(q->admin_sched, new_admin);
1992 call_rcu(&admin->rcu, taprio_free_sched_cb);
1994 spin_unlock_irqrestore(&q->current_entry_lock, flags);
1996 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1997 taprio_offload_config_changed(q);
2004 NL_SET_ERR_MSG_MOD(extack,
2005 "Size table not specified, frame length estimations may be inaccurate");
2008 spin_unlock_bh(qdisc_lock(sch));
2012 call_rcu(&new_admin->rcu, taprio_free_sched_cb);
2017 static void taprio_reset(struct Qdisc *sch)
2019 struct taprio_sched *q = qdisc_priv(sch);
2020 struct net_device *dev = qdisc_dev(sch);
2023 hrtimer_cancel(&q->advance_timer);
2026 for (i = 0; i < dev->num_tx_queues; i++)
2028 qdisc_reset(q->qdiscs[i]);
2032 static void taprio_destroy(struct Qdisc *sch)
2034 struct taprio_sched *q = qdisc_priv(sch);
2035 struct net_device *dev = qdisc_dev(sch);
2036 struct sched_gate_list *oper, *admin;
2039 list_del(&q->taprio_list);
2041 /* Note that taprio_reset() might not be called if an error
2042 * happens in qdisc_create(), after taprio_init() has been called.
2044 hrtimer_cancel(&q->advance_timer);
2045 qdisc_synchronize(sch);
2047 taprio_disable_offload(dev, q, NULL);
2050 for (i = 0; i < dev->num_tx_queues; i++)
2051 qdisc_put(q->qdiscs[i]);
2057 netdev_reset_tc(dev);
2059 oper = rtnl_dereference(q->oper_sched);
2060 admin = rtnl_dereference(q->admin_sched);
2063 call_rcu(&oper->rcu, taprio_free_sched_cb);
2066 call_rcu(&admin->rcu, taprio_free_sched_cb);
2068 taprio_cleanup_broken_mqprio(q);
2071 static int taprio_init(struct Qdisc *sch, struct nlattr *opt,
2072 struct netlink_ext_ack *extack)
2074 struct taprio_sched *q = qdisc_priv(sch);
2075 struct net_device *dev = qdisc_dev(sch);
2078 spin_lock_init(&q->current_entry_lock);
2080 hrtimer_init(&q->advance_timer, CLOCK_TAI, HRTIMER_MODE_ABS);
2081 q->advance_timer.function = advance_sched;
2085 /* We only support static clockids. Use an invalid value as default
2086 * and get the valid one on taprio_change().
2089 q->flags = TAPRIO_FLAGS_INVALID;
2091 list_add(&q->taprio_list, &taprio_list);
2093 if (sch->parent != TC_H_ROOT) {
2094 NL_SET_ERR_MSG_MOD(extack, "Can only be attached as root qdisc");
2098 if (!netif_is_multiqueue(dev)) {
2099 NL_SET_ERR_MSG_MOD(extack, "Multi-queue device is required");
2103 q->qdiscs = kcalloc(dev->num_tx_queues, sizeof(q->qdiscs[0]),
2111 for (i = 0; i < dev->num_tx_queues; i++) {
2112 struct netdev_queue *dev_queue;
2113 struct Qdisc *qdisc;
2115 dev_queue = netdev_get_tx_queue(dev, i);
2116 qdisc = qdisc_create_dflt(dev_queue,
2118 TC_H_MAKE(TC_H_MAJ(sch->handle),
2124 if (i < dev->real_num_tx_queues)
2125 qdisc_hash_add(qdisc, false);
2127 q->qdiscs[i] = qdisc;
2130 for (tc = 0; tc < TC_QOPT_MAX_QUEUE; tc++)
2131 q->fp[tc] = TC_FP_EXPRESS;
2133 taprio_detect_broken_mqprio(q);
2135 return taprio_change(sch, opt, extack);
2138 static void taprio_attach(struct Qdisc *sch)
2140 struct taprio_sched *q = qdisc_priv(sch);
2141 struct net_device *dev = qdisc_dev(sch);
2144 /* Attach underlying qdisc */
2145 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
2146 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, ntx);
2147 struct Qdisc *old, *dev_queue_qdisc;
2149 if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
2150 struct Qdisc *qdisc = q->qdiscs[ntx];
2152 /* In offload mode, the root taprio qdisc is bypassed
2153 * and the netdev TX queues see the children directly
2155 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
2156 dev_queue_qdisc = qdisc;
2158 /* In software mode, attach the root taprio qdisc
2159 * to all netdev TX queues, so that dev_qdisc_enqueue()
2160 * goes through taprio_enqueue().
2162 dev_queue_qdisc = sch;
2164 old = dev_graft_qdisc(dev_queue, dev_queue_qdisc);
2165 /* The qdisc's refcount requires to be elevated once
2166 * for each netdev TX queue it is grafted onto
2168 qdisc_refcount_inc(dev_queue_qdisc);
2174 static struct netdev_queue *taprio_queue_get(struct Qdisc *sch,
2177 struct net_device *dev = qdisc_dev(sch);
2178 unsigned long ntx = cl - 1;
2180 if (ntx >= dev->num_tx_queues)
2183 return netdev_get_tx_queue(dev, ntx);
2186 static int taprio_graft(struct Qdisc *sch, unsigned long cl,
2187 struct Qdisc *new, struct Qdisc **old,
2188 struct netlink_ext_ack *extack)
2190 struct taprio_sched *q = qdisc_priv(sch);
2191 struct net_device *dev = qdisc_dev(sch);
2192 struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
2197 if (dev->flags & IFF_UP)
2198 dev_deactivate(dev);
2200 /* In offload mode, the child Qdisc is directly attached to the netdev
2201 * TX queue, and thus, we need to keep its refcount elevated in order
2202 * to counteract qdisc_graft()'s call to qdisc_put() once per TX queue.
2203 * However, save the reference to the new qdisc in the private array in
2204 * both software and offload cases, to have an up-to-date reference to
2207 *old = q->qdiscs[cl - 1];
2208 if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
2209 WARN_ON_ONCE(dev_graft_qdisc(dev_queue, new) != *old);
2211 qdisc_refcount_inc(new);
2216 q->qdiscs[cl - 1] = new;
2218 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
2220 if (dev->flags & IFF_UP)
2226 static int dump_entry(struct sk_buff *msg,
2227 const struct sched_entry *entry)
2229 struct nlattr *item;
2231 item = nla_nest_start_noflag(msg, TCA_TAPRIO_SCHED_ENTRY);
2235 if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INDEX, entry->index))
2236 goto nla_put_failure;
2238 if (nla_put_u8(msg, TCA_TAPRIO_SCHED_ENTRY_CMD, entry->command))
2239 goto nla_put_failure;
2241 if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_GATE_MASK,
2243 goto nla_put_failure;
2245 if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INTERVAL,
2247 goto nla_put_failure;
2249 return nla_nest_end(msg, item);
2252 nla_nest_cancel(msg, item);
2256 static int dump_schedule(struct sk_buff *msg,
2257 const struct sched_gate_list *root)
2259 struct nlattr *entry_list;
2260 struct sched_entry *entry;
2262 if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_BASE_TIME,
2263 root->base_time, TCA_TAPRIO_PAD))
2266 if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME,
2267 root->cycle_time, TCA_TAPRIO_PAD))
2270 if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION,
2271 root->cycle_time_extension, TCA_TAPRIO_PAD))
2274 entry_list = nla_nest_start_noflag(msg,
2275 TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST);
2279 list_for_each_entry(entry, &root->entries, list) {
2280 if (dump_entry(msg, entry) < 0)
2284 nla_nest_end(msg, entry_list);
2288 nla_nest_cancel(msg, entry_list);
2292 static int taprio_dump_tc_entries(struct sk_buff *skb,
2293 struct taprio_sched *q,
2294 struct sched_gate_list *sched)
2299 for (tc = 0; tc < TC_MAX_QUEUE; tc++) {
2300 n = nla_nest_start(skb, TCA_TAPRIO_ATTR_TC_ENTRY);
2304 if (nla_put_u32(skb, TCA_TAPRIO_TC_ENTRY_INDEX, tc))
2305 goto nla_put_failure;
2307 if (nla_put_u32(skb, TCA_TAPRIO_TC_ENTRY_MAX_SDU,
2308 sched->max_sdu[tc]))
2309 goto nla_put_failure;
2311 if (nla_put_u32(skb, TCA_TAPRIO_TC_ENTRY_FP, q->fp[tc]))
2312 goto nla_put_failure;
2314 nla_nest_end(skb, n);
2320 nla_nest_cancel(skb, n);
2324 static int taprio_put_stat(struct sk_buff *skb, u64 val, u16 attrtype)
2326 if (val == TAPRIO_STAT_NOT_SET)
2328 if (nla_put_u64_64bit(skb, attrtype, val, TCA_TAPRIO_OFFLOAD_STATS_PAD))
2333 static int taprio_dump_xstats(struct Qdisc *sch, struct gnet_dump *d,
2334 struct tc_taprio_qopt_offload *offload,
2335 struct tc_taprio_qopt_stats *stats)
2337 struct net_device *dev = qdisc_dev(sch);
2338 const struct net_device_ops *ops;
2339 struct sk_buff *skb = d->skb;
2340 struct nlattr *xstats;
2343 ops = qdisc_dev(sch)->netdev_ops;
2345 /* FIXME I could use qdisc_offload_dump_helper(), but that messes
2346 * with sch->flags depending on whether the device reports taprio
2347 * stats, and I'm not sure whether that's a good idea, considering
2348 * that stats are optional to the offload itself
2350 if (!ops->ndo_setup_tc)
2353 memset(stats, 0xff, sizeof(*stats));
2355 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
2356 if (err == -EOPNOTSUPP)
2361 xstats = nla_nest_start(skb, TCA_STATS_APP);
2365 if (taprio_put_stat(skb, stats->window_drops,
2366 TCA_TAPRIO_OFFLOAD_STATS_WINDOW_DROPS) ||
2367 taprio_put_stat(skb, stats->tx_overruns,
2368 TCA_TAPRIO_OFFLOAD_STATS_TX_OVERRUNS))
2371 nla_nest_end(skb, xstats);
2376 nla_nest_cancel(skb, xstats);
2381 static int taprio_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
2383 struct tc_taprio_qopt_offload offload = {
2384 .cmd = TAPRIO_CMD_STATS,
2387 return taprio_dump_xstats(sch, d, &offload, &offload.stats);
2390 static int taprio_dump(struct Qdisc *sch, struct sk_buff *skb)
2392 struct taprio_sched *q = qdisc_priv(sch);
2393 struct net_device *dev = qdisc_dev(sch);
2394 struct sched_gate_list *oper, *admin;
2395 struct tc_mqprio_qopt opt = { 0 };
2396 struct nlattr *nest, *sched_nest;
2398 oper = rtnl_dereference(q->oper_sched);
2399 admin = rtnl_dereference(q->admin_sched);
2401 mqprio_qopt_reconstruct(dev, &opt);
2403 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
2407 if (nla_put(skb, TCA_TAPRIO_ATTR_PRIOMAP, sizeof(opt), &opt))
2410 if (!FULL_OFFLOAD_IS_ENABLED(q->flags) &&
2411 nla_put_s32(skb, TCA_TAPRIO_ATTR_SCHED_CLOCKID, q->clockid))
2414 if (q->flags && nla_put_u32(skb, TCA_TAPRIO_ATTR_FLAGS, q->flags))
2417 if (q->txtime_delay &&
2418 nla_put_u32(skb, TCA_TAPRIO_ATTR_TXTIME_DELAY, q->txtime_delay))
2421 if (oper && taprio_dump_tc_entries(skb, q, oper))
2424 if (oper && dump_schedule(skb, oper))
2430 sched_nest = nla_nest_start_noflag(skb, TCA_TAPRIO_ATTR_ADMIN_SCHED);
2434 if (dump_schedule(skb, admin))
2437 nla_nest_end(skb, sched_nest);
2440 return nla_nest_end(skb, nest);
2443 nla_nest_cancel(skb, sched_nest);
2446 nla_nest_cancel(skb, nest);
2452 static struct Qdisc *taprio_leaf(struct Qdisc *sch, unsigned long cl)
2454 struct taprio_sched *q = qdisc_priv(sch);
2455 struct net_device *dev = qdisc_dev(sch);
2456 unsigned int ntx = cl - 1;
2458 if (ntx >= dev->num_tx_queues)
2461 return q->qdiscs[ntx];
2464 static unsigned long taprio_find(struct Qdisc *sch, u32 classid)
2466 unsigned int ntx = TC_H_MIN(classid);
2468 if (!taprio_queue_get(sch, ntx))
2473 static int taprio_dump_class(struct Qdisc *sch, unsigned long cl,
2474 struct sk_buff *skb, struct tcmsg *tcm)
2476 struct Qdisc *child = taprio_leaf(sch, cl);
2478 tcm->tcm_parent = TC_H_ROOT;
2479 tcm->tcm_handle |= TC_H_MIN(cl);
2480 tcm->tcm_info = child->handle;
2485 static int taprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
2486 struct gnet_dump *d)
2490 struct Qdisc *child = taprio_leaf(sch, cl);
2491 struct tc_taprio_qopt_offload offload = {
2492 .cmd = TAPRIO_CMD_QUEUE_STATS,
2498 if (gnet_stats_copy_basic(d, NULL, &child->bstats, true) < 0 ||
2499 qdisc_qstats_copy(d, child) < 0)
2502 return taprio_dump_xstats(sch, d, &offload, &offload.queue_stats.stats);
2505 static void taprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
2507 struct net_device *dev = qdisc_dev(sch);
2513 arg->count = arg->skip;
2514 for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
2515 if (!tc_qdisc_stats_dump(sch, ntx + 1, arg))
2520 static struct netdev_queue *taprio_select_queue(struct Qdisc *sch,
2523 return taprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
2526 static const struct Qdisc_class_ops taprio_class_ops = {
2527 .graft = taprio_graft,
2528 .leaf = taprio_leaf,
2529 .find = taprio_find,
2530 .walk = taprio_walk,
2531 .dump = taprio_dump_class,
2532 .dump_stats = taprio_dump_class_stats,
2533 .select_queue = taprio_select_queue,
2536 static struct Qdisc_ops taprio_qdisc_ops __read_mostly = {
2537 .cl_ops = &taprio_class_ops,
2539 .priv_size = sizeof(struct taprio_sched),
2540 .init = taprio_init,
2541 .change = taprio_change,
2542 .destroy = taprio_destroy,
2543 .reset = taprio_reset,
2544 .attach = taprio_attach,
2545 .peek = taprio_peek,
2546 .dequeue = taprio_dequeue,
2547 .enqueue = taprio_enqueue,
2548 .dump = taprio_dump,
2549 .dump_stats = taprio_dump_stats,
2550 .owner = THIS_MODULE,
2553 static struct notifier_block taprio_device_notifier = {
2554 .notifier_call = taprio_dev_notifier,
2557 static int __init taprio_module_init(void)
2559 int err = register_netdevice_notifier(&taprio_device_notifier);
2564 return register_qdisc(&taprio_qdisc_ops);
2567 static void __exit taprio_module_exit(void)
2569 unregister_qdisc(&taprio_qdisc_ops);
2570 unregister_netdevice_notifier(&taprio_device_notifier);
2573 module_init(taprio_module_init);
2574 module_exit(taprio_module_exit);
2575 MODULE_LICENSE("GPL");
2576 MODULE_DESCRIPTION("Time Aware Priority qdisc");