~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/sched/sch_taprio.c

Version: ~ [ linux-5.15-rc6 ] ~ [ linux-5.14.14 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.75 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.155 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.213 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.252 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.287 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.289 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp