net/sched/sch_taprio.c

   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/ethtool.h>
  10 #include <linux/types.h>
  11 #include <linux/slab.h>
  12 #include <linux/kernel.h>
  13 #include <linux/string.h>
  14 #include <linux/list.h>
  15 #include <linux/errno.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/math64.h>
  18 #include <linux/module.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/rcupdate.h>
  21 #include <net/netlink.h>
  22 #include <net/pkt_sched.h>
  23 #include <net/pkt_cls.h>
  24 #include <net/sch_generic.h>
  25 #include <net/sock.h>
  26 #include <net/tcp.h>
  27
  28 static LIST_HEAD(taprio_list);
  29 static DEFINE_SPINLOCK(taprio_list_lock);
  30
  31 #define TAPRIO_ALL_GATES_OPEN -1
  32
  33 #define TXTIME_ASSIST_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST)
  34 #define FULL_OFFLOAD_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD)
  35 #define TAPRIO_FLAGS_INVALID U32_MAX
  36
  37 struct sched_entry {
  38         struct list_head list;
  39
  40         /* The instant that this entry "closes" and the next one
  41          * should open, the qdisc will make some effort so that no
  42          * packet leaves after this time.
  43          */
  44         ktime_t close_time;
  45         ktime_t next_txtime;
  46         atomic_t budget;
  47         int index;
  48         u32 gate_mask;
  49         u32 interval;
  50         u8 command;
  51 };
  52
  53 struct sched_gate_list {
  54         struct rcu_head rcu;
  55         struct list_head entries;
  56         size_t num_entries;
  57         ktime_t cycle_close_time;
  58         s64 cycle_time;
  59         s64 cycle_time_extension;
  60         s64 base_time;
  61 };
  62
  63 struct taprio_sched {
  64         struct Qdisc **qdiscs;
  65         struct Qdisc *root;
  66         u32 flags;
  67         enum tk_offsets tk_offset;
  68         int clockid;
  69         atomic64_t picos_per_byte; /* Using picoseconds because for 10Gbps+
  70                                     * speeds it's sub-nanoseconds per byte
  71                                     */
  72
  73         /* Protects the update side of the RCU protected current_entry */
  74         spinlock_t current_entry_lock;
  75         struct sched_entry __rcu *current_entry;
  76         struct sched_gate_list __rcu *oper_sched;
  77         struct sched_gate_list __rcu *admin_sched;
  78         struct hrtimer advance_timer;
  79         struct list_head taprio_list;
  80         struct sk_buff *(*dequeue)(struct Qdisc *sch);
  81         struct sk_buff *(*peek)(struct Qdisc *sch);
  82         u32 txtime_delay;
  83 };
  84
  85 struct __tc_taprio_qopt_offload {
  86         refcount_t users;
  87         struct tc_taprio_qopt_offload offload;
  88 };
  89
  90 static ktime_t sched_base_time(const struct sched_gate_list *sched)
  91 {
  92         if (!sched)
  93                 return KTIME_MAX;
  94
  95         return ns_to_ktime(sched->base_time);
  96 }
  97
  98 static ktime_t taprio_get_time(struct taprio_sched *q)
  99 {
 100         ktime_t mono = ktime_get();
 101
 102         switch (q->tk_offset) {
 103         case TK_OFFS_MAX:
 104                 return mono;
 105         default:
 106                 return ktime_mono_to_any(mono, q->tk_offset);
 107         }
 108
 109         return KTIME_MAX;
 110 }
 111
 112 static void taprio_free_sched_cb(struct rcu_head *head)
 113 {
 114         struct sched_gate_list *sched = container_of(head, struct sched_gate_list, rcu);
 115         struct sched_entry *entry, *n;
 116
 117         if (!sched)
 118                 return;
 119
 120         list_for_each_entry_safe(entry, n, &sched->entries, list) {
 121                 list_del(&entry->list);
 122                 kfree(entry);
 123         }
 124
 125         kfree(sched);
 126 }
 127
 128 static void switch_schedules(struct taprio_sched *q,
 129                              struct sched_gate_list **admin,
 130                              struct sched_gate_list **oper)
 131 {
 132         rcu_assign_pointer(q->oper_sched, *admin);
 133         rcu_assign_pointer(q->admin_sched, NULL);
 134
 135         if (*oper)
 136                 call_rcu(&(*oper)->rcu, taprio_free_sched_cb);
 137
 138         *oper = *admin;
 139         *admin = NULL;
 140 }
 141
 142 /* Get how much time has been already elapsed in the current cycle. */
 143 static s32 get_cycle_time_elapsed(struct sched_gate_list *sched, ktime_t time)
 144 {
 145         ktime_t time_since_sched_start;
 146         s32 time_elapsed;
 147
 148         time_since_sched_start = ktime_sub(time, sched->base_time);
 149         div_s64_rem(time_since_sched_start, sched->cycle_time, &time_elapsed);
 150
 151         return time_elapsed;
 152 }
 153
 154 static ktime_t get_interval_end_time(struct sched_gate_list *sched,
 155                                      struct sched_gate_list *admin,
 156                                      struct sched_entry *entry,
 157                                      ktime_t intv_start)
 158 {
 159         s32 cycle_elapsed = get_cycle_time_elapsed(sched, intv_start);
 160         ktime_t intv_end, cycle_ext_end, cycle_end;
 161
 162         cycle_end = ktime_add_ns(intv_start, sched->cycle_time - cycle_elapsed);
 163         intv_end = ktime_add_ns(intv_start, entry->interval);
 164         cycle_ext_end = ktime_add(cycle_end, sched->cycle_time_extension);
 165
 166         if (ktime_before(intv_end, cycle_end))
 167                 return intv_end;
 168         else if (admin && admin != sched &&
 169                  ktime_after(admin->base_time, cycle_end) &&
 170                  ktime_before(admin->base_time, cycle_ext_end))
 171                 return admin->base_time;
 172         else
 173                 return cycle_end;
 174 }
 175
 176 static int length_to_duration(struct taprio_sched *q, int len)
 177 {
 178         return div_u64(len * atomic64_read(&q->picos_per_byte), 1000);
 179 }
 180
 181 /* Returns the entry corresponding to next available interval. If
 182  * validate_interval is set, it only validates whether the timestamp occurs
 183  * when the gate corresponding to the skb's traffic class is open.
 184  */
 185 static struct sched_entry *find_entry_to_transmit(struct sk_buff *skb,
 186                                                   struct Qdisc *sch,
 187                                                   struct sched_gate_list *sched,
 188                                                   struct sched_gate_list *admin,
 189                                                   ktime_t time,
 190                                                   ktime_t *interval_start,
 191                                                   ktime_t *interval_end,
 192                                                   bool validate_interval)
 193 {
 194         ktime_t curr_intv_start, curr_intv_end, cycle_end, packet_transmit_time;
 195         ktime_t earliest_txtime = KTIME_MAX, txtime, cycle, transmit_end_time;
 196         struct sched_entry *entry = NULL, *entry_found = NULL;
 197         struct taprio_sched *q = qdisc_priv(sch);
 198         struct net_device *dev = qdisc_dev(sch);
 199         bool entry_available = false;
 200         s32 cycle_elapsed;
 201         int tc, n;
 202
 203         tc = netdev_get_prio_tc_map(dev, skb->priority);
 204         packet_transmit_time = length_to_duration(q, qdisc_pkt_len(skb));
 205
 206         *interval_start = 0;
 207         *interval_end = 0;
 208
 209         if (!sched)
 210                 return NULL;
 211
 212         cycle = sched->cycle_time;
 213         cycle_elapsed = get_cycle_time_elapsed(sched, time);
 214         curr_intv_end = ktime_sub_ns(time, cycle_elapsed);
 215         cycle_end = ktime_add_ns(curr_intv_end, cycle);
 216
 217         list_for_each_entry(entry, &sched->entries, list) {
 218                 curr_intv_start = curr_intv_end;
 219                 curr_intv_end = get_interval_end_time(sched, admin, entry,
 220                                                       curr_intv_start);
 221
 222                 if (ktime_after(curr_intv_start, cycle_end))
 223                         break;
 224
 225                 if (!(entry->gate_mask & BIT(tc)) ||
 226                     packet_transmit_time > entry->interval)
 227                         continue;
 228
 229                 txtime = entry->next_txtime;
 230
 231                 if (ktime_before(txtime, time) || validate_interval) {
 232                         transmit_end_time = ktime_add_ns(time, packet_transmit_time);
 233                         if ((ktime_before(curr_intv_start, time) &&
 234                              ktime_before(transmit_end_time, curr_intv_end)) ||
 235                             (ktime_after(curr_intv_start, time) && !validate_interval)) {
 236                                 entry_found = entry;
 237                                 *interval_start = curr_intv_start;
 238                                 *interval_end = curr_intv_end;
 239                                 break;
 240                         } else if (!entry_available && !validate_interval) {
 241                                 /* Here, we are just trying to find out the
 242                                  * first available interval in the next cycle.
 243                                  */
 244                                 entry_available = 1;
 245                                 entry_found = entry;
 246                                 *interval_start = ktime_add_ns(curr_intv_start, cycle);
 247                                 *interval_end = ktime_add_ns(curr_intv_end, cycle);
 248                         }
 249                 } else if (ktime_before(txtime, earliest_txtime) &&
 250                            !entry_available) {
 251                         earliest_txtime = txtime;
 252                         entry_found = entry;
 253                         n = div_s64(ktime_sub(txtime, curr_intv_start), cycle);
 254                         *interval_start = ktime_add(curr_intv_start, n * cycle);
 255                         *interval_end = ktime_add(curr_intv_end, n * cycle);
 256                 }
 257         }
 258
 259         return entry_found;
 260 }
 261
 262 static bool is_valid_interval(struct sk_buff *skb, struct Qdisc *sch)
 263 {
 264         struct taprio_sched *q = qdisc_priv(sch);
 265         struct sched_gate_list *sched, *admin;
 266         ktime_t interval_start, interval_end;
 267         struct sched_entry *entry;
 268
 269         rcu_read_lock();
 270         sched = rcu_dereference(q->oper_sched);
 271         admin = rcu_dereference(q->admin_sched);
 272
 273         entry = find_entry_to_transmit(skb, sch, sched, admin, skb->tstamp,
 274                                        &interval_start, &interval_end, true);
 275         rcu_read_unlock();
 276
 277         return entry;
 278 }
 279
 280 static bool taprio_flags_valid(u32 flags)
 281 {
 282         /* Make sure no other flag bits are set. */
 283         if (flags & ~(TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST |
 284                       TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
 285                 return false;
 286         /* txtime-assist and full offload are mutually exclusive */
 287         if ((flags & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST) &&
 288             (flags & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
 289                 return false;
 290         return true;
 291 }
 292
 293 /* This returns the tstamp value set by TCP in terms of the set clock. */
 294 static ktime_t get_tcp_tstamp(struct taprio_sched *q, struct sk_buff *skb)
 295 {
 296         unsigned int offset = skb_network_offset(skb);
 297         const struct ipv6hdr *ipv6h;
 298         const struct iphdr *iph;
 299         struct ipv6hdr _ipv6h;
 300
 301         ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h);
 302         if (!ipv6h)
 303                 return 0;
 304
 305         if (ipv6h->version == 4) {
 306                 iph = (struct iphdr *)ipv6h;
 307                 offset += iph->ihl * 4;
 308
 309                 /* special-case 6in4 tunnelling, as that is a common way to get
 310                  * v6 connectivity in the home
 311                  */
 312                 if (iph->protocol == IPPROTO_IPV6) {
 313                         ipv6h = skb_header_pointer(skb, offset,
 314                                                    sizeof(_ipv6h), &_ipv6h);
 315
 316                         if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP)
 317                                 return 0;
 318                 } else if (iph->protocol != IPPROTO_TCP) {
 319                         return 0;
 320                 }
 321         } else if (ipv6h->version == 6 && ipv6h->nexthdr != IPPROTO_TCP) {
 322                 return 0;
 323         }
 324
 325         return ktime_mono_to_any(skb->skb_mstamp_ns, q->tk_offset);
 326 }
 327
 328 /* There are a few scenarios where we will have to modify the txtime from
 329  * what is read from next_txtime in sched_entry. They are:
 330  * 1. If txtime is in the past,
 331  *    a. The gate for the traffic class is currently open and packet can be
 332  *       transmitted before it closes, schedule the packet right away.
 333  *    b. If the gate corresponding to the traffic class is going to open later
 334  *       in the cycle, set the txtime of packet to the interval start.
 335  * 2. If txtime is in the future, there are packets corresponding to the
 336  *    current traffic class waiting to be transmitted. So, the following
 337  *    possibilities exist:
 338  *    a. We can transmit the packet before the window containing the txtime
 339  *       closes.
 340  *    b. The window might close before the transmission can be completed
 341  *       successfully. So, schedule the packet in the next open window.
 342  */
 343 static long get_packet_txtime(struct sk_buff *skb, struct Qdisc *sch)
 344 {
 345         ktime_t transmit_end_time, interval_end, interval_start, tcp_tstamp;
 346         struct taprio_sched *q = qdisc_priv(sch);
 347         struct sched_gate_list *sched, *admin;
 348         ktime_t minimum_time, now, txtime;
 349         int len, packet_transmit_time;
 350         struct sched_entry *entry;
 351         bool sched_changed;
 352
 353         now = taprio_get_time(q);
 354         minimum_time = ktime_add_ns(now, q->txtime_delay);
 355
 356         tcp_tstamp = get_tcp_tstamp(q, skb);
 357         minimum_time = max_t(ktime_t, minimum_time, tcp_tstamp);
 358
 359         rcu_read_lock();
 360         admin = rcu_dereference(q->admin_sched);
 361         sched = rcu_dereference(q->oper_sched);
 362         if (admin && ktime_after(minimum_time, admin->base_time))
 363                 switch_schedules(q, &admin, &sched);
 364
 365         /* Until the schedule starts, all the queues are open */
 366         if (!sched || ktime_before(minimum_time, sched->base_time)) {
 367                 txtime = minimum_time;
 368                 goto done;
 369         }
 370
 371         len = qdisc_pkt_len(skb);
 372         packet_transmit_time = length_to_duration(q, len);
 373
 374         do {
 375                 sched_changed = 0;
 376
 377                 entry = find_entry_to_transmit(skb, sch, sched, admin,
 378                                                minimum_time,
 379                                                &interval_start, &interval_end,
 380                                                false);
 381                 if (!entry) {
 382                         txtime = 0;
 383                         goto done;
 384                 }
 385
 386                 txtime = entry->next_txtime;
 387                 txtime = max_t(ktime_t, txtime, minimum_time);
 388                 txtime = max_t(ktime_t, txtime, interval_start);
 389
 390                 if (admin && admin != sched &&
 391                     ktime_after(txtime, admin->base_time)) {
 392                         sched = admin;
 393                         sched_changed = 1;
 394                         continue;
 395                 }
 396
 397                 transmit_end_time = ktime_add(txtime, packet_transmit_time);
 398                 minimum_time = transmit_end_time;
 399
 400                 /* Update the txtime of current entry to the next time it's
 401                  * interval starts.
 402                  */
 403                 if (ktime_after(transmit_end_time, interval_end))
 404                         entry->next_txtime = ktime_add(interval_start, sched->cycle_time);
 405         } while (sched_changed || ktime_after(transmit_end_time, interval_end));
 406
 407         entry->next_txtime = transmit_end_time;
 408
 409 done:
 410         rcu_read_unlock();
 411         return txtime;
 412 }
 413
 414 static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 415                           struct sk_buff **to_free)
 416 {
 417         struct taprio_sched *q = qdisc_priv(sch);
 418         struct Qdisc *child;
 419         int queue;
 420
 421         queue = skb_get_queue_mapping(skb);
 422
 423         child = q->qdiscs[queue];
 424         if (unlikely(!child))
 425                 return qdisc_drop(skb, sch, to_free);
 426
 427         if (skb->sk && sock_flag(skb->sk, SOCK_TXTIME)) {
 428                 if (!is_valid_interval(skb, sch))
 429                         return qdisc_drop(skb, sch, to_free);
 430         } else if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
 431                 skb->tstamp = get_packet_txtime(skb, sch);
 432                 if (!skb->tstamp)
 433                         return qdisc_drop(skb, sch, to_free);
 434         }
 435
 436         qdisc_qstats_backlog_inc(sch, skb);
 437         sch->q.qlen++;
 438
 439         return qdisc_enqueue(skb, child, to_free);
 440 }
 441
 442 static struct sk_buff *taprio_peek_soft(struct Qdisc *sch)
 443 {
 444         struct taprio_sched *q = qdisc_priv(sch);
 445         struct net_device *dev = qdisc_dev(sch);
 446         struct sched_entry *entry;
 447         struct sk_buff *skb;
 448         u32 gate_mask;
 449         int i;
 450
 451         rcu_read_lock();
 452         entry = rcu_dereference(q->current_entry);
 453         gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
 454         rcu_read_unlock();
 455
 456         if (!gate_mask)
 457                 return NULL;
 458
 459         for (i = 0; i < dev->num_tx_queues; i++) {
 460                 struct Qdisc *child = q->qdiscs[i];
 461                 int prio;
 462                 u8 tc;
 463
 464                 if (unlikely(!child))
 465                         continue;
 466
 467                 skb = child->ops->peek(child);
 468                 if (!skb)
 469                         continue;
 470
 471                 if (TXTIME_ASSIST_IS_ENABLED(q->flags))
 472                         return skb;
 473
 474                 prio = skb->priority;
 475                 tc = netdev_get_prio_tc_map(dev, prio);
 476
 477                 if (!(gate_mask & BIT(tc)))
 478                         continue;
 479
 480                 return skb;
 481         }
 482
 483         return NULL;
 484 }
 485
 486 static struct sk_buff *taprio_peek_offload(struct Qdisc *sch)
 487 {
 488         struct taprio_sched *q = qdisc_priv(sch);
 489         struct net_device *dev = qdisc_dev(sch);
 490         struct sk_buff *skb;
 491         int i;
 492
 493         for (i = 0; i < dev->num_tx_queues; i++) {
 494                 struct Qdisc *child = q->qdiscs[i];
 495
 496                 if (unlikely(!child))
 497                         continue;
 498
 499                 skb = child->ops->peek(child);
 500                 if (!skb)
 501                         continue;
 502
 503                 return skb;
 504         }
 505
 506         return NULL;
 507 }
 508
 509 static struct sk_buff *taprio_peek(struct Qdisc *sch)
 510 {
 511         struct taprio_sched *q = qdisc_priv(sch);
 512
 513         return q->peek(sch);
 514 }
 515
 516 static void taprio_set_budget(struct taprio_sched *q, struct sched_entry *entry)
 517 {
 518         atomic_set(&entry->budget,
 519                    div64_u64((u64)entry->interval * 1000,
 520                              atomic64_read(&q->picos_per_byte)));
 521 }
 522
 523 static struct sk_buff *taprio_dequeue_soft(struct Qdisc *sch)
 524 {
 525         struct taprio_sched *q = qdisc_priv(sch);
 526         struct net_device *dev = qdisc_dev(sch);
 527         struct sk_buff *skb = NULL;
 528         struct sched_entry *entry;
 529         u32 gate_mask;
 530         int i;
 531
 532         rcu_read_lock();
 533         entry = rcu_dereference(q->current_entry);
 534         /* if there's no entry, it means that the schedule didn't
 535          * start yet, so force all gates to be open, this is in
 536          * accordance to IEEE 802.1Qbv-2015 Section 8.6.9.4.5
 537          * "AdminGateSates"
 538          */
 539         gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
 540
 541         if (!gate_mask)
 542                 goto done;
 543
 544         for (i = 0; i < dev->num_tx_queues; i++) {
 545                 struct Qdisc *child = q->qdiscs[i];
 546                 ktime_t guard;
 547                 int prio;
 548                 int len;
 549                 u8 tc;
 550
 551                 if (unlikely(!child))
 552                         continue;
 553
 554                 if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
 555                         skb = child->ops->dequeue(child);
 556                         if (!skb)
 557                                 continue;
 558                         goto skb_found;
 559                 }
 560
 561                 skb = child->ops->peek(child);
 562                 if (!skb)
 563                         continue;
 564
 565                 prio = skb->priority;
 566                 tc = netdev_get_prio_tc_map(dev, prio);
 567
 568                 if (!(gate_mask & BIT(tc))) {
 569                         skb = NULL;
 570                         continue;
 571                 }
 572
 573                 len = qdisc_pkt_len(skb);
 574                 guard = ktime_add_ns(taprio_get_time(q),
 575                                      length_to_duration(q, len));
 576
 577                 /* In the case that there's no gate entry, there's no
 578                  * guard band ...
 579                  */
 580                 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
 581                     ktime_after(guard, entry->close_time)) {
 582                         skb = NULL;
 583                         continue;
 584                 }
 585
 586                 /* ... and no budget. */
 587                 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
 588                     atomic_sub_return(len, &entry->budget) < 0) {
 589                         skb = NULL;
 590                         continue;
 591                 }
 592
 593                 skb = child->ops->dequeue(child);
 594                 if (unlikely(!skb))
 595                         goto done;
 596
 597 skb_found:
 598                 qdisc_bstats_update(sch, skb);
 599                 qdisc_qstats_backlog_dec(sch, skb);
 600                 sch->q.qlen--;
 601
 602                 goto done;
 603         }
 604
 605 done:
 606         rcu_read_unlock();
 607
 608         return skb;
 609 }
 610
 611 static struct sk_buff *taprio_dequeue_offload(struct Qdisc *sch)
 612 {
 613         struct taprio_sched *q = qdisc_priv(sch);
 614         struct net_device *dev = qdisc_dev(sch);
 615         struct sk_buff *skb;
 616         int i;
 617
 618         for (i = 0; i < dev->num_tx_queues; i++) {
 619                 struct Qdisc *child = q->qdiscs[i];
 620
 621                 if (unlikely(!child))
 622                         continue;
 623
 624                 skb = child->ops->dequeue(child);
 625                 if (unlikely(!skb))
 626                         continue;
 627
 628                 qdisc_bstats_update(sch, skb);
 629                 qdisc_qstats_backlog_dec(sch, skb);
 630                 sch->q.qlen--;
 631
 632                 return skb;
 633         }
 634
 635         return NULL;
 636 }
 637
 638 static struct sk_buff *taprio_dequeue(struct Qdisc *sch)
 639 {
 640         struct taprio_sched *q = qdisc_priv(sch);
 641
 642         return q->dequeue(sch);
 643 }
 644
 645 static bool should_restart_cycle(const struct sched_gate_list *oper,
 646                                  const struct sched_entry *entry)
 647 {
 648         if (list_is_last(&entry->list, &oper->entries))
 649                 return true;
 650
 651         if (ktime_compare(entry->close_time, oper->cycle_close_time) == 0)
 652                 return true;
 653
 654         return false;
 655 }
 656
 657 static bool should_change_schedules(const struct sched_gate_list *admin,
 658                                     const struct sched_gate_list *oper,
 659                                     ktime_t close_time)
 660 {
 661         ktime_t next_base_time, extension_time;
 662
 663         if (!admin)
 664                 return false;
 665
 666         next_base_time = sched_base_time(admin);
 667
 668         /* This is the simple case, the close_time would fall after
 669          * the next schedule base_time.
 670          */
 671         if (ktime_compare(next_base_time, close_time) <= 0)
 672                 return true;
 673
 674         /* This is the cycle_time_extension case, if the close_time
 675          * plus the amount that can be extended would fall after the
 676          * next schedule base_time, we can extend the current schedule
 677          * for that amount.
 678          */
 679         extension_time = ktime_add_ns(close_time, oper->cycle_time_extension);
 680
 681         /* FIXME: the IEEE 802.1Q-2018 Specification isn't clear about
 682          * how precisely the extension should be made. So after
 683          * conformance testing, this logic may change.
 684          */
 685         if (ktime_compare(next_base_time, extension_time) <= 0)
 686                 return true;
 687
 688         return false;
 689 }
 690
 691 static enum hrtimer_restart advance_sched(struct hrtimer *timer)
 692 {
 693         struct taprio_sched *q = container_of(timer, struct taprio_sched,
 694                                               advance_timer);
 695         struct sched_gate_list *oper, *admin;
 696         struct sched_entry *entry, *next;
 697         struct Qdisc *sch = q->root;
 698         ktime_t close_time;
 699
 700         spin_lock(&q->current_entry_lock);
 701         entry = rcu_dereference_protected(q->current_entry,
 702                                           lockdep_is_held(&q->current_entry_lock));
 703         oper = rcu_dereference_protected(q->oper_sched,
 704                                          lockdep_is_held(&q->current_entry_lock));
 705         admin = rcu_dereference_protected(q->admin_sched,
 706                                           lockdep_is_held(&q->current_entry_lock));
 707
 708         if (!oper)
 709                 switch_schedules(q, &admin, &oper);
 710
 711         /* This can happen in two cases: 1. this is the very first run
 712          * of this function (i.e. we weren't running any schedule
 713          * previously); 2. The previous schedule just ended. The first
 714          * entry of all schedules are pre-calculated during the
 715          * schedule initialization.
 716          */
 717         if (unlikely(!entry || entry->close_time == oper->base_time)) {
 718                 next = list_first_entry(&oper->entries, struct sched_entry,
 719                                         list);
 720                 close_time = next->close_time;
 721                 goto first_run;
 722         }
 723
 724         if (should_restart_cycle(oper, entry)) {
 725                 next = list_first_entry(&oper->entries, struct sched_entry,
 726                                         list);
 727                 oper->cycle_close_time = ktime_add_ns(oper->cycle_close_time,
 728                                                       oper->cycle_time);
 729         } else {
 730                 next = list_next_entry(entry, list);
 731         }
 732
 733         close_time = ktime_add_ns(entry->close_time, next->interval);
 734         close_time = min_t(ktime_t, close_time, oper->cycle_close_time);
 735
 736         if (should_change_schedules(admin, oper, close_time)) {
 737                 /* Set things so the next time this runs, the new
 738                  * schedule runs.
 739                  */
 740                 close_time = sched_base_time(admin);
 741                 switch_schedules(q, &admin, &oper);
 742         }
 743
 744         next->close_time = close_time;
 745         taprio_set_budget(q, next);
 746
 747 first_run:
 748         rcu_assign_pointer(q->current_entry, next);
 749         spin_unlock(&q->current_entry_lock);
 750
 751         hrtimer_set_expires(&q->advance_timer, close_time);
 752
 753         rcu_read_lock();
 754         __netif_schedule(sch);
 755         rcu_read_unlock();
 756
 757         return HRTIMER_RESTART;
 758 }
 759
 760 static const struct nla_policy entry_policy[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = {
 761         [TCA_TAPRIO_SCHED_ENTRY_INDEX]     = { .type = NLA_U32 },
 762         [TCA_TAPRIO_SCHED_ENTRY_CMD]       = { .type = NLA_U8 },
 763         [TCA_TAPRIO_SCHED_ENTRY_GATE_MASK] = { .type = NLA_U32 },
 764         [TCA_TAPRIO_SCHED_ENTRY_INTERVAL]  = { .type = NLA_U32 },
 765 };
 766
 767 static const struct nla_policy taprio_policy[TCA_TAPRIO_ATTR_MAX + 1] = {
 768         [TCA_TAPRIO_ATTR_PRIOMAP]              = {
 769                 .len = sizeof(struct tc_mqprio_qopt)
 770         },
 771         [TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST]           = { .type = NLA_NESTED },
 772         [TCA_TAPRIO_ATTR_SCHED_BASE_TIME]            = { .type = NLA_S64 },
 773         [TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]         = { .type = NLA_NESTED },
 774         [TCA_TAPRIO_ATTR_SCHED_CLOCKID]              = { .type = NLA_S32 },
 775         [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]           = { .type = NLA_S64 },
 776         [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION] = { .type = NLA_S64 },
 777         [TCA_TAPRIO_ATTR_FLAGS]                      = { .type = NLA_U32 },
 778         [TCA_TAPRIO_ATTR_TXTIME_DELAY]               = { .type = NLA_U32 },
 779 };
 780
 781 static int fill_sched_entry(struct taprio_sched *q, struct nlattr **tb,
 782                             struct sched_entry *entry,
 783                             struct netlink_ext_ack *extack)
 784 {
 785         int min_duration = length_to_duration(q, ETH_ZLEN);
 786         u32 interval = 0;
 787
 788         if (tb[TCA_TAPRIO_SCHED_ENTRY_CMD])
 789                 entry->command = nla_get_u8(
 790                         tb[TCA_TAPRIO_SCHED_ENTRY_CMD]);
 791
 792         if (tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK])
 793                 entry->gate_mask = nla_get_u32(
 794                         tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK]);
 795
 796         if (tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL])
 797                 interval = nla_get_u32(
 798                         tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL]);
 799
 800         /* The interval should allow at least the minimum ethernet
 801          * frame to go out.
 802          */
 803         if (interval < min_duration) {
 804                 NL_SET_ERR_MSG(extack, "Invalid interval for schedule entry");
 805                 return -EINVAL;
 806         }
 807
 808         entry->interval = interval;
 809
 810         return 0;
 811 }
 812
 813 static int parse_sched_entry(struct taprio_sched *q, struct nlattr *n,
 814                              struct sched_entry *entry, int index,
 815                              struct netlink_ext_ack *extack)
 816 {
 817         struct nlattr *tb[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = { };
 818         int err;
 819
 820         err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_SCHED_ENTRY_MAX, n,
 821                                           entry_policy, NULL);
 822         if (err < 0) {
 823                 NL_SET_ERR_MSG(extack, "Could not parse nested entry");
 824                 return -EINVAL;
 825         }
 826
 827         entry->index = index;
 828
 829         return fill_sched_entry(q, tb, entry, extack);
 830 }
 831
 832 static int parse_sched_list(struct taprio_sched *q, struct nlattr *list,
 833                             struct sched_gate_list *sched,
 834                             struct netlink_ext_ack *extack)
 835 {
 836         struct nlattr *n;
 837         int err, rem;
 838         int i = 0;
 839
 840         if (!list)
 841                 return -EINVAL;
 842
 843         nla_for_each_nested(n, list, rem) {
 844                 struct sched_entry *entry;
 845
 846                 if (nla_type(n) != TCA_TAPRIO_SCHED_ENTRY) {
 847                         NL_SET_ERR_MSG(extack, "Attribute is not of type 'entry'");
 848                         continue;
 849                 }
 850
 851                 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 852                 if (!entry) {
 853                         NL_SET_ERR_MSG(extack, "Not enough memory for entry");
 854                         return -ENOMEM;
 855                 }
 856
 857                 err = parse_sched_entry(q, n, entry, i, extack);
 858                 if (err < 0) {
 859                         kfree(entry);
 860                         return err;
 861                 }
 862
 863                 list_add_tail(&entry->list, &sched->entries);
 864                 i++;
 865         }
 866
 867         sched->num_entries = i;
 868
 869         return i;
 870 }
 871
 872 static int parse_taprio_schedule(struct taprio_sched *q, struct nlattr **tb,
 873                                  struct sched_gate_list *new,
 874                                  struct netlink_ext_ack *extack)
 875 {
 876         int err = 0;
 877
 878         if (tb[TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]) {
 879                 NL_SET_ERR_MSG(extack, "Adding a single entry is not supported");
 880                 return -ENOTSUPP;
 881         }
 882
 883         if (tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME])
 884                 new->base_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME]);
 885
 886         if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION])
 887                 new->cycle_time_extension = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION]);
 888
 889         if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME])
 890                 new->cycle_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]);
 891
 892         if (tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST])
 893                 err = parse_sched_list(q, tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST],
 894                                        new, extack);
 895         if (err < 0)
 896                 return err;
 897
 898         if (!new->cycle_time) {
 899                 struct sched_entry *entry;
 900                 ktime_t cycle = 0;
 901
 902                 list_for_each_entry(entry, &new->entries, list)
 903                         cycle = ktime_add_ns(cycle, entry->interval);
 904                 new->cycle_time = cycle;
 905         }
 906
 907         return 0;
 908 }
 909
 910 static int taprio_parse_mqprio_opt(struct net_device *dev,
 911                                    struct tc_mqprio_qopt *qopt,
 912                                    struct netlink_ext_ack *extack,
 913                                    u32 taprio_flags)
 914 {
 915         int i, j;
 916
 917         if (!qopt && !dev->num_tc) {
 918                 NL_SET_ERR_MSG(extack, "'mqprio' configuration is necessary");
 919                 return -EINVAL;
 920         }
 921
 922         /* If num_tc is already set, it means that the user already
 923          * configured the mqprio part
 924          */
 925         if (dev->num_tc)
 926                 return 0;
 927
 928         /* Verify num_tc is not out of max range */
 929         if (qopt->num_tc > TC_MAX_QUEUE) {
 930                 NL_SET_ERR_MSG(extack, "Number of traffic classes is outside valid range");
 931                 return -EINVAL;
 932         }
 933
 934         /* taprio imposes that traffic classes map 1:n to tx queues */
 935         if (qopt->num_tc > dev->num_tx_queues) {
 936                 NL_SET_ERR_MSG(extack, "Number of traffic classes is greater than number of HW queues");
 937                 return -EINVAL;
 938         }
 939
 940         /* Verify priority mapping uses valid tcs */
 941         for (i = 0; i <= TC_BITMASK; i++) {
 942                 if (qopt->prio_tc_map[i] >= qopt->num_tc) {
 943                         NL_SET_ERR_MSG(extack, "Invalid traffic class in priority to traffic class mapping");
 944                         return -EINVAL;
 945                 }
 946         }
 947
 948         for (i = 0; i < qopt->num_tc; i++) {
 949                 unsigned int last = qopt->offset[i] + qopt->count[i];
 950
 951                 /* Verify the queue count is in tx range being equal to the
 952                  * real_num_tx_queues indicates the last queue is in use.
 953                  */
 954                 if (qopt->offset[i] >= dev->num_tx_queues ||
 955                     !qopt->count[i] ||
 956                     last > dev->real_num_tx_queues) {
 957                         NL_SET_ERR_MSG(extack, "Invalid queue in traffic class to queue mapping");
 958                         return -EINVAL;
 959                 }
 960
 961                 if (TXTIME_ASSIST_IS_ENABLED(taprio_flags))
 962                         continue;
 963
 964                 /* Verify that the offset and counts do not overlap */
 965                 for (j = i + 1; j < qopt->num_tc; j++) {
 966                         if (last > qopt->offset[j]) {
 967                                 NL_SET_ERR_MSG(extack, "Detected overlap in the traffic class to queue mapping");
 968                                 return -EINVAL;
 969                         }
 970                 }
 971         }
 972
 973         return 0;
 974 }
 975
 976 static int taprio_get_start_time(struct Qdisc *sch,
 977                                  struct sched_gate_list *sched,
 978                                  ktime_t *start)
 979 {
 980         struct taprio_sched *q = qdisc_priv(sch);
 981         ktime_t now, base, cycle;
 982         s64 n;
 983
 984         base = sched_base_time(sched);
 985         now = taprio_get_time(q);
 986
 987         if (ktime_after(base, now)) {
 988                 *start = base;
 989                 return 0;
 990         }
 991
 992         cycle = sched->cycle_time;
 993
 994         /* The qdisc is expected to have at least one sched_entry.  Moreover,
 995          * any entry must have 'interval' > 0. Thus if the cycle time is zero,
 996          * something went really wrong. In that case, we should warn about this
 997          * inconsistent state and return error.
 998          */
 999         if (WARN_ON(!cycle))
1000                 return -EFAULT;
1001
1002         /* Schedule the start time for the beginning of the next
1003          * cycle.
1004          */
1005         n = div64_s64(ktime_sub_ns(now, base), cycle);
1006         *start = ktime_add_ns(base, (n + 1) * cycle);
1007         return 0;
1008 }
1009
1010 static void setup_first_close_time(struct taprio_sched *q,
1011                                    struct sched_gate_list *sched, ktime_t base)
1012 {
1013         struct sched_entry *first;
1014         ktime_t cycle;
1015
1016         first = list_first_entry(&sched->entries,
1017                                  struct sched_entry, list);
1018
1019         cycle = sched->cycle_time;
1020
1021         /* FIXME: find a better place to do this */
1022         sched->cycle_close_time = ktime_add_ns(base, cycle);
1023
1024         first->close_time = ktime_add_ns(base, first->interval);
1025         taprio_set_budget(q, first);
1026         rcu_assign_pointer(q->current_entry, NULL);
1027 }
1028
1029 static void taprio_start_sched(struct Qdisc *sch,
1030                                ktime_t start, struct sched_gate_list *new)
1031 {
1032         struct taprio_sched *q = qdisc_priv(sch);
1033         ktime_t expires;
1034
1035         if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1036                 return;
1037
1038         expires = hrtimer_get_expires(&q->advance_timer);
1039         if (expires == 0)
1040                 expires = KTIME_MAX;
1041
1042         /* If the new schedule starts before the next expiration, we
1043          * reprogram it to the earliest one, so we change the admin
1044          * schedule to the operational one at the right time.
1045          */
1046         start = min_t(ktime_t, start, expires);
1047
1048         hrtimer_start(&q->advance_timer, start, HRTIMER_MODE_ABS);
1049 }
1050
1051 static void taprio_set_picos_per_byte(struct net_device *dev,
1052                                       struct taprio_sched *q)
1053 {
1054         struct ethtool_link_ksettings ecmd;
1055         int speed = SPEED_10;
1056         int picos_per_byte;
1057         int err;
1058
1059         err = __ethtool_get_link_ksettings(dev, &ecmd);
1060         if (err < 0)
1061                 goto skip;
1062
1063         if (ecmd.base.speed && ecmd.base.speed != SPEED_UNKNOWN)
1064                 speed = ecmd.base.speed;
1065
1066 skip:
1067         picos_per_byte = (USEC_PER_SEC * 8) / speed;
1068
1069         atomic64_set(&q->picos_per_byte, picos_per_byte);
1070         netdev_dbg(dev, "taprio: set %s's picos_per_byte to: %lld, linkspeed: %d\n",
1071                    dev->name, (long long)atomic64_read(&q->picos_per_byte),
1072                    ecmd.base.speed);
1073 }
1074
1075 static int taprio_dev_notifier(struct notifier_block *nb, unsigned long event,
1076                                void *ptr)
1077 {
1078         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1079         struct net_device *qdev;
1080         struct taprio_sched *q;
1081         bool found = false;
1082
1083         ASSERT_RTNL();
1084
1085         if (event != NETDEV_UP && event != NETDEV_CHANGE)
1086                 return NOTIFY_DONE;
1087
1088         spin_lock(&taprio_list_lock);
1089         list_for_each_entry(q, &taprio_list, taprio_list) {
1090                 qdev = qdisc_dev(q->root);
1091                 if (qdev == dev) {
1092                         found = true;
1093                         break;
1094                 }
1095         }
1096         spin_unlock(&taprio_list_lock);
1097
1098         if (found)
1099                 taprio_set_picos_per_byte(dev, q);
1100
1101         return NOTIFY_DONE;
1102 }
1103
1104 static void setup_txtime(struct taprio_sched *q,
1105                          struct sched_gate_list *sched, ktime_t base)
1106 {
1107         struct sched_entry *entry;
1108         u32 interval = 0;
1109
1110         list_for_each_entry(entry, &sched->entries, list) {
1111                 entry->next_txtime = ktime_add_ns(base, interval);
1112                 interval += entry->interval;
1113         }
1114 }
1115
1116 static struct tc_taprio_qopt_offload *taprio_offload_alloc(int num_entries)
1117 {
1118         struct __tc_taprio_qopt_offload *__offload;
1119
1120         __offload = kzalloc(struct_size(__offload, offload.entries, num_entries),
1121                             GFP_KERNEL);
1122         if (!__offload)
1123                 return NULL;
1124
1125         refcount_set(&__offload->users, 1);
1126
1127         return &__offload->offload;
1128 }
1129
1130 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
1131                                                   *offload)
1132 {
1133         struct __tc_taprio_qopt_offload *__offload;
1134
1135         __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1136                                  offload);
1137
1138         refcount_inc(&__offload->users);
1139
1140         return offload;
1141 }
1142 EXPORT_SYMBOL_GPL(taprio_offload_get);
1143
1144 void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
1145 {
1146         struct __tc_taprio_qopt_offload *__offload;
1147
1148         __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1149                                  offload);
1150
1151         if (!refcount_dec_and_test(&__offload->users))
1152                 return;
1153
1154         kfree(__offload);
1155 }
1156 EXPORT_SYMBOL_GPL(taprio_offload_free);
1157
1158 /* The function will only serve to keep the pointers to the "oper" and "admin"
1159  * schedules valid in relation to their base times, so when calling dump() the
1160  * users looks at the right schedules.
1161  * When using full offload, the admin configuration is promoted to oper at the
1162  * base_time in the PHC time domain.  But because the system time is not
1163  * necessarily in sync with that, we can't just trigger a hrtimer to call
1164  * switch_schedules at the right hardware time.
1165  * At the moment we call this by hand right away from taprio, but in the future
1166  * it will be useful to create a mechanism for drivers to notify taprio of the
1167  * offload state (PENDING, ACTIVE, INACTIVE) so it can be visible in dump().
1168  * This is left as TODO.
1169  */
1170 static void taprio_offload_config_changed(struct taprio_sched *q)
1171 {
1172         struct sched_gate_list *oper, *admin;
1173
1174         spin_lock(&q->current_entry_lock);
1175
1176         oper = rcu_dereference_protected(q->oper_sched,
1177                                          lockdep_is_held(&q->current_entry_lock));
1178         admin = rcu_dereference_protected(q->admin_sched,
1179                                           lockdep_is_held(&q->current_entry_lock));
1180
1181         switch_schedules(q, &admin, &oper);
1182
1183         spin_unlock(&q->current_entry_lock);
1184 }
1185
1186 static u32 tc_map_to_queue_mask(struct net_device *dev, u32 tc_mask)
1187 {
1188         u32 i, queue_mask = 0;
1189
1190         for (i = 0; i < dev->num_tc; i++) {
1191                 u32 offset, count;
1192
1193                 if (!(tc_mask & BIT(i)))
1194                         continue;
1195
1196                 offset = dev->tc_to_txq[i].offset;
1197                 count = dev->tc_to_txq[i].count;
1198
1199                 queue_mask |= GENMASK(offset + count - 1, offset);
1200         }
1201
1202         return queue_mask;
1203 }
1204
1205 static void taprio_sched_to_offload(struct net_device *dev,
1206                                     struct sched_gate_list *sched,
1207                                     struct tc_taprio_qopt_offload *offload)
1208 {
1209         struct sched_entry *entry;
1210         int i = 0;
1211
1212         offload->base_time = sched->base_time;
1213         offload->cycle_time = sched->cycle_time;
1214         offload->cycle_time_extension = sched->cycle_time_extension;
1215
1216         list_for_each_entry(entry, &sched->entries, list) {
1217                 struct tc_taprio_sched_entry *e = &offload->entries[i];
1218
1219                 e->command = entry->command;
1220                 e->interval = entry->interval;
1221                 e->gate_mask = tc_map_to_queue_mask(dev, entry->gate_mask);
1222
1223                 i++;
1224         }
1225
1226         offload->num_entries = i;
1227 }
1228
1229 static int taprio_enable_offload(struct net_device *dev,
1230                                  struct taprio_sched *q,
1231                                  struct sched_gate_list *sched,
1232                                  struct netlink_ext_ack *extack)
1233 {
1234         const struct net_device_ops *ops = dev->netdev_ops;
1235         struct tc_taprio_qopt_offload *offload;
1236         int err = 0;
1237
1238         if (!ops->ndo_setup_tc) {
1239                 NL_SET_ERR_MSG(extack,
1240                                "Device does not support taprio offload");
1241                 return -EOPNOTSUPP;
1242         }
1243
1244         offload = taprio_offload_alloc(sched->num_entries);
1245         if (!offload) {
1246                 NL_SET_ERR_MSG(extack,
1247                                "Not enough memory for enabling offload mode");
1248                 return -ENOMEM;
1249         }
1250         offload->enable = 1;
1251         taprio_sched_to_offload(dev, sched, offload);
1252
1253         err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1254         if (err < 0) {
1255                 NL_SET_ERR_MSG(extack,
1256                                "Device failed to setup taprio offload");
1257                 goto done;
1258         }
1259
1260 done:
1261         taprio_offload_free(offload);
1262
1263         return err;
1264 }
1265
1266 static int taprio_disable_offload(struct net_device *dev,
1267                                   struct taprio_sched *q,
1268                                   struct netlink_ext_ack *extack)
1269 {
1270         const struct net_device_ops *ops = dev->netdev_ops;
1271         struct tc_taprio_qopt_offload *offload;
1272         int err;
1273
1274         if (!FULL_OFFLOAD_IS_ENABLED(q->flags))
1275                 return 0;
1276
1277         if (!ops->ndo_setup_tc)
1278                 return -EOPNOTSUPP;
1279
1280         offload = taprio_offload_alloc(0);
1281         if (!offload) {
1282                 NL_SET_ERR_MSG(extack,
1283                                "Not enough memory to disable offload mode");
1284                 return -ENOMEM;
1285         }
1286         offload->enable = 0;
1287
1288         err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1289         if (err < 0) {
1290                 NL_SET_ERR_MSG(extack,
1291                                "Device failed to disable offload");
1292                 goto out;
1293         }
1294
1295 out:
1296         taprio_offload_free(offload);
1297
1298         return err;
1299 }
1300
1301 /* If full offload is enabled, the only possible clockid is the net device's
1302  * PHC. For that reason, specifying a clockid through netlink is incorrect.
1303  * For txtime-assist, it is implicitly assumed that the device's PHC is kept
1304  * in sync with the specified clockid via a user space daemon such as phc2sys.
1305  * For both software taprio and txtime-assist, the clockid is used for the
1306  * hrtimer that advances the schedule and hence mandatory.
1307  */
1308 static int taprio_parse_clockid(struct Qdisc *sch, struct nlattr **tb,
1309                                 struct netlink_ext_ack *extack)
1310 {
1311         struct taprio_sched *q = qdisc_priv(sch);
1312         struct net_device *dev = qdisc_dev(sch);
1313         int err = -EINVAL;
1314
1315         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1316                 const struct ethtool_ops *ops = dev->ethtool_ops;
1317                 struct ethtool_ts_info info = {
1318                         .cmd = ETHTOOL_GET_TS_INFO,
1319                         .phc_index = -1,
1320                 };
1321
1322                 if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1323                         NL_SET_ERR_MSG(extack,
1324                                        "The 'clockid' cannot be specified for full offload");
1325                         goto out;
1326                 }
1327
1328                 if (ops && ops->get_ts_info)
1329                         err = ops->get_ts_info(dev, &info);
1330
1331                 if (err || info.phc_index < 0) {
1332                         NL_SET_ERR_MSG(extack,
1333                                        "Device does not have a PTP clock");
1334                         err = -ENOTSUPP;
1335                         goto out;
1336                 }
1337         } else if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1338                 int clockid = nla_get_s32(tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]);
1339
1340                 /* We only support static clockids and we don't allow
1341                  * for it to be modified after the first init.
1342                  */
1343                 if (clockid < 0 ||
1344                     (q->clockid != -1 && q->clockid != clockid)) {
1345                         NL_SET_ERR_MSG(extack,
1346                                        "Changing the 'clockid' of a running schedule is not supported");
1347                         err = -ENOTSUPP;
1348                         goto out;
1349                 }
1350
1351                 switch (clockid) {
1352                 case CLOCK_REALTIME:
1353                         q->tk_offset = TK_OFFS_REAL;
1354                         break;
1355                 case CLOCK_MONOTONIC:
1356                         q->tk_offset = TK_OFFS_MAX;
1357                         break;
1358                 case CLOCK_BOOTTIME:
1359                         q->tk_offset = TK_OFFS_BOOT;
1360                         break;
1361                 case CLOCK_TAI:
1362                         q->tk_offset = TK_OFFS_TAI;
1363                         break;
1364                 default:
1365                         NL_SET_ERR_MSG(extack, "Invalid 'clockid'");
1366                         err = -EINVAL;
1367                         goto out;
1368                 }
1369
1370                 q->clockid = clockid;
1371         } else {
1372                 NL_SET_ERR_MSG(extack, "Specifying a 'clockid' is mandatory");
1373                 goto out;
1374         }
1375
1376         /* Everything went ok, return success. */
1377         err = 0;
1378
1379 out:
1380         return err;
1381 }
1382
1383 static int taprio_mqprio_cmp(const struct net_device *dev,
1384                              const struct tc_mqprio_qopt *mqprio)
1385 {
1386         int i;
1387
1388         if (!mqprio || mqprio->num_tc != dev->num_tc)
1389                 return -1;
1390
1391         for (i = 0; i < mqprio->num_tc; i++)
1392                 if (dev->tc_to_txq[i].count != mqprio->count[i] ||
1393                     dev->tc_to_txq[i].offset != mqprio->offset[i])
1394                         return -1;
1395
1396         for (i = 0; i <= TC_BITMASK; i++)
1397                 if (dev->prio_tc_map[i] != mqprio->prio_tc_map[i])
1398                         return -1;
1399
1400         return 0;
1401 }
1402
1403 /* The semantics of the 'flags' argument in relation to 'change()'
1404  * requests, are interpreted following two rules (which are applied in
1405  * this order): (1) an omitted 'flags' argument is interpreted as
1406  * zero; (2) the 'flags' of a "running" taprio instance cannot be
1407  * changed.
1408  */
1409 static int taprio_new_flags(const struct nlattr *attr, u32 old,
1410                             struct netlink_ext_ack *extack)
1411 {
1412         u32 new = 0;
1413
1414         if (attr)
1415                 new = nla_get_u32(attr);
1416
1417         if (old != TAPRIO_FLAGS_INVALID && old != new) {
1418                 NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
1419                 return -EOPNOTSUPP;
1420         }
1421
1422         if (!taprio_flags_valid(new)) {
1423                 NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
1424                 return -EINVAL;
1425         }
1426
1427         return new;
1428 }
1429
1430 static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
1431                          struct netlink_ext_ack *extack)
1432 {
1433         struct nlattr *tb[TCA_TAPRIO_ATTR_MAX + 1] = { };
1434         struct sched_gate_list *oper, *admin, *new_admin;
1435         struct taprio_sched *q = qdisc_priv(sch);
1436         struct net_device *dev = qdisc_dev(sch);
1437         struct tc_mqprio_qopt *mqprio = NULL;
1438         unsigned long flags;
1439         ktime_t start;
1440         int i, err;
1441
1442         err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_ATTR_MAX, opt,
1443                                           taprio_policy, extack);
1444         if (err < 0)
1445                 return err;
1446
1447         if (tb[TCA_TAPRIO_ATTR_PRIOMAP])
1448                 mqprio = nla_data(tb[TCA_TAPRIO_ATTR_PRIOMAP]);
1449
1450         err = taprio_new_flags(tb[TCA_TAPRIO_ATTR_FLAGS],
1451                                q->flags, extack);
1452         if (err < 0)
1453                 return err;
1454
1455         q->flags = err;
1456
1457         err = taprio_parse_mqprio_opt(dev, mqprio, extack, q->flags);
1458         if (err < 0)
1459                 return err;
1460
1461         new_admin = kzalloc(sizeof(*new_admin), GFP_KERNEL);
1462         if (!new_admin) {
1463                 NL_SET_ERR_MSG(extack, "Not enough memory for a new schedule");
1464                 return -ENOMEM;
1465         }
1466         INIT_LIST_HEAD(&new_admin->entries);
1467
1468         rcu_read_lock();
1469         oper = rcu_dereference(q->oper_sched);
1470         admin = rcu_dereference(q->admin_sched);
1471         rcu_read_unlock();
1472
1473         /* no changes - no new mqprio settings */
1474         if (!taprio_mqprio_cmp(dev, mqprio))
1475                 mqprio = NULL;
1476
1477         if (mqprio && (oper || admin)) {
1478                 NL_SET_ERR_MSG(extack, "Changing the traffic mapping of a running schedule is not supported");
1479                 err = -ENOTSUPP;
1480                 goto free_sched;
1481         }
1482
1483         err = parse_taprio_schedule(q, tb, new_admin, extack);
1484         if (err < 0)
1485                 goto free_sched;
1486
1487         if (new_admin->num_entries == 0) {
1488                 NL_SET_ERR_MSG(extack, "There should be at least one entry in the schedule");
1489                 err = -EINVAL;
1490                 goto free_sched;
1491         }
1492
1493         err = taprio_parse_clockid(sch, tb, extack);
1494         if (err < 0)
1495                 goto free_sched;
1496
1497         taprio_set_picos_per_byte(dev, q);
1498
1499         if (mqprio) {
1500                 netdev_set_num_tc(dev, mqprio->num_tc);
1501                 for (i = 0; i < mqprio->num_tc; i++)
1502                         netdev_set_tc_queue(dev, i,
1503                                             mqprio->count[i],
1504                                             mqprio->offset[i]);
1505
1506                 /* Always use supplied priority mappings */
1507                 for (i = 0; i <= TC_BITMASK; i++)
1508                         netdev_set_prio_tc_map(dev, i,
1509                                                mqprio->prio_tc_map[i]);
1510         }
1511
1512         if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1513                 err = taprio_enable_offload(dev, q, new_admin, extack);
1514         else
1515                 err = taprio_disable_offload(dev, q, extack);
1516         if (err)
1517                 goto free_sched;
1518
1519         /* Protects against enqueue()/dequeue() */
1520         spin_lock_bh(qdisc_lock(sch));
1521
1522         if (tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]) {
1523                 if (!TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1524                         NL_SET_ERR_MSG_MOD(extack, "txtime-delay can only be set when txtime-assist mode is enabled");
1525                         err = -EINVAL;
1526                         goto unlock;
1527                 }
1528
1529                 q->txtime_delay = nla_get_u32(tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]);
1530         }
1531
1532         if (!TXTIME_ASSIST_IS_ENABLED(q->flags) &&
1533             !FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1534             !hrtimer_active(&q->advance_timer)) {
1535                 hrtimer_init(&q->advance_timer, q->clockid, HRTIMER_MODE_ABS);
1536                 q->advance_timer.function = advance_sched;
1537         }
1538
1539         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1540                 q->dequeue = taprio_dequeue_offload;
1541                 q->peek = taprio_peek_offload;
1542         } else {
1543                 /* Be sure to always keep the function pointers
1544                  * in a consistent state.
1545                  */
1546                 q->dequeue = taprio_dequeue_soft;
1547                 q->peek = taprio_peek_soft;
1548         }
1549
1550         err = taprio_get_start_time(sch, new_admin, &start);
1551         if (err < 0) {
1552                 NL_SET_ERR_MSG(extack, "Internal error: failed get start time");
1553                 goto unlock;
1554         }
1555
1556         setup_txtime(q, new_admin, start);
1557
1558         if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1559                 if (!oper) {
1560                         rcu_assign_pointer(q->oper_sched, new_admin);
1561                         err = 0;
1562                         new_admin = NULL;
1563                         goto unlock;
1564                 }
1565
1566                 rcu_assign_pointer(q->admin_sched, new_admin);
1567                 if (admin)
1568                         call_rcu(&admin->rcu, taprio_free_sched_cb);
1569         } else {
1570                 setup_first_close_time(q, new_admin, start);
1571
1572                 /* Protects against advance_sched() */
1573                 spin_lock_irqsave(&q->current_entry_lock, flags);
1574
1575                 taprio_start_sched(sch, start, new_admin);
1576
1577                 rcu_assign_pointer(q->admin_sched, new_admin);
1578                 if (admin)
1579                         call_rcu(&admin->rcu, taprio_free_sched_cb);
1580
1581                 spin_unlock_irqrestore(&q->current_entry_lock, flags);
1582
1583                 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1584                         taprio_offload_config_changed(q);
1585         }
1586
1587         new_admin = NULL;
1588         err = 0;
1589
1590 unlock:
1591         spin_unlock_bh(qdisc_lock(sch));
1592
1593 free_sched:
1594         if (new_admin)
1595                 call_rcu(&new_admin->rcu, taprio_free_sched_cb);
1596
1597         return err;
1598 }
1599
1600 static void taprio_reset(struct Qdisc *sch)
1601 {
1602         struct taprio_sched *q = qdisc_priv(sch);
1603         struct net_device *dev = qdisc_dev(sch);
1604         int i;
1605
1606         hrtimer_cancel(&q->advance_timer);
1607         if (q->qdiscs) {
1608                 for (i = 0; i < dev->num_tx_queues && q->qdiscs[i]; i++)
1609                         qdisc_reset(q->qdiscs[i]);
1610         }
1611         sch->qstats.backlog = 0;
1612         sch->q.qlen = 0;
1613 }
1614
1615 static void taprio_destroy(struct Qdisc *sch)
1616 {
1617         struct taprio_sched *q = qdisc_priv(sch);
1618         struct net_device *dev = qdisc_dev(sch);
1619         unsigned int i;
1620
1621         spin_lock(&taprio_list_lock);
1622         list_del(&q->taprio_list);
1623         spin_unlock(&taprio_list_lock);
1624
1625
1626         taprio_disable_offload(dev, q, NULL);
1627
1628         if (q->qdiscs) {
1629                 for (i = 0; i < dev->num_tx_queues && q->qdiscs[i]; i++)
1630                         qdisc_put(q->qdiscs[i]);
1631
1632                 kfree(q->qdiscs);
1633         }
1634         q->qdiscs = NULL;
1635
1636         netdev_reset_tc(dev);
1637
1638         if (q->oper_sched)
1639                 call_rcu(&q->oper_sched->rcu, taprio_free_sched_cb);
1640
1641         if (q->admin_sched)
1642                 call_rcu(&q->admin_sched->rcu, taprio_free_sched_cb);
1643 }
1644
1645 static int taprio_init(struct Qdisc *sch, struct nlattr *opt,
1646                        struct netlink_ext_ack *extack)
1647 {
1648         struct taprio_sched *q = qdisc_priv(sch);
1649         struct net_device *dev = qdisc_dev(sch);
1650         int i;
1651
1652         spin_lock_init(&q->current_entry_lock);
1653
1654         hrtimer_init(&q->advance_timer, CLOCK_TAI, HRTIMER_MODE_ABS);
1655         q->advance_timer.function = advance_sched;
1656
1657         q->dequeue = taprio_dequeue_soft;
1658         q->peek = taprio_peek_soft;
1659
1660         q->root = sch;
1661
1662         /* We only support static clockids. Use an invalid value as default
1663          * and get the valid one on taprio_change().
1664          */
1665         q->clockid = -1;
1666         q->flags = TAPRIO_FLAGS_INVALID;
1667
1668         spin_lock(&taprio_list_lock);
1669         list_add(&q->taprio_list, &taprio_list);
1670         spin_unlock(&taprio_list_lock);
1671
1672         if (sch->parent != TC_H_ROOT)
1673                 return -EOPNOTSUPP;
1674
1675         if (!netif_is_multiqueue(dev))
1676                 return -EOPNOTSUPP;
1677
1678         /* pre-allocate qdisc, attachment can't fail */
1679         q->qdiscs = kcalloc(dev->num_tx_queues,
1680                             sizeof(q->qdiscs[0]),
1681                             GFP_KERNEL);
1682
1683         if (!q->qdiscs)
1684                 return -ENOMEM;
1685
1686         if (!opt)
1687                 return -EINVAL;
1688
1689         for (i = 0; i < dev->num_tx_queues; i++) {
1690                 struct netdev_queue *dev_queue;
1691                 struct Qdisc *qdisc;
1692
1693                 dev_queue = netdev_get_tx_queue(dev, i);
1694                 qdisc = qdisc_create_dflt(dev_queue,
1695                                           &pfifo_qdisc_ops,
1696                                           TC_H_MAKE(TC_H_MAJ(sch->handle),
1697                                                     TC_H_MIN(i + 1)),
1698                                           extack);
1699                 if (!qdisc)
1700                         return -ENOMEM;
1701
1702                 if (i < dev->real_num_tx_queues)
1703                         qdisc_hash_add(qdisc, false);
1704
1705                 q->qdiscs[i] = qdisc;
1706         }
1707
1708         return taprio_change(sch, opt, extack);
1709 }
1710
1711 static struct netdev_queue *taprio_queue_get(struct Qdisc *sch,
1712                                              unsigned long cl)
1713 {
1714         struct net_device *dev = qdisc_dev(sch);
1715         unsigned long ntx = cl - 1;
1716
1717         if (ntx >= dev->num_tx_queues)
1718                 return NULL;
1719
1720         return netdev_get_tx_queue(dev, ntx);
1721 }
1722
1723 static int taprio_graft(struct Qdisc *sch, unsigned long cl,
1724                         struct Qdisc *new, struct Qdisc **old,
1725                         struct netlink_ext_ack *extack)
1726 {
1727         struct taprio_sched *q = qdisc_priv(sch);
1728         struct net_device *dev = qdisc_dev(sch);
1729         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1730
1731         if (!dev_queue)
1732                 return -EINVAL;
1733
1734         if (dev->flags & IFF_UP)
1735                 dev_deactivate(dev);
1736
1737         *old = q->qdiscs[cl - 1];
1738         q->qdiscs[cl - 1] = new;
1739
1740         if (new)
1741                 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1742
1743         if (dev->flags & IFF_UP)
1744                 dev_activate(dev);
1745
1746         return 0;
1747 }
1748
1749 static int dump_entry(struct sk_buff *msg,
1750                       const struct sched_entry *entry)
1751 {
1752         struct nlattr *item;
1753
1754         item = nla_nest_start_noflag(msg, TCA_TAPRIO_SCHED_ENTRY);
1755         if (!item)
1756                 return -ENOSPC;
1757
1758         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INDEX, entry->index))
1759                 goto nla_put_failure;
1760
1761         if (nla_put_u8(msg, TCA_TAPRIO_SCHED_ENTRY_CMD, entry->command))
1762                 goto nla_put_failure;
1763
1764         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_GATE_MASK,
1765                         entry->gate_mask))
1766                 goto nla_put_failure;
1767
1768         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INTERVAL,
1769                         entry->interval))
1770                 goto nla_put_failure;
1771
1772         return nla_nest_end(msg, item);
1773
1774 nla_put_failure:
1775         nla_nest_cancel(msg, item);
1776         return -1;
1777 }
1778
1779 static int dump_schedule(struct sk_buff *msg,
1780                          const struct sched_gate_list *root)
1781 {
1782         struct nlattr *entry_list;
1783         struct sched_entry *entry;
1784
1785         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_BASE_TIME,
1786                         root->base_time, TCA_TAPRIO_PAD))
1787                 return -1;
1788
1789         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME,
1790                         root->cycle_time, TCA_TAPRIO_PAD))
1791                 return -1;
1792
1793         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION,
1794                         root->cycle_time_extension, TCA_TAPRIO_PAD))
1795                 return -1;
1796
1797         entry_list = nla_nest_start_noflag(msg,
1798                                            TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST);
1799         if (!entry_list)
1800                 goto error_nest;
1801
1802         list_for_each_entry(entry, &root->entries, list) {
1803                 if (dump_entry(msg, entry) < 0)
1804                         goto error_nest;
1805         }
1806
1807         nla_nest_end(msg, entry_list);
1808         return 0;
1809
1810 error_nest:
1811         nla_nest_cancel(msg, entry_list);
1812         return -1;
1813 }
1814
1815 static int taprio_dump(struct Qdisc *sch, struct sk_buff *skb)
1816 {
1817         struct taprio_sched *q = qdisc_priv(sch);
1818         struct net_device *dev = qdisc_dev(sch);
1819         struct sched_gate_list *oper, *admin;
1820         struct tc_mqprio_qopt opt = { 0 };
1821         struct nlattr *nest, *sched_nest;
1822         unsigned int i;
1823
1824         rcu_read_lock();
1825         oper = rcu_dereference(q->oper_sched);
1826         admin = rcu_dereference(q->admin_sched);
1827
1828         opt.num_tc = netdev_get_num_tc(dev);
1829         memcpy(opt.prio_tc_map, dev->prio_tc_map, sizeof(opt.prio_tc_map));
1830
1831         for (i = 0; i < netdev_get_num_tc(dev); i++) {
1832                 opt.count[i] = dev->tc_to_txq[i].count;
1833                 opt.offset[i] = dev->tc_to_txq[i].offset;
1834         }
1835
1836         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1837         if (!nest)
1838                 goto start_error;
1839
1840         if (nla_put(skb, TCA_TAPRIO_ATTR_PRIOMAP, sizeof(opt), &opt))
1841                 goto options_error;
1842
1843         if (!FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1844             nla_put_s32(skb, TCA_TAPRIO_ATTR_SCHED_CLOCKID, q->clockid))
1845                 goto options_error;
1846
1847         if (q->flags && nla_put_u32(skb, TCA_TAPRIO_ATTR_FLAGS, q->flags))
1848                 goto options_error;
1849
1850         if (q->txtime_delay &&
1851             nla_put_u32(skb, TCA_TAPRIO_ATTR_TXTIME_DELAY, q->txtime_delay))
1852                 goto options_error;
1853
1854         if (oper && dump_schedule(skb, oper))
1855                 goto options_error;
1856
1857         if (!admin)
1858                 goto done;
1859
1860         sched_nest = nla_nest_start_noflag(skb, TCA_TAPRIO_ATTR_ADMIN_SCHED);
1861         if (!sched_nest)
1862                 goto options_error;
1863
1864         if (dump_schedule(skb, admin))
1865                 goto admin_error;
1866
1867         nla_nest_end(skb, sched_nest);
1868
1869 done:
1870         rcu_read_unlock();
1871
1872         return nla_nest_end(skb, nest);
1873
1874 admin_error:
1875         nla_nest_cancel(skb, sched_nest);
1876
1877 options_error:
1878         nla_nest_cancel(skb, nest);
1879
1880 start_error:
1881         rcu_read_unlock();
1882         return -ENOSPC;
1883 }
1884
1885 static struct Qdisc *taprio_leaf(struct Qdisc *sch, unsigned long cl)
1886 {
1887         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1888
1889         if (!dev_queue)
1890                 return NULL;
1891
1892         return dev_queue->qdisc_sleeping;
1893 }
1894
1895 static unsigned long taprio_find(struct Qdisc *sch, u32 classid)
1896 {
1897         unsigned int ntx = TC_H_MIN(classid);
1898
1899         if (!taprio_queue_get(sch, ntx))
1900                 return 0;
1901         return ntx;
1902 }
1903
1904 static int taprio_dump_class(struct Qdisc *sch, unsigned long cl,
1905                              struct sk_buff *skb, struct tcmsg *tcm)
1906 {
1907         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1908
1909         tcm->tcm_parent = TC_H_ROOT;
1910         tcm->tcm_handle |= TC_H_MIN(cl);
1911         tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
1912
1913         return 0;
1914 }
1915
1916 static int taprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
1917                                    struct gnet_dump *d)
1918         __releases(d->lock)
1919         __acquires(d->lock)
1920 {
1921         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1922
1923         sch = dev_queue->qdisc_sleeping;
1924         if (gnet_stats_copy_basic(&sch->running, d, NULL, &sch->bstats) < 0 ||
1925             qdisc_qstats_copy(d, sch) < 0)
1926                 return -1;
1927         return 0;
1928 }
1929
1930 static void taprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1931 {
1932         struct net_device *dev = qdisc_dev(sch);
1933         unsigned long ntx;
1934
1935         if (arg->stop)
1936                 return;
1937
1938         arg->count = arg->skip;
1939         for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
1940                 if (arg->fn(sch, ntx + 1, arg) < 0) {
1941                         arg->stop = 1;
1942                         break;
1943                 }
1944                 arg->count++;
1945         }
1946 }
1947
1948 static struct netdev_queue *taprio_select_queue(struct Qdisc *sch,
1949                                                 struct tcmsg *tcm)
1950 {
1951         return taprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
1952 }
1953
1954 static const struct Qdisc_class_ops taprio_class_ops = {
1955         .graft          = taprio_graft,
1956         .leaf           = taprio_leaf,
1957         .find           = taprio_find,
1958         .walk           = taprio_walk,
1959         .dump           = taprio_dump_class,
1960         .dump_stats     = taprio_dump_class_stats,
1961         .select_queue   = taprio_select_queue,
1962 };
1963
1964 static struct Qdisc_ops taprio_qdisc_ops __read_mostly = {
1965         .cl_ops         = &taprio_class_ops,
1966         .id             = "taprio",
1967         .priv_size      = sizeof(struct taprio_sched),
1968         .init           = taprio_init,
1969         .change         = taprio_change,
1970         .destroy        = taprio_destroy,
1971         .reset          = taprio_reset,
1972         .peek           = taprio_peek,
1973         .dequeue        = taprio_dequeue,
1974         .enqueue        = taprio_enqueue,
1975         .dump           = taprio_dump,
1976         .owner          = THIS_MODULE,
1977 };
1978
1979 static struct notifier_block taprio_device_notifier = {
1980         .notifier_call = taprio_dev_notifier,
1981 };
1982
1983 static int __init taprio_module_init(void)
1984 {
1985         int err = register_netdevice_notifier(&taprio_device_notifier);
1986
1987         if (err)
1988                 return err;
1989
1990         return register_qdisc(&taprio_qdisc_ops);
1991 }
1992
1993 static void __exit taprio_module_exit(void)
1994 {
1995         unregister_qdisc(&taprio_qdisc_ops);
1996         unregister_netdevice_notifier(&taprio_device_notifier);
1997 }
1998
1999 module_init(taprio_module_init);
2000 module_exit(taprio_module_exit);
2001 MODULE_LICENSE("GPL");