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/types.h>
  10 #include <linux/slab.h>
  11 #include <linux/kernel.h>
  12 #include <linux/string.h>
  13 #include <linux/list.h>
  14 #include <linux/errno.h>
  15 #include <linux/skbuff.h>
  16 #include <linux/math64.h>
  17 #include <linux/module.h>
  18 #include <linux/spinlock.h>
  19 #include <linux/rcupdate.h>
  20 #include <net/netlink.h>
  21 #include <net/pkt_sched.h>
  22 #include <net/pkt_cls.h>
  23 #include <net/sch_generic.h>
  24 #include <net/sock.h>
  25 #include <net/tcp.h>
  26
  27 static LIST_HEAD(taprio_list);
  28 static DEFINE_SPINLOCK(taprio_list_lock);
  29
  30 #define TAPRIO_ALL_GATES_OPEN -1
  31
  32 #define TXTIME_ASSIST_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST)
  33 #define FULL_OFFLOAD_IS_ENABLED(flags) ((flags) & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD)
  34 #define TAPRIO_FLAGS_INVALID U32_MAX
  35
  36 struct sched_entry {
  37         struct list_head list;
  38
  39         /* The instant that this entry "closes" and the next one
  40          * should open, the qdisc will make some effort so that no
  41          * packet leaves after this time.
  42          */
  43         ktime_t close_time;
  44         ktime_t next_txtime;
  45         atomic_t budget;
  46         int index;
  47         u32 gate_mask;
  48         u32 interval;
  49         u8 command;
  50 };
  51
  52 struct sched_gate_list {
  53         struct rcu_head rcu;
  54         struct list_head entries;
  55         size_t num_entries;
  56         ktime_t cycle_close_time;
  57         s64 cycle_time;
  58         s64 cycle_time_extension;
  59         s64 base_time;
  60 };
  61
  62 struct taprio_sched {
  63         struct Qdisc **qdiscs;
  64         struct Qdisc *root;
  65         u32 flags;
  66         enum tk_offsets tk_offset;
  67         int clockid;
  68         bool offloaded;
  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_mono_to_any(const struct taprio_sched *q, ktime_t mono)
  99 {
 100         /* This pairs with WRITE_ONCE() in taprio_parse_clockid() */
 101         enum tk_offsets tk_offset = READ_ONCE(q->tk_offset);
 102
 103         switch (tk_offset) {
 104         case TK_OFFS_MAX:
 105                 return mono;
 106         default:
 107                 return ktime_mono_to_any(mono, tk_offset);
 108         }
 109 }
 110
 111 static ktime_t taprio_get_time(const struct taprio_sched *q)
 112 {
 113         return taprio_mono_to_any(q, ktime_get());
 114 }
 115
 116 static void taprio_free_sched_cb(struct rcu_head *head)
 117 {
 118         struct sched_gate_list *sched = container_of(head, struct sched_gate_list, rcu);
 119         struct sched_entry *entry, *n;
 120
 121         if (!sched)
 122                 return;
 123
 124         list_for_each_entry_safe(entry, n, &sched->entries, list) {
 125                 list_del(&entry->list);
 126                 kfree(entry);
 127         }
 128
 129         kfree(sched);
 130 }
 131
 132 static void switch_schedules(struct taprio_sched *q,
 133                              struct sched_gate_list **admin,
 134                              struct sched_gate_list **oper)
 135 {
 136         rcu_assign_pointer(q->oper_sched, *admin);
 137         rcu_assign_pointer(q->admin_sched, NULL);
 138
 139         if (*oper)
 140                 call_rcu(&(*oper)->rcu, taprio_free_sched_cb);
 141
 142         *oper = *admin;
 143         *admin = NULL;
 144 }
 145
 146 /* Get how much time has been already elapsed in the current cycle. */
 147 static s32 get_cycle_time_elapsed(struct sched_gate_list *sched, ktime_t time)
 148 {
 149         ktime_t time_since_sched_start;
 150         s32 time_elapsed;
 151
 152         time_since_sched_start = ktime_sub(time, sched->base_time);
 153         div_s64_rem(time_since_sched_start, sched->cycle_time, &time_elapsed);
 154
 155         return time_elapsed;
 156 }
 157
 158 static ktime_t get_interval_end_time(struct sched_gate_list *sched,
 159                                      struct sched_gate_list *admin,
 160                                      struct sched_entry *entry,
 161                                      ktime_t intv_start)
 162 {
 163         s32 cycle_elapsed = get_cycle_time_elapsed(sched, intv_start);
 164         ktime_t intv_end, cycle_ext_end, cycle_end;
 165
 166         cycle_end = ktime_add_ns(intv_start, sched->cycle_time - cycle_elapsed);
 167         intv_end = ktime_add_ns(intv_start, entry->interval);
 168         cycle_ext_end = ktime_add(cycle_end, sched->cycle_time_extension);
 169
 170         if (ktime_before(intv_end, cycle_end))
 171                 return intv_end;
 172         else if (admin && admin != sched &&
 173                  ktime_after(admin->base_time, cycle_end) &&
 174                  ktime_before(admin->base_time, cycle_ext_end))
 175                 return admin->base_time;
 176         else
 177                 return cycle_end;
 178 }
 179
 180 static int length_to_duration(struct taprio_sched *q, int len)
 181 {
 182         return div_u64(len * atomic64_read(&q->picos_per_byte), 1000);
 183 }
 184
 185 /* Returns the entry corresponding to next available interval. If
 186  * validate_interval is set, it only validates whether the timestamp occurs
 187  * when the gate corresponding to the skb's traffic class is open.
 188  */
 189 static struct sched_entry *find_entry_to_transmit(struct sk_buff *skb,
 190                                                   struct Qdisc *sch,
 191                                                   struct sched_gate_list *sched,
 192                                                   struct sched_gate_list *admin,
 193                                                   ktime_t time,
 194                                                   ktime_t *interval_start,
 195                                                   ktime_t *interval_end,
 196                                                   bool validate_interval)
 197 {
 198         ktime_t curr_intv_start, curr_intv_end, cycle_end, packet_transmit_time;
 199         ktime_t earliest_txtime = KTIME_MAX, txtime, cycle, transmit_end_time;
 200         struct sched_entry *entry = NULL, *entry_found = NULL;
 201         struct taprio_sched *q = qdisc_priv(sch);
 202         struct net_device *dev = qdisc_dev(sch);
 203         bool entry_available = false;
 204         s32 cycle_elapsed;
 205         int tc, n;
 206
 207         tc = netdev_get_prio_tc_map(dev, skb->priority);
 208         packet_transmit_time = length_to_duration(q, qdisc_pkt_len(skb));
 209
 210         *interval_start = 0;
 211         *interval_end = 0;
 212
 213         if (!sched)
 214                 return NULL;
 215
 216         cycle = sched->cycle_time;
 217         cycle_elapsed = get_cycle_time_elapsed(sched, time);
 218         curr_intv_end = ktime_sub_ns(time, cycle_elapsed);
 219         cycle_end = ktime_add_ns(curr_intv_end, cycle);
 220
 221         list_for_each_entry(entry, &sched->entries, list) {
 222                 curr_intv_start = curr_intv_end;
 223                 curr_intv_end = get_interval_end_time(sched, admin, entry,
 224                                                       curr_intv_start);
 225
 226                 if (ktime_after(curr_intv_start, cycle_end))
 227                         break;
 228
 229                 if (!(entry->gate_mask & BIT(tc)) ||
 230                     packet_transmit_time > entry->interval)
 231                         continue;
 232
 233                 txtime = entry->next_txtime;
 234
 235                 if (ktime_before(txtime, time) || validate_interval) {
 236                         transmit_end_time = ktime_add_ns(time, packet_transmit_time);
 237                         if ((ktime_before(curr_intv_start, time) &&
 238                              ktime_before(transmit_end_time, curr_intv_end)) ||
 239                             (ktime_after(curr_intv_start, time) && !validate_interval)) {
 240                                 entry_found = entry;
 241                                 *interval_start = curr_intv_start;
 242                                 *interval_end = curr_intv_end;
 243                                 break;
 244                         } else if (!entry_available && !validate_interval) {
 245                                 /* Here, we are just trying to find out the
 246                                  * first available interval in the next cycle.
 247                                  */
 248                                 entry_available = 1;
 249                                 entry_found = entry;
 250                                 *interval_start = ktime_add_ns(curr_intv_start, cycle);
 251                                 *interval_end = ktime_add_ns(curr_intv_end, cycle);
 252                         }
 253                 } else if (ktime_before(txtime, earliest_txtime) &&
 254                            !entry_available) {
 255                         earliest_txtime = txtime;
 256                         entry_found = entry;
 257                         n = div_s64(ktime_sub(txtime, curr_intv_start), cycle);
 258                         *interval_start = ktime_add(curr_intv_start, n * cycle);
 259                         *interval_end = ktime_add(curr_intv_end, n * cycle);
 260                 }
 261         }
 262
 263         return entry_found;
 264 }
 265
 266 static bool is_valid_interval(struct sk_buff *skb, struct Qdisc *sch)
 267 {
 268         struct taprio_sched *q = qdisc_priv(sch);
 269         struct sched_gate_list *sched, *admin;
 270         ktime_t interval_start, interval_end;
 271         struct sched_entry *entry;
 272
 273         rcu_read_lock();
 274         sched = rcu_dereference(q->oper_sched);
 275         admin = rcu_dereference(q->admin_sched);
 276
 277         entry = find_entry_to_transmit(skb, sch, sched, admin, skb->tstamp,
 278                                        &interval_start, &interval_end, true);
 279         rcu_read_unlock();
 280
 281         return entry;
 282 }
 283
 284 static bool taprio_flags_valid(u32 flags)
 285 {
 286         /* Make sure no other flag bits are set. */
 287         if (flags & ~(TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST |
 288                       TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
 289                 return false;
 290         /* txtime-assist and full offload are mutually exclusive */
 291         if ((flags & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST) &&
 292             (flags & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
 293                 return false;
 294         return true;
 295 }
 296
 297 /* This returns the tstamp value set by TCP in terms of the set clock. */
 298 static ktime_t get_tcp_tstamp(struct taprio_sched *q, struct sk_buff *skb)
 299 {
 300         unsigned int offset = skb_network_offset(skb);
 301         const struct ipv6hdr *ipv6h;
 302         const struct iphdr *iph;
 303         struct ipv6hdr _ipv6h;
 304
 305         ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h);
 306         if (!ipv6h)
 307                 return 0;
 308
 309         if (ipv6h->version == 4) {
 310                 iph = (struct iphdr *)ipv6h;
 311                 offset += iph->ihl * 4;
 312
 313                 /* special-case 6in4 tunnelling, as that is a common way to get
 314                  * v6 connectivity in the home
 315                  */
 316                 if (iph->protocol == IPPROTO_IPV6) {
 317                         ipv6h = skb_header_pointer(skb, offset,
 318                                                    sizeof(_ipv6h), &_ipv6h);
 319
 320                         if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP)
 321                                 return 0;
 322                 } else if (iph->protocol != IPPROTO_TCP) {
 323                         return 0;
 324                 }
 325         } else if (ipv6h->version == 6 && ipv6h->nexthdr != IPPROTO_TCP) {
 326                 return 0;
 327         }
 328
 329         return taprio_mono_to_any(q, skb->skb_mstamp_ns);
 330 }
 331
 332 /* There are a few scenarios where we will have to modify the txtime from
 333  * what is read from next_txtime in sched_entry. They are:
 334  * 1. If txtime is in the past,
 335  *    a. The gate for the traffic class is currently open and packet can be
 336  *       transmitted before it closes, schedule the packet right away.
 337  *    b. If the gate corresponding to the traffic class is going to open later
 338  *       in the cycle, set the txtime of packet to the interval start.
 339  * 2. If txtime is in the future, there are packets corresponding to the
 340  *    current traffic class waiting to be transmitted. So, the following
 341  *    possibilities exist:
 342  *    a. We can transmit the packet before the window containing the txtime
 343  *       closes.
 344  *    b. The window might close before the transmission can be completed
 345  *       successfully. So, schedule the packet in the next open window.
 346  */
 347 static long get_packet_txtime(struct sk_buff *skb, struct Qdisc *sch)
 348 {
 349         ktime_t transmit_end_time, interval_end, interval_start, tcp_tstamp;
 350         struct taprio_sched *q = qdisc_priv(sch);
 351         struct sched_gate_list *sched, *admin;
 352         ktime_t minimum_time, now, txtime;
 353         int len, packet_transmit_time;
 354         struct sched_entry *entry;
 355         bool sched_changed;
 356
 357         now = taprio_get_time(q);
 358         minimum_time = ktime_add_ns(now, q->txtime_delay);
 359
 360         tcp_tstamp = get_tcp_tstamp(q, skb);
 361         minimum_time = max_t(ktime_t, minimum_time, tcp_tstamp);
 362
 363         rcu_read_lock();
 364         admin = rcu_dereference(q->admin_sched);
 365         sched = rcu_dereference(q->oper_sched);
 366         if (admin && ktime_after(minimum_time, admin->base_time))
 367                 switch_schedules(q, &admin, &sched);
 368
 369         /* Until the schedule starts, all the queues are open */
 370         if (!sched || ktime_before(minimum_time, sched->base_time)) {
 371                 txtime = minimum_time;
 372                 goto done;
 373         }
 374
 375         len = qdisc_pkt_len(skb);
 376         packet_transmit_time = length_to_duration(q, len);
 377
 378         do {
 379                 sched_changed = 0;
 380
 381                 entry = find_entry_to_transmit(skb, sch, sched, admin,
 382                                                minimum_time,
 383                                                &interval_start, &interval_end,
 384                                                false);
 385                 if (!entry) {
 386                         txtime = 0;
 387                         goto done;
 388                 }
 389
 390                 txtime = entry->next_txtime;
 391                 txtime = max_t(ktime_t, txtime, minimum_time);
 392                 txtime = max_t(ktime_t, txtime, interval_start);
 393
 394                 if (admin && admin != sched &&
 395                     ktime_after(txtime, admin->base_time)) {
 396                         sched = admin;
 397                         sched_changed = 1;
 398                         continue;
 399                 }
 400
 401                 transmit_end_time = ktime_add(txtime, packet_transmit_time);
 402                 minimum_time = transmit_end_time;
 403
 404                 /* Update the txtime of current entry to the next time it's
 405                  * interval starts.
 406                  */
 407                 if (ktime_after(transmit_end_time, interval_end))
 408                         entry->next_txtime = ktime_add(interval_start, sched->cycle_time);
 409         } while (sched_changed || ktime_after(transmit_end_time, interval_end));
 410
 411         entry->next_txtime = transmit_end_time;
 412
 413 done:
 414         rcu_read_unlock();
 415         return txtime;
 416 }
 417
 418 static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 419                           struct sk_buff **to_free)
 420 {
 421         struct taprio_sched *q = qdisc_priv(sch);
 422         struct Qdisc *child;
 423         int queue;
 424
 425         queue = skb_get_queue_mapping(skb);
 426
 427         child = q->qdiscs[queue];
 428         if (unlikely(!child))
 429                 return qdisc_drop(skb, sch, to_free);
 430
 431         /* sk_flags are only safe to use on full sockets. */
 432         if (skb->sk && sk_fullsock(skb->sk) && sock_flag(skb->sk, SOCK_TXTIME)) {
 433                 if (!is_valid_interval(skb, sch))
 434                         return qdisc_drop(skb, sch, to_free);
 435         } else if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
 436                 skb->tstamp = get_packet_txtime(skb, sch);
 437                 if (!skb->tstamp)
 438                         return qdisc_drop(skb, sch, to_free);
 439         }
 440
 441         qdisc_qstats_backlog_inc(sch, skb);
 442         sch->q.qlen++;
 443
 444         return qdisc_enqueue(skb, child, to_free);
 445 }
 446
 447 static struct sk_buff *taprio_peek_soft(struct Qdisc *sch)
 448 {
 449         struct taprio_sched *q = qdisc_priv(sch);
 450         struct net_device *dev = qdisc_dev(sch);
 451         struct sched_entry *entry;
 452         struct sk_buff *skb;
 453         u32 gate_mask;
 454         int i;
 455
 456         rcu_read_lock();
 457         entry = rcu_dereference(q->current_entry);
 458         gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
 459         rcu_read_unlock();
 460
 461         if (!gate_mask)
 462                 return NULL;
 463
 464         for (i = 0; i < dev->num_tx_queues; i++) {
 465                 struct Qdisc *child = q->qdiscs[i];
 466                 int prio;
 467                 u8 tc;
 468
 469                 if (unlikely(!child))
 470                         continue;
 471
 472                 skb = child->ops->peek(child);
 473                 if (!skb)
 474                         continue;
 475
 476                 if (TXTIME_ASSIST_IS_ENABLED(q->flags))
 477                         return skb;
 478
 479                 prio = skb->priority;
 480                 tc = netdev_get_prio_tc_map(dev, prio);
 481
 482                 if (!(gate_mask & BIT(tc)))
 483                         continue;
 484
 485                 return skb;
 486         }
 487
 488         return NULL;
 489 }
 490
 491 static struct sk_buff *taprio_peek_offload(struct Qdisc *sch)
 492 {
 493         struct taprio_sched *q = qdisc_priv(sch);
 494         struct net_device *dev = qdisc_dev(sch);
 495         struct sk_buff *skb;
 496         int i;
 497
 498         for (i = 0; i < dev->num_tx_queues; i++) {
 499                 struct Qdisc *child = q->qdiscs[i];
 500
 501                 if (unlikely(!child))
 502                         continue;
 503
 504                 skb = child->ops->peek(child);
 505                 if (!skb)
 506                         continue;
 507
 508                 return skb;
 509         }
 510
 511         return NULL;
 512 }
 513
 514 static struct sk_buff *taprio_peek(struct Qdisc *sch)
 515 {
 516         struct taprio_sched *q = qdisc_priv(sch);
 517
 518         return q->peek(sch);
 519 }
 520
 521 static void taprio_set_budget(struct taprio_sched *q, struct sched_entry *entry)
 522 {
 523         atomic_set(&entry->budget,
 524                    div64_u64((u64)entry->interval * 1000,
 525                              atomic64_read(&q->picos_per_byte)));
 526 }
 527
 528 static struct sk_buff *taprio_dequeue_soft(struct Qdisc *sch)
 529 {
 530         struct taprio_sched *q = qdisc_priv(sch);
 531         struct net_device *dev = qdisc_dev(sch);
 532         struct sk_buff *skb = NULL;
 533         struct sched_entry *entry;
 534         u32 gate_mask;
 535         int i;
 536
 537         rcu_read_lock();
 538         entry = rcu_dereference(q->current_entry);
 539         /* if there's no entry, it means that the schedule didn't
 540          * start yet, so force all gates to be open, this is in
 541          * accordance to IEEE 802.1Qbv-2015 Section 8.6.9.4.5
 542          * "AdminGateSates"
 543          */
 544         gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
 545
 546         if (!gate_mask)
 547                 goto done;
 548
 549         for (i = 0; i < dev->num_tx_queues; i++) {
 550                 struct Qdisc *child = q->qdiscs[i];
 551                 ktime_t guard;
 552                 int prio;
 553                 int len;
 554                 u8 tc;
 555
 556                 if (unlikely(!child))
 557                         continue;
 558
 559                 if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
 560                         skb = child->ops->dequeue(child);
 561                         if (!skb)
 562                                 continue;
 563                         goto skb_found;
 564                 }
 565
 566                 skb = child->ops->peek(child);
 567                 if (!skb)
 568                         continue;
 569
 570                 prio = skb->priority;
 571                 tc = netdev_get_prio_tc_map(dev, prio);
 572
 573                 if (!(gate_mask & BIT(tc))) {
 574                         skb = NULL;
 575                         continue;
 576                 }
 577
 578                 len = qdisc_pkt_len(skb);
 579                 guard = ktime_add_ns(taprio_get_time(q),
 580                                      length_to_duration(q, len));
 581
 582                 /* In the case that there's no gate entry, there's no
 583                  * guard band ...
 584                  */
 585                 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
 586                     ktime_after(guard, entry->close_time)) {
 587                         skb = NULL;
 588                         continue;
 589                 }
 590
 591                 /* ... and no budget. */
 592                 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
 593                     atomic_sub_return(len, &entry->budget) < 0) {
 594                         skb = NULL;
 595                         continue;
 596                 }
 597
 598                 skb = child->ops->dequeue(child);
 599                 if (unlikely(!skb))
 600                         goto done;
 601
 602 skb_found:
 603                 qdisc_bstats_update(sch, skb);
 604                 qdisc_qstats_backlog_dec(sch, skb);
 605                 sch->q.qlen--;
 606
 607                 goto done;
 608         }
 609
 610 done:
 611         rcu_read_unlock();
 612
 613         return skb;
 614 }
 615
 616 static struct sk_buff *taprio_dequeue_offload(struct Qdisc *sch)
 617 {
 618         struct taprio_sched *q = qdisc_priv(sch);
 619         struct net_device *dev = qdisc_dev(sch);
 620         struct sk_buff *skb;
 621         int i;
 622
 623         for (i = 0; i < dev->num_tx_queues; i++) {
 624                 struct Qdisc *child = q->qdiscs[i];
 625
 626                 if (unlikely(!child))
 627                         continue;
 628
 629                 skb = child->ops->dequeue(child);
 630                 if (unlikely(!skb))
 631                         continue;
 632
 633                 qdisc_bstats_update(sch, skb);
 634                 qdisc_qstats_backlog_dec(sch, skb);
 635                 sch->q.qlen--;
 636
 637                 return skb;
 638         }
 639
 640         return NULL;
 641 }
 642
 643 static struct sk_buff *taprio_dequeue(struct Qdisc *sch)
 644 {
 645         struct taprio_sched *q = qdisc_priv(sch);
 646
 647         return q->dequeue(sch);
 648 }
 649
 650 static bool should_restart_cycle(const struct sched_gate_list *oper,
 651                                  const struct sched_entry *entry)
 652 {
 653         if (list_is_last(&entry->list, &oper->entries))
 654                 return true;
 655
 656         if (ktime_compare(entry->close_time, oper->cycle_close_time) == 0)
 657                 return true;
 658
 659         return false;
 660 }
 661
 662 static bool should_change_schedules(const struct sched_gate_list *admin,
 663                                     const struct sched_gate_list *oper,
 664                                     ktime_t close_time)
 665 {
 666         ktime_t next_base_time, extension_time;
 667
 668         if (!admin)
 669                 return false;
 670
 671         next_base_time = sched_base_time(admin);
 672
 673         /* This is the simple case, the close_time would fall after
 674          * the next schedule base_time.
 675          */
 676         if (ktime_compare(next_base_time, close_time) <= 0)
 677                 return true;
 678
 679         /* This is the cycle_time_extension case, if the close_time
 680          * plus the amount that can be extended would fall after the
 681          * next schedule base_time, we can extend the current schedule
 682          * for that amount.
 683          */
 684         extension_time = ktime_add_ns(close_time, oper->cycle_time_extension);
 685
 686         /* FIXME: the IEEE 802.1Q-2018 Specification isn't clear about
 687          * how precisely the extension should be made. So after
 688          * conformance testing, this logic may change.
 689          */
 690         if (ktime_compare(next_base_time, extension_time) <= 0)
 691                 return true;
 692
 693         return false;
 694 }
 695
 696 static enum hrtimer_restart advance_sched(struct hrtimer *timer)
 697 {
 698         struct taprio_sched *q = container_of(timer, struct taprio_sched,
 699                                               advance_timer);
 700         struct sched_gate_list *oper, *admin;
 701         struct sched_entry *entry, *next;
 702         struct Qdisc *sch = q->root;
 703         ktime_t close_time;
 704
 705         spin_lock(&q->current_entry_lock);
 706         entry = rcu_dereference_protected(q->current_entry,
 707                                           lockdep_is_held(&q->current_entry_lock));
 708         oper = rcu_dereference_protected(q->oper_sched,
 709                                          lockdep_is_held(&q->current_entry_lock));
 710         admin = rcu_dereference_protected(q->admin_sched,
 711                                           lockdep_is_held(&q->current_entry_lock));
 712
 713         if (!oper)
 714                 switch_schedules(q, &admin, &oper);
 715
 716         /* This can happen in two cases: 1. this is the very first run
 717          * of this function (i.e. we weren't running any schedule
 718          * previously); 2. The previous schedule just ended. The first
 719          * entry of all schedules are pre-calculated during the
 720          * schedule initialization.
 721          */
 722         if (unlikely(!entry || entry->close_time == oper->base_time)) {
 723                 next = list_first_entry(&oper->entries, struct sched_entry,
 724                                         list);
 725                 close_time = next->close_time;
 726                 goto first_run;
 727         }
 728
 729         if (should_restart_cycle(oper, entry)) {
 730                 next = list_first_entry(&oper->entries, struct sched_entry,
 731                                         list);
 732                 oper->cycle_close_time = ktime_add_ns(oper->cycle_close_time,
 733                                                       oper->cycle_time);
 734         } else {
 735                 next = list_next_entry(entry, list);
 736         }
 737
 738         close_time = ktime_add_ns(entry->close_time, next->interval);
 739         close_time = min_t(ktime_t, close_time, oper->cycle_close_time);
 740
 741         if (should_change_schedules(admin, oper, close_time)) {
 742                 /* Set things so the next time this runs, the new
 743                  * schedule runs.
 744                  */
 745                 close_time = sched_base_time(admin);
 746                 switch_schedules(q, &admin, &oper);
 747         }
 748
 749         next->close_time = close_time;
 750         taprio_set_budget(q, next);
 751
 752 first_run:
 753         rcu_assign_pointer(q->current_entry, next);
 754         spin_unlock(&q->current_entry_lock);
 755
 756         hrtimer_set_expires(&q->advance_timer, close_time);
 757
 758         rcu_read_lock();
 759         __netif_schedule(sch);
 760         rcu_read_unlock();
 761
 762         return HRTIMER_RESTART;
 763 }
 764
 765 static const struct nla_policy entry_policy[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = {
 766         [TCA_TAPRIO_SCHED_ENTRY_INDEX]     = { .type = NLA_U32 },
 767         [TCA_TAPRIO_SCHED_ENTRY_CMD]       = { .type = NLA_U8 },
 768         [TCA_TAPRIO_SCHED_ENTRY_GATE_MASK] = { .type = NLA_U32 },
 769         [TCA_TAPRIO_SCHED_ENTRY_INTERVAL]  = { .type = NLA_U32 },
 770 };
 771
 772 static const struct nla_policy taprio_policy[TCA_TAPRIO_ATTR_MAX + 1] = {
 773         [TCA_TAPRIO_ATTR_PRIOMAP]              = {
 774                 .len = sizeof(struct tc_mqprio_qopt)
 775         },
 776         [TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST]           = { .type = NLA_NESTED },
 777         [TCA_TAPRIO_ATTR_SCHED_BASE_TIME]            = { .type = NLA_S64 },
 778         [TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]         = { .type = NLA_NESTED },
 779         [TCA_TAPRIO_ATTR_SCHED_CLOCKID]              = { .type = NLA_S32 },
 780         [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]           = { .type = NLA_S64 },
 781         [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION] = { .type = NLA_S64 },
 782         [TCA_TAPRIO_ATTR_FLAGS]                      = { .type = NLA_U32 },
 783         [TCA_TAPRIO_ATTR_TXTIME_DELAY]               = { .type = NLA_U32 },
 784 };
 785
 786 static int fill_sched_entry(struct taprio_sched *q, struct nlattr **tb,
 787                             struct sched_entry *entry,
 788                             struct netlink_ext_ack *extack)
 789 {
 790         int min_duration = length_to_duration(q, ETH_ZLEN);
 791         u32 interval = 0;
 792
 793         if (tb[TCA_TAPRIO_SCHED_ENTRY_CMD])
 794                 entry->command = nla_get_u8(
 795                         tb[TCA_TAPRIO_SCHED_ENTRY_CMD]);
 796
 797         if (tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK])
 798                 entry->gate_mask = nla_get_u32(
 799                         tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK]);
 800
 801         if (tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL])
 802                 interval = nla_get_u32(
 803                         tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL]);
 804
 805         /* The interval should allow at least the minimum ethernet
 806          * frame to go out.
 807          */
 808         if (interval < min_duration) {
 809                 NL_SET_ERR_MSG(extack, "Invalid interval for schedule entry");
 810                 return -EINVAL;
 811         }
 812
 813         entry->interval = interval;
 814
 815         return 0;
 816 }
 817
 818 static int parse_sched_entry(struct taprio_sched *q, struct nlattr *n,
 819                              struct sched_entry *entry, int index,
 820                              struct netlink_ext_ack *extack)
 821 {
 822         struct nlattr *tb[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = { };
 823         int err;
 824
 825         err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_SCHED_ENTRY_MAX, n,
 826                                           entry_policy, NULL);
 827         if (err < 0) {
 828                 NL_SET_ERR_MSG(extack, "Could not parse nested entry");
 829                 return -EINVAL;
 830         }
 831
 832         entry->index = index;
 833
 834         return fill_sched_entry(q, tb, entry, extack);
 835 }
 836
 837 static int parse_sched_list(struct taprio_sched *q, struct nlattr *list,
 838                             struct sched_gate_list *sched,
 839                             struct netlink_ext_ack *extack)
 840 {
 841         struct nlattr *n;
 842         int err, rem;
 843         int i = 0;
 844
 845         if (!list)
 846                 return -EINVAL;
 847
 848         nla_for_each_nested(n, list, rem) {
 849                 struct sched_entry *entry;
 850
 851                 if (nla_type(n) != TCA_TAPRIO_SCHED_ENTRY) {
 852                         NL_SET_ERR_MSG(extack, "Attribute is not of type 'entry'");
 853                         continue;
 854                 }
 855
 856                 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 857                 if (!entry) {
 858                         NL_SET_ERR_MSG(extack, "Not enough memory for entry");
 859                         return -ENOMEM;
 860                 }
 861
 862                 err = parse_sched_entry(q, n, entry, i, extack);
 863                 if (err < 0) {
 864                         kfree(entry);
 865                         return err;
 866                 }
 867
 868                 list_add_tail(&entry->list, &sched->entries);
 869                 i++;
 870         }
 871
 872         sched->num_entries = i;
 873
 874         return i;
 875 }
 876
 877 static int parse_taprio_schedule(struct taprio_sched *q, struct nlattr **tb,
 878                                  struct sched_gate_list *new,
 879                                  struct netlink_ext_ack *extack)
 880 {
 881         int err = 0;
 882
 883         if (tb[TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]) {
 884                 NL_SET_ERR_MSG(extack, "Adding a single entry is not supported");
 885                 return -ENOTSUPP;
 886         }
 887
 888         if (tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME])
 889                 new->base_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME]);
 890
 891         if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION])
 892                 new->cycle_time_extension = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION]);
 893
 894         if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME])
 895                 new->cycle_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]);
 896
 897         if (tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST])
 898                 err = parse_sched_list(q, tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST],
 899                                        new, extack);
 900         if (err < 0)
 901                 return err;
 902
 903         if (!new->cycle_time) {
 904                 struct sched_entry *entry;
 905                 ktime_t cycle = 0;
 906
 907                 list_for_each_entry(entry, &new->entries, list)
 908                         cycle = ktime_add_ns(cycle, entry->interval);
 909
 910                 if (!cycle) {
 911                         NL_SET_ERR_MSG(extack, "'cycle_time' can never be 0");
 912                         return -EINVAL;
 913                 }
 914
 915                 new->cycle_time = cycle;
 916         }
 917
 918         return 0;
 919 }
 920
 921 static int taprio_parse_mqprio_opt(struct net_device *dev,
 922                                    struct tc_mqprio_qopt *qopt,
 923                                    struct netlink_ext_ack *extack,
 924                                    u32 taprio_flags)
 925 {
 926         int i, j;
 927
 928         if (!qopt && !dev->num_tc) {
 929                 NL_SET_ERR_MSG(extack, "'mqprio' configuration is necessary");
 930                 return -EINVAL;
 931         }
 932
 933         /* If num_tc is already set, it means that the user already
 934          * configured the mqprio part
 935          */
 936         if (dev->num_tc)
 937                 return 0;
 938
 939         /* Verify num_tc is not out of max range */
 940         if (qopt->num_tc > TC_MAX_QUEUE) {
 941                 NL_SET_ERR_MSG(extack, "Number of traffic classes is outside valid range");
 942                 return -EINVAL;
 943         }
 944
 945         /* taprio imposes that traffic classes map 1:n to tx queues */
 946         if (qopt->num_tc > dev->num_tx_queues) {
 947                 NL_SET_ERR_MSG(extack, "Number of traffic classes is greater than number of HW queues");
 948                 return -EINVAL;
 949         }
 950
 951         /* Verify priority mapping uses valid tcs */
 952         for (i = 0; i <= TC_BITMASK; i++) {
 953                 if (qopt->prio_tc_map[i] >= qopt->num_tc) {
 954                         NL_SET_ERR_MSG(extack, "Invalid traffic class in priority to traffic class mapping");
 955                         return -EINVAL;
 956                 }
 957         }
 958
 959         for (i = 0; i < qopt->num_tc; i++) {
 960                 unsigned int last = qopt->offset[i] + qopt->count[i];
 961
 962                 /* Verify the queue count is in tx range being equal to the
 963                  * real_num_tx_queues indicates the last queue is in use.
 964                  */
 965                 if (qopt->offset[i] >= dev->num_tx_queues ||
 966                     !qopt->count[i] ||
 967                     last > dev->real_num_tx_queues) {
 968                         NL_SET_ERR_MSG(extack, "Invalid queue in traffic class to queue mapping");
 969                         return -EINVAL;
 970                 }
 971
 972                 if (TXTIME_ASSIST_IS_ENABLED(taprio_flags))
 973                         continue;
 974
 975                 /* Verify that the offset and counts do not overlap */
 976                 for (j = i + 1; j < qopt->num_tc; j++) {
 977                         if (last > qopt->offset[j]) {
 978                                 NL_SET_ERR_MSG(extack, "Detected overlap in the traffic class to queue mapping");
 979                                 return -EINVAL;
 980                         }
 981                 }
 982         }
 983
 984         return 0;
 985 }
 986
 987 static int taprio_get_start_time(struct Qdisc *sch,
 988                                  struct sched_gate_list *sched,
 989                                  ktime_t *start)
 990 {
 991         struct taprio_sched *q = qdisc_priv(sch);
 992         ktime_t now, base, cycle;
 993         s64 n;
 994
 995         base = sched_base_time(sched);
 996         now = taprio_get_time(q);
 997
 998         if (ktime_after(base, now)) {
 999                 *start = base;
1000                 return 0;
1001         }
1002
1003         cycle = sched->cycle_time;
1004
1005         /* The qdisc is expected to have at least one sched_entry.  Moreover,
1006          * any entry must have 'interval' > 0. Thus if the cycle time is zero,
1007          * something went really wrong. In that case, we should warn about this
1008          * inconsistent state and return error.
1009          */
1010         if (WARN_ON(!cycle))
1011                 return -EFAULT;
1012
1013         /* Schedule the start time for the beginning of the next
1014          * cycle.
1015          */
1016         n = div64_s64(ktime_sub_ns(now, base), cycle);
1017         *start = ktime_add_ns(base, (n + 1) * cycle);
1018         return 0;
1019 }
1020
1021 static void setup_first_close_time(struct taprio_sched *q,
1022                                    struct sched_gate_list *sched, ktime_t base)
1023 {
1024         struct sched_entry *first;
1025         ktime_t cycle;
1026
1027         first = list_first_entry(&sched->entries,
1028                                  struct sched_entry, list);
1029
1030         cycle = sched->cycle_time;
1031
1032         /* FIXME: find a better place to do this */
1033         sched->cycle_close_time = ktime_add_ns(base, cycle);
1034
1035         first->close_time = ktime_add_ns(base, first->interval);
1036         taprio_set_budget(q, first);
1037         rcu_assign_pointer(q->current_entry, NULL);
1038 }
1039
1040 static void taprio_start_sched(struct Qdisc *sch,
1041                                ktime_t start, struct sched_gate_list *new)
1042 {
1043         struct taprio_sched *q = qdisc_priv(sch);
1044         ktime_t expires;
1045
1046         if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1047                 return;
1048
1049         expires = hrtimer_get_expires(&q->advance_timer);
1050         if (expires == 0)
1051                 expires = KTIME_MAX;
1052
1053         /* If the new schedule starts before the next expiration, we
1054          * reprogram it to the earliest one, so we change the admin
1055          * schedule to the operational one at the right time.
1056          */
1057         start = min_t(ktime_t, start, expires);
1058
1059         hrtimer_start(&q->advance_timer, start, HRTIMER_MODE_ABS);
1060 }
1061
1062 static void taprio_set_picos_per_byte(struct net_device *dev,
1063                                       struct taprio_sched *q)
1064 {
1065         struct ethtool_link_ksettings ecmd;
1066         int speed = SPEED_10;
1067         int picos_per_byte;
1068         int err;
1069
1070         err = __ethtool_get_link_ksettings(dev, &ecmd);
1071         if (err < 0)
1072                 goto skip;
1073
1074         if (ecmd.base.speed && ecmd.base.speed != SPEED_UNKNOWN)
1075                 speed = ecmd.base.speed;
1076
1077 skip:
1078         picos_per_byte = (USEC_PER_SEC * 8) / speed;
1079
1080         atomic64_set(&q->picos_per_byte, picos_per_byte);
1081         netdev_dbg(dev, "taprio: set %s's picos_per_byte to: %lld, linkspeed: %d\n",
1082                    dev->name, (long long)atomic64_read(&q->picos_per_byte),
1083                    ecmd.base.speed);
1084 }
1085
1086 static int taprio_dev_notifier(struct notifier_block *nb, unsigned long event,
1087                                void *ptr)
1088 {
1089         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1090         struct net_device *qdev;
1091         struct taprio_sched *q;
1092         bool found = false;
1093
1094         ASSERT_RTNL();
1095
1096         if (event != NETDEV_UP && event != NETDEV_CHANGE)
1097                 return NOTIFY_DONE;
1098
1099         spin_lock(&taprio_list_lock);
1100         list_for_each_entry(q, &taprio_list, taprio_list) {
1101                 qdev = qdisc_dev(q->root);
1102                 if (qdev == dev) {
1103                         found = true;
1104                         break;
1105                 }
1106         }
1107         spin_unlock(&taprio_list_lock);
1108
1109         if (found)
1110                 taprio_set_picos_per_byte(dev, q);
1111
1112         return NOTIFY_DONE;
1113 }
1114
1115 static void setup_txtime(struct taprio_sched *q,
1116                          struct sched_gate_list *sched, ktime_t base)
1117 {
1118         struct sched_entry *entry;
1119         u32 interval = 0;
1120
1121         list_for_each_entry(entry, &sched->entries, list) {
1122                 entry->next_txtime = ktime_add_ns(base, interval);
1123                 interval += entry->interval;
1124         }
1125 }
1126
1127 static struct tc_taprio_qopt_offload *taprio_offload_alloc(int num_entries)
1128 {
1129         size_t size = sizeof(struct tc_taprio_sched_entry) * num_entries +
1130                       sizeof(struct __tc_taprio_qopt_offload);
1131         struct __tc_taprio_qopt_offload *__offload;
1132
1133         __offload = kzalloc(size, GFP_KERNEL);
1134         if (!__offload)
1135                 return NULL;
1136
1137         refcount_set(&__offload->users, 1);
1138
1139         return &__offload->offload;
1140 }
1141
1142 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
1143                                                   *offload)
1144 {
1145         struct __tc_taprio_qopt_offload *__offload;
1146
1147         __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1148                                  offload);
1149
1150         refcount_inc(&__offload->users);
1151
1152         return offload;
1153 }
1154 EXPORT_SYMBOL_GPL(taprio_offload_get);
1155
1156 void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
1157 {
1158         struct __tc_taprio_qopt_offload *__offload;
1159
1160         __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1161                                  offload);
1162
1163         if (!refcount_dec_and_test(&__offload->users))
1164                 return;
1165
1166         kfree(__offload);
1167 }
1168 EXPORT_SYMBOL_GPL(taprio_offload_free);
1169
1170 /* The function will only serve to keep the pointers to the "oper" and "admin"
1171  * schedules valid in relation to their base times, so when calling dump() the
1172  * users looks at the right schedules.
1173  * When using full offload, the admin configuration is promoted to oper at the
1174  * base_time in the PHC time domain.  But because the system time is not
1175  * necessarily in sync with that, we can't just trigger a hrtimer to call
1176  * switch_schedules at the right hardware time.
1177  * At the moment we call this by hand right away from taprio, but in the future
1178  * it will be useful to create a mechanism for drivers to notify taprio of the
1179  * offload state (PENDING, ACTIVE, INACTIVE) so it can be visible in dump().
1180  * This is left as TODO.
1181  */
1182 static void taprio_offload_config_changed(struct taprio_sched *q)
1183 {
1184         struct sched_gate_list *oper, *admin;
1185
1186         spin_lock(&q->current_entry_lock);
1187
1188         oper = rcu_dereference_protected(q->oper_sched,
1189                                          lockdep_is_held(&q->current_entry_lock));
1190         admin = rcu_dereference_protected(q->admin_sched,
1191                                           lockdep_is_held(&q->current_entry_lock));
1192
1193         switch_schedules(q, &admin, &oper);
1194
1195         spin_unlock(&q->current_entry_lock);
1196 }
1197
1198 static u32 tc_map_to_queue_mask(struct net_device *dev, u32 tc_mask)
1199 {
1200         u32 i, queue_mask = 0;
1201
1202         for (i = 0; i < dev->num_tc; i++) {
1203                 u32 offset, count;
1204
1205                 if (!(tc_mask & BIT(i)))
1206                         continue;
1207
1208                 offset = dev->tc_to_txq[i].offset;
1209                 count = dev->tc_to_txq[i].count;
1210
1211                 queue_mask |= GENMASK(offset + count - 1, offset);
1212         }
1213
1214         return queue_mask;
1215 }
1216
1217 static void taprio_sched_to_offload(struct net_device *dev,
1218                                     struct sched_gate_list *sched,
1219                                     struct tc_taprio_qopt_offload *offload)
1220 {
1221         struct sched_entry *entry;
1222         int i = 0;
1223
1224         offload->base_time = sched->base_time;
1225         offload->cycle_time = sched->cycle_time;
1226         offload->cycle_time_extension = sched->cycle_time_extension;
1227
1228         list_for_each_entry(entry, &sched->entries, list) {
1229                 struct tc_taprio_sched_entry *e = &offload->entries[i];
1230
1231                 e->command = entry->command;
1232                 e->interval = entry->interval;
1233                 e->gate_mask = tc_map_to_queue_mask(dev, entry->gate_mask);
1234
1235                 i++;
1236         }
1237
1238         offload->num_entries = i;
1239 }
1240
1241 static int taprio_enable_offload(struct net_device *dev,
1242                                  struct taprio_sched *q,
1243                                  struct sched_gate_list *sched,
1244                                  struct netlink_ext_ack *extack)
1245 {
1246         const struct net_device_ops *ops = dev->netdev_ops;
1247         struct tc_taprio_qopt_offload *offload;
1248         int err = 0;
1249
1250         if (!ops->ndo_setup_tc) {
1251                 NL_SET_ERR_MSG(extack,
1252                                "Device does not support taprio offload");
1253                 return -EOPNOTSUPP;
1254         }
1255
1256         offload = taprio_offload_alloc(sched->num_entries);
1257         if (!offload) {
1258                 NL_SET_ERR_MSG(extack,
1259                                "Not enough memory for enabling offload mode");
1260                 return -ENOMEM;
1261         }
1262         offload->enable = 1;
1263         taprio_sched_to_offload(dev, sched, offload);
1264
1265         err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1266         if (err < 0) {
1267                 NL_SET_ERR_MSG(extack,
1268                                "Device failed to setup taprio offload");
1269                 goto done;
1270         }
1271
1272         q->offloaded = true;
1273
1274 done:
1275         taprio_offload_free(offload);
1276
1277         return err;
1278 }
1279
1280 static int taprio_disable_offload(struct net_device *dev,
1281                                   struct taprio_sched *q,
1282                                   struct netlink_ext_ack *extack)
1283 {
1284         const struct net_device_ops *ops = dev->netdev_ops;
1285         struct tc_taprio_qopt_offload *offload;
1286         int err;
1287
1288         if (!q->offloaded)
1289                 return 0;
1290
1291         offload = taprio_offload_alloc(0);
1292         if (!offload) {
1293                 NL_SET_ERR_MSG(extack,
1294                                "Not enough memory to disable offload mode");
1295                 return -ENOMEM;
1296         }
1297         offload->enable = 0;
1298
1299         err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1300         if (err < 0) {
1301                 NL_SET_ERR_MSG(extack,
1302                                "Device failed to disable offload");
1303                 goto out;
1304         }
1305
1306         q->offloaded = false;
1307
1308 out:
1309         taprio_offload_free(offload);
1310
1311         return err;
1312 }
1313
1314 /* If full offload is enabled, the only possible clockid is the net device's
1315  * PHC. For that reason, specifying a clockid through netlink is incorrect.
1316  * For txtime-assist, it is implicitly assumed that the device's PHC is kept
1317  * in sync with the specified clockid via a user space daemon such as phc2sys.
1318  * For both software taprio and txtime-assist, the clockid is used for the
1319  * hrtimer that advances the schedule and hence mandatory.
1320  */
1321 static int taprio_parse_clockid(struct Qdisc *sch, struct nlattr **tb,
1322                                 struct netlink_ext_ack *extack)
1323 {
1324         struct taprio_sched *q = qdisc_priv(sch);
1325         struct net_device *dev = qdisc_dev(sch);
1326         int err = -EINVAL;
1327
1328         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1329                 const struct ethtool_ops *ops = dev->ethtool_ops;
1330                 struct ethtool_ts_info info = {
1331                         .cmd = ETHTOOL_GET_TS_INFO,
1332                         .phc_index = -1,
1333                 };
1334
1335                 if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1336                         NL_SET_ERR_MSG(extack,
1337                                        "The 'clockid' cannot be specified for full offload");
1338                         goto out;
1339                 }
1340
1341                 if (ops && ops->get_ts_info)
1342                         err = ops->get_ts_info(dev, &info);
1343
1344                 if (err || info.phc_index < 0) {
1345                         NL_SET_ERR_MSG(extack,
1346                                        "Device does not have a PTP clock");
1347                         err = -ENOTSUPP;
1348                         goto out;
1349                 }
1350         } else if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1351                 int clockid = nla_get_s32(tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]);
1352                 enum tk_offsets tk_offset;
1353
1354                 /* We only support static clockids and we don't allow
1355                  * for it to be modified after the first init.
1356                  */
1357                 if (clockid < 0 ||
1358                     (q->clockid != -1 && q->clockid != clockid)) {
1359                         NL_SET_ERR_MSG(extack,
1360                                        "Changing the 'clockid' of a running schedule is not supported");
1361                         err = -ENOTSUPP;
1362                         goto out;
1363                 }
1364
1365                 switch (clockid) {
1366                 case CLOCK_REALTIME:
1367                         tk_offset = TK_OFFS_REAL;
1368                         break;
1369                 case CLOCK_MONOTONIC:
1370                         tk_offset = TK_OFFS_MAX;
1371                         break;
1372                 case CLOCK_BOOTTIME:
1373                         tk_offset = TK_OFFS_BOOT;
1374                         break;
1375                 case CLOCK_TAI:
1376                         tk_offset = TK_OFFS_TAI;
1377                         break;
1378                 default:
1379                         NL_SET_ERR_MSG(extack, "Invalid 'clockid'");
1380                         err = -EINVAL;
1381                         goto out;
1382                 }
1383                 /* This pairs with READ_ONCE() in taprio_mono_to_any */
1384                 WRITE_ONCE(q->tk_offset, tk_offset);
1385
1386                 q->clockid = clockid;
1387         } else {
1388                 NL_SET_ERR_MSG(extack, "Specifying a 'clockid' is mandatory");
1389                 goto out;
1390         }
1391
1392         /* Everything went ok, return success. */
1393         err = 0;
1394
1395 out:
1396         return err;
1397 }
1398
1399 static int taprio_mqprio_cmp(const struct net_device *dev,
1400                              const struct tc_mqprio_qopt *mqprio)
1401 {
1402         int i;
1403
1404         if (!mqprio || mqprio->num_tc != dev->num_tc)
1405                 return -1;
1406
1407         for (i = 0; i < mqprio->num_tc; i++)
1408                 if (dev->tc_to_txq[i].count != mqprio->count[i] ||
1409                     dev->tc_to_txq[i].offset != mqprio->offset[i])
1410                         return -1;
1411
1412         for (i = 0; i <= TC_BITMASK; i++)
1413                 if (dev->prio_tc_map[i] != mqprio->prio_tc_map[i])
1414                         return -1;
1415
1416         return 0;
1417 }
1418
1419 /* The semantics of the 'flags' argument in relation to 'change()'
1420  * requests, are interpreted following two rules (which are applied in
1421  * this order): (1) an omitted 'flags' argument is interpreted as
1422  * zero; (2) the 'flags' of a "running" taprio instance cannot be
1423  * changed.
1424  */
1425 static int taprio_new_flags(const struct nlattr *attr, u32 old,
1426                             struct netlink_ext_ack *extack)
1427 {
1428         u32 new = 0;
1429
1430         if (attr)
1431                 new = nla_get_u32(attr);
1432
1433         if (old != TAPRIO_FLAGS_INVALID && old != new) {
1434                 NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
1435                 return -EOPNOTSUPP;
1436         }
1437
1438         if (!taprio_flags_valid(new)) {
1439                 NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
1440                 return -EINVAL;
1441         }
1442
1443         return new;
1444 }
1445
1446 static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
1447                          struct netlink_ext_ack *extack)
1448 {
1449         struct nlattr *tb[TCA_TAPRIO_ATTR_MAX + 1] = { };
1450         struct sched_gate_list *oper, *admin, *new_admin;
1451         struct taprio_sched *q = qdisc_priv(sch);
1452         struct net_device *dev = qdisc_dev(sch);
1453         struct tc_mqprio_qopt *mqprio = NULL;
1454         unsigned long flags;
1455         ktime_t start;
1456         int i, err;
1457
1458         err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_ATTR_MAX, opt,
1459                                           taprio_policy, extack);
1460         if (err < 0)
1461                 return err;
1462
1463         if (tb[TCA_TAPRIO_ATTR_PRIOMAP])
1464                 mqprio = nla_data(tb[TCA_TAPRIO_ATTR_PRIOMAP]);
1465
1466         err = taprio_new_flags(tb[TCA_TAPRIO_ATTR_FLAGS],
1467                                q->flags, extack);
1468         if (err < 0)
1469                 return err;
1470
1471         q->flags = err;
1472
1473         err = taprio_parse_mqprio_opt(dev, mqprio, extack, q->flags);
1474         if (err < 0)
1475                 return err;
1476
1477         new_admin = kzalloc(sizeof(*new_admin), GFP_KERNEL);
1478         if (!new_admin) {
1479                 NL_SET_ERR_MSG(extack, "Not enough memory for a new schedule");
1480                 return -ENOMEM;
1481         }
1482         INIT_LIST_HEAD(&new_admin->entries);
1483
1484         rcu_read_lock();
1485         oper = rcu_dereference(q->oper_sched);
1486         admin = rcu_dereference(q->admin_sched);
1487         rcu_read_unlock();
1488
1489         /* no changes - no new mqprio settings */
1490         if (!taprio_mqprio_cmp(dev, mqprio))
1491                 mqprio = NULL;
1492
1493         if (mqprio && (oper || admin)) {
1494                 NL_SET_ERR_MSG(extack, "Changing the traffic mapping of a running schedule is not supported");
1495                 err = -ENOTSUPP;
1496                 goto free_sched;
1497         }
1498
1499         err = parse_taprio_schedule(q, tb, new_admin, extack);
1500         if (err < 0)
1501                 goto free_sched;
1502
1503         if (new_admin->num_entries == 0) {
1504                 NL_SET_ERR_MSG(extack, "There should be at least one entry in the schedule");
1505                 err = -EINVAL;
1506                 goto free_sched;
1507         }
1508
1509         err = taprio_parse_clockid(sch, tb, extack);
1510         if (err < 0)
1511                 goto free_sched;
1512
1513         taprio_set_picos_per_byte(dev, q);
1514
1515         if (mqprio) {
1516                 err = netdev_set_num_tc(dev, mqprio->num_tc);
1517                 if (err)
1518                         goto free_sched;
1519                 for (i = 0; i < mqprio->num_tc; i++)
1520                         netdev_set_tc_queue(dev, i,
1521                                             mqprio->count[i],
1522                                             mqprio->offset[i]);
1523
1524                 /* Always use supplied priority mappings */
1525                 for (i = 0; i <= TC_BITMASK; i++)
1526                         netdev_set_prio_tc_map(dev, i,
1527                                                mqprio->prio_tc_map[i]);
1528         }
1529
1530         if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1531                 err = taprio_enable_offload(dev, q, new_admin, extack);
1532         else
1533                 err = taprio_disable_offload(dev, q, extack);
1534         if (err)
1535                 goto free_sched;
1536
1537         /* Protects against enqueue()/dequeue() */
1538         spin_lock_bh(qdisc_lock(sch));
1539
1540         if (tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]) {
1541                 if (!TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1542                         NL_SET_ERR_MSG_MOD(extack, "txtime-delay can only be set when txtime-assist mode is enabled");
1543                         err = -EINVAL;
1544                         goto unlock;
1545                 }
1546
1547                 q->txtime_delay = nla_get_u32(tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]);
1548         }
1549
1550         if (!TXTIME_ASSIST_IS_ENABLED(q->flags) &&
1551             !FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1552             !hrtimer_active(&q->advance_timer)) {
1553                 hrtimer_init(&q->advance_timer, q->clockid, HRTIMER_MODE_ABS);
1554                 q->advance_timer.function = advance_sched;
1555         }
1556
1557         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1558                 q->dequeue = taprio_dequeue_offload;
1559                 q->peek = taprio_peek_offload;
1560         } else {
1561                 /* Be sure to always keep the function pointers
1562                  * in a consistent state.
1563                  */
1564                 q->dequeue = taprio_dequeue_soft;
1565                 q->peek = taprio_peek_soft;
1566         }
1567
1568         err = taprio_get_start_time(sch, new_admin, &start);
1569         if (err < 0) {
1570                 NL_SET_ERR_MSG(extack, "Internal error: failed get start time");
1571                 goto unlock;
1572         }
1573
1574         setup_txtime(q, new_admin, start);
1575
1576         if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1577                 if (!oper) {
1578                         rcu_assign_pointer(q->oper_sched, new_admin);
1579                         err = 0;
1580                         new_admin = NULL;
1581                         goto unlock;
1582                 }
1583
1584                 rcu_assign_pointer(q->admin_sched, new_admin);
1585                 if (admin)
1586                         call_rcu(&admin->rcu, taprio_free_sched_cb);
1587         } else {
1588                 setup_first_close_time(q, new_admin, start);
1589
1590                 /* Protects against advance_sched() */
1591                 spin_lock_irqsave(&q->current_entry_lock, flags);
1592
1593                 taprio_start_sched(sch, start, new_admin);
1594
1595                 rcu_assign_pointer(q->admin_sched, new_admin);
1596                 if (admin)
1597                         call_rcu(&admin->rcu, taprio_free_sched_cb);
1598
1599                 spin_unlock_irqrestore(&q->current_entry_lock, flags);
1600
1601                 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1602                         taprio_offload_config_changed(q);
1603         }
1604
1605         new_admin = NULL;
1606         err = 0;
1607
1608 unlock:
1609         spin_unlock_bh(qdisc_lock(sch));
1610
1611 free_sched:
1612         if (new_admin)
1613                 call_rcu(&new_admin->rcu, taprio_free_sched_cb);
1614
1615         return err;
1616 }
1617
1618 static void taprio_reset(struct Qdisc *sch)
1619 {
1620         struct taprio_sched *q = qdisc_priv(sch);
1621         struct net_device *dev = qdisc_dev(sch);
1622         int i;
1623
1624         hrtimer_cancel(&q->advance_timer);
1625
1626         if (q->qdiscs) {
1627                 for (i = 0; i < dev->num_tx_queues && q->qdiscs[i]; i++)
1628                         qdisc_reset(q->qdiscs[i]);
1629         }
1630         sch->qstats.backlog = 0;
1631         sch->q.qlen = 0;
1632 }
1633
1634 static void taprio_destroy(struct Qdisc *sch)
1635 {
1636         struct taprio_sched *q = qdisc_priv(sch);
1637         struct net_device *dev = qdisc_dev(sch);
1638         unsigned int i;
1639
1640         spin_lock(&taprio_list_lock);
1641         list_del(&q->taprio_list);
1642         spin_unlock(&taprio_list_lock);
1643
1644         /* Note that taprio_reset() might not be called if an error
1645          * happens in qdisc_create(), after taprio_init() has been called.
1646          */
1647         hrtimer_cancel(&q->advance_timer);
1648         qdisc_synchronize(sch);
1649
1650         taprio_disable_offload(dev, q, NULL);
1651
1652         if (q->qdiscs) {
1653                 for (i = 0; i < dev->num_tx_queues; i++)
1654                         qdisc_put(q->qdiscs[i]);
1655
1656                 kfree(q->qdiscs);
1657         }
1658         q->qdiscs = NULL;
1659
1660         netdev_reset_tc(dev);
1661
1662         if (q->oper_sched)
1663                 call_rcu(&q->oper_sched->rcu, taprio_free_sched_cb);
1664
1665         if (q->admin_sched)
1666                 call_rcu(&q->admin_sched->rcu, taprio_free_sched_cb);
1667 }
1668
1669 static int taprio_init(struct Qdisc *sch, struct nlattr *opt,
1670                        struct netlink_ext_ack *extack)
1671 {
1672         struct taprio_sched *q = qdisc_priv(sch);
1673         struct net_device *dev = qdisc_dev(sch);
1674         int i;
1675
1676         spin_lock_init(&q->current_entry_lock);
1677
1678         hrtimer_init(&q->advance_timer, CLOCK_TAI, HRTIMER_MODE_ABS);
1679         q->advance_timer.function = advance_sched;
1680
1681         q->dequeue = taprio_dequeue_soft;
1682         q->peek = taprio_peek_soft;
1683
1684         q->root = sch;
1685
1686         /* We only support static clockids. Use an invalid value as default
1687          * and get the valid one on taprio_change().
1688          */
1689         q->clockid = -1;
1690         q->flags = TAPRIO_FLAGS_INVALID;
1691
1692         spin_lock(&taprio_list_lock);
1693         list_add(&q->taprio_list, &taprio_list);
1694         spin_unlock(&taprio_list_lock);
1695
1696         if (sch->parent != TC_H_ROOT)
1697                 return -EOPNOTSUPP;
1698
1699         if (!netif_is_multiqueue(dev))
1700                 return -EOPNOTSUPP;
1701
1702         /* pre-allocate qdisc, attachment can't fail */
1703         q->qdiscs = kcalloc(dev->num_tx_queues,
1704                             sizeof(q->qdiscs[0]),
1705                             GFP_KERNEL);
1706
1707         if (!q->qdiscs)
1708                 return -ENOMEM;
1709
1710         if (!opt)
1711                 return -EINVAL;
1712
1713         for (i = 0; i < dev->num_tx_queues; i++) {
1714                 struct netdev_queue *dev_queue;
1715                 struct Qdisc *qdisc;
1716
1717                 dev_queue = netdev_get_tx_queue(dev, i);
1718                 qdisc = qdisc_create_dflt(dev_queue,
1719                                           &pfifo_qdisc_ops,
1720                                           TC_H_MAKE(TC_H_MAJ(sch->handle),
1721                                                     TC_H_MIN(i + 1)),
1722                                           extack);
1723                 if (!qdisc)
1724                         return -ENOMEM;
1725
1726                 if (i < dev->real_num_tx_queues)
1727                         qdisc_hash_add(qdisc, false);
1728
1729                 q->qdiscs[i] = qdisc;
1730         }
1731
1732         return taprio_change(sch, opt, extack);
1733 }
1734
1735 static struct netdev_queue *taprio_queue_get(struct Qdisc *sch,
1736                                              unsigned long cl)
1737 {
1738         struct net_device *dev = qdisc_dev(sch);
1739         unsigned long ntx = cl - 1;
1740
1741         if (ntx >= dev->num_tx_queues)
1742                 return NULL;
1743
1744         return netdev_get_tx_queue(dev, ntx);
1745 }
1746
1747 static int taprio_graft(struct Qdisc *sch, unsigned long cl,
1748                         struct Qdisc *new, struct Qdisc **old,
1749                         struct netlink_ext_ack *extack)
1750 {
1751         struct taprio_sched *q = qdisc_priv(sch);
1752         struct net_device *dev = qdisc_dev(sch);
1753         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1754
1755         if (!dev_queue)
1756                 return -EINVAL;
1757
1758         if (dev->flags & IFF_UP)
1759                 dev_deactivate(dev);
1760
1761         *old = q->qdiscs[cl - 1];
1762         q->qdiscs[cl - 1] = new;
1763
1764         if (new)
1765                 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1766
1767         if (dev->flags & IFF_UP)
1768                 dev_activate(dev);
1769
1770         return 0;
1771 }
1772
1773 static int dump_entry(struct sk_buff *msg,
1774                       const struct sched_entry *entry)
1775 {
1776         struct nlattr *item;
1777
1778         item = nla_nest_start_noflag(msg, TCA_TAPRIO_SCHED_ENTRY);
1779         if (!item)
1780                 return -ENOSPC;
1781
1782         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INDEX, entry->index))
1783                 goto nla_put_failure;
1784
1785         if (nla_put_u8(msg, TCA_TAPRIO_SCHED_ENTRY_CMD, entry->command))
1786                 goto nla_put_failure;
1787
1788         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_GATE_MASK,
1789                         entry->gate_mask))
1790                 goto nla_put_failure;
1791
1792         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INTERVAL,
1793                         entry->interval))
1794                 goto nla_put_failure;
1795
1796         return nla_nest_end(msg, item);
1797
1798 nla_put_failure:
1799         nla_nest_cancel(msg, item);
1800         return -1;
1801 }
1802
1803 static int dump_schedule(struct sk_buff *msg,
1804                          const struct sched_gate_list *root)
1805 {
1806         struct nlattr *entry_list;
1807         struct sched_entry *entry;
1808
1809         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_BASE_TIME,
1810                         root->base_time, TCA_TAPRIO_PAD))
1811                 return -1;
1812
1813         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME,
1814                         root->cycle_time, TCA_TAPRIO_PAD))
1815                 return -1;
1816
1817         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION,
1818                         root->cycle_time_extension, TCA_TAPRIO_PAD))
1819                 return -1;
1820
1821         entry_list = nla_nest_start_noflag(msg,
1822                                            TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST);
1823         if (!entry_list)
1824                 goto error_nest;
1825
1826         list_for_each_entry(entry, &root->entries, list) {
1827                 if (dump_entry(msg, entry) < 0)
1828                         goto error_nest;
1829         }
1830
1831         nla_nest_end(msg, entry_list);
1832         return 0;
1833
1834 error_nest:
1835         nla_nest_cancel(msg, entry_list);
1836         return -1;
1837 }
1838
1839 static int taprio_dump(struct Qdisc *sch, struct sk_buff *skb)
1840 {
1841         struct taprio_sched *q = qdisc_priv(sch);
1842         struct net_device *dev = qdisc_dev(sch);
1843         struct sched_gate_list *oper, *admin;
1844         struct tc_mqprio_qopt opt = { 0 };
1845         struct nlattr *nest, *sched_nest;
1846         unsigned int i;
1847
1848         rcu_read_lock();
1849         oper = rcu_dereference(q->oper_sched);
1850         admin = rcu_dereference(q->admin_sched);
1851
1852         opt.num_tc = netdev_get_num_tc(dev);
1853         memcpy(opt.prio_tc_map, dev->prio_tc_map, sizeof(opt.prio_tc_map));
1854
1855         for (i = 0; i < netdev_get_num_tc(dev); i++) {
1856                 opt.count[i] = dev->tc_to_txq[i].count;
1857                 opt.offset[i] = dev->tc_to_txq[i].offset;
1858         }
1859
1860         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1861         if (!nest)
1862                 goto start_error;
1863
1864         if (nla_put(skb, TCA_TAPRIO_ATTR_PRIOMAP, sizeof(opt), &opt))
1865                 goto options_error;
1866
1867         if (!FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1868             nla_put_s32(skb, TCA_TAPRIO_ATTR_SCHED_CLOCKID, q->clockid))
1869                 goto options_error;
1870
1871         if (q->flags && nla_put_u32(skb, TCA_TAPRIO_ATTR_FLAGS, q->flags))
1872                 goto options_error;
1873
1874         if (q->txtime_delay &&
1875             nla_put_u32(skb, TCA_TAPRIO_ATTR_TXTIME_DELAY, q->txtime_delay))
1876                 goto options_error;
1877
1878         if (oper && dump_schedule(skb, oper))
1879                 goto options_error;
1880
1881         if (!admin)
1882                 goto done;
1883
1884         sched_nest = nla_nest_start_noflag(skb, TCA_TAPRIO_ATTR_ADMIN_SCHED);
1885         if (!sched_nest)
1886                 goto options_error;
1887
1888         if (dump_schedule(skb, admin))
1889                 goto admin_error;
1890
1891         nla_nest_end(skb, sched_nest);
1892
1893 done:
1894         rcu_read_unlock();
1895
1896         return nla_nest_end(skb, nest);
1897
1898 admin_error:
1899         nla_nest_cancel(skb, sched_nest);
1900
1901 options_error:
1902         nla_nest_cancel(skb, nest);
1903
1904 start_error:
1905         rcu_read_unlock();
1906         return -ENOSPC;
1907 }
1908
1909 static struct Qdisc *taprio_leaf(struct Qdisc *sch, unsigned long cl)
1910 {
1911         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1912
1913         if (!dev_queue)
1914                 return NULL;
1915
1916         return dev_queue->qdisc_sleeping;
1917 }
1918
1919 static unsigned long taprio_find(struct Qdisc *sch, u32 classid)
1920 {
1921         unsigned int ntx = TC_H_MIN(classid);
1922
1923         if (!taprio_queue_get(sch, ntx))
1924                 return 0;
1925         return ntx;
1926 }
1927
1928 static int taprio_dump_class(struct Qdisc *sch, unsigned long cl,
1929                              struct sk_buff *skb, struct tcmsg *tcm)
1930 {
1931         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1932
1933         tcm->tcm_parent = TC_H_ROOT;
1934         tcm->tcm_handle |= TC_H_MIN(cl);
1935         tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
1936
1937         return 0;
1938 }
1939
1940 static int taprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
1941                                    struct gnet_dump *d)
1942         __releases(d->lock)
1943         __acquires(d->lock)
1944 {
1945         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1946
1947         sch = dev_queue->qdisc_sleeping;
1948         if (gnet_stats_copy_basic(&sch->running, d, NULL, &sch->bstats) < 0 ||
1949             qdisc_qstats_copy(d, sch) < 0)
1950                 return -1;
1951         return 0;
1952 }
1953
1954 static void taprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1955 {
1956         struct net_device *dev = qdisc_dev(sch);
1957         unsigned long ntx;
1958
1959         if (arg->stop)
1960                 return;
1961
1962         arg->count = arg->skip;
1963         for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
1964                 if (arg->fn(sch, ntx + 1, arg) < 0) {
1965                         arg->stop = 1;
1966                         break;
1967                 }
1968                 arg->count++;
1969         }
1970 }
1971
1972 static struct netdev_queue *taprio_select_queue(struct Qdisc *sch,
1973                                                 struct tcmsg *tcm)
1974 {
1975         return taprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
1976 }
1977
1978 static const struct Qdisc_class_ops taprio_class_ops = {
1979         .graft          = taprio_graft,
1980         .leaf           = taprio_leaf,
1981         .find           = taprio_find,
1982         .walk           = taprio_walk,
1983         .dump           = taprio_dump_class,
1984         .dump_stats     = taprio_dump_class_stats,
1985         .select_queue   = taprio_select_queue,
1986 };
1987
1988 static struct Qdisc_ops taprio_qdisc_ops __read_mostly = {
1989         .cl_ops         = &taprio_class_ops,
1990         .id             = "taprio",
1991         .priv_size      = sizeof(struct taprio_sched),
1992         .init           = taprio_init,
1993         .change         = taprio_change,
1994         .destroy        = taprio_destroy,
1995         .reset          = taprio_reset,
1996         .peek           = taprio_peek,
1997         .dequeue        = taprio_dequeue,
1998         .enqueue        = taprio_enqueue,
1999         .dump           = taprio_dump,
2000         .owner          = THIS_MODULE,
2001 };
2002
2003 static struct notifier_block taprio_device_notifier = {
2004         .notifier_call = taprio_dev_notifier,
2005 };
2006
2007 static int __init taprio_module_init(void)
2008 {
2009         int err = register_netdevice_notifier(&taprio_device_notifier);
2010
2011         if (err)
2012                 return err;
2013
2014         return register_qdisc(&taprio_qdisc_ops);
2015 }
2016
2017 static void __exit taprio_module_exit(void)
2018 {
2019         unregister_qdisc(&taprio_qdisc_ops);
2020         unregister_netdevice_notifier(&taprio_device_notifier);
2021 }
2022
2023 module_init(taprio_module_init);
2024 module_exit(taprio_module_exit);
2025 MODULE_LICENSE("GPL");