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_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         list_for_each_entry_safe(entry, n, &sched->entries, list) {
 122                 list_del(&entry->list);
 123                 kfree(entry);
 124         }
 125
 126         kfree(sched);
 127 }
 128
 129 static void switch_schedules(struct taprio_sched *q,
 130                              struct sched_gate_list **admin,
 131                              struct sched_gate_list **oper)
 132 {
 133         rcu_assign_pointer(q->oper_sched, *admin);
 134         rcu_assign_pointer(q->admin_sched, NULL);
 135
 136         if (*oper)
 137                 call_rcu(&(*oper)->rcu, taprio_free_sched_cb);
 138
 139         *oper = *admin;
 140         *admin = NULL;
 141 }
 142
 143 /* Get how much time has been already elapsed in the current cycle. */
 144 static s32 get_cycle_time_elapsed(struct sched_gate_list *sched, ktime_t time)
 145 {
 146         ktime_t time_since_sched_start;
 147         s32 time_elapsed;
 148
 149         time_since_sched_start = ktime_sub(time, sched->base_time);
 150         div_s64_rem(time_since_sched_start, sched->cycle_time, &time_elapsed);
 151
 152         return time_elapsed;
 153 }
 154
 155 static ktime_t get_interval_end_time(struct sched_gate_list *sched,
 156                                      struct sched_gate_list *admin,
 157                                      struct sched_entry *entry,
 158                                      ktime_t intv_start)
 159 {
 160         s32 cycle_elapsed = get_cycle_time_elapsed(sched, intv_start);
 161         ktime_t intv_end, cycle_ext_end, cycle_end;
 162
 163         cycle_end = ktime_add_ns(intv_start, sched->cycle_time - cycle_elapsed);
 164         intv_end = ktime_add_ns(intv_start, entry->interval);
 165         cycle_ext_end = ktime_add(cycle_end, sched->cycle_time_extension);
 166
 167         if (ktime_before(intv_end, cycle_end))
 168                 return intv_end;
 169         else if (admin && admin != sched &&
 170                  ktime_after(admin->base_time, cycle_end) &&
 171                  ktime_before(admin->base_time, cycle_ext_end))
 172                 return admin->base_time;
 173         else
 174                 return cycle_end;
 175 }
 176
 177 static int length_to_duration(struct taprio_sched *q, int len)
 178 {
 179         return div_u64(len * atomic64_read(&q->picos_per_byte), 1000);
 180 }
 181
 182 /* Returns the entry corresponding to next available interval. If
 183  * validate_interval is set, it only validates whether the timestamp occurs
 184  * when the gate corresponding to the skb's traffic class is open.
 185  */
 186 static struct sched_entry *find_entry_to_transmit(struct sk_buff *skb,
 187                                                   struct Qdisc *sch,
 188                                                   struct sched_gate_list *sched,
 189                                                   struct sched_gate_list *admin,
 190                                                   ktime_t time,
 191                                                   ktime_t *interval_start,
 192                                                   ktime_t *interval_end,
 193                                                   bool validate_interval)
 194 {
 195         ktime_t curr_intv_start, curr_intv_end, cycle_end, packet_transmit_time;
 196         ktime_t earliest_txtime = KTIME_MAX, txtime, cycle, transmit_end_time;
 197         struct sched_entry *entry = NULL, *entry_found = NULL;
 198         struct taprio_sched *q = qdisc_priv(sch);
 199         struct net_device *dev = qdisc_dev(sch);
 200         bool entry_available = false;
 201         s32 cycle_elapsed;
 202         int tc, n;
 203
 204         tc = netdev_get_prio_tc_map(dev, skb->priority);
 205         packet_transmit_time = length_to_duration(q, qdisc_pkt_len(skb));
 206
 207         *interval_start = 0;
 208         *interval_end = 0;
 209
 210         if (!sched)
 211                 return NULL;
 212
 213         cycle = sched->cycle_time;
 214         cycle_elapsed = get_cycle_time_elapsed(sched, time);
 215         curr_intv_end = ktime_sub_ns(time, cycle_elapsed);
 216         cycle_end = ktime_add_ns(curr_intv_end, cycle);
 217
 218         list_for_each_entry(entry, &sched->entries, list) {
 219                 curr_intv_start = curr_intv_end;
 220                 curr_intv_end = get_interval_end_time(sched, admin, entry,
 221                                                       curr_intv_start);
 222
 223                 if (ktime_after(curr_intv_start, cycle_end))
 224                         break;
 225
 226                 if (!(entry->gate_mask & BIT(tc)) ||
 227                     packet_transmit_time > entry->interval)
 228                         continue;
 229
 230                 txtime = entry->next_txtime;
 231
 232                 if (ktime_before(txtime, time) || validate_interval) {
 233                         transmit_end_time = ktime_add_ns(time, packet_transmit_time);
 234                         if ((ktime_before(curr_intv_start, time) &&
 235                              ktime_before(transmit_end_time, curr_intv_end)) ||
 236                             (ktime_after(curr_intv_start, time) && !validate_interval)) {
 237                                 entry_found = entry;
 238                                 *interval_start = curr_intv_start;
 239                                 *interval_end = curr_intv_end;
 240                                 break;
 241                         } else if (!entry_available && !validate_interval) {
 242                                 /* Here, we are just trying to find out the
 243                                  * first available interval in the next cycle.
 244                                  */
 245                                 entry_available = true;
 246                                 entry_found = entry;
 247                                 *interval_start = ktime_add_ns(curr_intv_start, cycle);
 248                                 *interval_end = ktime_add_ns(curr_intv_end, cycle);
 249                         }
 250                 } else if (ktime_before(txtime, earliest_txtime) &&
 251                            !entry_available) {
 252                         earliest_txtime = txtime;
 253                         entry_found = entry;
 254                         n = div_s64(ktime_sub(txtime, curr_intv_start), cycle);
 255                         *interval_start = ktime_add(curr_intv_start, n * cycle);
 256                         *interval_end = ktime_add(curr_intv_end, n * cycle);
 257                 }
 258         }
 259
 260         return entry_found;
 261 }
 262
 263 static bool is_valid_interval(struct sk_buff *skb, struct Qdisc *sch)
 264 {
 265         struct taprio_sched *q = qdisc_priv(sch);
 266         struct sched_gate_list *sched, *admin;
 267         ktime_t interval_start, interval_end;
 268         struct sched_entry *entry;
 269
 270         rcu_read_lock();
 271         sched = rcu_dereference(q->oper_sched);
 272         admin = rcu_dereference(q->admin_sched);
 273
 274         entry = find_entry_to_transmit(skb, sch, sched, admin, skb->tstamp,
 275                                        &interval_start, &interval_end, true);
 276         rcu_read_unlock();
 277
 278         return entry;
 279 }
 280
 281 static bool taprio_flags_valid(u32 flags)
 282 {
 283         /* Make sure no other flag bits are set. */
 284         if (flags & ~(TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST |
 285                       TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
 286                 return false;
 287         /* txtime-assist and full offload are mutually exclusive */
 288         if ((flags & TCA_TAPRIO_ATTR_FLAG_TXTIME_ASSIST) &&
 289             (flags & TCA_TAPRIO_ATTR_FLAG_FULL_OFFLOAD))
 290                 return false;
 291         return true;
 292 }
 293
 294 /* This returns the tstamp value set by TCP in terms of the set clock. */
 295 static ktime_t get_tcp_tstamp(struct taprio_sched *q, struct sk_buff *skb)
 296 {
 297         unsigned int offset = skb_network_offset(skb);
 298         const struct ipv6hdr *ipv6h;
 299         const struct iphdr *iph;
 300         struct ipv6hdr _ipv6h;
 301
 302         ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h);
 303         if (!ipv6h)
 304                 return 0;
 305
 306         if (ipv6h->version == 4) {
 307                 iph = (struct iphdr *)ipv6h;
 308                 offset += iph->ihl * 4;
 309
 310                 /* special-case 6in4 tunnelling, as that is a common way to get
 311                  * v6 connectivity in the home
 312                  */
 313                 if (iph->protocol == IPPROTO_IPV6) {
 314                         ipv6h = skb_header_pointer(skb, offset,
 315                                                    sizeof(_ipv6h), &_ipv6h);
 316
 317                         if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP)
 318                                 return 0;
 319                 } else if (iph->protocol != IPPROTO_TCP) {
 320                         return 0;
 321                 }
 322         } else if (ipv6h->version == 6 && ipv6h->nexthdr != IPPROTO_TCP) {
 323                 return 0;
 324         }
 325
 326         return taprio_mono_to_any(q, skb->skb_mstamp_ns);
 327 }
 328
 329 /* There are a few scenarios where we will have to modify the txtime from
 330  * what is read from next_txtime in sched_entry. They are:
 331  * 1. If txtime is in the past,
 332  *    a. The gate for the traffic class is currently open and packet can be
 333  *       transmitted before it closes, schedule the packet right away.
 334  *    b. If the gate corresponding to the traffic class is going to open later
 335  *       in the cycle, set the txtime of packet to the interval start.
 336  * 2. If txtime is in the future, there are packets corresponding to the
 337  *    current traffic class waiting to be transmitted. So, the following
 338  *    possibilities exist:
 339  *    a. We can transmit the packet before the window containing the txtime
 340  *       closes.
 341  *    b. The window might close before the transmission can be completed
 342  *       successfully. So, schedule the packet in the next open window.
 343  */
 344 static long get_packet_txtime(struct sk_buff *skb, struct Qdisc *sch)
 345 {
 346         ktime_t transmit_end_time, interval_end, interval_start, tcp_tstamp;
 347         struct taprio_sched *q = qdisc_priv(sch);
 348         struct sched_gate_list *sched, *admin;
 349         ktime_t minimum_time, now, txtime;
 350         int len, packet_transmit_time;
 351         struct sched_entry *entry;
 352         bool sched_changed;
 353
 354         now = taprio_get_time(q);
 355         minimum_time = ktime_add_ns(now, q->txtime_delay);
 356
 357         tcp_tstamp = get_tcp_tstamp(q, skb);
 358         minimum_time = max_t(ktime_t, minimum_time, tcp_tstamp);
 359
 360         rcu_read_lock();
 361         admin = rcu_dereference(q->admin_sched);
 362         sched = rcu_dereference(q->oper_sched);
 363         if (admin && ktime_after(minimum_time, admin->base_time))
 364                 switch_schedules(q, &admin, &sched);
 365
 366         /* Until the schedule starts, all the queues are open */
 367         if (!sched || ktime_before(minimum_time, sched->base_time)) {
 368                 txtime = minimum_time;
 369                 goto done;
 370         }
 371
 372         len = qdisc_pkt_len(skb);
 373         packet_transmit_time = length_to_duration(q, len);
 374
 375         do {
 376                 sched_changed = false;
 377
 378                 entry = find_entry_to_transmit(skb, sch, sched, admin,
 379                                                minimum_time,
 380                                                &interval_start, &interval_end,
 381                                                false);
 382                 if (!entry) {
 383                         txtime = 0;
 384                         goto done;
 385                 }
 386
 387                 txtime = entry->next_txtime;
 388                 txtime = max_t(ktime_t, txtime, minimum_time);
 389                 txtime = max_t(ktime_t, txtime, interval_start);
 390
 391                 if (admin && admin != sched &&
 392                     ktime_after(txtime, admin->base_time)) {
 393                         sched = admin;
 394                         sched_changed = true;
 395                         continue;
 396                 }
 397
 398                 transmit_end_time = ktime_add(txtime, packet_transmit_time);
 399                 minimum_time = transmit_end_time;
 400
 401                 /* Update the txtime of current entry to the next time it's
 402                  * interval starts.
 403                  */
 404                 if (ktime_after(transmit_end_time, interval_end))
 405                         entry->next_txtime = ktime_add(interval_start, sched->cycle_time);
 406         } while (sched_changed || ktime_after(transmit_end_time, interval_end));
 407
 408         entry->next_txtime = transmit_end_time;
 409
 410 done:
 411         rcu_read_unlock();
 412         return txtime;
 413 }
 414
 415 static int taprio_enqueue_one(struct sk_buff *skb, struct Qdisc *sch,
 416                               struct Qdisc *child, struct sk_buff **to_free)
 417 {
 418         struct taprio_sched *q = qdisc_priv(sch);
 419
 420         /* sk_flags are only safe to use on full sockets. */
 421         if (skb->sk && sk_fullsock(skb->sk) && sock_flag(skb->sk, SOCK_TXTIME)) {
 422                 if (!is_valid_interval(skb, sch))
 423                         return qdisc_drop(skb, sch, to_free);
 424         } else if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
 425                 skb->tstamp = get_packet_txtime(skb, sch);
 426                 if (!skb->tstamp)
 427                         return qdisc_drop(skb, sch, to_free);
 428         }
 429
 430         qdisc_qstats_backlog_inc(sch, skb);
 431         sch->q.qlen++;
 432
 433         return qdisc_enqueue(skb, child, to_free);
 434 }
 435
 436 static int taprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
 437                           struct sk_buff **to_free)
 438 {
 439         struct taprio_sched *q = qdisc_priv(sch);
 440         struct Qdisc *child;
 441         int queue;
 442
 443         if (unlikely(FULL_OFFLOAD_IS_ENABLED(q->flags))) {
 444                 WARN_ONCE(1, "Trying to enqueue skb into the root of a taprio qdisc configured with full offload\n");
 445                 return qdisc_drop(skb, sch, to_free);
 446         }
 447
 448         queue = skb_get_queue_mapping(skb);
 449
 450         child = q->qdiscs[queue];
 451         if (unlikely(!child))
 452                 return qdisc_drop(skb, sch, to_free);
 453
 454         /* Large packets might not be transmitted when the transmission duration
 455          * exceeds any configured interval. Therefore, segment the skb into
 456          * smaller chunks. Skip it for the full offload case, as the driver
 457          * and/or the hardware is expected to handle this.
 458          */
 459         if (skb_is_gso(skb) && !FULL_OFFLOAD_IS_ENABLED(q->flags)) {
 460                 unsigned int slen = 0, numsegs = 0, len = qdisc_pkt_len(skb);
 461                 netdev_features_t features = netif_skb_features(skb);
 462                 struct sk_buff *segs, *nskb;
 463                 int ret;
 464
 465                 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
 466                 if (IS_ERR_OR_NULL(segs))
 467                         return qdisc_drop(skb, sch, to_free);
 468
 469                 skb_list_walk_safe(segs, segs, nskb) {
 470                         skb_mark_not_on_list(segs);
 471                         qdisc_skb_cb(segs)->pkt_len = segs->len;
 472                         slen += segs->len;
 473
 474                         ret = taprio_enqueue_one(segs, sch, child, to_free);
 475                         if (ret != NET_XMIT_SUCCESS) {
 476                                 if (net_xmit_drop_count(ret))
 477                                         qdisc_qstats_drop(sch);
 478                         } else {
 479                                 numsegs++;
 480                         }
 481                 }
 482
 483                 if (numsegs > 1)
 484                         qdisc_tree_reduce_backlog(sch, 1 - numsegs, len - slen);
 485                 consume_skb(skb);
 486
 487                 return numsegs > 0 ? NET_XMIT_SUCCESS : NET_XMIT_DROP;
 488         }
 489
 490         return taprio_enqueue_one(skb, sch, child, to_free);
 491 }
 492
 493 static struct sk_buff *taprio_peek_soft(struct Qdisc *sch)
 494 {
 495         struct taprio_sched *q = qdisc_priv(sch);
 496         struct net_device *dev = qdisc_dev(sch);
 497         struct sched_entry *entry;
 498         struct sk_buff *skb;
 499         u32 gate_mask;
 500         int i;
 501
 502         rcu_read_lock();
 503         entry = rcu_dereference(q->current_entry);
 504         gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
 505         rcu_read_unlock();
 506
 507         if (!gate_mask)
 508                 return NULL;
 509
 510         for (i = 0; i < dev->num_tx_queues; i++) {
 511                 struct Qdisc *child = q->qdiscs[i];
 512                 int prio;
 513                 u8 tc;
 514
 515                 if (unlikely(!child))
 516                         continue;
 517
 518                 skb = child->ops->peek(child);
 519                 if (!skb)
 520                         continue;
 521
 522                 if (TXTIME_ASSIST_IS_ENABLED(q->flags))
 523                         return skb;
 524
 525                 prio = skb->priority;
 526                 tc = netdev_get_prio_tc_map(dev, prio);
 527
 528                 if (!(gate_mask & BIT(tc)))
 529                         continue;
 530
 531                 return skb;
 532         }
 533
 534         return NULL;
 535 }
 536
 537 static struct sk_buff *taprio_peek_offload(struct Qdisc *sch)
 538 {
 539         WARN_ONCE(1, "Trying to peek into the root of a taprio qdisc configured with full offload\n");
 540
 541         return NULL;
 542 }
 543
 544 static struct sk_buff *taprio_peek(struct Qdisc *sch)
 545 {
 546         struct taprio_sched *q = qdisc_priv(sch);
 547
 548         return q->peek(sch);
 549 }
 550
 551 static void taprio_set_budget(struct taprio_sched *q, struct sched_entry *entry)
 552 {
 553         atomic_set(&entry->budget,
 554                    div64_u64((u64)entry->interval * 1000,
 555                              atomic64_read(&q->picos_per_byte)));
 556 }
 557
 558 static struct sk_buff *taprio_dequeue_soft(struct Qdisc *sch)
 559 {
 560         struct taprio_sched *q = qdisc_priv(sch);
 561         struct net_device *dev = qdisc_dev(sch);
 562         struct sk_buff *skb = NULL;
 563         struct sched_entry *entry;
 564         u32 gate_mask;
 565         int i;
 566
 567         rcu_read_lock();
 568         entry = rcu_dereference(q->current_entry);
 569         /* if there's no entry, it means that the schedule didn't
 570          * start yet, so force all gates to be open, this is in
 571          * accordance to IEEE 802.1Qbv-2015 Section 8.6.9.4.5
 572          * "AdminGateStates"
 573          */
 574         gate_mask = entry ? entry->gate_mask : TAPRIO_ALL_GATES_OPEN;
 575
 576         if (!gate_mask)
 577                 goto done;
 578
 579         for (i = 0; i < dev->num_tx_queues; i++) {
 580                 struct Qdisc *child = q->qdiscs[i];
 581                 ktime_t guard;
 582                 int prio;
 583                 int len;
 584                 u8 tc;
 585
 586                 if (unlikely(!child))
 587                         continue;
 588
 589                 if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
 590                         skb = child->ops->dequeue(child);
 591                         if (!skb)
 592                                 continue;
 593                         goto skb_found;
 594                 }
 595
 596                 skb = child->ops->peek(child);
 597                 if (!skb)
 598                         continue;
 599
 600                 prio = skb->priority;
 601                 tc = netdev_get_prio_tc_map(dev, prio);
 602
 603                 if (!(gate_mask & BIT(tc))) {
 604                         skb = NULL;
 605                         continue;
 606                 }
 607
 608                 len = qdisc_pkt_len(skb);
 609                 guard = ktime_add_ns(taprio_get_time(q),
 610                                      length_to_duration(q, len));
 611
 612                 /* In the case that there's no gate entry, there's no
 613                  * guard band ...
 614                  */
 615                 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
 616                     ktime_after(guard, entry->close_time)) {
 617                         skb = NULL;
 618                         continue;
 619                 }
 620
 621                 /* ... and no budget. */
 622                 if (gate_mask != TAPRIO_ALL_GATES_OPEN &&
 623                     atomic_sub_return(len, &entry->budget) < 0) {
 624                         skb = NULL;
 625                         continue;
 626                 }
 627
 628                 skb = child->ops->dequeue(child);
 629                 if (unlikely(!skb))
 630                         goto done;
 631
 632 skb_found:
 633                 qdisc_bstats_update(sch, skb);
 634                 qdisc_qstats_backlog_dec(sch, skb);
 635                 sch->q.qlen--;
 636
 637                 goto done;
 638         }
 639
 640 done:
 641         rcu_read_unlock();
 642
 643         return skb;
 644 }
 645
 646 static struct sk_buff *taprio_dequeue_offload(struct Qdisc *sch)
 647 {
 648         WARN_ONCE(1, "Trying to dequeue from the root of a taprio qdisc configured with full offload\n");
 649
 650         return NULL;
 651 }
 652
 653 static struct sk_buff *taprio_dequeue(struct Qdisc *sch)
 654 {
 655         struct taprio_sched *q = qdisc_priv(sch);
 656
 657         return q->dequeue(sch);
 658 }
 659
 660 static bool should_restart_cycle(const struct sched_gate_list *oper,
 661                                  const struct sched_entry *entry)
 662 {
 663         if (list_is_last(&entry->list, &oper->entries))
 664                 return true;
 665
 666         if (ktime_compare(entry->close_time, oper->cycle_close_time) == 0)
 667                 return true;
 668
 669         return false;
 670 }
 671
 672 static bool should_change_schedules(const struct sched_gate_list *admin,
 673                                     const struct sched_gate_list *oper,
 674                                     ktime_t close_time)
 675 {
 676         ktime_t next_base_time, extension_time;
 677
 678         if (!admin)
 679                 return false;
 680
 681         next_base_time = sched_base_time(admin);
 682
 683         /* This is the simple case, the close_time would fall after
 684          * the next schedule base_time.
 685          */
 686         if (ktime_compare(next_base_time, close_time) <= 0)
 687                 return true;
 688
 689         /* This is the cycle_time_extension case, if the close_time
 690          * plus the amount that can be extended would fall after the
 691          * next schedule base_time, we can extend the current schedule
 692          * for that amount.
 693          */
 694         extension_time = ktime_add_ns(close_time, oper->cycle_time_extension);
 695
 696         /* FIXME: the IEEE 802.1Q-2018 Specification isn't clear about
 697          * how precisely the extension should be made. So after
 698          * conformance testing, this logic may change.
 699          */
 700         if (ktime_compare(next_base_time, extension_time) <= 0)
 701                 return true;
 702
 703         return false;
 704 }
 705
 706 static enum hrtimer_restart advance_sched(struct hrtimer *timer)
 707 {
 708         struct taprio_sched *q = container_of(timer, struct taprio_sched,
 709                                               advance_timer);
 710         struct sched_gate_list *oper, *admin;
 711         struct sched_entry *entry, *next;
 712         struct Qdisc *sch = q->root;
 713         ktime_t close_time;
 714
 715         spin_lock(&q->current_entry_lock);
 716         entry = rcu_dereference_protected(q->current_entry,
 717                                           lockdep_is_held(&q->current_entry_lock));
 718         oper = rcu_dereference_protected(q->oper_sched,
 719                                          lockdep_is_held(&q->current_entry_lock));
 720         admin = rcu_dereference_protected(q->admin_sched,
 721                                           lockdep_is_held(&q->current_entry_lock));
 722
 723         if (!oper)
 724                 switch_schedules(q, &admin, &oper);
 725
 726         /* This can happen in two cases: 1. this is the very first run
 727          * of this function (i.e. we weren't running any schedule
 728          * previously); 2. The previous schedule just ended. The first
 729          * entry of all schedules are pre-calculated during the
 730          * schedule initialization.
 731          */
 732         if (unlikely(!entry || entry->close_time == oper->base_time)) {
 733                 next = list_first_entry(&oper->entries, struct sched_entry,
 734                                         list);
 735                 close_time = next->close_time;
 736                 goto first_run;
 737         }
 738
 739         if (should_restart_cycle(oper, entry)) {
 740                 next = list_first_entry(&oper->entries, struct sched_entry,
 741                                         list);
 742                 oper->cycle_close_time = ktime_add_ns(oper->cycle_close_time,
 743                                                       oper->cycle_time);
 744         } else {
 745                 next = list_next_entry(entry, list);
 746         }
 747
 748         close_time = ktime_add_ns(entry->close_time, next->interval);
 749         close_time = min_t(ktime_t, close_time, oper->cycle_close_time);
 750
 751         if (should_change_schedules(admin, oper, close_time)) {
 752                 /* Set things so the next time this runs, the new
 753                  * schedule runs.
 754                  */
 755                 close_time = sched_base_time(admin);
 756                 switch_schedules(q, &admin, &oper);
 757         }
 758
 759         next->close_time = close_time;
 760         taprio_set_budget(q, next);
 761
 762 first_run:
 763         rcu_assign_pointer(q->current_entry, next);
 764         spin_unlock(&q->current_entry_lock);
 765
 766         hrtimer_set_expires(&q->advance_timer, close_time);
 767
 768         rcu_read_lock();
 769         __netif_schedule(sch);
 770         rcu_read_unlock();
 771
 772         return HRTIMER_RESTART;
 773 }
 774
 775 static const struct nla_policy entry_policy[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = {
 776         [TCA_TAPRIO_SCHED_ENTRY_INDEX]     = { .type = NLA_U32 },
 777         [TCA_TAPRIO_SCHED_ENTRY_CMD]       = { .type = NLA_U8 },
 778         [TCA_TAPRIO_SCHED_ENTRY_GATE_MASK] = { .type = NLA_U32 },
 779         [TCA_TAPRIO_SCHED_ENTRY_INTERVAL]  = { .type = NLA_U32 },
 780 };
 781
 782 static const struct nla_policy taprio_policy[TCA_TAPRIO_ATTR_MAX + 1] = {
 783         [TCA_TAPRIO_ATTR_PRIOMAP]              = {
 784                 .len = sizeof(struct tc_mqprio_qopt)
 785         },
 786         [TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST]           = { .type = NLA_NESTED },
 787         [TCA_TAPRIO_ATTR_SCHED_BASE_TIME]            = { .type = NLA_S64 },
 788         [TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]         = { .type = NLA_NESTED },
 789         [TCA_TAPRIO_ATTR_SCHED_CLOCKID]              = { .type = NLA_S32 },
 790         [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]           = { .type = NLA_S64 },
 791         [TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION] = { .type = NLA_S64 },
 792         [TCA_TAPRIO_ATTR_FLAGS]                      = { .type = NLA_U32 },
 793         [TCA_TAPRIO_ATTR_TXTIME_DELAY]               = { .type = NLA_U32 },
 794 };
 795
 796 static int fill_sched_entry(struct taprio_sched *q, struct nlattr **tb,
 797                             struct sched_entry *entry,
 798                             struct netlink_ext_ack *extack)
 799 {
 800         int min_duration = length_to_duration(q, ETH_ZLEN);
 801         u32 interval = 0;
 802
 803         if (tb[TCA_TAPRIO_SCHED_ENTRY_CMD])
 804                 entry->command = nla_get_u8(
 805                         tb[TCA_TAPRIO_SCHED_ENTRY_CMD]);
 806
 807         if (tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK])
 808                 entry->gate_mask = nla_get_u32(
 809                         tb[TCA_TAPRIO_SCHED_ENTRY_GATE_MASK]);
 810
 811         if (tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL])
 812                 interval = nla_get_u32(
 813                         tb[TCA_TAPRIO_SCHED_ENTRY_INTERVAL]);
 814
 815         /* The interval should allow at least the minimum ethernet
 816          * frame to go out.
 817          */
 818         if (interval < min_duration) {
 819                 NL_SET_ERR_MSG(extack, "Invalid interval for schedule entry");
 820                 return -EINVAL;
 821         }
 822
 823         entry->interval = interval;
 824
 825         return 0;
 826 }
 827
 828 static int parse_sched_entry(struct taprio_sched *q, struct nlattr *n,
 829                              struct sched_entry *entry, int index,
 830                              struct netlink_ext_ack *extack)
 831 {
 832         struct nlattr *tb[TCA_TAPRIO_SCHED_ENTRY_MAX + 1] = { };
 833         int err;
 834
 835         err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_SCHED_ENTRY_MAX, n,
 836                                           entry_policy, NULL);
 837         if (err < 0) {
 838                 NL_SET_ERR_MSG(extack, "Could not parse nested entry");
 839                 return -EINVAL;
 840         }
 841
 842         entry->index = index;
 843
 844         return fill_sched_entry(q, tb, entry, extack);
 845 }
 846
 847 static int parse_sched_list(struct taprio_sched *q, struct nlattr *list,
 848                             struct sched_gate_list *sched,
 849                             struct netlink_ext_ack *extack)
 850 {
 851         struct nlattr *n;
 852         int err, rem;
 853         int i = 0;
 854
 855         if (!list)
 856                 return -EINVAL;
 857
 858         nla_for_each_nested(n, list, rem) {
 859                 struct sched_entry *entry;
 860
 861                 if (nla_type(n) != TCA_TAPRIO_SCHED_ENTRY) {
 862                         NL_SET_ERR_MSG(extack, "Attribute is not of type 'entry'");
 863                         continue;
 864                 }
 865
 866                 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 867                 if (!entry) {
 868                         NL_SET_ERR_MSG(extack, "Not enough memory for entry");
 869                         return -ENOMEM;
 870                 }
 871
 872                 err = parse_sched_entry(q, n, entry, i, extack);
 873                 if (err < 0) {
 874                         kfree(entry);
 875                         return err;
 876                 }
 877
 878                 list_add_tail(&entry->list, &sched->entries);
 879                 i++;
 880         }
 881
 882         sched->num_entries = i;
 883
 884         return i;
 885 }
 886
 887 static int parse_taprio_schedule(struct taprio_sched *q, struct nlattr **tb,
 888                                  struct sched_gate_list *new,
 889                                  struct netlink_ext_ack *extack)
 890 {
 891         int err = 0;
 892
 893         if (tb[TCA_TAPRIO_ATTR_SCHED_SINGLE_ENTRY]) {
 894                 NL_SET_ERR_MSG(extack, "Adding a single entry is not supported");
 895                 return -ENOTSUPP;
 896         }
 897
 898         if (tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME])
 899                 new->base_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_BASE_TIME]);
 900
 901         if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION])
 902                 new->cycle_time_extension = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION]);
 903
 904         if (tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME])
 905                 new->cycle_time = nla_get_s64(tb[TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME]);
 906
 907         if (tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST])
 908                 err = parse_sched_list(q, tb[TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST],
 909                                        new, extack);
 910         if (err < 0)
 911                 return err;
 912
 913         if (!new->cycle_time) {
 914                 struct sched_entry *entry;
 915                 ktime_t cycle = 0;
 916
 917                 list_for_each_entry(entry, &new->entries, list)
 918                         cycle = ktime_add_ns(cycle, entry->interval);
 919
 920                 if (!cycle) {
 921                         NL_SET_ERR_MSG(extack, "'cycle_time' can never be 0");
 922                         return -EINVAL;
 923                 }
 924
 925                 new->cycle_time = cycle;
 926         }
 927
 928         return 0;
 929 }
 930
 931 static int taprio_parse_mqprio_opt(struct net_device *dev,
 932                                    struct tc_mqprio_qopt *qopt,
 933                                    struct netlink_ext_ack *extack,
 934                                    u32 taprio_flags)
 935 {
 936         int i, j;
 937
 938         if (!qopt && !dev->num_tc) {
 939                 NL_SET_ERR_MSG(extack, "'mqprio' configuration is necessary");
 940                 return -EINVAL;
 941         }
 942
 943         /* If num_tc is already set, it means that the user already
 944          * configured the mqprio part
 945          */
 946         if (dev->num_tc)
 947                 return 0;
 948
 949         /* Verify num_tc is not out of max range */
 950         if (qopt->num_tc > TC_MAX_QUEUE) {
 951                 NL_SET_ERR_MSG(extack, "Number of traffic classes is outside valid range");
 952                 return -EINVAL;
 953         }
 954
 955         /* taprio imposes that traffic classes map 1:n to tx queues */
 956         if (qopt->num_tc > dev->num_tx_queues) {
 957                 NL_SET_ERR_MSG(extack, "Number of traffic classes is greater than number of HW queues");
 958                 return -EINVAL;
 959         }
 960
 961         /* Verify priority mapping uses valid tcs */
 962         for (i = 0; i <= TC_BITMASK; i++) {
 963                 if (qopt->prio_tc_map[i] >= qopt->num_tc) {
 964                         NL_SET_ERR_MSG(extack, "Invalid traffic class in priority to traffic class mapping");
 965                         return -EINVAL;
 966                 }
 967         }
 968
 969         for (i = 0; i < qopt->num_tc; i++) {
 970                 unsigned int last = qopt->offset[i] + qopt->count[i];
 971
 972                 /* Verify the queue count is in tx range being equal to the
 973                  * real_num_tx_queues indicates the last queue is in use.
 974                  */
 975                 if (qopt->offset[i] >= dev->num_tx_queues ||
 976                     !qopt->count[i] ||
 977                     last > dev->real_num_tx_queues) {
 978                         NL_SET_ERR_MSG(extack, "Invalid queue in traffic class to queue mapping");
 979                         return -EINVAL;
 980                 }
 981
 982                 if (TXTIME_ASSIST_IS_ENABLED(taprio_flags))
 983                         continue;
 984
 985                 /* Verify that the offset and counts do not overlap */
 986                 for (j = i + 1; j < qopt->num_tc; j++) {
 987                         if (last > qopt->offset[j]) {
 988                                 NL_SET_ERR_MSG(extack, "Detected overlap in the traffic class to queue mapping");
 989                                 return -EINVAL;
 990                         }
 991                 }
 992         }
 993
 994         return 0;
 995 }
 996
 997 static int taprio_get_start_time(struct Qdisc *sch,
 998                                  struct sched_gate_list *sched,
 999                                  ktime_t *start)
1000 {
1001         struct taprio_sched *q = qdisc_priv(sch);
1002         ktime_t now, base, cycle;
1003         s64 n;
1004
1005         base = sched_base_time(sched);
1006         now = taprio_get_time(q);
1007
1008         if (ktime_after(base, now)) {
1009                 *start = base;
1010                 return 0;
1011         }
1012
1013         cycle = sched->cycle_time;
1014
1015         /* The qdisc is expected to have at least one sched_entry.  Moreover,
1016          * any entry must have 'interval' > 0. Thus if the cycle time is zero,
1017          * something went really wrong. In that case, we should warn about this
1018          * inconsistent state and return error.
1019          */
1020         if (WARN_ON(!cycle))
1021                 return -EFAULT;
1022
1023         /* Schedule the start time for the beginning of the next
1024          * cycle.
1025          */
1026         n = div64_s64(ktime_sub_ns(now, base), cycle);
1027         *start = ktime_add_ns(base, (n + 1) * cycle);
1028         return 0;
1029 }
1030
1031 static void setup_first_close_time(struct taprio_sched *q,
1032                                    struct sched_gate_list *sched, ktime_t base)
1033 {
1034         struct sched_entry *first;
1035         ktime_t cycle;
1036
1037         first = list_first_entry(&sched->entries,
1038                                  struct sched_entry, list);
1039
1040         cycle = sched->cycle_time;
1041
1042         /* FIXME: find a better place to do this */
1043         sched->cycle_close_time = ktime_add_ns(base, cycle);
1044
1045         first->close_time = ktime_add_ns(base, first->interval);
1046         taprio_set_budget(q, first);
1047         rcu_assign_pointer(q->current_entry, NULL);
1048 }
1049
1050 static void taprio_start_sched(struct Qdisc *sch,
1051                                ktime_t start, struct sched_gate_list *new)
1052 {
1053         struct taprio_sched *q = qdisc_priv(sch);
1054         ktime_t expires;
1055
1056         if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1057                 return;
1058
1059         expires = hrtimer_get_expires(&q->advance_timer);
1060         if (expires == 0)
1061                 expires = KTIME_MAX;
1062
1063         /* If the new schedule starts before the next expiration, we
1064          * reprogram it to the earliest one, so we change the admin
1065          * schedule to the operational one at the right time.
1066          */
1067         start = min_t(ktime_t, start, expires);
1068
1069         hrtimer_start(&q->advance_timer, start, HRTIMER_MODE_ABS);
1070 }
1071
1072 static void taprio_set_picos_per_byte(struct net_device *dev,
1073                                       struct taprio_sched *q)
1074 {
1075         struct ethtool_link_ksettings ecmd;
1076         int speed = SPEED_10;
1077         int picos_per_byte;
1078         int err;
1079
1080         err = __ethtool_get_link_ksettings(dev, &ecmd);
1081         if (err < 0)
1082                 goto skip;
1083
1084         if (ecmd.base.speed && ecmd.base.speed != SPEED_UNKNOWN)
1085                 speed = ecmd.base.speed;
1086
1087 skip:
1088         picos_per_byte = (USEC_PER_SEC * 8) / speed;
1089
1090         atomic64_set(&q->picos_per_byte, picos_per_byte);
1091         netdev_dbg(dev, "taprio: set %s's picos_per_byte to: %lld, linkspeed: %d\n",
1092                    dev->name, (long long)atomic64_read(&q->picos_per_byte),
1093                    ecmd.base.speed);
1094 }
1095
1096 static int taprio_dev_notifier(struct notifier_block *nb, unsigned long event,
1097                                void *ptr)
1098 {
1099         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1100         struct net_device *qdev;
1101         struct taprio_sched *q;
1102         bool found = false;
1103
1104         ASSERT_RTNL();
1105
1106         if (event != NETDEV_UP && event != NETDEV_CHANGE)
1107                 return NOTIFY_DONE;
1108
1109         spin_lock(&taprio_list_lock);
1110         list_for_each_entry(q, &taprio_list, taprio_list) {
1111                 qdev = qdisc_dev(q->root);
1112                 if (qdev == dev) {
1113                         found = true;
1114                         break;
1115                 }
1116         }
1117         spin_unlock(&taprio_list_lock);
1118
1119         if (found)
1120                 taprio_set_picos_per_byte(dev, q);
1121
1122         return NOTIFY_DONE;
1123 }
1124
1125 static void setup_txtime(struct taprio_sched *q,
1126                          struct sched_gate_list *sched, ktime_t base)
1127 {
1128         struct sched_entry *entry;
1129         u32 interval = 0;
1130
1131         list_for_each_entry(entry, &sched->entries, list) {
1132                 entry->next_txtime = ktime_add_ns(base, interval);
1133                 interval += entry->interval;
1134         }
1135 }
1136
1137 static struct tc_taprio_qopt_offload *taprio_offload_alloc(int num_entries)
1138 {
1139         struct __tc_taprio_qopt_offload *__offload;
1140
1141         __offload = kzalloc(struct_size(__offload, offload.entries, num_entries),
1142                             GFP_KERNEL);
1143         if (!__offload)
1144                 return NULL;
1145
1146         refcount_set(&__offload->users, 1);
1147
1148         return &__offload->offload;
1149 }
1150
1151 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
1152                                                   *offload)
1153 {
1154         struct __tc_taprio_qopt_offload *__offload;
1155
1156         __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1157                                  offload);
1158
1159         refcount_inc(&__offload->users);
1160
1161         return offload;
1162 }
1163 EXPORT_SYMBOL_GPL(taprio_offload_get);
1164
1165 void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
1166 {
1167         struct __tc_taprio_qopt_offload *__offload;
1168
1169         __offload = container_of(offload, struct __tc_taprio_qopt_offload,
1170                                  offload);
1171
1172         if (!refcount_dec_and_test(&__offload->users))
1173                 return;
1174
1175         kfree(__offload);
1176 }
1177 EXPORT_SYMBOL_GPL(taprio_offload_free);
1178
1179 /* The function will only serve to keep the pointers to the "oper" and "admin"
1180  * schedules valid in relation to their base times, so when calling dump() the
1181  * users looks at the right schedules.
1182  * When using full offload, the admin configuration is promoted to oper at the
1183  * base_time in the PHC time domain.  But because the system time is not
1184  * necessarily in sync with that, we can't just trigger a hrtimer to call
1185  * switch_schedules at the right hardware time.
1186  * At the moment we call this by hand right away from taprio, but in the future
1187  * it will be useful to create a mechanism for drivers to notify taprio of the
1188  * offload state (PENDING, ACTIVE, INACTIVE) so it can be visible in dump().
1189  * This is left as TODO.
1190  */
1191 static void taprio_offload_config_changed(struct taprio_sched *q)
1192 {
1193         struct sched_gate_list *oper, *admin;
1194
1195         spin_lock(&q->current_entry_lock);
1196
1197         oper = rcu_dereference_protected(q->oper_sched,
1198                                          lockdep_is_held(&q->current_entry_lock));
1199         admin = rcu_dereference_protected(q->admin_sched,
1200                                           lockdep_is_held(&q->current_entry_lock));
1201
1202         switch_schedules(q, &admin, &oper);
1203
1204         spin_unlock(&q->current_entry_lock);
1205 }
1206
1207 static u32 tc_map_to_queue_mask(struct net_device *dev, u32 tc_mask)
1208 {
1209         u32 i, queue_mask = 0;
1210
1211         for (i = 0; i < dev->num_tc; i++) {
1212                 u32 offset, count;
1213
1214                 if (!(tc_mask & BIT(i)))
1215                         continue;
1216
1217                 offset = dev->tc_to_txq[i].offset;
1218                 count = dev->tc_to_txq[i].count;
1219
1220                 queue_mask |= GENMASK(offset + count - 1, offset);
1221         }
1222
1223         return queue_mask;
1224 }
1225
1226 static void taprio_sched_to_offload(struct net_device *dev,
1227                                     struct sched_gate_list *sched,
1228                                     struct tc_taprio_qopt_offload *offload)
1229 {
1230         struct sched_entry *entry;
1231         int i = 0;
1232
1233         offload->base_time = sched->base_time;
1234         offload->cycle_time = sched->cycle_time;
1235         offload->cycle_time_extension = sched->cycle_time_extension;
1236
1237         list_for_each_entry(entry, &sched->entries, list) {
1238                 struct tc_taprio_sched_entry *e = &offload->entries[i];
1239
1240                 e->command = entry->command;
1241                 e->interval = entry->interval;
1242                 e->gate_mask = tc_map_to_queue_mask(dev, entry->gate_mask);
1243
1244                 i++;
1245         }
1246
1247         offload->num_entries = i;
1248 }
1249
1250 static int taprio_enable_offload(struct net_device *dev,
1251                                  struct taprio_sched *q,
1252                                  struct sched_gate_list *sched,
1253                                  struct netlink_ext_ack *extack)
1254 {
1255         const struct net_device_ops *ops = dev->netdev_ops;
1256         struct tc_taprio_qopt_offload *offload;
1257         int err = 0;
1258
1259         if (!ops->ndo_setup_tc) {
1260                 NL_SET_ERR_MSG(extack,
1261                                "Device does not support taprio offload");
1262                 return -EOPNOTSUPP;
1263         }
1264
1265         offload = taprio_offload_alloc(sched->num_entries);
1266         if (!offload) {
1267                 NL_SET_ERR_MSG(extack,
1268                                "Not enough memory for enabling offload mode");
1269                 return -ENOMEM;
1270         }
1271         offload->enable = 1;
1272         taprio_sched_to_offload(dev, sched, offload);
1273
1274         err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1275         if (err < 0) {
1276                 NL_SET_ERR_MSG(extack,
1277                                "Device failed to setup taprio offload");
1278                 goto done;
1279         }
1280
1281 done:
1282         taprio_offload_free(offload);
1283
1284         return err;
1285 }
1286
1287 static int taprio_disable_offload(struct net_device *dev,
1288                                   struct taprio_sched *q,
1289                                   struct netlink_ext_ack *extack)
1290 {
1291         const struct net_device_ops *ops = dev->netdev_ops;
1292         struct tc_taprio_qopt_offload *offload;
1293         int err;
1294
1295         if (!FULL_OFFLOAD_IS_ENABLED(q->flags))
1296                 return 0;
1297
1298         if (!ops->ndo_setup_tc)
1299                 return -EOPNOTSUPP;
1300
1301         offload = taprio_offload_alloc(0);
1302         if (!offload) {
1303                 NL_SET_ERR_MSG(extack,
1304                                "Not enough memory to disable offload mode");
1305                 return -ENOMEM;
1306         }
1307         offload->enable = 0;
1308
1309         err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_TAPRIO, offload);
1310         if (err < 0) {
1311                 NL_SET_ERR_MSG(extack,
1312                                "Device failed to disable offload");
1313                 goto out;
1314         }
1315
1316 out:
1317         taprio_offload_free(offload);
1318
1319         return err;
1320 }
1321
1322 /* If full offload is enabled, the only possible clockid is the net device's
1323  * PHC. For that reason, specifying a clockid through netlink is incorrect.
1324  * For txtime-assist, it is implicitly assumed that the device's PHC is kept
1325  * in sync with the specified clockid via a user space daemon such as phc2sys.
1326  * For both software taprio and txtime-assist, the clockid is used for the
1327  * hrtimer that advances the schedule and hence mandatory.
1328  */
1329 static int taprio_parse_clockid(struct Qdisc *sch, struct nlattr **tb,
1330                                 struct netlink_ext_ack *extack)
1331 {
1332         struct taprio_sched *q = qdisc_priv(sch);
1333         struct net_device *dev = qdisc_dev(sch);
1334         int err = -EINVAL;
1335
1336         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1337                 const struct ethtool_ops *ops = dev->ethtool_ops;
1338                 struct ethtool_ts_info info = {
1339                         .cmd = ETHTOOL_GET_TS_INFO,
1340                         .phc_index = -1,
1341                 };
1342
1343                 if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1344                         NL_SET_ERR_MSG(extack,
1345                                        "The 'clockid' cannot be specified for full offload");
1346                         goto out;
1347                 }
1348
1349                 if (ops && ops->get_ts_info)
1350                         err = ops->get_ts_info(dev, &info);
1351
1352                 if (err || info.phc_index < 0) {
1353                         NL_SET_ERR_MSG(extack,
1354                                        "Device does not have a PTP clock");
1355                         err = -ENOTSUPP;
1356                         goto out;
1357                 }
1358         } else if (tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]) {
1359                 int clockid = nla_get_s32(tb[TCA_TAPRIO_ATTR_SCHED_CLOCKID]);
1360                 enum tk_offsets tk_offset;
1361
1362                 /* We only support static clockids and we don't allow
1363                  * for it to be modified after the first init.
1364                  */
1365                 if (clockid < 0 ||
1366                     (q->clockid != -1 && q->clockid != clockid)) {
1367                         NL_SET_ERR_MSG(extack,
1368                                        "Changing the 'clockid' of a running schedule is not supported");
1369                         err = -ENOTSUPP;
1370                         goto out;
1371                 }
1372
1373                 switch (clockid) {
1374                 case CLOCK_REALTIME:
1375                         tk_offset = TK_OFFS_REAL;
1376                         break;
1377                 case CLOCK_MONOTONIC:
1378                         tk_offset = TK_OFFS_MAX;
1379                         break;
1380                 case CLOCK_BOOTTIME:
1381                         tk_offset = TK_OFFS_BOOT;
1382                         break;
1383                 case CLOCK_TAI:
1384                         tk_offset = TK_OFFS_TAI;
1385                         break;
1386                 default:
1387                         NL_SET_ERR_MSG(extack, "Invalid 'clockid'");
1388                         err = -EINVAL;
1389                         goto out;
1390                 }
1391                 /* This pairs with READ_ONCE() in taprio_mono_to_any */
1392                 WRITE_ONCE(q->tk_offset, tk_offset);
1393
1394                 q->clockid = clockid;
1395         } else {
1396                 NL_SET_ERR_MSG(extack, "Specifying a 'clockid' is mandatory");
1397                 goto out;
1398         }
1399
1400         /* Everything went ok, return success. */
1401         err = 0;
1402
1403 out:
1404         return err;
1405 }
1406
1407 static int taprio_mqprio_cmp(const struct net_device *dev,
1408                              const struct tc_mqprio_qopt *mqprio)
1409 {
1410         int i;
1411
1412         if (!mqprio || mqprio->num_tc != dev->num_tc)
1413                 return -1;
1414
1415         for (i = 0; i < mqprio->num_tc; i++)
1416                 if (dev->tc_to_txq[i].count != mqprio->count[i] ||
1417                     dev->tc_to_txq[i].offset != mqprio->offset[i])
1418                         return -1;
1419
1420         for (i = 0; i <= TC_BITMASK; i++)
1421                 if (dev->prio_tc_map[i] != mqprio->prio_tc_map[i])
1422                         return -1;
1423
1424         return 0;
1425 }
1426
1427 /* The semantics of the 'flags' argument in relation to 'change()'
1428  * requests, are interpreted following two rules (which are applied in
1429  * this order): (1) an omitted 'flags' argument is interpreted as
1430  * zero; (2) the 'flags' of a "running" taprio instance cannot be
1431  * changed.
1432  */
1433 static int taprio_new_flags(const struct nlattr *attr, u32 old,
1434                             struct netlink_ext_ack *extack)
1435 {
1436         u32 new = 0;
1437
1438         if (attr)
1439                 new = nla_get_u32(attr);
1440
1441         if (old != TAPRIO_FLAGS_INVALID && old != new) {
1442                 NL_SET_ERR_MSG_MOD(extack, "Changing 'flags' of a running schedule is not supported");
1443                 return -EOPNOTSUPP;
1444         }
1445
1446         if (!taprio_flags_valid(new)) {
1447                 NL_SET_ERR_MSG_MOD(extack, "Specified 'flags' are not valid");
1448                 return -EINVAL;
1449         }
1450
1451         return new;
1452 }
1453
1454 static int taprio_change(struct Qdisc *sch, struct nlattr *opt,
1455                          struct netlink_ext_ack *extack)
1456 {
1457         struct nlattr *tb[TCA_TAPRIO_ATTR_MAX + 1] = { };
1458         struct sched_gate_list *oper, *admin, *new_admin;
1459         struct taprio_sched *q = qdisc_priv(sch);
1460         struct net_device *dev = qdisc_dev(sch);
1461         struct tc_mqprio_qopt *mqprio = NULL;
1462         unsigned long flags;
1463         ktime_t start;
1464         int i, err;
1465
1466         err = nla_parse_nested_deprecated(tb, TCA_TAPRIO_ATTR_MAX, opt,
1467                                           taprio_policy, extack);
1468         if (err < 0)
1469                 return err;
1470
1471         if (tb[TCA_TAPRIO_ATTR_PRIOMAP])
1472                 mqprio = nla_data(tb[TCA_TAPRIO_ATTR_PRIOMAP]);
1473
1474         err = taprio_new_flags(tb[TCA_TAPRIO_ATTR_FLAGS],
1475                                q->flags, extack);
1476         if (err < 0)
1477                 return err;
1478
1479         q->flags = err;
1480
1481         err = taprio_parse_mqprio_opt(dev, mqprio, extack, q->flags);
1482         if (err < 0)
1483                 return err;
1484
1485         new_admin = kzalloc(sizeof(*new_admin), GFP_KERNEL);
1486         if (!new_admin) {
1487                 NL_SET_ERR_MSG(extack, "Not enough memory for a new schedule");
1488                 return -ENOMEM;
1489         }
1490         INIT_LIST_HEAD(&new_admin->entries);
1491
1492         rcu_read_lock();
1493         oper = rcu_dereference(q->oper_sched);
1494         admin = rcu_dereference(q->admin_sched);
1495         rcu_read_unlock();
1496
1497         /* no changes - no new mqprio settings */
1498         if (!taprio_mqprio_cmp(dev, mqprio))
1499                 mqprio = NULL;
1500
1501         if (mqprio && (oper || admin)) {
1502                 NL_SET_ERR_MSG(extack, "Changing the traffic mapping of a running schedule is not supported");
1503                 err = -ENOTSUPP;
1504                 goto free_sched;
1505         }
1506
1507         err = parse_taprio_schedule(q, tb, new_admin, extack);
1508         if (err < 0)
1509                 goto free_sched;
1510
1511         if (new_admin->num_entries == 0) {
1512                 NL_SET_ERR_MSG(extack, "There should be at least one entry in the schedule");
1513                 err = -EINVAL;
1514                 goto free_sched;
1515         }
1516
1517         err = taprio_parse_clockid(sch, tb, extack);
1518         if (err < 0)
1519                 goto free_sched;
1520
1521         taprio_set_picos_per_byte(dev, q);
1522
1523         if (mqprio) {
1524                 err = netdev_set_num_tc(dev, mqprio->num_tc);
1525                 if (err)
1526                         goto free_sched;
1527                 for (i = 0; i < mqprio->num_tc; i++)
1528                         netdev_set_tc_queue(dev, i,
1529                                             mqprio->count[i],
1530                                             mqprio->offset[i]);
1531
1532                 /* Always use supplied priority mappings */
1533                 for (i = 0; i <= TC_BITMASK; i++)
1534                         netdev_set_prio_tc_map(dev, i,
1535                                                mqprio->prio_tc_map[i]);
1536         }
1537
1538         if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1539                 err = taprio_enable_offload(dev, q, new_admin, extack);
1540         else
1541                 err = taprio_disable_offload(dev, q, extack);
1542         if (err)
1543                 goto free_sched;
1544
1545         /* Protects against enqueue()/dequeue() */
1546         spin_lock_bh(qdisc_lock(sch));
1547
1548         if (tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]) {
1549                 if (!TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1550                         NL_SET_ERR_MSG_MOD(extack, "txtime-delay can only be set when txtime-assist mode is enabled");
1551                         err = -EINVAL;
1552                         goto unlock;
1553                 }
1554
1555                 q->txtime_delay = nla_get_u32(tb[TCA_TAPRIO_ATTR_TXTIME_DELAY]);
1556         }
1557
1558         if (!TXTIME_ASSIST_IS_ENABLED(q->flags) &&
1559             !FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1560             !hrtimer_active(&q->advance_timer)) {
1561                 hrtimer_init(&q->advance_timer, q->clockid, HRTIMER_MODE_ABS);
1562                 q->advance_timer.function = advance_sched;
1563         }
1564
1565         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1566                 q->dequeue = taprio_dequeue_offload;
1567                 q->peek = taprio_peek_offload;
1568         } else {
1569                 /* Be sure to always keep the function pointers
1570                  * in a consistent state.
1571                  */
1572                 q->dequeue = taprio_dequeue_soft;
1573                 q->peek = taprio_peek_soft;
1574         }
1575
1576         err = taprio_get_start_time(sch, new_admin, &start);
1577         if (err < 0) {
1578                 NL_SET_ERR_MSG(extack, "Internal error: failed get start time");
1579                 goto unlock;
1580         }
1581
1582         setup_txtime(q, new_admin, start);
1583
1584         if (TXTIME_ASSIST_IS_ENABLED(q->flags)) {
1585                 if (!oper) {
1586                         rcu_assign_pointer(q->oper_sched, new_admin);
1587                         err = 0;
1588                         new_admin = NULL;
1589                         goto unlock;
1590                 }
1591
1592                 rcu_assign_pointer(q->admin_sched, new_admin);
1593                 if (admin)
1594                         call_rcu(&admin->rcu, taprio_free_sched_cb);
1595         } else {
1596                 setup_first_close_time(q, new_admin, start);
1597
1598                 /* Protects against advance_sched() */
1599                 spin_lock_irqsave(&q->current_entry_lock, flags);
1600
1601                 taprio_start_sched(sch, start, new_admin);
1602
1603                 rcu_assign_pointer(q->admin_sched, new_admin);
1604                 if (admin)
1605                         call_rcu(&admin->rcu, taprio_free_sched_cb);
1606
1607                 spin_unlock_irqrestore(&q->current_entry_lock, flags);
1608
1609                 if (FULL_OFFLOAD_IS_ENABLED(q->flags))
1610                         taprio_offload_config_changed(q);
1611         }
1612
1613         new_admin = NULL;
1614         err = 0;
1615
1616 unlock:
1617         spin_unlock_bh(qdisc_lock(sch));
1618
1619 free_sched:
1620         if (new_admin)
1621                 call_rcu(&new_admin->rcu, taprio_free_sched_cb);
1622
1623         return err;
1624 }
1625
1626 static void taprio_reset(struct Qdisc *sch)
1627 {
1628         struct taprio_sched *q = qdisc_priv(sch);
1629         struct net_device *dev = qdisc_dev(sch);
1630         int i;
1631
1632         hrtimer_cancel(&q->advance_timer);
1633         if (q->qdiscs) {
1634                 for (i = 0; i < dev->num_tx_queues; i++)
1635                         if (q->qdiscs[i])
1636                                 qdisc_reset(q->qdiscs[i]);
1637         }
1638         sch->qstats.backlog = 0;
1639         sch->q.qlen = 0;
1640 }
1641
1642 static void taprio_destroy(struct Qdisc *sch)
1643 {
1644         struct taprio_sched *q = qdisc_priv(sch);
1645         struct net_device *dev = qdisc_dev(sch);
1646         unsigned int i;
1647
1648         spin_lock(&taprio_list_lock);
1649         list_del(&q->taprio_list);
1650         spin_unlock(&taprio_list_lock);
1651
1652         /* Note that taprio_reset() might not be called if an error
1653          * happens in qdisc_create(), after taprio_init() has been called.
1654          */
1655         hrtimer_cancel(&q->advance_timer);
1656
1657         taprio_disable_offload(dev, q, NULL);
1658
1659         if (q->qdiscs) {
1660                 for (i = 0; i < dev->num_tx_queues; i++)
1661                         qdisc_put(q->qdiscs[i]);
1662
1663                 kfree(q->qdiscs);
1664         }
1665         q->qdiscs = NULL;
1666
1667         netdev_reset_tc(dev);
1668
1669         if (q->oper_sched)
1670                 call_rcu(&q->oper_sched->rcu, taprio_free_sched_cb);
1671
1672         if (q->admin_sched)
1673                 call_rcu(&q->admin_sched->rcu, taprio_free_sched_cb);
1674 }
1675
1676 static int taprio_init(struct Qdisc *sch, struct nlattr *opt,
1677                        struct netlink_ext_ack *extack)
1678 {
1679         struct taprio_sched *q = qdisc_priv(sch);
1680         struct net_device *dev = qdisc_dev(sch);
1681         int i;
1682
1683         spin_lock_init(&q->current_entry_lock);
1684
1685         hrtimer_init(&q->advance_timer, CLOCK_TAI, HRTIMER_MODE_ABS);
1686         q->advance_timer.function = advance_sched;
1687
1688         q->dequeue = taprio_dequeue_soft;
1689         q->peek = taprio_peek_soft;
1690
1691         q->root = sch;
1692
1693         /* We only support static clockids. Use an invalid value as default
1694          * and get the valid one on taprio_change().
1695          */
1696         q->clockid = -1;
1697         q->flags = TAPRIO_FLAGS_INVALID;
1698
1699         spin_lock(&taprio_list_lock);
1700         list_add(&q->taprio_list, &taprio_list);
1701         spin_unlock(&taprio_list_lock);
1702
1703         if (sch->parent != TC_H_ROOT)
1704                 return -EOPNOTSUPP;
1705
1706         if (!netif_is_multiqueue(dev))
1707                 return -EOPNOTSUPP;
1708
1709         /* pre-allocate qdisc, attachment can't fail */
1710         q->qdiscs = kcalloc(dev->num_tx_queues,
1711                             sizeof(q->qdiscs[0]),
1712                             GFP_KERNEL);
1713
1714         if (!q->qdiscs)
1715                 return -ENOMEM;
1716
1717         if (!opt)
1718                 return -EINVAL;
1719
1720         for (i = 0; i < dev->num_tx_queues; i++) {
1721                 struct netdev_queue *dev_queue;
1722                 struct Qdisc *qdisc;
1723
1724                 dev_queue = netdev_get_tx_queue(dev, i);
1725                 qdisc = qdisc_create_dflt(dev_queue,
1726                                           &pfifo_qdisc_ops,
1727                                           TC_H_MAKE(TC_H_MAJ(sch->handle),
1728                                                     TC_H_MIN(i + 1)),
1729                                           extack);
1730                 if (!qdisc)
1731                         return -ENOMEM;
1732
1733                 if (i < dev->real_num_tx_queues)
1734                         qdisc_hash_add(qdisc, false);
1735
1736                 q->qdiscs[i] = qdisc;
1737         }
1738
1739         return taprio_change(sch, opt, extack);
1740 }
1741
1742 static void taprio_attach(struct Qdisc *sch)
1743 {
1744         struct taprio_sched *q = qdisc_priv(sch);
1745         struct net_device *dev = qdisc_dev(sch);
1746         unsigned int ntx;
1747
1748         /* Attach underlying qdisc */
1749         for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
1750                 struct Qdisc *qdisc = q->qdiscs[ntx];
1751                 struct Qdisc *old;
1752
1753                 if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1754                         qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1755                         old = dev_graft_qdisc(qdisc->dev_queue, qdisc);
1756                 } else {
1757                         old = dev_graft_qdisc(qdisc->dev_queue, sch);
1758                         qdisc_refcount_inc(sch);
1759                 }
1760                 if (old)
1761                         qdisc_put(old);
1762         }
1763
1764         /* access to the child qdiscs is not needed in offload mode */
1765         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1766                 kfree(q->qdiscs);
1767                 q->qdiscs = NULL;
1768         }
1769 }
1770
1771 static struct netdev_queue *taprio_queue_get(struct Qdisc *sch,
1772                                              unsigned long cl)
1773 {
1774         struct net_device *dev = qdisc_dev(sch);
1775         unsigned long ntx = cl - 1;
1776
1777         if (ntx >= dev->num_tx_queues)
1778                 return NULL;
1779
1780         return netdev_get_tx_queue(dev, ntx);
1781 }
1782
1783 static int taprio_graft(struct Qdisc *sch, unsigned long cl,
1784                         struct Qdisc *new, struct Qdisc **old,
1785                         struct netlink_ext_ack *extack)
1786 {
1787         struct taprio_sched *q = qdisc_priv(sch);
1788         struct net_device *dev = qdisc_dev(sch);
1789         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1790
1791         if (!dev_queue)
1792                 return -EINVAL;
1793
1794         if (dev->flags & IFF_UP)
1795                 dev_deactivate(dev);
1796
1797         if (FULL_OFFLOAD_IS_ENABLED(q->flags)) {
1798                 *old = dev_graft_qdisc(dev_queue, new);
1799         } else {
1800                 *old = q->qdiscs[cl - 1];
1801                 q->qdiscs[cl - 1] = new;
1802         }
1803
1804         if (new)
1805                 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1806
1807         if (dev->flags & IFF_UP)
1808                 dev_activate(dev);
1809
1810         return 0;
1811 }
1812
1813 static int dump_entry(struct sk_buff *msg,
1814                       const struct sched_entry *entry)
1815 {
1816         struct nlattr *item;
1817
1818         item = nla_nest_start_noflag(msg, TCA_TAPRIO_SCHED_ENTRY);
1819         if (!item)
1820                 return -ENOSPC;
1821
1822         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INDEX, entry->index))
1823                 goto nla_put_failure;
1824
1825         if (nla_put_u8(msg, TCA_TAPRIO_SCHED_ENTRY_CMD, entry->command))
1826                 goto nla_put_failure;
1827
1828         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_GATE_MASK,
1829                         entry->gate_mask))
1830                 goto nla_put_failure;
1831
1832         if (nla_put_u32(msg, TCA_TAPRIO_SCHED_ENTRY_INTERVAL,
1833                         entry->interval))
1834                 goto nla_put_failure;
1835
1836         return nla_nest_end(msg, item);
1837
1838 nla_put_failure:
1839         nla_nest_cancel(msg, item);
1840         return -1;
1841 }
1842
1843 static int dump_schedule(struct sk_buff *msg,
1844                          const struct sched_gate_list *root)
1845 {
1846         struct nlattr *entry_list;
1847         struct sched_entry *entry;
1848
1849         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_BASE_TIME,
1850                         root->base_time, TCA_TAPRIO_PAD))
1851                 return -1;
1852
1853         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME,
1854                         root->cycle_time, TCA_TAPRIO_PAD))
1855                 return -1;
1856
1857         if (nla_put_s64(msg, TCA_TAPRIO_ATTR_SCHED_CYCLE_TIME_EXTENSION,
1858                         root->cycle_time_extension, TCA_TAPRIO_PAD))
1859                 return -1;
1860
1861         entry_list = nla_nest_start_noflag(msg,
1862                                            TCA_TAPRIO_ATTR_SCHED_ENTRY_LIST);
1863         if (!entry_list)
1864                 goto error_nest;
1865
1866         list_for_each_entry(entry, &root->entries, list) {
1867                 if (dump_entry(msg, entry) < 0)
1868                         goto error_nest;
1869         }
1870
1871         nla_nest_end(msg, entry_list);
1872         return 0;
1873
1874 error_nest:
1875         nla_nest_cancel(msg, entry_list);
1876         return -1;
1877 }
1878
1879 static int taprio_dump(struct Qdisc *sch, struct sk_buff *skb)
1880 {
1881         struct taprio_sched *q = qdisc_priv(sch);
1882         struct net_device *dev = qdisc_dev(sch);
1883         struct sched_gate_list *oper, *admin;
1884         struct tc_mqprio_qopt opt = { 0 };
1885         struct nlattr *nest, *sched_nest;
1886         unsigned int i;
1887
1888         rcu_read_lock();
1889         oper = rcu_dereference(q->oper_sched);
1890         admin = rcu_dereference(q->admin_sched);
1891
1892         opt.num_tc = netdev_get_num_tc(dev);
1893         memcpy(opt.prio_tc_map, dev->prio_tc_map, sizeof(opt.prio_tc_map));
1894
1895         for (i = 0; i < netdev_get_num_tc(dev); i++) {
1896                 opt.count[i] = dev->tc_to_txq[i].count;
1897                 opt.offset[i] = dev->tc_to_txq[i].offset;
1898         }
1899
1900         nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
1901         if (!nest)
1902                 goto start_error;
1903
1904         if (nla_put(skb, TCA_TAPRIO_ATTR_PRIOMAP, sizeof(opt), &opt))
1905                 goto options_error;
1906
1907         if (!FULL_OFFLOAD_IS_ENABLED(q->flags) &&
1908             nla_put_s32(skb, TCA_TAPRIO_ATTR_SCHED_CLOCKID, q->clockid))
1909                 goto options_error;
1910
1911         if (q->flags && nla_put_u32(skb, TCA_TAPRIO_ATTR_FLAGS, q->flags))
1912                 goto options_error;
1913
1914         if (q->txtime_delay &&
1915             nla_put_u32(skb, TCA_TAPRIO_ATTR_TXTIME_DELAY, q->txtime_delay))
1916                 goto options_error;
1917
1918         if (oper && dump_schedule(skb, oper))
1919                 goto options_error;
1920
1921         if (!admin)
1922                 goto done;
1923
1924         sched_nest = nla_nest_start_noflag(skb, TCA_TAPRIO_ATTR_ADMIN_SCHED);
1925         if (!sched_nest)
1926                 goto options_error;
1927
1928         if (dump_schedule(skb, admin))
1929                 goto admin_error;
1930
1931         nla_nest_end(skb, sched_nest);
1932
1933 done:
1934         rcu_read_unlock();
1935
1936         return nla_nest_end(skb, nest);
1937
1938 admin_error:
1939         nla_nest_cancel(skb, sched_nest);
1940
1941 options_error:
1942         nla_nest_cancel(skb, nest);
1943
1944 start_error:
1945         rcu_read_unlock();
1946         return -ENOSPC;
1947 }
1948
1949 static struct Qdisc *taprio_leaf(struct Qdisc *sch, unsigned long cl)
1950 {
1951         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1952
1953         if (!dev_queue)
1954                 return NULL;
1955
1956         return dev_queue->qdisc_sleeping;
1957 }
1958
1959 static unsigned long taprio_find(struct Qdisc *sch, u32 classid)
1960 {
1961         unsigned int ntx = TC_H_MIN(classid);
1962
1963         if (!taprio_queue_get(sch, ntx))
1964                 return 0;
1965         return ntx;
1966 }
1967
1968 static int taprio_dump_class(struct Qdisc *sch, unsigned long cl,
1969                              struct sk_buff *skb, struct tcmsg *tcm)
1970 {
1971         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1972
1973         tcm->tcm_parent = TC_H_ROOT;
1974         tcm->tcm_handle |= TC_H_MIN(cl);
1975         tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
1976
1977         return 0;
1978 }
1979
1980 static int taprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
1981                                    struct gnet_dump *d)
1982         __releases(d->lock)
1983         __acquires(d->lock)
1984 {
1985         struct netdev_queue *dev_queue = taprio_queue_get(sch, cl);
1986
1987         sch = dev_queue->qdisc_sleeping;
1988         if (gnet_stats_copy_basic(d, NULL, &sch->bstats, true) < 0 ||
1989             qdisc_qstats_copy(d, sch) < 0)
1990                 return -1;
1991         return 0;
1992 }
1993
1994 static void taprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
1995 {
1996         struct net_device *dev = qdisc_dev(sch);
1997         unsigned long ntx;
1998
1999         if (arg->stop)
2000                 return;
2001
2002         arg->count = arg->skip;
2003         for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
2004                 if (arg->fn(sch, ntx + 1, arg) < 0) {
2005                         arg->stop = 1;
2006                         break;
2007                 }
2008                 arg->count++;
2009         }
2010 }
2011
2012 static struct netdev_queue *taprio_select_queue(struct Qdisc *sch,
2013                                                 struct tcmsg *tcm)
2014 {
2015         return taprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
2016 }
2017
2018 static const struct Qdisc_class_ops taprio_class_ops = {
2019         .graft          = taprio_graft,
2020         .leaf           = taprio_leaf,
2021         .find           = taprio_find,
2022         .walk           = taprio_walk,
2023         .dump           = taprio_dump_class,
2024         .dump_stats     = taprio_dump_class_stats,
2025         .select_queue   = taprio_select_queue,
2026 };
2027
2028 static struct Qdisc_ops taprio_qdisc_ops __read_mostly = {
2029         .cl_ops         = &taprio_class_ops,
2030         .id             = "taprio",
2031         .priv_size      = sizeof(struct taprio_sched),
2032         .init           = taprio_init,
2033         .change         = taprio_change,
2034         .destroy        = taprio_destroy,
2035         .reset          = taprio_reset,
2036         .attach         = taprio_attach,
2037         .peek           = taprio_peek,
2038         .dequeue        = taprio_dequeue,
2039         .enqueue        = taprio_enqueue,
2040         .dump           = taprio_dump,
2041         .owner          = THIS_MODULE,
2042 };
2043
2044 static struct notifier_block taprio_device_notifier = {
2045         .notifier_call = taprio_dev_notifier,
2046 };
2047
2048 static int __init taprio_module_init(void)
2049 {
2050         int err = register_netdevice_notifier(&taprio_device_notifier);
2051
2052         if (err)
2053                 return err;
2054
2055         return register_qdisc(&taprio_qdisc_ops);
2056 }
2057
2058 static void __exit taprio_module_exit(void)
2059 {
2060         unregister_qdisc(&taprio_qdisc_ops);
2061         unregister_netdevice_notifier(&taprio_device_notifier);
2062 }
2063
2064 module_init(taprio_module_init);
2065 module_exit(taprio_module_exit);
2066 MODULE_LICENSE("GPL");