1 /* 2 * net/sched/sch_qfq.c Quick Fair Queueing Plus Scheduler. 3 * 4 * Copyright (c) 2009 Fabio Checconi, Luigi Rizzo, and Paolo Valente. 5 * Copyright (c) 2012 Paolo Valente. 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License 9 * version 2 as published by the Free Software Foundation. 10 */ 11 12 #include <linux/module.h> 13 #include <linux/init.h> 14 #include <linux/bitops.h> 15 #include <linux/errno.h> 16 #include <linux/netdevice.h> 17 #include <linux/pkt_sched.h> 18 #include <net/sch_generic.h> 19 #include <net/pkt_sched.h> 20 #include <net/pkt_cls.h> 21 22 23 /* Quick Fair Queueing Plus 24 ======================== 25 26 Sources: 27 28 [1] Paolo Valente, 29 "Reducing the Execution Time of Fair-Queueing Schedulers." 30 http://algo.ing.unimo.it/people/paolo/agg-sched/agg-sched.pdf 31 32 Sources for QFQ: 33 34 [2] Fabio Checconi, Luigi Rizzo, and Paolo Valente: "QFQ: Efficient 35 Packet Scheduling with Tight Bandwidth Distribution Guarantees." 36 37 See also: 38 http://retis.sssup.it/~fabio/linux/qfq/ 39 */ 40 41 /* 42 43 QFQ+ divides classes into aggregates of at most MAX_AGG_CLASSES 44 classes. Each aggregate is timestamped with a virtual start time S 45 and a virtual finish time F, and scheduled according to its 46 timestamps. S and F are computed as a function of a system virtual 47 time function V. The classes within each aggregate are instead 48 scheduled with DRR. 49 50 To speed up operations, QFQ+ divides also aggregates into a limited 51 number of groups. Which group a class belongs to depends on the 52 ratio between the maximum packet length for the class and the weight 53 of the class. Groups have their own S and F. In the end, QFQ+ 54 schedules groups, then aggregates within groups, then classes within 55 aggregates. See [1] and [2] for a full description. 56 57 Virtual time computations. 58 59 S, F and V are all computed in fixed point arithmetic with 60 FRAC_BITS decimal bits. 61 62 QFQ_MAX_INDEX is the maximum index allowed for a group. We need 63 one bit per index. 64 QFQ_MAX_WSHIFT is the maximum power of two supported as a weight. 65 66 The layout of the bits is as below: 67 68 [ MTU_SHIFT ][ FRAC_BITS ] 69 [ MAX_INDEX ][ MIN_SLOT_SHIFT ] 70 ^.__grp->index = 0 71 *.__grp->slot_shift 72 73 where MIN_SLOT_SHIFT is derived by difference from the others. 74 75 The max group index corresponds to Lmax/w_min, where 76 Lmax=1<<MTU_SHIFT, w_min = 1 . 77 From this, and knowing how many groups (MAX_INDEX) we want, 78 we can derive the shift corresponding to each group. 79 80 Because we often need to compute 81 F = S + len/w_i and V = V + len/wsum 82 instead of storing w_i store the value 83 inv_w = (1<<FRAC_BITS)/w_i 84 so we can do F = S + len * inv_w * wsum. 85 We use W_TOT in the formulas so we can easily move between 86 static and adaptive weight sum. 87 88 The per-scheduler-instance data contain all the data structures 89 for the scheduler: bitmaps and bucket lists. 90 91 */ 92 93 /* 94 * Maximum number of consecutive slots occupied by backlogged classes 95 * inside a group. 96 */ 97 #define QFQ_MAX_SLOTS 32 98 99 /* 100 * Shifts used for aggregate<->group mapping. We allow class weights that are 101 * in the range [1, 2^MAX_WSHIFT], and we try to map each aggregate i to the 102 * group with the smallest index that can support the L_i / r_i configured 103 * for the classes in the aggregate. 104 * 105 * grp->index is the index of the group; and grp->slot_shift 106 * is the shift for the corresponding (scaled) sigma_i. 107 */ 108 #define QFQ_MAX_INDEX 24 109 #define QFQ_MAX_WSHIFT 10 110 111 #define QFQ_MAX_WEIGHT (1<<QFQ_MAX_WSHIFT) /* see qfq_slot_insert */ 112 #define QFQ_MAX_WSUM (64*QFQ_MAX_WEIGHT) 113 114 #define FRAC_BITS 30 /* fixed point arithmetic */ 115 #define ONE_FP (1UL << FRAC_BITS) 116 117 #define QFQ_MTU_SHIFT 16 /* to support TSO/GSO */ 118 #define QFQ_MIN_LMAX 512 /* see qfq_slot_insert */ 119 120 #define QFQ_MAX_AGG_CLASSES 8 /* max num classes per aggregate allowed */ 121 122 /* 123 * Possible group states. These values are used as indexes for the bitmaps 124 * array of struct qfq_queue. 125 */ 126 enum qfq_state { ER, IR, EB, IB, QFQ_MAX_STATE }; 127 128 struct qfq_group; 129 130 struct qfq_aggregate; 131 132 struct qfq_class { 133 struct Qdisc_class_common common; 134 135 unsigned int filter_cnt; 136 137 struct gnet_stats_basic_packed bstats; 138 struct gnet_stats_queue qstats; 139 struct net_rate_estimator __rcu *rate_est; 140 struct Qdisc *qdisc; 141 struct list_head alist; /* Link for active-classes list. */ 142 struct qfq_aggregate *agg; /* Parent aggregate. */ 143 int deficit; /* DRR deficit counter. */ 144 }; 145 146 struct qfq_aggregate { 147 struct hlist_node next; /* Link for the slot list. */ 148 u64 S, F; /* flow timestamps (exact) */ 149 150 /* group we belong to. In principle we would need the index, 151 * which is log_2(lmax/weight), but we never reference it 152 * directly, only the group. 153 */ 154 struct qfq_group *grp; 155 156 /* these are copied from the flowset. */ 157 u32 class_weight; /* Weight of each class in this aggregate. */ 158 /* Max pkt size for the classes in this aggregate, DRR quantum. */ 159 int lmax; 160 161 u32 inv_w; /* ONE_FP/(sum of weights of classes in aggr.). */ 162 u32 budgetmax; /* Max budget for this aggregate. */ 163 u32 initial_budget, budget; /* Initial and current budget. */ 164 165 int num_classes; /* Number of classes in this aggr. */ 166 struct list_head active; /* DRR queue of active classes. */ 167 168 struct hlist_node nonfull_next; /* See nonfull_aggs in qfq_sched. */ 169 }; 170 171 struct qfq_group { 172 u64 S, F; /* group timestamps (approx). */ 173 unsigned int slot_shift; /* Slot shift. */ 174 unsigned int index; /* Group index. */ 175 unsigned int front; /* Index of the front slot. */ 176 unsigned long full_slots; /* non-empty slots */ 177 178 /* Array of RR lists of active aggregates. */ 179 struct hlist_head slots[QFQ_MAX_SLOTS]; 180 }; 181 182 struct qfq_sched { 183 struct tcf_proto __rcu *filter_list; 184 struct tcf_block *block; 185 struct Qdisc_class_hash clhash; 186 187 u64 oldV, V; /* Precise virtual times. */ 188 struct qfq_aggregate *in_serv_agg; /* Aggregate being served. */ 189 u32 wsum; /* weight sum */ 190 u32 iwsum; /* inverse weight sum */ 191 192 unsigned long bitmaps[QFQ_MAX_STATE]; /* Group bitmaps. */ 193 struct qfq_group groups[QFQ_MAX_INDEX + 1]; /* The groups. */ 194 u32 min_slot_shift; /* Index of the group-0 bit in the bitmaps. */ 195 196 u32 max_agg_classes; /* Max number of classes per aggr. */ 197 struct hlist_head nonfull_aggs; /* Aggs with room for more classes. */ 198 }; 199 200 /* 201 * Possible reasons why the timestamps of an aggregate are updated 202 * enqueue: the aggregate switches from idle to active and must scheduled 203 * for service 204 * requeue: the aggregate finishes its budget, so it stops being served and 205 * must be rescheduled for service 206 */ 207 enum update_reason {enqueue, requeue}; 208 209 static struct qfq_class *qfq_find_class(struct Qdisc *sch, u32 classid) 210 { 211 struct qfq_sched *q = qdisc_priv(sch); 212 struct Qdisc_class_common *clc; 213 214 clc = qdisc_class_find(&q->clhash, classid); 215 if (clc == NULL) 216 return NULL; 217 return container_of(clc, struct qfq_class, common); 218 } 219 220 static void qfq_purge_queue(struct qfq_class *cl) 221 { 222 unsigned int len = cl->qdisc->q.qlen; 223 unsigned int backlog = cl->qdisc->qstats.backlog; 224 225 qdisc_reset(cl->qdisc); 226 qdisc_tree_reduce_backlog(cl->qdisc, len, backlog); 227 } 228 229 static const struct nla_policy qfq_policy[TCA_QFQ_MAX + 1] = { 230 [TCA_QFQ_WEIGHT] = { .type = NLA_U32 }, 231 [TCA_QFQ_LMAX] = { .type = NLA_U32 }, 232 }; 233 234 /* 235 * Calculate a flow index, given its weight and maximum packet length. 236 * index = log_2(maxlen/weight) but we need to apply the scaling. 237 * This is used only once at flow creation. 238 */ 239 static int qfq_calc_index(u32 inv_w, unsigned int maxlen, u32 min_slot_shift) 240 { 241 u64 slot_size = (u64)maxlen * inv_w; 242 unsigned long size_map; 243 int index = 0; 244 245 size_map = slot_size >> min_slot_shift; 246 if (!size_map) 247 goto out; 248 249 index = __fls(size_map) + 1; /* basically a log_2 */ 250 index -= !(slot_size - (1ULL << (index + min_slot_shift - 1))); 251 252 if (index < 0) 253 index = 0; 254 out: 255 pr_debug("qfq calc_index: W = %lu, L = %u, I = %d\n", 256 (unsigned long) ONE_FP/inv_w, maxlen, index); 257 258 return index; 259 } 260 261 static void qfq_deactivate_agg(struct qfq_sched *, struct qfq_aggregate *); 262 static void qfq_activate_agg(struct qfq_sched *, struct qfq_aggregate *, 263 enum update_reason); 264 265 static void qfq_init_agg(struct qfq_sched *q, struct qfq_aggregate *agg, 266 u32 lmax, u32 weight) 267 { 268 INIT_LIST_HEAD(&agg->active); 269 hlist_add_head(&agg->nonfull_next, &q->nonfull_aggs); 270 271 agg->lmax = lmax; 272 agg->class_weight = weight; 273 } 274 275 static struct qfq_aggregate *qfq_find_agg(struct qfq_sched *q, 276 u32 lmax, u32 weight) 277 { 278 struct qfq_aggregate *agg; 279 280 hlist_for_each_entry(agg, &q->nonfull_aggs, nonfull_next) 281 if (agg->lmax == lmax && agg->class_weight == weight) 282 return agg; 283 284 return NULL; 285 } 286 287 288 /* Update aggregate as a function of the new number of classes. */ 289 static void qfq_update_agg(struct qfq_sched *q, struct qfq_aggregate *agg, 290 int new_num_classes) 291 { 292 u32 new_agg_weight; 293 294 if (new_num_classes == q->max_agg_classes) 295 hlist_del_init(&agg->nonfull_next); 296 297 if (agg->num_classes > new_num_classes && 298 new_num_classes == q->max_agg_classes - 1) /* agg no more full */ 299 hlist_add_head(&agg->nonfull_next, &q->nonfull_aggs); 300 301 /* The next assignment may let 302 * agg->initial_budget > agg->budgetmax 303 * hold, we will take it into account in charge_actual_service(). 304 */ 305 agg->budgetmax = new_num_classes * agg->lmax; 306 new_agg_weight = agg->class_weight * new_num_classes; 307 agg->inv_w = ONE_FP/new_agg_weight; 308 309 if (agg->grp == NULL) { 310 int i = qfq_calc_index(agg->inv_w, agg->budgetmax, 311 q->min_slot_shift); 312 agg->grp = &q->groups[i]; 313 } 314 315 q->wsum += 316 (int) agg->class_weight * (new_num_classes - agg->num_classes); 317 q->iwsum = ONE_FP / q->wsum; 318 319 agg->num_classes = new_num_classes; 320 } 321 322 /* Add class to aggregate. */ 323 static void qfq_add_to_agg(struct qfq_sched *q, 324 struct qfq_aggregate *agg, 325 struct qfq_class *cl) 326 { 327 cl->agg = agg; 328 329 qfq_update_agg(q, agg, agg->num_classes+1); 330 if (cl->qdisc->q.qlen > 0) { /* adding an active class */ 331 list_add_tail(&cl->alist, &agg->active); 332 if (list_first_entry(&agg->active, struct qfq_class, alist) == 333 cl && q->in_serv_agg != agg) /* agg was inactive */ 334 qfq_activate_agg(q, agg, enqueue); /* schedule agg */ 335 } 336 } 337 338 static struct qfq_aggregate *qfq_choose_next_agg(struct qfq_sched *); 339 340 static void qfq_destroy_agg(struct qfq_sched *q, struct qfq_aggregate *agg) 341 { 342 hlist_del_init(&agg->nonfull_next); 343 q->wsum -= agg->class_weight; 344 if (q->wsum != 0) 345 q->iwsum = ONE_FP / q->wsum; 346 347 if (q->in_serv_agg == agg) 348 q->in_serv_agg = qfq_choose_next_agg(q); 349 kfree(agg); 350 } 351 352 /* Deschedule class from within its parent aggregate. */ 353 static void qfq_deactivate_class(struct qfq_sched *q, struct qfq_class *cl) 354 { 355 struct qfq_aggregate *agg = cl->agg; 356 357 358 list_del(&cl->alist); /* remove from RR queue of the aggregate */ 359 if (list_empty(&agg->active)) /* agg is now inactive */ 360 qfq_deactivate_agg(q, agg); 361 } 362 363 /* Remove class from its parent aggregate. */ 364 static void qfq_rm_from_agg(struct qfq_sched *q, struct qfq_class *cl) 365 { 366 struct qfq_aggregate *agg = cl->agg; 367 368 cl->agg = NULL; 369 if (agg->num_classes == 1) { /* agg being emptied, destroy it */ 370 qfq_destroy_agg(q, agg); 371 return; 372 } 373 qfq_update_agg(q, agg, agg->num_classes-1); 374 } 375 376 /* Deschedule class and remove it from its parent aggregate. */ 377 static void qfq_deact_rm_from_agg(struct qfq_sched *q, struct qfq_class *cl) 378 { 379 if (cl->qdisc->q.qlen > 0) /* class is active */ 380 qfq_deactivate_class(q, cl); 381 382 qfq_rm_from_agg(q, cl); 383 } 384 385 /* Move class to a new aggregate, matching the new class weight and/or lmax */ 386 static int qfq_change_agg(struct Qdisc *sch, struct qfq_class *cl, u32 weight, 387 u32 lmax) 388 { 389 struct qfq_sched *q = qdisc_priv(sch); 390 struct qfq_aggregate *new_agg = qfq_find_agg(q, lmax, weight); 391 392 if (new_agg == NULL) { /* create new aggregate */ 393 new_agg = kzalloc(sizeof(*new_agg), GFP_ATOMIC); 394 if (new_agg == NULL) 395 return -ENOBUFS; 396 qfq_init_agg(q, new_agg, lmax, weight); 397 } 398 qfq_deact_rm_from_agg(q, cl); 399 qfq_add_to_agg(q, new_agg, cl); 400 401 return 0; 402 } 403 404 static int qfq_change_class(struct Qdisc *sch, u32 classid, u32 parentid, 405 struct nlattr **tca, unsigned long *arg, 406 struct netlink_ext_ack *extack) 407 { 408 struct qfq_sched *q = qdisc_priv(sch); 409 struct qfq_class *cl = (struct qfq_class *)*arg; 410 bool existing = false; 411 struct nlattr *tb[TCA_QFQ_MAX + 1]; 412 struct qfq_aggregate *new_agg = NULL; 413 u32 weight, lmax, inv_w; 414 int err; 415 int delta_w; 416 417 if (tca[TCA_OPTIONS] == NULL) { 418 pr_notice("qfq: no options\n"); 419 return -EINVAL; 420 } 421 422 err = nla_parse_nested(tb, TCA_QFQ_MAX, tca[TCA_OPTIONS], qfq_policy, 423 NULL); 424 if (err < 0) 425 return err; 426 427 if (tb[TCA_QFQ_WEIGHT]) { 428 weight = nla_get_u32(tb[TCA_QFQ_WEIGHT]); 429 if (!weight || weight > (1UL << QFQ_MAX_WSHIFT)) { 430 pr_notice("qfq: invalid weight %u\n", weight); 431 return -EINVAL; 432 } 433 } else 434 weight = 1; 435 436 if (tb[TCA_QFQ_LMAX]) { 437 lmax = nla_get_u32(tb[TCA_QFQ_LMAX]); 438 if (lmax < QFQ_MIN_LMAX || lmax > (1UL << QFQ_MTU_SHIFT)) { 439 pr_notice("qfq: invalid max length %u\n", lmax); 440 return -EINVAL; 441 } 442 } else 443 lmax = psched_mtu(qdisc_dev(sch)); 444 445 inv_w = ONE_FP / weight; 446 weight = ONE_FP / inv_w; 447 448 if (cl != NULL && 449 lmax == cl->agg->lmax && 450 weight == cl->agg->class_weight) 451 return 0; /* nothing to change */ 452 453 delta_w = weight - (cl ? cl->agg->class_weight : 0); 454 455 if (q->wsum + delta_w > QFQ_MAX_WSUM) { 456 pr_notice("qfq: total weight out of range (%d + %u)\n", 457 delta_w, q->wsum); 458 return -EINVAL; 459 } 460 461 if (cl != NULL) { /* modify existing class */ 462 if (tca[TCA_RATE]) { 463 err = gen_replace_estimator(&cl->bstats, NULL, 464 &cl->rate_est, 465 NULL, 466 qdisc_root_sleeping_running(sch), 467 tca[TCA_RATE]); 468 if (err) 469 return err; 470 } 471 existing = true; 472 goto set_change_agg; 473 } 474 475 /* create and init new class */ 476 cl = kzalloc(sizeof(struct qfq_class), GFP_KERNEL); 477 if (cl == NULL) 478 return -ENOBUFS; 479 480 cl->common.classid = classid; 481 cl->deficit = lmax; 482 483 cl->qdisc = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 484 classid, NULL); 485 if (cl->qdisc == NULL) 486 cl->qdisc = &noop_qdisc; 487 488 if (tca[TCA_RATE]) { 489 err = gen_new_estimator(&cl->bstats, NULL, 490 &cl->rate_est, 491 NULL, 492 qdisc_root_sleeping_running(sch), 493 tca[TCA_RATE]); 494 if (err) 495 goto destroy_class; 496 } 497 498 if (cl->qdisc != &noop_qdisc) 499 qdisc_hash_add(cl->qdisc, true); 500 sch_tree_lock(sch); 501 qdisc_class_hash_insert(&q->clhash, &cl->common); 502 sch_tree_unlock(sch); 503 504 qdisc_class_hash_grow(sch, &q->clhash); 505 506 set_change_agg: 507 sch_tree_lock(sch); 508 new_agg = qfq_find_agg(q, lmax, weight); 509 if (new_agg == NULL) { /* create new aggregate */ 510 sch_tree_unlock(sch); 511 new_agg = kzalloc(sizeof(*new_agg), GFP_KERNEL); 512 if (new_agg == NULL) { 513 err = -ENOBUFS; 514 gen_kill_estimator(&cl->rate_est); 515 goto destroy_class; 516 } 517 sch_tree_lock(sch); 518 qfq_init_agg(q, new_agg, lmax, weight); 519 } 520 if (existing) 521 qfq_deact_rm_from_agg(q, cl); 522 qfq_add_to_agg(q, new_agg, cl); 523 sch_tree_unlock(sch); 524 525 *arg = (unsigned long)cl; 526 return 0; 527 528 destroy_class: 529 qdisc_put(cl->qdisc); 530 kfree(cl); 531 return err; 532 } 533 534 static void qfq_destroy_class(struct Qdisc *sch, struct qfq_class *cl) 535 { 536 struct qfq_sched *q = qdisc_priv(sch); 537 538 qfq_rm_from_agg(q, cl); 539 gen_kill_estimator(&cl->rate_est); 540 qdisc_put(cl->qdisc); 541 kfree(cl); 542 } 543 544 static int qfq_delete_class(struct Qdisc *sch, unsigned long arg) 545 { 546 struct qfq_sched *q = qdisc_priv(sch); 547 struct qfq_class *cl = (struct qfq_class *)arg; 548 549 if (cl->filter_cnt > 0) 550 return -EBUSY; 551 552 sch_tree_lock(sch); 553 554 qfq_purge_queue(cl); 555 qdisc_class_hash_remove(&q->clhash, &cl->common); 556 557 sch_tree_unlock(sch); 558 559 qfq_destroy_class(sch, cl); 560 return 0; 561 } 562 563 static unsigned long qfq_search_class(struct Qdisc *sch, u32 classid) 564 { 565 return (unsigned long)qfq_find_class(sch, classid); 566 } 567 568 static struct tcf_block *qfq_tcf_block(struct Qdisc *sch, unsigned long cl, 569 struct netlink_ext_ack *extack) 570 { 571 struct qfq_sched *q = qdisc_priv(sch); 572 573 if (cl) 574 return NULL; 575 576 return q->block; 577 } 578 579 static unsigned long qfq_bind_tcf(struct Qdisc *sch, unsigned long parent, 580 u32 classid) 581 { 582 struct qfq_class *cl = qfq_find_class(sch, classid); 583 584 if (cl != NULL) 585 cl->filter_cnt++; 586 587 return (unsigned long)cl; 588 } 589 590 static void qfq_unbind_tcf(struct Qdisc *sch, unsigned long arg) 591 { 592 struct qfq_class *cl = (struct qfq_class *)arg; 593 594 cl->filter_cnt--; 595 } 596 597 static int qfq_graft_class(struct Qdisc *sch, unsigned long arg, 598 struct Qdisc *new, struct Qdisc **old, 599 struct netlink_ext_ack *extack) 600 { 601 struct qfq_class *cl = (struct qfq_class *)arg; 602 603 if (new == NULL) { 604 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops, 605 cl->common.classid, NULL); 606 if (new == NULL) 607 new = &noop_qdisc; 608 } 609 610 *old = qdisc_replace(sch, new, &cl->qdisc); 611 return 0; 612 } 613 614 static struct Qdisc *qfq_class_leaf(struct Qdisc *sch, unsigned long arg) 615 { 616 struct qfq_class *cl = (struct qfq_class *)arg; 617 618 return cl->qdisc; 619 } 620 621 static int qfq_dump_class(struct Qdisc *sch, unsigned long arg, 622 struct sk_buff *skb, struct tcmsg *tcm) 623 { 624 struct qfq_class *cl = (struct qfq_class *)arg; 625 struct nlattr *nest; 626 627 tcm->tcm_parent = TC_H_ROOT; 628 tcm->tcm_handle = cl->common.classid; 629 tcm->tcm_info = cl->qdisc->handle; 630 631 nest = nla_nest_start(skb, TCA_OPTIONS); 632 if (nest == NULL) 633 goto nla_put_failure; 634 if (nla_put_u32(skb, TCA_QFQ_WEIGHT, cl->agg->class_weight) || 635 nla_put_u32(skb, TCA_QFQ_LMAX, cl->agg->lmax)) 636 goto nla_put_failure; 637 return nla_nest_end(skb, nest); 638 639 nla_put_failure: 640 nla_nest_cancel(skb, nest); 641 return -EMSGSIZE; 642 } 643 644 static int qfq_dump_class_stats(struct Qdisc *sch, unsigned long arg, 645 struct gnet_dump *d) 646 { 647 struct qfq_class *cl = (struct qfq_class *)arg; 648 struct tc_qfq_stats xstats; 649 650 memset(&xstats, 0, sizeof(xstats)); 651 652 xstats.weight = cl->agg->class_weight; 653 xstats.lmax = cl->agg->lmax; 654 655 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), 656 d, NULL, &cl->bstats) < 0 || 657 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 || 658 gnet_stats_copy_queue(d, NULL, 659 &cl->qdisc->qstats, cl->qdisc->q.qlen) < 0) 660 return -1; 661 662 return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); 663 } 664 665 static void qfq_walk(struct Qdisc *sch, struct qdisc_walker *arg) 666 { 667 struct qfq_sched *q = qdisc_priv(sch); 668 struct qfq_class *cl; 669 unsigned int i; 670 671 if (arg->stop) 672 return; 673 674 for (i = 0; i < q->clhash.hashsize; i++) { 675 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 676 if (arg->count < arg->skip) { 677 arg->count++; 678 continue; 679 } 680 if (arg->fn(sch, (unsigned long)cl, arg) < 0) { 681 arg->stop = 1; 682 return; 683 } 684 arg->count++; 685 } 686 } 687 } 688 689 static struct qfq_class *qfq_classify(struct sk_buff *skb, struct Qdisc *sch, 690 int *qerr) 691 { 692 struct qfq_sched *q = qdisc_priv(sch); 693 struct qfq_class *cl; 694 struct tcf_result res; 695 struct tcf_proto *fl; 696 int result; 697 698 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) { 699 pr_debug("qfq_classify: found %d\n", skb->priority); 700 cl = qfq_find_class(sch, skb->priority); 701 if (cl != NULL) 702 return cl; 703 } 704 705 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 706 fl = rcu_dereference_bh(q->filter_list); 707 result = tcf_classify(skb, fl, &res, false); 708 if (result >= 0) { 709 #ifdef CONFIG_NET_CLS_ACT 710 switch (result) { 711 case TC_ACT_QUEUED: 712 case TC_ACT_STOLEN: 713 case TC_ACT_TRAP: 714 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 715 /* fall through */ 716 case TC_ACT_SHOT: 717 return NULL; 718 } 719 #endif 720 cl = (struct qfq_class *)res.class; 721 if (cl == NULL) 722 cl = qfq_find_class(sch, res.classid); 723 return cl; 724 } 725 726 return NULL; 727 } 728 729 /* Generic comparison function, handling wraparound. */ 730 static inline int qfq_gt(u64 a, u64 b) 731 { 732 return (s64)(a - b) > 0; 733 } 734 735 /* Round a precise timestamp to its slotted value. */ 736 static inline u64 qfq_round_down(u64 ts, unsigned int shift) 737 { 738 return ts & ~((1ULL << shift) - 1); 739 } 740 741 /* return the pointer to the group with lowest index in the bitmap */ 742 static inline struct qfq_group *qfq_ffs(struct qfq_sched *q, 743 unsigned long bitmap) 744 { 745 int index = __ffs(bitmap); 746 return &q->groups[index]; 747 } 748 /* Calculate a mask to mimic what would be ffs_from(). */ 749 static inline unsigned long mask_from(unsigned long bitmap, int from) 750 { 751 return bitmap & ~((1UL << from) - 1); 752 } 753 754 /* 755 * The state computation relies on ER=0, IR=1, EB=2, IB=3 756 * First compute eligibility comparing grp->S, q->V, 757 * then check if someone is blocking us and possibly add EB 758 */ 759 static int qfq_calc_state(struct qfq_sched *q, const struct qfq_group *grp) 760 { 761 /* if S > V we are not eligible */ 762 unsigned int state = qfq_gt(grp->S, q->V); 763 unsigned long mask = mask_from(q->bitmaps[ER], grp->index); 764 struct qfq_group *next; 765 766 if (mask) { 767 next = qfq_ffs(q, mask); 768 if (qfq_gt(grp->F, next->F)) 769 state |= EB; 770 } 771 772 return state; 773 } 774 775 776 /* 777 * In principle 778 * q->bitmaps[dst] |= q->bitmaps[src] & mask; 779 * q->bitmaps[src] &= ~mask; 780 * but we should make sure that src != dst 781 */ 782 static inline void qfq_move_groups(struct qfq_sched *q, unsigned long mask, 783 int src, int dst) 784 { 785 q->bitmaps[dst] |= q->bitmaps[src] & mask; 786 q->bitmaps[src] &= ~mask; 787 } 788 789 static void qfq_unblock_groups(struct qfq_sched *q, int index, u64 old_F) 790 { 791 unsigned long mask = mask_from(q->bitmaps[ER], index + 1); 792 struct qfq_group *next; 793 794 if (mask) { 795 next = qfq_ffs(q, mask); 796 if (!qfq_gt(next->F, old_F)) 797 return; 798 } 799 800 mask = (1UL << index) - 1; 801 qfq_move_groups(q, mask, EB, ER); 802 qfq_move_groups(q, mask, IB, IR); 803 } 804 805 /* 806 * perhaps 807 * 808 old_V ^= q->V; 809 old_V >>= q->min_slot_shift; 810 if (old_V) { 811 ... 812 } 813 * 814 */ 815 static void qfq_make_eligible(struct qfq_sched *q) 816 { 817 unsigned long vslot = q->V >> q->min_slot_shift; 818 unsigned long old_vslot = q->oldV >> q->min_slot_shift; 819 820 if (vslot != old_vslot) { 821 unsigned long mask; 822 int last_flip_pos = fls(vslot ^ old_vslot); 823 824 if (last_flip_pos > 31) /* higher than the number of groups */ 825 mask = ~0UL; /* make all groups eligible */ 826 else 827 mask = (1UL << last_flip_pos) - 1; 828 829 qfq_move_groups(q, mask, IR, ER); 830 qfq_move_groups(q, mask, IB, EB); 831 } 832 } 833 834 /* 835 * The index of the slot in which the input aggregate agg is to be 836 * inserted must not be higher than QFQ_MAX_SLOTS-2. There is a '-2' 837 * and not a '-1' because the start time of the group may be moved 838 * backward by one slot after the aggregate has been inserted, and 839 * this would cause non-empty slots to be right-shifted by one 840 * position. 841 * 842 * QFQ+ fully satisfies this bound to the slot index if the parameters 843 * of the classes are not changed dynamically, and if QFQ+ never 844 * happens to postpone the service of agg unjustly, i.e., it never 845 * happens that the aggregate becomes backlogged and eligible, or just 846 * eligible, while an aggregate with a higher approximated finish time 847 * is being served. In particular, in this case QFQ+ guarantees that 848 * the timestamps of agg are low enough that the slot index is never 849 * higher than 2. Unfortunately, QFQ+ cannot provide the same 850 * guarantee if it happens to unjustly postpone the service of agg, or 851 * if the parameters of some class are changed. 852 * 853 * As for the first event, i.e., an out-of-order service, the 854 * upper bound to the slot index guaranteed by QFQ+ grows to 855 * 2 + 856 * QFQ_MAX_AGG_CLASSES * ((1<<QFQ_MTU_SHIFT)/QFQ_MIN_LMAX) * 857 * (current_max_weight/current_wsum) <= 2 + 8 * 128 * 1. 858 * 859 * The following function deals with this problem by backward-shifting 860 * the timestamps of agg, if needed, so as to guarantee that the slot 861 * index is never higher than QFQ_MAX_SLOTS-2. This backward-shift may 862 * cause the service of other aggregates to be postponed, yet the 863 * worst-case guarantees of these aggregates are not violated. In 864 * fact, in case of no out-of-order service, the timestamps of agg 865 * would have been even lower than they are after the backward shift, 866 * because QFQ+ would have guaranteed a maximum value equal to 2 for 867 * the slot index, and 2 < QFQ_MAX_SLOTS-2. Hence the aggregates whose 868 * service is postponed because of the backward-shift would have 869 * however waited for the service of agg before being served. 870 * 871 * The other event that may cause the slot index to be higher than 2 872 * for agg is a recent change of the parameters of some class. If the 873 * weight of a class is increased or the lmax (max_pkt_size) of the 874 * class is decreased, then a new aggregate with smaller slot size 875 * than the original parent aggregate of the class may happen to be 876 * activated. The activation of this aggregate should be properly 877 * delayed to when the service of the class has finished in the ideal 878 * system tracked by QFQ+. If the activation of the aggregate is not 879 * delayed to this reference time instant, then this aggregate may be 880 * unjustly served before other aggregates waiting for service. This 881 * may cause the above bound to the slot index to be violated for some 882 * of these unlucky aggregates. 883 * 884 * Instead of delaying the activation of the new aggregate, which is 885 * quite complex, the above-discussed capping of the slot index is 886 * used to handle also the consequences of a change of the parameters 887 * of a class. 888 */ 889 static void qfq_slot_insert(struct qfq_group *grp, struct qfq_aggregate *agg, 890 u64 roundedS) 891 { 892 u64 slot = (roundedS - grp->S) >> grp->slot_shift; 893 unsigned int i; /* slot index in the bucket list */ 894 895 if (unlikely(slot > QFQ_MAX_SLOTS - 2)) { 896 u64 deltaS = roundedS - grp->S - 897 ((u64)(QFQ_MAX_SLOTS - 2)<<grp->slot_shift); 898 agg->S -= deltaS; 899 agg->F -= deltaS; 900 slot = QFQ_MAX_SLOTS - 2; 901 } 902 903 i = (grp->front + slot) % QFQ_MAX_SLOTS; 904 905 hlist_add_head(&agg->next, &grp->slots[i]); 906 __set_bit(slot, &grp->full_slots); 907 } 908 909 /* Maybe introduce hlist_first_entry?? */ 910 static struct qfq_aggregate *qfq_slot_head(struct qfq_group *grp) 911 { 912 return hlist_entry(grp->slots[grp->front].first, 913 struct qfq_aggregate, next); 914 } 915 916 /* 917 * remove the entry from the slot 918 */ 919 static void qfq_front_slot_remove(struct qfq_group *grp) 920 { 921 struct qfq_aggregate *agg = qfq_slot_head(grp); 922 923 BUG_ON(!agg); 924 hlist_del(&agg->next); 925 if (hlist_empty(&grp->slots[grp->front])) 926 __clear_bit(0, &grp->full_slots); 927 } 928 929 /* 930 * Returns the first aggregate in the first non-empty bucket of the 931 * group. As a side effect, adjusts the bucket list so the first 932 * non-empty bucket is at position 0 in full_slots. 933 */ 934 static struct qfq_aggregate *qfq_slot_scan(struct qfq_group *grp) 935 { 936 unsigned int i; 937 938 pr_debug("qfq slot_scan: grp %u full %#lx\n", 939 grp->index, grp->full_slots); 940 941 if (grp->full_slots == 0) 942 return NULL; 943 944 i = __ffs(grp->full_slots); /* zero based */ 945 if (i > 0) { 946 grp->front = (grp->front + i) % QFQ_MAX_SLOTS; 947 grp->full_slots >>= i; 948 } 949 950 return qfq_slot_head(grp); 951 } 952 953 /* 954 * adjust the bucket list. When the start time of a group decreases, 955 * we move the index down (modulo QFQ_MAX_SLOTS) so we don't need to 956 * move the objects. The mask of occupied slots must be shifted 957 * because we use ffs() to find the first non-empty slot. 958 * This covers decreases in the group's start time, but what about 959 * increases of the start time ? 960 * Here too we should make sure that i is less than 32 961 */ 962 static void qfq_slot_rotate(struct qfq_group *grp, u64 roundedS) 963 { 964 unsigned int i = (grp->S - roundedS) >> grp->slot_shift; 965 966 grp->full_slots <<= i; 967 grp->front = (grp->front - i) % QFQ_MAX_SLOTS; 968 } 969 970 static void qfq_update_eligible(struct qfq_sched *q) 971 { 972 struct qfq_group *grp; 973 unsigned long ineligible; 974 975 ineligible = q->bitmaps[IR] | q->bitmaps[IB]; 976 if (ineligible) { 977 if (!q->bitmaps[ER]) { 978 grp = qfq_ffs(q, ineligible); 979 if (qfq_gt(grp->S, q->V)) 980 q->V = grp->S; 981 } 982 qfq_make_eligible(q); 983 } 984 } 985 986 /* Dequeue head packet of the head class in the DRR queue of the aggregate. */ 987 static void agg_dequeue(struct qfq_aggregate *agg, 988 struct qfq_class *cl, unsigned int len) 989 { 990 qdisc_dequeue_peeked(cl->qdisc); 991 992 cl->deficit -= (int) len; 993 994 if (cl->qdisc->q.qlen == 0) /* no more packets, remove from list */ 995 list_del(&cl->alist); 996 else if (cl->deficit < qdisc_pkt_len(cl->qdisc->ops->peek(cl->qdisc))) { 997 cl->deficit += agg->lmax; 998 list_move_tail(&cl->alist, &agg->active); 999 } 1000 } 1001 1002 static inline struct sk_buff *qfq_peek_skb(struct qfq_aggregate *agg, 1003 struct qfq_class **cl, 1004 unsigned int *len) 1005 { 1006 struct sk_buff *skb; 1007 1008 *cl = list_first_entry(&agg->active, struct qfq_class, alist); 1009 skb = (*cl)->qdisc->ops->peek((*cl)->qdisc); 1010 if (skb == NULL) 1011 WARN_ONCE(1, "qfq_dequeue: non-workconserving leaf\n"); 1012 else 1013 *len = qdisc_pkt_len(skb); 1014 1015 return skb; 1016 } 1017 1018 /* Update F according to the actual service received by the aggregate. */ 1019 static inline void charge_actual_service(struct qfq_aggregate *agg) 1020 { 1021 /* Compute the service received by the aggregate, taking into 1022 * account that, after decreasing the number of classes in 1023 * agg, it may happen that 1024 * agg->initial_budget - agg->budget > agg->bugdetmax 1025 */ 1026 u32 service_received = min(agg->budgetmax, 1027 agg->initial_budget - agg->budget); 1028 1029 agg->F = agg->S + (u64)service_received * agg->inv_w; 1030 } 1031 1032 /* Assign a reasonable start time for a new aggregate in group i. 1033 * Admissible values for \hat(F) are multiples of \sigma_i 1034 * no greater than V+\sigma_i . Larger values mean that 1035 * we had a wraparound so we consider the timestamp to be stale. 1036 * 1037 * If F is not stale and F >= V then we set S = F. 1038 * Otherwise we should assign S = V, but this may violate 1039 * the ordering in EB (see [2]). So, if we have groups in ER, 1040 * set S to the F_j of the first group j which would be blocking us. 1041 * We are guaranteed not to move S backward because 1042 * otherwise our group i would still be blocked. 1043 */ 1044 static void qfq_update_start(struct qfq_sched *q, struct qfq_aggregate *agg) 1045 { 1046 unsigned long mask; 1047 u64 limit, roundedF; 1048 int slot_shift = agg->grp->slot_shift; 1049 1050 roundedF = qfq_round_down(agg->F, slot_shift); 1051 limit = qfq_round_down(q->V, slot_shift) + (1ULL << slot_shift); 1052 1053 if (!qfq_gt(agg->F, q->V) || qfq_gt(roundedF, limit)) { 1054 /* timestamp was stale */ 1055 mask = mask_from(q->bitmaps[ER], agg->grp->index); 1056 if (mask) { 1057 struct qfq_group *next = qfq_ffs(q, mask); 1058 if (qfq_gt(roundedF, next->F)) { 1059 if (qfq_gt(limit, next->F)) 1060 agg->S = next->F; 1061 else /* preserve timestamp correctness */ 1062 agg->S = limit; 1063 return; 1064 } 1065 } 1066 agg->S = q->V; 1067 } else /* timestamp is not stale */ 1068 agg->S = agg->F; 1069 } 1070 1071 /* Update the timestamps of agg before scheduling/rescheduling it for 1072 * service. In particular, assign to agg->F its maximum possible 1073 * value, i.e., the virtual finish time with which the aggregate 1074 * should be labeled if it used all its budget once in service. 1075 */ 1076 static inline void 1077 qfq_update_agg_ts(struct qfq_sched *q, 1078 struct qfq_aggregate *agg, enum update_reason reason) 1079 { 1080 if (reason != requeue) 1081 qfq_update_start(q, agg); 1082 else /* just charge agg for the service received */ 1083 agg->S = agg->F; 1084 1085 agg->F = agg->S + (u64)agg->budgetmax * agg->inv_w; 1086 } 1087 1088 static void qfq_schedule_agg(struct qfq_sched *q, struct qfq_aggregate *agg); 1089 1090 static struct sk_buff *qfq_dequeue(struct Qdisc *sch) 1091 { 1092 struct qfq_sched *q = qdisc_priv(sch); 1093 struct qfq_aggregate *in_serv_agg = q->in_serv_agg; 1094 struct qfq_class *cl; 1095 struct sk_buff *skb = NULL; 1096 /* next-packet len, 0 means no more active classes in in-service agg */ 1097 unsigned int len = 0; 1098 1099 if (in_serv_agg == NULL) 1100 return NULL; 1101 1102 if (!list_empty(&in_serv_agg->active)) 1103 skb = qfq_peek_skb(in_serv_agg, &cl, &len); 1104 1105 /* 1106 * If there are no active classes in the in-service aggregate, 1107 * or if the aggregate has not enough budget to serve its next 1108 * class, then choose the next aggregate to serve. 1109 */ 1110 if (len == 0 || in_serv_agg->budget < len) { 1111 charge_actual_service(in_serv_agg); 1112 1113 /* recharge the budget of the aggregate */ 1114 in_serv_agg->initial_budget = in_serv_agg->budget = 1115 in_serv_agg->budgetmax; 1116 1117 if (!list_empty(&in_serv_agg->active)) { 1118 /* 1119 * Still active: reschedule for 1120 * service. Possible optimization: if no other 1121 * aggregate is active, then there is no point 1122 * in rescheduling this aggregate, and we can 1123 * just keep it as the in-service one. This 1124 * should be however a corner case, and to 1125 * handle it, we would need to maintain an 1126 * extra num_active_aggs field. 1127 */ 1128 qfq_update_agg_ts(q, in_serv_agg, requeue); 1129 qfq_schedule_agg(q, in_serv_agg); 1130 } else if (sch->q.qlen == 0) { /* no aggregate to serve */ 1131 q->in_serv_agg = NULL; 1132 return NULL; 1133 } 1134 1135 /* 1136 * If we get here, there are other aggregates queued: 1137 * choose the new aggregate to serve. 1138 */ 1139 in_serv_agg = q->in_serv_agg = qfq_choose_next_agg(q); 1140 skb = qfq_peek_skb(in_serv_agg, &cl, &len); 1141 } 1142 if (!skb) 1143 return NULL; 1144 1145 qdisc_qstats_backlog_dec(sch, skb); 1146 sch->q.qlen--; 1147 qdisc_bstats_update(sch, skb); 1148 1149 agg_dequeue(in_serv_agg, cl, len); 1150 /* If lmax is lowered, through qfq_change_class, for a class 1151 * owning pending packets with larger size than the new value 1152 * of lmax, then the following condition may hold. 1153 */ 1154 if (unlikely(in_serv_agg->budget < len)) 1155 in_serv_agg->budget = 0; 1156 else 1157 in_serv_agg->budget -= len; 1158 1159 q->V += (u64)len * q->iwsum; 1160 pr_debug("qfq dequeue: len %u F %lld now %lld\n", 1161 len, (unsigned long long) in_serv_agg->F, 1162 (unsigned long long) q->V); 1163 1164 return skb; 1165 } 1166 1167 static struct qfq_aggregate *qfq_choose_next_agg(struct qfq_sched *q) 1168 { 1169 struct qfq_group *grp; 1170 struct qfq_aggregate *agg, *new_front_agg; 1171 u64 old_F; 1172 1173 qfq_update_eligible(q); 1174 q->oldV = q->V; 1175 1176 if (!q->bitmaps[ER]) 1177 return NULL; 1178 1179 grp = qfq_ffs(q, q->bitmaps[ER]); 1180 old_F = grp->F; 1181 1182 agg = qfq_slot_head(grp); 1183 1184 /* agg starts to be served, remove it from schedule */ 1185 qfq_front_slot_remove(grp); 1186 1187 new_front_agg = qfq_slot_scan(grp); 1188 1189 if (new_front_agg == NULL) /* group is now inactive, remove from ER */ 1190 __clear_bit(grp->index, &q->bitmaps[ER]); 1191 else { 1192 u64 roundedS = qfq_round_down(new_front_agg->S, 1193 grp->slot_shift); 1194 unsigned int s; 1195 1196 if (grp->S == roundedS) 1197 return agg; 1198 grp->S = roundedS; 1199 grp->F = roundedS + (2ULL << grp->slot_shift); 1200 __clear_bit(grp->index, &q->bitmaps[ER]); 1201 s = qfq_calc_state(q, grp); 1202 __set_bit(grp->index, &q->bitmaps[s]); 1203 } 1204 1205 qfq_unblock_groups(q, grp->index, old_F); 1206 1207 return agg; 1208 } 1209 1210 static int qfq_enqueue(struct sk_buff *skb, struct Qdisc *sch, 1211 struct sk_buff **to_free) 1212 { 1213 struct qfq_sched *q = qdisc_priv(sch); 1214 struct qfq_class *cl; 1215 struct qfq_aggregate *agg; 1216 int err = 0; 1217 1218 cl = qfq_classify(skb, sch, &err); 1219 if (cl == NULL) { 1220 if (err & __NET_XMIT_BYPASS) 1221 qdisc_qstats_drop(sch); 1222 __qdisc_drop(skb, to_free); 1223 return err; 1224 } 1225 pr_debug("qfq_enqueue: cl = %x\n", cl->common.classid); 1226 1227 if (unlikely(cl->agg->lmax < qdisc_pkt_len(skb))) { 1228 pr_debug("qfq: increasing maxpkt from %u to %u for class %u", 1229 cl->agg->lmax, qdisc_pkt_len(skb), cl->common.classid); 1230 err = qfq_change_agg(sch, cl, cl->agg->class_weight, 1231 qdisc_pkt_len(skb)); 1232 if (err) { 1233 cl->qstats.drops++; 1234 return qdisc_drop(skb, sch, to_free); 1235 } 1236 } 1237 1238 err = qdisc_enqueue(skb, cl->qdisc, to_free); 1239 if (unlikely(err != NET_XMIT_SUCCESS)) { 1240 pr_debug("qfq_enqueue: enqueue failed %d\n", err); 1241 if (net_xmit_drop_count(err)) { 1242 cl->qstats.drops++; 1243 qdisc_qstats_drop(sch); 1244 } 1245 return err; 1246 } 1247 1248 bstats_update(&cl->bstats, skb); 1249 qdisc_qstats_backlog_inc(sch, skb); 1250 ++sch->q.qlen; 1251 1252 agg = cl->agg; 1253 /* if the queue was not empty, then done here */ 1254 if (cl->qdisc->q.qlen != 1) { 1255 if (unlikely(skb == cl->qdisc->ops->peek(cl->qdisc)) && 1256 list_first_entry(&agg->active, struct qfq_class, alist) 1257 == cl && cl->deficit < qdisc_pkt_len(skb)) 1258 list_move_tail(&cl->alist, &agg->active); 1259 1260 return err; 1261 } 1262 1263 /* schedule class for service within the aggregate */ 1264 cl->deficit = agg->lmax; 1265 list_add_tail(&cl->alist, &agg->active); 1266 1267 if (list_first_entry(&agg->active, struct qfq_class, alist) != cl || 1268 q->in_serv_agg == agg) 1269 return err; /* non-empty or in service, nothing else to do */ 1270 1271 qfq_activate_agg(q, agg, enqueue); 1272 1273 return err; 1274 } 1275 1276 /* 1277 * Schedule aggregate according to its timestamps. 1278 */ 1279 static void qfq_schedule_agg(struct qfq_sched *q, struct qfq_aggregate *agg) 1280 { 1281 struct qfq_group *grp = agg->grp; 1282 u64 roundedS; 1283 int s; 1284 1285 roundedS = qfq_round_down(agg->S, grp->slot_shift); 1286 1287 /* 1288 * Insert agg in the correct bucket. 1289 * If agg->S >= grp->S we don't need to adjust the 1290 * bucket list and simply go to the insertion phase. 1291 * Otherwise grp->S is decreasing, we must make room 1292 * in the bucket list, and also recompute the group state. 1293 * Finally, if there were no flows in this group and nobody 1294 * was in ER make sure to adjust V. 1295 */ 1296 if (grp->full_slots) { 1297 if (!qfq_gt(grp->S, agg->S)) 1298 goto skip_update; 1299 1300 /* create a slot for this agg->S */ 1301 qfq_slot_rotate(grp, roundedS); 1302 /* group was surely ineligible, remove */ 1303 __clear_bit(grp->index, &q->bitmaps[IR]); 1304 __clear_bit(grp->index, &q->bitmaps[IB]); 1305 } else if (!q->bitmaps[ER] && qfq_gt(roundedS, q->V) && 1306 q->in_serv_agg == NULL) 1307 q->V = roundedS; 1308 1309 grp->S = roundedS; 1310 grp->F = roundedS + (2ULL << grp->slot_shift); 1311 s = qfq_calc_state(q, grp); 1312 __set_bit(grp->index, &q->bitmaps[s]); 1313 1314 pr_debug("qfq enqueue: new state %d %#lx S %lld F %lld V %lld\n", 1315 s, q->bitmaps[s], 1316 (unsigned long long) agg->S, 1317 (unsigned long long) agg->F, 1318 (unsigned long long) q->V); 1319 1320 skip_update: 1321 qfq_slot_insert(grp, agg, roundedS); 1322 } 1323 1324 1325 /* Update agg ts and schedule agg for service */ 1326 static void qfq_activate_agg(struct qfq_sched *q, struct qfq_aggregate *agg, 1327 enum update_reason reason) 1328 { 1329 agg->initial_budget = agg->budget = agg->budgetmax; /* recharge budg. */ 1330 1331 qfq_update_agg_ts(q, agg, reason); 1332 if (q->in_serv_agg == NULL) { /* no aggr. in service or scheduled */ 1333 q->in_serv_agg = agg; /* start serving this aggregate */ 1334 /* update V: to be in service, agg must be eligible */ 1335 q->oldV = q->V = agg->S; 1336 } else if (agg != q->in_serv_agg) 1337 qfq_schedule_agg(q, agg); 1338 } 1339 1340 static void qfq_slot_remove(struct qfq_sched *q, struct qfq_group *grp, 1341 struct qfq_aggregate *agg) 1342 { 1343 unsigned int i, offset; 1344 u64 roundedS; 1345 1346 roundedS = qfq_round_down(agg->S, grp->slot_shift); 1347 offset = (roundedS - grp->S) >> grp->slot_shift; 1348 1349 i = (grp->front + offset) % QFQ_MAX_SLOTS; 1350 1351 hlist_del(&agg->next); 1352 if (hlist_empty(&grp->slots[i])) 1353 __clear_bit(offset, &grp->full_slots); 1354 } 1355 1356 /* 1357 * Called to forcibly deschedule an aggregate. If the aggregate is 1358 * not in the front bucket, or if the latter has other aggregates in 1359 * the front bucket, we can simply remove the aggregate with no other 1360 * side effects. 1361 * Otherwise we must propagate the event up. 1362 */ 1363 static void qfq_deactivate_agg(struct qfq_sched *q, struct qfq_aggregate *agg) 1364 { 1365 struct qfq_group *grp = agg->grp; 1366 unsigned long mask; 1367 u64 roundedS; 1368 int s; 1369 1370 if (agg == q->in_serv_agg) { 1371 charge_actual_service(agg); 1372 q->in_serv_agg = qfq_choose_next_agg(q); 1373 return; 1374 } 1375 1376 agg->F = agg->S; 1377 qfq_slot_remove(q, grp, agg); 1378 1379 if (!grp->full_slots) { 1380 __clear_bit(grp->index, &q->bitmaps[IR]); 1381 __clear_bit(grp->index, &q->bitmaps[EB]); 1382 __clear_bit(grp->index, &q->bitmaps[IB]); 1383 1384 if (test_bit(grp->index, &q->bitmaps[ER]) && 1385 !(q->bitmaps[ER] & ~((1UL << grp->index) - 1))) { 1386 mask = q->bitmaps[ER] & ((1UL << grp->index) - 1); 1387 if (mask) 1388 mask = ~((1UL << __fls(mask)) - 1); 1389 else 1390 mask = ~0UL; 1391 qfq_move_groups(q, mask, EB, ER); 1392 qfq_move_groups(q, mask, IB, IR); 1393 } 1394 __clear_bit(grp->index, &q->bitmaps[ER]); 1395 } else if (hlist_empty(&grp->slots[grp->front])) { 1396 agg = qfq_slot_scan(grp); 1397 roundedS = qfq_round_down(agg->S, grp->slot_shift); 1398 if (grp->S != roundedS) { 1399 __clear_bit(grp->index, &q->bitmaps[ER]); 1400 __clear_bit(grp->index, &q->bitmaps[IR]); 1401 __clear_bit(grp->index, &q->bitmaps[EB]); 1402 __clear_bit(grp->index, &q->bitmaps[IB]); 1403 grp->S = roundedS; 1404 grp->F = roundedS + (2ULL << grp->slot_shift); 1405 s = qfq_calc_state(q, grp); 1406 __set_bit(grp->index, &q->bitmaps[s]); 1407 } 1408 } 1409 } 1410 1411 static void qfq_qlen_notify(struct Qdisc *sch, unsigned long arg) 1412 { 1413 struct qfq_sched *q = qdisc_priv(sch); 1414 struct qfq_class *cl = (struct qfq_class *)arg; 1415 1416 qfq_deactivate_class(q, cl); 1417 } 1418 1419 static int qfq_init_qdisc(struct Qdisc *sch, struct nlattr *opt, 1420 struct netlink_ext_ack *extack) 1421 { 1422 struct qfq_sched *q = qdisc_priv(sch); 1423 struct qfq_group *grp; 1424 int i, j, err; 1425 u32 max_cl_shift, maxbudg_shift, max_classes; 1426 1427 err = tcf_block_get(&q->block, &q->filter_list, sch, extack); 1428 if (err) 1429 return err; 1430 1431 err = qdisc_class_hash_init(&q->clhash); 1432 if (err < 0) 1433 return err; 1434 1435 if (qdisc_dev(sch)->tx_queue_len + 1 > QFQ_MAX_AGG_CLASSES) 1436 max_classes = QFQ_MAX_AGG_CLASSES; 1437 else 1438 max_classes = qdisc_dev(sch)->tx_queue_len + 1; 1439 /* max_cl_shift = floor(log_2(max_classes)) */ 1440 max_cl_shift = __fls(max_classes); 1441 q->max_agg_classes = 1<<max_cl_shift; 1442 1443 /* maxbudg_shift = log2(max_len * max_classes_per_agg) */ 1444 maxbudg_shift = QFQ_MTU_SHIFT + max_cl_shift; 1445 q->min_slot_shift = FRAC_BITS + maxbudg_shift - QFQ_MAX_INDEX; 1446 1447 for (i = 0; i <= QFQ_MAX_INDEX; i++) { 1448 grp = &q->groups[i]; 1449 grp->index = i; 1450 grp->slot_shift = q->min_slot_shift + i; 1451 for (j = 0; j < QFQ_MAX_SLOTS; j++) 1452 INIT_HLIST_HEAD(&grp->slots[j]); 1453 } 1454 1455 INIT_HLIST_HEAD(&q->nonfull_aggs); 1456 1457 return 0; 1458 } 1459 1460 static void qfq_reset_qdisc(struct Qdisc *sch) 1461 { 1462 struct qfq_sched *q = qdisc_priv(sch); 1463 struct qfq_class *cl; 1464 unsigned int i; 1465 1466 for (i = 0; i < q->clhash.hashsize; i++) { 1467 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 1468 if (cl->qdisc->q.qlen > 0) 1469 qfq_deactivate_class(q, cl); 1470 1471 qdisc_reset(cl->qdisc); 1472 } 1473 } 1474 sch->qstats.backlog = 0; 1475 sch->q.qlen = 0; 1476 } 1477 1478 static void qfq_destroy_qdisc(struct Qdisc *sch) 1479 { 1480 struct qfq_sched *q = qdisc_priv(sch); 1481 struct qfq_class *cl; 1482 struct hlist_node *next; 1483 unsigned int i; 1484 1485 tcf_block_put(q->block); 1486 1487 for (i = 0; i < q->clhash.hashsize; i++) { 1488 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], 1489 common.hnode) { 1490 qfq_destroy_class(sch, cl); 1491 } 1492 } 1493 qdisc_class_hash_destroy(&q->clhash); 1494 } 1495 1496 static const struct Qdisc_class_ops qfq_class_ops = { 1497 .change = qfq_change_class, 1498 .delete = qfq_delete_class, 1499 .find = qfq_search_class, 1500 .tcf_block = qfq_tcf_block, 1501 .bind_tcf = qfq_bind_tcf, 1502 .unbind_tcf = qfq_unbind_tcf, 1503 .graft = qfq_graft_class, 1504 .leaf = qfq_class_leaf, 1505 .qlen_notify = qfq_qlen_notify, 1506 .dump = qfq_dump_class, 1507 .dump_stats = qfq_dump_class_stats, 1508 .walk = qfq_walk, 1509 }; 1510 1511 static struct Qdisc_ops qfq_qdisc_ops __read_mostly = { 1512 .cl_ops = &qfq_class_ops, 1513 .id = "qfq", 1514 .priv_size = sizeof(struct qfq_sched), 1515 .enqueue = qfq_enqueue, 1516 .dequeue = qfq_dequeue, 1517 .peek = qdisc_peek_dequeued, 1518 .init = qfq_init_qdisc, 1519 .reset = qfq_reset_qdisc, 1520 .destroy = qfq_destroy_qdisc, 1521 .owner = THIS_MODULE, 1522 }; 1523 1524 static int __init qfq_init(void) 1525 { 1526 return register_qdisc(&qfq_qdisc_ops); 1527 } 1528 1529 static void __exit qfq_exit(void) 1530 { 1531 unregister_qdisc(&qfq_qdisc_ops); 1532 } 1533 1534 module_init(qfq_init); 1535 module_exit(qfq_exit); 1536 MODULE_LICENSE("GPL"); 1537