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#define IWSUM (ONE_FP/QFQ_MAX_WSUM) 117 118#define QFQ_MTU_SHIFT 16 /* to support TSO/GSO */ 119#define QFQ_MIN_LMAX 512 /* see qfq_slot_insert */ 120 121#define QFQ_MAX_AGG_CLASSES 8 /* max num classes per aggregate allowed */ 122 123/* 124 * Possible group states. These values are used as indexes for the bitmaps 125 * array of struct qfq_queue. 126 */ 127enum qfq_state { ER, IR, EB, IB, QFQ_MAX_STATE }; 128 129struct qfq_group; 130 131struct qfq_aggregate; 132 133struct qfq_class { 134 struct Qdisc_class_common common; 135 136 unsigned int refcnt; 137 unsigned int filter_cnt; 138 139 struct gnet_stats_basic_packed bstats; 140 struct gnet_stats_queue qstats; 141 struct gnet_stats_rate_est rate_est; 142 struct Qdisc *qdisc; 143 struct list_head alist; /* Link for active-classes list. */ 144 struct qfq_aggregate *agg; /* Parent aggregate. */ 145 int deficit; /* DRR deficit counter. */ 146}; 147 148struct qfq_aggregate { 149 struct hlist_node next; /* Link for the slot list. */ 150 u64 S, F; /* flow timestamps (exact) */ 151 152 /* group we belong to. In principle we would need the index, 153 * which is log_2(lmax/weight), but we never reference it 154 * directly, only the group. 155 */ 156 struct qfq_group *grp; 157 158 /* these are copied from the flowset. */ 159 u32 class_weight; /* Weight of each class in this aggregate. */ 160 /* Max pkt size for the classes in this aggregate, DRR quantum. */ 161 int lmax; 162 163 u32 inv_w; /* ONE_FP/(sum of weights of classes in aggr.). */ 164 u32 budgetmax; /* Max budget for this aggregate. */ 165 u32 initial_budget, budget; /* Initial and current budget. */ 166 167 int num_classes; /* Number of classes in this aggr. */ 168 struct list_head active; /* DRR queue of active classes. */ 169 170 struct hlist_node nonfull_next; /* See nonfull_aggs in qfq_sched. */ 171}; 172 173struct qfq_group { 174 u64 S, F; /* group timestamps (approx). */ 175 unsigned int slot_shift; /* Slot shift. */ 176 unsigned int index; /* Group index. */ 177 unsigned int front; /* Index of the front slot. */ 178 unsigned long full_slots; /* non-empty slots */ 179 180 /* Array of RR lists of active aggregates. */ 181 struct hlist_head slots[QFQ_MAX_SLOTS]; 182}; 183 184struct qfq_sched { 185 struct tcf_proto *filter_list; 186 struct Qdisc_class_hash clhash; 187 188 u64 oldV, V; /* Precise virtual times. */ 189 struct qfq_aggregate *in_serv_agg; /* Aggregate being served. */ 190 u32 num_active_agg; /* Num. of active aggregates */ 191 u32 wsum; /* weight sum */ 192 193 unsigned long bitmaps[QFQ_MAX_STATE]; /* Group bitmaps. */ 194 struct qfq_group groups[QFQ_MAX_INDEX + 1]; /* The groups. */ 195 u32 min_slot_shift; /* Index of the group-0 bit in the bitmaps. */ 196 197 u32 max_agg_classes; /* Max number of classes per aggr. */ 198 struct hlist_head nonfull_aggs; /* Aggs with room for more classes. */ 199}; 200 201/* 202 * Possible reasons why the timestamps of an aggregate are updated 203 * enqueue: the aggregate switches from idle to active and must scheduled 204 * for service 205 * requeue: the aggregate finishes its budget, so it stops being served and 206 * must be rescheduled for service 207 */ 208enum update_reason {enqueue, requeue}; 209 210static struct qfq_class *qfq_find_class(struct Qdisc *sch, u32 classid) 211{ 212 struct qfq_sched *q = qdisc_priv(sch); 213 struct Qdisc_class_common *clc; 214 215 clc = qdisc_class_find(&q->clhash, classid); 216 if (clc == NULL) 217 return NULL; 218 return container_of(clc, struct qfq_class, common); 219} 220 221static void qfq_purge_queue(struct qfq_class *cl) 222{ 223 unsigned int len = cl->qdisc->q.qlen; 224 225 qdisc_reset(cl->qdisc); 226 qdisc_tree_decrease_qlen(cl->qdisc, len); 227} 228 229static 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 */ 239static 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; 254out: 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 261static void qfq_deactivate_agg(struct qfq_sched *, struct qfq_aggregate *); 262static void qfq_activate_agg(struct qfq_sched *, struct qfq_aggregate *, 263 enum update_reason); 264 265static 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 275static 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. */ 289static 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); 317 318 agg->num_classes = new_num_classes; 319} 320 321/* Add class to aggregate. */ 322static void qfq_add_to_agg(struct qfq_sched *q, 323 struct qfq_aggregate *agg, 324 struct qfq_class *cl) 325{ 326 cl->agg = agg; 327 328 qfq_update_agg(q, agg, agg->num_classes+1); 329 if (cl->qdisc->q.qlen > 0) { /* adding an active class */ 330 list_add_tail(&cl->alist, &agg->active); 331 if (list_first_entry(&agg->active, struct qfq_class, alist) == 332 cl && q->in_serv_agg != agg) /* agg was inactive */ 333 qfq_activate_agg(q, agg, enqueue); /* schedule agg */ 334 } 335} 336 337static struct qfq_aggregate *qfq_choose_next_agg(struct qfq_sched *); 338 339static void qfq_destroy_agg(struct qfq_sched *q, struct qfq_aggregate *agg) 340{ 341 if (!hlist_unhashed(&agg->nonfull_next)) 342 hlist_del_init(&agg->nonfull_next); 343 if (q->in_serv_agg == agg) 344 q->in_serv_agg = qfq_choose_next_agg(q); 345 kfree(agg); 346} 347 348/* Deschedule class from within its parent aggregate. */ 349static void qfq_deactivate_class(struct qfq_sched *q, struct qfq_class *cl) 350{ 351 struct qfq_aggregate *agg = cl->agg; 352 353 354 list_del(&cl->alist); /* remove from RR queue of the aggregate */ 355 if (list_empty(&agg->active)) /* agg is now inactive */ 356 qfq_deactivate_agg(q, agg); 357} 358 359/* Remove class from its parent aggregate. */ 360static void qfq_rm_from_agg(struct qfq_sched *q, struct qfq_class *cl) 361{ 362 struct qfq_aggregate *agg = cl->agg; 363 364 cl->agg = NULL; 365 if (agg->num_classes == 1) { /* agg being emptied, destroy it */ 366 qfq_destroy_agg(q, agg); 367 return; 368 } 369 qfq_update_agg(q, agg, agg->num_classes-1); 370} 371 372/* Deschedule class and remove it from its parent aggregate. */ 373static void qfq_deact_rm_from_agg(struct qfq_sched *q, struct qfq_class *cl) 374{ 375 if (cl->qdisc->q.qlen > 0) /* class is active */ 376 qfq_deactivate_class(q, cl); 377 378 qfq_rm_from_agg(q, cl); 379} 380 381/* Move class to a new aggregate, matching the new class weight and/or lmax */ 382static int qfq_change_agg(struct Qdisc *sch, struct qfq_class *cl, u32 weight, 383 u32 lmax) 384{ 385 struct qfq_sched *q = qdisc_priv(sch); 386 struct qfq_aggregate *new_agg = qfq_find_agg(q, lmax, weight); 387 388 if (new_agg == NULL) { /* create new aggregate */ 389 new_agg = kzalloc(sizeof(*new_agg), GFP_ATOMIC); 390 if (new_agg == NULL) 391 return -ENOBUFS; 392 qfq_init_agg(q, new_agg, lmax, weight); 393 } 394 qfq_deact_rm_from_agg(q, cl); 395 qfq_add_to_agg(q, new_agg, cl); 396 397 return 0; 398} 399 400static int qfq_change_class(struct Qdisc *sch, u32 classid, u32 parentid, 401 struct nlattr **tca, unsigned long *arg) 402{ 403 struct qfq_sched *q = qdisc_priv(sch); 404 struct qfq_class *cl = (struct qfq_class *)*arg; 405 bool existing = false; 406 struct nlattr *tb[TCA_QFQ_MAX + 1]; 407 struct qfq_aggregate *new_agg = NULL; 408 u32 weight, lmax, inv_w; 409 int err; 410 int delta_w; 411 412 if (tca[TCA_OPTIONS] == NULL) { 413 pr_notice("qfq: no options\n"); 414 return -EINVAL; 415 } 416 417 err = nla_parse_nested(tb, TCA_QFQ_MAX, tca[TCA_OPTIONS], qfq_policy); 418 if (err < 0) 419 return err; 420 421 if (tb[TCA_QFQ_WEIGHT]) { 422 weight = nla_get_u32(tb[TCA_QFQ_WEIGHT]); 423 if (!weight || weight > (1UL << QFQ_MAX_WSHIFT)) { 424 pr_notice("qfq: invalid weight %u\n", weight); 425 return -EINVAL; 426 } 427 } else 428 weight = 1; 429 430 if (tb[TCA_QFQ_LMAX]) { 431 lmax = nla_get_u32(tb[TCA_QFQ_LMAX]); 432 if (lmax < QFQ_MIN_LMAX || lmax > (1UL << QFQ_MTU_SHIFT)) { 433 pr_notice("qfq: invalid max length %u\n", lmax); 434 return -EINVAL; 435 } 436 } else 437 lmax = psched_mtu(qdisc_dev(sch)); 438 439 inv_w = ONE_FP / weight; 440 weight = ONE_FP / inv_w; 441 442 if (cl != NULL && 443 lmax == cl->agg->lmax && 444 weight == cl->agg->class_weight) 445 return 0; /* nothing to change */ 446 447 delta_w = weight - (cl ? cl->agg->class_weight : 0); 448 449 if (q->wsum + delta_w > QFQ_MAX_WSUM) { 450 pr_notice("qfq: total weight out of range (%d + %u)\n", 451 delta_w, q->wsum); 452 return -EINVAL; 453 } 454 455 if (cl != NULL) { /* modify existing class */ 456 if (tca[TCA_RATE]) { 457 err = gen_replace_estimator(&cl->bstats, &cl->rate_est, 458 qdisc_root_sleeping_lock(sch), 459 tca[TCA_RATE]); 460 if (err) 461 return err; 462 } 463 existing = true; 464 goto set_change_agg; 465 } 466 467 /* create and init new class */ 468 cl = kzalloc(sizeof(struct qfq_class), GFP_KERNEL); 469 if (cl == NULL) 470 return -ENOBUFS; 471 472 cl->refcnt = 1; 473 cl->common.classid = classid; 474 cl->deficit = lmax; 475 476 cl->qdisc = qdisc_create_dflt(sch->dev_queue, 477 &pfifo_qdisc_ops, classid); 478 if (cl->qdisc == NULL) 479 cl->qdisc = &noop_qdisc; 480 481 if (tca[TCA_RATE]) { 482 err = gen_new_estimator(&cl->bstats, &cl->rate_est, 483 qdisc_root_sleeping_lock(sch), 484 tca[TCA_RATE]); 485 if (err) 486 goto destroy_class; 487 } 488 489 sch_tree_lock(sch); 490 qdisc_class_hash_insert(&q->clhash, &cl->common); 491 sch_tree_unlock(sch); 492 493 qdisc_class_hash_grow(sch, &q->clhash); 494 495set_change_agg: 496 sch_tree_lock(sch); 497 new_agg = qfq_find_agg(q, lmax, weight); 498 if (new_agg == NULL) { /* create new aggregate */ 499 sch_tree_unlock(sch); 500 new_agg = kzalloc(sizeof(*new_agg), GFP_KERNEL); 501 if (new_agg == NULL) { 502 err = -ENOBUFS; 503 gen_kill_estimator(&cl->bstats, &cl->rate_est); 504 goto destroy_class; 505 } 506 sch_tree_lock(sch); 507 qfq_init_agg(q, new_agg, lmax, weight); 508 } 509 if (existing) 510 qfq_deact_rm_from_agg(q, cl); 511 qfq_add_to_agg(q, new_agg, cl); 512 sch_tree_unlock(sch); 513 514 *arg = (unsigned long)cl; 515 return 0; 516 517destroy_class: 518 qdisc_destroy(cl->qdisc); 519 kfree(cl); 520 return err; 521} 522 523static void qfq_destroy_class(struct Qdisc *sch, struct qfq_class *cl) 524{ 525 struct qfq_sched *q = qdisc_priv(sch); 526 527 qfq_rm_from_agg(q, cl); 528 gen_kill_estimator(&cl->bstats, &cl->rate_est); 529 qdisc_destroy(cl->qdisc); 530 kfree(cl); 531} 532 533static int qfq_delete_class(struct Qdisc *sch, unsigned long arg) 534{ 535 struct qfq_sched *q = qdisc_priv(sch); 536 struct qfq_class *cl = (struct qfq_class *)arg; 537 538 if (cl->filter_cnt > 0) 539 return -EBUSY; 540 541 sch_tree_lock(sch); 542 543 qfq_purge_queue(cl); 544 qdisc_class_hash_remove(&q->clhash, &cl->common); 545 546 BUG_ON(--cl->refcnt == 0); 547 /* 548 * This shouldn't happen: we "hold" one cops->get() when called 549 * from tc_ctl_tclass; the destroy method is done from cops->put(). 550 */ 551 552 sch_tree_unlock(sch); 553 return 0; 554} 555 556static unsigned long qfq_get_class(struct Qdisc *sch, u32 classid) 557{ 558 struct qfq_class *cl = qfq_find_class(sch, classid); 559 560 if (cl != NULL) 561 cl->refcnt++; 562 563 return (unsigned long)cl; 564} 565 566static void qfq_put_class(struct Qdisc *sch, unsigned long arg) 567{ 568 struct qfq_class *cl = (struct qfq_class *)arg; 569 570 if (--cl->refcnt == 0) 571 qfq_destroy_class(sch, cl); 572} 573 574static struct tcf_proto **qfq_tcf_chain(struct Qdisc *sch, unsigned long cl) 575{ 576 struct qfq_sched *q = qdisc_priv(sch); 577 578 if (cl) 579 return NULL; 580 581 return &q->filter_list; 582} 583 584static unsigned long qfq_bind_tcf(struct Qdisc *sch, unsigned long parent, 585 u32 classid) 586{ 587 struct qfq_class *cl = qfq_find_class(sch, classid); 588 589 if (cl != NULL) 590 cl->filter_cnt++; 591 592 return (unsigned long)cl; 593} 594 595static void qfq_unbind_tcf(struct Qdisc *sch, unsigned long arg) 596{ 597 struct qfq_class *cl = (struct qfq_class *)arg; 598 599 cl->filter_cnt--; 600} 601 602static int qfq_graft_class(struct Qdisc *sch, unsigned long arg, 603 struct Qdisc *new, struct Qdisc **old) 604{ 605 struct qfq_class *cl = (struct qfq_class *)arg; 606 607 if (new == NULL) { 608 new = qdisc_create_dflt(sch->dev_queue, 609 &pfifo_qdisc_ops, cl->common.classid); 610 if (new == NULL) 611 new = &noop_qdisc; 612 } 613 614 sch_tree_lock(sch); 615 qfq_purge_queue(cl); 616 *old = cl->qdisc; 617 cl->qdisc = new; 618 sch_tree_unlock(sch); 619 return 0; 620} 621 622static struct Qdisc *qfq_class_leaf(struct Qdisc *sch, unsigned long arg) 623{ 624 struct qfq_class *cl = (struct qfq_class *)arg; 625 626 return cl->qdisc; 627} 628 629static int qfq_dump_class(struct Qdisc *sch, unsigned long arg, 630 struct sk_buff *skb, struct tcmsg *tcm) 631{ 632 struct qfq_class *cl = (struct qfq_class *)arg; 633 struct nlattr *nest; 634 635 tcm->tcm_parent = TC_H_ROOT; 636 tcm->tcm_handle = cl->common.classid; 637 tcm->tcm_info = cl->qdisc->handle; 638 639 nest = nla_nest_start(skb, TCA_OPTIONS); 640 if (nest == NULL) 641 goto nla_put_failure; 642 if (nla_put_u32(skb, TCA_QFQ_WEIGHT, cl->agg->class_weight) || 643 nla_put_u32(skb, TCA_QFQ_LMAX, cl->agg->lmax)) 644 goto nla_put_failure; 645 return nla_nest_end(skb, nest); 646 647nla_put_failure: 648 nla_nest_cancel(skb, nest); 649 return -EMSGSIZE; 650} 651 652static int qfq_dump_class_stats(struct Qdisc *sch, unsigned long arg, 653 struct gnet_dump *d) 654{ 655 struct qfq_class *cl = (struct qfq_class *)arg; 656 struct tc_qfq_stats xstats; 657 658 memset(&xstats, 0, sizeof(xstats)); 659 cl->qdisc->qstats.qlen = cl->qdisc->q.qlen; 660 661 xstats.weight = cl->agg->class_weight; 662 xstats.lmax = cl->agg->lmax; 663 664 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 || 665 gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 || 666 gnet_stats_copy_queue(d, &cl->qdisc->qstats) < 0) 667 return -1; 668 669 return gnet_stats_copy_app(d, &xstats, sizeof(xstats)); 670} 671 672static void qfq_walk(struct Qdisc *sch, struct qdisc_walker *arg) 673{ 674 struct qfq_sched *q = qdisc_priv(sch); 675 struct qfq_class *cl; 676 unsigned int i; 677 678 if (arg->stop) 679 return; 680 681 for (i = 0; i < q->clhash.hashsize; i++) { 682 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 683 if (arg->count < arg->skip) { 684 arg->count++; 685 continue; 686 } 687 if (arg->fn(sch, (unsigned long)cl, arg) < 0) { 688 arg->stop = 1; 689 return; 690 } 691 arg->count++; 692 } 693 } 694} 695 696static struct qfq_class *qfq_classify(struct sk_buff *skb, struct Qdisc *sch, 697 int *qerr) 698{ 699 struct qfq_sched *q = qdisc_priv(sch); 700 struct qfq_class *cl; 701 struct tcf_result res; 702 int result; 703 704 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) { 705 pr_debug("qfq_classify: found %d\n", skb->priority); 706 cl = qfq_find_class(sch, skb->priority); 707 if (cl != NULL) 708 return cl; 709 } 710 711 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 712 result = tc_classify(skb, q->filter_list, &res); 713 if (result >= 0) { 714#ifdef CONFIG_NET_CLS_ACT 715 switch (result) { 716 case TC_ACT_QUEUED: 717 case TC_ACT_STOLEN: 718 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 719 case TC_ACT_SHOT: 720 return NULL; 721 } 722#endif 723 cl = (struct qfq_class *)res.class; 724 if (cl == NULL) 725 cl = qfq_find_class(sch, res.classid); 726 return cl; 727 } 728 729 return NULL; 730} 731 732/* Generic comparison function, handling wraparound. */ 733static inline int qfq_gt(u64 a, u64 b) 734{ 735 return (s64)(a - b) > 0; 736} 737 738/* Round a precise timestamp to its slotted value. */ 739static inline u64 qfq_round_down(u64 ts, unsigned int shift) 740{ 741 return ts & ~((1ULL << shift) - 1); 742} 743 744/* return the pointer to the group with lowest index in the bitmap */ 745static inline struct qfq_group *qfq_ffs(struct qfq_sched *q, 746 unsigned long bitmap) 747{ 748 int index = __ffs(bitmap); 749 return &q->groups[index]; 750} 751/* Calculate a mask to mimic what would be ffs_from(). */ 752static inline unsigned long mask_from(unsigned long bitmap, int from) 753{ 754 return bitmap & ~((1UL << from) - 1); 755} 756 757/* 758 * The state computation relies on ER=0, IR=1, EB=2, IB=3 759 * First compute eligibility comparing grp->S, q->V, 760 * then check if someone is blocking us and possibly add EB 761 */ 762static int qfq_calc_state(struct qfq_sched *q, const struct qfq_group *grp) 763{ 764 /* if S > V we are not eligible */ 765 unsigned int state = qfq_gt(grp->S, q->V); 766 unsigned long mask = mask_from(q->bitmaps[ER], grp->index); 767 struct qfq_group *next; 768 769 if (mask) { 770 next = qfq_ffs(q, mask); 771 if (qfq_gt(grp->F, next->F)) 772 state |= EB; 773 } 774 775 return state; 776} 777 778 779/* 780 * In principle 781 * q->bitmaps[dst] |= q->bitmaps[src] & mask; 782 * q->bitmaps[src] &= ~mask; 783 * but we should make sure that src != dst 784 */ 785static inline void qfq_move_groups(struct qfq_sched *q, unsigned long mask, 786 int src, int dst) 787{ 788 q->bitmaps[dst] |= q->bitmaps[src] & mask; 789 q->bitmaps[src] &= ~mask; 790} 791 792static void qfq_unblock_groups(struct qfq_sched *q, int index, u64 old_F) 793{ 794 unsigned long mask = mask_from(q->bitmaps[ER], index + 1); 795 struct qfq_group *next; 796 797 if (mask) { 798 next = qfq_ffs(q, mask); 799 if (!qfq_gt(next->F, old_F)) 800 return; 801 } 802 803 mask = (1UL << index) - 1; 804 qfq_move_groups(q, mask, EB, ER); 805 qfq_move_groups(q, mask, IB, IR); 806} 807 808/* 809 * perhaps 810 * 811 old_V ^= q->V; 812 old_V >>= q->min_slot_shift; 813 if (old_V) { 814 ... 815 } 816 * 817 */ 818static void qfq_make_eligible(struct qfq_sched *q) 819{ 820 unsigned long vslot = q->V >> q->min_slot_shift; 821 unsigned long old_vslot = q->oldV >> q->min_slot_shift; 822 823 if (vslot != old_vslot) { 824 unsigned long mask = (1ULL << fls(vslot ^ old_vslot)) - 1; 825 qfq_move_groups(q, mask, IR, ER); 826 qfq_move_groups(q, mask, IB, EB); 827 } 828} 829 830 831/* 832 * The index of the slot in which the aggregate is to be inserted must 833 * not be higher than QFQ_MAX_SLOTS-2. There is a '-2' and not a '-1' 834 * because the start time of the group may be moved backward by one 835 * slot after the aggregate has been inserted, and this would cause 836 * non-empty slots to be right-shifted by one position. 837 * 838 * If the weight and lmax (max_pkt_size) of the classes do not change, 839 * then QFQ+ does meet the above contraint according to the current 840 * values of its parameters. In fact, if the weight and lmax of the 841 * classes do not change, then, from the theory, QFQ+ guarantees that 842 * the slot index is never higher than 843 * 2 + QFQ_MAX_AGG_CLASSES * ((1<<QFQ_MTU_SHIFT)/QFQ_MIN_LMAX) * 844 * (QFQ_MAX_WEIGHT/QFQ_MAX_WSUM) = 2 + 8 * 128 * (1 / 64) = 18 845 * 846 * When the weight of a class is increased or the lmax of the class is 847 * decreased, a new aggregate with smaller slot size than the original 848 * parent aggregate of the class may happen to be activated. The 849 * activation of this aggregate should be properly delayed to when the 850 * service of the class has finished in the ideal system tracked by 851 * QFQ+. If the activation of the aggregate is not delayed to this 852 * reference time instant, then this aggregate may be unjustly served 853 * before other aggregates waiting for service. This may cause the 854 * above bound to the slot index to be violated for some of these 855 * unlucky aggregates. 856 * 857 * Instead of delaying the activation of the new aggregate, which is 858 * quite complex, the following inaccurate but simple solution is used: 859 * if the slot index is higher than QFQ_MAX_SLOTS-2, then the 860 * timestamps of the aggregate are shifted backward so as to let the 861 * slot index become equal to QFQ_MAX_SLOTS-2. 862 */ 863static void qfq_slot_insert(struct qfq_group *grp, struct qfq_aggregate *agg, 864 u64 roundedS) 865{ 866 u64 slot = (roundedS - grp->S) >> grp->slot_shift; 867 unsigned int i; /* slot index in the bucket list */ 868 869 if (unlikely(slot > QFQ_MAX_SLOTS - 2)) { 870 u64 deltaS = roundedS - grp->S - 871 ((u64)(QFQ_MAX_SLOTS - 2)<<grp->slot_shift); 872 agg->S -= deltaS; 873 agg->F -= deltaS; 874 slot = QFQ_MAX_SLOTS - 2; 875 } 876 877 i = (grp->front + slot) % QFQ_MAX_SLOTS; 878 879 hlist_add_head(&agg->next, &grp->slots[i]); 880 __set_bit(slot, &grp->full_slots); 881} 882 883/* Maybe introduce hlist_first_entry?? */ 884static struct qfq_aggregate *qfq_slot_head(struct qfq_group *grp) 885{ 886 return hlist_entry(grp->slots[grp->front].first, 887 struct qfq_aggregate, next); 888} 889 890/* 891 * remove the entry from the slot 892 */ 893static void qfq_front_slot_remove(struct qfq_group *grp) 894{ 895 struct qfq_aggregate *agg = qfq_slot_head(grp); 896 897 BUG_ON(!agg); 898 hlist_del(&agg->next); 899 if (hlist_empty(&grp->slots[grp->front])) 900 __clear_bit(0, &grp->full_slots); 901} 902 903/* 904 * Returns the first aggregate in the first non-empty bucket of the 905 * group. As a side effect, adjusts the bucket list so the first 906 * non-empty bucket is at position 0 in full_slots. 907 */ 908static struct qfq_aggregate *qfq_slot_scan(struct qfq_group *grp) 909{ 910 unsigned int i; 911 912 pr_debug("qfq slot_scan: grp %u full %#lx\n", 913 grp->index, grp->full_slots); 914 915 if (grp->full_slots == 0) 916 return NULL; 917 918 i = __ffs(grp->full_slots); /* zero based */ 919 if (i > 0) { 920 grp->front = (grp->front + i) % QFQ_MAX_SLOTS; 921 grp->full_slots >>= i; 922 } 923 924 return qfq_slot_head(grp); 925} 926 927/* 928 * adjust the bucket list. When the start time of a group decreases, 929 * we move the index down (modulo QFQ_MAX_SLOTS) so we don't need to 930 * move the objects. The mask of occupied slots must be shifted 931 * because we use ffs() to find the first non-empty slot. 932 * This covers decreases in the group's start time, but what about 933 * increases of the start time ? 934 * Here too we should make sure that i is less than 32 935 */ 936static void qfq_slot_rotate(struct qfq_group *grp, u64 roundedS) 937{ 938 unsigned int i = (grp->S - roundedS) >> grp->slot_shift; 939 940 grp->full_slots <<= i; 941 grp->front = (grp->front - i) % QFQ_MAX_SLOTS; 942} 943 944static void qfq_update_eligible(struct qfq_sched *q) 945{ 946 struct qfq_group *grp; 947 unsigned long ineligible; 948 949 ineligible = q->bitmaps[IR] | q->bitmaps[IB]; 950 if (ineligible) { 951 if (!q->bitmaps[ER]) { 952 grp = qfq_ffs(q, ineligible); 953 if (qfq_gt(grp->S, q->V)) 954 q->V = grp->S; 955 } 956 qfq_make_eligible(q); 957 } 958} 959 960/* Dequeue head packet of the head class in the DRR queue of the aggregate. */ 961static void agg_dequeue(struct qfq_aggregate *agg, 962 struct qfq_class *cl, unsigned int len) 963{ 964 qdisc_dequeue_peeked(cl->qdisc); 965 966 cl->deficit -= (int) len; 967 968 if (cl->qdisc->q.qlen == 0) /* no more packets, remove from list */ 969 list_del(&cl->alist); 970 else if (cl->deficit < qdisc_pkt_len(cl->qdisc->ops->peek(cl->qdisc))) { 971 cl->deficit += agg->lmax; 972 list_move_tail(&cl->alist, &agg->active); 973 } 974} 975 976static inline struct sk_buff *qfq_peek_skb(struct qfq_aggregate *agg, 977 struct qfq_class **cl, 978 unsigned int *len) 979{ 980 struct sk_buff *skb; 981 982 *cl = list_first_entry(&agg->active, struct qfq_class, alist); 983 skb = (*cl)->qdisc->ops->peek((*cl)->qdisc); 984 if (skb == NULL) 985 WARN_ONCE(1, "qfq_dequeue: non-workconserving leaf\n"); 986 else 987 *len = qdisc_pkt_len(skb); 988 989 return skb; 990} 991 992/* Update F according to the actual service received by the aggregate. */ 993static inline void charge_actual_service(struct qfq_aggregate *agg) 994{ 995 /* Compute the service received by the aggregate, taking into 996 * account that, after decreasing the number of classes in 997 * agg, it may happen that 998 * agg->initial_budget - agg->budget > agg->bugdetmax 999 */ 1000 u32 service_received = min(agg->budgetmax, 1001 agg->initial_budget - agg->budget); 1002 1003 agg->F = agg->S + (u64)service_received * agg->inv_w; 1004} 1005 1006static inline void qfq_update_agg_ts(struct qfq_sched *q, 1007 struct qfq_aggregate *agg, 1008 enum update_reason reason); 1009 1010static void qfq_schedule_agg(struct qfq_sched *q, struct qfq_aggregate *agg); 1011 1012static struct sk_buff *qfq_dequeue(struct Qdisc *sch) 1013{ 1014 struct qfq_sched *q = qdisc_priv(sch); 1015 struct qfq_aggregate *in_serv_agg = q->in_serv_agg; 1016 struct qfq_class *cl; 1017 struct sk_buff *skb = NULL; 1018 /* next-packet len, 0 means no more active classes in in-service agg */ 1019 unsigned int len = 0; 1020 1021 if (in_serv_agg == NULL) 1022 return NULL; 1023 1024 if (!list_empty(&in_serv_agg->active)) 1025 skb = qfq_peek_skb(in_serv_agg, &cl, &len); 1026 1027 /* 1028 * If there are no active classes in the in-service aggregate, 1029 * or if the aggregate has not enough budget to serve its next 1030 * class, then choose the next aggregate to serve. 1031 */ 1032 if (len == 0 || in_serv_agg->budget < len) { 1033 charge_actual_service(in_serv_agg); 1034 1035 /* recharge the budget of the aggregate */ 1036 in_serv_agg->initial_budget = in_serv_agg->budget = 1037 in_serv_agg->budgetmax; 1038 1039 if (!list_empty(&in_serv_agg->active)) { 1040 /* 1041 * Still active: reschedule for 1042 * service. Possible optimization: if no other 1043 * aggregate is active, then there is no point 1044 * in rescheduling this aggregate, and we can 1045 * just keep it as the in-service one. This 1046 * should be however a corner case, and to 1047 * handle it, we would need to maintain an 1048 * extra num_active_aggs field. 1049 */ 1050 qfq_update_agg_ts(q, in_serv_agg, requeue); 1051 qfq_schedule_agg(q, in_serv_agg); 1052 } else if (sch->q.qlen == 0) { /* no aggregate to serve */ 1053 q->in_serv_agg = NULL; 1054 return NULL; 1055 } 1056 1057 /* 1058 * If we get here, there are other aggregates queued: 1059 * choose the new aggregate to serve. 1060 */ 1061 in_serv_agg = q->in_serv_agg = qfq_choose_next_agg(q); 1062 skb = qfq_peek_skb(in_serv_agg, &cl, &len); 1063 } 1064 if (!skb) 1065 return NULL; 1066 1067 sch->q.qlen--; 1068 qdisc_bstats_update(sch, skb); 1069 1070 agg_dequeue(in_serv_agg, cl, len); 1071 /* If lmax is lowered, through qfq_change_class, for a class 1072 * owning pending packets with larger size than the new value 1073 * of lmax, then the following condition may hold. 1074 */ 1075 if (unlikely(in_serv_agg->budget < len)) 1076 in_serv_agg->budget = 0; 1077 else 1078 in_serv_agg->budget -= len; 1079 1080 q->V += (u64)len * IWSUM; 1081 pr_debug("qfq dequeue: len %u F %lld now %lld\n", 1082 len, (unsigned long long) in_serv_agg->F, 1083 (unsigned long long) q->V); 1084 1085 return skb; 1086} 1087 1088static struct qfq_aggregate *qfq_choose_next_agg(struct qfq_sched *q) 1089{ 1090 struct qfq_group *grp; 1091 struct qfq_aggregate *agg, *new_front_agg; 1092 u64 old_F; 1093 1094 qfq_update_eligible(q); 1095 q->oldV = q->V; 1096 1097 if (!q->bitmaps[ER]) 1098 return NULL; 1099 1100 grp = qfq_ffs(q, q->bitmaps[ER]); 1101 old_F = grp->F; 1102 1103 agg = qfq_slot_head(grp); 1104 1105 /* agg starts to be served, remove it from schedule */ 1106 qfq_front_slot_remove(grp); 1107 1108 new_front_agg = qfq_slot_scan(grp); 1109 1110 if (new_front_agg == NULL) /* group is now inactive, remove from ER */ 1111 __clear_bit(grp->index, &q->bitmaps[ER]); 1112 else { 1113 u64 roundedS = qfq_round_down(new_front_agg->S, 1114 grp->slot_shift); 1115 unsigned int s; 1116 1117 if (grp->S == roundedS) 1118 return agg; 1119 grp->S = roundedS; 1120 grp->F = roundedS + (2ULL << grp->slot_shift); 1121 __clear_bit(grp->index, &q->bitmaps[ER]); 1122 s = qfq_calc_state(q, grp); 1123 __set_bit(grp->index, &q->bitmaps[s]); 1124 } 1125 1126 qfq_unblock_groups(q, grp->index, old_F); 1127 1128 return agg; 1129} 1130 1131/* 1132 * Assign a reasonable start time for a new aggregate in group i. 1133 * Admissible values for \hat(F) are multiples of \sigma_i 1134 * no greater than V+\sigma_i . Larger values mean that 1135 * we had a wraparound so we consider the timestamp to be stale. 1136 * 1137 * If F is not stale and F >= V then we set S = F. 1138 * Otherwise we should assign S = V, but this may violate 1139 * the ordering in EB (see [2]). So, if we have groups in ER, 1140 * set S to the F_j of the first group j which would be blocking us. 1141 * We are guaranteed not to move S backward because 1142 * otherwise our group i would still be blocked. 1143 */ 1144static void qfq_update_start(struct qfq_sched *q, struct qfq_aggregate *agg) 1145{ 1146 unsigned long mask; 1147 u64 limit, roundedF; 1148 int slot_shift = agg->grp->slot_shift; 1149 1150 roundedF = qfq_round_down(agg->F, slot_shift); 1151 limit = qfq_round_down(q->V, slot_shift) + (1ULL << slot_shift); 1152 1153 if (!qfq_gt(agg->F, q->V) || qfq_gt(roundedF, limit)) { 1154 /* timestamp was stale */ 1155 mask = mask_from(q->bitmaps[ER], agg->grp->index); 1156 if (mask) { 1157 struct qfq_group *next = qfq_ffs(q, mask); 1158 if (qfq_gt(roundedF, next->F)) { 1159 if (qfq_gt(limit, next->F)) 1160 agg->S = next->F; 1161 else /* preserve timestamp correctness */ 1162 agg->S = limit; 1163 return; 1164 } 1165 } 1166 agg->S = q->V; 1167 } else /* timestamp is not stale */ 1168 agg->S = agg->F; 1169} 1170 1171/* 1172 * Update the timestamps of agg before scheduling/rescheduling it for 1173 * service. In particular, assign to agg->F its maximum possible 1174 * value, i.e., the virtual finish time with which the aggregate 1175 * should be labeled if it used all its budget once in service. 1176 */ 1177static inline void 1178qfq_update_agg_ts(struct qfq_sched *q, 1179 struct qfq_aggregate *agg, enum update_reason reason) 1180{ 1181 if (reason != requeue) 1182 qfq_update_start(q, agg); 1183 else /* just charge agg for the service received */ 1184 agg->S = agg->F; 1185 1186 agg->F = agg->S + (u64)agg->budgetmax * agg->inv_w; 1187} 1188 1189static void qfq_schedule_agg(struct qfq_sched *, struct qfq_aggregate *); 1190 1191static int qfq_enqueue(struct sk_buff *skb, struct Qdisc *sch) 1192{ 1193 struct qfq_sched *q = qdisc_priv(sch); 1194 struct qfq_class *cl; 1195 struct qfq_aggregate *agg; 1196 int err = 0; 1197 1198 cl = qfq_classify(skb, sch, &err); 1199 if (cl == NULL) { 1200 if (err & __NET_XMIT_BYPASS) 1201 sch->qstats.drops++; 1202 kfree_skb(skb); 1203 return err; 1204 } 1205 pr_debug("qfq_enqueue: cl = %x\n", cl->common.classid); 1206 1207 if (unlikely(cl->agg->lmax < qdisc_pkt_len(skb))) { 1208 pr_debug("qfq: increasing maxpkt from %u to %u for class %u", 1209 cl->agg->lmax, qdisc_pkt_len(skb), cl->common.classid); 1210 err = qfq_change_agg(sch, cl, cl->agg->class_weight, 1211 qdisc_pkt_len(skb)); 1212 if (err) 1213 return err; 1214 } 1215 1216 err = qdisc_enqueue(skb, cl->qdisc); 1217 if (unlikely(err != NET_XMIT_SUCCESS)) { 1218 pr_debug("qfq_enqueue: enqueue failed %d\n", err); 1219 if (net_xmit_drop_count(err)) { 1220 cl->qstats.drops++; 1221 sch->qstats.drops++; 1222 } 1223 return err; 1224 } 1225 1226 bstats_update(&cl->bstats, skb); 1227 ++sch->q.qlen; 1228 1229 agg = cl->agg; 1230 /* if the queue was not empty, then done here */ 1231 if (cl->qdisc->q.qlen != 1) { 1232 if (unlikely(skb == cl->qdisc->ops->peek(cl->qdisc)) && 1233 list_first_entry(&agg->active, struct qfq_class, alist) 1234 == cl && cl->deficit < qdisc_pkt_len(skb)) 1235 list_move_tail(&cl->alist, &agg->active); 1236 1237 return err; 1238 } 1239 1240 /* schedule class for service within the aggregate */ 1241 cl->deficit = agg->lmax; 1242 list_add_tail(&cl->alist, &agg->active); 1243 1244 if (list_first_entry(&agg->active, struct qfq_class, alist) != cl || 1245 q->in_serv_agg == agg) 1246 return err; /* non-empty or in service, nothing else to do */ 1247 1248 qfq_activate_agg(q, agg, enqueue); 1249 1250 return err; 1251} 1252 1253/* 1254 * Schedule aggregate according to its timestamps. 1255 */ 1256static void qfq_schedule_agg(struct qfq_sched *q, struct qfq_aggregate *agg) 1257{ 1258 struct qfq_group *grp = agg->grp; 1259 u64 roundedS; 1260 int s; 1261 1262 roundedS = qfq_round_down(agg->S, grp->slot_shift); 1263 1264 /* 1265 * Insert agg in the correct bucket. 1266 * If agg->S >= grp->S we don't need to adjust the 1267 * bucket list and simply go to the insertion phase. 1268 * Otherwise grp->S is decreasing, we must make room 1269 * in the bucket list, and also recompute the group state. 1270 * Finally, if there were no flows in this group and nobody 1271 * was in ER make sure to adjust V. 1272 */ 1273 if (grp->full_slots) { 1274 if (!qfq_gt(grp->S, agg->S)) 1275 goto skip_update; 1276 1277 /* create a slot for this agg->S */ 1278 qfq_slot_rotate(grp, roundedS); 1279 /* group was surely ineligible, remove */ 1280 __clear_bit(grp->index, &q->bitmaps[IR]); 1281 __clear_bit(grp->index, &q->bitmaps[IB]); 1282 } else if (!q->bitmaps[ER] && qfq_gt(roundedS, q->V) && 1283 q->in_serv_agg == NULL) 1284 q->V = roundedS; 1285 1286 grp->S = roundedS; 1287 grp->F = roundedS + (2ULL << grp->slot_shift); 1288 s = qfq_calc_state(q, grp); 1289 __set_bit(grp->index, &q->bitmaps[s]); 1290 1291 pr_debug("qfq enqueue: new state %d %#lx S %lld F %lld V %lld\n", 1292 s, q->bitmaps[s], 1293 (unsigned long long) agg->S, 1294 (unsigned long long) agg->F, 1295 (unsigned long long) q->V); 1296 1297skip_update: 1298 qfq_slot_insert(grp, agg, roundedS); 1299} 1300 1301 1302/* Update agg ts and schedule agg for service */ 1303static void qfq_activate_agg(struct qfq_sched *q, struct qfq_aggregate *agg, 1304 enum update_reason reason) 1305{ 1306 agg->initial_budget = agg->budget = agg->budgetmax; /* recharge budg. */ 1307 1308 qfq_update_agg_ts(q, agg, reason); 1309 if (q->in_serv_agg == NULL) { /* no aggr. in service or scheduled */ 1310 q->in_serv_agg = agg; /* start serving this aggregate */ 1311 /* update V: to be in service, agg must be eligible */ 1312 q->oldV = q->V = agg->S; 1313 } else if (agg != q->in_serv_agg) 1314 qfq_schedule_agg(q, agg); 1315} 1316 1317static void qfq_slot_remove(struct qfq_sched *q, struct qfq_group *grp, 1318 struct qfq_aggregate *agg) 1319{ 1320 unsigned int i, offset; 1321 u64 roundedS; 1322 1323 roundedS = qfq_round_down(agg->S, grp->slot_shift); 1324 offset = (roundedS - grp->S) >> grp->slot_shift; 1325 1326 i = (grp->front + offset) % QFQ_MAX_SLOTS; 1327 1328 hlist_del(&agg->next); 1329 if (hlist_empty(&grp->slots[i])) 1330 __clear_bit(offset, &grp->full_slots); 1331} 1332 1333/* 1334 * Called to forcibly deschedule an aggregate. If the aggregate is 1335 * not in the front bucket, or if the latter has other aggregates in 1336 * the front bucket, we can simply remove the aggregate with no other 1337 * side effects. 1338 * Otherwise we must propagate the event up. 1339 */ 1340static void qfq_deactivate_agg(struct qfq_sched *q, struct qfq_aggregate *agg) 1341{ 1342 struct qfq_group *grp = agg->grp; 1343 unsigned long mask; 1344 u64 roundedS; 1345 int s; 1346 1347 if (agg == q->in_serv_agg) { 1348 charge_actual_service(agg); 1349 q->in_serv_agg = qfq_choose_next_agg(q); 1350 return; 1351 } 1352 1353 agg->F = agg->S; 1354 qfq_slot_remove(q, grp, agg); 1355 1356 if (!grp->full_slots) { 1357 __clear_bit(grp->index, &q->bitmaps[IR]); 1358 __clear_bit(grp->index, &q->bitmaps[EB]); 1359 __clear_bit(grp->index, &q->bitmaps[IB]); 1360 1361 if (test_bit(grp->index, &q->bitmaps[ER]) && 1362 !(q->bitmaps[ER] & ~((1UL << grp->index) - 1))) { 1363 mask = q->bitmaps[ER] & ((1UL << grp->index) - 1); 1364 if (mask) 1365 mask = ~((1UL << __fls(mask)) - 1); 1366 else 1367 mask = ~0UL; 1368 qfq_move_groups(q, mask, EB, ER); 1369 qfq_move_groups(q, mask, IB, IR); 1370 } 1371 __clear_bit(grp->index, &q->bitmaps[ER]); 1372 } else if (hlist_empty(&grp->slots[grp->front])) { 1373 agg = qfq_slot_scan(grp); 1374 roundedS = qfq_round_down(agg->S, grp->slot_shift); 1375 if (grp->S != roundedS) { 1376 __clear_bit(grp->index, &q->bitmaps[ER]); 1377 __clear_bit(grp->index, &q->bitmaps[IR]); 1378 __clear_bit(grp->index, &q->bitmaps[EB]); 1379 __clear_bit(grp->index, &q->bitmaps[IB]); 1380 grp->S = roundedS; 1381 grp->F = roundedS + (2ULL << grp->slot_shift); 1382 s = qfq_calc_state(q, grp); 1383 __set_bit(grp->index, &q->bitmaps[s]); 1384 } 1385 } 1386} 1387 1388static void qfq_qlen_notify(struct Qdisc *sch, unsigned long arg) 1389{ 1390 struct qfq_sched *q = qdisc_priv(sch); 1391 struct qfq_class *cl = (struct qfq_class *)arg; 1392 1393 if (cl->qdisc->q.qlen == 0) 1394 qfq_deactivate_class(q, cl); 1395} 1396 1397static unsigned int qfq_drop_from_slot(struct qfq_sched *q, 1398 struct hlist_head *slot) 1399{ 1400 struct qfq_aggregate *agg; 1401 struct qfq_class *cl; 1402 unsigned int len; 1403 1404 hlist_for_each_entry(agg, slot, next) { 1405 list_for_each_entry(cl, &agg->active, alist) { 1406 1407 if (!cl->qdisc->ops->drop) 1408 continue; 1409 1410 len = cl->qdisc->ops->drop(cl->qdisc); 1411 if (len > 0) { 1412 if (cl->qdisc->q.qlen == 0) 1413 qfq_deactivate_class(q, cl); 1414 1415 return len; 1416 } 1417 } 1418 } 1419 return 0; 1420} 1421 1422static unsigned int qfq_drop(struct Qdisc *sch) 1423{ 1424 struct qfq_sched *q = qdisc_priv(sch); 1425 struct qfq_group *grp; 1426 unsigned int i, j, len; 1427 1428 for (i = 0; i <= QFQ_MAX_INDEX; i++) { 1429 grp = &q->groups[i]; 1430 for (j = 0; j < QFQ_MAX_SLOTS; j++) { 1431 len = qfq_drop_from_slot(q, &grp->slots[j]); 1432 if (len > 0) { 1433 sch->q.qlen--; 1434 return len; 1435 } 1436 } 1437 1438 } 1439 1440 return 0; 1441} 1442 1443static int qfq_init_qdisc(struct Qdisc *sch, struct nlattr *opt) 1444{ 1445 struct qfq_sched *q = qdisc_priv(sch); 1446 struct qfq_group *grp; 1447 int i, j, err; 1448 u32 max_cl_shift, maxbudg_shift, max_classes; 1449 1450 err = qdisc_class_hash_init(&q->clhash); 1451 if (err < 0) 1452 return err; 1453 1454 if (qdisc_dev(sch)->tx_queue_len + 1 > QFQ_MAX_AGG_CLASSES) 1455 max_classes = QFQ_MAX_AGG_CLASSES; 1456 else 1457 max_classes = qdisc_dev(sch)->tx_queue_len + 1; 1458 /* max_cl_shift = floor(log_2(max_classes)) */ 1459 max_cl_shift = __fls(max_classes); 1460 q->max_agg_classes = 1<<max_cl_shift; 1461 1462 /* maxbudg_shift = log2(max_len * max_classes_per_agg) */ 1463 maxbudg_shift = QFQ_MTU_SHIFT + max_cl_shift; 1464 q->min_slot_shift = FRAC_BITS + maxbudg_shift - QFQ_MAX_INDEX; 1465 1466 for (i = 0; i <= QFQ_MAX_INDEX; i++) { 1467 grp = &q->groups[i]; 1468 grp->index = i; 1469 grp->slot_shift = q->min_slot_shift + i; 1470 for (j = 0; j < QFQ_MAX_SLOTS; j++) 1471 INIT_HLIST_HEAD(&grp->slots[j]); 1472 } 1473 1474 INIT_HLIST_HEAD(&q->nonfull_aggs); 1475 1476 return 0; 1477} 1478 1479static void qfq_reset_qdisc(struct Qdisc *sch) 1480{ 1481 struct qfq_sched *q = qdisc_priv(sch); 1482 struct qfq_class *cl; 1483 unsigned int i; 1484 1485 for (i = 0; i < q->clhash.hashsize; i++) { 1486 hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) { 1487 if (cl->qdisc->q.qlen > 0) 1488 qfq_deactivate_class(q, cl); 1489 1490 qdisc_reset(cl->qdisc); 1491 } 1492 } 1493 sch->q.qlen = 0; 1494} 1495 1496static void qfq_destroy_qdisc(struct Qdisc *sch) 1497{ 1498 struct qfq_sched *q = qdisc_priv(sch); 1499 struct qfq_class *cl; 1500 struct hlist_node *next; 1501 unsigned int i; 1502 1503 tcf_destroy_chain(&q->filter_list); 1504 1505 for (i = 0; i < q->clhash.hashsize; i++) { 1506 hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i], 1507 common.hnode) { 1508 qfq_destroy_class(sch, cl); 1509 } 1510 } 1511 qdisc_class_hash_destroy(&q->clhash); 1512} 1513 1514static const struct Qdisc_class_ops qfq_class_ops = { 1515 .change = qfq_change_class, 1516 .delete = qfq_delete_class, 1517 .get = qfq_get_class, 1518 .put = qfq_put_class, 1519 .tcf_chain = qfq_tcf_chain, 1520 .bind_tcf = qfq_bind_tcf, 1521 .unbind_tcf = qfq_unbind_tcf, 1522 .graft = qfq_graft_class, 1523 .leaf = qfq_class_leaf, 1524 .qlen_notify = qfq_qlen_notify, 1525 .dump = qfq_dump_class, 1526 .dump_stats = qfq_dump_class_stats, 1527 .walk = qfq_walk, 1528}; 1529 1530static struct Qdisc_ops qfq_qdisc_ops __read_mostly = { 1531 .cl_ops = &qfq_class_ops, 1532 .id = "qfq", 1533 .priv_size = sizeof(struct qfq_sched), 1534 .enqueue = qfq_enqueue, 1535 .dequeue = qfq_dequeue, 1536 .peek = qdisc_peek_dequeued, 1537 .drop = qfq_drop, 1538 .init = qfq_init_qdisc, 1539 .reset = qfq_reset_qdisc, 1540 .destroy = qfq_destroy_qdisc, 1541 .owner = THIS_MODULE, 1542}; 1543 1544static int __init qfq_init(void) 1545{ 1546 return register_qdisc(&qfq_qdisc_ops); 1547} 1548 1549static void __exit qfq_exit(void) 1550{ 1551 unregister_qdisc(&qfq_qdisc_ops); 1552} 1553 1554module_init(qfq_init); 1555module_exit(qfq_exit); 1556MODULE_LICENSE("GPL"); 1557