xref: /openbmc/linux/net/sched/sch_qfq.c (revision e3d786a3)
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