2 * net/sched/sch_cbs.c Credit Based Shaper
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Vinicius Costa Gomes <vinicius.gomes@intel.com>
13 /* Credit Based Shaper (CBS)
14 * =========================
16 * This is a simple rate-limiting shaper aimed at TSN applications on
17 * systems with known traffic workloads.
19 * Its algorithm is defined by the IEEE 802.1Q-2014 Specification,
20 * Section 8.6.8.2, and explained in more detail in the Annex L of the
23 * There are four tunables to be considered:
25 * 'idleslope': Idleslope is the rate of credits that is
26 * accumulated (in kilobits per second) when there is at least
27 * one packet waiting for transmission. Packets are transmitted
28 * when the current value of credits is equal or greater than
29 * zero. When there is no packet to be transmitted the amount of
30 * credits is set to zero. This is the main tunable of the CBS
34 * Sendslope is the rate of credits that is depleted (it should be a
35 * negative number of kilobits per second) when a transmission is
36 * ocurring. It can be calculated as follows, (IEEE 802.1Q-2014 Section
39 * sendslope = idleslope - port_transmit_rate
41 * 'hicredit': Hicredit defines the maximum amount of credits (in
42 * bytes) that can be accumulated. Hicredit depends on the
43 * characteristics of interfering traffic,
44 * 'max_interference_size' is the maximum size of any burst of
45 * traffic that can delay the transmission of a frame that is
46 * available for transmission for this traffic class, (IEEE
47 * 802.1Q-2014 Annex L, Equation L-3):
49 * hicredit = max_interference_size * (idleslope / port_transmit_rate)
51 * 'locredit': Locredit is the minimum amount of credits that can
52 * be reached. It is a function of the traffic flowing through
53 * this qdisc (IEEE 802.1Q-2014 Annex L, Equation L-2):
55 * locredit = max_frame_size * (sendslope / port_transmit_rate)
58 #include <linux/module.h>
59 #include <linux/types.h>
60 #include <linux/kernel.h>
61 #include <linux/string.h>
62 #include <linux/errno.h>
63 #include <linux/skbuff.h>
64 #include <net/netevent.h>
65 #include <net/netlink.h>
66 #include <net/sch_generic.h>
67 #include <net/pkt_sched.h>
69 static LIST_HEAD(cbs_list);
70 static DEFINE_SPINLOCK(cbs_list_lock);
72 #define BYTES_PER_KBIT (1000LL / 8)
74 struct cbs_sched_data {
77 atomic64_t port_rate; /* in bytes/s */
78 s64 last; /* timestamp in ns */
79 s64 credits; /* in bytes */
80 s32 locredit; /* in bytes */
81 s32 hicredit; /* in bytes */
82 s64 sendslope; /* in bytes/s */
83 s64 idleslope; /* in bytes/s */
84 struct qdisc_watchdog watchdog;
85 int (*enqueue)(struct sk_buff *skb, struct Qdisc *sch,
86 struct sk_buff **to_free);
87 struct sk_buff *(*dequeue)(struct Qdisc *sch);
89 struct list_head cbs_list;
92 static int cbs_child_enqueue(struct sk_buff *skb, struct Qdisc *sch,
94 struct sk_buff **to_free)
96 unsigned int len = qdisc_pkt_len(skb);
99 err = child->ops->enqueue(skb, child, to_free);
100 if (err != NET_XMIT_SUCCESS)
103 sch->qstats.backlog += len;
106 return NET_XMIT_SUCCESS;
109 static int cbs_enqueue_offload(struct sk_buff *skb, struct Qdisc *sch,
110 struct sk_buff **to_free)
112 struct cbs_sched_data *q = qdisc_priv(sch);
113 struct Qdisc *qdisc = q->qdisc;
115 return cbs_child_enqueue(skb, sch, qdisc, to_free);
118 static int cbs_enqueue_soft(struct sk_buff *skb, struct Qdisc *sch,
119 struct sk_buff **to_free)
121 struct cbs_sched_data *q = qdisc_priv(sch);
122 struct Qdisc *qdisc = q->qdisc;
124 if (sch->q.qlen == 0 && q->credits > 0) {
125 /* We need to stop accumulating credits when there's
126 * no enqueued packets and q->credits is positive.
129 q->last = ktime_get_ns();
132 return cbs_child_enqueue(skb, sch, qdisc, to_free);
135 static int cbs_enqueue(struct sk_buff *skb, struct Qdisc *sch,
136 struct sk_buff **to_free)
138 struct cbs_sched_data *q = qdisc_priv(sch);
140 return q->enqueue(skb, sch, to_free);
143 /* timediff is in ns, slope is in bytes/s */
144 static s64 timediff_to_credits(s64 timediff, s64 slope)
146 return div64_s64(timediff * slope, NSEC_PER_SEC);
149 static s64 delay_from_credits(s64 credits, s64 slope)
151 if (unlikely(slope == 0))
154 return div64_s64(-credits * NSEC_PER_SEC, slope);
157 static s64 credits_from_len(unsigned int len, s64 slope, s64 port_rate)
159 if (unlikely(port_rate == 0))
162 return div64_s64(len * slope, port_rate);
165 static struct sk_buff *cbs_child_dequeue(struct Qdisc *sch, struct Qdisc *child)
169 skb = child->ops->dequeue(child);
173 qdisc_qstats_backlog_dec(sch, skb);
174 qdisc_bstats_update(sch, skb);
180 static struct sk_buff *cbs_dequeue_soft(struct Qdisc *sch)
182 struct cbs_sched_data *q = qdisc_priv(sch);
183 struct Qdisc *qdisc = q->qdisc;
184 s64 now = ktime_get_ns();
189 if (atomic64_read(&q->port_rate) == -1) {
190 WARN_ONCE(1, "cbs: dequeue() called with unknown port rate.");
194 if (q->credits < 0) {
195 credits = timediff_to_credits(now - q->last, q->idleslope);
197 credits = q->credits + credits;
198 q->credits = min_t(s64, credits, q->hicredit);
200 if (q->credits < 0) {
203 delay = delay_from_credits(q->credits, q->idleslope);
204 qdisc_watchdog_schedule_ns(&q->watchdog, now + delay);
211 skb = cbs_child_dequeue(sch, qdisc);
215 len = qdisc_pkt_len(skb);
217 /* As sendslope is a negative number, this will decrease the
218 * amount of q->credits.
220 credits = credits_from_len(len, q->sendslope,
221 atomic64_read(&q->port_rate));
222 credits += q->credits;
224 q->credits = max_t(s64, credits, q->locredit);
230 static struct sk_buff *cbs_dequeue_offload(struct Qdisc *sch)
232 struct cbs_sched_data *q = qdisc_priv(sch);
233 struct Qdisc *qdisc = q->qdisc;
235 return cbs_child_dequeue(sch, qdisc);
238 static struct sk_buff *cbs_dequeue(struct Qdisc *sch)
240 struct cbs_sched_data *q = qdisc_priv(sch);
242 return q->dequeue(sch);
245 static const struct nla_policy cbs_policy[TCA_CBS_MAX + 1] = {
246 [TCA_CBS_PARMS] = { .len = sizeof(struct tc_cbs_qopt) },
249 static void cbs_disable_offload(struct net_device *dev,
250 struct cbs_sched_data *q)
252 struct tc_cbs_qopt_offload cbs = { };
253 const struct net_device_ops *ops;
259 q->enqueue = cbs_enqueue_soft;
260 q->dequeue = cbs_dequeue_soft;
262 ops = dev->netdev_ops;
263 if (!ops->ndo_setup_tc)
266 cbs.queue = q->queue;
269 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_CBS, &cbs);
271 pr_warn("Couldn't disable CBS offload for queue %d\n",
275 static int cbs_enable_offload(struct net_device *dev, struct cbs_sched_data *q,
276 const struct tc_cbs_qopt *opt,
277 struct netlink_ext_ack *extack)
279 const struct net_device_ops *ops = dev->netdev_ops;
280 struct tc_cbs_qopt_offload cbs = { };
283 if (!ops->ndo_setup_tc) {
284 NL_SET_ERR_MSG(extack, "Specified device does not support cbs offload");
288 cbs.queue = q->queue;
291 cbs.hicredit = opt->hicredit;
292 cbs.locredit = opt->locredit;
293 cbs.idleslope = opt->idleslope;
294 cbs.sendslope = opt->sendslope;
296 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_CBS, &cbs);
298 NL_SET_ERR_MSG(extack, "Specified device failed to setup cbs hardware offload");
302 q->enqueue = cbs_enqueue_offload;
303 q->dequeue = cbs_dequeue_offload;
308 static void cbs_set_port_rate(struct net_device *dev, struct cbs_sched_data *q)
310 struct ethtool_link_ksettings ecmd;
313 if (!__ethtool_get_link_ksettings(dev, &ecmd) &&
314 ecmd.base.speed != SPEED_UNKNOWN)
315 port_rate = ecmd.base.speed * 1000 * BYTES_PER_KBIT;
317 atomic64_set(&q->port_rate, port_rate);
318 netdev_dbg(dev, "cbs: set %s's port_rate to: %lld, linkspeed: %d\n",
319 dev->name, (long long)atomic64_read(&q->port_rate),
323 static int cbs_dev_notifier(struct notifier_block *nb, unsigned long event,
326 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
327 struct cbs_sched_data *q;
328 struct net_device *qdev;
333 if (event != NETDEV_UP && event != NETDEV_CHANGE)
336 spin_lock(&cbs_list_lock);
337 list_for_each_entry(q, &cbs_list, cbs_list) {
338 qdev = qdisc_dev(q->qdisc);
344 spin_unlock(&cbs_list_lock);
347 cbs_set_port_rate(dev, q);
352 static int cbs_change(struct Qdisc *sch, struct nlattr *opt,
353 struct netlink_ext_ack *extack)
355 struct cbs_sched_data *q = qdisc_priv(sch);
356 struct net_device *dev = qdisc_dev(sch);
357 struct nlattr *tb[TCA_CBS_MAX + 1];
358 struct tc_cbs_qopt *qopt;
361 err = nla_parse_nested_deprecated(tb, TCA_CBS_MAX, opt, cbs_policy,
366 if (!tb[TCA_CBS_PARMS]) {
367 NL_SET_ERR_MSG(extack, "Missing CBS parameter which are mandatory");
371 qopt = nla_data(tb[TCA_CBS_PARMS]);
373 if (!qopt->offload) {
374 cbs_set_port_rate(dev, q);
375 cbs_disable_offload(dev, q);
377 err = cbs_enable_offload(dev, q, qopt, extack);
382 /* Everything went OK, save the parameters used. */
383 q->hicredit = qopt->hicredit;
384 q->locredit = qopt->locredit;
385 q->idleslope = qopt->idleslope * BYTES_PER_KBIT;
386 q->sendslope = qopt->sendslope * BYTES_PER_KBIT;
387 q->offload = qopt->offload;
392 static int cbs_init(struct Qdisc *sch, struct nlattr *opt,
393 struct netlink_ext_ack *extack)
395 struct cbs_sched_data *q = qdisc_priv(sch);
396 struct net_device *dev = qdisc_dev(sch);
400 NL_SET_ERR_MSG(extack, "Missing CBS qdisc options which are mandatory");
404 q->qdisc = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
405 sch->handle, extack);
409 qdisc_hash_add(q->qdisc, false);
411 q->queue = sch->dev_queue - netdev_get_tx_queue(dev, 0);
413 q->enqueue = cbs_enqueue_soft;
414 q->dequeue = cbs_dequeue_soft;
416 qdisc_watchdog_init(&q->watchdog, sch);
418 err = cbs_change(sch, opt, extack);
423 spin_lock(&cbs_list_lock);
424 list_add(&q->cbs_list, &cbs_list);
425 spin_unlock(&cbs_list_lock);
431 static void cbs_destroy(struct Qdisc *sch)
433 struct cbs_sched_data *q = qdisc_priv(sch);
434 struct net_device *dev = qdisc_dev(sch);
436 spin_lock(&cbs_list_lock);
437 list_del(&q->cbs_list);
438 spin_unlock(&cbs_list_lock);
440 qdisc_watchdog_cancel(&q->watchdog);
441 cbs_disable_offload(dev, q);
447 static int cbs_dump(struct Qdisc *sch, struct sk_buff *skb)
449 struct cbs_sched_data *q = qdisc_priv(sch);
450 struct tc_cbs_qopt opt = { };
453 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
455 goto nla_put_failure;
457 opt.hicredit = q->hicredit;
458 opt.locredit = q->locredit;
459 opt.sendslope = div64_s64(q->sendslope, BYTES_PER_KBIT);
460 opt.idleslope = div64_s64(q->idleslope, BYTES_PER_KBIT);
461 opt.offload = q->offload;
463 if (nla_put(skb, TCA_CBS_PARMS, sizeof(opt), &opt))
464 goto nla_put_failure;
466 return nla_nest_end(skb, nest);
469 nla_nest_cancel(skb, nest);
473 static int cbs_dump_class(struct Qdisc *sch, unsigned long cl,
474 struct sk_buff *skb, struct tcmsg *tcm)
476 struct cbs_sched_data *q = qdisc_priv(sch);
478 if (cl != 1 || !q->qdisc) /* only one class */
481 tcm->tcm_handle |= TC_H_MIN(1);
482 tcm->tcm_info = q->qdisc->handle;
487 static int cbs_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
488 struct Qdisc **old, struct netlink_ext_ack *extack)
490 struct cbs_sched_data *q = qdisc_priv(sch);
493 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
499 *old = qdisc_replace(sch, new, &q->qdisc);
503 static struct Qdisc *cbs_leaf(struct Qdisc *sch, unsigned long arg)
505 struct cbs_sched_data *q = qdisc_priv(sch);
510 static unsigned long cbs_find(struct Qdisc *sch, u32 classid)
515 static void cbs_walk(struct Qdisc *sch, struct qdisc_walker *walker)
518 if (walker->count >= walker->skip) {
519 if (walker->fn(sch, 1, walker) < 0) {
528 static const struct Qdisc_class_ops cbs_class_ops = {
533 .dump = cbs_dump_class,
536 static struct Qdisc_ops cbs_qdisc_ops __read_mostly = {
538 .cl_ops = &cbs_class_ops,
539 .priv_size = sizeof(struct cbs_sched_data),
540 .enqueue = cbs_enqueue,
541 .dequeue = cbs_dequeue,
542 .peek = qdisc_peek_dequeued,
544 .reset = qdisc_reset_queue,
545 .destroy = cbs_destroy,
546 .change = cbs_change,
548 .owner = THIS_MODULE,
551 static struct notifier_block cbs_device_notifier = {
552 .notifier_call = cbs_dev_notifier,
555 static int __init cbs_module_init(void)
557 int err = register_netdevice_notifier(&cbs_device_notifier);
562 return register_qdisc(&cbs_qdisc_ops);
565 static void __exit cbs_module_exit(void)
567 unregister_qdisc(&cbs_qdisc_ops);
568 unregister_netdevice_notifier(&cbs_device_notifier);
570 module_init(cbs_module_init)
571 module_exit(cbs_module_exit)
572 MODULE_LICENSE("GPL");