1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net/sched/sch_cbs.c Credit Based Shaper
5 * Authors: Vinicius Costa Gomes <vinicius.gomes@intel.com>
8 /* Credit Based Shaper (CBS)
9 * =========================
11 * This is a simple rate-limiting shaper aimed at TSN applications on
12 * systems with known traffic workloads.
14 * Its algorithm is defined by the IEEE 802.1Q-2014 Specification,
15 * Section 8.6.8.2, and explained in more detail in the Annex L of the
18 * There are four tunables to be considered:
20 * 'idleslope': Idleslope is the rate of credits that is
21 * accumulated (in kilobits per second) when there is at least
22 * one packet waiting for transmission. Packets are transmitted
23 * when the current value of credits is equal or greater than
24 * zero. When there is no packet to be transmitted the amount of
25 * credits is set to zero. This is the main tunable of the CBS
29 * Sendslope is the rate of credits that is depleted (it should be a
30 * negative number of kilobits per second) when a transmission is
31 * ocurring. It can be calculated as follows, (IEEE 802.1Q-2014 Section
34 * sendslope = idleslope - port_transmit_rate
36 * 'hicredit': Hicredit defines the maximum amount of credits (in
37 * bytes) that can be accumulated. Hicredit depends on the
38 * characteristics of interfering traffic,
39 * 'max_interference_size' is the maximum size of any burst of
40 * traffic that can delay the transmission of a frame that is
41 * available for transmission for this traffic class, (IEEE
42 * 802.1Q-2014 Annex L, Equation L-3):
44 * hicredit = max_interference_size * (idleslope / port_transmit_rate)
46 * 'locredit': Locredit is the minimum amount of credits that can
47 * be reached. It is a function of the traffic flowing through
48 * this qdisc (IEEE 802.1Q-2014 Annex L, Equation L-2):
50 * locredit = max_frame_size * (sendslope / port_transmit_rate)
53 #include <linux/module.h>
54 #include <linux/types.h>
55 #include <linux/kernel.h>
56 #include <linux/string.h>
57 #include <linux/errno.h>
58 #include <linux/skbuff.h>
59 #include <net/netevent.h>
60 #include <net/netlink.h>
61 #include <net/sch_generic.h>
62 #include <net/pkt_sched.h>
64 static LIST_HEAD(cbs_list);
65 static DEFINE_SPINLOCK(cbs_list_lock);
67 #define BYTES_PER_KBIT (1000LL / 8)
69 struct cbs_sched_data {
72 atomic64_t port_rate; /* in bytes/s */
73 s64 last; /* timestamp in ns */
74 s64 credits; /* in bytes */
75 s32 locredit; /* in bytes */
76 s32 hicredit; /* in bytes */
77 s64 sendslope; /* in bytes/s */
78 s64 idleslope; /* in bytes/s */
79 struct qdisc_watchdog watchdog;
80 int (*enqueue)(struct sk_buff *skb, struct Qdisc *sch,
81 struct sk_buff **to_free);
82 struct sk_buff *(*dequeue)(struct Qdisc *sch);
84 struct list_head cbs_list;
87 static int cbs_child_enqueue(struct sk_buff *skb, struct Qdisc *sch,
89 struct sk_buff **to_free)
91 unsigned int len = qdisc_pkt_len(skb);
94 err = child->ops->enqueue(skb, child, to_free);
95 if (err != NET_XMIT_SUCCESS)
98 sch->qstats.backlog += len;
101 return NET_XMIT_SUCCESS;
104 static int cbs_enqueue_offload(struct sk_buff *skb, struct Qdisc *sch,
105 struct sk_buff **to_free)
107 struct cbs_sched_data *q = qdisc_priv(sch);
108 struct Qdisc *qdisc = q->qdisc;
110 return cbs_child_enqueue(skb, sch, qdisc, to_free);
113 static int cbs_enqueue_soft(struct sk_buff *skb, struct Qdisc *sch,
114 struct sk_buff **to_free)
116 struct cbs_sched_data *q = qdisc_priv(sch);
117 struct Qdisc *qdisc = q->qdisc;
119 if (sch->q.qlen == 0 && q->credits > 0) {
120 /* We need to stop accumulating credits when there's
121 * no enqueued packets and q->credits is positive.
124 q->last = ktime_get_ns();
127 return cbs_child_enqueue(skb, sch, qdisc, to_free);
130 static int cbs_enqueue(struct sk_buff *skb, struct Qdisc *sch,
131 struct sk_buff **to_free)
133 struct cbs_sched_data *q = qdisc_priv(sch);
135 return q->enqueue(skb, sch, to_free);
138 /* timediff is in ns, slope is in bytes/s */
139 static s64 timediff_to_credits(s64 timediff, s64 slope)
141 return div64_s64(timediff * slope, NSEC_PER_SEC);
144 static s64 delay_from_credits(s64 credits, s64 slope)
146 if (unlikely(slope == 0))
149 return div64_s64(-credits * NSEC_PER_SEC, slope);
152 static s64 credits_from_len(unsigned int len, s64 slope, s64 port_rate)
154 if (unlikely(port_rate == 0))
157 return div64_s64(len * slope, port_rate);
160 static struct sk_buff *cbs_child_dequeue(struct Qdisc *sch, struct Qdisc *child)
164 skb = child->ops->dequeue(child);
168 qdisc_qstats_backlog_dec(sch, skb);
169 qdisc_bstats_update(sch, skb);
175 static struct sk_buff *cbs_dequeue_soft(struct Qdisc *sch)
177 struct cbs_sched_data *q = qdisc_priv(sch);
178 struct Qdisc *qdisc = q->qdisc;
179 s64 now = ktime_get_ns();
184 if (q->credits < 0) {
185 credits = timediff_to_credits(now - q->last, q->idleslope);
187 credits = q->credits + credits;
188 q->credits = min_t(s64, credits, q->hicredit);
190 if (q->credits < 0) {
193 delay = delay_from_credits(q->credits, q->idleslope);
194 qdisc_watchdog_schedule_ns(&q->watchdog, now + delay);
201 skb = cbs_child_dequeue(sch, qdisc);
205 len = qdisc_pkt_len(skb);
207 /* As sendslope is a negative number, this will decrease the
208 * amount of q->credits.
210 credits = credits_from_len(len, q->sendslope,
211 atomic64_read(&q->port_rate));
212 credits += q->credits;
214 q->credits = max_t(s64, credits, q->locredit);
220 static struct sk_buff *cbs_dequeue_offload(struct Qdisc *sch)
222 struct cbs_sched_data *q = qdisc_priv(sch);
223 struct Qdisc *qdisc = q->qdisc;
225 return cbs_child_dequeue(sch, qdisc);
228 static struct sk_buff *cbs_dequeue(struct Qdisc *sch)
230 struct cbs_sched_data *q = qdisc_priv(sch);
232 return q->dequeue(sch);
235 static const struct nla_policy cbs_policy[TCA_CBS_MAX + 1] = {
236 [TCA_CBS_PARMS] = { .len = sizeof(struct tc_cbs_qopt) },
239 static void cbs_disable_offload(struct net_device *dev,
240 struct cbs_sched_data *q)
242 struct tc_cbs_qopt_offload cbs = { };
243 const struct net_device_ops *ops;
249 q->enqueue = cbs_enqueue_soft;
250 q->dequeue = cbs_dequeue_soft;
252 ops = dev->netdev_ops;
253 if (!ops->ndo_setup_tc)
256 cbs.queue = q->queue;
259 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_CBS, &cbs);
261 pr_warn("Couldn't disable CBS offload for queue %d\n",
265 static int cbs_enable_offload(struct net_device *dev, struct cbs_sched_data *q,
266 const struct tc_cbs_qopt *opt,
267 struct netlink_ext_ack *extack)
269 const struct net_device_ops *ops = dev->netdev_ops;
270 struct tc_cbs_qopt_offload cbs = { };
273 if (!ops->ndo_setup_tc) {
274 NL_SET_ERR_MSG(extack, "Specified device does not support cbs offload");
278 cbs.queue = q->queue;
281 cbs.hicredit = opt->hicredit;
282 cbs.locredit = opt->locredit;
283 cbs.idleslope = opt->idleslope;
284 cbs.sendslope = opt->sendslope;
286 err = ops->ndo_setup_tc(dev, TC_SETUP_QDISC_CBS, &cbs);
288 NL_SET_ERR_MSG(extack, "Specified device failed to setup cbs hardware offload");
292 q->enqueue = cbs_enqueue_offload;
293 q->dequeue = cbs_dequeue_offload;
298 static void cbs_set_port_rate(struct net_device *dev, struct cbs_sched_data *q)
300 struct ethtool_link_ksettings ecmd;
301 int speed = SPEED_10;
305 err = __ethtool_get_link_ksettings(dev, &ecmd);
309 if (ecmd.base.speed != SPEED_UNKNOWN)
310 speed = ecmd.base.speed;
313 port_rate = speed * 1000 * BYTES_PER_KBIT;
315 atomic64_set(&q->port_rate, port_rate);
316 netdev_dbg(dev, "cbs: set %s's port_rate to: %lld, linkspeed: %d\n",
317 dev->name, (long long)atomic64_read(&q->port_rate),
321 static int cbs_dev_notifier(struct notifier_block *nb, unsigned long event,
324 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
325 struct cbs_sched_data *q;
326 struct net_device *qdev;
331 if (event != NETDEV_UP && event != NETDEV_CHANGE)
334 spin_lock(&cbs_list_lock);
335 list_for_each_entry(q, &cbs_list, cbs_list) {
336 qdev = qdisc_dev(q->qdisc);
342 spin_unlock(&cbs_list_lock);
345 cbs_set_port_rate(dev, q);
350 static int cbs_change(struct Qdisc *sch, struct nlattr *opt,
351 struct netlink_ext_ack *extack)
353 struct cbs_sched_data *q = qdisc_priv(sch);
354 struct net_device *dev = qdisc_dev(sch);
355 struct nlattr *tb[TCA_CBS_MAX + 1];
356 struct tc_cbs_qopt *qopt;
359 err = nla_parse_nested_deprecated(tb, TCA_CBS_MAX, opt, cbs_policy,
364 if (!tb[TCA_CBS_PARMS]) {
365 NL_SET_ERR_MSG(extack, "Missing CBS parameter which are mandatory");
369 qopt = nla_data(tb[TCA_CBS_PARMS]);
371 if (!qopt->offload) {
372 cbs_set_port_rate(dev, q);
373 cbs_disable_offload(dev, q);
375 err = cbs_enable_offload(dev, q, qopt, extack);
380 /* Everything went OK, save the parameters used. */
381 q->hicredit = qopt->hicredit;
382 q->locredit = qopt->locredit;
383 q->idleslope = qopt->idleslope * BYTES_PER_KBIT;
384 q->sendslope = qopt->sendslope * BYTES_PER_KBIT;
385 q->offload = qopt->offload;
390 static int cbs_init(struct Qdisc *sch, struct nlattr *opt,
391 struct netlink_ext_ack *extack)
393 struct cbs_sched_data *q = qdisc_priv(sch);
394 struct net_device *dev = qdisc_dev(sch);
398 NL_SET_ERR_MSG(extack, "Missing CBS qdisc options which are mandatory");
402 q->qdisc = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
403 sch->handle, extack);
407 qdisc_hash_add(q->qdisc, false);
409 q->queue = sch->dev_queue - netdev_get_tx_queue(dev, 0);
411 q->enqueue = cbs_enqueue_soft;
412 q->dequeue = cbs_dequeue_soft;
414 qdisc_watchdog_init(&q->watchdog, sch);
416 err = cbs_change(sch, opt, extack);
421 spin_lock(&cbs_list_lock);
422 list_add(&q->cbs_list, &cbs_list);
423 spin_unlock(&cbs_list_lock);
429 static void cbs_destroy(struct Qdisc *sch)
431 struct cbs_sched_data *q = qdisc_priv(sch);
432 struct net_device *dev = qdisc_dev(sch);
434 spin_lock(&cbs_list_lock);
435 list_del(&q->cbs_list);
436 spin_unlock(&cbs_list_lock);
438 qdisc_watchdog_cancel(&q->watchdog);
439 cbs_disable_offload(dev, q);
445 static int cbs_dump(struct Qdisc *sch, struct sk_buff *skb)
447 struct cbs_sched_data *q = qdisc_priv(sch);
448 struct tc_cbs_qopt opt = { };
451 nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
453 goto nla_put_failure;
455 opt.hicredit = q->hicredit;
456 opt.locredit = q->locredit;
457 opt.sendslope = div64_s64(q->sendslope, BYTES_PER_KBIT);
458 opt.idleslope = div64_s64(q->idleslope, BYTES_PER_KBIT);
459 opt.offload = q->offload;
461 if (nla_put(skb, TCA_CBS_PARMS, sizeof(opt), &opt))
462 goto nla_put_failure;
464 return nla_nest_end(skb, nest);
467 nla_nest_cancel(skb, nest);
471 static int cbs_dump_class(struct Qdisc *sch, unsigned long cl,
472 struct sk_buff *skb, struct tcmsg *tcm)
474 struct cbs_sched_data *q = qdisc_priv(sch);
476 if (cl != 1 || !q->qdisc) /* only one class */
479 tcm->tcm_handle |= TC_H_MIN(1);
480 tcm->tcm_info = q->qdisc->handle;
485 static int cbs_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
486 struct Qdisc **old, struct netlink_ext_ack *extack)
488 struct cbs_sched_data *q = qdisc_priv(sch);
491 new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
497 *old = qdisc_replace(sch, new, &q->qdisc);
501 static struct Qdisc *cbs_leaf(struct Qdisc *sch, unsigned long arg)
503 struct cbs_sched_data *q = qdisc_priv(sch);
508 static unsigned long cbs_find(struct Qdisc *sch, u32 classid)
513 static void cbs_walk(struct Qdisc *sch, struct qdisc_walker *walker)
516 if (walker->count >= walker->skip) {
517 if (walker->fn(sch, 1, walker) < 0) {
526 static const struct Qdisc_class_ops cbs_class_ops = {
531 .dump = cbs_dump_class,
534 static struct Qdisc_ops cbs_qdisc_ops __read_mostly = {
536 .cl_ops = &cbs_class_ops,
537 .priv_size = sizeof(struct cbs_sched_data),
538 .enqueue = cbs_enqueue,
539 .dequeue = cbs_dequeue,
540 .peek = qdisc_peek_dequeued,
542 .reset = qdisc_reset_queue,
543 .destroy = cbs_destroy,
544 .change = cbs_change,
546 .owner = THIS_MODULE,
549 static struct notifier_block cbs_device_notifier = {
550 .notifier_call = cbs_dev_notifier,
553 static int __init cbs_module_init(void)
557 err = register_netdevice_notifier(&cbs_device_notifier);
561 err = register_qdisc(&cbs_qdisc_ops);
563 unregister_netdevice_notifier(&cbs_device_notifier);
568 static void __exit cbs_module_exit(void)
570 unregister_qdisc(&cbs_qdisc_ops);
571 unregister_netdevice_notifier(&cbs_device_notifier);
573 module_init(cbs_module_init)
574 module_exit(cbs_module_exit)
575 MODULE_LICENSE("GPL");