projects/linux·blob master · refs

336 lines · 8.4 KiB

1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __NET_PKT_SCHED_H
3#define __NET_PKT_SCHED_H
4
5#include <linux/jiffies.h>
6#include <linux/ktime.h>
7#include <linux/if_vlan.h>
8#include <linux/netdevice.h>
9#include <net/sch_generic.h>
10#include <net/net_namespace.h>
11#include <uapi/linux/pkt_sched.h>
12
13#define DEFAULT_TX_QUEUE_LEN 1000
14#define STAB_SIZE_LOG_MAX 30
15#define QDISC_PKT_LEN_MAX (1 << 20) /* 1 MiB */
16
17struct qdisc_walker {
18 int stop;
19 int skip;
20 int count;
21 int (*fn)(struct Qdisc *, unsigned long cl, struct qdisc_walker *);
22};
23
24#define qdisc_priv(q) \
25 _Generic(q, \
26 const struct Qdisc * : (const void *)&q->privdata, \
27 struct Qdisc * : (void *)&q->privdata)
28
29/*
30 Timer resolution MUST BE < 10% of min_schedulable_packet_size/bandwidth
31
32 Normal IP packet size ~ 512byte, hence:
33
34 0.5Kbyte/1Mbyte/sec = 0.5msec, so that we need 50usec timer for
35 10Mbit ethernet.
36
37 10msec resolution -> <50Kbit/sec.
38
39 The result: [34]86 is not good choice for QoS router :-(
40
41 The things are not so bad, because we may use artificial
42 clock evaluated by integration of network data flow
43 in the most critical places.
44 */
45
46typedef u64 psched_time_t;
47
48/* Avoid doing 64 bit divide */
49#define PSCHED_SHIFT 6
50#define PSCHED_TICKS2NS(x) ((s64)(x) << PSCHED_SHIFT)
51#define PSCHED_NS2TICKS(x) ((x) >> PSCHED_SHIFT)
52
53#define PSCHED_TICKS_PER_SEC PSCHED_NS2TICKS(NSEC_PER_SEC)
54#define PSCHED_PASTPERFECT 0
55
56static inline psched_time_t psched_get_time(void)
57{
58 return PSCHED_NS2TICKS(ktime_get_ns());
59}
60
61struct qdisc_watchdog {
62 struct hrtimer timer;
63 struct Qdisc *qdisc;
64};
65
66void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
67 clockid_t clockid);
68void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);
69
70void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
71 u64 delta_ns);
72
73static inline void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd,
74 u64 expires)
75{
76 return qdisc_watchdog_schedule_range_ns(wd, expires, 0ULL);
77}
78
79static inline void qdisc_watchdog_schedule(struct qdisc_watchdog *wd,
80 psched_time_t expires)
81{
82 qdisc_watchdog_schedule_ns(wd, PSCHED_TICKS2NS(expires));
83}
84
85void qdisc_watchdog_cancel(struct qdisc_watchdog *wd);
86
87extern struct Qdisc_ops pfifo_qdisc_ops;
88extern struct Qdisc_ops bfifo_qdisc_ops;
89extern struct Qdisc_ops pfifo_head_drop_qdisc_ops;
90
91int fifo_set_limit(struct Qdisc *q, unsigned int limit);
92struct Qdisc *fifo_create_dflt(struct Qdisc *sch, struct Qdisc_ops *ops,
93 unsigned int limit,
94 struct netlink_ext_ack *extack);
95
96int register_qdisc(struct Qdisc_ops *qops);
97void unregister_qdisc(struct Qdisc_ops *qops);
98#define NET_SCH_ALIAS_PREFIX "net-sch-"
99#define MODULE_ALIAS_NET_SCH(id) MODULE_ALIAS(NET_SCH_ALIAS_PREFIX id)
100void qdisc_get_default(char *id, size_t len);
101int qdisc_set_default(const char *id);
102
103void qdisc_hash_add(struct Qdisc *q, bool invisible);
104void qdisc_hash_del(struct Qdisc *q);
105struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
106struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle);
107struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
108 struct nlattr *tab,
109 struct netlink_ext_ack *extack);
110void qdisc_put_rtab(struct qdisc_rate_table *tab);
111void qdisc_put_stab(struct qdisc_size_table *tab);
112bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
113 struct net_device *dev, struct netdev_queue *txq,
114 spinlock_t *root_lock, bool validate);
115
116void __qdisc_run(struct Qdisc *q);
117
118static inline struct sk_buff *qdisc_run(struct Qdisc *q)
119{
120 if (qdisc_run_begin(q)) {
121 __qdisc_run(q);
122 return qdisc_run_end(q);
123 }
124 return NULL;
125}
126
127extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
128
129/* Calculate maximal size of packet seen by hard_start_xmit
130 routine of this device.
131 */
132static inline unsigned int psched_mtu(const struct net_device *dev)
133{
134 return READ_ONCE(dev->mtu) + dev->hard_header_len;
135}
136
137static inline struct net *qdisc_net(struct Qdisc *q)
138{
139 return dev_net(q->dev_queue->dev);
140}
141
142struct tc_query_caps_base {
143 enum tc_setup_type type;
144 void *caps;
145};
146
147struct tc_cbs_qopt_offload {
148 u8 enable;
149 s32 queue;
150 s32 hicredit;
151 s32 locredit;
152 s32 idleslope;
153 s32 sendslope;
154};
155
156struct tc_etf_qopt_offload {
157 u8 enable;
158 s32 queue;
159};
160
161struct tc_mqprio_caps {
162 bool validate_queue_counts:1;
163};
164
165struct tc_mqprio_qopt_offload {
166 /* struct tc_mqprio_qopt must always be the first element */
167 struct tc_mqprio_qopt qopt;
168 struct netlink_ext_ack *extack;
169 u16 mode;
170 u16 shaper;
171 u32 flags;
172 u64 min_rate[TC_QOPT_MAX_QUEUE];
173 u64 max_rate[TC_QOPT_MAX_QUEUE];
174 unsigned long preemptible_tcs;
175};
176
177struct tc_taprio_caps {
178 bool supports_queue_max_sdu:1;
179 bool gate_mask_per_txq:1;
180 /* Device expects lower TXQ numbers to have higher priority over higher
181 * TXQs, regardless of their TC mapping. DO NOT USE FOR NEW DRIVERS,
182 * INSTEAD ENFORCE A PROPER TC:TXQ MAPPING COMING FROM USER SPACE.
183 */
184 bool broken_mqprio:1;
185};
186
187enum tc_taprio_qopt_cmd {
188 TAPRIO_CMD_REPLACE,
189 TAPRIO_CMD_DESTROY,
190 TAPRIO_CMD_STATS,
191 TAPRIO_CMD_QUEUE_STATS,
192};
193
194/**
195 * struct tc_taprio_qopt_stats - IEEE 802.1Qbv statistics
196 * @window_drops: Frames that were dropped because they were too large to be
197 * transmitted in any of the allotted time windows (open gates) for their
198 * traffic class.
199 * @tx_overruns: Frames still being transmitted by the MAC after the
200 * transmission gate associated with their traffic class has closed.
201 * Equivalent to `12.29.1.1.2 TransmissionOverrun` from 802.1Q-2018.
202 */
203struct tc_taprio_qopt_stats {
204 u64 window_drops;
205 u64 tx_overruns;
206};
207
208struct tc_taprio_qopt_queue_stats {
209 int queue;
210 struct tc_taprio_qopt_stats stats;
211};
212
213struct tc_taprio_sched_entry {
214 u8 command; /* TC_TAPRIO_CMD_* */
215
216 /* The gate_mask in the offloading side refers to traffic classes */
217 u32 gate_mask;
218 u32 interval;
219};
220
221struct tc_taprio_qopt_offload {
222 enum tc_taprio_qopt_cmd cmd;
223
224 union {
225 /* TAPRIO_CMD_STATS */
226 struct tc_taprio_qopt_stats stats;
227 /* TAPRIO_CMD_QUEUE_STATS */
228 struct tc_taprio_qopt_queue_stats queue_stats;
229 /* TAPRIO_CMD_REPLACE */
230 struct {
231 struct tc_mqprio_qopt_offload mqprio;
232 struct netlink_ext_ack *extack;
233 ktime_t base_time;
234 u64 cycle_time;
235 u64 cycle_time_extension;
236 u32 max_sdu[TC_MAX_QUEUE];
237
238 size_t num_entries;
239 struct tc_taprio_sched_entry entries[];
240 };
241 };
242};
243
244#if IS_ENABLED(CONFIG_NET_SCH_TAPRIO)
245
246/* Reference counting */
247struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
248 *offload);
249void taprio_offload_free(struct tc_taprio_qopt_offload *offload);
250
251#else
252
253/* Reference counting */
254static inline struct tc_taprio_qopt_offload *
255taprio_offload_get(struct tc_taprio_qopt_offload *offload)
256{
257 return NULL;
258}
259
260static inline void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
261{
262}
263
264#endif
265
266/* Ensure skb_mstamp_ns, which might have been populated with the txtime, is
267 * not mistaken for a software timestamp, because this will otherwise prevent
268 * the dispatch of hardware timestamps to the socket.
269 */
270static inline void skb_txtime_consumed(struct sk_buff *skb)
271{
272 skb->tstamp = ktime_set(0, 0);
273}
274
275static inline bool tc_qdisc_stats_dump(struct Qdisc *sch,
276 unsigned long cl,
277 struct qdisc_walker *arg)
278{
279 if (arg->count >= arg->skip && arg->fn(sch, cl, arg) < 0) {
280 arg->stop = 1;
281 return false;
282 }
283
284 arg->count++;
285 return true;
286}
287
288static inline void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
289{
290 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
291 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
292 txt, qdisc->ops->id, qdisc->handle >> 16);
293 qdisc->flags |= TCQ_F_WARN_NONWC;
294 }
295}
296
297static inline unsigned int qdisc_peek_len(struct Qdisc *sch)
298{
299 struct sk_buff *skb;
300 unsigned int len;
301
302 skb = sch->ops->peek(sch);
303 if (unlikely(skb == NULL)) {
304 qdisc_warn_nonwc("qdisc_peek_len", sch);
305 return 0;
306 }
307 len = qdisc_pkt_len(skb);
308
309 return len;
310}
311
312static inline void qdisc_lock_init(struct Qdisc *sch,
313 const struct Qdisc_ops *ops)
314{
315 spin_lock_init(&sch->q.lock);
316
317 /* Skip dynamic keys if nesting is not possible */
318 if (ops->static_flags & TCQ_F_INGRESS ||
319 ops == &noqueue_qdisc_ops)
320 return;
321
322 lockdep_register_key(&sch->root_lock_key);
323 lockdep_set_class(&sch->q.lock, &sch->root_lock_key);
324}
325
326static inline void qdisc_lock_uninit(struct Qdisc *sch,
327 const struct Qdisc_ops *ops)
328{
329 if (ops->static_flags & TCQ_F_INGRESS ||
330 ops == &noqueue_qdisc_ops)
331 return;
332
333 lockdep_unregister_key(&sch->root_lock_key);
334}
335
336#endif