xref: /linux/include/net/pkt_sched.h (revision bc9ff192a6c940d9a26e21a0a82f2667067aaf5f)
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 
16 struct qdisc_walker {
17 	int	stop;
18 	int	skip;
19 	int	count;
20 	int	(*fn)(struct Qdisc *, unsigned long cl, struct qdisc_walker *);
21 };
22 
23 #define qdisc_priv(q)							\
24 	_Generic(q,							\
25 		 const struct Qdisc * : (const void *)&q->privdata,	\
26 		 struct Qdisc * : (void *)&q->privdata)
27 
qdisc_from_priv(void * priv)28 static inline struct Qdisc *qdisc_from_priv(void *priv)
29 {
30 	return container_of(priv, struct Qdisc, privdata);
31 }
32 
33 /*
34    Timer resolution MUST BE < 10% of min_schedulable_packet_size/bandwidth
35 
36    Normal IP packet size ~ 512byte, hence:
37 
38    0.5Kbyte/1Mbyte/sec = 0.5msec, so that we need 50usec timer for
39    10Mbit ethernet.
40 
41    10msec resolution -> <50Kbit/sec.
42 
43    The result: [34]86 is not good choice for QoS router :-(
44 
45    The things are not so bad, because we may use artificial
46    clock evaluated by integration of network data flow
47    in the most critical places.
48  */
49 
50 typedef u64	psched_time_t;
51 typedef long	psched_tdiff_t;
52 
53 /* Avoid doing 64 bit divide */
54 #define PSCHED_SHIFT			6
55 #define PSCHED_TICKS2NS(x)		((s64)(x) << PSCHED_SHIFT)
56 #define PSCHED_NS2TICKS(x)		((x) >> PSCHED_SHIFT)
57 
58 #define PSCHED_TICKS_PER_SEC		PSCHED_NS2TICKS(NSEC_PER_SEC)
59 #define PSCHED_PASTPERFECT		0
60 
psched_get_time(void)61 static inline psched_time_t psched_get_time(void)
62 {
63 	return PSCHED_NS2TICKS(ktime_get_ns());
64 }
65 
66 struct qdisc_watchdog {
67 	struct hrtimer	timer;
68 	struct Qdisc	*qdisc;
69 };
70 
71 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
72 				 clockid_t clockid);
73 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc);
74 
75 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
76 				      u64 delta_ns);
77 
qdisc_watchdog_schedule_ns(struct qdisc_watchdog * wd,u64 expires)78 static inline void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd,
79 					      u64 expires)
80 {
81 	return qdisc_watchdog_schedule_range_ns(wd, expires, 0ULL);
82 }
83 
qdisc_watchdog_schedule(struct qdisc_watchdog * wd,psched_time_t expires)84 static inline void qdisc_watchdog_schedule(struct qdisc_watchdog *wd,
85 					   psched_time_t expires)
86 {
87 	qdisc_watchdog_schedule_ns(wd, PSCHED_TICKS2NS(expires));
88 }
89 
90 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd);
91 
92 extern struct Qdisc_ops pfifo_qdisc_ops;
93 extern struct Qdisc_ops bfifo_qdisc_ops;
94 extern struct Qdisc_ops pfifo_head_drop_qdisc_ops;
95 
96 int fifo_set_limit(struct Qdisc *q, unsigned int limit);
97 struct Qdisc *fifo_create_dflt(struct Qdisc *sch, struct Qdisc_ops *ops,
98 			       unsigned int limit,
99 			       struct netlink_ext_ack *extack);
100 
101 int register_qdisc(struct Qdisc_ops *qops);
102 void unregister_qdisc(struct Qdisc_ops *qops);
103 #define NET_SCH_ALIAS_PREFIX "net-sch-"
104 #define MODULE_ALIAS_NET_SCH(id)	MODULE_ALIAS(NET_SCH_ALIAS_PREFIX id)
105 void qdisc_get_default(char *id, size_t len);
106 int qdisc_set_default(const char *id);
107 
108 void qdisc_hash_add(struct Qdisc *q, bool invisible);
109 void qdisc_hash_del(struct Qdisc *q);
110 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
111 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle);
112 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
113 					struct nlattr *tab,
114 					struct netlink_ext_ack *extack);
115 void qdisc_put_rtab(struct qdisc_rate_table *tab);
116 void qdisc_put_stab(struct qdisc_size_table *tab);
117 bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
118 		     struct net_device *dev, struct netdev_queue *txq,
119 		     spinlock_t *root_lock, bool validate);
120 
121 void __qdisc_run(struct Qdisc *q);
122 
qdisc_run(struct Qdisc * q)123 static inline void qdisc_run(struct Qdisc *q)
124 {
125 	if (qdisc_run_begin(q)) {
126 		__qdisc_run(q);
127 		qdisc_run_end(q);
128 	}
129 }
130 
131 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
132 
133 /* Calculate maximal size of packet seen by hard_start_xmit
134    routine of this device.
135  */
psched_mtu(const struct net_device * dev)136 static inline unsigned int psched_mtu(const struct net_device *dev)
137 {
138 	return READ_ONCE(dev->mtu) + dev->hard_header_len;
139 }
140 
qdisc_net(struct Qdisc * q)141 static inline struct net *qdisc_net(struct Qdisc *q)
142 {
143 	return dev_net(q->dev_queue->dev);
144 }
145 
146 struct tc_query_caps_base {
147 	enum tc_setup_type type;
148 	void *caps;
149 };
150 
151 struct tc_cbs_qopt_offload {
152 	u8 enable;
153 	s32 queue;
154 	s32 hicredit;
155 	s32 locredit;
156 	s32 idleslope;
157 	s32 sendslope;
158 };
159 
160 struct tc_etf_qopt_offload {
161 	u8 enable;
162 	s32 queue;
163 };
164 
165 struct tc_mqprio_caps {
166 	bool validate_queue_counts:1;
167 };
168 
169 struct tc_mqprio_qopt_offload {
170 	/* struct tc_mqprio_qopt must always be the first element */
171 	struct tc_mqprio_qopt qopt;
172 	struct netlink_ext_ack *extack;
173 	u16 mode;
174 	u16 shaper;
175 	u32 flags;
176 	u64 min_rate[TC_QOPT_MAX_QUEUE];
177 	u64 max_rate[TC_QOPT_MAX_QUEUE];
178 	unsigned long preemptible_tcs;
179 };
180 
181 struct tc_taprio_caps {
182 	bool supports_queue_max_sdu:1;
183 	bool gate_mask_per_txq:1;
184 	/* Device expects lower TXQ numbers to have higher priority over higher
185 	 * TXQs, regardless of their TC mapping. DO NOT USE FOR NEW DRIVERS,
186 	 * INSTEAD ENFORCE A PROPER TC:TXQ MAPPING COMING FROM USER SPACE.
187 	 */
188 	bool broken_mqprio:1;
189 };
190 
191 enum tc_taprio_qopt_cmd {
192 	TAPRIO_CMD_REPLACE,
193 	TAPRIO_CMD_DESTROY,
194 	TAPRIO_CMD_STATS,
195 	TAPRIO_CMD_QUEUE_STATS,
196 };
197 
198 /**
199  * struct tc_taprio_qopt_stats - IEEE 802.1Qbv statistics
200  * @window_drops: Frames that were dropped because they were too large to be
201  *	transmitted in any of the allotted time windows (open gates) for their
202  *	traffic class.
203  * @tx_overruns: Frames still being transmitted by the MAC after the
204  *	transmission gate associated with their traffic class has closed.
205  *	Equivalent to `12.29.1.1.2 TransmissionOverrun` from 802.1Q-2018.
206  */
207 struct tc_taprio_qopt_stats {
208 	u64 window_drops;
209 	u64 tx_overruns;
210 };
211 
212 struct tc_taprio_qopt_queue_stats {
213 	int queue;
214 	struct tc_taprio_qopt_stats stats;
215 };
216 
217 struct tc_taprio_sched_entry {
218 	u8 command; /* TC_TAPRIO_CMD_* */
219 
220 	/* The gate_mask in the offloading side refers to traffic classes */
221 	u32 gate_mask;
222 	u32 interval;
223 };
224 
225 struct tc_taprio_qopt_offload {
226 	enum tc_taprio_qopt_cmd cmd;
227 
228 	union {
229 		/* TAPRIO_CMD_STATS */
230 		struct tc_taprio_qopt_stats stats;
231 		/* TAPRIO_CMD_QUEUE_STATS */
232 		struct tc_taprio_qopt_queue_stats queue_stats;
233 		/* TAPRIO_CMD_REPLACE */
234 		struct {
235 			struct tc_mqprio_qopt_offload mqprio;
236 			struct netlink_ext_ack *extack;
237 			ktime_t base_time;
238 			u64 cycle_time;
239 			u64 cycle_time_extension;
240 			u32 max_sdu[TC_MAX_QUEUE];
241 
242 			size_t num_entries;
243 			struct tc_taprio_sched_entry entries[];
244 		};
245 	};
246 };
247 
248 #if IS_ENABLED(CONFIG_NET_SCH_TAPRIO)
249 
250 /* Reference counting */
251 struct tc_taprio_qopt_offload *taprio_offload_get(struct tc_taprio_qopt_offload
252 						  *offload);
253 void taprio_offload_free(struct tc_taprio_qopt_offload *offload);
254 
255 #else
256 
257 /* Reference counting */
258 static inline struct tc_taprio_qopt_offload *
taprio_offload_get(struct tc_taprio_qopt_offload * offload)259 taprio_offload_get(struct tc_taprio_qopt_offload *offload)
260 {
261 	return NULL;
262 }
263 
taprio_offload_free(struct tc_taprio_qopt_offload * offload)264 static inline void taprio_offload_free(struct tc_taprio_qopt_offload *offload)
265 {
266 }
267 
268 #endif
269 
270 /* Ensure skb_mstamp_ns, which might have been populated with the txtime, is
271  * not mistaken for a software timestamp, because this will otherwise prevent
272  * the dispatch of hardware timestamps to the socket.
273  */
skb_txtime_consumed(struct sk_buff * skb)274 static inline void skb_txtime_consumed(struct sk_buff *skb)
275 {
276 	skb->tstamp = ktime_set(0, 0);
277 }
278 
tc_qdisc_stats_dump(struct Qdisc * sch,unsigned long cl,struct qdisc_walker * arg)279 static inline bool tc_qdisc_stats_dump(struct Qdisc *sch,
280 				       unsigned long cl,
281 				       struct qdisc_walker *arg)
282 {
283 	if (arg->count >= arg->skip && arg->fn(sch, cl, arg) < 0) {
284 		arg->stop = 1;
285 		return false;
286 	}
287 
288 	arg->count++;
289 	return true;
290 }
291 
qdisc_warn_nonwc(const char * txt,struct Qdisc * qdisc)292 static inline void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
293 {
294 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
295 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
296 			txt, qdisc->ops->id, qdisc->handle >> 16);
297 		qdisc->flags |= TCQ_F_WARN_NONWC;
298 	}
299 }
300 
qdisc_peek_len(struct Qdisc * sch)301 static inline unsigned int qdisc_peek_len(struct Qdisc *sch)
302 {
303 	struct sk_buff *skb;
304 	unsigned int len;
305 
306 	skb = sch->ops->peek(sch);
307 	if (unlikely(skb == NULL)) {
308 		qdisc_warn_nonwc("qdisc_peek_len", sch);
309 		return 0;
310 	}
311 	len = qdisc_pkt_len(skb);
312 
313 	return len;
314 }
315 
316 #endif
317