1 #ifndef __NET_PKT_SCHED_H
2 #define __NET_PKT_SCHED_H
4 #define PSCHED_GETTIMEOFDAY 1
5 #define PSCHED_JIFFIES 2
8 #define PSCHED_CLOCK_SOURCE PSCHED_JIFFIES
10 #include <linux/config.h>
11 #include <linux/netdevice.h>
12 #include <linux/types.h>
13 #include <linux/pkt_sched.h>
14 #include <net/pkt_cls.h>
28 int (*fn)(struct Qdisc *, unsigned long cl, struct qdisc_walker *);
31 struct Qdisc_class_ops
33 /* Child qdisc manipulation */
34 int (*graft)(struct Qdisc *, unsigned long cl, struct Qdisc *, struct Qdisc **);
35 struct Qdisc * (*leaf)(struct Qdisc *, unsigned long cl);
37 /* Class manipulation routines */
38 unsigned long (*get)(struct Qdisc *, u32 classid);
39 void (*put)(struct Qdisc *, unsigned long);
40 int (*change)(struct Qdisc *, u32, u32, struct rtattr **, unsigned long *);
41 int (*delete)(struct Qdisc *, unsigned long);
42 void (*walk)(struct Qdisc *, struct qdisc_walker * arg);
44 /* Filter manipulation */
45 struct tcf_proto ** (*tcf_chain)(struct Qdisc *, unsigned long);
46 unsigned long (*bind_tcf)(struct Qdisc *, unsigned long, u32 classid);
47 void (*unbind_tcf)(struct Qdisc *, unsigned long);
49 /* rtnetlink specific */
50 int (*dump)(struct Qdisc *, unsigned long, struct sk_buff *skb, struct tcmsg*);
57 struct Qdisc_ops *next;
58 struct Qdisc_class_ops *cl_ops;
62 int (*enqueue)(struct sk_buff *, struct Qdisc *);
63 struct sk_buff * (*dequeue)(struct Qdisc *);
64 int (*requeue)(struct sk_buff *, struct Qdisc *);
65 unsigned int (*drop)(struct Qdisc *);
67 int (*init)(struct Qdisc *, struct rtattr *arg);
68 void (*reset)(struct Qdisc *);
69 void (*destroy)(struct Qdisc *);
70 int (*change)(struct Qdisc *, struct rtattr *arg);
72 int (*dump)(struct Qdisc *, struct sk_buff *);
77 extern rwlock_t qdisc_tree_lock;
81 int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
82 struct sk_buff * (*dequeue)(struct Qdisc *dev);
84 #define TCQ_F_BUILTIN 1
85 #define TCQ_F_THROTTLED 2
86 #define TCQ_F_INGRES 4
87 struct Qdisc_ops *ops;
91 struct sk_buff_head q;
92 struct net_device *dev;
94 struct tc_stats stats;
95 int (*reshape_fail)(struct sk_buff *skb, struct Qdisc *q);
97 /* This field is deprecated, but it is still used by CBQ
98 * and it will live until better solution will be invented.
100 struct Qdisc *__parent;
105 struct qdisc_rate_table
107 struct tc_ratespec rate;
109 struct qdisc_rate_table *next;
113 static inline void sch_tree_lock(struct Qdisc *q)
115 write_lock(&qdisc_tree_lock);
116 spin_lock_bh(&q->dev->queue_lock);
119 static inline void sch_tree_unlock(struct Qdisc *q)
121 spin_unlock_bh(&q->dev->queue_lock);
122 write_unlock(&qdisc_tree_lock);
125 static inline void tcf_tree_lock(struct tcf_proto *tp)
127 write_lock(&qdisc_tree_lock);
128 spin_lock_bh(&tp->q->dev->queue_lock);
131 static inline void tcf_tree_unlock(struct tcf_proto *tp)
133 spin_unlock_bh(&tp->q->dev->queue_lock);
134 write_unlock(&qdisc_tree_lock);
138 static inline unsigned long
139 cls_set_class(struct tcf_proto *tp, unsigned long *clp, unsigned long cl)
141 unsigned long old_cl;
150 static inline unsigned long
151 __cls_set_class(unsigned long *clp, unsigned long cl)
153 unsigned long old_cl;
162 Timer resolution MUST BE < 10% of min_schedulable_packet_size/bandwidth
164 Normal IP packet size ~ 512byte, hence:
166 0.5Kbyte/1Mbyte/sec = 0.5msec, so that we need 50usec timer for
169 10msec resolution -> <50Kbit/sec.
171 The result: [34]86 is not good choice for QoS router :-(
173 The things are not so bad, because we may use artifical
174 clock evaluated by integration of network data flow
175 in the most critical places.
177 Note: we do not use fastgettimeofday.
178 The reason is that, when it is not the same thing as
179 gettimeofday, it returns invalid timestamp, which is
180 not updated, when net_bh is active.
182 So, use PSCHED_CLOCK_SOURCE = PSCHED_CPU on alpha and pentiums
183 with rtdsc. And PSCHED_JIFFIES on all other architectures, including [34]86
184 and pentiums without rtdsc.
185 You can use PSCHED_GETTIMEOFDAY on another architectures,
186 which have fast and precise clock source, but it is too expensive.
189 /* General note about internal clock.
191 Any clock source returns time intervals, measured in units
192 close to 1usec. With source PSCHED_GETTIMEOFDAY it is precisely
193 microseconds, otherwise something close but different chosen to minimize
194 arithmetic cost. Ratio usec/internal untis in form nominator/denominator
195 may be read from /proc/net/psched.
199 #if PSCHED_CLOCK_SOURCE == PSCHED_GETTIMEOFDAY
201 typedef struct timeval psched_time_t;
202 typedef long psched_tdiff_t;
204 #define PSCHED_GET_TIME(stamp) do_gettimeofday(&(stamp))
205 #define PSCHED_US2JIFFIE(usecs) (((usecs)+(1000000/HZ-1))/(1000000/HZ))
206 #define PSCHED_JIFFIE2US(delay) ((delay)*(1000000/HZ))
208 #define PSCHED_EXPORTLIST EXPORT_SYMBOL(psched_tod_diff);
210 #else /* PSCHED_CLOCK_SOURCE != PSCHED_GETTIMEOFDAY */
212 #define PSCHED_EXPORTLIST PSCHED_EXPORTLIST_1 PSCHED_EXPORTLIST_2
214 typedef u64 psched_time_t;
215 typedef long psched_tdiff_t;
217 extern psched_time_t psched_time_base;
219 #if PSCHED_CLOCK_SOURCE == PSCHED_JIFFIES
222 #define PSCHED_JSCALE 14
223 #elif HZ >= 96 && HZ < 192
224 #define PSCHED_JSCALE 13
225 #elif HZ >= 192 && HZ < 384
226 #define PSCHED_JSCALE 12
227 #elif HZ >= 384 && HZ < 768
228 #define PSCHED_JSCALE 11
230 #define PSCHED_JSCALE 10
233 #define PSCHED_EXPORTLIST_2
235 #if BITS_PER_LONG <= 32
237 #define PSCHED_WATCHER unsigned long
239 extern PSCHED_WATCHER psched_time_mark;
241 #define PSCHED_GET_TIME(stamp) ((stamp) = psched_time_base + (((unsigned long)(jiffies-psched_time_mark))<<PSCHED_JSCALE))
243 #define PSCHED_EXPORTLIST_1 EXPORT_SYMBOL(psched_time_base); \
244 EXPORT_SYMBOL(psched_time_mark);
248 #define PSCHED_GET_TIME(stamp) ((stamp) = (jiffies<<PSCHED_JSCALE))
250 #define PSCHED_EXPORTLIST_1
254 #define PSCHED_US2JIFFIE(delay) (((delay)+(1<<PSCHED_JSCALE)-1)>>PSCHED_JSCALE)
255 #define PSCHED_JIFFIE2US(delay) ((delay)<<PSCHED_JSCALE)
257 #elif PSCHED_CLOCK_SOURCE == PSCHED_CPU
259 extern psched_tdiff_t psched_clock_per_hz;
260 extern int psched_clock_scale;
262 #define PSCHED_EXPORTLIST_2 EXPORT_SYMBOL(psched_clock_per_hz); \
263 EXPORT_SYMBOL(psched_clock_scale);
265 #define PSCHED_US2JIFFIE(delay) (((delay)+psched_clock_per_hz-1)/psched_clock_per_hz)
266 #define PSCHED_JIFFIE2US(delay) ((delay)*psched_clock_per_hz)
268 #ifdef CONFIG_X86_TSC
270 #define PSCHED_GET_TIME(stamp) \
273 (stamp) = __cur>>psched_clock_scale; \
276 #define PSCHED_EXPORTLIST_1
278 #elif defined (__alpha__)
280 #define PSCHED_WATCHER u32
282 extern PSCHED_WATCHER psched_time_mark;
284 #define PSCHED_GET_TIME(stamp) \
286 __asm__ __volatile__ ("rpcc %0" : "r="(__res)); \
287 if (__res <= psched_time_mark) psched_time_base += 0x100000000UL; \
288 psched_time_mark = __res; \
289 (stamp) = (psched_time_base + __res)>>psched_clock_scale; \
292 #define PSCHED_EXPORTLIST_1 EXPORT_SYMBOL(psched_time_base); \
293 EXPORT_SYMBOL(psched_time_mark);
297 #error PSCHED_CLOCK_SOURCE=PSCHED_CPU is not supported on this arch.
301 #endif /* PSCHED_CLOCK_SOURCE == PSCHED_JIFFIES */
303 #endif /* PSCHED_CLOCK_SOURCE == PSCHED_GETTIMEOFDAY */
305 #if PSCHED_CLOCK_SOURCE == PSCHED_GETTIMEOFDAY
306 #define PSCHED_TDIFF(tv1, tv2) \
308 int __delta_sec = (tv1).tv_sec - (tv2).tv_sec; \
309 int __delta = (tv1).tv_usec - (tv2).tv_usec; \
311 switch (__delta_sec) { \
315 __delta += 1000000; \
317 __delta += 1000000; \
323 extern int psched_tod_diff(int delta_sec, int bound);
325 #define PSCHED_TDIFF_SAFE(tv1, tv2, bound, guard) \
327 int __delta_sec = (tv1).tv_sec - (tv2).tv_sec; \
328 int __delta = (tv1).tv_usec - (tv2).tv_usec; \
329 switch (__delta_sec) { \
331 __delta = psched_tod_diff(__delta_sec, bound); guard; break; \
333 __delta += 1000000; \
335 __delta += 1000000; \
341 #define PSCHED_TLESS(tv1, tv2) (((tv1).tv_usec < (tv2).tv_usec && \
342 (tv1).tv_sec <= (tv2).tv_sec) || \
343 (tv1).tv_sec < (tv2).tv_sec)
345 #define PSCHED_TADD2(tv, delta, tv_res) \
347 int __delta = (tv).tv_usec + (delta); \
348 (tv_res).tv_sec = (tv).tv_sec; \
349 if (__delta > 1000000) { (tv_res).tv_sec++; __delta -= 1000000; } \
350 (tv_res).tv_usec = __delta; \
353 #define PSCHED_TADD(tv, delta) \
355 (tv).tv_usec += (delta); \
356 if ((tv).tv_usec > 1000000) { (tv).tv_sec++; \
357 (tv).tv_usec -= 1000000; } \
360 /* Set/check that time is in the "past perfect";
361 it depends on concrete representation of system time
364 #define PSCHED_SET_PASTPERFECT(t) ((t).tv_sec = 0)
365 #define PSCHED_IS_PASTPERFECT(t) ((t).tv_sec == 0)
367 #define PSCHED_AUDIT_TDIFF(t) ({ if ((t) > 2000000) (t) = 2000000; })
371 #define PSCHED_TDIFF(tv1, tv2) (long)((tv1) - (tv2))
372 #define PSCHED_TDIFF_SAFE(tv1, tv2, bound, guard) \
374 long long __delta = (tv1) - (tv2); \
375 if ( __delta > (long long)(bound)) { __delta = (bound); guard; } \
380 #define PSCHED_TLESS(tv1, tv2) ((tv1) < (tv2))
381 #define PSCHED_TADD2(tv, delta, tv_res) ((tv_res) = (tv) + (delta))
382 #define PSCHED_TADD(tv, delta) ((tv) += (delta))
383 #define PSCHED_SET_PASTPERFECT(t) ((t) = 0)
384 #define PSCHED_IS_PASTPERFECT(t) ((t) == 0)
385 #define PSCHED_AUDIT_TDIFF(t)
391 struct tcf_police *next;
405 struct qdisc_rate_table *R_tab;
406 struct qdisc_rate_table *P_tab;
408 struct tc_stats stats;
411 extern int qdisc_copy_stats(struct sk_buff *skb, struct tc_stats *st);
412 extern void tcf_police_destroy(struct tcf_police *p);
413 extern struct tcf_police * tcf_police_locate(struct rtattr *rta, struct rtattr *est);
414 extern int tcf_police_dump(struct sk_buff *skb, struct tcf_police *p);
415 extern int tcf_police(struct sk_buff *skb, struct tcf_police *p);
417 static inline void tcf_police_release(struct tcf_police *p)
419 if (p && --p->refcnt == 0)
420 tcf_police_destroy(p);
423 extern struct Qdisc noop_qdisc;
424 extern struct Qdisc_ops noop_qdisc_ops;
425 extern struct Qdisc_ops pfifo_qdisc_ops;
426 extern struct Qdisc_ops bfifo_qdisc_ops;
428 int register_qdisc(struct Qdisc_ops *qops);
429 int unregister_qdisc(struct Qdisc_ops *qops);
430 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle);
431 struct Qdisc *qdisc_lookup_class(struct net_device *dev, u32 handle);
432 void dev_init_scheduler(struct net_device *dev);
433 void dev_shutdown(struct net_device *dev);
434 void dev_activate(struct net_device *dev);
435 void dev_deactivate(struct net_device *dev);
436 void qdisc_reset(struct Qdisc *qdisc);
437 void qdisc_destroy(struct Qdisc *qdisc);
438 struct Qdisc * qdisc_create_dflt(struct net_device *dev, struct Qdisc_ops *ops);
439 int qdisc_new_estimator(struct tc_stats *stats, struct rtattr *opt);
440 void qdisc_kill_estimator(struct tc_stats *stats);
441 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct rtattr *tab);
442 void qdisc_put_rtab(struct qdisc_rate_table *tab);
444 extern int qdisc_restart(struct net_device *dev);
446 static inline void qdisc_run(struct net_device *dev)
448 while (!netif_queue_stopped(dev) &&
449 qdisc_restart(dev)<0)
453 /* Calculate maximal size of packet seen by hard_start_xmit
454 routine of this device.
456 static inline unsigned psched_mtu(struct net_device *dev)
458 unsigned mtu = dev->mtu;
459 return dev->hard_header ? mtu + dev->hard_header_len : mtu;