netprio_cgroup.h: dont include module.h from other includes
[deliverable/linux.git] / include / net / red.h
CommitLineData
a7834745
TG
1#ifndef __NET_SCHED_RED_H
2#define __NET_SCHED_RED_H
3
a7834745
TG
4#include <linux/types.h>
5#include <net/pkt_sched.h>
6#include <net/inet_ecn.h>
7#include <net/dsfield.h>
8af2a218 8#include <linux/reciprocal_div.h>
a7834745
TG
9
10/* Random Early Detection (RED) algorithm.
11 =======================================
12
13 Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways
14 for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
15
16 This file codes a "divisionless" version of RED algorithm
17 as written down in Fig.17 of the paper.
18
19 Short description.
20 ------------------
21
22 When a new packet arrives we calculate the average queue length:
23
24 avg = (1-W)*avg + W*current_queue_len,
25
26 W is the filter time constant (chosen as 2^(-Wlog)), it controls
27 the inertia of the algorithm. To allow larger bursts, W should be
28 decreased.
29
30 if (avg > th_max) -> packet marked (dropped).
31 if (avg < th_min) -> packet passes.
32 if (th_min < avg < th_max) we calculate probability:
33
34 Pb = max_P * (avg - th_min)/(th_max-th_min)
35
36 and mark (drop) packet with this probability.
37 Pb changes from 0 (at avg==th_min) to max_P (avg==th_max).
38 max_P should be small (not 1), usually 0.01..0.02 is good value.
39
40 max_P is chosen as a number, so that max_P/(th_max-th_min)
41 is a negative power of two in order arithmetics to contain
42 only shifts.
43
44
45 Parameters, settable by user:
46 -----------------------------
47
48 qth_min - bytes (should be < qth_max/2)
49 qth_max - bytes (should be at least 2*qth_min and less limit)
50 Wlog - bits (<32) log(1/W).
51 Plog - bits (<32)
52
53 Plog is related to max_P by formula:
54
55 max_P = (qth_max-qth_min)/2^Plog;
56
57 F.e. if qth_max=128K and qth_min=32K, then Plog=22
58 corresponds to max_P=0.02
59
60 Scell_log
61 Stab
62
63 Lookup table for log((1-W)^(t/t_ave).
64
65
66 NOTES:
67
68 Upper bound on W.
69 -----------------
70
71 If you want to allow bursts of L packets of size S,
72 you should choose W:
73
74 L + 1 - th_min/S < (1-(1-W)^L)/W
75
76 th_min/S = 32 th_min/S = 4
77
78 log(W) L
79 -1 33
80 -2 35
81 -3 39
82 -4 46
83 -5 57
84 -6 75
85 -7 101
86 -8 135
87 -9 190
88 etc.
89 */
90
8af2a218
ED
91/*
92 * Adaptative RED : An Algorithm for Increasing the Robustness of RED's AQM
93 * (Sally FLoyd, Ramakrishna Gummadi, and Scott Shenker) August 2001
94 *
95 * Every 500 ms:
96 * if (avg > target and max_p <= 0.5)
97 * increase max_p : max_p += alpha;
98 * else if (avg < target and max_p >= 0.01)
99 * decrease max_p : max_p *= beta;
100 *
101 * target :[qth_min + 0.4*(qth_min - qth_max),
102 * qth_min + 0.6*(qth_min - qth_max)].
103 * alpha : min(0.01, max_p / 4)
104 * beta : 0.9
105 * max_P is a Q0.32 fixed point number (with 32 bits mantissa)
106 * max_P between 0.01 and 0.5 (1% - 50%) [ Its no longer a negative power of two ]
107 */
108#define RED_ONE_PERCENT ((u32)DIV_ROUND_CLOSEST(1ULL<<32, 100))
109
110#define MAX_P_MIN (1 * RED_ONE_PERCENT)
111#define MAX_P_MAX (50 * RED_ONE_PERCENT)
112#define MAX_P_ALPHA(val) min(MAX_P_MIN, val / 4)
113
a7834745
TG
114#define RED_STAB_SIZE 256
115#define RED_STAB_MASK (RED_STAB_SIZE - 1)
116
fd2c3ef7 117struct red_stats {
a7834745
TG
118 u32 prob_drop; /* Early probability drops */
119 u32 prob_mark; /* Early probability marks */
120 u32 forced_drop; /* Forced drops, qavg > max_thresh */
121 u32 forced_mark; /* Forced marks, qavg > max_thresh */
122 u32 pdrop; /* Drops due to queue limits */
123 u32 other; /* Drops due to drop() calls */
a7834745
TG
124};
125
fd2c3ef7 126struct red_parms {
a7834745 127 /* Parameters */
8af2a218
ED
128 u32 qth_min; /* Min avg length threshold: Wlog scaled */
129 u32 qth_max; /* Max avg length threshold: Wlog scaled */
a7834745 130 u32 Scell_max;
8af2a218
ED
131 u32 max_P; /* probability, [0 .. 1.0] 32 scaled */
132 u32 max_P_reciprocal; /* reciprocal_value(max_P / qth_delta) */
133 u32 qth_delta; /* max_th - min_th */
134 u32 target_min; /* min_th + 0.4*(max_th - min_th) */
135 u32 target_max; /* min_th + 0.6*(max_th - min_th) */
a7834745
TG
136 u8 Scell_log;
137 u8 Wlog; /* log(W) */
138 u8 Plog; /* random number bits */
139 u8 Stab[RED_STAB_SIZE];
eeca6688 140};
a7834745 141
eeca6688 142struct red_vars {
a7834745
TG
143 /* Variables */
144 int qcount; /* Number of packets since last random
145 number generation */
146 u32 qR; /* Cached random number */
147
8af2a218 148 unsigned long qavg; /* Average queue length: Wlog scaled */
ea6a5d3b 149 ktime_t qidlestart; /* Start of current idle period */
a7834745
TG
150};
151
8af2a218 152static inline u32 red_maxp(u8 Plog)
a7834745 153{
8af2a218 154 return Plog < 32 ? (~0U >> Plog) : ~0U;
a7834745
TG
155}
156
eeca6688
ED
157static inline void red_set_vars(struct red_vars *v)
158{
159 /* Reset average queue length, the value is strictly bound
160 * to the parameters below, reseting hurts a bit but leaving
161 * it might result in an unreasonable qavg for a while. --TGR
162 */
163 v->qavg = 0;
164
165 v->qcount = -1;
166}
8af2a218 167
a7834745
TG
168static inline void red_set_parms(struct red_parms *p,
169 u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog,
a73ed26b 170 u8 Scell_log, u8 *stab, u32 max_P)
a7834745 171{
8af2a218 172 int delta = qth_max - qth_min;
a73ed26b 173 u32 max_p_delta;
8af2a218 174
a7834745
TG
175 p->qth_min = qth_min << Wlog;
176 p->qth_max = qth_max << Wlog;
177 p->Wlog = Wlog;
178 p->Plog = Plog;
8af2a218
ED
179 if (delta < 0)
180 delta = 1;
181 p->qth_delta = delta;
a73ed26b
ED
182 if (!max_P) {
183 max_P = red_maxp(Plog);
184 max_P *= delta; /* max_P = (qth_max - qth_min)/2^Plog */
185 }
186 p->max_P = max_P;
187 max_p_delta = max_P / delta;
188 max_p_delta = max(max_p_delta, 1U);
189 p->max_P_reciprocal = reciprocal_value(max_p_delta);
8af2a218
ED
190
191 /* RED Adaptative target :
192 * [min_th + 0.4*(min_th - max_th),
193 * min_th + 0.6*(min_th - max_th)].
194 */
195 delta /= 5;
196 p->target_min = qth_min + 2*delta;
197 p->target_max = qth_min + 3*delta;
198
a7834745
TG
199 p->Scell_log = Scell_log;
200 p->Scell_max = (255 << Scell_log);
201
ddecf0f4
ED
202 if (stab)
203 memcpy(p->Stab, stab, sizeof(p->Stab));
a7834745
TG
204}
205
eeca6688 206static inline int red_is_idling(const struct red_vars *v)
a7834745 207{
eeca6688 208 return v->qidlestart.tv64 != 0;
a7834745
TG
209}
210
eeca6688 211static inline void red_start_of_idle_period(struct red_vars *v)
a7834745 212{
eeca6688 213 v->qidlestart = ktime_get();
a7834745
TG
214}
215
eeca6688 216static inline void red_end_of_idle_period(struct red_vars *v)
a7834745 217{
eeca6688 218 v->qidlestart.tv64 = 0;
a7834745
TG
219}
220
eeca6688 221static inline void red_restart(struct red_vars *v)
a7834745 222{
eeca6688
ED
223 red_end_of_idle_period(v);
224 v->qavg = 0;
225 v->qcount = -1;
a7834745
TG
226}
227
eeca6688
ED
228static inline unsigned long red_calc_qavg_from_idle_time(const struct red_parms *p,
229 const struct red_vars *v)
a7834745 230{
eeca6688 231 s64 delta = ktime_us_delta(ktime_get(), v->qidlestart);
ea6a5d3b 232 long us_idle = min_t(s64, delta, p->Scell_max);
a7834745
TG
233 int shift;
234
a7834745
TG
235 /*
236 * The problem: ideally, average length queue recalcultion should
237 * be done over constant clock intervals. This is too expensive, so
238 * that the calculation is driven by outgoing packets.
239 * When the queue is idle we have to model this clock by hand.
240 *
241 * SF+VJ proposed to "generate":
242 *
243 * m = idletime / (average_pkt_size / bandwidth)
244 *
245 * dummy packets as a burst after idle time, i.e.
246 *
247 * p->qavg *= (1-W)^m
248 *
249 * This is an apparently overcomplicated solution (f.e. we have to
250 * precompute a table to make this calculation in reasonable time)
251 * I believe that a simpler model may be used here,
252 * but it is field for experiments.
253 */
254
255 shift = p->Stab[(us_idle >> p->Scell_log) & RED_STAB_MASK];
256
257 if (shift)
eeca6688 258 return v->qavg >> shift;
a7834745
TG
259 else {
260 /* Approximate initial part of exponent with linear function:
261 *
262 * (1-W)^m ~= 1-mW + ...
263 *
264 * Seems, it is the best solution to
265 * problem of too coarse exponent tabulation.
266 */
eeca6688 267 us_idle = (v->qavg * (u64)us_idle) >> p->Scell_log;
a7834745 268
eeca6688
ED
269 if (us_idle < (v->qavg >> 1))
270 return v->qavg - us_idle;
a7834745 271 else
eeca6688 272 return v->qavg >> 1;
a7834745
TG
273 }
274}
275
8af2a218 276static inline unsigned long red_calc_qavg_no_idle_time(const struct red_parms *p,
eeca6688 277 const struct red_vars *v,
a7834745
TG
278 unsigned int backlog)
279{
280 /*
281 * NOTE: p->qavg is fixed point number with point at Wlog.
282 * The formula below is equvalent to floating point
283 * version:
284 *
285 * qavg = qavg*(1-W) + backlog*W;
286 *
287 * --ANK (980924)
288 */
eeca6688 289 return v->qavg + (backlog - (v->qavg >> p->Wlog));
a7834745
TG
290}
291
8af2a218 292static inline unsigned long red_calc_qavg(const struct red_parms *p,
eeca6688 293 const struct red_vars *v,
a7834745
TG
294 unsigned int backlog)
295{
eeca6688
ED
296 if (!red_is_idling(v))
297 return red_calc_qavg_no_idle_time(p, v, backlog);
a7834745 298 else
eeca6688 299 return red_calc_qavg_from_idle_time(p, v);
a7834745
TG
300}
301
8af2a218
ED
302
303static inline u32 red_random(const struct red_parms *p)
a7834745 304{
8af2a218 305 return reciprocal_divide(net_random(), p->max_P_reciprocal);
a7834745
TG
306}
307
eeca6688
ED
308static inline int red_mark_probability(const struct red_parms *p,
309 const struct red_vars *v,
310 unsigned long qavg)
a7834745
TG
311{
312 /* The formula used below causes questions.
313
8af2a218
ED
314 OK. qR is random number in the interval
315 (0..1/max_P)*(qth_max-qth_min)
a7834745
TG
316 i.e. 0..(2^Plog). If we used floating point
317 arithmetics, it would be: (2^Plog)*rnd_num,
318 where rnd_num is less 1.
319
320 Taking into account, that qavg have fixed
8af2a218 321 point at Wlog, two lines
a7834745
TG
322 below have the following floating point equivalent:
323
324 max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount
325
326 Any questions? --ANK (980924)
327 */
eeca6688 328 return !(((qavg - p->qth_min) >> p->Wlog) * v->qcount < v->qR);
a7834745
TG
329}
330
331enum {
332 RED_BELOW_MIN_THRESH,
333 RED_BETWEEN_TRESH,
334 RED_ABOVE_MAX_TRESH,
335};
336
eeca6688 337static inline int red_cmp_thresh(const struct red_parms *p, unsigned long qavg)
a7834745
TG
338{
339 if (qavg < p->qth_min)
340 return RED_BELOW_MIN_THRESH;
341 else if (qavg >= p->qth_max)
342 return RED_ABOVE_MAX_TRESH;
343 else
344 return RED_BETWEEN_TRESH;
345}
346
347enum {
348 RED_DONT_MARK,
349 RED_PROB_MARK,
350 RED_HARD_MARK,
351};
352
eeca6688
ED
353static inline int red_action(const struct red_parms *p,
354 struct red_vars *v,
355 unsigned long qavg)
a7834745
TG
356{
357 switch (red_cmp_thresh(p, qavg)) {
358 case RED_BELOW_MIN_THRESH:
eeca6688 359 v->qcount = -1;
a7834745
TG
360 return RED_DONT_MARK;
361
362 case RED_BETWEEN_TRESH:
eeca6688
ED
363 if (++v->qcount) {
364 if (red_mark_probability(p, v, qavg)) {
365 v->qcount = 0;
366 v->qR = red_random(p);
a7834745
TG
367 return RED_PROB_MARK;
368 }
369 } else
eeca6688 370 v->qR = red_random(p);
a7834745
TG
371
372 return RED_DONT_MARK;
373
374 case RED_ABOVE_MAX_TRESH:
eeca6688 375 v->qcount = -1;
a7834745
TG
376 return RED_HARD_MARK;
377 }
378
379 BUG();
380 return RED_DONT_MARK;
381}
382
eeca6688 383static inline void red_adaptative_algo(struct red_parms *p, struct red_vars *v)
8af2a218
ED
384{
385 unsigned long qavg;
386 u32 max_p_delta;
387
eeca6688
ED
388 qavg = v->qavg;
389 if (red_is_idling(v))
390 qavg = red_calc_qavg_from_idle_time(p, v);
8af2a218
ED
391
392 /* p->qavg is fixed point number with point at Wlog */
393 qavg >>= p->Wlog;
394
395 if (qavg > p->target_max && p->max_P <= MAX_P_MAX)
396 p->max_P += MAX_P_ALPHA(p->max_P); /* maxp = maxp + alpha */
397 else if (qavg < p->target_min && p->max_P >= MAX_P_MIN)
398 p->max_P = (p->max_P/10)*9; /* maxp = maxp * Beta */
399
400 max_p_delta = DIV_ROUND_CLOSEST(p->max_P, p->qth_delta);
a73ed26b 401 max_p_delta = max(max_p_delta, 1U);
8af2a218
ED
402 p->max_P_reciprocal = reciprocal_value(max_p_delta);
403}
a7834745 404#endif
This page took 0.562058 seconds and 5 git commands to generate.