1 #ifndef __NET_SCHED_RED_H 2 #define __NET_SCHED_RED_H 3 4 #include <linux/types.h> 5 #include <net/pkt_sched.h> 6 #include <net/inet_ecn.h> 7 #include <net/dsfield.h> 8 9 /* Random Early Detection (RED) algorithm. 10 ======================================= 11 12 Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways 13 for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking. 14 15 This file codes a "divisionless" version of RED algorithm 16 as written down in Fig.17 of the paper. 17 18 Short description. 19 ------------------ 20 21 When a new packet arrives we calculate the average queue length: 22 23 avg = (1-W)*avg + W*current_queue_len, 24 25 W is the filter time constant (chosen as 2^(-Wlog)), it controls 26 the inertia of the algorithm. To allow larger bursts, W should be 27 decreased. 28 29 if (avg > th_max) -> packet marked (dropped). 30 if (avg < th_min) -> packet passes. 31 if (th_min < avg < th_max) we calculate probability: 32 33 Pb = max_P * (avg - th_min)/(th_max-th_min) 34 35 and mark (drop) packet with this probability. 36 Pb changes from 0 (at avg==th_min) to max_P (avg==th_max). 37 max_P should be small (not 1), usually 0.01..0.02 is good value. 38 39 max_P is chosen as a number, so that max_P/(th_max-th_min) 40 is a negative power of two in order arithmetics to contain 41 only shifts. 42 43 44 Parameters, settable by user: 45 ----------------------------- 46 47 qth_min - bytes (should be < qth_max/2) 48 qth_max - bytes (should be at least 2*qth_min and less limit) 49 Wlog - bits (<32) log(1/W). 50 Plog - bits (<32) 51 52 Plog is related to max_P by formula: 53 54 max_P = (qth_max-qth_min)/2^Plog; 55 56 F.e. if qth_max=128K and qth_min=32K, then Plog=22 57 corresponds to max_P=0.02 58 59 Scell_log 60 Stab 61 62 Lookup table for log((1-W)^(t/t_ave). 63 64 65 NOTES: 66 67 Upper bound on W. 68 ----------------- 69 70 If you want to allow bursts of L packets of size S, 71 you should choose W: 72 73 L + 1 - th_min/S < (1-(1-W)^L)/W 74 75 th_min/S = 32 th_min/S = 4 76 77 log(W) L 78 -1 33 79 -2 35 80 -3 39 81 -4 46 82 -5 57 83 -6 75 84 -7 101 85 -8 135 86 -9 190 87 etc. 88 */ 89 90 #define RED_STAB_SIZE 256 91 #define RED_STAB_MASK (RED_STAB_SIZE - 1) 92 93 struct red_stats { 94 u32 prob_drop; /* Early probability drops */ 95 u32 prob_mark; /* Early probability marks */ 96 u32 forced_drop; /* Forced drops, qavg > max_thresh */ 97 u32 forced_mark; /* Forced marks, qavg > max_thresh */ 98 u32 pdrop; /* Drops due to queue limits */ 99 u32 other; /* Drops due to drop() calls */ 100 }; 101 102 struct red_parms { 103 /* Parameters */ 104 u32 qth_min; /* Min avg length threshold: A scaled */ 105 u32 qth_max; /* Max avg length threshold: A scaled */ 106 u32 Scell_max; 107 u32 Rmask; /* Cached random mask, see red_rmask */ 108 u8 Scell_log; 109 u8 Wlog; /* log(W) */ 110 u8 Plog; /* random number bits */ 111 u8 Stab[RED_STAB_SIZE]; 112 113 /* Variables */ 114 int qcount; /* Number of packets since last random 115 number generation */ 116 u32 qR; /* Cached random number */ 117 118 unsigned long qavg; /* Average queue length: A scaled */ 119 ktime_t qidlestart; /* Start of current idle period */ 120 }; 121 122 static inline u32 red_rmask(u8 Plog) 123 { 124 return Plog < 32 ? ((1 << Plog) - 1) : ~0UL; 125 } 126 127 static inline void red_set_parms(struct red_parms *p, 128 u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog, 129 u8 Scell_log, u8 *stab) 130 { 131 /* Reset average queue length, the value is strictly bound 132 * to the parameters below, reseting hurts a bit but leaving 133 * it might result in an unreasonable qavg for a while. --TGR 134 */ 135 p->qavg = 0; 136 137 p->qcount = -1; 138 p->qth_min = qth_min << Wlog; 139 p->qth_max = qth_max << Wlog; 140 p->Wlog = Wlog; 141 p->Plog = Plog; 142 p->Rmask = red_rmask(Plog); 143 p->Scell_log = Scell_log; 144 p->Scell_max = (255 << Scell_log); 145 146 memcpy(p->Stab, stab, sizeof(p->Stab)); 147 } 148 149 static inline int red_is_idling(struct red_parms *p) 150 { 151 return p->qidlestart.tv64 != 0; 152 } 153 154 static inline void red_start_of_idle_period(struct red_parms *p) 155 { 156 p->qidlestart = ktime_get(); 157 } 158 159 static inline void red_end_of_idle_period(struct red_parms *p) 160 { 161 p->qidlestart.tv64 = 0; 162 } 163 164 static inline void red_restart(struct red_parms *p) 165 { 166 red_end_of_idle_period(p); 167 p->qavg = 0; 168 p->qcount = -1; 169 } 170 171 static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p) 172 { 173 s64 delta = ktime_us_delta(ktime_get(), p->qidlestart); 174 long us_idle = min_t(s64, delta, p->Scell_max); 175 int shift; 176 177 /* 178 * The problem: ideally, average length queue recalcultion should 179 * be done over constant clock intervals. This is too expensive, so 180 * that the calculation is driven by outgoing packets. 181 * When the queue is idle we have to model this clock by hand. 182 * 183 * SF+VJ proposed to "generate": 184 * 185 * m = idletime / (average_pkt_size / bandwidth) 186 * 187 * dummy packets as a burst after idle time, i.e. 188 * 189 * p->qavg *= (1-W)^m 190 * 191 * This is an apparently overcomplicated solution (f.e. we have to 192 * precompute a table to make this calculation in reasonable time) 193 * I believe that a simpler model may be used here, 194 * but it is field for experiments. 195 */ 196 197 shift = p->Stab[(us_idle >> p->Scell_log) & RED_STAB_MASK]; 198 199 if (shift) 200 return p->qavg >> shift; 201 else { 202 /* Approximate initial part of exponent with linear function: 203 * 204 * (1-W)^m ~= 1-mW + ... 205 * 206 * Seems, it is the best solution to 207 * problem of too coarse exponent tabulation. 208 */ 209 us_idle = (p->qavg * (u64)us_idle) >> p->Scell_log; 210 211 if (us_idle < (p->qavg >> 1)) 212 return p->qavg - us_idle; 213 else 214 return p->qavg >> 1; 215 } 216 } 217 218 static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p, 219 unsigned int backlog) 220 { 221 /* 222 * NOTE: p->qavg is fixed point number with point at Wlog. 223 * The formula below is equvalent to floating point 224 * version: 225 * 226 * qavg = qavg*(1-W) + backlog*W; 227 * 228 * --ANK (980924) 229 */ 230 return p->qavg + (backlog - (p->qavg >> p->Wlog)); 231 } 232 233 static inline unsigned long red_calc_qavg(struct red_parms *p, 234 unsigned int backlog) 235 { 236 if (!red_is_idling(p)) 237 return red_calc_qavg_no_idle_time(p, backlog); 238 else 239 return red_calc_qavg_from_idle_time(p); 240 } 241 242 static inline u32 red_random(struct red_parms *p) 243 { 244 return net_random() & p->Rmask; 245 } 246 247 static inline int red_mark_probability(struct red_parms *p, unsigned long qavg) 248 { 249 /* The formula used below causes questions. 250 251 OK. qR is random number in the interval 0..Rmask 252 i.e. 0..(2^Plog). If we used floating point 253 arithmetics, it would be: (2^Plog)*rnd_num, 254 where rnd_num is less 1. 255 256 Taking into account, that qavg have fixed 257 point at Wlog, and Plog is related to max_P by 258 max_P = (qth_max-qth_min)/2^Plog; two lines 259 below have the following floating point equivalent: 260 261 max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount 262 263 Any questions? --ANK (980924) 264 */ 265 return !(((qavg - p->qth_min) >> p->Wlog) * p->qcount < p->qR); 266 } 267 268 enum { 269 RED_BELOW_MIN_THRESH, 270 RED_BETWEEN_TRESH, 271 RED_ABOVE_MAX_TRESH, 272 }; 273 274 static inline int red_cmp_thresh(struct red_parms *p, unsigned long qavg) 275 { 276 if (qavg < p->qth_min) 277 return RED_BELOW_MIN_THRESH; 278 else if (qavg >= p->qth_max) 279 return RED_ABOVE_MAX_TRESH; 280 else 281 return RED_BETWEEN_TRESH; 282 } 283 284 enum { 285 RED_DONT_MARK, 286 RED_PROB_MARK, 287 RED_HARD_MARK, 288 }; 289 290 static inline int red_action(struct red_parms *p, unsigned long qavg) 291 { 292 switch (red_cmp_thresh(p, qavg)) { 293 case RED_BELOW_MIN_THRESH: 294 p->qcount = -1; 295 return RED_DONT_MARK; 296 297 case RED_BETWEEN_TRESH: 298 if (++p->qcount) { 299 if (red_mark_probability(p, qavg)) { 300 p->qcount = 0; 301 p->qR = red_random(p); 302 return RED_PROB_MARK; 303 } 304 } else 305 p->qR = red_random(p); 306 307 return RED_DONT_MARK; 308 309 case RED_ABOVE_MAX_TRESH: 310 p->qcount = -1; 311 return RED_HARD_MARK; 312 } 313 314 BUG(); 315 return RED_DONT_MARK; 316 } 317 318 #endif 319