1 /* 2 * Copyright (c) 2016-2021 Apple Inc. All rights reserved. 3 * 4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ 5 * 6 * This file contains Original Code and/or Modifications of Original Code 7 * as defined in and that are subject to the Apple Public Source License 8 * Version 2.0 (the 'License'). You may not use this file except in 9 * compliance with the License. The rights granted to you under the License 10 * may not be used to create, or enable the creation or redistribution of, 11 * unlawful or unlicensed copies of an Apple operating system, or to 12 * circumvent, violate, or enable the circumvention or violation of, any 13 * terms of an Apple operating system software license agreement. 14 * 15 * Please obtain a copy of the License at 16 * http://www.opensource.apple.com/apsl/ and read it before using this file. 17 * 18 * The Original Code and all software distributed under the License are 19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER 20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, 21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, 22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. 23 * Please see the License for the specific language governing rights and 24 * limitations under the License. 25 * 26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ 27 */ 28 29 #ifndef _NET_PKTSCHED_FQ_CODEL_H_ 30 #define _NET_PKTSCHED_FQ_CODEL_H_ 31 32 #ifdef PRIVATE 33 #include <sys/types.h> 34 #include <sys/param.h> 35 36 #ifdef BSD_KERNEL_PRIVATE 37 #include <net/flowadv.h> 38 #include <net/pktsched/pktsched.h> 39 #endif /* BSD_KERNEL_PRIVATE */ 40 41 #ifdef __cplusplus 42 extern "C" { 43 #endif 44 45 #ifdef BSD_KERNEL_PRIVATE 46 struct fcl_stat { 47 u_int32_t fcl_flow_control; 48 u_int32_t fcl_flow_feedback; 49 u_int32_t fcl_dequeue_stall; 50 u_int32_t fcl_flow_control_fail; 51 u_int64_t fcl_drop_overflow; 52 u_int64_t fcl_drop_early; 53 u_int32_t fcl_drop_memfailure; 54 u_int32_t fcl_flows_cnt; 55 u_int32_t fcl_newflows_cnt; 56 u_int32_t fcl_oldflows_cnt; 57 u_int64_t fcl_pkt_cnt; 58 u_int64_t fcl_dequeue; 59 u_int64_t fcl_dequeue_bytes; 60 u_int64_t fcl_byte_cnt; 61 u_int32_t fcl_throttle_on; 62 u_int32_t fcl_throttle_off; 63 u_int32_t fcl_throttle_drops; 64 u_int32_t fcl_dup_rexmts; 65 u_int32_t fcl_pkts_compressible; 66 u_int32_t fcl_pkts_compressed; 67 uint64_t fcl_min_qdelay; 68 uint64_t fcl_max_qdelay; 69 uint64_t fcl_avg_qdelay; 70 uint32_t fcl_overwhelming; 71 uint64_t fcl_ce_marked; 72 uint64_t fcl_ce_mark_failures; 73 uint64_t fcl_l4s_pkts; 74 }; 75 76 /* 77 * Use 8 bits from the flow id as the tag for set associative 78 * hashing 79 * NOTE: The first 2 bits of the flow id is being used to encode the flow 80 * domain information, so don't use the top 8 bits as it won't have a uniform 81 * distribution. 82 */ 83 84 #define FQ_IF_HASH_TAG_SIZE 8 85 #define FQ_IF_HASH_TAG_SHIFT 16 86 #define FQ_IF_HASH_TAG_MASK 0xFF 87 #define FQ_IF_HASH_TABLE_SIZE (1 << FQ_IF_HASH_TAG_SIZE) 88 89 /* Set the quantum to be one MTU */ 90 #define FQ_IF_DEFAULT_QUANTUM 1500 91 92 /* Max number of service classes currently supported */ 93 #define FQ_IF_MAX_CLASSES 10 94 _Static_assert(FQ_IF_MAX_CLASSES < 127, 95 "maximum number of classes needs to fit in a single byte"); 96 97 #define FQ_IF_LARGE_FLOW_BYTE_LIMIT 15000 98 99 /* Max number of classq groups currently supported */ 100 #define FQ_IF_MAX_GROUPS 16 101 102 typedef enum : uint8_t { 103 FQ_TFC_C = 0, /* classic traffic */ 104 FQ_TFC_L4S = 1, /* L4S traffic */ 105 FQ_TFC_CNT = 2, 106 } fq_tfc_type_t; 107 108 struct flowq; 109 typedef u_int32_t pktsched_bitmap_t; 110 struct if_ifclassq_stats; 111 112 typedef enum : uint8_t { 113 FQ_IF_ER = 0, /* eligible, ready */ 114 FQ_IF_IR = 1, /* ineligible, ready */ 115 FQ_IF_EB = 2, /* eligible blocked */ 116 FQ_IF_IB = 3, /* ineligible, blocked */ 117 FQ_IF_MAX_STATE 118 } fq_if_state; 119 120 /* 121 * This priority index is used for QFQ state bitmaps, lower index gets 122 * higher priority 123 */ 124 #define FQ_IF_BK_SYS_INDEX 9 125 #define FQ_IF_BK_INDEX 8 126 #define FQ_IF_BE_INDEX 7 127 #define FQ_IF_RD_INDEX 6 128 #define FQ_IF_OAM_INDEX 5 129 #define FQ_IF_AV_INDEX 4 130 #define FQ_IF_RV_INDEX 3 131 #define FQ_IF_VI_INDEX 2 132 #define FQ_IF_SIG_INDEX 2 133 #define FQ_IF_VO_INDEX 1 134 #define FQ_IF_CTL_INDEX 0 135 136 typedef SLIST_HEAD(, flowq) flowq_list_t; 137 typedef STAILQ_HEAD(, flowq) flowq_stailq_t; 138 typedef struct fq_if_classq { 139 uint32_t fcl_pri; /* class priority, lower the better */ 140 uint32_t fcl_service_class; /* service class */ 141 uint32_t fcl_quantum; /* quantum in bytes */ 142 uint32_t fcl_drr_max; /* max flows per class for DRR */ 143 int64_t fcl_budget; /* budget for this classq */ 144 flowq_stailq_t fcl_new_flows; /* List of new flows */ 145 flowq_stailq_t fcl_old_flows; /* List of old flows */ 146 struct fcl_stat fcl_stat; 147 } fq_if_classq_t; 148 typedef struct fq_codel_classq_group { 149 /* Target queue delays (ns) */ 150 uint64_t fqg_target_qdelays[FQ_TFC_CNT]; 151 /* update intervals (ns) */ 152 uint64_t fqg_update_intervals[FQ_TFC_CNT]; 153 /* classq bitmaps */ 154 pktsched_bitmap_t fqg_bitmaps[FQ_IF_MAX_STATE]; 155 TAILQ_ENTRY(fq_codel_classq_group) fqg_grp_link; 156 uint32_t fqg_bytes; /* bytes count */ 157 uint32_t fqg_len; /* pkts count */ 158 uint8_t fqg_flags; /* flags */ 159 #define FQ_IF_DEFAULT_GRP 0x1 160 uint8_t fqg_index; /* group index */ 161 fq_if_classq_t fqg_classq[FQ_IF_MAX_CLASSES]; /* class queues */ 162 struct flowq *fqg_large_flow; /* flow has highest number of bytes */ 163 } fq_if_group_t; 164 165 #define FQG_LEN(_fqg) ((_fqg)->fqg_len) 166 #define FQG_IS_EMPTY(_fqg) (FQG_LEN(_fqg) == 0) 167 #define FQG_INC_LEN(_fqg) (FQG_LEN(_fqg)++) 168 #define FQG_DEC_LEN(_fqg) (FQG_LEN(_fqg)--) 169 #define FQG_ADD_LEN(_fqg, _len) (FQG_LEN(_fqg) += (_len)) 170 #define FQG_SUB_LEN(_fqg, _len) (FQG_LEN(_fqg) -= (_len)) 171 #define FQG_BYTES(_fqg) ((_fqg)->fqg_bytes) 172 173 #define FQG_INC_BYTES(_fqg, _len) \ 174 ((_fqg)->fqg_bytes = (_fqg)->fqg_bytes + (_len)) 175 #define FQG_DEC_BYTES(_fqg, _len) \ 176 ((_fqg)->fqg_bytes = (_fqg)->fqg_bytes - (_len)) 177 178 typedef TAILQ_HEAD(, fq_codel_classq_group) fq_grp_tailq_t; 179 180 typedef int (* fq_if_bitmaps_ffs)(fq_grp_tailq_t *, int, fq_if_state, fq_if_group_t **); 181 typedef boolean_t (* fq_if_bitmaps_zeros)(fq_grp_tailq_t *, int, fq_if_state); 182 typedef void (* fq_if_bitmaps_cpy)(fq_grp_tailq_t *, int, fq_if_state, fq_if_state); 183 typedef void (* fq_if_bitmaps_clr)(fq_grp_tailq_t *, int, fq_if_state); 184 185 /* 186 * Functions that are used to look at groups' 187 * bitmaps and decide which pri and group are the 188 * next one to dequeue from. 189 */ 190 typedef struct fq_if_bitmap_ops { 191 fq_if_bitmaps_ffs ffs; 192 fq_if_bitmaps_zeros zeros; 193 fq_if_bitmaps_cpy cpy; 194 fq_if_bitmaps_clr clr; 195 } bitmap_ops_t; 196 197 typedef struct fq_codel_sched_data { 198 struct ifclassq *fqs_ifq; /* back pointer to ifclassq */ 199 flowq_list_t fqs_flows[FQ_IF_HASH_TABLE_SIZE]; /* flows table */ 200 uint32_t fqs_pkt_droplimit; /* drop limit */ 201 uint8_t fqs_throttle; /* throttle on or off */ 202 uint8_t fqs_flags; /* flags */ 203 #define FQS_DRIVER_MANAGED 0x1 204 struct flowadv_fclist fqs_fclist; /* flow control state */ 205 struct flowq *fqs_large_flow; /* flow has highest number of bytes */ 206 TAILQ_HEAD(, flowq) fqs_empty_list; /* list of empty flows */ 207 /* list of groups in combined mode */ 208 fq_grp_tailq_t fqs_combined_grp_list; 209 uint32_t fqs_empty_list_cnt; 210 /* bitmap indicating which grp is in combined mode */ 211 pktsched_bitmap_t fqs_combined_grp_bitmap; 212 classq_pkt_type_t fqs_ptype; 213 bitmap_ops_t *fqs_bm_ops; 214 #define grp_bitmaps_ffs fqs_bm_ops->ffs 215 #define grp_bitmaps_zeros fqs_bm_ops->zeros 216 #define grp_bitmaps_cpy fqs_bm_ops->cpy 217 #define grp_bitmaps_clr fqs_bm_ops->clr 218 fq_if_group_t *fqs_classq_groups[FQ_IF_MAX_GROUPS]; 219 } fq_if_t; 220 221 #define FQS_GROUP(_fqs, _group_idx) \ 222 (fq_if_find_grp((_fqs), (_group_idx))) 223 224 #define FQS_CLASSQ(_fqs, _group_idx, _sc_idx) \ 225 (FQS_GROUP((_fqs), (_group_idx))->fqg_classq[_sc_idx]) 226 227 #define FQ_GROUP(_fq) \ 228 ((_fq)->fq_group) 229 230 #define FQ_GRP_LEN(_fq) \ 231 (FQ_GROUP((_fq))->fqg_len) 232 #define FQ_GRP_IS_EMPTY(_fq) \ 233 (FQ_GRP_LEN((_fq)) == 0) 234 #define FQ_GRP_INC_LEN(_fq) \ 235 (FQ_GRP_LEN((_fq))++) 236 #define FQ_GRP_DEC_LEN(_fq) \ 237 (FQ_GRP_LEN((_fq))--) 238 #define FQ_GRP_ADD_LEN(_fq, _len) \ 239 (FQ_GRP_LEN((_fq)) += (_len)) 240 #define FQ_GRP_SUB_LEN(_fq, _len) \ 241 (FQ_GRP_LEN((_fq)) -= (_len)) 242 243 #define FQS_GRP_ADD_LEN(_fqs, _grp_idx, _len) \ 244 (FQS_GROUP(_fqs, grp_idx)->fqg_len += (_len)) 245 246 247 #define FQ_GRP_BYTES(_fq) \ 248 (FQ_GROUP((_fq))->fqg_bytes) 249 #define FQ_GRP_INC_BYTES(_fq, _len) \ 250 (FQ_GRP_BYTES((_fq)) += (_len)) 251 #define FQ_GRP_DEC_BYTES(_fq, _len) \ 252 (FQ_GRP_BYTES((_fq)) -= (_len)) 253 254 #define FQS_GRP_INC_BYTES(_fqs, grp_idx, _len) \ 255 (FQS_GROUP(_fqs, grp_idx)->fqg_bytes += (_len)) 256 257 #define FQ_CLASSQ(_fq) \ 258 (FQ_GROUP((_fq))->fqg_classq[(_fq)->fq_sc_index]) 259 260 #define FQ_TARGET_DELAY(_fq) \ 261 (FQ_GROUP((_fq))->fqg_target_qdelays[(_fq)->fq_tfc_type]) 262 #define FQ_UPDATE_INTERVAL(_fq) \ 263 (FQ_GROUP((_fq))->fqg_update_intervals[(_fq)->fq_tfc_type]) 264 265 #endif /* BSD_KERNEL_PRIVATE */ 266 267 struct fq_codel_flowstats { 268 u_int32_t fqst_min_qdelay; 269 #define FQ_FLOWSTATS_OLD_FLOW 0x1 270 #define FQ_FLOWSTATS_NEW_FLOW 0x2 271 #define FQ_FLOWSTATS_LARGE_FLOW 0x4 272 #define FQ_FLOWSTATS_DELAY_HIGH 0x8 273 #define FQ_FLOWSTATS_FLOWCTL_ON 0x10 274 u_int32_t fqst_flags; 275 u_int32_t fqst_bytes; 276 u_int32_t fqst_flowhash; 277 }; 278 279 #define FQ_IF_MAX_FLOWSTATS 20 280 281 struct fq_codel_classstats { 282 u_int32_t fcls_pri; 283 u_int32_t fcls_service_class; 284 u_int32_t fcls_quantum; 285 u_int32_t fcls_drr_max; 286 int64_t fcls_budget; 287 u_int64_t fcls_target_qdelay; 288 u_int64_t fcls_l4s_target_qdelay; 289 u_int64_t fcls_update_interval; 290 u_int32_t fcls_flow_control; 291 u_int32_t fcls_flow_feedback; 292 u_int32_t fcls_dequeue_stall; 293 u_int32_t fcls_flow_control_fail; 294 u_int64_t fcls_drop_overflow; 295 u_int64_t fcls_drop_early; 296 u_int32_t fcls_drop_memfailure; 297 u_int32_t fcls_flows_cnt; 298 u_int32_t fcls_newflows_cnt; 299 u_int32_t fcls_oldflows_cnt; 300 u_int64_t fcls_pkt_cnt; 301 u_int64_t fcls_dequeue; 302 u_int64_t fcls_dequeue_bytes; 303 u_int64_t fcls_byte_cnt; 304 u_int32_t fcls_throttle_on; 305 u_int32_t fcls_throttle_off; 306 u_int32_t fcls_throttle_drops; 307 u_int32_t fcls_dup_rexmts; 308 u_int32_t fcls_flowstats_cnt; 309 struct fq_codel_flowstats fcls_flowstats[FQ_IF_MAX_FLOWSTATS]; 310 u_int32_t fcls_pkts_compressible; 311 u_int32_t fcls_pkts_compressed; 312 uint64_t fcls_min_qdelay; 313 uint64_t fcls_max_qdelay; 314 uint64_t fcls_avg_qdelay; 315 uint32_t fcls_overwhelming; 316 uint64_t fcls_ce_marked; 317 uint64_t fcls_ce_mark_failures; 318 uint64_t fcls_l4s_pkts; 319 }; 320 321 #ifdef BSD_KERNEL_PRIVATE 322 extern void pktsched_fq_init(void); 323 extern void fq_codel_scheduler_init(void); 324 extern int fq_if_enqueue_classq(struct ifclassq *ifq, classq_pkt_t *h, 325 classq_pkt_t *t, uint32_t cnt, uint32_t bytes, boolean_t *pdrop); 326 extern void fq_if_dequeue_classq(struct ifclassq *ifq, classq_pkt_t *pkt, 327 uint8_t grp_idx); 328 extern void fq_if_dequeue_sc_classq(struct ifclassq *ifq, mbuf_svc_class_t svc, 329 classq_pkt_t *pkt, uint8_t grp_idx); 330 extern int fq_if_dequeue_classq_multi(struct ifclassq *ifq, u_int32_t maxpktcnt, 331 u_int32_t maxbytecnt, classq_pkt_t *first_packet, classq_pkt_t *last_packet, 332 u_int32_t *retpktcnt, u_int32_t *retbytecnt, uint8_t grp_idx); 333 extern int fq_if_dequeue_sc_classq_multi(struct ifclassq *ifq, 334 mbuf_svc_class_t svc, u_int32_t maxpktcnt, u_int32_t maxbytecnt, 335 classq_pkt_t *first_packet, classq_pkt_t *last_packet, u_int32_t *retpktcnt, 336 u_int32_t *retbytecnt, uint8_t grp_idx); 337 extern int fq_if_request_classq(struct ifclassq *ifq, cqrq_t rq, void *arg); 338 extern struct flowq *fq_if_hash_pkt(fq_if_t *, fq_if_group_t *, 339 u_int32_t, mbuf_svc_class_t, u_int64_t, bool, fq_tfc_type_t); 340 extern boolean_t fq_if_at_drop_limit(fq_if_t *); 341 extern boolean_t fq_if_almost_at_drop_limit(fq_if_t *fqs); 342 extern void fq_if_drop_packet(fq_if_t *, uint64_t); 343 extern void fq_if_is_flow_heavy(fq_if_t *, struct flowq *); 344 extern boolean_t fq_if_add_fcentry(fq_if_t *, pktsched_pkt_t *, uint8_t, 345 struct flowq *, fq_if_classq_t *); 346 extern void fq_if_flow_feedback(fq_if_t *, struct flowq *, fq_if_classq_t *); 347 extern int fq_if_setup_ifclassq(struct ifclassq *ifq, u_int32_t flags, 348 classq_pkt_type_t ptype); 349 extern void fq_if_teardown_ifclassq(struct ifclassq *ifq); 350 extern int fq_if_getqstats_ifclassq(struct ifclassq *ifq, uint8_t gid, 351 u_int32_t qid, struct if_ifclassq_stats *ifqs); 352 extern void fq_if_destroy_flow(fq_if_t *, fq_if_classq_t *, struct flowq *); 353 extern void fq_if_move_to_empty_flow(fq_if_t *, fq_if_classq_t *, 354 struct flowq *, uint64_t); 355 extern int fq_if_create_grp(struct ifclassq *ifcq, uint8_t qset_idx, uint8_t flags); 356 extern void fq_if_set_grp_combined(struct ifclassq *ifcq, uint8_t qset_idx); 357 extern void fq_if_set_grp_separated(struct ifclassq *ifcq, uint8_t qset_idx); 358 extern fq_if_group_t *fq_if_find_grp(fq_if_t *fqs, uint8_t grp_idx); 359 #endif /* BSD_KERNEL_PRIVATE */ 360 361 #ifdef __cplusplus 362 } 363 #endif 364 365 #endif /* PRIVATE */ 366 #endif /* _NET_PKTSCHED_PKTSCHED_FQ_CODEL_H_ */ 367