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 }; 72 73 /* 74 * Use 8 bits from the flow id as the tag for set associative 75 * hashing 76 * NOTE: The first 2 bits of the flow id is being used to encode the flow 77 * domain information, so don't use the top 8 bits as it won't have a uniform 78 * distribution. 79 */ 80 81 #define FQ_IF_HASH_TAG_SIZE 8 82 #define FQ_IF_HASH_TAG_SHIFT 16 83 #define FQ_IF_HASH_TAG_MASK 0xFF 84 #define FQ_IF_HASH_TABLE_SIZE (1 << FQ_IF_HASH_TAG_SIZE) 85 86 /* Set the quantum to be one MTU */ 87 #define FQ_IF_DEFAULT_QUANTUM 1500 88 89 /* Max number of service classes currently supported */ 90 #define FQ_IF_MAX_CLASSES 10 91 _Static_assert(FQ_IF_MAX_CLASSES < 127, 92 "maximum number of classes needs to fit in a single byte"); 93 94 #define FQ_IF_LARGE_FLOW_BYTE_LIMIT 15000 95 96 /* Max number of classq groups currently supported */ 97 #define FQ_IF_MAX_GROUPS 16 98 99 typedef enum : uint8_t { 100 FQ_TFC_C = 0, /* classic traffic */ 101 FQ_TFC_L4S = 1, /* L4S traffic */ 102 FQ_TFC_CNT = 2, 103 } fq_tfc_type_t; 104 105 struct flowq; 106 typedef u_int32_t pktsched_bitmap_t; 107 struct if_ifclassq_stats; 108 109 typedef enum : uint8_t { 110 FQ_IF_ER = 0, /* eligible, ready */ 111 FQ_IF_IR = 1, /* ineligible, ready */ 112 FQ_IF_EB = 2, /* eligible blocked */ 113 FQ_IF_IB = 3, /* ineligible, blocked */ 114 FQ_IF_MAX_STATE 115 } fq_if_state; 116 117 /* 118 * This priority index is used for QFQ state bitmaps, lower index gets 119 * higher priority 120 */ 121 #define FQ_IF_BK_SYS_INDEX 9 122 #define FQ_IF_BK_INDEX 8 123 #define FQ_IF_BE_INDEX 7 124 #define FQ_IF_RD_INDEX 6 125 #define FQ_IF_OAM_INDEX 5 126 #define FQ_IF_AV_INDEX 4 127 #define FQ_IF_RV_INDEX 3 128 #define FQ_IF_VI_INDEX 2 129 #define FQ_IF_SIG_INDEX 2 130 #define FQ_IF_VO_INDEX 1 131 #define FQ_IF_CTL_INDEX 0 132 133 typedef SLIST_HEAD(, flowq) flowq_list_t; 134 typedef STAILQ_HEAD(, flowq) flowq_stailq_t; 135 typedef struct fq_if_classq { 136 uint32_t fcl_pri; /* class priority, lower the better */ 137 uint32_t fcl_service_class; /* service class */ 138 uint32_t fcl_quantum; /* quantum in bytes */ 139 uint32_t fcl_drr_max; /* max flows per class for DRR */ 140 int64_t fcl_budget; /* budget for this classq */ 141 flowq_stailq_t fcl_new_flows; /* List of new flows */ 142 flowq_stailq_t fcl_old_flows; /* List of old flows */ 143 struct fcl_stat fcl_stat; 144 } fq_if_classq_t; 145 typedef struct fq_codel_classq_group { 146 /* Target queue delays (ns) */ 147 uint64_t fqg_target_qdelays[FQ_TFC_CNT]; 148 /* update intervals (ns) */ 149 uint64_t fqg_update_intervals[FQ_TFC_CNT]; 150 /* classq bitmaps */ 151 pktsched_bitmap_t fqg_bitmaps[FQ_IF_MAX_STATE]; 152 TAILQ_ENTRY(fq_codel_classq_group) fqg_grp_link; 153 uint32_t fqg_bytes; /* bytes count */ 154 uint32_t fqg_len; /* pkts count */ 155 uint8_t fqg_flags; /* flags */ 156 #define FQ_IF_DEFAULT_GRP 0x1 157 uint8_t fqg_index; /* group index */ 158 fq_if_classq_t fqg_classq[FQ_IF_MAX_CLASSES]; /* class queues */ 159 struct flowq *fqg_large_flow; /* flow has highest number of bytes */ 160 } fq_if_group_t; 161 162 #define FQG_LEN(_fqg) ((_fqg)->fqg_len) 163 #define FQG_IS_EMPTY(_fqg) (FQG_LEN(_fqg) == 0) 164 #define FQG_INC_LEN(_fqg) (FQG_LEN(_fqg)++) 165 #define FQG_DEC_LEN(_fqg) (FQG_LEN(_fqg)--) 166 #define FQG_ADD_LEN(_fqg, _len) (FQG_LEN(_fqg) += (_len)) 167 #define FQG_SUB_LEN(_fqg, _len) (FQG_LEN(_fqg) -= (_len)) 168 #define FQG_BYTES(_fqg) ((_fqg)->fqg_bytes) 169 170 #define FQG_INC_BYTES(_fqg, _len) \ 171 ((_fqg)->fqg_bytes = (_fqg)->fqg_bytes + (_len)) 172 #define FQG_DEC_BYTES(_fqg, _len) \ 173 ((_fqg)->fqg_bytes = (_fqg)->fqg_bytes - (_len)) 174 175 typedef TAILQ_HEAD(, fq_codel_classq_group) fq_grp_tailq_t; 176 177 typedef int (* fq_if_bitmaps_ffs)(fq_grp_tailq_t *, int, fq_if_state, fq_if_group_t **); 178 typedef boolean_t (* fq_if_bitmaps_zeros)(fq_grp_tailq_t *, int, fq_if_state); 179 typedef void (* fq_if_bitmaps_cpy)(fq_grp_tailq_t *, int, fq_if_state, fq_if_state); 180 typedef void (* fq_if_bitmaps_clr)(fq_grp_tailq_t *, int, fq_if_state); 181 182 /* 183 * Functions that are used to look at groups' 184 * bitmaps and decide which pri and group are the 185 * next one to dequeue from. 186 */ 187 typedef struct fq_if_bitmap_ops { 188 fq_if_bitmaps_ffs ffs; 189 fq_if_bitmaps_zeros zeros; 190 fq_if_bitmaps_cpy cpy; 191 fq_if_bitmaps_clr clr; 192 } bitmap_ops_t; 193 194 typedef struct fq_codel_sched_data { 195 struct ifclassq *fqs_ifq; /* back pointer to ifclassq */ 196 flowq_list_t fqs_flows[FQ_IF_HASH_TABLE_SIZE]; /* flows table */ 197 uint32_t fqs_pkt_droplimit; /* drop limit */ 198 uint8_t fqs_throttle; /* throttle on or off */ 199 uint8_t fqs_flags; /* flags */ 200 #define FQS_DRIVER_MANAGED 0x1 201 struct flowadv_fclist fqs_fclist; /* flow control state */ 202 struct flowq *fqs_large_flow; /* flow has highest number of bytes */ 203 TAILQ_HEAD(, flowq) fqs_empty_list; /* list of empty flows */ 204 /* list of groups in combined mode */ 205 fq_grp_tailq_t fqs_combined_grp_list; 206 uint32_t fqs_empty_list_cnt; 207 /* bitmap indicating which grp is in combined mode */ 208 pktsched_bitmap_t fqs_combined_grp_bitmap; 209 classq_pkt_type_t fqs_ptype; 210 bitmap_ops_t *fqs_bm_ops; 211 #define grp_bitmaps_ffs fqs_bm_ops->ffs 212 #define grp_bitmaps_zeros fqs_bm_ops->zeros 213 #define grp_bitmaps_cpy fqs_bm_ops->cpy 214 #define grp_bitmaps_clr fqs_bm_ops->clr 215 fq_if_group_t *fqs_classq_groups[FQ_IF_MAX_GROUPS]; 216 } fq_if_t; 217 218 #define FQS_GROUP(_fqs, _group_idx) \ 219 (fq_if_find_grp((_fqs), (_group_idx))) 220 221 #define FQS_CLASSQ(_fqs, _group_idx, _sc_idx) \ 222 (FQS_GROUP((_fqs), (_group_idx))->fqg_classq[_sc_idx]) 223 224 #define FQ_GROUP(_fq) \ 225 ((_fq)->fq_group) 226 227 #define FQ_GRP_LEN(_fq) \ 228 (FQ_GROUP((_fq))->fqg_len) 229 #define FQ_GRP_IS_EMPTY(_fq) \ 230 (FQ_GRP_LEN((_fq)) == 0) 231 #define FQ_GRP_INC_LEN(_fq) \ 232 (FQ_GRP_LEN((_fq))++) 233 #define FQ_GRP_DEC_LEN(_fq) \ 234 (FQ_GRP_LEN((_fq))--) 235 #define FQ_GRP_ADD_LEN(_fq, _len) \ 236 (FQ_GRP_LEN((_fq)) += (_len)) 237 #define FQ_GRP_SUB_LEN(_fq, _len) \ 238 (FQ_GRP_LEN((_fq)) -= (_len)) 239 240 #define FQS_GRP_ADD_LEN(_fqs, _grp_idx, _len) \ 241 (FQS_GROUP(_fqs, grp_idx)->fqg_len += (_len)) 242 243 244 #define FQ_GRP_BYTES(_fq) \ 245 (FQ_GROUP((_fq))->fqg_bytes) 246 #define FQ_GRP_INC_BYTES(_fq, _len) \ 247 (FQ_GRP_BYTES((_fq)) += (_len)) 248 #define FQ_GRP_DEC_BYTES(_fq, _len) \ 249 (FQ_GRP_BYTES((_fq)) -= (_len)) 250 251 #define FQS_GRP_INC_BYTES(_fqs, grp_idx, _len) \ 252 (FQS_GROUP(_fqs, grp_idx)->fqg_bytes += (_len)) 253 254 #define FQ_CLASSQ(_fq) \ 255 (FQ_GROUP((_fq))->fqg_classq[(_fq)->fq_sc_index]) 256 257 #define FQ_TARGET_DELAY(_fq) \ 258 (FQ_GROUP((_fq))->fqg_target_qdelays[(_fq)->fq_tfc_type]) 259 #define FQ_UPDATE_INTERVAL(_fq) \ 260 (FQ_GROUP((_fq))->fqg_update_intervals[(_fq)->fq_tfc_type]) 261 262 #endif /* BSD_KERNEL_PRIVATE */ 263 264 struct fq_codel_flowstats { 265 u_int32_t fqst_min_qdelay; 266 #define FQ_FLOWSTATS_OLD_FLOW 0x1 267 #define FQ_FLOWSTATS_NEW_FLOW 0x2 268 #define FQ_FLOWSTATS_LARGE_FLOW 0x4 269 #define FQ_FLOWSTATS_DELAY_HIGH 0x8 270 #define FQ_FLOWSTATS_FLOWCTL_ON 0x10 271 u_int32_t fqst_flags; 272 u_int32_t fqst_bytes; 273 u_int32_t fqst_flowhash; 274 }; 275 276 #define FQ_IF_MAX_FLOWSTATS 20 277 278 struct fq_codel_classstats { 279 u_int32_t fcls_pri; 280 u_int32_t fcls_service_class; 281 u_int32_t fcls_quantum; 282 u_int32_t fcls_drr_max; 283 int64_t fcls_budget; 284 u_int64_t fcls_target_qdelay; 285 u_int64_t fcls_update_interval; 286 u_int32_t fcls_flow_control; 287 u_int32_t fcls_flow_feedback; 288 u_int32_t fcls_dequeue_stall; 289 u_int32_t fcls_flow_control_fail; 290 u_int64_t fcls_drop_overflow; 291 u_int64_t fcls_drop_early; 292 u_int32_t fcls_drop_memfailure; 293 u_int32_t fcls_flows_cnt; 294 u_int32_t fcls_newflows_cnt; 295 u_int32_t fcls_oldflows_cnt; 296 u_int64_t fcls_pkt_cnt; 297 u_int64_t fcls_dequeue; 298 u_int64_t fcls_dequeue_bytes; 299 u_int64_t fcls_byte_cnt; 300 u_int32_t fcls_throttle_on; 301 u_int32_t fcls_throttle_off; 302 u_int32_t fcls_throttle_drops; 303 u_int32_t fcls_dup_rexmts; 304 u_int32_t fcls_flowstats_cnt; 305 struct fq_codel_flowstats fcls_flowstats[FQ_IF_MAX_FLOWSTATS]; 306 u_int32_t fcls_pkts_compressible; 307 u_int32_t fcls_pkts_compressed; 308 uint64_t fcls_min_qdelay; 309 uint64_t fcls_max_qdelay; 310 uint64_t fcls_avg_qdelay; 311 uint32_t fcls_overwhelming; 312 }; 313 314 #ifdef BSD_KERNEL_PRIVATE 315 extern void pktsched_fq_init(void); 316 extern void fq_codel_scheduler_init(void); 317 extern int fq_if_enqueue_classq(struct ifclassq *ifq, classq_pkt_t *h, 318 classq_pkt_t *t, uint32_t cnt, uint32_t bytes, boolean_t *pdrop); 319 extern void fq_if_dequeue_classq(struct ifclassq *ifq, classq_pkt_t *pkt, 320 uint8_t grp_idx); 321 extern void fq_if_dequeue_sc_classq(struct ifclassq *ifq, mbuf_svc_class_t svc, 322 classq_pkt_t *pkt, uint8_t grp_idx); 323 extern int fq_if_dequeue_classq_multi(struct ifclassq *ifq, u_int32_t maxpktcnt, 324 u_int32_t maxbytecnt, classq_pkt_t *first_packet, classq_pkt_t *last_packet, 325 u_int32_t *retpktcnt, u_int32_t *retbytecnt, uint8_t grp_idx); 326 extern int fq_if_dequeue_sc_classq_multi(struct ifclassq *ifq, 327 mbuf_svc_class_t svc, u_int32_t maxpktcnt, u_int32_t maxbytecnt, 328 classq_pkt_t *first_packet, classq_pkt_t *last_packet, u_int32_t *retpktcnt, 329 u_int32_t *retbytecnt, uint8_t grp_idx); 330 extern int fq_if_request_classq(struct ifclassq *ifq, cqrq_t rq, void *arg); 331 extern struct flowq *fq_if_hash_pkt(fq_if_t *, fq_if_group_t *, 332 u_int32_t, mbuf_svc_class_t, u_int64_t, bool, fq_tfc_type_t); 333 extern boolean_t fq_if_at_drop_limit(fq_if_t *); 334 extern boolean_t fq_if_almost_at_drop_limit(fq_if_t *fqs); 335 extern void fq_if_drop_packet(fq_if_t *, uint64_t); 336 extern void fq_if_is_flow_heavy(fq_if_t *, struct flowq *); 337 extern boolean_t fq_if_add_fcentry(fq_if_t *, pktsched_pkt_t *, uint8_t, 338 struct flowq *, fq_if_classq_t *); 339 extern void fq_if_flow_feedback(fq_if_t *, struct flowq *, fq_if_classq_t *); 340 extern int fq_if_setup_ifclassq(struct ifclassq *ifq, u_int32_t flags, 341 classq_pkt_type_t ptype); 342 extern void fq_if_teardown_ifclassq(struct ifclassq *ifq); 343 extern int fq_if_getqstats_ifclassq(struct ifclassq *ifq, uint8_t gid, 344 u_int32_t qid, struct if_ifclassq_stats *ifqs); 345 extern void fq_if_destroy_flow(fq_if_t *, fq_if_classq_t *, struct flowq *); 346 extern void fq_if_move_to_empty_flow(fq_if_t *, fq_if_classq_t *, 347 struct flowq *, uint64_t); 348 extern int fq_if_create_grp(struct ifclassq *ifcq, uint8_t qset_idx, uint8_t flags); 349 extern void fq_if_set_grp_combined(struct ifclassq *ifcq, uint8_t qset_idx); 350 extern void fq_if_set_grp_separated(struct ifclassq *ifcq, uint8_t qset_idx); 351 extern fq_if_group_t *fq_if_find_grp(fq_if_t *fqs, uint8_t grp_idx); 352 #endif /* BSD_KERNEL_PRIVATE */ 353 354 #ifdef __cplusplus 355 } 356 #endif 357 358 #endif /* PRIVATE */ 359 #endif /* _NET_PKTSCHED_PKTSCHED_FQ_CODEL_H_ */ 360