| 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_CLASSQ_CLASSQ_FQ_CODEL_H |
| 30 | #define _NET_CLASSQ_CLASSQ_FQ_CODEL_H |
| 31 | #ifdef PRIVATE |
| 32 | #ifdef BSD_KERNEL_PRIVATE |
| 33 | #include <stdbool.h> |
| 34 | #include <sys/time.h> |
| 35 | #include <net/flowadv.h> |
| 36 | #include <net/classq/if_classq.h> |
| 37 | #if SKYWALK |
| 38 | #include <skywalk/os_skywalk_private.h> |
| 39 | #endif /* SKYWALK */ |
| 40 | |
| 41 | #ifdef __cplusplus |
| 42 | extern "C" { |
| 43 | #endif |
| 44 | |
| 45 | #define AQM_KTRACE_AON_FLOW_HIGH_DELAY AQMDBG_CODE(DBG_AQM_ALWAYSON, 0x001) |
| 46 | #define AQM_KTRACE_AON_THROTTLE AQMDBG_CODE(DBG_AQM_ALWAYSON, 0x002) |
| 47 | #define AQM_KTRACE_AON_FLOW_OVERWHELMING AQMDBG_CODE(DBG_AQM_ALWAYSON, 0x003) |
| 48 | #define AQM_KTRACE_AON_FLOW_DQ_STALL AQMDBG_CODE(DBG_AQM_ALWAYSON, 0x004) |
| 49 | |
| 50 | #define AQM_KTRACE_STATS_FLOW_ENQUEUE AQMDBG_CODE(DBG_AQM_STATS, 0x001) |
| 51 | #define AQM_KTRACE_STATS_FLOW_DEQUEUE AQMDBG_CODE(DBG_AQM_STATS, 0x002) |
| 52 | #define AQM_KTRACE_STATS_FLOW_CTL AQMDBG_CODE(DBG_AQM_STATS, 0x003) |
| 53 | #define AQM_KTRACE_STATS_FLOW_ALLOC AQMDBG_CODE(DBG_AQM_STATS, 0x004) |
| 54 | #define AQM_KTRACE_STATS_FLOW_DESTROY AQMDBG_CODE(DBG_AQM_STATS, 0x005) |
| 55 | #define AQM_KTRACE_STATS_FLOW_REPORT_CE AQMDBG_CODE(DBG_AQM_STATS, 0x006) |
| 56 | #define AQM_KTRACE_STATS_GET_QLEN AQMDBG_CODE(DBG_AQM_STATS, 0x007) |
| 57 | #define AQM_KTRACE_TX_NOT_READY AQMDBG_CODE(DBG_AQM_STATS, 0x008) |
| 58 | #define AQM_KTRACE_TX_PACEMAKER AQMDBG_CODE(DBG_AQM_STATS, 0x009) |
| 59 | |
| 60 | #define AQM_KTRACE_FQ_GRP_SC_IDX(_fq_) \ |
| 61 | ((_fq_)->fq_group->fqg_index << 4 | (_fq_)->fq_sc_index) |
| 62 | |
| 63 | #define FQ_MIN_FC_THRESHOLD_BYTES 7500 |
| 64 | #define FQ_IS_DELAY_HIGH(_fq_) ((_fq_)->fq_flags & FQF_DELAY_HIGH) |
| 65 | #define FQ_SET_DELAY_HIGH(_fq_) do { \ |
| 66 | if (!FQ_IS_DELAY_HIGH(_fq_)) { \ |
| 67 | KDBG(AQM_KTRACE_AON_FLOW_HIGH_DELAY | DBG_FUNC_START, \ |
| 68 | (_fq_)->fq_flowhash, AQM_KTRACE_FQ_GRP_SC_IDX((_fq_)), \ |
| 69 | (_fq_)->fq_bytes, (_fq_)->fq_min_qdelay); \ |
| 70 | } \ |
| 71 | (_fq_)->fq_flags |= FQF_DELAY_HIGH; \ |
| 72 | } while (0) |
| 73 | #define FQ_CLEAR_DELAY_HIGH(_fq_) do { \ |
| 74 | if (FQ_IS_DELAY_HIGH(_fq_)) { \ |
| 75 | KDBG(AQM_KTRACE_AON_FLOW_HIGH_DELAY | DBG_FUNC_END, \ |
| 76 | (_fq_)->fq_flowhash, AQM_KTRACE_FQ_GRP_SC_IDX((_fq_)), \ |
| 77 | (_fq_)->fq_bytes, (_fq_)->fq_min_qdelay); \ |
| 78 | } \ |
| 79 | (_fq_)->fq_flags &= ~FQF_DELAY_HIGH; \ |
| 80 | } while (0) |
| 81 | |
| 82 | #define FQ_IS_OVERWHELMING(_fq_) ((_fq_)->fq_flags & FQF_OVERWHELMING) |
| 83 | #define FQ_SET_OVERWHELMING(_fq_) do { \ |
| 84 | if (!FQ_IS_OVERWHELMING(_fq_)) { \ |
| 85 | KDBG(AQM_KTRACE_AON_FLOW_OVERWHELMING | DBG_FUNC_START, \ |
| 86 | (_fq_)->fq_flowhash, AQM_KTRACE_FQ_GRP_SC_IDX((_fq_)), \ |
| 87 | (_fq_)->fq_bytes, (_fq_)->fq_min_qdelay); \ |
| 88 | } \ |
| 89 | (_fq_)->fq_flags |= FQF_OVERWHELMING; \ |
| 90 | } while (0) |
| 91 | #define FQ_CLEAR_OVERWHELMING(_fq_) do { \ |
| 92 | if (FQ_IS_OVERWHELMING(_fq_)) { \ |
| 93 | KDBG(AQM_KTRACE_AON_FLOW_OVERWHELMING | DBG_FUNC_END, \ |
| 94 | (_fq_)->fq_flowhash, AQM_KTRACE_FQ_GRP_SC_IDX((_fq_)), \ |
| 95 | (_fq_)->fq_bytes, (_fq_)->fq_min_qdelay); \ |
| 96 | } \ |
| 97 | (_fq_)->fq_flags &= ~FQF_OVERWHELMING; \ |
| 98 | } while (0) |
| 99 | |
| 100 | /* |
| 101 | * time (in ns) the flow queue can stay in empty state. |
| 102 | */ |
| 103 | #define FQ_EMPTY_PURGE_DELAY (3ULL * 1000 * 1000 * 1000) |
| 104 | |
| 105 | /* |
| 106 | * maximum number of flow queues which can be purged during a dequeue. |
| 107 | */ |
| 108 | #define FQ_EMPTY_PURGE_MAX 4 |
| 109 | |
| 110 | #define FQ_INVALID_TX_TS UINT64_MAX |
| 111 | |
| 112 | struct flowq { |
| 113 | #pragma pack(push,1) |
| 114 | union { |
| 115 | MBUFQ_HEAD(mbufq_head) __mbufq; /* mbuf packet queue */ |
| 116 | #if SKYWALK |
| 117 | KPKTQ_HEAD(kpktq_head) __kpktq; /* skywalk packet queue */ |
| 118 | #endif /* SKYWALK */ |
| 119 | } __fq_pktq_u; |
| 120 | #pragma pack(pop) |
| 121 | uint32_t fq_flowhash; /* Flow hash */ |
| 122 | uint32_t fq_bytes; /* Number of bytes in the queue */ |
| 123 | int32_t fq_deficit; /* Deficit for scheduling */ |
| 124 | fq_if_group_t *fq_group; /* Back pointer to the group */ |
| 125 | #define FQF_FLOWCTL_CAPABLE 0x01 /* Use flow control instead of drop */ |
| 126 | #define FQF_DELAY_HIGH 0x02 /* Min delay is greater than target */ |
| 127 | #define FQF_NEW_FLOW 0x04 /* Currently on new flows queue */ |
| 128 | #define FQF_OLD_FLOW 0x08 /* Currently on old flows queue */ |
| 129 | #define FQF_FLOWCTL_ON 0x10 /* Currently flow controlled */ |
| 130 | #define FQF_EMPTY_FLOW 0x20 /* Currently on empty flows queue */ |
| 131 | #define FQF_OVERWHELMING 0x40 /* The largest flow when AQM hits queue limit */ |
| 132 | #define FQF_FRESH_FLOW 0x80 /* The flow queue has just been allocated */ |
| 133 | uint8_t fq_flags; /* flags */ |
| 134 | uint8_t fq_sc_index; /* service_class index */ |
| 135 | bool fq_in_dqlist; |
| 136 | fq_tfc_type_t fq_tfc_type; |
| 137 | uint8_t __fq_pad_uint8[4]; |
| 138 | uint64_t fq_min_qdelay; /* min queue delay for Codel */ |
| 139 | uint64_t fq_getqtime; /* last dequeue time */ |
| 140 | /* total pkt count since last congestion event report */ |
| 141 | uint32_t fq_pkts_since_last_report; |
| 142 | /* the next time that a paced packet is ready to go*/ |
| 143 | uint64_t fq_next_tx_time; |
| 144 | union { |
| 145 | uint64_t fq_updatetime; /* next update interval */ |
| 146 | /* empty list purge time (in nanoseconds) */ |
| 147 | uint64_t fq_empty_purge_time; |
| 148 | }; |
| 149 | SLIST_ENTRY(flowq) fq_hashlink; /* for flow queue hash table */ |
| 150 | /* |
| 151 | * flow queue will only be on either one of the lists. |
| 152 | */ |
| 153 | union { |
| 154 | STAILQ_ENTRY(flowq) fq_actlink; /* for new/old flow queues */ |
| 155 | /* entry on empty flow queue list */ |
| 156 | TAILQ_ENTRY(flowq) fq_empty_link; |
| 157 | }; |
| 158 | /* entry on dequeue flow list */ |
| 159 | STAILQ_ENTRY(flowq) fq_dqlink; |
| 160 | /* temporary packet queue for dequeued packets */ |
| 161 | classq_pkt_t fq_dq_head; |
| 162 | classq_pkt_t fq_dq_tail; |
| 163 | }; |
| 164 | |
| 165 | typedef struct flowq fq_t; |
| 166 | |
| 167 | #define FQF_FLOW_STATE_MASK (FQF_DELAY_HIGH | FQF_NEW_FLOW | \ |
| 168 | FQF_OLD_FLOW | FQF_FLOWCTL_ON | FQF_EMPTY_FLOW) |
| 169 | |
| 170 | #define fq_mbufq __fq_pktq_u.__mbufq |
| 171 | #if SKYWALK |
| 172 | #define fq_kpktq __fq_pktq_u.__kpktq |
| 173 | #endif /* SKYWALK */ |
| 174 | |
| 175 | #if SKYWALK |
| 176 | #define fq_empty(_q, _ptype) (((_ptype) == QP_MBUF) ? \ |
| 177 | MBUFQ_EMPTY(&(_q)->fq_mbufq) : KPKTQ_EMPTY(&(_q)->fq_kpktq)) |
| 178 | #else /* !SKYWALK */ |
| 179 | #define fq_empty(_q, _ptype) MBUFQ_EMPTY(&(_q)->fq_mbufq) |
| 180 | #endif /* !SKYWALK */ |
| 181 | |
| 182 | #if SKYWALK |
| 183 | #define fq_enqueue(_q, _h, _t, _c, _ptype) do { \ |
| 184 | switch (_ptype) { \ |
| 185 | case QP_MBUF: \ |
| 186 | ASSERT((_h).cp_ptype == QP_MBUF); \ |
| 187 | ASSERT((_t).cp_ptype == QP_MBUF); \ |
| 188 | MBUFQ_ENQUEUE_MULTI(&(_q)->fq_mbufq, (_h).cp_mbuf, \ |
| 189 | (_t).cp_mbuf); \ |
| 190 | MBUFQ_ADD_CRUMB_MULTI(&(_q)->fq_mbufq, (_h).cp_mbuf, \ |
| 191 | (_t).cp_mbuf, PKT_CRUMB_FQ_ENQUEUE); \ |
| 192 | break; \ |
| 193 | case QP_PACKET: \ |
| 194 | ASSERT((_h).cp_ptype == QP_PACKET); \ |
| 195 | ASSERT((_t).cp_ptype == QP_PACKET); \ |
| 196 | KPKTQ_ENQUEUE_MULTI(&(_q)->fq_kpktq, (_h).cp_kpkt, \ |
| 197 | (_t).cp_kpkt, (_c)); \ |
| 198 | break; \ |
| 199 | default: \ |
| 200 | VERIFY(0); \ |
| 201 | __builtin_unreachable(); \ |
| 202 | break; \ |
| 203 | } \ |
| 204 | } while (0) |
| 205 | #else /* !SKYWALK */ |
| 206 | #define fq_enqueue(_q, _h, _t, _c, _ptype) { \ |
| 207 | MBUFQ_ENQUEUE_MULTI(&(_q)->fq_mbufq, (_h).cp_mbuf, (_t).cp_mbuf); \ |
| 208 | MBUFQ_ADD_CRUMB_MULTI(&(_q)->fq_mbufq, (_h).cp_mbuf, \ |
| 209 | (_t).cp_mbuf, PKT_CRUMB_FQ_ENQUEUE); \ |
| 210 | } while (0) |
| 211 | #endif /* !SKYWALK */ |
| 212 | |
| 213 | #if SKYWALK |
| 214 | #define fq_dequeue(_q, _p, _ptype) do { \ |
| 215 | switch (_ptype) { \ |
| 216 | case QP_MBUF: { \ |
| 217 | MBUFQ_DEQUEUE(&(_q)->fq_mbufq, (_p)->cp_mbuf); \ |
| 218 | if (__probable((_p)->cp_mbuf != NULL)) { \ |
| 219 | CLASSQ_PKT_INIT_MBUF((_p), (_p)->cp_mbuf); \ |
| 220 | m_add_crumb((_p)->cp_mbuf, \ |
| 221 | PKT_CRUMB_FQ_DEQUEUE); \ |
| 222 | } \ |
| 223 | break; \ |
| 224 | } \ |
| 225 | case QP_PACKET: { \ |
| 226 | KPKTQ_DEQUEUE(&(_q)->fq_kpktq, (_p)->cp_kpkt); \ |
| 227 | if (__probable((_p)->cp_kpkt != NULL)) { \ |
| 228 | CLASSQ_PKT_INIT_PACKET((_p), (_p)->cp_kpkt); \ |
| 229 | } \ |
| 230 | break; \ |
| 231 | } \ |
| 232 | default: \ |
| 233 | VERIFY(0); \ |
| 234 | __builtin_unreachable(); \ |
| 235 | break; \ |
| 236 | } \ |
| 237 | } while (0) |
| 238 | #else /* !SKYWALK */ |
| 239 | #define fq_dequeue(_q, _p, _ptype) do { \ |
| 240 | MBUFQ_DEQUEUE(&(_q)->fq_mbufq, (_p)->cp_mbuf); \ |
| 241 | if (__probable((_p)->cp_mbuf != NULL)) { \ |
| 242 | CLASSQ_PKT_INIT_MBUF((_p), (_p)->cp_mbuf); \ |
| 243 | m_add_crumb((_p)->cp_mbuf, PKT_CRUMB_FQ_DEQUEUE); \ |
| 244 | } \ |
| 245 | } while (0) |
| 246 | #endif /* !SKYWALK */ |
| 247 | |
| 248 | struct fq_codel_sched_data; |
| 249 | struct fq_if_classq; |
| 250 | |
| 251 | /* Function definitions */ |
| 252 | extern void fq_codel_init(void); |
| 253 | extern fq_t *fq_alloc(classq_pkt_type_t); |
| 254 | extern void fq_destroy(fq_t *, classq_pkt_type_t); |
| 255 | extern int fq_addq(struct fq_codel_sched_data *, fq_if_group_t *, |
| 256 | pktsched_pkt_t *, struct fq_if_classq *); |
| 257 | extern void fq_getq_flow(struct fq_codel_sched_data *, fq_t *, |
| 258 | pktsched_pkt_t *, uint64_t now); |
| 259 | extern void fq_codel_dequeue(fq_if_t *fqs, fq_t *fq, |
| 260 | pktsched_pkt_t *pkt, uint64_t now); |
| 261 | extern void fq_getq_flow_internal(struct fq_codel_sched_data *, |
| 262 | fq_t *, pktsched_pkt_t *); |
| 263 | extern void fq_head_drop(struct fq_codel_sched_data *, fq_t *); |
| 264 | extern boolean_t fq_tx_time_ready(fq_if_t *fqs, fq_t *fq, uint64_t now, |
| 265 | uint64_t *ready_time); |
| 266 | |
| 267 | #ifdef __cplusplus |
| 268 | } |
| 269 | #endif |
| 270 | #endif /* BSD_KERNEL_PRIVATE */ |
| 271 | #endif /* PRIVATE */ |
| 272 | #endif /* _NET_CLASSQ_CLASSQ_FQ_CODEL_H */ |
| 273 | |