]>
Commit | Line | Data |
---|---|---|
316670eb | 1 | /* |
f427ee49 | 2 | * Copyright (c) 2011-2020 Apple Inc. All rights reserved. |
316670eb A |
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 | #include <sys/cdefs.h> | |
30 | #include <sys/param.h> | |
31 | #include <sys/mbuf.h> | |
32 | #include <sys/errno.h> | |
33 | #include <sys/random.h> | |
34 | #include <sys/kernel_types.h> | |
35 | #include <sys/sysctl.h> | |
36 | ||
37 | #include <kern/zalloc.h> | |
38 | ||
39 | #include <net/if.h> | |
40 | #include <net/net_osdep.h> | |
41 | #include <net/classq/classq.h> | |
39037602 | 42 | #include <pexpert/pexpert.h> |
316670eb | 43 | #include <net/classq/classq_sfb.h> |
a39ff7e2 | 44 | #include <net/classq/classq_fq_codel.h> |
316670eb | 45 | #include <net/pktsched/pktsched.h> |
39037602 | 46 | #include <net/pktsched/pktsched_fq_codel.h> |
f427ee49 | 47 | #include <net/flowadv.h> |
316670eb A |
48 | |
49 | #include <libkern/libkern.h> | |
50 | ||
316670eb A |
51 | |
52 | static errno_t ifclassq_dequeue_common(struct ifclassq *, mbuf_svc_class_t, | |
cb323159 A |
53 | u_int32_t, u_int32_t, classq_pkt_t *, classq_pkt_t *, u_int32_t *, |
54 | u_int32_t *, boolean_t); | |
55 | static void ifclassq_tbr_dequeue_common(struct ifclassq *, mbuf_svc_class_t, | |
56 | boolean_t, classq_pkt_t *); | |
316670eb | 57 | |
39037602 | 58 | static u_int64_t ifclassq_target_qdelay = 0; |
0a7de745 | 59 | SYSCTL_QUAD(_net_classq, OID_AUTO, target_qdelay, CTLFLAG_RW | CTLFLAG_LOCKED, |
39037602 A |
60 | &ifclassq_target_qdelay, "target queue delay in nanoseconds"); |
61 | ||
62 | static u_int64_t ifclassq_update_interval = 0; | |
63 | SYSCTL_QUAD(_net_classq, OID_AUTO, update_interval, | |
0a7de745 | 64 | CTLFLAG_RW | CTLFLAG_LOCKED, &ifclassq_update_interval, |
39037602 A |
65 | "update interval in nanoseconds"); |
66 | ||
316670eb A |
67 | void |
68 | classq_init(void) | |
69 | { | |
70 | _CASSERT(MBUF_TC_BE == 0); | |
71 | _CASSERT(MBUF_SC_BE == 0); | |
72 | _CASSERT(IFCQ_SC_MAX == MBUF_SC_MAX_CLASSES); | |
73 | ||
f427ee49 | 74 | fq_codel_init(); |
316670eb A |
75 | } |
76 | ||
77 | int | |
78 | ifclassq_setup(struct ifnet *ifp, u_int32_t sflags, boolean_t reuse) | |
79 | { | |
80 | #pragma unused(reuse) | |
81 | struct ifclassq *ifq = &ifp->if_snd; | |
82 | int err = 0; | |
83 | ||
84 | IFCQ_LOCK(ifq); | |
85 | VERIFY(IFCQ_IS_EMPTY(ifq)); | |
86 | ifq->ifcq_ifp = ifp; | |
87 | IFCQ_LEN(ifq) = 0; | |
3e170ce0 | 88 | IFCQ_BYTES(ifq) = 0; |
0a7de745 A |
89 | bzero(&ifq->ifcq_xmitcnt, sizeof(ifq->ifcq_xmitcnt)); |
90 | bzero(&ifq->ifcq_dropcnt, sizeof(ifq->ifcq_dropcnt)); | |
316670eb A |
91 | |
92 | VERIFY(!IFCQ_TBR_IS_ENABLED(ifq)); | |
93 | VERIFY(ifq->ifcq_type == PKTSCHEDT_NONE); | |
94 | VERIFY(ifq->ifcq_flags == 0); | |
95 | VERIFY(ifq->ifcq_sflags == 0); | |
96 | VERIFY(ifq->ifcq_disc == NULL); | |
316670eb A |
97 | |
98 | if (ifp->if_eflags & IFEF_TXSTART) { | |
99 | u_int32_t maxlen = 0; | |
100 | ||
0a7de745 | 101 | if ((maxlen = IFCQ_MAXLEN(ifq)) == 0) { |
316670eb | 102 | maxlen = if_sndq_maxlen; |
0a7de745 | 103 | } |
316670eb A |
104 | IFCQ_SET_MAXLEN(ifq, maxlen); |
105 | ||
fe8ab488 A |
106 | if (IFCQ_MAXLEN(ifq) != if_sndq_maxlen && |
107 | IFCQ_TARGET_QDELAY(ifq) == 0) { | |
108 | /* | |
109 | * Choose static queues because the interface has | |
110 | * maximum queue size set | |
111 | */ | |
112 | sflags &= ~PKTSCHEDF_QALG_DELAYBASED; | |
113 | } | |
316670eb A |
114 | ifq->ifcq_sflags = sflags; |
115 | err = ifclassq_pktsched_setup(ifq); | |
0a7de745 | 116 | if (err == 0) { |
316670eb | 117 | ifq->ifcq_flags = (IFCQF_READY | IFCQF_ENABLED); |
0a7de745 | 118 | } |
316670eb | 119 | } |
316670eb | 120 | IFCQ_UNLOCK(ifq); |
0a7de745 | 121 | return err; |
316670eb A |
122 | } |
123 | ||
124 | void | |
125 | ifclassq_teardown(struct ifnet *ifp) | |
126 | { | |
127 | struct ifclassq *ifq = &ifp->if_snd; | |
128 | ||
129 | IFCQ_LOCK(ifq); | |
316670eb A |
130 | |
131 | if (IFCQ_IS_READY(ifq)) { | |
132 | if (IFCQ_TBR_IS_ENABLED(ifq)) { | |
cb323159 | 133 | struct tb_profile tb = { .rate = 0, .percent = 0, .depth = 0 }; |
316670eb A |
134 | (void) ifclassq_tbr_set(ifq, &tb, FALSE); |
135 | } | |
f427ee49 | 136 | pktsched_teardown(ifq); |
316670eb A |
137 | ifq->ifcq_flags = 0; |
138 | } | |
139 | ifq->ifcq_sflags = 0; | |
140 | ||
141 | VERIFY(IFCQ_IS_EMPTY(ifq)); | |
142 | VERIFY(!IFCQ_TBR_IS_ENABLED(ifq)); | |
143 | VERIFY(ifq->ifcq_type == PKTSCHEDT_NONE); | |
144 | VERIFY(ifq->ifcq_flags == 0); | |
145 | VERIFY(ifq->ifcq_sflags == 0); | |
146 | VERIFY(ifq->ifcq_disc == NULL); | |
316670eb | 147 | IFCQ_LEN(ifq) = 0; |
3e170ce0 | 148 | IFCQ_BYTES(ifq) = 0; |
316670eb | 149 | IFCQ_MAXLEN(ifq) = 0; |
0a7de745 A |
150 | bzero(&ifq->ifcq_xmitcnt, sizeof(ifq->ifcq_xmitcnt)); |
151 | bzero(&ifq->ifcq_dropcnt, sizeof(ifq->ifcq_dropcnt)); | |
316670eb A |
152 | |
153 | IFCQ_UNLOCK(ifq); | |
154 | } | |
155 | ||
156 | int | |
157 | ifclassq_pktsched_setup(struct ifclassq *ifq) | |
158 | { | |
159 | struct ifnet *ifp = ifq->ifcq_ifp; | |
5ba3f43e | 160 | classq_pkt_type_t ptype = QP_MBUF; |
316670eb A |
161 | int err = 0; |
162 | ||
163 | IFCQ_LOCK_ASSERT_HELD(ifq); | |
164 | VERIFY(ifp->if_eflags & IFEF_TXSTART); | |
165 | ||
f427ee49 | 166 | err = pktsched_setup(ifq, PKTSCHEDT_FQ_CODEL, ifq->ifcq_sflags, ptype); |
316670eb | 167 | |
0a7de745 | 168 | return err; |
316670eb A |
169 | } |
170 | ||
171 | void | |
172 | ifclassq_set_maxlen(struct ifclassq *ifq, u_int32_t maxqlen) | |
173 | { | |
174 | IFCQ_LOCK(ifq); | |
0a7de745 | 175 | if (maxqlen == 0) { |
316670eb | 176 | maxqlen = if_sndq_maxlen; |
0a7de745 | 177 | } |
316670eb A |
178 | IFCQ_SET_MAXLEN(ifq, maxqlen); |
179 | IFCQ_UNLOCK(ifq); | |
180 | } | |
181 | ||
182 | u_int32_t | |
183 | ifclassq_get_maxlen(struct ifclassq *ifq) | |
184 | { | |
0a7de745 | 185 | return IFCQ_MAXLEN(ifq); |
316670eb A |
186 | } |
187 | ||
39236c6e A |
188 | int |
189 | ifclassq_get_len(struct ifclassq *ifq, mbuf_svc_class_t sc, u_int32_t *packets, | |
190 | u_int32_t *bytes) | |
316670eb | 191 | { |
39236c6e A |
192 | int err = 0; |
193 | ||
194 | IFCQ_LOCK(ifq); | |
195 | if (sc == MBUF_SC_UNSPEC) { | |
196 | VERIFY(packets != NULL); | |
197 | *packets = IFCQ_LEN(ifq); | |
198 | } else { | |
f427ee49 A |
199 | cqrq_stat_sc_t req = { sc, 0, 0 }; |
200 | ||
39236c6e A |
201 | VERIFY(MBUF_VALID_SC(sc)); |
202 | VERIFY(packets != NULL && bytes != NULL); | |
f427ee49 A |
203 | |
204 | err = fq_if_request_classq(ifq, CLASSQRQ_STAT_SC, &req); | |
205 | if (packets != NULL) { | |
206 | *packets = req.packets; | |
207 | } | |
208 | if (bytes != NULL) { | |
209 | *bytes = req.bytes; | |
210 | } | |
39236c6e A |
211 | } |
212 | IFCQ_UNLOCK(ifq); | |
213 | ||
cb323159 | 214 | |
0a7de745 | 215 | return err; |
316670eb A |
216 | } |
217 | ||
5ba3f43e A |
218 | inline void |
219 | ifclassq_set_packet_metadata(struct ifclassq *ifq, struct ifnet *ifp, | |
cb323159 | 220 | classq_pkt_t *p) |
5ba3f43e | 221 | { |
0a7de745 | 222 | if (!IFNET_IS_CELLULAR(ifp)) { |
5ba3f43e | 223 | return; |
0a7de745 | 224 | } |
5ba3f43e | 225 | |
cb323159 | 226 | switch (p->cp_ptype) { |
5ba3f43e | 227 | case QP_MBUF: { |
cb323159 | 228 | struct mbuf *m = p->cp_mbuf; |
5ba3f43e A |
229 | m->m_pkthdr.pkt_flags |= PKTF_VALID_UNSENT_DATA; |
230 | m->m_pkthdr.bufstatus_if = IFCQ_BYTES(ifq); | |
f427ee49 | 231 | m->m_pkthdr.bufstatus_sndbuf = (uint32_t)ifp->if_sndbyte_unsent; |
5ba3f43e A |
232 | break; |
233 | } | |
234 | ||
235 | ||
236 | default: | |
237 | VERIFY(0); | |
238 | /* NOTREACHED */ | |
cb323159 | 239 | __builtin_unreachable(); |
5ba3f43e A |
240 | } |
241 | } | |
242 | ||
316670eb | 243 | errno_t |
f427ee49 A |
244 | ifclassq_enqueue(struct ifclassq *ifq, classq_pkt_t *head, classq_pkt_t *tail, |
245 | u_int32_t cnt, u_int32_t bytes, boolean_t *pdrop) | |
316670eb | 246 | { |
f427ee49 | 247 | return fq_if_enqueue_classq(ifq, head, tail, cnt, bytes, pdrop); |
316670eb A |
248 | } |
249 | ||
250 | errno_t | |
39037602 | 251 | ifclassq_dequeue(struct ifclassq *ifq, u_int32_t pkt_limit, |
cb323159 A |
252 | u_int32_t byte_limit, classq_pkt_t *head, classq_pkt_t *tail, |
253 | u_int32_t *cnt, u_int32_t *len) | |
316670eb | 254 | { |
0a7de745 | 255 | return ifclassq_dequeue_common(ifq, MBUF_SC_UNSPEC, pkt_limit, |
cb323159 | 256 | byte_limit, head, tail, cnt, len, FALSE); |
316670eb A |
257 | } |
258 | ||
259 | errno_t | |
260 | ifclassq_dequeue_sc(struct ifclassq *ifq, mbuf_svc_class_t sc, | |
cb323159 A |
261 | u_int32_t pkt_limit, u_int32_t byte_limit, classq_pkt_t *head, |
262 | classq_pkt_t *tail, u_int32_t *cnt, u_int32_t *len) | |
316670eb | 263 | { |
0a7de745 | 264 | return ifclassq_dequeue_common(ifq, sc, pkt_limit, byte_limit, |
cb323159 | 265 | head, tail, cnt, len, TRUE); |
316670eb A |
266 | } |
267 | ||
268 | static errno_t | |
cb323159 A |
269 | ifclassq_dequeue_common_default(struct ifclassq *ifq, mbuf_svc_class_t sc, |
270 | u_int32_t pkt_limit, u_int32_t byte_limit, classq_pkt_t *head, | |
271 | classq_pkt_t *tail, u_int32_t *cnt, u_int32_t *len, boolean_t drvmgt) | |
316670eb A |
272 | { |
273 | struct ifnet *ifp = ifq->ifcq_ifp; | |
f427ee49 | 274 | u_int32_t i = 0, l = 0; |
cb323159 A |
275 | classq_pkt_t first = CLASSQ_PKT_INITIALIZER(first); |
276 | classq_pkt_t last = CLASSQ_PKT_INITIALIZER(last); | |
316670eb A |
277 | |
278 | VERIFY(!drvmgt || MBUF_VALID_SC(sc)); | |
279 | ||
0a7de745 | 280 | if (IFCQ_TBR_IS_ENABLED(ifq)) { |
5ba3f43e | 281 | goto dequeue_loop; |
0a7de745 | 282 | } |
5ba3f43e | 283 | |
39037602 A |
284 | /* |
285 | * If the scheduler support dequeueing multiple packets at the | |
286 | * same time, call that one instead. | |
287 | */ | |
f427ee49 | 288 | if (drvmgt) { |
5ba3f43e | 289 | int err; |
39037602 | 290 | |
f427ee49 A |
291 | IFCQ_LOCK_SPIN(ifq); |
292 | err = fq_if_dequeue_sc_classq_multi(ifq, sc, pkt_limit, | |
cb323159 | 293 | byte_limit, head, tail, cnt, len); |
5ba3f43e A |
294 | IFCQ_UNLOCK(ifq); |
295 | ||
cb323159 | 296 | if (err == 0 && head->cp_mbuf == NULL) { |
5ba3f43e | 297 | err = EAGAIN; |
0a7de745 A |
298 | } |
299 | return err; | |
f427ee49 | 300 | } else { |
39037602 | 301 | int err; |
5ba3f43e | 302 | |
f427ee49 A |
303 | IFCQ_LOCK_SPIN(ifq); |
304 | err = fq_if_dequeue_classq_multi(ifq, pkt_limit, byte_limit, | |
cb323159 | 305 | head, tail, cnt, len); |
39037602 A |
306 | IFCQ_UNLOCK(ifq); |
307 | ||
cb323159 | 308 | if (err == 0 && head->cp_mbuf == NULL) { |
39037602 | 309 | err = EAGAIN; |
0a7de745 A |
310 | } |
311 | return err; | |
39037602 A |
312 | } |
313 | ||
5ba3f43e | 314 | dequeue_loop: |
316670eb | 315 | |
f427ee49 | 316 | IFCQ_LOCK_SPIN(ifq); |
316670eb | 317 | |
39037602 | 318 | while (i < pkt_limit && l < byte_limit) { |
316670eb | 319 | if (drvmgt) { |
0a7de745 | 320 | if (IFCQ_TBR_IS_ENABLED(ifq)) { |
cb323159 | 321 | IFCQ_TBR_DEQUEUE_SC(ifq, sc, head); |
0a7de745 | 322 | } else { |
f427ee49 | 323 | fq_if_dequeue_sc_classq(ifq, sc, head); |
0a7de745 | 324 | } |
316670eb | 325 | } else { |
0a7de745 | 326 | if (IFCQ_TBR_IS_ENABLED(ifq)) { |
cb323159 | 327 | IFCQ_TBR_DEQUEUE(ifq, head); |
0a7de745 | 328 | } else { |
f427ee49 | 329 | fq_if_dequeue_classq(ifq, head); |
0a7de745 | 330 | } |
316670eb | 331 | } |
316670eb | 332 | |
cb323159 | 333 | if (head->cp_mbuf == NULL) { |
316670eb | 334 | break; |
0a7de745 | 335 | } |
316670eb | 336 | |
cb323159 A |
337 | if (first.cp_mbuf == NULL) { |
338 | first = *head; | |
339 | } | |
340 | ||
341 | switch (head->cp_ptype) { | |
5ba3f43e | 342 | case QP_MBUF: |
cb323159 A |
343 | head->cp_mbuf->m_nextpkt = NULL; |
344 | l += head->cp_mbuf->m_pkthdr.len; | |
345 | ifclassq_set_packet_metadata(ifq, ifp, head); | |
346 | if (last.cp_mbuf != NULL) { | |
347 | last.cp_mbuf->m_nextpkt = head->cp_mbuf; | |
348 | } | |
5ba3f43e | 349 | break; |
316670eb | 350 | |
316670eb | 351 | |
5ba3f43e A |
352 | default: |
353 | VERIFY(0); | |
354 | /* NOTREACHED */ | |
cb323159 | 355 | __builtin_unreachable(); |
3e170ce0 | 356 | } |
5ba3f43e | 357 | |
cb323159 | 358 | last = *head; |
316670eb A |
359 | i++; |
360 | } | |
361 | ||
362 | IFCQ_UNLOCK(ifq); | |
363 | ||
0a7de745 | 364 | if (tail != NULL) { |
316670eb | 365 | *tail = last; |
0a7de745 A |
366 | } |
367 | if (cnt != NULL) { | |
316670eb | 368 | *cnt = i; |
0a7de745 A |
369 | } |
370 | if (len != NULL) { | |
316670eb | 371 | *len = l; |
0a7de745 | 372 | } |
316670eb | 373 | |
cb323159 A |
374 | *head = first; |
375 | return (first.cp_mbuf != NULL) ? 0 : EAGAIN; | |
376 | } | |
377 | ||
378 | static errno_t | |
379 | ifclassq_dequeue_common(struct ifclassq *ifq, mbuf_svc_class_t sc, | |
380 | u_int32_t pkt_limit, u_int32_t byte_limit, classq_pkt_t *head, | |
381 | classq_pkt_t *tail, u_int32_t *cnt, u_int32_t *len, boolean_t drvmgt) | |
382 | { | |
383 | return ifclassq_dequeue_common_default(ifq, sc, | |
384 | pkt_limit, byte_limit, head, tail, cnt, len, drvmgt); | |
316670eb A |
385 | } |
386 | ||
316670eb A |
387 | void |
388 | ifclassq_update(struct ifclassq *ifq, cqev_t ev) | |
389 | { | |
390 | IFCQ_LOCK_ASSERT_HELD(ifq); | |
391 | VERIFY(IFCQ_IS_READY(ifq)); | |
f427ee49 | 392 | fq_if_request_classq(ifq, CLASSQRQ_EVENT, (void *)ev); |
316670eb A |
393 | } |
394 | ||
395 | int | |
f427ee49 | 396 | ifclassq_attach(struct ifclassq *ifq, u_int32_t type, void *discipline) |
316670eb A |
397 | { |
398 | IFCQ_LOCK_ASSERT_HELD(ifq); | |
399 | ||
400 | VERIFY(ifq->ifcq_disc == NULL); | |
316670eb A |
401 | |
402 | ifq->ifcq_type = type; | |
403 | ifq->ifcq_disc = discipline; | |
316670eb | 404 | |
0a7de745 | 405 | return 0; |
316670eb A |
406 | } |
407 | ||
f427ee49 | 408 | void |
316670eb A |
409 | ifclassq_detach(struct ifclassq *ifq) |
410 | { | |
411 | IFCQ_LOCK_ASSERT_HELD(ifq); | |
412 | ||
413 | VERIFY(ifq->ifcq_disc == NULL); | |
414 | ||
415 | ifq->ifcq_type = PKTSCHEDT_NONE; | |
316670eb A |
416 | } |
417 | ||
418 | int | |
419 | ifclassq_getqstats(struct ifclassq *ifq, u_int32_t qid, void *ubuf, | |
420 | u_int32_t *nbytes) | |
421 | { | |
422 | struct if_ifclassq_stats *ifqs; | |
423 | int err; | |
424 | ||
0a7de745 A |
425 | if (*nbytes < sizeof(*ifqs)) { |
426 | return EINVAL; | |
427 | } | |
316670eb | 428 | |
0a7de745 A |
429 | ifqs = _MALLOC(sizeof(*ifqs), M_TEMP, M_WAITOK | M_ZERO); |
430 | if (ifqs == NULL) { | |
431 | return ENOMEM; | |
432 | } | |
316670eb A |
433 | |
434 | IFCQ_LOCK(ifq); | |
435 | if (!IFCQ_IS_READY(ifq)) { | |
436 | IFCQ_UNLOCK(ifq); | |
437 | _FREE(ifqs, M_TEMP); | |
0a7de745 | 438 | return ENXIO; |
316670eb A |
439 | } |
440 | ||
441 | ifqs->ifqs_len = IFCQ_LEN(ifq); | |
442 | ifqs->ifqs_maxlen = IFCQ_MAXLEN(ifq); | |
443 | *(&ifqs->ifqs_xmitcnt) = *(&ifq->ifcq_xmitcnt); | |
444 | *(&ifqs->ifqs_dropcnt) = *(&ifq->ifcq_dropcnt); | |
445 | ifqs->ifqs_scheduler = ifq->ifcq_type; | |
446 | ||
447 | err = pktsched_getqstats(ifq, qid, ifqs); | |
448 | IFCQ_UNLOCK(ifq); | |
449 | ||
450 | if (err == 0 && (err = copyout((caddr_t)ifqs, | |
0a7de745 A |
451 | (user_addr_t)(uintptr_t)ubuf, sizeof(*ifqs))) == 0) { |
452 | *nbytes = sizeof(*ifqs); | |
453 | } | |
316670eb A |
454 | |
455 | _FREE(ifqs, M_TEMP); | |
456 | ||
0a7de745 | 457 | return err; |
316670eb A |
458 | } |
459 | ||
460 | const char * | |
461 | ifclassq_ev2str(cqev_t ev) | |
462 | { | |
463 | const char *c; | |
464 | ||
465 | switch (ev) { | |
39236c6e A |
466 | case CLASSQ_EV_LINK_BANDWIDTH: |
467 | c = "LINK_BANDWIDTH"; | |
468 | break; | |
469 | ||
470 | case CLASSQ_EV_LINK_LATENCY: | |
471 | c = "LINK_LATENCY"; | |
316670eb A |
472 | break; |
473 | ||
474 | case CLASSQ_EV_LINK_MTU: | |
475 | c = "LINK_MTU"; | |
476 | break; | |
477 | ||
478 | case CLASSQ_EV_LINK_UP: | |
479 | c = "LINK_UP"; | |
480 | break; | |
481 | ||
482 | case CLASSQ_EV_LINK_DOWN: | |
483 | c = "LINK_DOWN"; | |
484 | break; | |
485 | ||
486 | default: | |
487 | c = "UNKNOWN"; | |
488 | break; | |
489 | } | |
490 | ||
0a7de745 | 491 | return c; |
316670eb A |
492 | } |
493 | ||
494 | /* | |
495 | * internal representation of token bucket parameters | |
496 | * rate: byte_per_unittime << 32 | |
497 | * (((bits_per_sec) / 8) << 32) / machclk_freq | |
498 | * depth: byte << 32 | |
499 | * | |
500 | */ | |
0a7de745 A |
501 | #define TBR_SHIFT 32 |
502 | #define TBR_SCALE(x) ((int64_t)(x) << TBR_SHIFT) | |
503 | #define TBR_UNSCALE(x) ((x) >> TBR_SHIFT) | |
316670eb | 504 | |
cb323159 A |
505 | void |
506 | ifclassq_tbr_dequeue(struct ifclassq *ifq, classq_pkt_t *pkt) | |
316670eb | 507 | { |
cb323159 | 508 | ifclassq_tbr_dequeue_common(ifq, MBUF_SC_UNSPEC, FALSE, pkt); |
316670eb A |
509 | } |
510 | ||
cb323159 | 511 | void |
5ba3f43e | 512 | ifclassq_tbr_dequeue_sc(struct ifclassq *ifq, mbuf_svc_class_t sc, |
cb323159 | 513 | classq_pkt_t *pkt) |
316670eb | 514 | { |
cb323159 | 515 | ifclassq_tbr_dequeue_common(ifq, sc, TRUE, pkt); |
316670eb A |
516 | } |
517 | ||
cb323159 | 518 | static void |
5ba3f43e | 519 | ifclassq_tbr_dequeue_common(struct ifclassq *ifq, mbuf_svc_class_t sc, |
cb323159 | 520 | boolean_t drvmgt, classq_pkt_t *pkt) |
316670eb A |
521 | { |
522 | struct tb_regulator *tbr; | |
316670eb A |
523 | int64_t interval; |
524 | u_int64_t now; | |
525 | ||
526 | IFCQ_LOCK_ASSERT_HELD(ifq); | |
527 | ||
528 | VERIFY(!drvmgt || MBUF_VALID_SC(sc)); | |
529 | VERIFY(IFCQ_TBR_IS_ENABLED(ifq)); | |
530 | ||
cb323159 | 531 | *pkt = CLASSQ_PKT_INITIALIZER(*pkt); |
316670eb | 532 | tbr = &ifq->ifcq_tbr; |
5ba3f43e A |
533 | /* update token only when it is negative */ |
534 | if (tbr->tbr_token <= 0) { | |
535 | now = read_machclk(); | |
536 | interval = now - tbr->tbr_last; | |
537 | if (interval >= tbr->tbr_filluptime) { | |
538 | tbr->tbr_token = tbr->tbr_depth; | |
539 | } else { | |
540 | tbr->tbr_token += interval * tbr->tbr_rate; | |
0a7de745 | 541 | if (tbr->tbr_token > tbr->tbr_depth) { |
316670eb | 542 | tbr->tbr_token = tbr->tbr_depth; |
0a7de745 | 543 | } |
316670eb | 544 | } |
5ba3f43e | 545 | tbr->tbr_last = now; |
316670eb | 546 | } |
5ba3f43e | 547 | /* if token is still negative, don't allow dequeue */ |
0a7de745 | 548 | if (tbr->tbr_token <= 0) { |
cb323159 | 549 | return; |
0a7de745 | 550 | } |
316670eb A |
551 | |
552 | /* | |
553 | * ifclassq takes precedence over ALTQ queue; | |
554 | * ifcq_drain count is adjusted by the caller. | |
555 | */ | |
0a7de745 | 556 | if (drvmgt) { |
f427ee49 | 557 | fq_if_dequeue_sc_classq(ifq, sc, pkt); |
0a7de745 | 558 | } else { |
f427ee49 | 559 | fq_if_dequeue_classq(ifq, pkt); |
0a7de745 | 560 | } |
5ba3f43e | 561 | |
cb323159 A |
562 | if (pkt->cp_mbuf != NULL) { |
563 | switch (pkt->cp_ptype) { | |
5ba3f43e | 564 | case QP_MBUF: |
cb323159 | 565 | tbr->tbr_token -= TBR_SCALE(m_pktlen(pkt->cp_mbuf)); |
5ba3f43e A |
566 | break; |
567 | ||
568 | ||
569 | default: | |
570 | VERIFY(0); | |
571 | /* NOTREACHED */ | |
316670eb A |
572 | } |
573 | } | |
316670eb A |
574 | } |
575 | ||
576 | /* | |
577 | * set a token bucket regulator. | |
578 | * if the specified rate is zero, the token bucket regulator is deleted. | |
579 | */ | |
580 | int | |
581 | ifclassq_tbr_set(struct ifclassq *ifq, struct tb_profile *profile, | |
582 | boolean_t update) | |
583 | { | |
584 | struct tb_regulator *tbr; | |
585 | struct ifnet *ifp = ifq->ifcq_ifp; | |
586 | u_int64_t rate, old_rate; | |
587 | ||
588 | IFCQ_LOCK_ASSERT_HELD(ifq); | |
589 | VERIFY(IFCQ_IS_READY(ifq)); | |
590 | ||
591 | VERIFY(machclk_freq != 0); | |
592 | ||
593 | tbr = &ifq->ifcq_tbr; | |
594 | old_rate = tbr->tbr_rate_raw; | |
595 | ||
596 | rate = profile->rate; | |
597 | if (profile->percent > 0) { | |
598 | u_int64_t eff_rate; | |
599 | ||
0a7de745 A |
600 | if (profile->percent > 100) { |
601 | return EINVAL; | |
602 | } | |
603 | if ((eff_rate = ifp->if_output_bw.eff_bw) == 0) { | |
604 | return ENODEV; | |
605 | } | |
316670eb A |
606 | rate = (eff_rate * profile->percent) / 100; |
607 | } | |
608 | ||
609 | if (rate == 0) { | |
0a7de745 | 610 | if (!IFCQ_TBR_IS_ENABLED(ifq)) { |
cb323159 | 611 | return 0; |
0a7de745 | 612 | } |
316670eb | 613 | |
0a7de745 | 614 | if (pktsched_verbose) { |
316670eb | 615 | printf("%s: TBR disabled\n", if_name(ifp)); |
0a7de745 | 616 | } |
316670eb A |
617 | |
618 | /* disable this TBR */ | |
619 | ifq->ifcq_flags &= ~IFCQF_TBR; | |
0a7de745 | 620 | bzero(tbr, sizeof(*tbr)); |
316670eb | 621 | ifnet_set_start_cycle(ifp, NULL); |
0a7de745 | 622 | if (update) { |
39236c6e | 623 | ifclassq_update(ifq, CLASSQ_EV_LINK_BANDWIDTH); |
0a7de745 A |
624 | } |
625 | return 0; | |
316670eb A |
626 | } |
627 | ||
628 | if (pktsched_verbose) { | |
629 | printf("%s: TBR %s (rate %llu bps depth %u)\n", if_name(ifp), | |
630 | (ifq->ifcq_flags & IFCQF_TBR) ? "reconfigured" : | |
631 | "enabled", rate, profile->depth); | |
632 | } | |
633 | ||
634 | /* set the new TBR */ | |
0a7de745 | 635 | bzero(tbr, sizeof(*tbr)); |
316670eb A |
636 | tbr->tbr_rate_raw = rate; |
637 | tbr->tbr_percent = profile->percent; | |
638 | ifq->ifcq_flags |= IFCQF_TBR; | |
639 | ||
640 | /* | |
641 | * Note that the TBR fill up time (hence the ifnet restart time) | |
642 | * is directly related to the specified TBR depth. The ideal | |
643 | * depth value should be computed such that the interval time | |
644 | * between each successive wakeup is adequately spaced apart, | |
645 | * in order to reduce scheduling overheads. A target interval | |
646 | * of 10 ms seems to provide good performance balance. This can be | |
647 | * overridden by specifying the depth profile. Values smaller than | |
648 | * the ideal depth will reduce delay at the expense of CPU cycles. | |
649 | */ | |
650 | tbr->tbr_rate = TBR_SCALE(rate / 8) / machclk_freq; | |
651 | if (tbr->tbr_rate > 0) { | |
652 | u_int32_t mtu = ifp->if_mtu; | |
653 | int64_t ival, idepth = 0; | |
654 | int i; | |
655 | ||
0a7de745 | 656 | if (mtu < IF_MINMTU) { |
316670eb | 657 | mtu = IF_MINMTU; |
0a7de745 | 658 | } |
316670eb A |
659 | |
660 | ival = pktsched_nsecs_to_abstime(10 * NSEC_PER_MSEC); /* 10ms */ | |
661 | ||
0a7de745 | 662 | for (i = 1;; i++) { |
316670eb | 663 | idepth = TBR_SCALE(i * mtu); |
0a7de745 | 664 | if ((idepth / tbr->tbr_rate) > ival) { |
316670eb | 665 | break; |
0a7de745 | 666 | } |
316670eb A |
667 | } |
668 | VERIFY(idepth > 0); | |
669 | ||
670 | tbr->tbr_depth = TBR_SCALE(profile->depth); | |
671 | if (tbr->tbr_depth == 0) { | |
672 | tbr->tbr_filluptime = idepth / tbr->tbr_rate; | |
673 | /* a little fudge factor to get closer to rate */ | |
674 | tbr->tbr_depth = idepth + (idepth >> 3); | |
675 | } else { | |
676 | tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate; | |
677 | } | |
678 | } else { | |
679 | tbr->tbr_depth = TBR_SCALE(profile->depth); | |
680 | tbr->tbr_filluptime = 0xffffffffffffffffLL; | |
681 | } | |
682 | tbr->tbr_token = tbr->tbr_depth; | |
683 | tbr->tbr_last = read_machclk(); | |
316670eb A |
684 | |
685 | if (tbr->tbr_rate > 0 && (ifp->if_flags & IFF_UP)) { | |
686 | struct timespec ts = | |
f427ee49 | 687 | { 0, (long)pktsched_abs_to_nsecs(tbr->tbr_filluptime) }; |
316670eb A |
688 | if (pktsched_verbose) { |
689 | printf("%s: TBR calculated tokens %lld " | |
690 | "filluptime %llu ns\n", if_name(ifp), | |
691 | TBR_UNSCALE(tbr->tbr_token), | |
692 | pktsched_abs_to_nsecs(tbr->tbr_filluptime)); | |
693 | } | |
694 | ifnet_set_start_cycle(ifp, &ts); | |
695 | } else { | |
696 | if (pktsched_verbose) { | |
697 | if (tbr->tbr_rate == 0) { | |
698 | printf("%s: TBR calculated tokens %lld " | |
699 | "infinite filluptime\n", if_name(ifp), | |
700 | TBR_UNSCALE(tbr->tbr_token)); | |
701 | } else if (!(ifp->if_flags & IFF_UP)) { | |
702 | printf("%s: TBR suspended (link is down)\n", | |
703 | if_name(ifp)); | |
704 | } | |
705 | } | |
706 | ifnet_set_start_cycle(ifp, NULL); | |
707 | } | |
0a7de745 | 708 | if (update && tbr->tbr_rate_raw != old_rate) { |
39236c6e | 709 | ifclassq_update(ifq, CLASSQ_EV_LINK_BANDWIDTH); |
0a7de745 | 710 | } |
316670eb | 711 | |
0a7de745 | 712 | return 0; |
316670eb | 713 | } |
39037602 A |
714 | |
715 | void | |
716 | ifclassq_calc_target_qdelay(struct ifnet *ifp, u_int64_t *if_target_qdelay) | |
717 | { | |
5ba3f43e A |
718 | u_int64_t qdelay = 0; |
719 | qdelay = IFCQ_TARGET_QDELAY(&ifp->if_snd); | |
39037602 | 720 | |
0a7de745 | 721 | if (ifclassq_target_qdelay != 0) { |
5ba3f43e | 722 | qdelay = ifclassq_target_qdelay; |
0a7de745 | 723 | } |
39037602 A |
724 | |
725 | /* | |
726 | * If we do not know the effective bandwidth, use the default | |
727 | * target queue delay. | |
728 | */ | |
0a7de745 | 729 | if (qdelay == 0) { |
5ba3f43e | 730 | qdelay = IFQ_TARGET_DELAY; |
0a7de745 | 731 | } |
39037602 A |
732 | |
733 | /* | |
734 | * If a delay has been added to ifnet start callback for | |
735 | * coalescing, we have to add that to the pre-set target delay | |
736 | * because the packets can be in the queue longer. | |
737 | */ | |
738 | if ((ifp->if_eflags & IFEF_ENQUEUE_MULTI) && | |
0a7de745 | 739 | ifp->if_start_delay_timeout > 0) { |
5ba3f43e | 740 | qdelay += ifp->if_start_delay_timeout; |
0a7de745 | 741 | } |
39037602 | 742 | |
5ba3f43e | 743 | *(if_target_qdelay) = qdelay; |
39037602 A |
744 | } |
745 | ||
746 | void | |
747 | ifclassq_calc_update_interval(u_int64_t *update_interval) | |
748 | { | |
749 | u_int64_t uint = 0; | |
750 | ||
751 | /* If the system level override is set, use it */ | |
0a7de745 | 752 | if (ifclassq_update_interval != 0) { |
39037602 | 753 | uint = ifclassq_update_interval; |
0a7de745 | 754 | } |
39037602 A |
755 | |
756 | /* Otherwise use the default value */ | |
0a7de745 | 757 | if (uint == 0) { |
39037602 | 758 | uint = IFQ_UPDATE_INTERVAL; |
0a7de745 | 759 | } |
39037602 A |
760 | |
761 | *update_interval = uint; | |
762 | } | |
a39ff7e2 A |
763 | |
764 | void | |
765 | ifclassq_reap_caches(boolean_t purge) | |
766 | { | |
767 | fq_codel_reap_caches(purge); | |
f427ee49 | 768 | flowadv_reap_caches(purge); |
a39ff7e2 | 769 | } |