1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/sched/sch_mqprio.c
4 *
5 * Copyright (c) 2010 John Fastabend <john.r.fastabend@intel.com>
6 */
7
8#include <linux/ethtool_netlink.h>
9#include <linux/types.h>
10#include <linux/slab.h>
11#include <linux/kernel.h>
12#include <linux/string.h>
13#include <linux/errno.h>
14#include <linux/skbuff.h>
15#include <linux/module.h>
16#include <net/netlink.h>
17#include <net/pkt_sched.h>
18#include <net/sch_generic.h>
19#include <net/pkt_cls.h>
20
21#include "sch_mqprio_lib.h"
22
23struct mqprio_sched {
24 struct Qdisc **qdiscs;
25 u16 mode;
26 u16 shaper;
27 int hw_offload;
28 u32 flags;
29 u64 min_rate[TC_QOPT_MAX_QUEUE];
30 u64 max_rate[TC_QOPT_MAX_QUEUE];
31 u32 fp[TC_QOPT_MAX_QUEUE];
32};
33
34static int mqprio_enable_offload(struct Qdisc *sch,
35 const struct tc_mqprio_qopt *qopt,
36 struct netlink_ext_ack *extack)
37{
38 struct mqprio_sched *priv = qdisc_priv(sch);
39 struct net_device *dev = qdisc_dev(qdisc: sch);
40 struct tc_mqprio_qopt_offload mqprio = {
41 .qopt = *qopt,
42 .extack = extack,
43 };
44 int err, i;
45
46 switch (priv->mode) {
47 case TC_MQPRIO_MODE_DCB:
48 if (priv->shaper != TC_MQPRIO_SHAPER_DCB)
49 return -EINVAL;
50 break;
51 case TC_MQPRIO_MODE_CHANNEL:
52 mqprio.flags = priv->flags;
53 if (priv->flags & TC_MQPRIO_F_MODE)
54 mqprio.mode = priv->mode;
55 if (priv->flags & TC_MQPRIO_F_SHAPER)
56 mqprio.shaper = priv->shaper;
57 if (priv->flags & TC_MQPRIO_F_MIN_RATE)
58 for (i = 0; i < mqprio.qopt.num_tc; i++)
59 mqprio.min_rate[i] = priv->min_rate[i];
60 if (priv->flags & TC_MQPRIO_F_MAX_RATE)
61 for (i = 0; i < mqprio.qopt.num_tc; i++)
62 mqprio.max_rate[i] = priv->max_rate[i];
63 break;
64 default:
65 return -EINVAL;
66 }
67
68 mqprio_fp_to_offload(fp: priv->fp, mqprio: &mqprio);
69
70 err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_MQPRIO,
71 &mqprio);
72 if (err)
73 return err;
74
75 priv->hw_offload = mqprio.qopt.hw;
76
77 return 0;
78}
79
80static void mqprio_disable_offload(struct Qdisc *sch)
81{
82 struct tc_mqprio_qopt_offload mqprio = { { 0 } };
83 struct mqprio_sched *priv = qdisc_priv(sch);
84 struct net_device *dev = qdisc_dev(qdisc: sch);
85
86 switch (priv->mode) {
87 case TC_MQPRIO_MODE_DCB:
88 case TC_MQPRIO_MODE_CHANNEL:
89 dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_MQPRIO,
90 &mqprio);
91 break;
92 }
93}
94
95static void mqprio_destroy(struct Qdisc *sch)
96{
97 struct net_device *dev = qdisc_dev(qdisc: sch);
98 struct mqprio_sched *priv = qdisc_priv(sch);
99 unsigned int ntx;
100
101 if (priv->qdiscs) {
102 for (ntx = 0;
103 ntx < dev->num_tx_queues && priv->qdiscs[ntx];
104 ntx++)
105 qdisc_put(qdisc: priv->qdiscs[ntx]);
106 kfree(objp: priv->qdiscs);
107 }
108
109 if (priv->hw_offload && dev->netdev_ops->ndo_setup_tc)
110 mqprio_disable_offload(sch);
111 else
112 netdev_set_num_tc(dev, num_tc: 0);
113}
114
115static int mqprio_parse_opt(struct net_device *dev, struct tc_mqprio_qopt *qopt,
116 const struct tc_mqprio_caps *caps,
117 struct netlink_ext_ack *extack)
118{
119 int err;
120
121 /* Limit qopt->hw to maximum supported offload value. Drivers have
122 * the option of overriding this later if they don't support the a
123 * given offload type.
124 */
125 if (qopt->hw > TC_MQPRIO_HW_OFFLOAD_MAX)
126 qopt->hw = TC_MQPRIO_HW_OFFLOAD_MAX;
127
128 /* If hardware offload is requested, we will leave 3 options to the
129 * device driver:
130 * - populate the queue counts itself (and ignore what was requested)
131 * - validate the provided queue counts by itself (and apply them)
132 * - request queue count validation here (and apply them)
133 */
134 err = mqprio_validate_qopt(dev, qopt,
135 validate_queue_counts: !qopt->hw || caps->validate_queue_counts,
136 allow_overlapping_txqs: false, extack);
137 if (err)
138 return err;
139
140 /* If ndo_setup_tc is not present then hardware doesn't support offload
141 * and we should return an error.
142 */
143 if (qopt->hw && !dev->netdev_ops->ndo_setup_tc) {
144 NL_SET_ERR_MSG(extack,
145 "Device does not support hardware offload");
146 return -EINVAL;
147 }
148
149 return 0;
150}
151
152static const struct
153nla_policy mqprio_tc_entry_policy[TCA_MQPRIO_TC_ENTRY_MAX + 1] = {
154 [TCA_MQPRIO_TC_ENTRY_INDEX] = NLA_POLICY_MAX(NLA_U32,
155 TC_QOPT_MAX_QUEUE),
156 [TCA_MQPRIO_TC_ENTRY_FP] = NLA_POLICY_RANGE(NLA_U32,
157 TC_FP_EXPRESS,
158 TC_FP_PREEMPTIBLE),
159};
160
161static const struct nla_policy mqprio_policy[TCA_MQPRIO_MAX + 1] = {
162 [TCA_MQPRIO_MODE] = { .len = sizeof(u16) },
163 [TCA_MQPRIO_SHAPER] = { .len = sizeof(u16) },
164 [TCA_MQPRIO_MIN_RATE64] = { .type = NLA_NESTED },
165 [TCA_MQPRIO_MAX_RATE64] = { .type = NLA_NESTED },
166 [TCA_MQPRIO_TC_ENTRY] = { .type = NLA_NESTED },
167};
168
169static int mqprio_parse_tc_entry(u32 fp[TC_QOPT_MAX_QUEUE],
170 struct nlattr *opt,
171 unsigned long *seen_tcs,
172 struct netlink_ext_ack *extack)
173{
174 struct nlattr *tb[TCA_MQPRIO_TC_ENTRY_MAX + 1];
175 int err, tc;
176
177 err = nla_parse_nested(tb, maxtype: TCA_MQPRIO_TC_ENTRY_MAX, nla: opt,
178 policy: mqprio_tc_entry_policy, extack);
179 if (err < 0)
180 return err;
181
182 if (NL_REQ_ATTR_CHECK(extack, opt, tb, TCA_MQPRIO_TC_ENTRY_INDEX)) {
183 NL_SET_ERR_MSG(extack, "TC entry index missing");
184 return -EINVAL;
185 }
186
187 tc = nla_get_u32(nla: tb[TCA_MQPRIO_TC_ENTRY_INDEX]);
188 if (*seen_tcs & BIT(tc)) {
189 NL_SET_ERR_MSG_ATTR(extack, tb[TCA_MQPRIO_TC_ENTRY_INDEX],
190 "Duplicate tc entry");
191 return -EINVAL;
192 }
193
194 *seen_tcs |= BIT(tc);
195
196 if (tb[TCA_MQPRIO_TC_ENTRY_FP])
197 fp[tc] = nla_get_u32(nla: tb[TCA_MQPRIO_TC_ENTRY_FP]);
198
199 return 0;
200}
201
202static int mqprio_parse_tc_entries(struct Qdisc *sch, struct nlattr *nlattr_opt,
203 int nlattr_opt_len,
204 struct netlink_ext_ack *extack)
205{
206 struct mqprio_sched *priv = qdisc_priv(sch);
207 struct net_device *dev = qdisc_dev(qdisc: sch);
208 bool have_preemption = false;
209 unsigned long seen_tcs = 0;
210 u32 fp[TC_QOPT_MAX_QUEUE];
211 struct nlattr *n;
212 int tc, rem;
213 int err = 0;
214
215 for (tc = 0; tc < TC_QOPT_MAX_QUEUE; tc++)
216 fp[tc] = priv->fp[tc];
217
218 nla_for_each_attr(n, nlattr_opt, nlattr_opt_len, rem) {
219 if (nla_type(nla: n) != TCA_MQPRIO_TC_ENTRY)
220 continue;
221
222 err = mqprio_parse_tc_entry(fp, opt: n, seen_tcs: &seen_tcs, extack);
223 if (err)
224 goto out;
225 }
226
227 for (tc = 0; tc < TC_QOPT_MAX_QUEUE; tc++) {
228 priv->fp[tc] = fp[tc];
229 if (fp[tc] == TC_FP_PREEMPTIBLE)
230 have_preemption = true;
231 }
232
233 if (have_preemption && !ethtool_dev_mm_supported(dev)) {
234 NL_SET_ERR_MSG(extack, "Device does not support preemption");
235 return -EOPNOTSUPP;
236 }
237out:
238 return err;
239}
240
241/* Parse the other netlink attributes that represent the payload of
242 * TCA_OPTIONS, which are appended right after struct tc_mqprio_qopt.
243 */
244static int mqprio_parse_nlattr(struct Qdisc *sch, struct tc_mqprio_qopt *qopt,
245 struct nlattr *opt,
246 struct netlink_ext_ack *extack)
247{
248 struct nlattr *nlattr_opt = nla_data(nla: opt) + NLA_ALIGN(sizeof(*qopt));
249 int nlattr_opt_len = nla_len(nla: opt) - NLA_ALIGN(sizeof(*qopt));
250 struct mqprio_sched *priv = qdisc_priv(sch);
251 struct nlattr *tb[TCA_MQPRIO_MAX + 1] = {};
252 struct nlattr *attr;
253 int i, rem, err;
254
255 if (nlattr_opt_len >= nla_attr_size(payload: 0)) {
256 err = nla_parse_deprecated(tb, TCA_MQPRIO_MAX, head: nlattr_opt,
257 len: nlattr_opt_len, policy: mqprio_policy,
258 NULL);
259 if (err < 0)
260 return err;
261 }
262
263 if (!qopt->hw) {
264 NL_SET_ERR_MSG(extack,
265 "mqprio TCA_OPTIONS can only contain netlink attributes in hardware mode");
266 return -EINVAL;
267 }
268
269 if (tb[TCA_MQPRIO_MODE]) {
270 priv->flags |= TC_MQPRIO_F_MODE;
271 priv->mode = nla_get_u16(nla: tb[TCA_MQPRIO_MODE]);
272 }
273
274 if (tb[TCA_MQPRIO_SHAPER]) {
275 priv->flags |= TC_MQPRIO_F_SHAPER;
276 priv->shaper = nla_get_u16(nla: tb[TCA_MQPRIO_SHAPER]);
277 }
278
279 if (tb[TCA_MQPRIO_MIN_RATE64]) {
280 if (priv->shaper != TC_MQPRIO_SHAPER_BW_RATE) {
281 NL_SET_ERR_MSG_ATTR(extack, tb[TCA_MQPRIO_MIN_RATE64],
282 "min_rate accepted only when shaper is in bw_rlimit mode");
283 return -EINVAL;
284 }
285 i = 0;
286 nla_for_each_nested(attr, tb[TCA_MQPRIO_MIN_RATE64],
287 rem) {
288 if (nla_type(nla: attr) != TCA_MQPRIO_MIN_RATE64) {
289 NL_SET_ERR_MSG_ATTR(extack, attr,
290 "Attribute type expected to be TCA_MQPRIO_MIN_RATE64");
291 return -EINVAL;
292 }
293
294 if (nla_len(nla: attr) != sizeof(u64)) {
295 NL_SET_ERR_MSG_ATTR(extack, attr,
296 "Attribute TCA_MQPRIO_MIN_RATE64 expected to have 8 bytes length");
297 return -EINVAL;
298 }
299
300 if (i >= qopt->num_tc)
301 break;
302 priv->min_rate[i] = nla_get_u64(nla: attr);
303 i++;
304 }
305 priv->flags |= TC_MQPRIO_F_MIN_RATE;
306 }
307
308 if (tb[TCA_MQPRIO_MAX_RATE64]) {
309 if (priv->shaper != TC_MQPRIO_SHAPER_BW_RATE) {
310 NL_SET_ERR_MSG_ATTR(extack, tb[TCA_MQPRIO_MAX_RATE64],
311 "max_rate accepted only when shaper is in bw_rlimit mode");
312 return -EINVAL;
313 }
314 i = 0;
315 nla_for_each_nested(attr, tb[TCA_MQPRIO_MAX_RATE64],
316 rem) {
317 if (nla_type(nla: attr) != TCA_MQPRIO_MAX_RATE64) {
318 NL_SET_ERR_MSG_ATTR(extack, attr,
319 "Attribute type expected to be TCA_MQPRIO_MAX_RATE64");
320 return -EINVAL;
321 }
322
323 if (nla_len(nla: attr) != sizeof(u64)) {
324 NL_SET_ERR_MSG_ATTR(extack, attr,
325 "Attribute TCA_MQPRIO_MAX_RATE64 expected to have 8 bytes length");
326 return -EINVAL;
327 }
328
329 if (i >= qopt->num_tc)
330 break;
331 priv->max_rate[i] = nla_get_u64(nla: attr);
332 i++;
333 }
334 priv->flags |= TC_MQPRIO_F_MAX_RATE;
335 }
336
337 if (tb[TCA_MQPRIO_TC_ENTRY]) {
338 err = mqprio_parse_tc_entries(sch, nlattr_opt, nlattr_opt_len,
339 extack);
340 if (err)
341 return err;
342 }
343
344 return 0;
345}
346
347static int mqprio_init(struct Qdisc *sch, struct nlattr *opt,
348 struct netlink_ext_ack *extack)
349{
350 struct net_device *dev = qdisc_dev(qdisc: sch);
351 struct mqprio_sched *priv = qdisc_priv(sch);
352 struct netdev_queue *dev_queue;
353 struct Qdisc *qdisc;
354 int i, err = -EOPNOTSUPP;
355 struct tc_mqprio_qopt *qopt = NULL;
356 struct tc_mqprio_caps caps;
357 int len, tc;
358
359 BUILD_BUG_ON(TC_MAX_QUEUE != TC_QOPT_MAX_QUEUE);
360 BUILD_BUG_ON(TC_BITMASK != TC_QOPT_BITMASK);
361
362 if (sch->parent != TC_H_ROOT)
363 return -EOPNOTSUPP;
364
365 if (!netif_is_multiqueue(dev))
366 return -EOPNOTSUPP;
367
368 /* make certain can allocate enough classids to handle queues */
369 if (dev->num_tx_queues >= TC_H_MIN_PRIORITY)
370 return -ENOMEM;
371
372 if (!opt || nla_len(nla: opt) < sizeof(*qopt))
373 return -EINVAL;
374
375 for (tc = 0; tc < TC_QOPT_MAX_QUEUE; tc++)
376 priv->fp[tc] = TC_FP_EXPRESS;
377
378 qdisc_offload_query_caps(dev, type: TC_SETUP_QDISC_MQPRIO,
379 caps: &caps, caps_len: sizeof(caps));
380
381 qopt = nla_data(nla: opt);
382 if (mqprio_parse_opt(dev, qopt, caps: &caps, extack))
383 return -EINVAL;
384
385 len = nla_len(nla: opt) - NLA_ALIGN(sizeof(*qopt));
386 if (len > 0) {
387 err = mqprio_parse_nlattr(sch, qopt, opt, extack);
388 if (err)
389 return err;
390 }
391
392 /* pre-allocate qdisc, attachment can't fail */
393 priv->qdiscs = kcalloc(n: dev->num_tx_queues, size: sizeof(priv->qdiscs[0]),
394 GFP_KERNEL);
395 if (!priv->qdiscs)
396 return -ENOMEM;
397
398 for (i = 0; i < dev->num_tx_queues; i++) {
399 dev_queue = netdev_get_tx_queue(dev, index: i);
400 qdisc = qdisc_create_dflt(dev_queue,
401 ops: get_default_qdisc_ops(dev, ntx: i),
402 TC_H_MAKE(TC_H_MAJ(sch->handle),
403 TC_H_MIN(i + 1)), extack);
404 if (!qdisc)
405 return -ENOMEM;
406
407 priv->qdiscs[i] = qdisc;
408 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
409 }
410
411 /* If the mqprio options indicate that hardware should own
412 * the queue mapping then run ndo_setup_tc otherwise use the
413 * supplied and verified mapping
414 */
415 if (qopt->hw) {
416 err = mqprio_enable_offload(sch, qopt, extack);
417 if (err)
418 return err;
419 } else {
420 netdev_set_num_tc(dev, num_tc: qopt->num_tc);
421 for (i = 0; i < qopt->num_tc; i++)
422 netdev_set_tc_queue(dev, tc: i,
423 count: qopt->count[i], offset: qopt->offset[i]);
424 }
425
426 /* Always use supplied priority mappings */
427 for (i = 0; i < TC_BITMASK + 1; i++)
428 netdev_set_prio_tc_map(dev, prio: i, tc: qopt->prio_tc_map[i]);
429
430 sch->flags |= TCQ_F_MQROOT;
431 return 0;
432}
433
434static void mqprio_attach(struct Qdisc *sch)
435{
436 struct net_device *dev = qdisc_dev(qdisc: sch);
437 struct mqprio_sched *priv = qdisc_priv(sch);
438 struct Qdisc *qdisc, *old;
439 unsigned int ntx;
440
441 /* Attach underlying qdisc */
442 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
443 qdisc = priv->qdiscs[ntx];
444 old = dev_graft_qdisc(dev_queue: qdisc->dev_queue, qdisc);
445 if (old)
446 qdisc_put(qdisc: old);
447 if (ntx < dev->real_num_tx_queues)
448 qdisc_hash_add(q: qdisc, invisible: false);
449 }
450 kfree(objp: priv->qdiscs);
451 priv->qdiscs = NULL;
452}
453
454static struct netdev_queue *mqprio_queue_get(struct Qdisc *sch,
455 unsigned long cl)
456{
457 struct net_device *dev = qdisc_dev(qdisc: sch);
458 unsigned long ntx = cl - 1;
459
460 if (ntx >= dev->num_tx_queues)
461 return NULL;
462 return netdev_get_tx_queue(dev, index: ntx);
463}
464
465static int mqprio_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
466 struct Qdisc **old, struct netlink_ext_ack *extack)
467{
468 struct net_device *dev = qdisc_dev(qdisc: sch);
469 struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
470
471 if (!dev_queue)
472 return -EINVAL;
473
474 if (dev->flags & IFF_UP)
475 dev_deactivate(dev);
476
477 *old = dev_graft_qdisc(dev_queue, qdisc: new);
478
479 if (new)
480 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
481
482 if (dev->flags & IFF_UP)
483 dev_activate(dev);
484
485 return 0;
486}
487
488static int dump_rates(struct mqprio_sched *priv,
489 struct tc_mqprio_qopt *opt, struct sk_buff *skb)
490{
491 struct nlattr *nest;
492 int i;
493
494 if (priv->flags & TC_MQPRIO_F_MIN_RATE) {
495 nest = nla_nest_start_noflag(skb, attrtype: TCA_MQPRIO_MIN_RATE64);
496 if (!nest)
497 goto nla_put_failure;
498
499 for (i = 0; i < opt->num_tc; i++) {
500 if (nla_put(skb, attrtype: TCA_MQPRIO_MIN_RATE64,
501 attrlen: sizeof(priv->min_rate[i]),
502 data: &priv->min_rate[i]))
503 goto nla_put_failure;
504 }
505 nla_nest_end(skb, start: nest);
506 }
507
508 if (priv->flags & TC_MQPRIO_F_MAX_RATE) {
509 nest = nla_nest_start_noflag(skb, attrtype: TCA_MQPRIO_MAX_RATE64);
510 if (!nest)
511 goto nla_put_failure;
512
513 for (i = 0; i < opt->num_tc; i++) {
514 if (nla_put(skb, attrtype: TCA_MQPRIO_MAX_RATE64,
515 attrlen: sizeof(priv->max_rate[i]),
516 data: &priv->max_rate[i]))
517 goto nla_put_failure;
518 }
519 nla_nest_end(skb, start: nest);
520 }
521 return 0;
522
523nla_put_failure:
524 nla_nest_cancel(skb, start: nest);
525 return -1;
526}
527
528static int mqprio_dump_tc_entries(struct mqprio_sched *priv,
529 struct sk_buff *skb)
530{
531 struct nlattr *n;
532 int tc;
533
534 for (tc = 0; tc < TC_QOPT_MAX_QUEUE; tc++) {
535 n = nla_nest_start(skb, attrtype: TCA_MQPRIO_TC_ENTRY);
536 if (!n)
537 return -EMSGSIZE;
538
539 if (nla_put_u32(skb, attrtype: TCA_MQPRIO_TC_ENTRY_INDEX, value: tc))
540 goto nla_put_failure;
541
542 if (nla_put_u32(skb, attrtype: TCA_MQPRIO_TC_ENTRY_FP, value: priv->fp[tc]))
543 goto nla_put_failure;
544
545 nla_nest_end(skb, start: n);
546 }
547
548 return 0;
549
550nla_put_failure:
551 nla_nest_cancel(skb, start: n);
552 return -EMSGSIZE;
553}
554
555static int mqprio_dump(struct Qdisc *sch, struct sk_buff *skb)
556{
557 struct net_device *dev = qdisc_dev(qdisc: sch);
558 struct mqprio_sched *priv = qdisc_priv(sch);
559 struct nlattr *nla = (struct nlattr *)skb_tail_pointer(skb);
560 struct tc_mqprio_qopt opt = { 0 };
561 struct Qdisc *qdisc;
562 unsigned int ntx;
563
564 sch->q.qlen = 0;
565 gnet_stats_basic_sync_init(b: &sch->bstats);
566 memset(&sch->qstats, 0, sizeof(sch->qstats));
567
568 /* MQ supports lockless qdiscs. However, statistics accounting needs
569 * to account for all, none, or a mix of locked and unlocked child
570 * qdiscs. Percpu stats are added to counters in-band and locking
571 * qdisc totals are added at end.
572 */
573 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
574 qdisc = rtnl_dereference(netdev_get_tx_queue(dev, ntx)->qdisc_sleeping);
575 spin_lock_bh(lock: qdisc_lock(qdisc));
576
577 gnet_stats_add_basic(bstats: &sch->bstats, cpu: qdisc->cpu_bstats,
578 b: &qdisc->bstats, running: false);
579 gnet_stats_add_queue(qstats: &sch->qstats, cpu_q: qdisc->cpu_qstats,
580 q: &qdisc->qstats);
581 sch->q.qlen += qdisc_qlen(q: qdisc);
582
583 spin_unlock_bh(lock: qdisc_lock(qdisc));
584 }
585
586 mqprio_qopt_reconstruct(dev, qopt: &opt);
587 opt.hw = priv->hw_offload;
588
589 if (nla_put(skb, attrtype: TCA_OPTIONS, attrlen: sizeof(opt), data: &opt))
590 goto nla_put_failure;
591
592 if ((priv->flags & TC_MQPRIO_F_MODE) &&
593 nla_put_u16(skb, attrtype: TCA_MQPRIO_MODE, value: priv->mode))
594 goto nla_put_failure;
595
596 if ((priv->flags & TC_MQPRIO_F_SHAPER) &&
597 nla_put_u16(skb, attrtype: TCA_MQPRIO_SHAPER, value: priv->shaper))
598 goto nla_put_failure;
599
600 if ((priv->flags & TC_MQPRIO_F_MIN_RATE ||
601 priv->flags & TC_MQPRIO_F_MAX_RATE) &&
602 (dump_rates(priv, opt: &opt, skb) != 0))
603 goto nla_put_failure;
604
605 if (mqprio_dump_tc_entries(priv, skb))
606 goto nla_put_failure;
607
608 return nla_nest_end(skb, start: nla);
609nla_put_failure:
610 nlmsg_trim(skb, mark: nla);
611 return -1;
612}
613
614static struct Qdisc *mqprio_leaf(struct Qdisc *sch, unsigned long cl)
615{
616 struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
617
618 if (!dev_queue)
619 return NULL;
620
621 return rtnl_dereference(dev_queue->qdisc_sleeping);
622}
623
624static unsigned long mqprio_find(struct Qdisc *sch, u32 classid)
625{
626 struct net_device *dev = qdisc_dev(qdisc: sch);
627 unsigned int ntx = TC_H_MIN(classid);
628
629 /* There are essentially two regions here that have valid classid
630 * values. The first region will have a classid value of 1 through
631 * num_tx_queues. All of these are backed by actual Qdiscs.
632 */
633 if (ntx < TC_H_MIN_PRIORITY)
634 return (ntx <= dev->num_tx_queues) ? ntx : 0;
635
636 /* The second region represents the hardware traffic classes. These
637 * are represented by classid values of TC_H_MIN_PRIORITY through
638 * TC_H_MIN_PRIORITY + netdev_get_num_tc - 1
639 */
640 return ((ntx - TC_H_MIN_PRIORITY) < netdev_get_num_tc(dev)) ? ntx : 0;
641}
642
643static int mqprio_dump_class(struct Qdisc *sch, unsigned long cl,
644 struct sk_buff *skb, struct tcmsg *tcm)
645{
646 if (cl < TC_H_MIN_PRIORITY) {
647 struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
648 struct net_device *dev = qdisc_dev(qdisc: sch);
649 int tc = netdev_txq_to_tc(dev, txq: cl - 1);
650
651 tcm->tcm_parent = (tc < 0) ? 0 :
652 TC_H_MAKE(TC_H_MAJ(sch->handle),
653 TC_H_MIN(tc + TC_H_MIN_PRIORITY));
654 tcm->tcm_info = rtnl_dereference(dev_queue->qdisc_sleeping)->handle;
655 } else {
656 tcm->tcm_parent = TC_H_ROOT;
657 tcm->tcm_info = 0;
658 }
659 tcm->tcm_handle |= TC_H_MIN(cl);
660 return 0;
661}
662
663static int mqprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
664 struct gnet_dump *d)
665 __releases(d->lock)
666 __acquires(d->lock)
667{
668 if (cl >= TC_H_MIN_PRIORITY) {
669 int i;
670 __u32 qlen;
671 struct gnet_stats_queue qstats = {0};
672 struct gnet_stats_basic_sync bstats;
673 struct net_device *dev = qdisc_dev(qdisc: sch);
674 struct netdev_tc_txq tc = dev->tc_to_txq[cl & TC_BITMASK];
675
676 gnet_stats_basic_sync_init(b: &bstats);
677 /* Drop lock here it will be reclaimed before touching
678 * statistics this is required because the d->lock we
679 * hold here is the look on dev_queue->qdisc_sleeping
680 * also acquired below.
681 */
682 if (d->lock)
683 spin_unlock_bh(lock: d->lock);
684
685 for (i = tc.offset; i < tc.offset + tc.count; i++) {
686 struct netdev_queue *q = netdev_get_tx_queue(dev, index: i);
687 struct Qdisc *qdisc = rtnl_dereference(q->qdisc);
688
689 spin_lock_bh(lock: qdisc_lock(qdisc));
690
691 gnet_stats_add_basic(bstats: &bstats, cpu: qdisc->cpu_bstats,
692 b: &qdisc->bstats, running: false);
693 gnet_stats_add_queue(qstats: &qstats, cpu_q: qdisc->cpu_qstats,
694 q: &qdisc->qstats);
695 sch->q.qlen += qdisc_qlen(q: qdisc);
696
697 spin_unlock_bh(lock: qdisc_lock(qdisc));
698 }
699 qlen = qdisc_qlen(q: sch) + qstats.qlen;
700
701 /* Reclaim root sleeping lock before completing stats */
702 if (d->lock)
703 spin_lock_bh(lock: d->lock);
704 if (gnet_stats_copy_basic(d, NULL, b: &bstats, running: false) < 0 ||
705 gnet_stats_copy_queue(d, NULL, q: &qstats, qlen) < 0)
706 return -1;
707 } else {
708 struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
709
710 sch = rtnl_dereference(dev_queue->qdisc_sleeping);
711 if (gnet_stats_copy_basic(d, cpu: sch->cpu_bstats,
712 b: &sch->bstats, running: true) < 0 ||
713 qdisc_qstats_copy(d, sch) < 0)
714 return -1;
715 }
716 return 0;
717}
718
719static void mqprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
720{
721 struct net_device *dev = qdisc_dev(qdisc: sch);
722 unsigned long ntx;
723
724 if (arg->stop)
725 return;
726
727 /* Walk hierarchy with a virtual class per tc */
728 arg->count = arg->skip;
729 for (ntx = arg->skip; ntx < netdev_get_num_tc(dev); ntx++) {
730 if (!tc_qdisc_stats_dump(sch, cl: ntx + TC_H_MIN_PRIORITY, arg))
731 return;
732 }
733
734 /* Pad the values and skip over unused traffic classes */
735 if (ntx < TC_MAX_QUEUE) {
736 arg->count = TC_MAX_QUEUE;
737 ntx = TC_MAX_QUEUE;
738 }
739
740 /* Reset offset, sort out remaining per-queue qdiscs */
741 for (ntx -= TC_MAX_QUEUE; ntx < dev->num_tx_queues; ntx++) {
742 if (arg->fn(sch, ntx + 1, arg) < 0) {
743 arg->stop = 1;
744 return;
745 }
746 arg->count++;
747 }
748}
749
750static struct netdev_queue *mqprio_select_queue(struct Qdisc *sch,
751 struct tcmsg *tcm)
752{
753 return mqprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
754}
755
756static const struct Qdisc_class_ops mqprio_class_ops = {
757 .graft = mqprio_graft,
758 .leaf = mqprio_leaf,
759 .find = mqprio_find,
760 .walk = mqprio_walk,
761 .dump = mqprio_dump_class,
762 .dump_stats = mqprio_dump_class_stats,
763 .select_queue = mqprio_select_queue,
764};
765
766static struct Qdisc_ops mqprio_qdisc_ops __read_mostly = {
767 .cl_ops = &mqprio_class_ops,
768 .id = "mqprio",
769 .priv_size = sizeof(struct mqprio_sched),
770 .init = mqprio_init,
771 .destroy = mqprio_destroy,
772 .attach = mqprio_attach,
773 .change_real_num_tx = mq_change_real_num_tx,
774 .dump = mqprio_dump,
775 .owner = THIS_MODULE,
776};
777
778static int __init mqprio_module_init(void)
779{
780 return register_qdisc(qops: &mqprio_qdisc_ops);
781}
782
783static void __exit mqprio_module_exit(void)
784{
785 unregister_qdisc(qops: &mqprio_qdisc_ops);
786}
787
788module_init(mqprio_module_init);
789module_exit(mqprio_module_exit);
790
791MODULE_LICENSE("GPL");
792

source code of linux/net/sched/sch_mqprio.c