• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/types.h>
9 #include <linux/slab.h>
10 #include <linux/kernel.h>
11 #include <linux/string.h>
12 #include <linux/errno.h>
13 #include <linux/skbuff.h>
14 #include <linux/module.h>
15 #include <net/netlink.h>
16 #include <net/pkt_sched.h>
17 #include <net/sch_generic.h>
18 #include <net/pkt_cls.h>
19 
20 struct mqprio_sched {
21 	struct Qdisc		**qdiscs;
22 	u16 mode;
23 	u16 shaper;
24 	int hw_offload;
25 	u32 flags;
26 	u64 min_rate[TC_QOPT_MAX_QUEUE];
27 	u64 max_rate[TC_QOPT_MAX_QUEUE];
28 };
29 
mqprio_destroy(struct Qdisc * sch)30 static void mqprio_destroy(struct Qdisc *sch)
31 {
32 	struct net_device *dev = qdisc_dev(sch);
33 	struct mqprio_sched *priv = qdisc_priv(sch);
34 	unsigned int ntx;
35 
36 	if (priv->qdiscs) {
37 		for (ntx = 0;
38 		     ntx < dev->num_tx_queues && priv->qdiscs[ntx];
39 		     ntx++)
40 			qdisc_put(priv->qdiscs[ntx]);
41 		kfree(priv->qdiscs);
42 	}
43 
44 	if (priv->hw_offload && dev->netdev_ops->ndo_setup_tc) {
45 		struct tc_mqprio_qopt_offload mqprio = { { 0 } };
46 
47 		switch (priv->mode) {
48 		case TC_MQPRIO_MODE_DCB:
49 		case TC_MQPRIO_MODE_CHANNEL:
50 			dev->netdev_ops->ndo_setup_tc(dev,
51 						      TC_SETUP_QDISC_MQPRIO,
52 						      &mqprio);
53 			break;
54 		default:
55 			return;
56 		}
57 	} else {
58 		netdev_set_num_tc(dev, 0);
59 	}
60 }
61 
mqprio_parse_opt(struct net_device * dev,struct tc_mqprio_qopt * qopt)62 static int mqprio_parse_opt(struct net_device *dev, struct tc_mqprio_qopt *qopt)
63 {
64 	int i, j;
65 
66 	/* Verify num_tc is not out of max range */
67 	if (qopt->num_tc > TC_MAX_QUEUE)
68 		return -EINVAL;
69 
70 	/* Verify priority mapping uses valid tcs */
71 	for (i = 0; i < TC_BITMASK + 1; i++) {
72 		if (qopt->prio_tc_map[i] >= qopt->num_tc)
73 			return -EINVAL;
74 	}
75 
76 	/* Limit qopt->hw to maximum supported offload value.  Drivers have
77 	 * the option of overriding this later if they don't support the a
78 	 * given offload type.
79 	 */
80 	if (qopt->hw > TC_MQPRIO_HW_OFFLOAD_MAX)
81 		qopt->hw = TC_MQPRIO_HW_OFFLOAD_MAX;
82 
83 	/* If hardware offload is requested we will leave it to the device
84 	 * to either populate the queue counts itself or to validate the
85 	 * provided queue counts.  If ndo_setup_tc is not present then
86 	 * hardware doesn't support offload and we should return an error.
87 	 */
88 	if (qopt->hw)
89 		return dev->netdev_ops->ndo_setup_tc ? 0 : -EINVAL;
90 
91 	for (i = 0; i < qopt->num_tc; i++) {
92 		unsigned int last = qopt->offset[i] + qopt->count[i];
93 
94 		/* Verify the queue count is in tx range being equal to the
95 		 * real_num_tx_queues indicates the last queue is in use.
96 		 */
97 		if (qopt->offset[i] >= dev->real_num_tx_queues ||
98 		    !qopt->count[i] ||
99 		    last > dev->real_num_tx_queues)
100 			return -EINVAL;
101 
102 		/* Verify that the offset and counts do not overlap */
103 		for (j = i + 1; j < qopt->num_tc; j++) {
104 			if (last > qopt->offset[j])
105 				return -EINVAL;
106 		}
107 	}
108 
109 	return 0;
110 }
111 
112 static const struct nla_policy mqprio_policy[TCA_MQPRIO_MAX + 1] = {
113 	[TCA_MQPRIO_MODE]	= { .len = sizeof(u16) },
114 	[TCA_MQPRIO_SHAPER]	= { .len = sizeof(u16) },
115 	[TCA_MQPRIO_MIN_RATE64]	= { .type = NLA_NESTED },
116 	[TCA_MQPRIO_MAX_RATE64]	= { .type = NLA_NESTED },
117 };
118 
parse_attr(struct nlattr * tb[],int maxtype,struct nlattr * nla,const struct nla_policy * policy,int len)119 static int parse_attr(struct nlattr *tb[], int maxtype, struct nlattr *nla,
120 		      const struct nla_policy *policy, int len)
121 {
122 	int nested_len = nla_len(nla) - NLA_ALIGN(len);
123 
124 	if (nested_len >= nla_attr_size(0))
125 		return nla_parse_deprecated(tb, maxtype,
126 					    nla_data(nla) + NLA_ALIGN(len),
127 					    nested_len, policy, NULL);
128 
129 	memset(tb, 0, sizeof(struct nlattr *) * (maxtype + 1));
130 	return 0;
131 }
132 
mqprio_init(struct Qdisc * sch,struct nlattr * opt,struct netlink_ext_ack * extack)133 static int mqprio_init(struct Qdisc *sch, struct nlattr *opt,
134 		       struct netlink_ext_ack *extack)
135 {
136 	struct net_device *dev = qdisc_dev(sch);
137 	struct mqprio_sched *priv = qdisc_priv(sch);
138 	struct netdev_queue *dev_queue;
139 	struct Qdisc *qdisc;
140 	int i, err = -EOPNOTSUPP;
141 	struct tc_mqprio_qopt *qopt = NULL;
142 	struct nlattr *tb[TCA_MQPRIO_MAX + 1];
143 	struct nlattr *attr;
144 	int rem;
145 	int len;
146 
147 	BUILD_BUG_ON(TC_MAX_QUEUE != TC_QOPT_MAX_QUEUE);
148 	BUILD_BUG_ON(TC_BITMASK != TC_QOPT_BITMASK);
149 
150 	if (sch->parent != TC_H_ROOT)
151 		return -EOPNOTSUPP;
152 
153 	if (!netif_is_multiqueue(dev))
154 		return -EOPNOTSUPP;
155 
156 	/* make certain can allocate enough classids to handle queues */
157 	if (dev->num_tx_queues >= TC_H_MIN_PRIORITY)
158 		return -ENOMEM;
159 
160 	if (!opt || nla_len(opt) < sizeof(*qopt))
161 		return -EINVAL;
162 
163 	qopt = nla_data(opt);
164 	if (mqprio_parse_opt(dev, qopt))
165 		return -EINVAL;
166 
167 	len = nla_len(opt) - NLA_ALIGN(sizeof(*qopt));
168 	if (len > 0) {
169 		err = parse_attr(tb, TCA_MQPRIO_MAX, opt, mqprio_policy,
170 				 sizeof(*qopt));
171 		if (err < 0)
172 			return err;
173 
174 		if (!qopt->hw)
175 			return -EINVAL;
176 
177 		if (tb[TCA_MQPRIO_MODE]) {
178 			priv->flags |= TC_MQPRIO_F_MODE;
179 			priv->mode = *(u16 *)nla_data(tb[TCA_MQPRIO_MODE]);
180 		}
181 
182 		if (tb[TCA_MQPRIO_SHAPER]) {
183 			priv->flags |= TC_MQPRIO_F_SHAPER;
184 			priv->shaper = *(u16 *)nla_data(tb[TCA_MQPRIO_SHAPER]);
185 		}
186 
187 		if (tb[TCA_MQPRIO_MIN_RATE64]) {
188 			if (priv->shaper != TC_MQPRIO_SHAPER_BW_RATE)
189 				return -EINVAL;
190 			i = 0;
191 			nla_for_each_nested(attr, tb[TCA_MQPRIO_MIN_RATE64],
192 					    rem) {
193 				if (nla_type(attr) != TCA_MQPRIO_MIN_RATE64)
194 					return -EINVAL;
195 				if (i >= qopt->num_tc)
196 					break;
197 				priv->min_rate[i] = *(u64 *)nla_data(attr);
198 				i++;
199 			}
200 			priv->flags |= TC_MQPRIO_F_MIN_RATE;
201 		}
202 
203 		if (tb[TCA_MQPRIO_MAX_RATE64]) {
204 			if (priv->shaper != TC_MQPRIO_SHAPER_BW_RATE)
205 				return -EINVAL;
206 			i = 0;
207 			nla_for_each_nested(attr, tb[TCA_MQPRIO_MAX_RATE64],
208 					    rem) {
209 				if (nla_type(attr) != TCA_MQPRIO_MAX_RATE64)
210 					return -EINVAL;
211 				if (i >= qopt->num_tc)
212 					break;
213 				priv->max_rate[i] = *(u64 *)nla_data(attr);
214 				i++;
215 			}
216 			priv->flags |= TC_MQPRIO_F_MAX_RATE;
217 		}
218 	}
219 
220 	/* pre-allocate qdisc, attachment can't fail */
221 	priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
222 			       GFP_KERNEL);
223 	if (!priv->qdiscs)
224 		return -ENOMEM;
225 
226 	for (i = 0; i < dev->num_tx_queues; i++) {
227 		dev_queue = netdev_get_tx_queue(dev, i);
228 		qdisc = qdisc_create_dflt(dev_queue,
229 					  get_default_qdisc_ops(dev, i),
230 					  TC_H_MAKE(TC_H_MAJ(sch->handle),
231 						    TC_H_MIN(i + 1)), extack);
232 		if (!qdisc)
233 			return -ENOMEM;
234 
235 		priv->qdiscs[i] = qdisc;
236 		qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
237 	}
238 
239 	/* If the mqprio options indicate that hardware should own
240 	 * the queue mapping then run ndo_setup_tc otherwise use the
241 	 * supplied and verified mapping
242 	 */
243 	if (qopt->hw) {
244 		struct tc_mqprio_qopt_offload mqprio = {.qopt = *qopt};
245 
246 		switch (priv->mode) {
247 		case TC_MQPRIO_MODE_DCB:
248 			if (priv->shaper != TC_MQPRIO_SHAPER_DCB)
249 				return -EINVAL;
250 			break;
251 		case TC_MQPRIO_MODE_CHANNEL:
252 			mqprio.flags = priv->flags;
253 			if (priv->flags & TC_MQPRIO_F_MODE)
254 				mqprio.mode = priv->mode;
255 			if (priv->flags & TC_MQPRIO_F_SHAPER)
256 				mqprio.shaper = priv->shaper;
257 			if (priv->flags & TC_MQPRIO_F_MIN_RATE)
258 				for (i = 0; i < mqprio.qopt.num_tc; i++)
259 					mqprio.min_rate[i] = priv->min_rate[i];
260 			if (priv->flags & TC_MQPRIO_F_MAX_RATE)
261 				for (i = 0; i < mqprio.qopt.num_tc; i++)
262 					mqprio.max_rate[i] = priv->max_rate[i];
263 			break;
264 		default:
265 			return -EINVAL;
266 		}
267 		err = dev->netdev_ops->ndo_setup_tc(dev,
268 						    TC_SETUP_QDISC_MQPRIO,
269 						    &mqprio);
270 		if (err)
271 			return err;
272 
273 		priv->hw_offload = mqprio.qopt.hw;
274 	} else {
275 		netdev_set_num_tc(dev, qopt->num_tc);
276 		for (i = 0; i < qopt->num_tc; i++)
277 			netdev_set_tc_queue(dev, i,
278 					    qopt->count[i], qopt->offset[i]);
279 	}
280 
281 	/* Always use supplied priority mappings */
282 	for (i = 0; i < TC_BITMASK + 1; i++)
283 		netdev_set_prio_tc_map(dev, i, qopt->prio_tc_map[i]);
284 
285 	sch->flags |= TCQ_F_MQROOT;
286 	return 0;
287 }
288 
mqprio_attach(struct Qdisc * sch)289 static void mqprio_attach(struct Qdisc *sch)
290 {
291 	struct net_device *dev = qdisc_dev(sch);
292 	struct mqprio_sched *priv = qdisc_priv(sch);
293 	struct Qdisc *qdisc, *old;
294 	unsigned int ntx;
295 
296 	/* Attach underlying qdisc */
297 	for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
298 		qdisc = priv->qdiscs[ntx];
299 		old = dev_graft_qdisc(qdisc->dev_queue, qdisc);
300 		if (old)
301 			qdisc_put(old);
302 		if (ntx < dev->real_num_tx_queues)
303 			qdisc_hash_add(qdisc, false);
304 	}
305 	kfree(priv->qdiscs);
306 	priv->qdiscs = NULL;
307 }
308 
mqprio_queue_get(struct Qdisc * sch,unsigned long cl)309 static struct netdev_queue *mqprio_queue_get(struct Qdisc *sch,
310 					     unsigned long cl)
311 {
312 	struct net_device *dev = qdisc_dev(sch);
313 	unsigned long ntx = cl - 1;
314 
315 	if (ntx >= dev->num_tx_queues)
316 		return NULL;
317 	return netdev_get_tx_queue(dev, ntx);
318 }
319 
mqprio_graft(struct Qdisc * sch,unsigned long cl,struct Qdisc * new,struct Qdisc ** old,struct netlink_ext_ack * extack)320 static int mqprio_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
321 			struct Qdisc **old, struct netlink_ext_ack *extack)
322 {
323 	struct net_device *dev = qdisc_dev(sch);
324 	struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
325 
326 	if (!dev_queue)
327 		return -EINVAL;
328 
329 	if (dev->flags & IFF_UP)
330 		dev_deactivate(dev);
331 
332 	*old = dev_graft_qdisc(dev_queue, new);
333 
334 	if (new)
335 		new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
336 
337 	if (dev->flags & IFF_UP)
338 		dev_activate(dev);
339 
340 	return 0;
341 }
342 
dump_rates(struct mqprio_sched * priv,struct tc_mqprio_qopt * opt,struct sk_buff * skb)343 static int dump_rates(struct mqprio_sched *priv,
344 		      struct tc_mqprio_qopt *opt, struct sk_buff *skb)
345 {
346 	struct nlattr *nest;
347 	int i;
348 
349 	if (priv->flags & TC_MQPRIO_F_MIN_RATE) {
350 		nest = nla_nest_start_noflag(skb, TCA_MQPRIO_MIN_RATE64);
351 		if (!nest)
352 			goto nla_put_failure;
353 
354 		for (i = 0; i < opt->num_tc; i++) {
355 			if (nla_put(skb, TCA_MQPRIO_MIN_RATE64,
356 				    sizeof(priv->min_rate[i]),
357 				    &priv->min_rate[i]))
358 				goto nla_put_failure;
359 		}
360 		nla_nest_end(skb, nest);
361 	}
362 
363 	if (priv->flags & TC_MQPRIO_F_MAX_RATE) {
364 		nest = nla_nest_start_noflag(skb, TCA_MQPRIO_MAX_RATE64);
365 		if (!nest)
366 			goto nla_put_failure;
367 
368 		for (i = 0; i < opt->num_tc; i++) {
369 			if (nla_put(skb, TCA_MQPRIO_MAX_RATE64,
370 				    sizeof(priv->max_rate[i]),
371 				    &priv->max_rate[i]))
372 				goto nla_put_failure;
373 		}
374 		nla_nest_end(skb, nest);
375 	}
376 	return 0;
377 
378 nla_put_failure:
379 	nla_nest_cancel(skb, nest);
380 	return -1;
381 }
382 
mqprio_dump(struct Qdisc * sch,struct sk_buff * skb)383 static int mqprio_dump(struct Qdisc *sch, struct sk_buff *skb)
384 {
385 	struct net_device *dev = qdisc_dev(sch);
386 	struct mqprio_sched *priv = qdisc_priv(sch);
387 	struct nlattr *nla = (struct nlattr *)skb_tail_pointer(skb);
388 	struct tc_mqprio_qopt opt = { 0 };
389 	struct Qdisc *qdisc;
390 	unsigned int ntx, tc;
391 
392 	sch->q.qlen = 0;
393 	memset(&sch->bstats, 0, sizeof(sch->bstats));
394 	memset(&sch->qstats, 0, sizeof(sch->qstats));
395 
396 	/* MQ supports lockless qdiscs. However, statistics accounting needs
397 	 * to account for all, none, or a mix of locked and unlocked child
398 	 * qdiscs. Percpu stats are added to counters in-band and locking
399 	 * qdisc totals are added at end.
400 	 */
401 	for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
402 		qdisc = netdev_get_tx_queue(dev, ntx)->qdisc_sleeping;
403 		spin_lock_bh(qdisc_lock(qdisc));
404 
405 		if (qdisc_is_percpu_stats(qdisc)) {
406 			__u32 qlen = qdisc_qlen_sum(qdisc);
407 
408 			__gnet_stats_copy_basic(NULL, &sch->bstats,
409 						qdisc->cpu_bstats,
410 						&qdisc->bstats);
411 			__gnet_stats_copy_queue(&sch->qstats,
412 						qdisc->cpu_qstats,
413 						&qdisc->qstats, qlen);
414 			sch->q.qlen		+= qlen;
415 		} else {
416 			sch->q.qlen		+= qdisc->q.qlen;
417 			sch->bstats.bytes	+= qdisc->bstats.bytes;
418 			sch->bstats.packets	+= qdisc->bstats.packets;
419 			sch->qstats.backlog	+= qdisc->qstats.backlog;
420 			sch->qstats.drops	+= qdisc->qstats.drops;
421 			sch->qstats.requeues	+= qdisc->qstats.requeues;
422 			sch->qstats.overlimits	+= qdisc->qstats.overlimits;
423 		}
424 
425 		spin_unlock_bh(qdisc_lock(qdisc));
426 	}
427 
428 	opt.num_tc = netdev_get_num_tc(dev);
429 	memcpy(opt.prio_tc_map, dev->prio_tc_map, sizeof(opt.prio_tc_map));
430 	opt.hw = priv->hw_offload;
431 
432 	for (tc = 0; tc < netdev_get_num_tc(dev); tc++) {
433 		opt.count[tc] = dev->tc_to_txq[tc].count;
434 		opt.offset[tc] = dev->tc_to_txq[tc].offset;
435 	}
436 
437 	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
438 		goto nla_put_failure;
439 
440 	if ((priv->flags & TC_MQPRIO_F_MODE) &&
441 	    nla_put_u16(skb, TCA_MQPRIO_MODE, priv->mode))
442 		goto nla_put_failure;
443 
444 	if ((priv->flags & TC_MQPRIO_F_SHAPER) &&
445 	    nla_put_u16(skb, TCA_MQPRIO_SHAPER, priv->shaper))
446 		goto nla_put_failure;
447 
448 	if ((priv->flags & TC_MQPRIO_F_MIN_RATE ||
449 	     priv->flags & TC_MQPRIO_F_MAX_RATE) &&
450 	    (dump_rates(priv, &opt, skb) != 0))
451 		goto nla_put_failure;
452 
453 	return nla_nest_end(skb, nla);
454 nla_put_failure:
455 	nlmsg_trim(skb, nla);
456 	return -1;
457 }
458 
mqprio_leaf(struct Qdisc * sch,unsigned long cl)459 static struct Qdisc *mqprio_leaf(struct Qdisc *sch, unsigned long cl)
460 {
461 	struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
462 
463 	if (!dev_queue)
464 		return NULL;
465 
466 	return dev_queue->qdisc_sleeping;
467 }
468 
mqprio_find(struct Qdisc * sch,u32 classid)469 static unsigned long mqprio_find(struct Qdisc *sch, u32 classid)
470 {
471 	struct net_device *dev = qdisc_dev(sch);
472 	unsigned int ntx = TC_H_MIN(classid);
473 
474 	/* There are essentially two regions here that have valid classid
475 	 * values. The first region will have a classid value of 1 through
476 	 * num_tx_queues. All of these are backed by actual Qdiscs.
477 	 */
478 	if (ntx < TC_H_MIN_PRIORITY)
479 		return (ntx <= dev->num_tx_queues) ? ntx : 0;
480 
481 	/* The second region represents the hardware traffic classes. These
482 	 * are represented by classid values of TC_H_MIN_PRIORITY through
483 	 * TC_H_MIN_PRIORITY + netdev_get_num_tc - 1
484 	 */
485 	return ((ntx - TC_H_MIN_PRIORITY) < netdev_get_num_tc(dev)) ? ntx : 0;
486 }
487 
mqprio_dump_class(struct Qdisc * sch,unsigned long cl,struct sk_buff * skb,struct tcmsg * tcm)488 static int mqprio_dump_class(struct Qdisc *sch, unsigned long cl,
489 			 struct sk_buff *skb, struct tcmsg *tcm)
490 {
491 	if (cl < TC_H_MIN_PRIORITY) {
492 		struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
493 		struct net_device *dev = qdisc_dev(sch);
494 		int tc = netdev_txq_to_tc(dev, cl - 1);
495 
496 		tcm->tcm_parent = (tc < 0) ? 0 :
497 			TC_H_MAKE(TC_H_MAJ(sch->handle),
498 				  TC_H_MIN(tc + TC_H_MIN_PRIORITY));
499 		tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
500 	} else {
501 		tcm->tcm_parent = TC_H_ROOT;
502 		tcm->tcm_info = 0;
503 	}
504 	tcm->tcm_handle |= TC_H_MIN(cl);
505 	return 0;
506 }
507 
mqprio_dump_class_stats(struct Qdisc * sch,unsigned long cl,struct gnet_dump * d)508 static int mqprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
509 				   struct gnet_dump *d)
510 	__releases(d->lock)
511 	__acquires(d->lock)
512 {
513 	if (cl >= TC_H_MIN_PRIORITY) {
514 		int i;
515 		__u32 qlen = 0;
516 		struct gnet_stats_queue qstats = {0};
517 		struct gnet_stats_basic_packed bstats = {0};
518 		struct net_device *dev = qdisc_dev(sch);
519 		struct netdev_tc_txq tc = dev->tc_to_txq[cl & TC_BITMASK];
520 
521 		/* Drop lock here it will be reclaimed before touching
522 		 * statistics this is required because the d->lock we
523 		 * hold here is the look on dev_queue->qdisc_sleeping
524 		 * also acquired below.
525 		 */
526 		if (d->lock)
527 			spin_unlock_bh(d->lock);
528 
529 		for (i = tc.offset; i < tc.offset + tc.count; i++) {
530 			struct netdev_queue *q = netdev_get_tx_queue(dev, i);
531 			struct Qdisc *qdisc = rtnl_dereference(q->qdisc);
532 			struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
533 			struct gnet_stats_queue __percpu *cpu_qstats = NULL;
534 
535 			spin_lock_bh(qdisc_lock(qdisc));
536 			if (qdisc_is_percpu_stats(qdisc)) {
537 				cpu_bstats = qdisc->cpu_bstats;
538 				cpu_qstats = qdisc->cpu_qstats;
539 			}
540 
541 			qlen = qdisc_qlen_sum(qdisc);
542 			__gnet_stats_copy_basic(NULL, &sch->bstats,
543 						cpu_bstats, &qdisc->bstats);
544 			__gnet_stats_copy_queue(&sch->qstats,
545 						cpu_qstats,
546 						&qdisc->qstats,
547 						qlen);
548 			spin_unlock_bh(qdisc_lock(qdisc));
549 		}
550 
551 		/* Reclaim root sleeping lock before completing stats */
552 		if (d->lock)
553 			spin_lock_bh(d->lock);
554 		if (gnet_stats_copy_basic(NULL, d, NULL, &bstats) < 0 ||
555 		    gnet_stats_copy_queue(d, NULL, &qstats, qlen) < 0)
556 			return -1;
557 	} else {
558 		struct netdev_queue *dev_queue = mqprio_queue_get(sch, cl);
559 
560 		sch = dev_queue->qdisc_sleeping;
561 		if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch), d,
562 					  sch->cpu_bstats, &sch->bstats) < 0 ||
563 		    qdisc_qstats_copy(d, sch) < 0)
564 			return -1;
565 	}
566 	return 0;
567 }
568 
mqprio_walk(struct Qdisc * sch,struct qdisc_walker * arg)569 static void mqprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
570 {
571 	struct net_device *dev = qdisc_dev(sch);
572 	unsigned long ntx;
573 
574 	if (arg->stop)
575 		return;
576 
577 	/* Walk hierarchy with a virtual class per tc */
578 	arg->count = arg->skip;
579 	for (ntx = arg->skip; ntx < netdev_get_num_tc(dev); ntx++) {
580 		if (arg->fn(sch, ntx + TC_H_MIN_PRIORITY, arg) < 0) {
581 			arg->stop = 1;
582 			return;
583 		}
584 		arg->count++;
585 	}
586 
587 	/* Pad the values and skip over unused traffic classes */
588 	if (ntx < TC_MAX_QUEUE) {
589 		arg->count = TC_MAX_QUEUE;
590 		ntx = TC_MAX_QUEUE;
591 	}
592 
593 	/* Reset offset, sort out remaining per-queue qdiscs */
594 	for (ntx -= TC_MAX_QUEUE; ntx < dev->num_tx_queues; ntx++) {
595 		if (arg->fn(sch, ntx + 1, arg) < 0) {
596 			arg->stop = 1;
597 			return;
598 		}
599 		arg->count++;
600 	}
601 }
602 
mqprio_select_queue(struct Qdisc * sch,struct tcmsg * tcm)603 static struct netdev_queue *mqprio_select_queue(struct Qdisc *sch,
604 						struct tcmsg *tcm)
605 {
606 	return mqprio_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
607 }
608 
609 static const struct Qdisc_class_ops mqprio_class_ops = {
610 	.graft		= mqprio_graft,
611 	.leaf		= mqprio_leaf,
612 	.find		= mqprio_find,
613 	.walk		= mqprio_walk,
614 	.dump		= mqprio_dump_class,
615 	.dump_stats	= mqprio_dump_class_stats,
616 	.select_queue	= mqprio_select_queue,
617 };
618 
619 static struct Qdisc_ops mqprio_qdisc_ops __read_mostly = {
620 	.cl_ops		= &mqprio_class_ops,
621 	.id		= "mqprio",
622 	.priv_size	= sizeof(struct mqprio_sched),
623 	.init		= mqprio_init,
624 	.destroy	= mqprio_destroy,
625 	.attach		= mqprio_attach,
626 	.dump		= mqprio_dump,
627 	.owner		= THIS_MODULE,
628 };
629 
mqprio_module_init(void)630 static int __init mqprio_module_init(void)
631 {
632 	return register_qdisc(&mqprio_qdisc_ops);
633 }
634 
mqprio_module_exit(void)635 static void __exit mqprio_module_exit(void)
636 {
637 	unregister_qdisc(&mqprio_qdisc_ops);
638 }
639 
640 module_init(mqprio_module_init);
641 module_exit(mqprio_module_exit);
642 
643 MODULE_LICENSE("GPL");
644