aboutsummaryrefslogtreecommitdiffstats
path: root/net/sched
diff options
context:
space:
mode:
authorPeter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>2007-06-29 00:04:31 -0400
committerDavid S. Miller <davem@sunset.davemloft.net>2007-07-11 01:16:22 -0400
commitd62733c8e437fdb58325617c4b3331769ba82d70 (patch)
treefcbef07b66cda79b047092e7d53dfd005046838a /net/sched
parentf25f4e44808f0f6c9875d94ef1c41ef86c288eb2 (diff)
[SCHED]: Qdisc changes and sch_rr added for multiqueue
Add the new sch_rr qdisc for multiqueue network device support. Allow sch_prio and sch_rr to be compiled with or without multiqueue hardware support. sch_rr is part of sch_prio, and is referenced from MODULE_ALIAS. This was done since sch_prio and sch_rr only differ in their dequeue routine. Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com> Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/sched')
-rw-r--r--net/sched/Kconfig11
-rw-r--r--net/sched/sch_prio.c129
2 files changed, 125 insertions, 15 deletions
diff --git a/net/sched/Kconfig b/net/sched/Kconfig
index 475df8449be9..f3217942ca87 100644
--- a/net/sched/Kconfig
+++ b/net/sched/Kconfig
@@ -111,6 +111,17 @@ config NET_SCH_PRIO
111 To compile this code as a module, choose M here: the 111 To compile this code as a module, choose M here: the
112 module will be called sch_prio. 112 module will be called sch_prio.
113 113
114config NET_SCH_RR
115 tristate "Multi Band Round Robin Queuing (RR)"
116 select NET_SCH_PRIO
117 ---help---
118 Say Y here if you want to use an n-band round robin packet
119 scheduler.
120
121 The module uses sch_prio for its framework and is aliased as
122 sch_rr, so it will load sch_prio, although it is referred
123 to using sch_rr.
124
114config NET_SCH_RED 125config NET_SCH_RED
115 tristate "Random Early Detection (RED)" 126 tristate "Random Early Detection (RED)"
116 ---help--- 127 ---help---
diff --git a/net/sched/sch_prio.c b/net/sched/sch_prio.c
index 6d7542c26e47..404522046289 100644
--- a/net/sched/sch_prio.c
+++ b/net/sched/sch_prio.c
@@ -40,9 +40,11 @@
40struct prio_sched_data 40struct prio_sched_data
41{ 41{
42 int bands; 42 int bands;
43 int curband; /* for round-robin */
43 struct tcf_proto *filter_list; 44 struct tcf_proto *filter_list;
44 u8 prio2band[TC_PRIO_MAX+1]; 45 u8 prio2band[TC_PRIO_MAX+1];
45 struct Qdisc *queues[TCQ_PRIO_BANDS]; 46 struct Qdisc *queues[TCQ_PRIO_BANDS];
47 int mq;
46}; 48};
47 49
48 50
@@ -70,14 +72,17 @@ prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
70#endif 72#endif
71 if (TC_H_MAJ(band)) 73 if (TC_H_MAJ(band))
72 band = 0; 74 band = 0;
73 return q->queues[q->prio2band[band&TC_PRIO_MAX]]; 75 band = q->prio2band[band&TC_PRIO_MAX];
76 goto out;
74 } 77 }
75 band = res.classid; 78 band = res.classid;
76 } 79 }
77 band = TC_H_MIN(band) - 1; 80 band = TC_H_MIN(band) - 1;
78 if (band >= q->bands) 81 if (band >= q->bands)
79 return q->queues[q->prio2band[0]]; 82 band = q->prio2band[0];
80 83out:
84 if (q->mq)
85 skb_set_queue_mapping(skb, band);
81 return q->queues[band]; 86 return q->queues[band];
82} 87}
83 88
@@ -144,17 +149,58 @@ prio_dequeue(struct Qdisc* sch)
144 struct Qdisc *qdisc; 149 struct Qdisc *qdisc;
145 150
146 for (prio = 0; prio < q->bands; prio++) { 151 for (prio = 0; prio < q->bands; prio++) {
147 qdisc = q->queues[prio]; 152 /* Check if the target subqueue is available before
148 skb = qdisc->dequeue(qdisc); 153 * pulling an skb. This way we avoid excessive requeues
149 if (skb) { 154 * for slower queues.
150 sch->q.qlen--; 155 */
151 return skb; 156 if (!netif_subqueue_stopped(sch->dev, (q->mq ? prio : 0))) {
157 qdisc = q->queues[prio];
158 skb = qdisc->dequeue(qdisc);
159 if (skb) {
160 sch->q.qlen--;
161 return skb;
162 }
152 } 163 }
153 } 164 }
154 return NULL; 165 return NULL;
155 166
156} 167}
157 168
169static struct sk_buff *rr_dequeue(struct Qdisc* sch)
170{
171 struct sk_buff *skb;
172 struct prio_sched_data *q = qdisc_priv(sch);
173 struct Qdisc *qdisc;
174 int bandcount;
175
176 /* Only take one pass through the queues. If nothing is available,
177 * return nothing.
178 */
179 for (bandcount = 0; bandcount < q->bands; bandcount++) {
180 /* Check if the target subqueue is available before
181 * pulling an skb. This way we avoid excessive requeues
182 * for slower queues. If the queue is stopped, try the
183 * next queue.
184 */
185 if (!netif_subqueue_stopped(sch->dev,
186 (q->mq ? q->curband : 0))) {
187 qdisc = q->queues[q->curband];
188 skb = qdisc->dequeue(qdisc);
189 if (skb) {
190 sch->q.qlen--;
191 q->curband++;
192 if (q->curband >= q->bands)
193 q->curband = 0;
194 return skb;
195 }
196 }
197 q->curband++;
198 if (q->curband >= q->bands)
199 q->curband = 0;
200 }
201 return NULL;
202}
203
158static unsigned int prio_drop(struct Qdisc* sch) 204static unsigned int prio_drop(struct Qdisc* sch)
159{ 205{
160 struct prio_sched_data *q = qdisc_priv(sch); 206 struct prio_sched_data *q = qdisc_priv(sch);
@@ -198,21 +244,41 @@ prio_destroy(struct Qdisc* sch)
198static int prio_tune(struct Qdisc *sch, struct rtattr *opt) 244static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
199{ 245{
200 struct prio_sched_data *q = qdisc_priv(sch); 246 struct prio_sched_data *q = qdisc_priv(sch);
201 struct tc_prio_qopt *qopt = RTA_DATA(opt); 247 struct tc_prio_qopt *qopt;
248 struct rtattr *tb[TCA_PRIO_MAX];
202 int i; 249 int i;
203 250
204 if (opt->rta_len < RTA_LENGTH(sizeof(*qopt))) 251 if (rtattr_parse_nested_compat(tb, TCA_PRIO_MAX, opt, qopt,
252 sizeof(*qopt)))
205 return -EINVAL; 253 return -EINVAL;
206 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2) 254 q->bands = qopt->bands;
255 /* If we're multiqueue, make sure the number of incoming bands
256 * matches the number of queues on the device we're associating with.
257 * If the number of bands requested is zero, then set q->bands to
258 * dev->egress_subqueue_count.
259 */
260 q->mq = RTA_GET_FLAG(tb[TCA_PRIO_MQ - 1]);
261 if (q->mq) {
262 if (sch->handle != TC_H_ROOT)
263 return -EINVAL;
264 if (netif_is_multiqueue(sch->dev)) {
265 if (q->bands == 0)
266 q->bands = sch->dev->egress_subqueue_count;
267 else if (q->bands != sch->dev->egress_subqueue_count)
268 return -EINVAL;
269 } else
270 return -EOPNOTSUPP;
271 }
272
273 if (q->bands > TCQ_PRIO_BANDS || q->bands < 2)
207 return -EINVAL; 274 return -EINVAL;
208 275
209 for (i=0; i<=TC_PRIO_MAX; i++) { 276 for (i=0; i<=TC_PRIO_MAX; i++) {
210 if (qopt->priomap[i] >= qopt->bands) 277 if (qopt->priomap[i] >= q->bands)
211 return -EINVAL; 278 return -EINVAL;
212 } 279 }
213 280
214 sch_tree_lock(sch); 281 sch_tree_lock(sch);
215 q->bands = qopt->bands;
216 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1); 282 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
217 283
218 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) { 284 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
@@ -268,11 +334,17 @@ static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
268{ 334{
269 struct prio_sched_data *q = qdisc_priv(sch); 335 struct prio_sched_data *q = qdisc_priv(sch);
270 unsigned char *b = skb_tail_pointer(skb); 336 unsigned char *b = skb_tail_pointer(skb);
337 struct rtattr *nest;
271 struct tc_prio_qopt opt; 338 struct tc_prio_qopt opt;
272 339
273 opt.bands = q->bands; 340 opt.bands = q->bands;
274 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1); 341 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
275 RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); 342
343 nest = RTA_NEST_COMPAT(skb, TCA_OPTIONS, sizeof(opt), &opt);
344 if (q->mq)
345 RTA_PUT_FLAG(skb, TCA_PRIO_MQ);
346 RTA_NEST_COMPAT_END(skb, nest);
347
276 return skb->len; 348 return skb->len;
277 349
278rtattr_failure: 350rtattr_failure:
@@ -443,17 +515,44 @@ static struct Qdisc_ops prio_qdisc_ops = {
443 .owner = THIS_MODULE, 515 .owner = THIS_MODULE,
444}; 516};
445 517
518static struct Qdisc_ops rr_qdisc_ops = {
519 .next = NULL,
520 .cl_ops = &prio_class_ops,
521 .id = "rr",
522 .priv_size = sizeof(struct prio_sched_data),
523 .enqueue = prio_enqueue,
524 .dequeue = rr_dequeue,
525 .requeue = prio_requeue,
526 .drop = prio_drop,
527 .init = prio_init,
528 .reset = prio_reset,
529 .destroy = prio_destroy,
530 .change = prio_tune,
531 .dump = prio_dump,
532 .owner = THIS_MODULE,
533};
534
446static int __init prio_module_init(void) 535static int __init prio_module_init(void)
447{ 536{
448 return register_qdisc(&prio_qdisc_ops); 537 int err;
538
539 err = register_qdisc(&prio_qdisc_ops);
540 if (err < 0)
541 return err;
542 err = register_qdisc(&rr_qdisc_ops);
543 if (err < 0)
544 unregister_qdisc(&prio_qdisc_ops);
545 return err;
449} 546}
450 547
451static void __exit prio_module_exit(void) 548static void __exit prio_module_exit(void)
452{ 549{
453 unregister_qdisc(&prio_qdisc_ops); 550 unregister_qdisc(&prio_qdisc_ops);
551 unregister_qdisc(&rr_qdisc_ops);
454} 552}
455 553
456module_init(prio_module_init) 554module_init(prio_module_init)
457module_exit(prio_module_exit) 555module_exit(prio_module_exit)
458 556
459MODULE_LICENSE("GPL"); 557MODULE_LICENSE("GPL");
558MODULE_ALIAS("sch_rr");