blob: a74e278654aa2290a0ef3061d1392c5cf405e5f5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/sch_generic.c Generic packet scheduler routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * Jamal Hadi Salim, <hadi@cyberus.ca> 990601
11 * - Ingress support
12 */
13
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/module.h>
16#include <linux/types.h>
17#include <linux/kernel.h>
18#include <linux/sched.h>
19#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/netdevice.h>
22#include <linux/skbuff.h>
23#include <linux/rtnetlink.h>
24#include <linux/init.h>
25#include <linux/rcupdate.h>
26#include <linux/list.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090027#include <linux/slab.h>
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +020028#include <linux/if_vlan.h>
Jiri Pirko292f1c72013-02-12 00:12:03 +000029#include <net/sch_generic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <net/pkt_sched.h>
Eric Dumazet7fee2262010-05-11 23:19:48 +000031#include <net/dst.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
stephen hemminger34aedd32013-08-31 10:15:33 -070033/* Qdisc to use by default */
34const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
35EXPORT_SYMBOL(default_qdisc_ops);
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/* Main transmission queue. */
38
Patrick McHardy0463d4a2007-04-16 17:02:10 -070039/* Modifications to data participating in scheduling must be protected with
David S. Miller5fb66222008-08-02 20:02:43 -070040 * qdisc_lock(qdisc) spinlock.
Patrick McHardy0463d4a2007-04-16 17:02:10 -070041 *
42 * The idea is the following:
David S. Millerc7e4f3b2008-07-16 03:22:39 -070043 * - enqueue, dequeue are serialized via qdisc root lock
44 * - ingress filtering is also serialized via qdisc root lock
Patrick McHardy0463d4a2007-04-16 17:02:10 -070045 * - updates to tree and tree walking are only done under the rtnl mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
David S. Miller37437bb2008-07-16 02:15:04 -070048static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070049{
Eric Dumazet7fee2262010-05-11 23:19:48 +000050 skb_dst_force(skb);
Jarek Poplawski62523522008-10-06 10:41:50 -070051 q->gso_skb = skb;
Jarek Poplawski53e91502008-10-08 11:36:22 -070052 q->qstats.requeues++;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +000053 q->q.qlen++; /* it's still part of the queue */
David S. Miller37437bb2008-07-16 02:15:04 -070054 __netif_schedule(q);
Jarek Poplawski62523522008-10-06 10:41:50 -070055
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070056 return 0;
57}
58
David S. Millerd3b753d2008-07-15 20:14:35 -070059static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070060{
Jarek Poplawski554794d2008-10-06 09:54:39 -070061 struct sk_buff *skb = q->gso_skb;
Eric Dumazet1abbe132012-12-11 15:54:33 +000062 const struct netdev_queue *txq = q->dev_queue;
Jarek Poplawski554794d2008-10-06 09:54:39 -070063
Jarek Poplawskiebf05982008-09-22 22:16:23 -070064 if (unlikely(skb)) {
Jarek Poplawskiebf05982008-09-22 22:16:23 -070065 /* check the reason of requeuing without tx lock first */
Eric Dumazet1abbe132012-12-11 15:54:33 +000066 txq = netdev_get_tx_queue(txq->dev, skb_get_queue_mapping(skb));
Tom Herbert734664982011-11-28 16:32:44 +000067 if (!netif_xmit_frozen_or_stopped(txq)) {
Jarek Poplawski62523522008-10-06 10:41:50 -070068 q->gso_skb = NULL;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +000069 q->q.qlen--;
70 } else
Jarek Poplawskiebf05982008-09-22 22:16:23 -070071 skb = NULL;
72 } else {
Eric Dumazet1abbe132012-12-11 15:54:33 +000073 if (!(q->flags & TCQ_F_ONETXQUEUE) || !netif_xmit_frozen_or_stopped(txq))
74 skb = q->dequeue(q);
Jarek Poplawskiebf05982008-09-22 22:16:23 -070075 }
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070076
77 return skb;
78}
79
Krishna Kumar6c1361a2007-06-24 19:56:09 -070080static inline int handle_dev_cpu_collision(struct sk_buff *skb,
David S. Miller970565b2008-07-08 23:10:33 -070081 struct netdev_queue *dev_queue,
Krishna Kumar6c1361a2007-06-24 19:56:09 -070082 struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070083{
Krishna Kumar6c1361a2007-06-24 19:56:09 -070084 int ret;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070085
David S. Millerc773e842008-07-08 23:13:53 -070086 if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
Krishna Kumar6c1361a2007-06-24 19:56:09 -070087 /*
88 * Same CPU holding the lock. It may be a transient
89 * configuration error, when hard_start_xmit() recurses. We
90 * detect it by checking xmit owner and drop the packet when
91 * deadloop is detected. Return OK to try the next skb.
92 */
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070093 kfree_skb(skb);
Joe Perchese87cc472012-05-13 21:56:26 +000094 net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n",
95 dev_queue->dev->name);
Krishna Kumar6c1361a2007-06-24 19:56:09 -070096 ret = qdisc_qlen(q);
97 } else {
98 /*
99 * Another cpu is holding lock, requeue & delay xmits for
100 * some time.
101 */
Eric Dumazetd6d9ca02010-07-19 10:48:49 +0000102 __this_cpu_inc(softnet_data.cpu_collision);
David S. Miller37437bb2008-07-16 02:15:04 -0700103 ret = dev_requeue_skb(skb, q);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700104 }
105
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700106 return ret;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700107}
108
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900109/*
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000110 * Transmit one skb, and handle the return status as required. Holding the
111 * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this
112 * function.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700113 *
114 * Returns to the caller:
115 * 0 - queue is empty or throttled.
116 * >0 - queue is not empty.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700117 */
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000118int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
119 struct net_device *dev, struct netdev_queue *txq,
120 spinlock_t *root_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Peter P Waskiewicz Jr5f1a4852007-11-13 20:40:55 -0800122 int ret = NETDEV_TX_BUSY;
David S. Miller7698b4f2008-07-16 01:42:40 -0700123
124 /* And release qdisc */
125 spin_unlock(root_lock);
Herbert Xud90df3a2007-05-10 04:55:14 -0700126
David S. Millerc773e842008-07-08 23:13:53 -0700127 HARD_TX_LOCK(dev, txq, smp_processor_id());
Tom Herbert734664982011-11-28 16:32:44 +0000128 if (!netif_xmit_frozen_or_stopped(txq))
David S. Millerfd2ea0a2008-07-17 01:56:23 -0700129 ret = dev_hard_start_xmit(skb, dev, txq);
Patrick McHardy572a9d72009-11-10 06:14:14 +0000130
David S. Millerc773e842008-07-08 23:13:53 -0700131 HARD_TX_UNLOCK(dev, txq);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700132
David S. Miller7698b4f2008-07-16 01:42:40 -0700133 spin_lock(root_lock);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700134
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000135 if (dev_xmit_complete(ret)) {
136 /* Driver sent out skb successfully or skb was consumed */
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700137 ret = qdisc_qlen(q);
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000138 } else if (ret == NETDEV_TX_LOCKED) {
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700139 /* Driver try lock failed */
David S. Millereb6aafe2008-07-08 23:12:38 -0700140 ret = handle_dev_cpu_collision(skb, txq, q);
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000141 } else {
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700142 /* Driver returned NETDEV_TX_BUSY - requeue skb */
Joe Perchese87cc472012-05-13 21:56:26 +0000143 if (unlikely(ret != NETDEV_TX_BUSY))
144 net_warn_ratelimited("BUG %s code %d qlen %d\n",
145 dev->name, ret, q->q.qlen);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700146
David S. Miller37437bb2008-07-16 02:15:04 -0700147 ret = dev_requeue_skb(skb, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700148 }
149
Tom Herbert734664982011-11-28 16:32:44 +0000150 if (ret && netif_xmit_frozen_or_stopped(txq))
David S. Miller37437bb2008-07-16 02:15:04 -0700151 ret = 0;
152
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700153 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
155
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000156/*
157 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
158 *
159 * __QDISC_STATE_RUNNING guarantees only one CPU can process
160 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
161 * this queue.
162 *
163 * netif_tx_lock serializes accesses to device driver.
164 *
165 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
166 * if one is grabbed, another must be free.
167 *
168 * Note, that this procedure can be called by a watchdog timer
169 *
170 * Returns to the caller:
171 * 0 - queue is empty or throttled.
172 * >0 - queue is not empty.
173 *
174 */
175static inline int qdisc_restart(struct Qdisc *q)
176{
177 struct netdev_queue *txq;
178 struct net_device *dev;
179 spinlock_t *root_lock;
180 struct sk_buff *skb;
181
182 /* Dequeue packet */
183 skb = dequeue_skb(q);
184 if (unlikely(!skb))
185 return 0;
Eric Dumazet7fee2262010-05-11 23:19:48 +0000186 WARN_ON_ONCE(skb_dst_is_noref(skb));
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000187 root_lock = qdisc_lock(q);
188 dev = qdisc_dev(q);
189 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
190
191 return sch_direct_xmit(skb, q, dev, txq, root_lock);
192}
193
David S. Miller37437bb2008-07-16 02:15:04 -0700194void __qdisc_run(struct Qdisc *q)
Herbert Xu48d83322006-06-19 23:57:59 -0700195{
jamald5b8aa12011-06-26 08:13:54 +0000196 int quota = weight_p;
Herbert Xu2ba25062008-03-28 16:25:26 -0700197
David S. Miller37437bb2008-07-16 02:15:04 -0700198 while (qdisc_restart(q)) {
Herbert Xu2ba25062008-03-28 16:25:26 -0700199 /*
jamald5b8aa12011-06-26 08:13:54 +0000200 * Ordered by possible occurrence: Postpone processing if
201 * 1. we've exceeded packet quota
202 * 2. another process needs the CPU;
Herbert Xu2ba25062008-03-28 16:25:26 -0700203 */
Krishna Kumarf0c50c72011-07-14 23:16:21 +0000204 if (--quota <= 0 || need_resched()) {
David S. Miller37437bb2008-07-16 02:15:04 -0700205 __netif_schedule(q);
Herbert Xu2ba25062008-03-28 16:25:26 -0700206 break;
207 }
208 }
Herbert Xu48d83322006-06-19 23:57:59 -0700209
Eric Dumazetbc135b22010-06-02 03:23:51 -0700210 qdisc_run_end(q);
Herbert Xu48d83322006-06-19 23:57:59 -0700211}
212
Eric Dumazet9d214932009-05-17 20:55:16 -0700213unsigned long dev_trans_start(struct net_device *dev)
214{
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200215 unsigned long val, res;
Eric Dumazet9d214932009-05-17 20:55:16 -0700216 unsigned int i;
217
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200218 if (is_vlan_dev(dev))
219 dev = vlan_dev_real_dev(dev);
220 res = dev->trans_start;
Eric Dumazet9d214932009-05-17 20:55:16 -0700221 for (i = 0; i < dev->num_tx_queues; i++) {
222 val = netdev_get_tx_queue(dev, i)->trans_start;
223 if (val && time_after(val, res))
224 res = val;
225 }
226 dev->trans_start = res;
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200227
Eric Dumazet9d214932009-05-17 20:55:16 -0700228 return res;
229}
230EXPORT_SYMBOL(dev_trans_start);
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232static void dev_watchdog(unsigned long arg)
233{
234 struct net_device *dev = (struct net_device *)arg;
235
Herbert Xu932ff272006-06-09 12:20:56 -0700236 netif_tx_lock(dev);
David S. Millere8a04642008-07-17 00:34:19 -0700237 if (!qdisc_tx_is_noop(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 if (netif_device_present(dev) &&
239 netif_running(dev) &&
240 netif_carrier_ok(dev)) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700241 int some_queue_timedout = 0;
David S. Millere8a04642008-07-17 00:34:19 -0700242 unsigned int i;
Eric Dumazet9d214932009-05-17 20:55:16 -0700243 unsigned long trans_start;
Stephen Hemminger338f7562006-05-16 15:02:12 -0700244
David S. Millere8a04642008-07-17 00:34:19 -0700245 for (i = 0; i < dev->num_tx_queues; i++) {
246 struct netdev_queue *txq;
247
248 txq = netdev_get_tx_queue(dev, i);
Eric Dumazet9d214932009-05-17 20:55:16 -0700249 /*
250 * old device drivers set dev->trans_start
251 */
252 trans_start = txq->trans_start ? : dev->trans_start;
Tom Herbert734664982011-11-28 16:32:44 +0000253 if (netif_xmit_stopped(txq) &&
Eric Dumazet9d214932009-05-17 20:55:16 -0700254 time_after(jiffies, (trans_start +
255 dev->watchdog_timeo))) {
256 some_queue_timedout = 1;
david decotignyccf5ff62011-11-16 12:15:10 +0000257 txq->trans_timeout++;
David S. Millere8a04642008-07-17 00:34:19 -0700258 break;
259 }
260 }
261
Eric Dumazet9d214932009-05-17 20:55:16 -0700262 if (some_queue_timedout) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700263 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
David S. Miller3019de12011-06-06 16:41:33 -0700264 dev->name, netdev_drivername(dev), i);
Stephen Hemmingerd3147742008-11-19 21:32:24 -0800265 dev->netdev_ops->ndo_tx_timeout(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 }
David S. Millere8a04642008-07-17 00:34:19 -0700267 if (!mod_timer(&dev->watchdog_timer,
268 round_jiffies(jiffies +
269 dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 dev_hold(dev);
271 }
272 }
Herbert Xu932ff272006-06-09 12:20:56 -0700273 netif_tx_unlock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275 dev_put(dev);
276}
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278void __netdev_watchdog_up(struct net_device *dev)
279{
Stephen Hemmingerd3147742008-11-19 21:32:24 -0800280 if (dev->netdev_ops->ndo_tx_timeout) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 if (dev->watchdog_timeo <= 0)
282 dev->watchdog_timeo = 5*HZ;
Venkatesh Pallipadi60468d52007-05-31 21:28:44 -0700283 if (!mod_timer(&dev->watchdog_timer,
284 round_jiffies(jiffies + dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 dev_hold(dev);
286 }
287}
288
289static void dev_watchdog_up(struct net_device *dev)
290{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 __netdev_watchdog_up(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292}
293
294static void dev_watchdog_down(struct net_device *dev)
295{
Herbert Xu932ff272006-06-09 12:20:56 -0700296 netif_tx_lock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 if (del_timer(&dev->watchdog_timer))
Stephen Hemminger15333062006-03-20 22:32:28 -0800298 dev_put(dev);
Herbert Xu932ff272006-06-09 12:20:56 -0700299 netif_tx_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
301
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700302/**
303 * netif_carrier_on - set carrier
304 * @dev: network device
305 *
306 * Device has detected that carrier.
307 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700308void netif_carrier_on(struct net_device *dev)
309{
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700310 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
David S. Millerb4730012008-11-19 15:33:54 -0800311 if (dev->reg_state == NETREG_UNINITIALIZED)
312 return;
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700313 linkwatch_fire_event(dev);
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700314 if (netif_running(dev))
315 __netdev_watchdog_up(dev);
316 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700317}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800318EXPORT_SYMBOL(netif_carrier_on);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700319
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700320/**
321 * netif_carrier_off - clear carrier
322 * @dev: network device
323 *
324 * Device has detected loss of carrier.
325 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700326void netif_carrier_off(struct net_device *dev)
327{
David S. Millerb4730012008-11-19 15:33:54 -0800328 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
329 if (dev->reg_state == NETREG_UNINITIALIZED)
330 return;
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700331 linkwatch_fire_event(dev);
David S. Millerb4730012008-11-19 15:33:54 -0800332 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700333}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800334EXPORT_SYMBOL(netif_carrier_off);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700335
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
337 under all circumstances. It is difficult to invent anything faster or
338 cheaper.
339 */
340
Thomas Graf94df1092005-06-18 22:59:08 -0700341static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
343 kfree_skb(skb);
344 return NET_XMIT_CN;
345}
346
Thomas Graf94df1092005-06-18 22:59:08 -0700347static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
349 return NULL;
350}
351
Eric Dumazet20fea082007-11-14 01:44:41 -0800352struct Qdisc_ops noop_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 .id = "noop",
354 .priv_size = 0,
355 .enqueue = noop_enqueue,
356 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700357 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 .owner = THIS_MODULE,
359};
360
David S. Miller7698b4f2008-07-16 01:42:40 -0700361static struct netdev_queue noop_netdev_queue = {
David S. Miller7698b4f2008-07-16 01:42:40 -0700362 .qdisc = &noop_qdisc,
Jarek Poplawski9f3ffae2008-10-19 23:37:47 -0700363 .qdisc_sleeping = &noop_qdisc,
David S. Miller7698b4f2008-07-16 01:42:40 -0700364};
365
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366struct Qdisc noop_qdisc = {
367 .enqueue = noop_enqueue,
368 .dequeue = noop_dequeue,
369 .flags = TCQ_F_BUILTIN,
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900370 .ops = &noop_qdisc_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 .list = LIST_HEAD_INIT(noop_qdisc.list),
David S. Miller83874002008-07-17 00:53:03 -0700372 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
David S. Miller7698b4f2008-07-16 01:42:40 -0700373 .dev_queue = &noop_netdev_queue,
Eric Dumazet7b5edbc2010-10-15 19:22:34 +0000374 .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375};
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800376EXPORT_SYMBOL(noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Eric Dumazet20fea082007-11-14 01:44:41 -0800378static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 .id = "noqueue",
380 .priv_size = 0,
381 .enqueue = noop_enqueue,
382 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700383 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 .owner = THIS_MODULE,
385};
386
David S. Miller30ee42b2008-07-18 23:00:11 -0700387static struct Qdisc noqueue_qdisc;
388static struct netdev_queue noqueue_netdev_queue = {
389 .qdisc = &noqueue_qdisc,
Jarek Poplawski9f3ffae2008-10-19 23:37:47 -0700390 .qdisc_sleeping = &noqueue_qdisc,
David S. Miller30ee42b2008-07-18 23:00:11 -0700391};
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393static struct Qdisc noqueue_qdisc = {
394 .enqueue = NULL,
395 .dequeue = noop_dequeue,
396 .flags = TCQ_F_BUILTIN,
397 .ops = &noqueue_qdisc_ops,
398 .list = LIST_HEAD_INIT(noqueue_qdisc.list),
David S. Miller30ee42b2008-07-18 23:00:11 -0700399 .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
400 .dev_queue = &noqueue_netdev_queue,
Eric Dumazet7b5edbc2010-10-15 19:22:34 +0000401 .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402};
403
404
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000405static const u8 prio2band[TC_PRIO_MAX + 1] = {
406 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
407};
Thomas Graf321090e2005-06-18 22:58:35 -0700408
David S. Millerd3678b42008-07-21 09:56:13 -0700409/* 3-band FIFO queue: old style, but should be a bit faster than
410 generic prio+fifo combination.
411 */
412
413#define PFIFO_FAST_BANDS 3
414
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000415/*
416 * Private data for a pfifo_fast scheduler containing:
417 * - queues for the three band
418 * - bitmap indicating which of the bands contain skbs
419 */
420struct pfifo_fast_priv {
421 u32 bitmap;
422 struct sk_buff_head q[PFIFO_FAST_BANDS];
423};
424
425/*
426 * Convert a bitmap to the first band number where an skb is queued, where:
427 * bitmap=0 means there are no skbs on any band.
428 * bitmap=1 means there is an skb on band 0.
429 * bitmap=7 means there are skbs on all 3 bands, etc.
430 */
431static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
432
433static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
434 int band)
David S. Millerd3678b42008-07-21 09:56:13 -0700435{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000436 return priv->q + band;
David S. Millerd3678b42008-07-21 09:56:13 -0700437}
438
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000439static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc)
David S. Millerd3678b42008-07-21 09:56:13 -0700440{
Krishna Kumara453e062009-08-30 22:20:28 -0700441 if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) {
442 int band = prio2band[skb->priority & TC_PRIO_MAX];
443 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
444 struct sk_buff_head *list = band2list(priv, band);
David S. Millerd3678b42008-07-21 09:56:13 -0700445
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000446 priv->bitmap |= (1 << band);
David S. Millerd3678b42008-07-21 09:56:13 -0700447 qdisc->q.qlen++;
Thomas Graf821d24a2005-06-18 22:58:15 -0700448 return __qdisc_enqueue_tail(skb, qdisc, list);
David S. Millerd3678b42008-07-21 09:56:13 -0700449 }
Thomas Graf821d24a2005-06-18 22:58:15 -0700450
451 return qdisc_drop(skb, qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000454static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000456 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
457 int band = bitmap2band[priv->bitmap];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000459 if (likely(band >= 0)) {
460 struct sk_buff_head *list = band2list(priv, band);
461 struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);
462
463 qdisc->q.qlen--;
464 if (skb_queue_empty(list))
465 priv->bitmap &= ~(1 << band);
466
467 return skb;
David S. Millerd3678b42008-07-21 09:56:13 -0700468 }
Thomas Graff87a9c32005-06-18 22:58:53 -0700469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 return NULL;
471}
472
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000473static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700474{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000475 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
476 int band = bitmap2band[priv->bitmap];
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700477
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000478 if (band >= 0) {
479 struct sk_buff_head *list = band2list(priv, band);
480
481 return skb_peek(list);
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700482 }
483
484 return NULL;
485}
486
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000487static void pfifo_fast_reset(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
David S. Millerd3678b42008-07-21 09:56:13 -0700489 int prio;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000490 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
David S. Millerd3678b42008-07-21 09:56:13 -0700491
492 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000493 __qdisc_reset_queue(qdisc, band2list(priv, prio));
David S. Millerd3678b42008-07-21 09:56:13 -0700494
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000495 priv->bitmap = 0;
Thomas Graf821d24a2005-06-18 22:58:15 -0700496 qdisc->qstats.backlog = 0;
David S. Millerd3678b42008-07-21 09:56:13 -0700497 qdisc->q.qlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
499
David S. Millerd3678b42008-07-21 09:56:13 -0700500static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
501{
502 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
503
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000504 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
David S. Miller1b34ec42012-03-29 05:11:39 -0400505 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
506 goto nla_put_failure;
David S. Millerd3678b42008-07-21 09:56:13 -0700507 return skb->len;
508
509nla_put_failure:
510 return -1;
511}
512
513static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
514{
515 int prio;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000516 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
David S. Millerd3678b42008-07-21 09:56:13 -0700517
518 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000519 skb_queue_head_init(band2list(priv, prio));
David S. Millerd3678b42008-07-21 09:56:13 -0700520
Eric Dumazet23624932011-01-21 16:26:09 -0800521 /* Can by-pass the queue discipline */
522 qdisc->flags |= TCQ_F_CAN_BYPASS;
David S. Millerd3678b42008-07-21 09:56:13 -0700523 return 0;
524}
525
David S. Miller6ec1c692009-09-06 01:58:51 -0700526struct Qdisc_ops pfifo_fast_ops __read_mostly = {
David S. Millerd3678b42008-07-21 09:56:13 -0700527 .id = "pfifo_fast",
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000528 .priv_size = sizeof(struct pfifo_fast_priv),
David S. Millerd3678b42008-07-21 09:56:13 -0700529 .enqueue = pfifo_fast_enqueue,
530 .dequeue = pfifo_fast_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700531 .peek = pfifo_fast_peek,
David S. Millerd3678b42008-07-21 09:56:13 -0700532 .init = pfifo_fast_init,
533 .reset = pfifo_fast_reset,
534 .dump = pfifo_fast_dump,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 .owner = THIS_MODULE,
536};
537
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000538static struct lock_class_key qdisc_tx_busylock;
539
David S. Miller5ce2d482008-07-08 17:06:30 -0700540struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
stephen hemmingerd2a7f262013-08-31 10:15:50 -0700541 const struct Qdisc_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
543 void *p;
544 struct Qdisc *sch;
Eric Dumazetd2760552011-03-03 11:10:02 -0800545 unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
Thomas Graf3d54b822005-07-05 14:15:09 -0700546 int err = -ENOBUFS;
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000547 struct net_device *dev = dev_queue->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
Eric Dumazetf2cd2d32010-11-29 08:14:37 +0000549 p = kzalloc_node(size, GFP_KERNEL,
550 netdev_queue_numa_node_read(dev_queue));
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (!p)
Thomas Graf3d54b822005-07-05 14:15:09 -0700553 goto errout;
Thomas Graf3d54b822005-07-05 14:15:09 -0700554 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
Eric Dumazetd2760552011-03-03 11:10:02 -0800555 /* if we got non aligned memory, ask more and do alignment ourself */
556 if (sch != p) {
557 kfree(p);
558 p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
559 netdev_queue_numa_node_read(dev_queue));
560 if (!p)
561 goto errout;
562 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
563 sch->padded = (char *) sch - (char *) p;
564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 INIT_LIST_HEAD(&sch->list);
566 skb_queue_head_init(&sch->q);
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000567
Eric Dumazet79640a42010-06-02 05:09:29 -0700568 spin_lock_init(&sch->busylock);
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000569 lockdep_set_class(&sch->busylock,
570 dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 sch->ops = ops;
573 sch->enqueue = ops->enqueue;
574 sch->dequeue = ops->dequeue;
David S. Millerbb949fb2008-07-08 16:55:56 -0700575 sch->dev_queue = dev_queue;
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000576 dev_hold(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 atomic_set(&sch->refcnt, 1);
Thomas Graf3d54b822005-07-05 14:15:09 -0700578
579 return sch;
580errout:
WANG Cong01e123d2008-06-27 19:51:35 -0700581 return ERR_PTR(err);
Thomas Graf3d54b822005-07-05 14:15:09 -0700582}
583
Changli Gao3511c912010-10-16 13:04:08 +0000584struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
stephen hemmingerd2a7f262013-08-31 10:15:50 -0700585 const struct Qdisc_ops *ops,
586 unsigned int parentid)
Thomas Graf3d54b822005-07-05 14:15:09 -0700587{
588 struct Qdisc *sch;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900589
stephen hemminger6da7c8f2013-08-27 16:19:08 -0700590 if (!try_module_get(ops->owner))
591 goto errout;
592
David S. Miller5ce2d482008-07-08 17:06:30 -0700593 sch = qdisc_alloc(dev_queue, ops);
Thomas Graf3d54b822005-07-05 14:15:09 -0700594 if (IS_ERR(sch))
595 goto errout;
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800596 sch->parent = parentid;
Thomas Graf3d54b822005-07-05 14:15:09 -0700597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 if (!ops->init || ops->init(sch, NULL) == 0)
599 return sch;
600
Thomas Graf0fbbeb12005-08-23 10:12:44 -0700601 qdisc_destroy(sch);
Thomas Graf3d54b822005-07-05 14:15:09 -0700602errout:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 return NULL;
604}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800605EXPORT_SYMBOL(qdisc_create_dflt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
David S. Miller5fb66222008-08-02 20:02:43 -0700607/* Under qdisc_lock(qdisc) and BH! */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609void qdisc_reset(struct Qdisc *qdisc)
610{
Eric Dumazet20fea082007-11-14 01:44:41 -0800611 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
613 if (ops->reset)
614 ops->reset(qdisc);
Jarek Poplawski67305eb2008-11-03 02:52:50 -0800615
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000616 if (qdisc->gso_skb) {
617 kfree_skb(qdisc->gso_skb);
618 qdisc->gso_skb = NULL;
619 qdisc->q.qlen = 0;
620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800622EXPORT_SYMBOL(qdisc_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
Eric Dumazet5d944c62010-03-31 07:06:04 +0000624static void qdisc_rcu_free(struct rcu_head *head)
625{
626 struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head);
627
628 kfree((char *) qdisc - qdisc->padded);
629}
630
David S. Miller1e0d5a52008-08-17 22:31:26 -0700631void qdisc_destroy(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Eric Dumazet20fea082007-11-14 01:44:41 -0800633 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
David S. Miller1e0d5a52008-08-17 22:31:26 -0700635 if (qdisc->flags & TCQ_F_BUILTIN ||
636 !atomic_dec_and_test(&qdisc->refcnt))
637 return;
638
David S. Miller3a682fb2008-07-20 18:13:01 -0700639#ifdef CONFIG_NET_SCHED
Jarek Poplawskif6e0b232008-08-22 03:24:05 -0700640 qdisc_list_del(qdisc);
641
Eric Dumazeta2da5702011-01-20 03:48:19 +0000642 qdisc_put_stab(rtnl_dereference(qdisc->stab));
David S. Miller3a682fb2008-07-20 18:13:01 -0700643#endif
Patrick McHardy85670cc2006-09-27 16:45:45 -0700644 gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
Patrick McHardy85670cc2006-09-27 16:45:45 -0700645 if (ops->reset)
646 ops->reset(qdisc);
647 if (ops->destroy)
648 ops->destroy(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
Patrick McHardy85670cc2006-09-27 16:45:45 -0700650 module_put(ops->owner);
David S. Miller5ce2d482008-07-08 17:06:30 -0700651 dev_put(qdisc_dev(qdisc));
David S. Miller8a34c5d2008-07-17 00:47:45 -0700652
Jarek Poplawski554794d2008-10-06 09:54:39 -0700653 kfree_skb(qdisc->gso_skb);
Eric Dumazet5d944c62010-03-31 07:06:04 +0000654 /*
655 * gen_estimator est_timer() might access qdisc->q.lock,
656 * wait a RCU grace period before freeing qdisc.
657 */
658 call_rcu(&qdisc->rcu_head, qdisc_rcu_free);
David S. Miller8a34c5d2008-07-17 00:47:45 -0700659}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800660EXPORT_SYMBOL(qdisc_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Patrick McHardy589983c2009-09-04 06:41:20 +0000662/* Attach toplevel qdisc to device queue. */
663struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
664 struct Qdisc *qdisc)
665{
666 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
667 spinlock_t *root_lock;
668
669 root_lock = qdisc_lock(oqdisc);
670 spin_lock_bh(root_lock);
671
672 /* Prune old scheduler */
673 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
674 qdisc_reset(oqdisc);
675
676 /* ... and graft new one */
677 if (qdisc == NULL)
678 qdisc = &noop_qdisc;
679 dev_queue->qdisc_sleeping = qdisc;
680 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
681
682 spin_unlock_bh(root_lock);
683
684 return oqdisc;
685}
John Fastabendb8970f02011-01-17 08:06:09 +0000686EXPORT_SYMBOL(dev_graft_qdisc);
Patrick McHardy589983c2009-09-04 06:41:20 +0000687
David S. Millere8a04642008-07-17 00:34:19 -0700688static void attach_one_default_qdisc(struct net_device *dev,
689 struct netdev_queue *dev_queue,
690 void *_unused)
691{
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000692 struct Qdisc *qdisc = &noqueue_qdisc;
David S. Millere8a04642008-07-17 00:34:19 -0700693
694 if (dev->tx_queue_len) {
Changli Gao3511c912010-10-16 13:04:08 +0000695 qdisc = qdisc_create_dflt(dev_queue,
stephen hemminger6da7c8f2013-08-27 16:19:08 -0700696 default_qdisc_ops, TC_H_ROOT);
David S. Millere8a04642008-07-17 00:34:19 -0700697 if (!qdisc) {
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000698 netdev_info(dev, "activation failed\n");
David S. Millere8a04642008-07-17 00:34:19 -0700699 return;
700 }
Eric Dumazet1abbe132012-12-11 15:54:33 +0000701 if (!netif_is_multiqueue(dev))
702 qdisc->flags |= TCQ_F_ONETXQUEUE;
David S. Millere8a04642008-07-17 00:34:19 -0700703 }
704 dev_queue->qdisc_sleeping = qdisc;
705}
706
David S. Miller6ec1c692009-09-06 01:58:51 -0700707static void attach_default_qdiscs(struct net_device *dev)
708{
709 struct netdev_queue *txq;
710 struct Qdisc *qdisc;
711
712 txq = netdev_get_tx_queue(dev, 0);
713
714 if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) {
715 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
716 dev->qdisc = txq->qdisc_sleeping;
717 atomic_inc(&dev->qdisc->refcnt);
718 } else {
Changli Gao3511c912010-10-16 13:04:08 +0000719 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT);
David S. Miller6ec1c692009-09-06 01:58:51 -0700720 if (qdisc) {
721 qdisc->ops->attach(qdisc);
722 dev->qdisc = qdisc;
723 }
724 }
725}
726
David S. Millere8a04642008-07-17 00:34:19 -0700727static void transition_one_qdisc(struct net_device *dev,
728 struct netdev_queue *dev_queue,
729 void *_need_watchdog)
730{
David S. Miller83874002008-07-17 00:53:03 -0700731 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700732 int *need_watchdog_p = _need_watchdog;
733
David S. Millera9312ae2008-08-17 21:51:03 -0700734 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
735 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
736
David S. Miller83874002008-07-17 00:53:03 -0700737 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
Eric Dumazet9d214932009-05-17 20:55:16 -0700738 if (need_watchdog_p && new_qdisc != &noqueue_qdisc) {
739 dev_queue->trans_start = 0;
David S. Millere8a04642008-07-17 00:34:19 -0700740 *need_watchdog_p = 1;
Eric Dumazet9d214932009-05-17 20:55:16 -0700741 }
David S. Millere8a04642008-07-17 00:34:19 -0700742}
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744void dev_activate(struct net_device *dev)
745{
David S. Millere8a04642008-07-17 00:34:19 -0700746 int need_watchdog;
David S. Millerb0e1e642008-07-08 17:42:10 -0700747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 /* No queueing discipline is attached to device;
stephen hemminger6da7c8f2013-08-27 16:19:08 -0700749 * create default one for devices, which need queueing
750 * and noqueue_qdisc for virtual interfaces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 */
752
David S. Miller6ec1c692009-09-06 01:58:51 -0700753 if (dev->qdisc == &noop_qdisc)
754 attach_default_qdiscs(dev);
Patrick McHardyaf356af2009-09-04 06:41:18 +0000755
Tommy S. Christensencacaddf2005-05-03 16:18:52 -0700756 if (!netif_carrier_ok(dev))
757 /* Delay activation until next carrier-on event */
758 return;
759
David S. Millere8a04642008-07-17 00:34:19 -0700760 need_watchdog = 0;
761 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
Eric Dumazet24824a02010-10-02 06:11:55 +0000762 if (dev_ingress_queue(dev))
763 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
David S. Millere8a04642008-07-17 00:34:19 -0700764
765 if (need_watchdog) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 dev->trans_start = jiffies;
767 dev_watchdog_up(dev);
768 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700769}
John Fastabendb8970f02011-01-17 08:06:09 +0000770EXPORT_SYMBOL(dev_activate);
David S. Millerb0e1e642008-07-08 17:42:10 -0700771
David S. Millere8a04642008-07-17 00:34:19 -0700772static void dev_deactivate_queue(struct net_device *dev,
773 struct netdev_queue *dev_queue,
774 void *_qdisc_default)
David S. Millerb0e1e642008-07-08 17:42:10 -0700775{
David S. Millere8a04642008-07-17 00:34:19 -0700776 struct Qdisc *qdisc_default = _qdisc_default;
David S. Miller970565b2008-07-08 23:10:33 -0700777 struct Qdisc *qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700778
David S. Miller970565b2008-07-08 23:10:33 -0700779 qdisc = dev_queue->qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700780 if (qdisc) {
David S. Miller83874002008-07-17 00:53:03 -0700781 spin_lock_bh(qdisc_lock(qdisc));
782
David S. Millera9312ae2008-08-17 21:51:03 -0700783 if (!(qdisc->flags & TCQ_F_BUILTIN))
784 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
785
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700786 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700787 qdisc_reset(qdisc);
David S. Millerd3b753d2008-07-15 20:14:35 -0700788
David S. Miller83874002008-07-17 00:53:03 -0700789 spin_unlock_bh(qdisc_lock(qdisc));
David S. Millerb0e1e642008-07-08 17:42:10 -0700790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
David S. Miller4335cd22008-08-17 21:58:07 -0700793static bool some_qdisc_is_busy(struct net_device *dev)
David S. Millere8a04642008-07-17 00:34:19 -0700794{
795 unsigned int i;
796
797 for (i = 0; i < dev->num_tx_queues; i++) {
798 struct netdev_queue *dev_queue;
David S. Miller7698b4f2008-07-16 01:42:40 -0700799 spinlock_t *root_lock;
David S. Millere2627c82008-07-16 00:56:32 -0700800 struct Qdisc *q;
David S. Millere8a04642008-07-17 00:34:19 -0700801 int val;
802
803 dev_queue = netdev_get_tx_queue(dev, i);
David S. Millerb9a3b112008-08-13 15:18:38 -0700804 q = dev_queue->qdisc_sleeping;
David S. Miller5fb66222008-08-02 20:02:43 -0700805 root_lock = qdisc_lock(q);
David S. Millere8a04642008-07-17 00:34:19 -0700806
David S. Miller4335cd22008-08-17 21:58:07 -0700807 spin_lock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700808
Eric Dumazetbc135b22010-06-02 03:23:51 -0700809 val = (qdisc_is_running(q) ||
David S. Millerb9a3b112008-08-13 15:18:38 -0700810 test_bit(__QDISC_STATE_SCHED, &q->state));
David S. Millere8a04642008-07-17 00:34:19 -0700811
David S. Miller4335cd22008-08-17 21:58:07 -0700812 spin_unlock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700813
814 if (val)
815 return true;
816 }
817 return false;
818}
819
Eric Dumazet31376632011-05-19 23:42:09 +0000820/**
821 * dev_deactivate_many - deactivate transmissions on several devices
822 * @head: list of devices to deactivate
823 *
824 * This function returns only when all outstanding transmissions
825 * have completed, unless all devices are in dismantle phase.
826 */
Octavian Purdila44345722010-12-13 12:44:07 +0000827void dev_deactivate_many(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
Octavian Purdila44345722010-12-13 12:44:07 +0000829 struct net_device *dev;
Eric Dumazet31376632011-05-19 23:42:09 +0000830 bool sync_needed = false;
Herbert Xu41a23b02007-05-10 14:12:47 -0700831
Octavian Purdila44345722010-12-13 12:44:07 +0000832 list_for_each_entry(dev, head, unreg_list) {
833 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
834 &noop_qdisc);
835 if (dev_ingress_queue(dev))
836 dev_deactivate_queue(dev, dev_ingress_queue(dev),
837 &noop_qdisc);
838
839 dev_watchdog_down(dev);
Eric Dumazet31376632011-05-19 23:42:09 +0000840 sync_needed |= !dev->dismantle;
Octavian Purdila44345722010-12-13 12:44:07 +0000841 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
Eric Dumazet31376632011-05-19 23:42:09 +0000843 /* Wait for outstanding qdisc-less dev_queue_xmit calls.
844 * This is avoided if all devices are in dismantle phase :
845 * Caller will call synchronize_net() for us
846 */
847 if (sync_needed)
848 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Herbert Xud4828d82006-06-22 02:28:18 -0700850 /* Wait for outstanding qdisc_run calls. */
Octavian Purdila44345722010-12-13 12:44:07 +0000851 list_for_each_entry(dev, head, unreg_list)
852 while (some_qdisc_is_busy(dev))
853 yield();
854}
855
856void dev_deactivate(struct net_device *dev)
857{
858 LIST_HEAD(single);
859
860 list_add(&dev->unreg_list, &single);
861 dev_deactivate_many(&single);
Eric W. Biederman5f04d502011-02-20 11:49:45 -0800862 list_del(&single);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863}
John Fastabendb8970f02011-01-17 08:06:09 +0000864EXPORT_SYMBOL(dev_deactivate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
David S. Millerb0e1e642008-07-08 17:42:10 -0700866static void dev_init_scheduler_queue(struct net_device *dev,
867 struct netdev_queue *dev_queue,
David S. Millere8a04642008-07-17 00:34:19 -0700868 void *_qdisc)
David S. Millerb0e1e642008-07-08 17:42:10 -0700869{
David S. Millere8a04642008-07-17 00:34:19 -0700870 struct Qdisc *qdisc = _qdisc;
871
David S. Millerb0e1e642008-07-08 17:42:10 -0700872 dev_queue->qdisc = qdisc;
873 dev_queue->qdisc_sleeping = qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700874}
875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876void dev_init_scheduler(struct net_device *dev)
877{
Patrick McHardyaf356af2009-09-04 06:41:18 +0000878 dev->qdisc = &noop_qdisc;
David S. Millere8a04642008-07-17 00:34:19 -0700879 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +0000880 if (dev_ingress_queue(dev))
881 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800883 setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884}
885
David S. Millere8a04642008-07-17 00:34:19 -0700886static void shutdown_scheduler_queue(struct net_device *dev,
887 struct netdev_queue *dev_queue,
888 void *_qdisc_default)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
David S. Millerb0e1e642008-07-08 17:42:10 -0700890 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700891 struct Qdisc *qdisc_default = _qdisc_default;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
David S. Millerb0e1e642008-07-08 17:42:10 -0700893 if (qdisc) {
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700894 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700895 dev_queue->qdisc_sleeping = qdisc_default;
896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 qdisc_destroy(qdisc);
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900898 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700899}
900
901void dev_shutdown(struct net_device *dev)
902{
David S. Millere8a04642008-07-17 00:34:19 -0700903 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +0000904 if (dev_ingress_queue(dev))
905 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Patrick McHardyaf356af2009-09-04 06:41:18 +0000906 qdisc_destroy(dev->qdisc);
907 dev->qdisc = &noop_qdisc;
908
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700909 WARN_ON(timer_pending(&dev->watchdog_timer));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}
Jiri Pirko292f1c72013-02-12 00:12:03 +0000911
Eric Dumazet01cb71d2013-06-02 13:55:05 +0000912void psched_ratecfg_precompute(struct psched_ratecfg *r,
913 const struct tc_ratespec *conf)
Jiri Pirko292f1c72013-02-12 00:12:03 +0000914{
Eric Dumazet01cb71d2013-06-02 13:55:05 +0000915 memset(r, 0, sizeof(*r));
916 r->overhead = conf->overhead;
Eric Dumazet130d3d62013-06-06 13:56:19 -0700917 r->rate_bytes_ps = conf->rate;
Jesper Dangaard Brouer8a8e3d82013-08-14 23:47:11 +0200918 r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
Jiri Pirko292f1c72013-02-12 00:12:03 +0000919 r->mult = 1;
920 /*
Eric Dumazet130d3d62013-06-06 13:56:19 -0700921 * The deal here is to replace a divide by a reciprocal one
922 * in fast path (a reciprocal divide is a multiply and a shift)
923 *
924 * Normal formula would be :
925 * time_in_ns = (NSEC_PER_SEC * len) / rate_bps
926 *
927 * We compute mult/shift to use instead :
928 * time_in_ns = (len * mult) >> shift;
929 *
930 * We try to get the highest possible mult value for accuracy,
931 * but have to make sure no overflows will ever happen.
Jiri Pirko292f1c72013-02-12 00:12:03 +0000932 */
Eric Dumazet130d3d62013-06-06 13:56:19 -0700933 if (r->rate_bytes_ps > 0) {
934 u64 factor = NSEC_PER_SEC;
Jiri Pirko292f1c72013-02-12 00:12:03 +0000935
Eric Dumazet130d3d62013-06-06 13:56:19 -0700936 for (;;) {
937 r->mult = div64_u64(factor, r->rate_bytes_ps);
938 if (r->mult & (1U << 31) || factor & (1ULL << 63))
939 break;
940 factor <<= 1;
941 r->shift++;
942 }
Jiri Pirko292f1c72013-02-12 00:12:03 +0000943 }
944}
945EXPORT_SYMBOL(psched_ratecfg_precompute);