From fd245a4adb5288eac37250875f237c40a20a1944 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Thu, 20 Jan 2011 05:27:16 +0000 Subject: net_sched: move TCQ_F_THROTTLED flag In commit 371121057607e (net: QDISC_STATE_RUNNING dont need atomic bit ops) I moved QDISC_STATE_RUNNING flag to __state container, located in the cache line containing qdisc lock and often dirtied fields. I now move TCQ_F_THROTTLED bit too, so that we let first cache line read mostly, and shared by all cpus. This should speedup HTB/CBQ for example. Not using test_bit()/__clear_bit()/__test_and_set_bit allows to use an "unsigned int" for __state container, reducing by 8 bytes Qdisc size. Introduce helpers to hide implementation details. Signed-off-by: Eric Dumazet CC: Patrick McHardy CC: Jesper Dangaard Brouer CC: Jarek Poplawski CC: Jamal Hadi Salim CC: Stephen Hemminger Signed-off-by: David S. Miller --- include/net/sch_generic.h | 38 ++++++++++++++++++++++++++++---------- 1 file changed, 28 insertions(+), 10 deletions(-) (limited to 'include/net/sch_generic.h') diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h index e9eee99d8b1f..f6345f55041c 100644 --- a/include/net/sch_generic.h +++ b/include/net/sch_generic.h @@ -31,7 +31,8 @@ enum qdisc_state_t { * following bits are only changed while qdisc lock is held */ enum qdisc___state_t { - __QDISC___STATE_RUNNING, + __QDISC___STATE_RUNNING = 1, + __QDISC___STATE_THROTTLED = 2, }; struct qdisc_size_table { @@ -46,10 +47,9 @@ struct Qdisc { struct sk_buff * (*dequeue)(struct Qdisc *dev); unsigned flags; #define TCQ_F_BUILTIN 1 -#define TCQ_F_THROTTLED 2 -#define TCQ_F_INGRESS 4 -#define TCQ_F_CAN_BYPASS 8 -#define TCQ_F_MQROOT 16 +#define TCQ_F_INGRESS 2 +#define TCQ_F_CAN_BYPASS 4 +#define TCQ_F_MQROOT 8 #define TCQ_F_WARN_NONWC (1 << 16) int padded; struct Qdisc_ops *ops; @@ -78,25 +78,43 @@ struct Qdisc { unsigned long state; struct sk_buff_head q; struct gnet_stats_basic_packed bstats; - unsigned long __state; + unsigned int __state; struct gnet_stats_queue qstats; struct rcu_head rcu_head; spinlock_t busylock; }; -static inline bool qdisc_is_running(struct Qdisc *qdisc) +static inline bool qdisc_is_running(const struct Qdisc *qdisc) { - return test_bit(__QDISC___STATE_RUNNING, &qdisc->__state); + return (qdisc->__state & __QDISC___STATE_RUNNING) ? true : false; } static inline bool qdisc_run_begin(struct Qdisc *qdisc) { - return !__test_and_set_bit(__QDISC___STATE_RUNNING, &qdisc->__state); + if (qdisc_is_running(qdisc)) + return false; + qdisc->__state |= __QDISC___STATE_RUNNING; + return true; } static inline void qdisc_run_end(struct Qdisc *qdisc) { - __clear_bit(__QDISC___STATE_RUNNING, &qdisc->__state); + qdisc->__state &= ~__QDISC___STATE_RUNNING; +} + +static inline bool qdisc_is_throttled(const struct Qdisc *qdisc) +{ + return (qdisc->__state & __QDISC___STATE_THROTTLED) ? true : false; +} + +static inline void qdisc_throttled(struct Qdisc *qdisc) +{ + qdisc->__state |= __QDISC___STATE_THROTTLED; +} + +static inline void qdisc_unthrottled(struct Qdisc *qdisc) +{ + qdisc->__state &= ~__QDISC___STATE_THROTTLED; } struct Qdisc_class_ops { -- cgit v1.2.3 From a2da570d62fcb9e8816f6920e1ec02c706b289fa Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Thu, 20 Jan 2011 03:48:19 +0000 Subject: net_sched: RCU conversion of stab This patch converts stab qdisc management to RCU, so that we can perform the qdisc_calculate_pkt_len() call before getting qdisc lock. This shortens the lock's held time in __dev_xmit_skb(). This permits more qdiscs to get TCQ_F_CAN_BYPASS status, avoiding lot of cache misses and so reducing latencies. Signed-off-by: Eric Dumazet CC: Patrick McHardy CC: Jesper Dangaard Brouer CC: Jarek Poplawski CC: Jamal Hadi Salim CC: Stephen Hemminger Signed-off-by: David S. Miller --- include/net/sch_generic.h | 21 +++++++++++++++------ net/core/dev.c | 8 +++++--- net/sched/sch_api.c | 26 +++++++++++++++++--------- net/sched/sch_generic.c | 2 +- 4 files changed, 38 insertions(+), 19 deletions(-) (limited to 'include/net/sch_generic.h') diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h index f6345f55041c..d531baa2506a 100644 --- a/include/net/sch_generic.h +++ b/include/net/sch_generic.h @@ -36,6 +36,7 @@ enum qdisc___state_t { }; struct qdisc_size_table { + struct rcu_head rcu; struct list_head list; struct tc_sizespec szopts; int refcnt; @@ -53,7 +54,7 @@ struct Qdisc { #define TCQ_F_WARN_NONWC (1 << 16) int padded; struct Qdisc_ops *ops; - struct qdisc_size_table *stab; + struct qdisc_size_table __rcu *stab; struct list_head list; u32 handle; u32 parent; @@ -349,8 +350,8 @@ extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, struct Qdisc_ops *ops); extern struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, struct Qdisc_ops *ops, u32 parentid); -extern void qdisc_calculate_pkt_len(struct sk_buff *skb, - struct qdisc_size_table *stab); +extern void __qdisc_calculate_pkt_len(struct sk_buff *skb, + const struct qdisc_size_table *stab); extern void tcf_destroy(struct tcf_proto *tp); extern void tcf_destroy_chain(struct tcf_proto **fl); @@ -429,12 +430,20 @@ enum net_xmit_qdisc_t { #define net_xmit_drop_count(e) (1) #endif -static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch) +static inline void qdisc_calculate_pkt_len(struct sk_buff *skb, + const struct Qdisc *sch) { #ifdef CONFIG_NET_SCHED - if (sch->stab) - qdisc_calculate_pkt_len(skb, sch->stab); + struct qdisc_size_table *stab = rcu_dereference_bh(sch->stab); + + if (stab) + __qdisc_calculate_pkt_len(skb, stab); #endif +} + +static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch) +{ + qdisc_calculate_pkt_len(skb, sch); return sch->enqueue(skb, sch); } diff --git a/net/core/dev.c b/net/core/dev.c index a4ccd47f3196..2730352d2ccc 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -2325,15 +2325,18 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, struct netdev_queue *txq) { spinlock_t *root_lock = qdisc_lock(q); - bool contended = qdisc_is_running(q); + bool contended; int rc; + qdisc_skb_cb(skb)->pkt_len = skb->len; + qdisc_calculate_pkt_len(skb, q); /* * Heuristic to force contended enqueues to serialize on a * separate lock before trying to get qdisc main lock. * This permits __QDISC_STATE_RUNNING owner to get the lock more often * and dequeue packets faster. */ + contended = qdisc_is_running(q); if (unlikely(contended)) spin_lock(&q->busylock); @@ -2351,7 +2354,6 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE)) skb_dst_force(skb); - qdisc_skb_cb(skb)->pkt_len = skb->len; qdisc_bstats_update(q, skb); if (sch_direct_xmit(skb, q, dev, txq, root_lock)) { @@ -2366,7 +2368,7 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, rc = NET_XMIT_SUCCESS; } else { skb_dst_force(skb); - rc = qdisc_enqueue_root(skb, q); + rc = q->enqueue(skb, q) & NET_XMIT_MASK; if (qdisc_run_begin(q)) { if (unlikely(contended)) { spin_unlock(&q->busylock); diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c index 374fcbef80e8..150741579408 100644 --- a/net/sched/sch_api.c +++ b/net/sched/sch_api.c @@ -398,6 +398,11 @@ static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) return stab; } +static void stab_kfree_rcu(struct rcu_head *head) +{ + kfree(container_of(head, struct qdisc_size_table, rcu)); +} + void qdisc_put_stab(struct qdisc_size_table *tab) { if (!tab) @@ -407,7 +412,7 @@ void qdisc_put_stab(struct qdisc_size_table *tab) if (--tab->refcnt == 0) { list_del(&tab->list); - kfree(tab); + call_rcu_bh(&tab->rcu, stab_kfree_rcu); } spin_unlock(&qdisc_stab_lock); @@ -430,7 +435,7 @@ nla_put_failure: return -1; } -void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab) +void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab) { int pkt_len, slot; @@ -456,7 +461,7 @@ out: pkt_len = 1; qdisc_skb_cb(skb)->pkt_len = pkt_len; } -EXPORT_SYMBOL(qdisc_calculate_pkt_len); +EXPORT_SYMBOL(__qdisc_calculate_pkt_len); void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc) { @@ -835,7 +840,7 @@ qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, err = PTR_ERR(stab); goto err_out4; } - sch->stab = stab; + rcu_assign_pointer(sch->stab, stab); } if (tca[TCA_RATE]) { spinlock_t *root_lock; @@ -875,7 +880,7 @@ err_out4: * Any broken qdiscs that would require a ops->reset() here? * The qdisc was never in action so it shouldn't be necessary. */ - qdisc_put_stab(sch->stab); + qdisc_put_stab(rtnl_dereference(sch->stab)); if (ops->destroy) ops->destroy(sch); goto err_out3; @@ -883,7 +888,7 @@ err_out4: static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) { - struct qdisc_size_table *stab = NULL; + struct qdisc_size_table *ostab, *stab = NULL; int err = 0; if (tca[TCA_OPTIONS]) { @@ -900,8 +905,9 @@ static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) return PTR_ERR(stab); } - qdisc_put_stab(sch->stab); - sch->stab = stab; + ostab = rtnl_dereference(sch->stab); + rcu_assign_pointer(sch->stab, stab); + qdisc_put_stab(ostab); if (tca[TCA_RATE]) { /* NB: ignores errors from replace_estimator @@ -1180,6 +1186,7 @@ static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, struct nlmsghdr *nlh; unsigned char *b = skb_tail_pointer(skb); struct gnet_dump d; + struct qdisc_size_table *stab; nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); tcm = NLMSG_DATA(nlh); @@ -1195,7 +1202,8 @@ static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, goto nla_put_failure; q->qstats.qlen = q->q.qlen; - if (q->stab && qdisc_dump_stab(skb, q->stab) < 0) + stab = rtnl_dereference(q->stab); + if (stab && qdisc_dump_stab(skb, stab) < 0) goto nla_put_failure; if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c index 2f1cb62130da..cc17e794c41e 100644 --- a/net/sched/sch_generic.c +++ b/net/sched/sch_generic.c @@ -632,7 +632,7 @@ void qdisc_destroy(struct Qdisc *qdisc) #ifdef CONFIG_NET_SCHED qdisc_list_del(qdisc); - qdisc_put_stab(qdisc->stab); + qdisc_put_stab(rtnl_dereference(qdisc->stab)); #endif gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); if (ops->reset) -- cgit v1.2.3 From d276055c4e90a7278cd5167ba9755c9b214bcff7 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Thu, 3 Mar 2011 11:10:02 -0800 Subject: net_sched: reduce fifo qdisc size Because of various alignements [SLUB / qdisc], we use 512 bytes of memory for one {p|b}fifo qdisc, instead of 256 bytes on 64bit arches and 192 bytes on 32bit ones. Move the "u32 limit" inside "struct Qdisc" (no impact on other qdiscs) Change qdisc_alloc(), first trying a regular allocation before an oversized one. Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- include/net/sch_generic.h | 1 + net/sched/sch_fifo.c | 34 +++++++++++----------------------- net/sched/sch_generic.c | 18 +++++++++++------- 3 files changed, 23 insertions(+), 30 deletions(-) (limited to 'include/net/sch_generic.h') diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h index 16626a04cb03..1934634f8896 100644 --- a/include/net/sch_generic.h +++ b/include/net/sch_generic.h @@ -83,6 +83,7 @@ struct Qdisc { struct gnet_stats_queue qstats; struct rcu_head rcu_head; spinlock_t busylock; + u32 limit; }; static inline bool qdisc_is_running(const struct Qdisc *qdisc) diff --git a/net/sched/sch_fifo.c b/net/sched/sch_fifo.c index be33f9ddf9dd..66effe2da8e0 100644 --- a/net/sched/sch_fifo.c +++ b/net/sched/sch_fifo.c @@ -19,15 +19,9 @@ /* 1 band FIFO pseudo-"scheduler" */ -struct fifo_sched_data { - u32 limit; -}; - static int bfifo_enqueue(struct sk_buff *skb, struct Qdisc *sch) { - struct fifo_sched_data *q = qdisc_priv(sch); - - if (likely(sch->qstats.backlog + qdisc_pkt_len(skb) <= q->limit)) + if (likely(sch->qstats.backlog + qdisc_pkt_len(skb) <= sch->limit)) return qdisc_enqueue_tail(skb, sch); return qdisc_reshape_fail(skb, sch); @@ -35,9 +29,7 @@ static int bfifo_enqueue(struct sk_buff *skb, struct Qdisc *sch) static int pfifo_enqueue(struct sk_buff *skb, struct Qdisc *sch) { - struct fifo_sched_data *q = qdisc_priv(sch); - - if (likely(skb_queue_len(&sch->q) < q->limit)) + if (likely(skb_queue_len(&sch->q) < sch->limit)) return qdisc_enqueue_tail(skb, sch); return qdisc_reshape_fail(skb, sch); @@ -45,9 +37,7 @@ static int pfifo_enqueue(struct sk_buff *skb, struct Qdisc *sch) static int pfifo_tail_enqueue(struct sk_buff *skb, struct Qdisc *sch) { - struct fifo_sched_data *q = qdisc_priv(sch); - - if (likely(skb_queue_len(&sch->q) < q->limit)) + if (likely(skb_queue_len(&sch->q) < sch->limit)) return qdisc_enqueue_tail(skb, sch); /* queue full, remove one skb to fulfill the limit */ @@ -60,7 +50,6 @@ static int pfifo_tail_enqueue(struct sk_buff *skb, struct Qdisc *sch) static int fifo_init(struct Qdisc *sch, struct nlattr *opt) { - struct fifo_sched_data *q = qdisc_priv(sch); bool bypass; bool is_bfifo = sch->ops == &bfifo_qdisc_ops; @@ -70,20 +59,20 @@ static int fifo_init(struct Qdisc *sch, struct nlattr *opt) if (is_bfifo) limit *= psched_mtu(qdisc_dev(sch)); - q->limit = limit; + sch->limit = limit; } else { struct tc_fifo_qopt *ctl = nla_data(opt); if (nla_len(opt) < sizeof(*ctl)) return -EINVAL; - q->limit = ctl->limit; + sch->limit = ctl->limit; } if (is_bfifo) - bypass = q->limit >= psched_mtu(qdisc_dev(sch)); + bypass = sch->limit >= psched_mtu(qdisc_dev(sch)); else - bypass = q->limit >= 1; + bypass = sch->limit >= 1; if (bypass) sch->flags |= TCQ_F_CAN_BYPASS; @@ -94,8 +83,7 @@ static int fifo_init(struct Qdisc *sch, struct nlattr *opt) static int fifo_dump(struct Qdisc *sch, struct sk_buff *skb) { - struct fifo_sched_data *q = qdisc_priv(sch); - struct tc_fifo_qopt opt = { .limit = q->limit }; + struct tc_fifo_qopt opt = { .limit = sch->limit }; NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); return skb->len; @@ -106,7 +94,7 @@ nla_put_failure: struct Qdisc_ops pfifo_qdisc_ops __read_mostly = { .id = "pfifo", - .priv_size = sizeof(struct fifo_sched_data), + .priv_size = 0, .enqueue = pfifo_enqueue, .dequeue = qdisc_dequeue_head, .peek = qdisc_peek_head, @@ -121,7 +109,7 @@ EXPORT_SYMBOL(pfifo_qdisc_ops); struct Qdisc_ops bfifo_qdisc_ops __read_mostly = { .id = "bfifo", - .priv_size = sizeof(struct fifo_sched_data), + .priv_size = 0, .enqueue = bfifo_enqueue, .dequeue = qdisc_dequeue_head, .peek = qdisc_peek_head, @@ -136,7 +124,7 @@ EXPORT_SYMBOL(bfifo_qdisc_ops); struct Qdisc_ops pfifo_head_drop_qdisc_ops __read_mostly = { .id = "pfifo_head_drop", - .priv_size = sizeof(struct fifo_sched_data), + .priv_size = 0, .enqueue = pfifo_tail_enqueue, .dequeue = qdisc_dequeue_head, .peek = qdisc_peek_head, diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c index 0da09d508737..a854cab03f1e 100644 --- a/net/sched/sch_generic.c +++ b/net/sched/sch_generic.c @@ -550,21 +550,25 @@ struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, { void *p; struct Qdisc *sch; - unsigned int size; + unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; int err = -ENOBUFS; - /* ensure that the Qdisc and the private data are 64-byte aligned */ - size = QDISC_ALIGN(sizeof(*sch)); - size += ops->priv_size + (QDISC_ALIGNTO - 1); - p = kzalloc_node(size, GFP_KERNEL, netdev_queue_numa_node_read(dev_queue)); if (!p) goto errout; sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); - sch->padded = (char *) sch - (char *) p; - + /* if we got non aligned memory, ask more and do alignment ourself */ + if (sch != p) { + kfree(p); + p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, + netdev_queue_numa_node_read(dev_queue)); + if (!p) + goto errout; + sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); + sch->padded = (char *) sch - (char *) p; + } INIT_LIST_HEAD(&sch->list); skb_queue_head_init(&sch->q); spin_lock_init(&sch->busylock); -- cgit v1.2.3