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 <eric.dumazet@gmail.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Jesper Dangaard Brouer <hawk@diku.dk>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Jamal Hadi Salim <hadi@cyberus.ca>
CC: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
Eric Dumazet 2011-01-20 03:48:19 +00:00 committed by David S. Miller
parent fd245a4adb
commit a2da570d62
4 changed files with 38 additions and 19 deletions

View File

@ -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);
}

View File

@ -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);

View File

@ -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 @@ static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
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 @@ void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
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 @@ qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
* 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 @@ qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
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,

View File

@ -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)