diff options
| author |   Eric Dumazet <edumazet@google.com> | 2025-11-21 08:32:55 +0000 |
|---|---|---|
| committer |   Paolo Abeni <pabeni@redhat.com> | 2025-11-25 16:10:32 +0100 |
| commit | 191ff13e42a7b7824fec5b2ed84fd6481356754d (patch) | |
| tree | ddd4c7cdcd4ca88654791f78ade112b11432f9a7 /net/core/dev.c | |
| parent | net_sched: add tcf_kfree_skb_list() helper (diff) | |
| download | wireguard-linux-191ff13e42a7b7824fec5b2ed84fd6481356754d.tar.xz wireguard-linux-191ff13e42a7b7824fec5b2ed84fd6481356754d.zip | |
net_sched: add qdisc_dequeue_drop() helper
Some qdisc like cake, codel, fq_codel might drop packets
in their dequeue() method.
This is currently problematic because dequeue() runs with
the qdisc spinlock held. Freeing skbs can be extremely expensive.
Add qdisc_dequeue_drop() method and a new TCQ_F_DEQUEUE_DROPS
so that these qdiscs can opt-in to defer the skb frees
after the socket spinlock is released.
TCQ_F_DEQUEUE_DROPS is an attempt to not penalize other qdiscs
with an extra cache line miss.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251121083256.674562-14-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Diffstat (limited to '')
| -rw-r--r-- | net/core/dev.c | 22 |
1 files changed, 13 insertions, 9 deletions
diff --git a/net/core/dev.c b/net/core/dev.c index e865cdb9b696..9094c0fb8c68 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -4141,7 +4141,7 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, struct net_device *dev, struct netdev_queue *txq) { - struct sk_buff *next, *to_free = NULL; + struct sk_buff *next, *to_free = NULL, *to_free2 = NULL; spinlock_t *root_lock = qdisc_lock(q); struct llist_node *ll_list, *first_n; unsigned long defer_count = 0; @@ -4160,7 +4160,7 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, if (unlikely(!nolock_qdisc_is_empty(q))) { rc = dev_qdisc_enqueue(skb, q, &to_free, txq); __qdisc_run(q); - qdisc_run_end(q); + to_free2 = qdisc_run_end(q); goto free_skbs; } @@ -4170,12 +4170,13 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, !nolock_qdisc_is_empty(q)) __qdisc_run(q); - qdisc_run_end(q); - return NET_XMIT_SUCCESS; + to_free2 = qdisc_run_end(q); + rc = NET_XMIT_SUCCESS; + goto free_skbs; } rc = dev_qdisc_enqueue(skb, q, &to_free, txq); - qdisc_run(q); + to_free2 = qdisc_run(q); goto free_skbs; } @@ -4234,7 +4235,7 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, qdisc_bstats_update(q, skb); if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) __qdisc_run(q); - qdisc_run_end(q); + to_free2 = qdisc_run_end(q); rc = NET_XMIT_SUCCESS; } else { int count = 0; @@ -4246,7 +4247,7 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, rc = dev_qdisc_enqueue(skb, q, &to_free, txq); count++; } - qdisc_run(q); + to_free2 = qdisc_run(q); if (count != 1) rc = NET_XMIT_SUCCESS; } @@ -4255,6 +4256,7 @@ unlock: free_skbs: tcf_kfree_skb_list(to_free); + tcf_kfree_skb_list(to_free2); return rc; } @@ -5747,8 +5749,9 @@ static __latent_entropy void net_tx_action(void) rcu_read_lock(); while (head) { - struct Qdisc *q = head; spinlock_t *root_lock = NULL; + struct sk_buff *to_free; + struct Qdisc *q = head; head = head->next_sched; @@ -5775,9 +5778,10 @@ static __latent_entropy void net_tx_action(void) } clear_bit(__QDISC_STATE_SCHED, &q->state); - qdisc_run(q); + to_free = qdisc_run(q); if (root_lock) spin_unlock(root_lock); + tcf_kfree_skb_list(to_free); } rcu_read_unlock(); |