diff options
| -rw-r--r-- | Documentation/admin-guide/sysctl/net.rst | 8 | ||||
| -rw-r--r-- | net/core/dev.c | 1 | ||||
| -rw-r--r-- | net/core/dev.h | 2 | ||||
| -rw-r--r-- | net/core/skbuff.c | 15 | ||||
| -rw-r--r-- | net/core/sysctl_net_core.c | 8 |
5 files changed, 27 insertions, 7 deletions
diff --git a/Documentation/admin-guide/sysctl/net.rst b/Documentation/admin-guide/sysctl/net.rst index f86b5e1623c6..fa4dcdb283cf 100644 --- a/Documentation/admin-guide/sysctl/net.rst +++ b/Documentation/admin-guide/sysctl/net.rst @@ -322,6 +322,14 @@ a leaked reference faster. A larger value may be useful to prevent false warnings on slow/loaded systems. Default value is 10, minimum 1, maximum 3600. +skb_defer_max +------------- + +Max size (in skbs) of the per-cpu list of skbs being freed +by the cpu which allocated them. Used by TCP stack so far. + +Default: 64 + optmem_max ---------- diff --git a/net/core/dev.c b/net/core/dev.c index d0b34bc50706..6359f8953269 100644 --- a/net/core/dev.c +++ b/net/core/dev.c @@ -4330,6 +4330,7 @@ int netdev_max_backlog __read_mostly = 1000; EXPORT_SYMBOL(netdev_max_backlog); int netdev_tstamp_prequeue __read_mostly = 1; +unsigned int sysctl_skb_defer_max __read_mostly = 64; int netdev_budget __read_mostly = 300; /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */ unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ; diff --git a/net/core/dev.h b/net/core/dev.h index 328b37af90ba..cbb8a925175a 100644 --- a/net/core/dev.h +++ b/net/core/dev.h @@ -39,7 +39,7 @@ void dev_addr_check(struct net_device *dev); /* sysctls not referred to from outside net/core/ */ extern int netdev_budget; extern unsigned int netdev_budget_usecs; - +extern unsigned int sysctl_skb_defer_max; extern int netdev_tstamp_prequeue; extern int netdev_unregister_timeout_secs; extern int weight_p; diff --git a/net/core/skbuff.c b/net/core/skbuff.c index b40c8cdf4785..1d10bb4adec1 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -80,6 +80,7 @@ #include <linux/user_namespace.h> #include <linux/indirect_call_wrapper.h> +#include "dev.h" #include "sock_destructor.h" struct kmem_cache *skbuff_head_cache __ro_after_init; @@ -6496,16 +6497,21 @@ void skb_attempt_defer_free(struct sk_buff *skb) int cpu = skb->alloc_cpu; struct softnet_data *sd; unsigned long flags; + unsigned int defer_max; bool kick; if (WARN_ON_ONCE(cpu >= nr_cpu_ids) || !cpu_online(cpu) || cpu == raw_smp_processor_id()) { - __kfree_skb(skb); +nodefer: __kfree_skb(skb); return; } sd = &per_cpu(softnet_data, cpu); + defer_max = READ_ONCE(sysctl_skb_defer_max); + if (READ_ONCE(sd->defer_count) >= defer_max) + goto nodefer; + /* We do not send an IPI or any signal. * Remote cpu will eventually call skb_defer_free_flush() */ @@ -6515,11 +6521,8 @@ void skb_attempt_defer_free(struct sk_buff *skb) WRITE_ONCE(sd->defer_list, skb); sd->defer_count++; - /* kick every time queue length reaches 128. - * This condition should hardly be hit under normal conditions, - * unless cpu suddenly stopped to receive NIC interrupts. - */ - kick = sd->defer_count == 128; + /* Send an IPI every time queue reaches half capacity. */ + kick = sd->defer_count == (defer_max >> 1); spin_unlock_irqrestore(&sd->defer_lock, flags); diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c index 195ca5c28771..ca8d38325e1e 100644 --- a/net/core/sysctl_net_core.c +++ b/net/core/sysctl_net_core.c @@ -578,6 +578,14 @@ static struct ctl_table net_core_table[] = { .extra1 = SYSCTL_ONE, .extra2 = &int_3600, }, + { + .procname = "skb_defer_max", + .data = &sysctl_skb_defer_max, + .maxlen = sizeof(unsigned int), + .mode = 0644, + .proc_handler = proc_dointvec_minmax, + .extra1 = SYSCTL_ZERO, + }, { } }; |