net: reorder struct netns_ct for better cache-line usage
Subsystems (3)
RFC PATCH 1/1 net: reorder struct netns_ct for better cache-line usage
Reorder struct netns_ct so that atomic_t "count" changes don't slowdown users of read mostly fields.
This is based on Eric Dumazet's proposed patch:
"netfilter: conntrack: remove the central spinlock"
http://thread.gmane.org/gmane.linux.network/268758/focus=47306
The tricky part of cache-aligning this structure, that it is getting inlined in struct net (include/net/net_namespace.h), thus changes to other netns_xxx structures affects our alignment.
Eric's original patch contained an ambiguity on 32-bit regarding alignment in struct net. This patch also takes 32-bit into account, and in case of changed (struct net) alignment sysctl_xxx entries have been ordered according to how often they are accessed.
---
Yes, I know it's not the time to submit this patch, but I just want to let Eric know that I'm planning on extracting part of his proposed patch, which have been stalled for months. I'll resubmit once net-next/nf-next opens for changes again.
include/net/netns/conntrack.h | 33 +++++++++++++++++---------------- 1 files changed, 17 insertions(+), 16 deletions(-)
include/net/netns/conntrack.h +17 -16
@@ -65,6 +65,23 @@ struct nf_ip_net {struct netns_ct {atomic_t count;unsigned int expect_count; +#ifdef CONFIG_SYSCTL +struct ctl_table_header *sysctl_header; +struct ctl_table_header *acct_sysctl_header; +struct ctl_table_header *tstamp_sysctl_header; +struct ctl_table_header *event_sysctl_header; +struct ctl_table_header *helper_sysctl_header; +#endif +char *slabname; +unsigned int sysctl_log_invalid;/* Log invalid packets */ +unsigned int sysctl_events_retry_timeout; +int sysctl_events; +int sysctl_acct; +int sysctl_auto_assign_helper; +bool auto_assign_helper_warned; +int sysctl_tstamp; +int sysctl_checksum; +unsigned int htable_size;struct kmem_cache *nf_conntrack_cachep;struct hlist_nulls_head *hash;
@@ -75,14 +92,6 @@ struct netns_ct {struct ip_conntrack_stat __percpu *stat;struct nf_ct_event_notifier __rcu *nf_conntrack_event_cb;struct nf_exp_event_notifier __rcu *nf_expect_event_cb; -int sysctl_events; -unsigned int sysctl_events_retry_timeout; -int sysctl_acct; -int sysctl_tstamp; -int sysctl_checksum; -unsigned int sysctl_log_invalid;/* Log invalid packets */ -int sysctl_auto_assign_helper; -bool auto_assign_helper_warned;struct nf_ip_net nf_ct_proto;#if defined(CONFIG_NF_CONNTRACK_LABELS )unsigned int labels_used;
@@ -92,13 +101,5 @@ struct netns_ct {struct hlist_head *nat_bysource;unsigned int nat_htable_size; #endif -#ifdef CONFIG_SYSCTL -struct ctl_table_header *sysctl_header; -struct ctl_table_header *acct_sysctl_header; -struct ctl_table_header *tstamp_sysctl_header; -struct ctl_table_header *event_sysctl_header; -struct ctl_table_header *helper_sysctl_header; -#endif - char *slabname; };#endif
Quoted 16 lines · Jesper Dangaard Brouer
Reorder struct
netns_ctso thatatomic_t"count" changes don't slowdown users of read mostly fields.This is based on Eric Dumazet's proposed patch:
"netfilter: conntrack: remove the central spinlock"
http://thread.gmane.org/gmane.linux.network/268758/focus=47306The tricky part of cache-aligning this structure, that it is getting inlined in struct net (
include/net/net_namespace.h), thus changes to othernetns_xxxstructures affects our alignment.Eric's original patch contained an ambiguity on 32-bit regarding alignment in struct net. This patch also takes 32-bit into account, and in case of changed (struct net) alignment
sysctl_xxxentries have been ordered according to how often they are accessed.
Quoted 7 lines
Signed-off-by Jesper Dangaard Brouer <[email protected]>Reviewed-by Jiri Benc <[email protected]>---
Yes, I know it's not the time to submit this patch, but I just want to let Eric know that I'm planning on extracting part of his proposed patch, which have been stalled for months. I'll resubmit once net-next/nf-next opens for changes again.
I have no plan working on conntrack in the near future.
I cannot really cope with a 3 months response time to a patch.
Feel free to grab and continue the work, thanks !