Currently, all sorts of skb allocation always do allocate
skbuff_heads one by one via kmem_cache_alloc().
On the other hand, we have percpu napi_alloc_cache to store
skbuff_heads queued up for freeing and flush them by bulks.
We can use this cache not only for bulk-wiping, but also to obtain
heads for new skbs and avoid unconditional allocations, as well as
for bulk-allocating.
As accessing napi_alloc_cache implies NAPI softirq context, do this
only for __napi_alloc_skb() and its derivatives (napi_alloc_skb()
and napi_get_frags()). The rough amount of their call sites are 69,
which is quite a number.
iperf3 showed 35-50 Mbps bumps for both TCP and UDP while performing
VLAN NAT on 1.2 GHz MIPS board. The boost is likely to be way bigger
on more powerful hosts and NICs with tens of Mpps.
Since v1 [0]:
- use one unified cache instead of two separate to greatly simplify
the logics and reduce hotpath overhead (Edward Cree);
- new: recycle also GRO_MERGED_FREE skbs instead of immediate
freeing;
- correct performance numbers after optimizations and performing
lots of tests for different use cases.
[0] https://lore.kernel.org/netdev/[email protected]
Alexander Lobakin (3):
skbuff: open-code __build_skb() inside __napi_alloc_skb()
skbuff: (re)use NAPI skb cache on allocation path
skbuff: recycle GRO_MERGED_FREE skbs into NAPI skb cache
include/linux/skbuff.h | 1 +
net/core/dev.c | 9 +----
net/core/skbuff.c | 74 +++++++++++++++++++++++++++++-------------
3 files changed, 54 insertions(+), 30 deletions(-)
--
2.30.0