The XDP CPU destination array/set, configured via xdp-cpu-redirect,
will always be fairly small. My different benchmarking showed that
the current modulo hashing into the CPU array can easily result in bad
distribution, expecially if the number of CPU is an even number.
This patch uses a proper hashing function on the input key. The key
used for hashing is inspired by the ippair hashing code in
src/tmqh-flow.c, and is based on the IP src + dst.
An important property is that the hashing is flow symmetric, meaning
that if the source and destintation gets swapped then the selected CPU
will remain the same. This is important for Suricate.
That hashing INITVAL (15485863 the 10^6th prime number) was fairly
arbitrary choosen, but experiments with kernel tree pktgen scripts
(pktgen_sample04_many_flows.sh +pktgen_sample05_flow_per_thread.sh)
showed this improved the distribution.
Signed-off-by: Jesper Dangaard Brouer <netoptimizer@brouer.com>
Adjusted function call API to take an initval. This allow the API
user to set the initial value, as a seed. This could also be used for
inputting the previous hash.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Main objective of the function is to be able to bypass a flow on
other interfaces. This is necessary in AF_PACKET case as the flow
table are per interface.
This patch adds a boolean option "xdp-cpu-redirect" to af-packet
interface configuration. If set, then the XDP filter will load
balance the skb creation on specified CPUs instead of doing the
creation on the CPU handling the packet. In the case of a card
with asymetric hashing this will allow to avoid saturating the
single CPU handling the trafic.
The XDP filter must contains a set of map allowing load balancing.
This is the case of xdp_filter.bpf.
Fixed-by: Jesper Dangaard Brouer <netoptimizer@brouer.com>
Device storage requires the devices to be created after storage
is finalized so we need to first get the list of devices then
create them when the storage is finalized.
This patch introduces the LiveDeviceName structure that is a list
of device name used during registration.
Code uses LiveRegisterDeviceName for pre registration and keep
using the LiveRegisterDevice function for part of the code that
create the interface during the runmode creation.
eBPF has a data type which is a per CPU array. By adding one element
to the array it is in fact added to all per CPU arrays in the kernel.
This allows to have a lockless structure in the kernel even when doing
counter update.
In userspace, we need to update the flow bypass code to fetch all
elements of the per CPU arrays.