On 08/28/2018 02:44 PM, Björn Töpel wrote:
> From: Björn Töpel <bjorn.to...@intel.com>
>
> This patch set introduces zero-copy AF_XDP support for Intel's i40e
> driver. In the first preparatory patch we also add support for
> XDP_REDIRECT for zero-copy allocated frames so that XDP programs can
> redirect them. This was a ToDo from the first AF_XDP zero-copy patch
> set from early June. Special thanks to Alex Duyck and Jesper Dangaard
> Brouer for reviewing earlier versions of this patch set.
>
> The i40e zero-copy code is located in its own file i40e_xsk.[ch]. Note
> that in the interest of time, to get an AF_XDP zero-copy implementation
> out there for people to try, some code paths have been copied from the
> XDP path to the zero-copy path. It is out goal to merge the two paths
> in later patch sets.
>
> In contrast to the implementation from beginning of June, this patch
> set does not require any extra HW queues for AF_XDP zero-copy
> TX. Instead, the XDP TX HW queue is used for both XDP_REDIRECT and
> AF_XDP zero-copy TX.
>
> Jeff, given that most of changes are in i40e, it is up to you how you
> would like to route these patches. The set is tagged bpf-next, but
> if taking it via the Intel driver tree is easier, let us know.
>
> We have run some benchmarks on a dual socket system with two Broadwell
> E5 2660 @ 2.0 GHz with hyperthreading turned off. Each socket has 14
> cores which gives a total of 28, but only two cores are used in these
> experiments. One for TR/RX and one for the user space application. The
> memory is DDR4 @ 2133 MT/s (1067 MHz) and the size of each DIMM is
> 8192MB and with 8 of those DIMMs in the system we have 64 GB of total
> memory. The compiler used is gcc (Ubuntu 7.3.0-16ubuntu3) 7.3.0. The
> NIC is Intel I40E 40Gbit/s using the i40e driver.
>
> Below are the results in Mpps of the I40E NIC benchmark runs for 64
> and 1500 byte packets, generated by a commercial packet generator HW
> outputing packets at full 40 Gbit/s line rate. The results are with
> retpoline and all other spectre and meltdown fixes, so these results
> are not comparable to the ones from the zero-copy patch set in June.
>
> AF_XDP performance 64 byte packets.
> Benchmark XDP_SKB XDP_DRV XDP_DRV with zerocopy
> rxdrop 2.6 8.2 15.0
> txpush 2.2 - 21.9
> l2fwd 1.7 2.3 11.3
>
> AF_XDP performance 1500 byte packets:
> Benchmark XDP_SKB XDP_DRV XDP_DRV with zerocopy
> rxdrop 2.0 3.3 3.3
> l2fwd 1.3 1.7 3.1
>
> XDP performance on our system as a base line:
>
> 64 byte packets:
> XDP stats CPU pps issue-pps
> XDP-RX CPU 16 18.4M 0
>
> 1500 byte packets:
> XDP stats CPU pps issue-pps
> XDP-RX CPU 16 3.3M 0
>
> The structure of the patch set is as follows:
>
> Patch 1: Add support for XDP_REDIRECT of zero-copy allocated frames
> Patches 2-4: Preparatory patches to common xsk and net code
> Patches 5-7: Preparatory patches to i40e driver code for RX
> Patch 8: i40e zero-copy support for RX
> Patch 9: Preparatory patch to i40e driver code for TX
> Patch 10: i40e zero-copy support for TX
> Patch 11: Add flags to sample application to force zero-copy/copy mode
>
> We based this patch set on bpf-next commit 050cdc6c9501 ("Merge
> git://git.kernel.org/pub/scm/linux/kernel/git/davem/net")
>
>
> Magnus & Björn
>
> Björn Töpel (8):
> xdp: implement convert_to_xdp_frame for MEM_TYPE_ZERO_COPY
> xdp: export xdp_rxq_info_unreg_mem_model
> xsk: expose xdp_umem_get_{data,dma} to drivers
> i40e: added queue pair disable/enable functions
> i40e: refactor Rx path for re-use
> i40e: move common Rx functions to i40e_txrx_common.h
> i40e: add AF_XDP zero-copy Rx support
> samples/bpf: add -c/--copy -z/--zero-copy flags to xdpsock
>
> Magnus Karlsson (3):
> net: add napi_if_scheduled_mark_missed
> i40e: move common Tx functions to i40e_txrx_common.h
> i40e: add AF_XDP zero-copy Tx support
Thanks for working on this, LGTM! Are you also planning to get ixgbe
out after that?
For the series:
Acked-by: Daniel Borkmann <dan...@iogearbox.net>
Thanks,
Daniel
