mbox series

[bpf-next,v3,0/3] xdp: Introduce bulking for non-map XDP_REDIRECT

Message ID 157918768284.1458396.8128704597152008763.stgit@toke.dk
Headers show
Series xdp: Introduce bulking for non-map XDP_REDIRECT | expand

Message

Toke Høiland-Jørgensen Jan. 16, 2020, 3:14 p.m. UTC
Since commit 96360004b862 ("xdp: Make devmap flush_list common for all map
instances"), devmap flushing is a global operation instead of tied to a
particular map. This means that with a bit of refactoring, we can finally fix
the performance delta between the bpf_redirect_map() and bpf_redirect() helper
functions, by introducing bulking for the latter as well.

This series makes this change by moving the data structure used for the bulking
into struct net_device itself, so we can access it even when there is not
devmap. Once this is done, moving the bpf_redirect() helper to use the bulking
mechanism becomes quite trivial, and brings bpf_redirect() up to the same as
bpf_redirect_map():

                       Before:   After:
1 CPU:
bpf_redirect_map:      8.4 Mpps  8.4 Mpps  (no change)
bpf_redirect:          5.0 Mpps  8.4 Mpps  (+68%)
2 CPUs:
bpf_redirect_map:     15.9 Mpps  16.1 Mpps  (+1% or ~no change)
bpf_redirect:          9.5 Mpps  15.9 Mpps  (+67%)

After this patch series, the only semantics different between the two variants
of the bpf() helper (apart from the absence of a map argument, obviously) is
that the _map() variant will return an error if passed an invalid map index,
whereas the bpf_redirect() helper will succeed, but drop packets on
xdp_do_redirect(). This is because the helper has no reference to the calling
netdev, so unfortunately we can't do the ifindex lookup directly in the helper.

Changelog:

v3:
  - Switch two more fields to avoid a list_head spanning two cache lines
  - Include Jesper's tracepoint patch
  - Also rename xdp_do_flush_map()
  - Fix a few nits from Maciej

v2:
  - Consolidate code paths and tracepoints for map and non-map redirect variants
    (Björn)
  - Add performance data for 2-CPU test (Jesper)
  - Move fields to avoid shifting cache lines in struct net_device (Eric)

---

Jesper Dangaard Brouer (1):
      devmap: Adjust tracepoint for map-less queue flush

Toke Høiland-Jørgensen (2):
      xdp: Move devmap bulk queue into struct net_device
      xdp: Use bulking for non-map XDP_REDIRECT and consolidate code paths


 drivers/net/tun.c              |    4 +
 drivers/net/veth.c             |    2 -
 drivers/net/virtio_net.c       |    2 -
 include/linux/bpf.h            |   13 +++-
 include/linux/filter.h         |   10 ++-
 include/linux/netdevice.h      |   13 ++--
 include/trace/events/xdp.h     |  130 +++++++++++++++++-----------------------
 kernel/bpf/devmap.c            |   95 +++++++++++++++--------------
 net/core/dev.c                 |    2 +
 net/core/filter.c              |   90 ++++++----------------------
 samples/bpf/xdp_monitor_kern.c |    8 +-
 11 files changed, 160 insertions(+), 209 deletions(-)

Comments

Alexei Starovoitov Jan. 17, 2020, 4:06 a.m. UTC | #1
On Thu, Jan 16, 2020 at 7:14 AM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>
> Since commit 96360004b862 ("xdp: Make devmap flush_list common for all map
> instances"), devmap flushing is a global operation instead of tied to a
> particular map. This means that with a bit of refactoring, we can finally fix
> the performance delta between the bpf_redirect_map() and bpf_redirect() helper
> functions, by introducing bulking for the latter as well.
>
> This series makes this change by moving the data structure used for the bulking
> into struct net_device itself, so we can access it even when there is not
> devmap. Once this is done, moving the bpf_redirect() helper to use the bulking
> mechanism becomes quite trivial, and brings bpf_redirect() up to the same as
> bpf_redirect_map():
>
>                        Before:   After:
> 1 CPU:
> bpf_redirect_map:      8.4 Mpps  8.4 Mpps  (no change)
> bpf_redirect:          5.0 Mpps  8.4 Mpps  (+68%)
> 2 CPUs:
> bpf_redirect_map:     15.9 Mpps  16.1 Mpps  (+1% or ~no change)
> bpf_redirect:          9.5 Mpps  15.9 Mpps  (+67%)
>
> After this patch series, the only semantics different between the two variants
> of the bpf() helper (apart from the absence of a map argument, obviously) is
> that the _map() variant will return an error if passed an invalid map index,
> whereas the bpf_redirect() helper will succeed, but drop packets on
> xdp_do_redirect(). This is because the helper has no reference to the calling
> netdev, so unfortunately we can't do the ifindex lookup directly in the helper.
>
> Changelog:
>
> v3:
>   - Switch two more fields to avoid a list_head spanning two cache lines
>   - Include Jesper's tracepoint patch
>   - Also rename xdp_do_flush_map()
>   - Fix a few nits from Maciej

Applied. Thanks