netdev - [PATCH net 0/5] nfp: flower fixes and typo in ethtool stats name

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <20171117010643.18308-1-jakub.kicinski@netronome.com>
Date:   Thu, 16 Nov 2017 17:06:38 -0800
From:   Jakub Kicinski <jakub.kicinski@...ronome.com>
To:     netdev@...r.kernel.org
Cc:     oss-drivers@...ronome.com,
        Jakub Kicinski <jakub.kicinski@...ronome.com>
Subject: [PATCH net 0/5] nfp: flower fixes and typo in ethtool stats name

Hi!

This set comes from the flower offload team.  From Pieter we have a fix
to the semantics of the flag telling FW whether to allocate or free
a mask and correction of a typo in name of one of the MAC statistics 
(reveive -> received, we use past participle to match HW docs).

Dirk fixes propagation of max MTU to representors.

John improves VXLAN offload.  The old code was not using egress_dev at
all, so Jiri missed it in his conversion.  The validation of ingress
port is still not perfect, we will have to wait for shared block dust
to settle to tackle it.  This is how John explains the cases:

The following example rule is now correctly offloaded in net-next kernel:

tc filter add dev vxlan0 ... enc_dst_port 4789 ... skip_sw \
  action redirect dev nfp_p0

The following rule will not be offloaded to the NFP (previously it
incorrectly matched vxlan packets - it shouldn't as ingress dev is not
a vxlan netdev):

tc filter add dev nfp_p0 ... enc_dst_port 4789 ... skip_sw \
  action redirect dev nfp_p0

Rules that are not matching on tunnels and are an egress offload are
rejected. The standard match code assumes the offloaded repr is the
ingress port. Rejecting egress offloads removes the chances of false
interpretation of the rules on the NFP.

A know issue is that the following rule example could still be offloaded
and incorrectly match tunnel data:

tc filter add dev dummy ... enc_dst_port 4789 ... skip_sw \
  action redirect dev nfp_p0

Because the egress register callback does not give information on the
ingress netdev, the patch assumes that if it is not a repr then it is 
the correct tunnel netdev. This may not be the case. The chances of this
happening is reduced as it is enforced that the rule match on the well
known vxlan port but it is still possible.

Dirk van der Merwe (1):
  nfp: inherit the max_mtu from the PF netdev

John Hurley (2):
  nfp: register flower reprs for egress dev offload
  nfp: remove false positive offloads in flower vxlan

Pieter Jansen van Vuuren (2):
  nfp: fix flower offload metadata flag usage
  nfp: fix vlan receive MAC statistics typo

 drivers/net/ethernet/netronome/nfp/flower/main.c   | 18 +++++++++++
 drivers/net/ethernet/netronome/nfp/flower/main.h   |  5 +--
 .../net/ethernet/netronome/nfp/flower/metadata.c   |  7 +++--
 .../net/ethernet/netronome/nfp/flower/offload.c    | 36 ++++++++++++++++++----
 drivers/net/ethernet/netronome/nfp/nfp_app.h       | 20 ++++++++++++
 .../net/ethernet/netronome/nfp/nfp_net_ethtool.c   |  2 +-
 drivers/net/ethernet/netronome/nfp/nfp_net_repr.c  | 11 ++++++-
 drivers/net/ethernet/netronome/nfp/nfp_port.h      |  2 +-
 8 files changed, 88 insertions(+), 13 deletions(-)

-- 
2.14.1