During driver probing the following code path is triggered.
igb_probe
->igb_sw_init
  ->igb_probe_vfs
    ->igb_pci_enable_sriov
      ->igb_sriov_reinit

Doing the SR-IOV re-init is not necessary during probing since we're
starting from scratch. Here we can call igb_enable_sriov() right away.

Running igb_sriov_reinit() during igb_probe() also seems to cause
occasional packet loss on some onboard 82576 NICs. Reproduced on
Dell and HP servers with onboard 82576 NICs.
Example:
Intel Corporation 82576 Gigabit Network Connection [8086:10c9] (rev 01)
Subsystem: Dell Device [1028:0481]
Signed-off-by: Stefan Assmann <sassmann@kpanic.de>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

When initializing igb driver (e.g. 82576, I350), IGB_FLAG_QUEUE_PAIRS is
set if adapter->rss_queues exceeds half of max_rss_queues in
igb_init_queue_configuration().
On the other hand, IGB_FLAG_QUEUE_PAIRS is not set even if the number of
queues exceeds half of max_combined in igb_set_channels() when changing
the number of queues by "ethtool -L".
In this case, if numvecs is larger than MAX_MSIX_ENTRIES (10), the size
of adapter->msix_entries[], an overflow can occur in
igb_set_interrupt_capability(), which in turn leads to an oops.

Fix this problem as follows:
 - When changing the number of queues by "ethtool -L", set
   IGB_FLAG_QUEUE_PAIRS in the same way as initializing igb driver.
 - When increasing the size of q_vector, reallocate it appropriately.
   (With IGB_FLAG_QUEUE_PAIRS set, the size of q_vector gets larger.)

Another possible way to fix this problem is to cap the queues at its
initial number, which is the number of the initial online cpus. But this
is not the optimal way because we cannot increase queues when another
cpu becomes online.

Note that before commit cd14ef54 ("igb: Change to use statically
allocated array for MSIx entries"), this problem did not cause oops
but just made the number of queues become 1 because of entering msi_only
mode in igb_set_interrupt_capability().

Fixes: 907b7835 ("igb: Add ethtool support to configure number of channels")
CC: stable <stable@vger.kernel.org>
Signed-off-by: Shota Suzuki <suzuki_shota_t3@lab.ntt.co.jp>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Phil Sutter says:

====================
net: Convert drivers to IFF_NO_QUEUE and cleanup afterwards

This series converts in-tree users away from the old and deprecated
'tx_queue_len = 0' idiom, adds a warning to notify out-of-tree driver
maintainers that there is need for action on their behalf and finally drops any
workarounds in scheduling algorithm implementations.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Those were all workarounds for the formerly double meaning of
tx_queue_len, which broke scheduling algorithms if untreated.

Now that all in-tree drivers have been converted away from setting
tx_queue_len = 0, it should be safe to drop these workarounds for
categorically broken setups.
Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Due to the introduction of IFF_NO_QUEUE, there is a better way for
drivers to indicate that no qdisc should be attached by default. Though,
the old convention can't be dropped since ignoring that setting would
break drivers still using it. Instead, add a warning so out-of-tree
driver maintainers get a chance to adjust their code before we finally
get rid of any special handling of tx_queue_len == 0.
Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Johnny Kim <johnny.kim@atmel.com>
Cc: Rachel Kim <rachel.kim@atmel.com>
Cc: Dean Lee <dean.lee@atmel.com>
Cc: Chris Park <chris.park@atmel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Dmitry Tarnyagin <dmitry.tarnyagin@lockless.no>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Arvid Brodin <arvid.brodin@alten.se>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Marek Lindner <mareklindner@neomailbox.ch>
Cc: Simon Wunderlich <sw@simonwunderlich.de>
Cc: Antonio Quartulli <antonio@meshcoding.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Jouni Malinen <j@w1.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Jay Vosburgh <j.vosburgh@gmail.com>
Cc: Veaceslav Falico <vfalico@gmail.com>
Cc: Andy Gospodarek <gospo@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Alexander Aring <alex.aring@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Tom Herbert says:

====================
net: Identifier Locator Addressing - Part I

This patch set provides rudimentary support for Identifier Locator
Addressing or ILA. The basic concept of ILA is that we split an IPv6
address into a 64 bit locator and 64 bit identifier. The identifier is
the identity of an entity in communication ("who"), and the locator
expresses the location of the entity ("where"). Applications
use externally visible address that contains the identifier.
When a packet is actually sent, a translation is done that
overwrites the first 64 bits of the address with a locator.
The packet can then be forwarded over the network to the host where
the addressed entity is located. At the receiver, the reverse
translation is done so the that the application sees the original,
untranslated address. Presumably an external control plane will
provide identifier->locator mappings.

v2:
  - Fix compilation erros when LWT not configured
  - Consolidate ILA into a single ila.c

v3:
  - Change pseudohdr argument od inet_proto_csum_replace functions to
    be a bool

v4:
  - In ila_build_state check locator being in netlink params before
    allocating tunnel state

The data path for ILA is a simple NAT translation that only operates
on the upper 64 bits of a destination address in IPv6 packets. The
basic process is:

   1) Lookup 64 bit identifier (lower 64 bits of destination)
   2) If a match is found
      a) Overwrite locator (upper 64 bits of destination) with
         the new locator
      b) Adjust any checksum that has destination address included in
         pseudo header
   3) Send or receive packet

ILA is a means to implement tunnels or network virtualization without
encapsulation. Since there is no encapsulation involved, we assume that
stateless support in the network for IPv6 (e.g. RSS, ECMP, TSO, etc.)
just works. Also, since we're minimally changing the packet many of
the worries about encapsulation (MTU, checksum, fragmentation) are
not relevant. The downside is that, ILA is not extensible like other
encapsulations (GUE for instance) so it might not be appropriate for
all use cases. Also, this only makes sense to do in IPv6!

A key aspect of ILA is performance. The intent is that ILA would be
used in data centers in virtualizing tasks or jobs. In the fullest
incarnation all intra data center communications might be targeted to
virtual ILA addresses. This is basically adding a new virtualization
capability to the existing services in a datacenter, so there is a
strong expectation is that this does not degrade performance for
existing applications.

Performance seems to be dependent on how ILA is hooked into kernel.
ILA can be implemented under some different models:

  - Mechanically it is a form a stateless DNAT
  - It can be thought of as a type of (source) routing
  - As a functional replacement of encapsulation

In this patch set we hook into the data path using Light Weight
Tunnels (LWT) infrastructure. As part of that, we add support in LWT
to redirect dst input. iproute will be modified to take a new ila encap
type. ILA can be configured like:

ip route add 3333:0:0:1:5555:0:2:0/128 \
   encap ila 2001:0:0:2 via 2401:db00:20:911a:face:0:27:0

ip -6 addr add 3333:0:0:1:5555:0:1:0/128 dev eth0

ip route add table local local 2001:0:0:1:5555:0:1:0/128
   encap ila 3333:0:0:1 dev lo

So sending to destination 3333:0:0:1:5555:0:2:0 will have destination
of 2001:0:0:2:5555:0:2:0 on the wire.

Performance results are below. With ILA we see about a 10% drop in
pps compared to non-ILA. Much of this drop can be attributed to the
loss of early demux on input (translation occurs after it is attempted).
We will address this in the next patch set. Also, IPvlan input path
does not work with ILA since the routing is bypassed-- this will
be addressed in a future patch.

Performance testing:

Performing netperf TCP_RR with 200 clients:

Non-ILA baseline
  84.92% CPU utilization
  1861922.9 tps
  93/163/330 50/90/99% latencies

ILA single destination
  83.16% CPU utilization
  1679683.4 tps
  105/180/332 50/90/99% latencies

References:

Slides from netconf:
http://vger.kernel.org/netconf2015Herbert-ILA.pdf

Slides from presentation at IETF:
https://www.ietf.org/proceedings/92/slides/slides-92-nvo3-1.pdf

I-D:
https://tools.ietf.org/html/draft-herbert-nvo3-ila-00
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Adding new module name ila. This implements ILA translation. Light
weight tunnel redirection is used to perform the translation in
the data path. This is configured by the "ip -6 route" command
using the "encap ila <locator>" option, where <locator> is the
value to set in destination locator of the packet. e.g.

ip -6 route add 3333:0:0:1:5555:0:1:0/128 \
      encap ila 2001:0:0:1 via 2401:db00:20:911a:face:0:25:0

Sets a route where 3333:0:0:1 will be overwritten by
2001:0:0:1 on output.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This function updates a checksum field value and skb->csum based on
a value which is the difference between the old and new checksum.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

inet_proto_csum_replace4,2,16 take a pseudohdr argument which indicates
the checksum field carries a pseudo header. This argument should be a
boolean instead of an int.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds the capability to redirect dst input in the same way
that dst output is redirected by LWT.

Also, save the original dst.input and and dst.out when setting up
lwtunnel redirection. These can be called by the client as a pass-
through.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

>> drivers/net/ethernet/cisco/enic/vnic_dev.c:1095:13: sparse: incorrect type in assignment (different address spaces)
   drivers/net/ethernet/cisco/enic/vnic_dev.c:1095:13:    expected void *res
   drivers/net/ethernet/cisco/enic/vnic_dev.c:1095:13:    got void [noderef] <asn:2>*
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

>> drivers/net/ethernet/mellanox/mlx5/core/en_rx.c:173:44: sparse: incorrect type in argument 1 (different base types)
   drivers/net/ethernet/mellanox/mlx5/core/en_rx.c:173:44:    expected restricted __sum16 [usertype] n
   drivers/net/ethernet/mellanox/mlx5/core/en_rx.c:173:44:    got restricted __be16 [usertype] check_sum
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Table lookup compiles out when VRF is not enabled.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

kbuild test robot reported:
tree:   git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git master
head:   d52736e2
commit: 4e3c8992 [751/762] net: Introduce VRF related flags and helpers
reproduce: make htmldocs

>> Warning(include/linux/netdevice.h:1293): Enum value 'IFF_VRF_MASTER' not described in enum 'netdev_priv_flags'
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

As Eric noted netif_index_is_vrf is not called with rcu_read_lock held,
so wrap the dev_get_by_index_rcu in rcu_read_lock and unlock.

If VRF is not enabled or oif is 0 skip the device lookup. In both cases
index cannot be the VRF master.
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Achiad Shochat says:

====================
Driver updates 16-Aug-2015

This patchset contains bug fixes, new RSS and pause parameters ethtool
options, and support for RX CHECKSUM_COMPLETE.

Patchset was applied and tested over commit adc6310 ("Merge branch
'mv88e6xxx-switchdev-fdb'").
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Only for packets with first ethertype set to IPv4/6 for now.
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Only rx/tx pause settings.
Autoneg setting is currently not supported.
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

- Port speed settings are applied by the device only upon
  port admin status transition from DOWN to UP.
  So we enforce this transition regardless of the port's
  current operation state (which may be occasionally DOWN if
  for example the network cable is disconnected).
- Fix the PORT_UP/DOWN device interface enum
- Set the local_port bit in the device PAOS register
- EXPORT the PAOS (Port Administrative and Operational Status)
  register set/query access functions.
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

- Change the maximum LRO session size from 16KB to 64KB
- Reduce the LRO session timeout from 512us to 32us in
  order to reduce the TCP latency of non-LRO'ed flows.
- Fix skb_shinfo(skb)->gso_size and set skb_shinfo(skb)->gso_type.
- Fix a bug accessing un-initialized mdev pointer.
Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>