A driver may use build_skb() for received packets.
These SKBs then have a head_frag.

Since commit d7e8883c ("net: make GRO aware of
skb->head_frag"), GRO may build frag_list SKBs out of
head_frag received SKBs.
In such a case, the chained SKBs end up with a head_frag.

Commit 07b26c94

 ("gso: Support partial splitting at
the frag_list pointer") adds partial segmentation of frag_list
SKB chains into individual SKBs.
However, this is not done if the chained SKBs have any
linear part, because the device may not be able to DMA
the private linear buffer.

A chained frag_list SKB with head_frag is wrongfully
detected in this case as having a private linear part
and thus falls back to software GSO, while in fact the
linear part is backed by a DMA page just like any other frag.

This causes low performance when forwarding those packets
that were built with build_skb()

Allow partial segmentation at the frag_list pointer for
chained SKBs with head_frag.

Note that such SKBs can only be created by GRO, when applied
to received packets with head_frag.
Also note that this change only affects the data path that
performs the partial segmentation at frag_list pointer, and
not any of the other more common data paths.

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Guillaume Nault says:

====================
l2tp: constify l2tp_session_get*() and l2tp_tunnel_find*()

Declare parameters of these functions as "const" where possible.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

l2tp_tunnel_find() and l2tp_tunnel_find_nth() don't modify "net".

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make l2tp_pernet()'s parameter constant, so that l2tp_session_get*() can
declare their "net" variable as "const".
Also constify "ifname" in l2tp_session_get_by_ifname().

Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

Benjamin Herrenschmidt says:

====================
ftgmac100: Rework batch 4 - Misc

This is v2 of the fourth batch of updates to the ftgmac100 driver.

This is a bunch of misc cleanups and fixes, such as properly
disabling HW checksum generation on AST2400 where it's known
to be broken and some chip init updates.

This also adds the ability to turn HW checksum on/off and
configure the ring sizes via ethtool.

v2 Fixes patch 1/10 (NETIF_F_HW_CSUM conversion)

The next (and last) batch will add a few more "features" such
as netpoll, multicast/promist, vlan offload...
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

I haven't seen any improvement above that size on the machines
I've tested with.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

We set an arbitrary max at 1024 since we pre-allocate the actual
descriptor arrays and skb arrays to the full size to keep the
code a bit simpler and avoid allocation failures in the reset
task.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Clear stale interrupts on entry, configure FIFO sizes, set FIFO
thresholds, configure interrupt mitigation.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The helpers just take space but don't provide much value. Simple
one line comments are more explanatory.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

To remove more confusion. This function is about obtaining the
initial MAC address at driver probe time.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

To avoid confusion with the ndo callback and generally be
clearer about the purpose of that function

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

So features can be turned on/off via ethtool

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

We found out that HW checksum generation only works from AST2500
onward. This disables it on AST2400 and removes the "no-hw-checksum"
properties in the device-trees. The problem we had wasn't related
to NC-SI.

Also rework the logic testing for that property so it can be used
to disable HW checksum generation and checking regardless of whether
NC-SI is used or not in case other variants out there need this.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

We test for aspeed chips to handle a couple of special cases,
but we do that by checking the machine type which isn't right.

Instead check the actual device compatible property. This also
updates the dtsi files for the aspeed SoC to match.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The documentation describes NETIF_F_IP_CSUM as deprecated
so let's switch to NETIF_F_HW_CSUM and use the helper to
handle unhandled protocols.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Ursula Braun says:

====================
net/smc: patches for net-next

here are some patches for net/smc. Most important are
improvements for socket closing.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

smc specifies IB_SEND_INLINE for IB_WR_SEND ib_post_send calls, but
provides a mapped buffer to be sent. This is inconsistent, since
IB_SEND_INLINE works without mapped buffer. Problem has not been
detected in the past, because tests had been limited to Connect X3 cards
from Mellanox, whose mlx4 driver just ignored the IB_SEND_INLINE flag.
For now, the IB_SEND_INLINE flag is removed.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make sure sockets never accepted are removed cleanly.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

unhash is already called in sock_put_work. Remove the second call.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

State SMC_CLOSED should be reached only, if ConnClosed has been sent to
the peer. If ConnClosed is received from the peer, a socket with
shutdown SHUT_WR done, switches errorneously to state SMC_CLOSED, which
means the peer socket is dangling. The local SMC socket is supposed to
switch to state APPFINCLOSEWAIT to make sure smc_close_final() is called
during socket close.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Several state changes occur during SMC socket closing. Currently
state changes triggered locally occur in process context with
lock_sock() taken while state changes triggered by peer occur in
tasklet context with bh_lock_sock() taken. bh_lock_sock() does not
wait till a lock_sock(() task in process context is finished. This
may lead to races in socket state transitions resulting in dangling
SMC-sockets, or it may lead to duplicate SMC socket freeing.
This patch introduces a closing worker to run all state changes under
lock_sock().

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Reported-by: Dave Jones <davej@codemonkey.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

Wake up reading file descriptors for a closing socket as well, otherwise
some socket applications may stall.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If peer indicates write_blocked, the cursor state of the received data
should be send to the peer immediately (in smc_tx_consumer_update()).
Afterwards the write_blocked indicator is cleared.

If there is no free slot for another write request, sending is postponed
to worker smc_tx_work, and the write_blocked indicator is not cleared.
Therefore another clearing check is needed in smc_tx_work().

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

SMC requires an active ib port on the RoCE device.
smc_pnet_find_roce_resource() determines the matching RoCE device port
according to the configured PNET table. Do not return the found
RoCE device port, if it is not flagged active.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The global event handler is created only, if the ib_device has already
been used by at least one link group. It is guaranteed that there exists
the corresponding entry in the smc_ib_devices list. Get rid of this
superfluous check.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch removes an outdated comment.

Signed-off-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In the submission of the lastest multiple buffer patch set, this fix was lost.
I am sending this patch to put it right again. The fix was originally proposed
by Arnd Bergmann.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Joao Pinto <jpinto@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

BPF helper functions access socket fields through skb->sk. This is not
set in ingress cgroup and socket filters. The association is only made
in skb_set_owner_r once the filter has accepted the packet. Sk is
available as socket lookup has taken place.

Temporarily set skb->sk to sk in these cases.

Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix the return value check which testing the wrong variable
in devlink_dpipe_header_put().

Fixes: 1555d204

 ("devlink: Support for pipeline debug (dpipe)")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Julian Wiedmann says:

====================
more s390/net updates

here's a second batch of s390/net patches for net-next.
A mixed bunch of qeth cleanups, and a few patches to add support for
ETHTOOL_GLINKSETTINGS.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

prepare() and complete() are not implemented by any discipline, so just
drop all the indirection.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Reviewed-by: Hans Wippel <hwippel@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

LINK_MODE_* replaces the u32-limited SUPPORTED_* / ENABLED_*
definitions.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Reviewed-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

get_settings() is deprecated and lacks support for higher link
speeds, so implement get_link_ksettings() instead.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Reviewed-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In preparation for moving to get_link_ksettings(), clean up how
we build the supported and advertised port/speed masks.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Reviewed-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Reviewed-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1. a buffer has 16 is_header flags, because that's its # of elements
2. replace the last occurrence of QETH_HEADER_SIZE, and remove it

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Acked-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ndo_start_xmit() expects us to return netdev_tx_t here...

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Acked-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

'elements_needed' is not used in qeth_do_send_packet_fast(),
so consequently remove it.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Acked-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Duplicated code.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Acked-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Identical code, we just need to call a layer-specific hook
to process any received buffer.

qeth_buffer_reclaim_work() is shuffled around to avoid a
forward declaration for qeth_queue_input_buffer().

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Acked-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There's a number of layer-independent ioctls that we can handle
in core, and reduce code duplication. For layer-specific ioctls,
add a do_ioctl() discipline hook.

Signed-off-by: Julian Wiedmann <jwi@linux.vnet.ibm.com>
Acked-by: Ursula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>