TC filters when used as classifiers are bound to TC classes.
However, there is a hidden difference when adding them in different
orders:

1. If we add tc classes before its filters, everything is fine.
   Logically, the classes exist before we specify their ID's in
   filters, it is easy to bind them together, just as in the current
   code base.

2. If we add tc filters before the tc classes they bind, we have to
   do dynamic lookup in fast path. What's worse, this happens all
   the time not just once, because on fast path tcf_result is passed
   on stack, there is no way to propagate back to the one in tc filters.

This hidden difference hurts performance silently if we have many tc
classes in hierarchy.

This patch intends to close this gap by doing the reverse binding when
we create a new class, in this case we can actually search all the
filters in its parent, match and fixup by classid. And because
tcf_result is specific to each type of tc filter, we have to introduce
a new ops for each filter to tell how to bind the class.

Note, we still can NOT totally get rid of those class lookup in
->enqueue() because cgroup and flow filters have no way to determine
the classid at setup time, they still have to go through dynamic lookup.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Saeed Mahameed says:

====================
mlx5-updates-2017-08-31 (GRE Offloads support)

This series provides the support for MPLS RSS and GRE TX offloads and
RSS support.

The first patch from Gal and Ariel provides the mlx5 driver support for
ConnectX capability to perform IP version identification and matching in
order to distinguish between IPv4 and IPv6 without the need to specify the
encapsulation type, thus perform RSS in MPLS automatically without
specifying MPLS ethertyoe. This patch will also serve for inner GRE IPv4/6
classification for inner GRE RSS.

2nd patch from Gal, Adds the TX offloads support for GRE tunneled packets,
by reporting the needed netdev features.

3rd patch from Gal, Adds GRE inner RSS support by creating the needed device
resources (Steering Tables/rules and traffic classifiers) to Match GRE traffic
and perform RSS hashing on the inner headers.

Improvement:
Testing 8 TCP streams bandwidth over GRE:
    System: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
    NIC: Mellanox Technologies MT28800 Family [ConnectX-5 Ex]
    Before: 21.3 Gbps (Single RQ)
    Now   : 90.5 Gbps (RSS spread on 8 RQs)
====================

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

Fix crash in linux PF driver when BARs have been cleared/de-programmed;
fail early init (prior to mapping BARs) if the BAR0 or
BAR1 registers are zero.

This situation can arise when the PF is added to a VM (PCI pass-through),
then a PF FLR is issued (in the VM).  After this occurs, the BAR registers
will be zero. If we attempt to load the PF driver in the host
(after VM has been shutdown), the host can reset.

Signed-off-by: Rick Farrington <ricardo.farrington@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

IPv4 name uses "destination ip" as does the IPv6 patch set.
Make the mac field consistent.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There are two typos in the document, netvsc.txt,
regarding UDP hashing level. This patch fixes them.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

xennet_start_xmit() might copy skb with inappropriate layout
into a fresh one.

Old skb is freed, and at this point it is not a drop, but
a consume. New skb will then be either consumed or dropped.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Introduce a new flow table and indirect TIRs which are used to hash the
inner packet headers of GRE tunneled packets.

When a GRE tunneled packet is received, the TTC flow table will match
the new IPv4/6->GRE rules which will forward it to the inner TTC table.
The inner TTC is similar to its counterpart outer TTC table, but
matching the inner packet headers instead of the outer ones (and does
not include the new IPv4/6->GRE rules).
The new rules will not add steering hops since they are added to an
already existing flow group which will be matched regardless of this
patch. Non GRE traffic will not be affected.

The inner flow table will forward the packet to inner indirect TIRs
which hash the inner packet and thus result in RSS for the tunneled
packets.

Testing 8 TCP streams bandwidth over GRE:
System: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
NIC: Mellanox Technologies MT28800 Family [ConnectX-5 Ex]
Before: 21.3 Gbps (Single RQ)
Now   : 90.5 Gbps (RSS spread on 8 RQs)

Signed-off-by: Gal Pressman <galp@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Add TX offloads support for GRE tunneled packets by reporting the needed
netdev features.

Signed-off-by: Gal Pressman <galp@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

This change adds the ability for flow steering to classify IPv4/6
packets with MPLS tag (Ethertype 0x8847 and 0x8848) as standard IP
packets and hit IPv4/6 classification steering rules.

Since IP packets with MPLS tag header have MPLS ethertype, they
missed the IPv4/6 ethertype rule and ended up hitting the default
filter forwarding all the packets to the same single RQ (No RSS).

Since our device is able to look past the MPLS tag and identify the
next protocol we introduce this solution which replaces ethertype
matching by the device's capability to perform IP version
identification and matching in order to distinguish between IPv4 and
IPv6.
Therefore, when driver is performing flow steering configuration on the
device it will use IP version matching in IP classified rules instead
of ethertype matching which will cause relevant MPLS tagged packets to
hit this rule as well.

If the device doesn't support IP version matching the driver will fall back
to use legacy ethertype matching in the steering as before.

Signed-off-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Ariel Levkovich <lariel@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Trivial fix to typos in printf error messages:
"conenct" -> "connect"
"listeen" -> "listen"

thanks to Daniel Borkmann for spotting one of these mistakes

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Shuah Khan <shuahkh@osg.samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Trivial fix to spelling mistake in DP_NOTICE message

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch removes the wrong check being done for the phy device being
returned by the mdiobus_get_phy() function. This function never returns
the error pointers.

Fixes: 256727da ("net: hns3: Add MDIO support to HNS3 Ethernet
Driver for hip08 SoC")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Salil Mehta <salil.mehta@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make this const as it is never modified.

Signed-off-by: Bhumika Goyal <bhumirks@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

IPv6 packet may carry more than one extension header, and IPv6 nodes must
accept and attempt to process extension headers in any order and occurring
any number of times in the same packet. Hence, there should be no
assumption that Segment Routing extension header is to appear immediately
after the IPv6 header.

Moreover, section 4.1 of RFC 8200 gives a recommendation on the order of
appearance of those extension headers within an IPv6 packet. According to
this recommendation, Segment Routing extension header should appear after
Hop-by-Hop and Destination Options headers (if they present).

This patch fixes the get_srh(), so it gets the segment routing header
regardless of its position in the chain of the extension headers in IPv6
packet, and makes sure that the IPv6 routing extension header is of Type 4.

Signed-off-by: Ahmed Abdelsalam <amsalam20@gmail.com>
Acked-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>

Antoine Tenart says:

====================
net: mvpp2: comphy configuration

This series, following up the one one the GoP/MAC configuration, aims at
stopping to depend on the firmware/bootloader configuration when using
the PPv2 engine. With this series the PPv2 driver does not need to rely
on a previous configuration, and dynamic reconfiguration while the
kernel is running can be done (i.e. switch one port from SGMII to 10G,
or the opposite). A port can now be configured in a different mode than
what's done in the firmware/bootloader as well.

The series first contain patches in the generic PHY framework to support
what is called the comphy (common PHYs), which is an h/w block providing
PHYs that can be configured in various modes ranging from SGMII, 10G
to SATA and others. As of now only the SGMII and 10G modes are
supported by the comphy driver.

Then patches are modifying the PPv2 driver to first add the comphy
initialization sequence (i.e. calls to the generic PHY framework) and to
then take advantage of this to allow dynamic reconfiguration (i.e.
configuring the mode of a port given what's connected, between sgmii and
10G). Note the use of the comphy in the PPv2 driver is kept optional
(i.e. if not described in dt the driver still as before an relies on the
firmware/bootloader configuration).

Finally there are dt/defconfig patches to describe and take advantage of
this.

This was tested on a range of devices: 8040-db, 8040-mcbin and 7040-db.

@Dave: the dt patches should go through the mvebu tree (patches 9-13).

Thanks!
Antoine

Since v3:
  - Now use of_phy_simple_xlate() to retrieve the phy.
  - Added an owner in the phy_ops structure.
  - Now allow the module to be selected with COMPILE_TEST.
  - Removed unused parameter in the comphy set_mode functions.
  - Added Kishon Acked-by in patch 1.

Since v2:
  - Kept the link mode enforcement.
  - Removed the netif_running() check.
  - Reworded the "dynamic reconfiguration of the PHY mode" commit log.
  - Added one patch not to force the GMAC autoneg parameters when using
    the XLG MAC.

Since v1:
  - Updated the mode settings variable name in the comphy driver to
    have 'cp110' in it.
  - Documented the PHY cell argument in the dt documentation.
  - New patch adding comphy phandles for the 7040-db board.
  - Checked if the carrier_on/off functions were needed. They are.
  - s/PHY/generic PHY/ in commit log of patch 1.
  - Rebased on the latest net-next/master.
====================

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

This patch adds logic to reconfigure the comphy/GoP/MAC when the link
state is updated at runtime. This is very useful on boards where many
link speed are supported: depending on what is negotiated the PPv2
driver will automatically reconfigures the link between the PHY and the
MAC.

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When using the XLG MAC, it does not make sense to force the GMAC autoneg
parameters. This patch adds checks to only set the GMAC autoneg
parameters when needed (i.e. when not using the XLG MAC).

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When the link status changes, the phylib calls the link_event function
in the mvpp2 driver. Before this patch only the egress/ingress transmit
was enabled/disabled. This patch adds more functionality to the link
status management code by enabling/disabling the port per-cpu
interrupts, and the port itself. The queues are now stopped as well, and
the netif carrier helpers are called.

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The link_event function is somewhat complicated. This cosmetic patch
simplifies it.

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On some platforms, the comphy is between the MAC GoP and the PHYs. The
mvpp2 driver currently relies on the firmware/bootloader to configure
the comphy. As a comphy driver was added to the generic PHY framework,
this patch uses it in the mvpp2 driver to configure the comphy at boot
time to avoid relying on the bootloader.

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The Marvell Armada 7K/8K SoCs contains an hardware block called COMPHY
that provides a number of shared PHYs used by various interfaces in the
SoC: network, SATA, PCIe, etc. This Device Tree binding allows to
describe this COMPHY hardware block.

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On the CP110 unit, which can be found on various Marvell platforms such
as the 7k and 8k (currently), a comphy (common PHYs) hardware block can
be found. This block provides a number of PHYs which can be used in
various modes by other controllers (network, SATA ...). These common
PHYs must be configured for the controllers using them to work correctly
either at boot time, or when the system runs to switch the mode used.
This patch adds a driver for this comphy hardware block, providing
callbacks for the its PHYs so that consumers can configure the modes
used.

As of this commit, two modes are supported by the comphy driver: sgmii
and 10gkr.

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Acked-by: Kishon Vijay Abraham I <kishon@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds more generic PHY modes to the phy_mode enum, to
allow configuring generic PHYs to the SGMII and/or the 10GKR mode
by using the set_mode callback.

Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Acked-by: Kishon Vijay Abraham I <kishon@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We should not hold a spinlock while pushing the skb into the networking
stack, so move the call to netif_rx_ni out of the critical region to where
we have dropped the spinlock.

Signed-off-by: Stefan Sørensen <stefan.sorensen@spectralink.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Chris Mi says:

====================
net/sched: Improve getting objects by indexes

Using current TC code, it is very slow to insert a lot of rules.

In order to improve the rules update rate in TC,
we introduced the following two changes:
        1) changed cls_flower to use IDR to manage the filters.
        2) changed all act_xxx modules to use IDR instead of
           a small hash table

But IDR has a limitation that it uses int. TC handle uses u32.
To make sure there is no regression, we add several new IDR APIs
to support unsigned long.

v2
==

Addressed Hannes's comment:
express idr_alloc in terms of idr_alloc_ext and most of the other functions
====================

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

Typically, each TC filter has its own action. All the actions of the
same type are saved in its hash table. But the hash buckets are too
small that it degrades to a list. And the performance is greatly
affected. For example, it takes about 0m11.914s to insert 64K rules.
If we convert the hash table to IDR, it only takes about 0m1.500s.
The improvement is huge.

But please note that the test result is based on previous patch that
cls_flower uses IDR.

Signed-off-by: Chris Mi <chrism@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, all filters with the same priority are linked in a doubly
linked list. Every filter should have a unique handle. To make the
handle unique, we need to iterate the list every time to see if the
handle exists or not when inserting a new filter. It is time-consuming.
For example, it takes about 5m3.169s to insert 64K rules.

This patch changes cls_flower to use IDR. With this patch, it
takes about 0m1.127s to insert 64K rules. The improvement is huge.

But please note that in this testing, all filters share the same action.
If every filter has a unique action, that is another bottleneck.
Follow-up patch in this patchset addresses that.

Signed-off-by: Chris Mi <chrism@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The following new APIs are added:

int idr_alloc_ext(struct idr *idr, void *ptr, unsigned long *index,
                  unsigned long start, unsigned long end, gfp_t gfp);
void *idr_remove_ext(struct idr *idr, unsigned long id);
void *idr_find_ext(const struct idr *idr, unsigned long id);
void *idr_replace_ext(struct idr *idr, void *ptr, unsigned long id);
void *idr_get_next_ext(struct idr *idr, unsigned long *nextid);

Signed-off-by: Chris Mi <chrism@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Subash Abhinov Kasiviswanathan says:

====================
net: Add support for rmnet driver

This patch series adds support for the rmnet driver which is required to
support recent chipsets using Qualcomm Technologies, Inc. modems. The data
from hardware follows the multiplexing and aggregation protocol (MAP).

This driver can be used to register onto any physical network device in
IP mode. Physical transports include USB, HSIC, PCIe and IP accelerator.

rmnet driver helps to decode these packets and queue them to network
stack (and encode and transmit it to the physical device).

v1: Same as the RFC patch with some minor fixes for issues reported by
kbuild test robot.

v1->v2: Change datatypes and remove config IOCTL as mentioned by David.
Also fix checkpatch issues and remove some unused code.

v2->v3: Move location to drivers/net and rename to rmnet. Change the
userspace - netlink communication from custom netlink to rtnl_link_ops.
Refactor some code. Use a fixed config for ingress and egress.

v3->v4: Move location to drivers/net/ethernet/qualcomm/.
Fix comments from Stephen and Jiri -
Split the ether and arp type changes into seperate patches.
Remove debug and custom logging and switch to standard netdevice log.
Remove module parameters. Refactor and change some code style issues.

v4->v5: Rename some structs and variables. Move the initializer
before the for loop start. Put the arp type in correct sequence.

v5->v6: Fix comments from Dan -
Use the upper link API. As a result, remove all the refcounting logic.
Device refcount is explicitly held on real_dev on rx_handler
registration only. Modifiy the flow control struct. Remove the unused
ethernet mode handling.

v6->v7: Fix comments from David - Add newline to end of Makefile. Remove
inline from .c files. Move the module init/exit to rmnet config. Fix an
error reported by kbuild test robot for an unused file.

v7->v8: Use a smaller value for ETH_P_MAP as mentioned by David. Change
netdev_info to netdev_dbg as mentioned by Andew. Fix comments from
Stephen regarding netdev_priv and sparse related errors of using 0 as NULL

v8->v9: Fix comments from David - Remove the CFLAG rule. Change the way
rmnet devices are freed. Instead of using a workqueue to unregister devices
individually, go through the list and free all devices within the rtnl_lock().

v9->v10: Actually fix the locking as mentioned by David. The locking scheme is
mentioned in a comment in rmnet_config.c. Change comment near MAP type
definition as mentioned by Dan. Refactor some code.

v10->v11: Allow RMNET to compile as a module as mentioned by David
====================

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

RmNet driver provides a transport agnostic MAP (multiplexing and
aggregation protocol) support in embedded module. Module provides
virtual network devices which can be attached to any IP-mode
physical device. This will be used to provide all MAP functionality
on future hardware in a single consistent location.

Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Define the raw IP type. This is needed for raw IP net devices
like rmnet.

Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Define the Qualcomm multiplexing and aggregation (MAP) ether type 0x00F9.
This is needed for receiving data in the MAP protocol like RMNET. This is
not an officially registered ID.

Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Florian Westphal says:

====================
tcp: re-add header prediction

Eric reported a performance regression caused by header prediction
removal.

We now call tcp_ack() much more frequently, for some workloads
this brings in enough cache line misses to become noticeable.

We could possibly still kill HP provided we find a different
way to suppress unneeded tcp_ack, but given we're late in
the cycle it seems preferable to revert.
====================

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

This reverts commit 45f119bf.

Eric Dumazet says:
  We found at Google a significant regression caused by
  45f119bf tcp: remove header prediction

  In typical RPC  (TCP_RR), when a TCP socket receives data, we now call
  tcp_ack() while we used to not call it.

  This touches enough cache lines to cause a slowdown.

so problem does not seem to be HP removal itself but the tcp_ack()
call.  Therefore, it might be possible to remove HP after all, provided
one finds a way to elide tcp_ack for most cases.

Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

This change was a followup to the header prediction removal,
so first revert this as a prerequisite to back out hp removal.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

When moving the IRDA code out of net/ into drivers/staging/irda/net, the
link order changes when IRDA is built into the kernel.  That causes a
kernel crash at boot time as netfilter isn't initialized yet.

To fix this, move the init call level of the irda core to be
device_initcall() as the link order keeps this being initialized at the
correct time.

Reported-by: kernel test robot <fengguang.wu@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

A stray return was added in the macro bcmgenet_##name##_writel where it
should not, drop it.

Reported-by: kbuild test robot <fengguang.wu@intel.com>
Fixes: 69d2ea9c ("net: bcmgenet: Use correct I/O accessors")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Florian reported UDP xmit drops that could be root caused to the
too small neigh limit.

Current limit is 64 KB, meaning that even a single UDP socket would hit
it, since its default sk_sndbuf comes from net.core.wmem_default
(~212992 bytes on 64bit arches).

Once ARP/ND resolution is in progress, we should allow a little more
packets to be queued, at least for one producer.

Once neigh arp_queue is filled, a rogue socket should hit its sk_sndbuf
limit and either block in sendmsg() or return -EAGAIN.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since the removal of NET_DMA, dmaengine.h header file shouldn't be needed
by netdevice.h anymore.

Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The GENET driver currently uses __raw_{read,write}l which means
native I/O endian. This works correctly for an ARM LE kernel (default)
but fails miserably on an ARM BE (BE8) kernel where registers are kept
little endian, so replace uses with {read,write}l_relaxed here which is
what we want because this is all performance sensitive code.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>