Add more statistics to be collected from the HW periodically. These stats
are tclass based (beside ECN marked packet, that exist only port based).
They are needed to expose RED qdisc stats and xstats correctly.
Signed-off-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This adds the counter group definitions for 2 new counter groups
which are necessary for gaining ECN & wred counters.
Signed-off-by: Yuval Mintz <yuvalm@mellanox.com>
Signed-off-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support for ndo_setup_tc with enum tc_setup_type value of TC_SETUP_RED.
This call sets RED qdisc on a traffic class.
This patch supports RED qdisc only as a root qdisc and set in on the
default tclass. It can be set with or without ECN.
Signed-off-by: Yuval Mintz <yuvalm@mellanox.com>
Signed-off-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds 2 new registers:
 - Congestion WRED ECN TClass Profile Register [CWTP]
 - Congestion WRED ECN TClass and Pool Mapping Register [CWTPM]

These registers would later be needed to offload RED-related
functionality to the HW.
Signed-off-by: Yuval Mintz <yuvalm@mellanox.com>
Signed-off-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Change TC_SETUP_CBS to TC_SETUP_QDISC_CBS to match the new convention..
Signed-off-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Change TC_SETUP_MQPRIO to TC_SETUP_QDISC_MQPRIO to match the new
convention.
Signed-off-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add the ability to offload RED qdisc by using ndo_setup_tc.
There are four commands for RED offloading:
* TC_RED_SET: handles set and change.
* TC_RED_DESTROY: handle qdisc destroy.
* TC_RED_STATS: update the qdiscs counters (given as reference)
* TC_RED_XSTAT: returns red xstats.

Whether RED is being offloaded is being determined every time dump action
is being called because parent change of this qdisc could change its
offload state but doesn't require any RED function to be called.
Signed-off-by: Nogah Frankel <nogahf@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The original patch had the wrong filename.

Fixes: bfdf7569 ("bpf: create samples/bpf/tcp_bpf.readme")
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Tom Herbert says:

====================
ila: make identifier format optional and other fixes

The identifier type and checksum neutral mapping bits are optional
in identifier formats. This patch set fixes the implementation to
make them optional and configurable.

Specific items:

  - Clean up checksum diff code in ILA
  - Add checksum neutral mapping auto so that checksum neutral
    mapping can be configured without requiring use of the C-bit
  - Add identifier type configuration and allow identifier
    type to be configured so that the identifier type field does
    not need to be present
  - Added ILA documention: ila.txt

I have fixes for ILA in iproute2 that will be poseted separately.

Tested: Ran netperf TCP_RR on various combinations of checksum
mode and the two supported identifier types.

v2:
  - Add proper sign off
  - In ILA LWT, only check prefix length includes identifier type
    if identifier type is enabled (ILA_ATYPE_USE_FORMAT).
  - Add a hook type so that it can be specified whether ILA
    translation is done on input or output route funciton in
    LWT.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add documenation for kernel ILA. This describes ILA, features,
configuration gives some examples.
Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

In LWT tunnels both an input and output route method is defined.
If both of these are executed in the same path then double translation
happens and the effect is not correct.

This patch adds a new attribute that indicates the hook type. Two
values are defined for route output and route output. ILA
translation is only done for the one that is set. The default is
to enable ILA on route output.
Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Allow identifier to be explicitly configured for a mapping.
This can either be one of the identifier types specified in the
ILA draft or a value of ILA_ATYPE_USE_FORMAT which means the
identifier type is inferred from the identifier type field.
If a value other than ILA_ATYPE_USE_FORMAT is set for a
mapping then it is assumed that the identifier type field is
not present in an identifier.
Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add checksum neutral auto that performs checksum neutral mapping
without using the C-bit. This is enabled by configuration of
a mapping.

The checksum neutral function has been split into
ila_csum_do_neutral_fmt and ila_csum_do_neutral_nofmt. The former
handles the C-bit and includes it in the adjustment value. The latter
just sets the adjustment value on the locator diff only.

Added configuration for checksum neutral map aut in ila_lwt
and ila_xlat.
Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Consolidate computing checksum diff into one function.

Add get_csum_diff_iaddr that computes the checksum diff between
an address argument and locator being written. get_csum_diff
calls this using the destination address in the IP header as
the argument.

Also moved ila_init_saved_csum to be close to the checksum

diff functions.
Signed-off-by: Tom Herbert <tom@quantonium.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Implements port to port forwarding with route table and arp table
lookup for ipv4 packets using bpf_redirect helper function and
lpm_trie  map.
Signed-off-by: Christina Jacob <Christina.Jacob@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is better for code locality and should slightly
speed up normal interrupts.

This also allows PPS clock output to start working for
i.mx7. This is because i.mx7 was already using the limit
of 3 interrupts, and needed another.
Signed-off-by: Troy Kisky <troy.kisky@boundarydevices.com>
Acked-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Vitaly Kuznetsov says:

====================
hv_netvsc: fix a hang on channel/mtu changes

It was found that netvsc driver doesn't survive e.g.

test. I was able to identify a hang in guest/host communication, it is
fixed by PATCH1 of this series. PATCH2 is a cosmetic change masking
unneeded messages.

Changes since v1:
- Throw away patches 2 and 3 of the original series as one is unneeded and
  the other is not justified [Eric Dumazet, Stephen Hemminger] so I'm only
  fixing the hang now, the crash doesn't reproduce. Will keep an eye on it.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Hyper-V hosts are known to send RNDIS messages even after we halt the
device in rndis_filter_halt_device(). Remove user visible messages
as they are not really useful.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

It was found that in some cases host refuses to teardown GPADL for send/
receive buffers (probably when some work with these buffere is scheduled or
ongoing). Change the teardown logic to be:
1) Send NVSP_MSG1_TYPE_REVOKE_* messages
2) Close the channel
3) Teardown GPADLs.
This seems to work reliably.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, we create a LED trigger for any link speed known to a PHY.
These triggers only fire when their exact link speed had been negotiated
(they aren't cumulative, that is, they don't fire for "their or any higher"
link speed).

What we are missing, however, is a trigger which will fire on any link
speed known to the PHY. Such trigger can then be used for implementing a
poor man's substitute of the "link" LED on boards that lack it.
Let's add it.
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, phy_led_trigger_change_speed() is handling a "no link" condition
like it was some kind of an error (using "goto" to a code at the function
end).

However, having no link at PHY is an ordinary operational state, so let's
handle it in an appropriately named separate function so it is more obvious
what the code is doing.
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fixes a case where GFP_ATOMIC allocation must be used instead of
GFP_KERNEL one.

[   54.891146]  lock_acquire+0xb3/0x2f0
[   54.891153]  ? fs_reclaim_acquire.part.60+0x5/0x30
[   54.891165]  fs_reclaim_acquire.part.60+0x29/0x30
[   54.891170]  ? fs_reclaim_acquire.part.60+0x5/0x30
[   54.891178]  kmem_cache_alloc_trace+0x3f/0x500
[   54.891186]  ? cyc2ns_read_end+0x1e/0x30
[   54.891196]  ipv6_add_addr+0x15a/0xc30
[   54.891217]  ? ipv6_create_tempaddr+0x2ea/0x5d0
[   54.891223]  ipv6_create_tempaddr+0x2ea/0x5d0
[   54.891238]  ? manage_tempaddrs+0x195/0x220
[   54.891249]  ? addrconf_prefix_rcv_add_addr+0x1c0/0x4f0
[   54.891255]  addrconf_prefix_rcv_add_addr+0x1c0/0x4f0
[   54.891268]  addrconf_prefix_rcv+0x2e5/0x9b0
[   54.891279]  ? neigh_update+0x446/0xb90
[   54.891298]  ? ndisc_router_discovery+0x5ab/0xf00
[   54.891303]  ndisc_router_discovery+0x5ab/0xf00
[   54.891311]  ? retint_kernel+0x2d/0x2d
[   54.891331]  ndisc_rcv+0x1b6/0x270
[   54.891340]  icmpv6_rcv+0x6aa/0x9f0
[   54.891345]  ? ipv6_chk_mcast_addr+0x176/0x530
[   54.891351]  ? do_csum+0x17b/0x260
[   54.891360]  ip6_input_finish+0x194/0xb20
[   54.891372]  ip6_input+0x5b/0x2c0
[   54.891380]  ? ip6_rcv_finish+0x320/0x320
[   54.891389]  ip6_mc_input+0x15a/0x250
[   54.891396]  ipv6_rcv+0x772/0x1050
[   54.891403]  ? consume_skb+0xbe/0x2d0
[   54.891412]  ? ip6_make_skb+0x2a0/0x2a0
[   54.891418]  ? ip6_input+0x2c0/0x2c0
[   54.891425]  __netif_receive_skb_core+0xa0f/0x1600
[   54.891436]  ? process_backlog+0xac/0x400
[   54.891441]  process_backlog+0xfa/0x400
[   54.891448]  ? net_rx_action+0x145/0x1130
[   54.891456]  net_rx_action+0x310/0x1130
[   54.891524]  ? RTUSBBulkReceive+0x11d/0x190 [mt7610u_sta]
[   54.891538]  __do_softirq+0x140/0xaba
[   54.891553]  irq_exit+0x10b/0x160
[   54.891561]  do_IRQ+0xbb/0x1b0

Fixes: f3d9832e ("ipv6: addrconf: cleanup locking in ipv6_add_addr")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Acked-by: David Ahern <dsahern@gmail.com>
Tested-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Roman Gushchin says:

====================
eBPF-based device cgroup controller

This patchset introduces an eBPF-based device controller for cgroup v2.

Patches (1) and (2) are a preparational work required to share some code
  with the existing device controller implementation.
Patch (3) is the main patch, which introduces a new bpf prog type
  and all necessary infrastructure.
Patch (4) moves cgroup_helpers.c/h to use them by patch (4).
Patch (5) implements an example of eBPF program which controls access
  to device files and corresponding userspace test.

v3:
  Renamed constants introduced by patch (3) to BPF_DEVCG_*

v2:
  Added patch (1).

v1:
  https://lkml.org/lkml/2017/11/1/363
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add a test for device cgroup controller.

The test loads a simple bpf program which logs all
device access attempts using trace_printk() and forbids
all operations except operations with /dev/zero and
/dev/urandom.

Then the test creates and joins a test cgroup, and attaches
the bpf program to it.

Then it tries to perform some simple device operations
and checks the result:

  create /dev/null (should fail)
  create /dev/zero (should pass)
  copy data from /dev/urandom to /dev/zero (should pass)
  copy data from /dev/urandom to /dev/full (should fail)
  copy data from /dev/random to /dev/zero (should fail)
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

The purpose of this move is to use these files in bpf tests.
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Cgroup v2 lacks the device controller, provided by cgroup v1.
This patch adds a new eBPF program type, which in combination
of previously added ability to attach multiple eBPF programs
to a cgroup, will provide a similar functionality, but with some
additional flexibility.

This patch introduces a BPF_PROG_TYPE_CGROUP_DEVICE program type.
A program takes major and minor device numbers, device type
(block/character) and access type (mknod/read/write) as parameters
and returns an integer which defines if the operation should be
allowed or terminated with -EPERM.
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is non-functional change to prepare the device cgroup code
for adding eBPF-based controller for cgroups v2.

The patch performs the following changes:
1) __devcgroup_inode_permission() and devcgroup_inode_mknod()
   are moving to the device-cgroup.h and converting into static inline.
2) __devcgroup_check_permission() is exported.
3) devcgroup_check_permission() wrapper is introduced to be used
   by both existing and new bpf-based implementations.
Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Rename device type and access type constants defined in
security/device_cgroup.c by adding the DEVCG_ prefix.

The reason behind this renaming is to make them global namespace
friendly, as they will be moved to the corresponding header file
by following patches.
Signed-off-by: Roman Gushchin <guro@fb.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Tejun Heo <tj@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

Saeed Mahameed says:

====================
mlx5-updates-2017-11-04

This series includes:

From Huy: dscp to priority mapping for Ethernet packet.

===================================================
First six patches enable differentiated services code point (dscp) to
priority mapping for Ethernet packet. Once this feature is
enabled, the packet is routed to the corresponding priority based on its
dscp. User can combine this feature with priority flow control (pfc)
feature to have priority flow control based on the dscp.

Firmware interface:
Mellanox firmware provides two control knobs for this feature:
  QPTS register allow changing the trust state between dscp and
  pcp mode. The default is pcp mode. Once in dscp mode, firmware will
  route the packet based on its dscp value if the dscp field exists.

  QPDPM register allow mapping a specific dscp (0 to 63) to a
  specific priority (0 to 7). By default, all the dscps are mapped to
  priority zero.

Software interface:
This feature is controlled via application priority TLV. IEEE
specification P802.1Qcd/D2.1 defines priority selector id 5 for
application priority TLV. This APP TLV selector defines DSCP to priority
map. This APP TLV can be sent by the switch or can be set locally using
software such as lldptool. In mlx5 drivers, we add the support for net
dcb's getapp and setapp call back. Mlx5 driver only handles the selector
id 5 application entry (dscp application priority application entry).
If user sends multiple dscp to priority APP TLV entries on the same
dscp, the last sent one will take effect. All the previous sent will be
deleted.

The firmware trust state (in QPTS register) is changed based on the
number of dscp to priority application entries. When the first dscp to
priority application entry is added by the user, the trust state is
changed to dscp. When the last dscp to priority application entry is
deleted by the user, the trust state is changed to pcp.

When the port is in DSCP trust state, the transmit queue is selected
based on the dscp of the skb.

When the port is in DSCP trust state and vport inline mode is not NONE,
firmware requires mlx5 driver to copy the IP header to the
wqe ethernet segment inline header if the skb has it.
This is done by changing the transmit queue sq's min inline mode to L3.
Note that the min inline mode of sqs that belong to other features
such as xdpsq, icosq are not modified.
===================================================

Plus to the dscp series, some small misc changes are include as well:

From Inbar, Ethtool msglvl support and some debug prints in DCBNL logic
From Or Gerlitz, Enlarge the NIC TC offload table size
From Rabie, Initialize destination_flow struct to 0
From Feras, Add inner TTC table to IPoIB flow steering
From Tal, Enable CQE based moderation on TX CQ
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Simon Horman says:

====================
nfp: ethtool and related improvements

Dirk van der Merwe says:

This patch series throws a couple of loosely related items into a single
series.

Patch 1: Clang compilation fix reported by
  Matthias Kaehlcke <mka@chromium.org>

Patch 2: Driver can now do MAC reinit on load when there has been a
  media override set in the NSP.

Patch 3: Refactor the nfp_app_reprs_set API.

Patch 4: Similar to vNICs, representors must be able to deal with media
  override changes in the NSP.

Patch 5: Since representors can now handle media overrides, we can
  allocate the get/set link ndo's to them.

Patch 6 & 7: Add support for FEC mode modification.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support in the driver ethtool ops to modify the NFP FEC modes.

The FEC modes can be set for vNIC associated with physical ports or
for MAC representor netdevs.
Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Implement helpers to determine and modify FEC modes via the NSP.
The NSP advertises FEC capabilities on a per port basis and provides
support for:
* Auto mode selection
* Reed Solomon
* BaseR
* None/Off
Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since it is now safe to modify link settings for representors, we can
attach the get/set link settings ndos to it. The get/set link settings
are nfp_port based operations.

If a port becomes invalid, the representor will be removed in the same
way a vnic would be.
Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If the NSP port table has been refreshed, resync the representor state
with the new port information. At the moment, this only entails looking
for invalid ports and killing off representors associated with them.

The repr instance becomes NULL which is safe since the app accessor
function for reprs returns NULL when it cannot access a repr.
Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The criteria that reprs cannot be replaced with another new set of reprs
has been removed. This check is not needed since the only use case that
could exercise this at the moment, would be to modify the number of
SRIOV VFs without first disabling them. This case is explicitly
disallowed in any case and subsequent patches in this series
need to be able to replace the running set of reprs.

All cases where the return code used to be checked for the
nfp_app_reprs_set function have been removed.
As stated above, it is not possible for the current code to encounter a
case where reprs exist and need to be replaced.
Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Recent management FW images can perform full reinit of MAC cores
without requiring a reboot.  When loading the driver check if there
are changes pending and if so call NSP MAC reinit.  Full application
FW reload is still required, and all MACs need to be reinited at the
same time (not only the ones which have been reconfigured, and thus
potentially causing disruption to unrelated netdevs) therefore for
now changing MAC config without reloading the driver still remains
future work.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Tested-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Matthias reports:

  nfp_eth_set_bit_config() is marked as __always_inline to allow gcc to
  identify the 'mask' parameter as known to be constant at compile time,
  which is required to use the FIELD_GET() macro.

  The forced inlining does the trick for gcc, but for kernel builds with
  clang it results in undefined symbols:

  drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.o: In function
    `__nfp_eth_set_aneg':

drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c:(.text+0x787):
    undefined reference to `__compiletime_assert_492'

drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp_eth.c:(.text+0x7b1):
    undefined reference to `__compiletime_assert_496'

  These __compiletime_assert_xyx() calls would have been optimized away
if
  the compiler had seen 'mask' as a constant.

Add a macro to extract the mask and shift and pass those to
nfp_eth_set_bit_config() separately.
Reported-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Tested-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently TCP RACK loss detection does not work well if packets are
being reordered beyond its static reordering window (min_rtt/4).Under
such reordering it may falsely trigger loss recoveries and reduce TCP
throughput significantly.

This patch improves that by increasing and reducing the reordering
window based on DSACK, which is now supported in major TCP implementations.
It makes RACK's reo_wnd adaptive based on DSACK and no. of recoveries.

- If DSACK is received, increment reo_wnd by min_rtt/4 (upper bounded
  by srtt), since there is possibility that spurious retransmission was
  due to reordering delay longer than reo_wnd.

- Persist the current reo_wnd value for TCP_RACK_RECOVERY_THRESH (16)
  no. of successful recoveries (accounts for full DSACK-based loss
  recovery undo). After that, reset it to default (min_rtt/4).

- At max, reo_wnd is incremented only once per rtt. So that the new
  DSACK on which we are reacting, is due to the spurious retx (approx)
  after the reo_wnd has been updated last time.

- reo_wnd is tracked in terms of steps (of min_rtt/4), rather than
  absolute value to account for change in rtt.

In our internal testing, we observed significant increase in throughput,
in scenarios where reordering exceeds min_rtt/4 (previous static value).
Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Vivien Didelot says:

====================
net: dsa: parsing stage

When registering a DSA switch, there is basically two stages.

The first stage is the parsing of the switch device, from either device
tree or platform data. It fetches the DSA tree to which it belongs, and
validates its ports. The switch device is then added to the tree, and
the second stage is called if this was the last switch of the tree.

The second stage is the setup of the tree, which validates that the tree
is complete, sets up the routing tables, the default CPU port for user
ports, sets up the switch drivers and finally the master interfaces,
which makes the whole switch fabric functional.

This patch series covers the first parsing stage. It fixes the type of
the switch and tree indexes to unsigned int, simplifies the tree
reference counting and the switch and CPU ports parsing.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Extend the dsa_port_parse_cpu() function to resolve the tagging protocol
at port parsing time, instead of waiting for the whole tree to be
complete.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>