Merge "net: aquantia: support for 32-bit ARM target"

	---help---

	  This enables the support for the aQuantia AQtion(tm) Ethernet card.

config AQFWD

	tristate "aQuantia Forwarding driver"

	depends on PCI && (X86_64 || ARM64 || ARM )

	---help---

	  This enables the support for forwarding driver for the aQuantia AQtion(tm) Ethernet card.

source "drivers/net/ethernet/aquantia/atlantic-fwd/Kconfig"

endif # NET_VENDOR_AQUANTIA

#

obj-$(CONFIG_AQTION) += atlantic/

obj-$(CONFIG_AQFWD)  += atlantic-fwd/

Supported hardware

------------------

AQC-107 / -108 / -109, revisions B0 / B1. FW versions 1.x / 2.x / 3.x

supported

Features / acceleration

-----------------------

 - Message-signalled interrupts

 - Hardware VLAN tag offloads

 - Hardware checksum offloads:

    - IPv4 header checksums

    - UDP / TCP checksums with IPv4 and IPv6

 - Hardware LSO for TCP v4/v6

 - Hardware LRO for TCP v4/v6

   - Coalescing TCP segments with TCP options not supported due to a

     HW limitation

 - Software GRO using the Linux NAPI GRO interface

 - Hardware Rx header splitting

 - RSS over up to 8 queues using MSI-X

 - Linear skb receive mode

 - N-tuple Rx filters

 - Runtime power management

 - Hardware MACSEC offload

Building and installation

-------------------------

1. Unpack the archive into a directory of your choice.

2. Change to that directory and compile the module:

   	$  make [KDIR=<path to kernel headers>]

   Explicitly specifying KDIR is only necessary when the headers can't

   be found automatically via the

   /lib/modules/`uname -r`/{build,source} symlinks

3. Load the module

   	# insmod atlnew.ko [<module parameters>]

4. Unload the module

   	# rmmod atlnew

Device configuration

--------------------

 # ip link set up dev <ifname>

 # ip addr add <IP addr>/<CIDR prefix length> brd + dev	<ifname>

Add a VLAN ID:

 # ip link add link <ifname> name <ifname>.<VLAN id> type vlan id <VLAN id>

 # ip link set up dev <ifname>.<VLAN id>

Remove a VLAN ID:

 # ip link del dev <ifname>.<VLAN id>

Module parameters

-----------------

Module parameters may be specified at module load time on the insmod

command line or via modprobe.conf files. Some parameters marked as

[RW] below can also be changed at run time via files in

/sys/module/atlnew/parameters directory

 - msi = < 1 | 0 > (default: 1) [RO]

       	Enable use of MSI interrupts.

 - rx_refill_batch = <num> (default: 16) [RW]

   	Number of cleaned Rx ring descriptors to trigger Rx refill

 - tx_clean_budget = <num> (default: 256) [RW]

   	Max number of processed Tx descriptors to clean in one pass of

   	NAPI poll

 - max_queues = <num> (default: 8) [RO]

	Max number of RSS queues. The actual number of queues will the

	minimum of this parameter and the number of logical CPUs in

	the system.

 - rx_linear = <0 | 1> (default: 0)

	Enables linear skb receive mode. Frames larger than

	ATL_RX_BUF_SIZE defined in atl_ring.h will be dropped.

 - min_intr_delay (default: 10) [RW]

	When an interrupt is about to be asserted on an LRO-enabled

	queue, LRO requires some lead time to close and flush

	in-progress flows. This parameter sets the LRO interrupt delay

	and provides the lower boundary for interrupt coalescing

	delays settable via ethtool. When modified, the new value

	takes effect on interface down/up cycle or on setting an

	interrupt coalescing delay via ethtool.

 - rx_mod_hyst (default: 10) [RW]

 - tx_mod_hyst (default: 10) [RW]

	Each ring has two interrupt coalescing timers. When an

	interrupt-causing event happens, they're initialized with

	values called minimum and maximum respectively. The former

	timer is restarted each time another such event happens, while

	the latter keeps running. When either timer reaches zero, the

	interrupt is reported to the host. The minimum value is set

	via ethtool. These module parameters set the difference

	between the maximum and minimum values for Rx and Tx

	paths. When modified, the new max values go into effect upon

	interface down/up cycle or upon modifying ethtool interrupt

	coalescing settings.

- keep_link = < 1 | 0 > (default: 0) [RO]

	When set to 1, the ethernet link is established at PCI probe

	time, and is not dropped on ifdown. This allows the rings

	allocated to forwarding engines to handle traffic even when

	linux network interface is down.

- sleep_delay = (default: 10000) [RW]

  Delay in ms after link goes down to suspend the device if runtime power

  management is turned on.

- macsec_bridge = < 1 | 0 > (default: 1) [RO]

  When set to 0, several SecY can be created, macsec will filter by source mac

  address to apply apropriate SecY.

  By default, only 1 SecY can be created, macsec will handle all packets.

- rx/tx_ring_size = <8..8184> (default: 4096) [RO]

  Default size for the rings.

VLAN filters

------------

Sixteen VLAN filters are programmable via the ethtool -N interface using

locations 0 through 15. The only flow-type supported is ether and only

VLAN id can be matched.

For VLAN filters only, it's possible not to specify an explicit

filter location. In this case an appropriate location will be assigned

automatically.

These filters are also used for HW filter acceleration for Linux VLAN

subdevices. Filters automatically added by Linux VLAN code can be

overridden using ethtool subject to restrictions described below:

* A special ring number 32 is used in ethtool interface to signify

  'any ring'. An ethtool filter rule with ring set to 32 accepts a

  packet leaving the ring to be determined using the standard RSS

  mechanism. VLAN code adds rules in this manner. Such a rule can be

  modified using ethool to set a specific destination ring.

* There can't be two filters with the same VID. An attempt to add a

  duplicate filter using an explicit location different from the

  location of existing filter with the same VID will fail. When the

  location is not specified, the existing filter will be modified.

* An attempt to change the VID of a filter added by VLAN code using an

  explicit location will fail. Either use a different explicit

  location or no location.

* An attempt to delete a rule created by VLAN code will return

  success, but the rule will remain in place. If a specific ring was

  set in the rule, it will be reset to 'any ring'.

* If a VLAN sub-device is added when there're no unused filters, VLAN

  promiscuous mode will be enabled. This will result in all VIDs being

  accepted. Filters can still be used to direct specific VIDs to

  specific rings.

* Even when all 16 filters are used, attempting to add a rule with a

  new VID via ethtool will succeed as long as at least one existing

  filter was set by VLAN code and still has 'any ring' as

  destination. This filter will be re-used for the new rule and VLAN

  promiscuous mode will be enabled.

* When a filter becomes available as a result of a deletion of a VLAN

  subdevice or a deletion of an ethtool filter rule, and a VLAN

  promiscuous mode is enabled, the vacated filter will be re-used for

  another enabled VLAN subdevice. This may result in VLAN promiscuous

  mode being turned off if that VLAN subdevice was the last one with

  no filter allocated.

Ethertype filters

-----------------

Fifteen ethertype filters are programmable via the ethtool -N

interface using locations 16 through 30. The only flow-type supported

is ether and only the ethertype field can be matched.

N-tuple filters

---------------

Eight IPv4 n-tuple filters are supported, programmable using ethtool

-N command with filter locations 32 through 39.

Fields that can be matched are source / destination IP addresses, and,

for TCP / UDP / SCTP protocols, additionally source / destination

ports can be matched. Partial (masked) matches are not supported, a

mask for each field must consist of either all ones or all zeroes.

Each aligned group of four filters (locations 32 through 35 or 36

through 39) can alternatively be used as one IPv6 filter, using filter

locations 32 or 36.

Testing forwarding performance

------------------------------

Since this is a design for a router, the best way to test would be

between our aquantia port and another high speed port on the same

DUT. However, in the situation where the Aquantia interface is the

only high speed one, we have devised the following testing strategy

to emulate these two ports on same DUT.

One of the simplest ways is to set up two VLAN interfaces and forward

traffic between them as in the following example:

Set up two VLANs with examlple ids 42 and 43 on the Aquantia

interface, assign the switch port connected to it to both of those

VLANs.

# ip link set up dev <iface>

# ip link add link <iface> name <iface>.42 type vlan id 42

# ip link add link <iface> name <iface>.43 type vlan id 43

# ip link set up dev <iface>.42

# ip link set up dev <iface>.43

# ip addr add 10.42.42.1/24 brd + dev <iface>.42

# ip addr add 10.42.43.1/24 brd + dev <iface>.43

# sysctl net.ipv4.ip_forward=1

This sets up two VLAN interfaces with VLAN ids 42 and 43 and enables

IP routing. Two hosts should be set up to generate and terminate the

traffic:

On a host on the 42 VLAN with IP 10.42.42.xx run

 $ iperf3 -s

On a host on the 43 VLAN run

 $ iperf3 -c 10.42.42.xx

Another possibility is to use a single host running Linux as a link

partner both generating and terminating the traffic. Two VLAN

sub-interfaces should be configured on a high-speed port on that host,

belonging to VLAN 42 / IP network 10.42.42/24 and VLAN 43 / IP network

10.42.43/24 respectively.

Since, by default, traffic addressed to one of the local IPs is

received locally, a special configuration of Linux policy routing is

required to make such traffic actually go out on the wire to the DUT,

as described below, where <lp_iface> stands for the chosen network interface:

# ip rule add table local prio 100

# ip rule del table local prio 0

# ip rule add iif lo table 42 prio 42

# ip link set up dev <lp_iface>

# ip link add link <lp_iface> name <lp_iface>.42 type vlan id 42

# ip link add link <lp_iface> name <lp_iface>.43 type vlan id 43

# ip link set up dev <lp_iface>.42

# ip link set up dev <lp_iface>.43

# ip addr add 10.42.42.10/24 brd + dev <lp_iface>.42

# ip addr add 10.42.43.10/24 brd + dev <lp_iface>.43

# ip route add 10.42.42.10 via 10.42.43.1 table 42

# ip route add 10.42.43.10 via 10.42.42.1 table 42

This routes the locally-originated traffic, that is addressed to the local IP

of one of the VLAN interfaces, out via the other VLAN interface, instead of

being immediately looped back locally.

$ iperf3 -s

$ iperf3 -c 10.42.42.10

The following command can be used to restore the default routing

behavior:

# ip route flush table 42

#

# Aquantia atlantic-forwarding driver config

#

if AQFWD

config ATLFWD_FWD

       bool "Enable forwarding-engine API"

       default n

       help

         Say Y to enable the forwarding-engine API

config ATLFWD_FWD_RXBUF

       int

       prompt "Rx buffer space for forwarding engine's rings" if ATLFWD_FWD

       range 0 320

       default 160 if ATLFWD_FWD

       default 0

       help

         Amount of Rx buffer to reserve for the forwarding-engine

         rings. This sets the default value of the fwd_rx_buf_reserve

         module option.

         Value in kiB, 0 to 320, defaults to 160 if forwarding-engine

         API enabled, 0 otherwise.

config ATLFWD_FWD_TXBUF

       int

       prompt "Tx buffer space for forwarding engine's rings" if ATLFWD_FWD

       range 0 160

       default 80 if ATLFWD_FWD

       default 0

       help

         Amount of Tx buffer to reserve for the forwarding-engine

         rings. This sets the default value of the fwd_tx_buf_reserve

         module option.

         Value in kiB, 0 to 320, defaults to 80 if forwarding-engine

         API enabled, 0 otherwise.

config ATLFWD_FWD_NETLINK

        bool "Enable netlink control socket for forwarding engine"

        default n

        depends on ATLFWD_FWD

        help

          Expose forwarding engine APIs over the netlink socket.

endif

# SPDX-License-Identifier: GPL-2.0-only

# Atlantic Network Driver

#

# Copyright(c) 2014-2019 aQuantia Corporation.

# Copyright(c) 2019-2020 Marvell International Ltd.

#

# This program is free software; you can redistribute it and/or modify

# it under the terms of the GNU General Public License version 2 as

# published by the Free Software Foundation.

obj-$(CONFIG_AQFWD) += atlantic-fwd.o

atlantic-fwd-objs := atl_fw.o \

		     atl2_fw.o \

		     atl_hw.o \

		     atl_main.o \

		     atl_ring.o \

		     atl_ethtool.o \

		     atl_trace.o \

		     atl_compat.o \

		     atl_hwmon.o \

		     atl_ptp.o \

		     atl_hw_ptp.o

atlantic-fwd-$(CONFIG_ATLFWD_FWD) += atl_fwd.o

atlantic-fwd-$(CONFIG_ATLFWD_FWD_NETLINK) += atl_fwdnl.o \

					      atl_fwdnl_params.o

atlantic-fwd-$(CONFIG_MACSEC) += atl_macsec.o macsec/macsec_api.o

ccflags-y += -I$(src)