net: dsa: make assisted_learning_on_cpu_port bypass offloaded LAG interfaces

Given the following topology, and focusing only on Box A:

         Box A
         +----------------------------------+
         | Board 1         br0              |
         |             +---------+          |
         |            /           \         |
         |            |           |         |
         |            |         bond0       |
         |            |        +-----+      |
         |192.168.1.1 |       /       \     |
         |  eno0     swp0    swp1    swp2   |
         +---|--------|-------|-------|-----+
             |        |       |       |
             +--------+       |       |
               Cable          |       |
                         Cable|       |Cable
               Cable          |       |
             +--------+       |       |
             |        |       |       |
         +---|--------|-------|-------|-----+
         |  eno0     swp0    swp1    swp2   |
         |192.168.1.2 |       \       /     |
         |            |        +-----+      |
         |            |         bond0       |
         |            |           |         |
         |            \           /         |
         |             +---------+          |
         | Board 2         br0              |
         +----------------------------------+
         Box B

The assisted_learning_on_cpu_port logic will see that swp0 is bridged
with a "foreign interface" (bond0) and will therefore install all
addresses learnt by the software bridge towards bond0 (including the
address of eno0 on Box B) as static addresses towards the CPU port.

But that's not what we want - bond0 is not really a "foreign interface"
but one we can offload including L2 forwarding from/towards it. So we
need to refine our logic for assisted learning such that, whenever we
see an address learnt on a non-DSA interface, we search through the tree
for any port that offloads that non-DSA interface.

Some confusion might arise as to why we search through the whole tree
instead of just the local switch returned by dsa_slave_dev_lower_find.
Or a different angle of the same confusion: why does
dsa_slave_dev_lower_find(br_dev) return a single dp that's under br_dev
instead of the whole list of bridged DSA ports?

To answer the second question, it should be enough to install the static
FDB entry on the CPU port of a single switch in the tree, because
dsa_port_fdb_add uses DSA_NOTIFIER_FDB_ADD which ensures that all other
switches in the tree get notified of that address, and add the entry
themselves using dsa_towards_port().

This should help understand the answer to the first question: the port
returned by dsa_slave_dev_lower_find may not be on the same switch as
the ports that offload the LAG. Nonetheless, if the driver implements
.crosschip_lag_join and .crosschip_bridge_join as mv88e6xxx does, there
still isn't any reason for trapping addresses learnt on the remote LAG
towards the CPU, and we should prevent that.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: rebalance LAGs on link up/down events

At present there is an issue when ocelot is offloading a bonding
interface, but one of the links of the physical ports goes down. Traffic
keeps being hashed towards that destination, and of course gets dropped
on egress.

Monitor the netdev notifier events emitted by the bonding driver for
changes in the physical state of lower interfaces, to determine which
ports are active and which ones are no longer.

Then extend ocelot_get_bond_mask to return either the configured bonding
interfaces, or the active ones, depending on a boolean argument. The
code that does rebalancing only needs to do so among the active ports,
whereas the bridge forwarding mask and the logical port IDs still need
to look at the permanently bonded ports.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: rename aggr_count to num_ports_in_lag

It makes it a bit easier to read and understand the code that deals with
balancing the 16 aggregation codes among the ports in a certain LAG.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: drop the use of the "lags" array

We can now simplify the implementation by always using ocelot_get_bond_mask
to look up the other ports that are offloading the same bonding interface
as us.

In ocelot_set_aggr_pgids, the code had a way to uniquely iterate through
LAGs. We need to achieve the same behavior by marking each LAG as visited,
which we do now by using a temporary 32-bit "visited" bitmask. This is
ok and we do not need dynamic memory allocation, because we know that
this switch architecture will not have more than 32 ports (the PGID port
masks are 32-bit anyway).

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: set up logical port IDs centrally

The setup of logical port IDs is done in two places: from the inconclusively
named ocelot_setup_lag and from ocelot_port_lag_leave, a function that
also calls ocelot_setup_lag (which apparently does an incomplete setup
of the LAG).

To improve this situation, we can rename ocelot_setup_lag into
ocelot_setup_logical_port_ids, and drop the "lag" argument. It will now
set up the logical port IDs of all switch ports, which may be just
slightly more inefficient but more maintainable.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: avoid unneeded "lp" variable in LAG join

The index of the LAG is equal to the logical port ID that all the
physical port members have, which is further equal to the index of the
first physical port that is a member of the LAG.

The code gets a bit carried away with logic like this:

if (a == b)
c = a;
else
c = b;

which can be simplified, of course, into:

c = b;

(with a being port, b being lp, c being lag)

This further makes the "lp" variable redundant, since we can use "lag"
everywhere where "lp" (logical port) was used. So instead of a "c = b"
assignment, we can do a complete deletion of b. Only one comment here:

if (bond_mask) {
lp = __ffs(bond_mask);
ocelot->lags[lp] = 0;
}

lp was clobbered before, because it was used as a temporary variable to
hold the new smallest port ID from the bond. Now that we don't have "lp"
any longer, we'll just avoid the temporary variable and zeroize the
bonding mask directly.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: set up the bonding mask in a way that avoids a net_device

Since this code should be called from pure switchdev as well as from
DSA, we must find a way to determine the bonding mask not by looking
directly at the net_device lowers of the bonding interface, since those
could have different private structures.

We keep a pointer to the bonding upper interface, if present, in struct
ocelot_port. Then the bonding mask becomes the bitwise OR of all ports
that have the same bonding upper interface. This adds a duplication of
functionality with the current "lags" array, but the duplication will be
short-lived, since further patches will remove the latter completely.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: use ipv6 in the aggregation code

IPv6 header information is not currently part of the entropy source for
the 4-bit aggregation code used for LAG offload, even though it could be.
The hardware reference manual says about these fields:

ANA::AGGR_CFG.AC_IP6_TCPUDP_PORT_ENA
Use IPv6 TCP/UDP port when calculating aggregation code. Configure
identically for all ports. Recommended value is 1.

ANA::AGGR_CFG.AC_IP6_FLOW_LBL_ENA
Use IPv6 flow label when calculating AC. Configure identically for all
ports. Recommended value is 1.

Integration with the xmit_hash_policy of the bonding interface is TBD.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: don't refuse bonding interfaces we can't offload

Since switchdev/DSA exposes network interfaces that fulfill many of the
same user space expectations that dedicated NICs do, it makes sense to
not deny bonding interfaces with a bonding policy that we cannot offload,
but instead allow the bonding driver to select the egress interface in
software.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: use a switch-case statement in ocelot_netdevice_event

Make ocelot's net device event handler more streamlined by structuring
it in a similar way with others. The inspiration here was
dsa_slave_netdevice_event.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mscc: ocelot: rename ocelot_netdevice_port_event to ocelot_netdevice_changeupper

ocelot_netdevice_port_event treats a single event, NETDEV_CHANGEUPPER.
So we can remove the check for the type of event, and rename the
function to be more suggestive, since there already is a function with a
very similar name of ocelot_netdevice_event.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'automatically-manage-dsa-master-interface-state'

Vladimir Oltean says:

====================
Automatically manage DSA master interface state

This patch series adds code that makes DSA open the master interface
automatically whenever one user interface gets opened, either by the
user, or by various networking subsystems: netconsole, nfsroot.
With that in place, we can remove some of the places in the network
stack where DSA-specific code was sprinkled.
====================

Link: https://lore.kernel.org/r/20210205133713.4172846-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Revert "net: ipv4: handle DSA enabled master network devices"

This reverts commit 728c02089a0e3eefb02e9927bfae50490f40e72e.

Since 2015 DSA has gained more integration with the network stack, we
can now have the same functionality without explicitly open-coding for
it:
- It now opens the DSA master netdevice automatically whenever a user
  netdevice is opened.
- The master and switch interfaces are coupled in an upper/lower
  hierarchy using the netdev adjacency lists.

In the nfsroot example below, the interface chosen by autoconfig was
swp3, and every interface except that and the DSA master, eth1, was
brought down afterwards:

[    8.714215] mscc_felix 0000:00:00.5 swp0 (uninitialized): PHY [0000:00:00.3:10] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[    8.978041] mscc_felix 0000:00:00.5 swp1 (uninitialized): PHY [0000:00:00.3:11] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[    9.246134] mscc_felix 0000:00:00.5 swp2 (uninitialized): PHY [0000:00:00.3:12] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[    9.486203] mscc_felix 0000:00:00.5 swp3 (uninitialized): PHY [0000:00:00.3:13] driver [Microsemi GE VSC8514 SyncE] (irq=POLL)
[    9.512827] mscc_felix 0000:00:00.5: configuring for fixed/internal link mode
[    9.521047] mscc_felix 0000:00:00.5: Link is Up - 2.5Gbps/Full - flow control off
[    9.530382] device eth1 entered promiscuous mode
[    9.535452] DSA: tree 0 setup
[    9.539777] printk: console [netcon0] enabled
[    9.544504] netconsole: network logging started
[    9.555047] fsl_enetc 0000:00:00.2 eth1: configuring for fixed/internal link mode
[    9.562790] fsl_enetc 0000:00:00.2 eth1: Link is Up - 1Gbps/Full - flow control off
[    9.564661] 8021q: adding VLAN 0 to HW filter on device bond0
[    9.637681] fsl_enetc 0000:00:00.0 eth0: PHY [0000:00:00.0:02] driver [Qualcomm Atheros AR8031/AR8033] (irq=POLL)
[    9.655679] fsl_enetc 0000:00:00.0 eth0: configuring for inband/sgmii link mode
[    9.666611] mscc_felix 0000:00:00.5 swp0: configuring for inband/qsgmii link mode
[    9.676216] 8021q: adding VLAN 0 to HW filter on device swp0
[    9.682086] mscc_felix 0000:00:00.5 swp1: configuring for inband/qsgmii link mode
[    9.690700] 8021q: adding VLAN 0 to HW filter on device swp1
[    9.696538] mscc_felix 0000:00:00.5 swp2: configuring for inband/qsgmii link mode
[    9.705131] 8021q: adding VLAN 0 to HW filter on device swp2
[    9.710964] mscc_felix 0000:00:00.5 swp3: configuring for inband/qsgmii link mode
[    9.719548] 8021q: adding VLAN 0 to HW filter on device swp3
[    9.747811] Sending DHCP requests ..
[   12.742899] mscc_felix 0000:00:00.5 swp1: Link is Up - 1Gbps/Full - flow control rx/tx
[   12.743828] mscc_felix 0000:00:00.5 swp0: Link is Up - 1Gbps/Full - flow control off
[   12.747062] IPv6: ADDRCONF(NETDEV_CHANGE): swp1: link becomes ready
[   12.755216] fsl_enetc 0000:00:00.0 eth0: Link is Up - 1Gbps/Full - flow control rx/tx
[   12.766603] IPv6: ADDRCONF(NETDEV_CHANGE): swp0: link becomes ready
[   12.783188] mscc_felix 0000:00:00.5 swp2: Link is Up - 1Gbps/Full - flow control rx/tx
[   12.785354] IPv6: ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
[   12.799535] IPv6: ADDRCONF(NETDEV_CHANGE): swp2: link becomes ready
[   13.803141] mscc_felix 0000:00:00.5 swp3: Link is Up - 1Gbps/Full - flow control rx/tx
[   13.811646] IPv6: ADDRCONF(NETDEV_CHANGE): swp3: link becomes ready
[   15.452018] ., OK
[   15.470336] IP-Config: Got DHCP answer from 10.0.0.1, my address is 10.0.0.39
[   15.477887] IP-Config: Complete:
[   15.481330]      device=swp3, hwaddr=00:04:9f:05:de:0a, ipaddr=10.0.0.39, mask=255.255.255.0, gw=10.0.0.1
[   15.491846]      host=10.0.0.39, domain=(none), nis-domain=(none)
[   15.498429]      bootserver=10.0.0.1, rootserver=10.0.0.1, rootpath=
[   15.498481]      nameserver0=8.8.8.8
[   15.627542] fsl_enetc 0000:00:00.0 eth0: Link is Down
[   15.690903] mscc_felix 0000:00:00.5 swp0: Link is Down
[   15.745216] mscc_felix 0000:00:00.5 swp1: Link is Down
[   15.800498] mscc_felix 0000:00:00.5 swp2: Link is Down

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Revert "net: Have netpoll bring-up DSA management interface"

This reverts commit 1532b9778478577152201adbafa7738b1e844868.

The above commit is good and it works, however it was meant as a bugfix
for stable kernels and now we have more self-contained ways in DSA to
handle the situation where the DSA master must be brought up.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: automatically bring user ports down when master goes down

This is not fixing any actual bug that I know of, but having a DSA
interface that is up even when its lower (master) interface is down is
one of those things that just do not sound right.

Yes, DSA checks if the master is up before actually bringing the
user interface up, but nobody prevents bringing the master interface
down immediately afterwards... Then the user ports would attempt
dev_queue_xmit on an interface that is down, and wonder what's wrong.

This patch prevents that from happening. NETDEV_GOING_DOWN is the
notification emitted _before_ the master actually goes down, and we are
protected by the rtnl_mutex, so all is well.

For those of you reading this because you were doing switch testing
such as latency measurements for autonomously forwarded traffic, and you
needed a controlled environment with no extra packets sent by the
network stack, this patch breaks that, because now the user ports go
down too, which may shut down the PHY etc. But please don't do it like
that, just do instead:

tc qdisc add dev eno2 clsact
tc filter add dev eno2 egress flower action drop

Tested with two cascaded DSA switches:
$ ip link set eno2 down
sja1105 spi2.0 sw0p2: Link is Down
mscc_felix 0000:00:00.5 swp0: Link is Down
fsl_enetc 0000:00:00.2 eno2: Link is Down

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: automatically bring up DSA master when opening user port

DSA wants the master interface to be open before the user port is due to
historical reasons. The promiscuity of interfaces that are down used to
have issues, as referenced Lennert Buytenhek in commit df02c6ff2e39
("dsa: fix master interface allmulti/promisc handling").

The bugfix mentioned there, commit b6c40d68ff64 ("net: only invoke
dev->change_rx_flags when device is UP"), was basically a "don't do
that" approach to working around the promiscuity while down issue.

Further work done by Vlad Yasevich in commit d2615bf45069 ("net: core:
Always propagate flag changes to interfaces") has resolved the
underlying issue, and it is strictly up to the DSA and 8021q drivers
now, it is no longer mandated by the networking core that the master
interface must be up when changing its promiscuity.

From DSA's point of view, deciding to error out in dsa_slave_open
because the master isn't up is
(a) a bad user experience and
(b) knocking at an open door.
Even if there still was an issue with promiscuity while down, DSA could
still just open the master and avoid it.

Doing it this way has the additional benefit that user space can now
remove DSA-specific workarounds, like systemd-networkd with BindCarrier:
https://github.com/systemd/systemd/issues/7478

And we can finally remove one of the 2 bullets in the "Common pitfalls
using DSA setups" chapter.

Tested with two cascaded DSA switches:

$ ip link set sw0p2 up
fsl_enetc 0000:00:00.2 eno2: configuring for fixed/internal link mode
fsl_enetc 0000:00:00.2 eno2: Link is Up - 1Gbps/Full - flow control rx/tx
mscc_felix 0000:00:00.5 swp0: configuring for fixed/sgmii link mode
mscc_felix 0000:00:00.5 swp0: Link is Up - 1Gbps/Full - flow control off
8021q: adding VLAN 0 to HW filter on device swp0
sja1105 spi2.0 sw0p2: configuring for phy/rgmii-id link mode
IPv6: ADDRCONF(NETDEV_CHANGE): eno2: link becomes ready
IPv6: ADDRCONF(NETDEV_CHANGE): swp0: link becomes ready

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-hns3-updates-for-next'

Huazhong Tan says:

====================
net: hns3: updates for -next

This series adds some code optimizations and compatibility
handlings for the HNS3 ethernet driver.

change log:
V2: refactor #2 as Jukub Kicinski reported and remove the part
    about RSS size which will not be different in different hw.
    updates netdev->max_mtu as well in #4 reported by Jakub Kicinski.

previous version:
V1: https://patchwork.kernel.org/project/netdevbpf/cover/1612269593-18691-1-git-send-email-tanhuazhong@huawei.com/
====================

Link: https://lore.kernel.org/r/1612513969-9278-1-git-send-email-tanhuazhong@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns3: replace macro of max qset number with specification

The max qset number is a fixed value now and it is defined by a macro.
In order to support other value in different kinds of device, it is
better to use specification queried from firmware to replace macro.

Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns3: debugfs add max tm rate specification print

In order to add a method to check the specification of max tm rate
for debugging, function hns3_dbg_dev_specs() adds this value print.

Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns3: add support for obtaining the maximum frame size

Since the newer hardware may supports different frame size,
so add support to obtain the capability from the firmware
instead of the fixed value.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns3: optimize the code when update the tc info

When update the TC info for NIC, there are some differences
between PF and VF. Currently, four "vport->vport_id" are
used to distinguish PF or VF. So merge them into one to
improve readability and maintainability of code.

Signed-off-by: GuoJia Liao <liaoguojia@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns3: RSS indirection table use device specification

As RSS indirection table size may be different in different
hardware. Instead of using macro, this value is better to use
device specification which querying from firmware.

BTW, RSS indirection table should be allocated by the queried
size instead the static array.

.get_rss_indir_size in struct hnae3_ae_ops is not used now,
so remove it as well.

Signed-off-by: Guangbin Huang <huangguangbin2@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns3: add api capability bits for firmware

To improve the compatibility of firmware for driver, help firmware
to deal with different api commands, add api capability bits when
initialize the command queue.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'mptcp-misc-updates-for-tests-lock-annotation'

Mat Martineau says:

====================
mptcp: Misc. updates for tests & lock annotation

Here are two fixes we've collected in the mptcp tree.

Patch 1 refactors a MPTCP selftest script to allow running a subset of
the tests.

Patch 2 adds some locking & might_sleep assertations.
====================

Link: https://lore.kernel.org/r/20210204232330.202441-1-mathew.j.martineau@linux.intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mptcp: pm: add lockdep assertions

Add a few assertions to make sure functions are called with the needed
locks held.
Two functions gain might_sleep annotations because they contain
conditional calls to functions that sleep.

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: mptcp: add command line arguments for mptcp_join.sh

Since the mptcp_join script is becoming too big, this patch splits it
into several smaller chunks, each of them has been defined in a function
as a individual test group for several related testcases.

Using bash getopts function to parse command line arguments, and invoke
each function to do the individual test group.

Here are all the arguments:
  -f subflows_tests
  -s signal_address_tests
  -l link_failure_tests
  -t add_addr_timeout_tests
  -r remove_tests
  -a add_tests
  -6 ipv6_tests
  -4 v4mapped_tests
  -b backup_tests
  -p add_addr_ports_tests
  -c syncookies_tests
  -h help

Run mptcp_join.sh with no argument will execute all testcases.

Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'dpaa2-add-1000base-x-support'

Russell King says:

====================
dpaa2: add 1000base-X support

This patch series adds 1000base-X support to pcs-lynx and DPAA2,
allowing runtime switching between SGMII and 1000base-X. This is
a pre-requisit for SFP module support on the SolidRun ComExpress 7.

v2: updated with Ioana's r-b's, and comment on backplane support
====================

Link: https://lore.kernel.org/r/20210205103859.GH1463@shell.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dpaa2-mac: add backplane link mode support

Add support for backplane link mode, which is, according to discussions
with NXP earlier in the year, is a mode where the OS (Linux) is able to
manage the PCS and Serdes itself.

This commit prepares the ground work for allowing 1G fiber connections
to be used with DPAA2 on the SolidRun CEX7 platforms.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dpaa2-mac: add 1000BASE-X support

Now that pcs-lynx supports 1000BASE-X, add support for this interface
mode to dpaa2-mac. pcs-lynx can be switched at runtime between SGMII
and 1000BASE-X mode, so allow dpaa2-mac to switch between these as
well.

This commit prepares the ground work for allowing 1G fiber connections
to be used with DPAA2 on the SolidRun CEX7 platforms.

Reviewed-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: pcs: add pcs-lynx 1000BASE-X support

Add support for 1000BASE-X to pcs-lynx for the LX2160A.

This commit prepares the ground work for allowing 1G fiber connections
to be used with DPAA2 on the SolidRun CEX7 platforms.

Reviewed-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: wan: farsync: use new tasklet API

This converts the driver to use the new tasklet API introduced in
commit 12cc923f1ccc ("tasklet: Introduce new initialization API")

The new API changes the argument passed to callback functions,
but fortunately it is unused so it is straight forward to use
DECLARE_TASKLET rather than DECLARE_TASLKLET_OLD.

Signed-off-by: Emil Renner Berthing <kernel@esmil.dk>
Link: https://lore.kernel.org/r/20210204173947.92884-1-kernel@esmil.dk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-avoid-the-memory-waste-in-some-ethernet-drivers'

Kevin Hao says:

====================
net: Avoid the memory waste in some Ethernet drivers

In the current implementation of napi_alloc_frag(), it doesn't have any
align guarantee for the returned buffer address. We would have to use
some ugly workarounds to make sure that we can get a align buffer
address for some Ethernet drivers. This patch series tries to introduce
some helper functions to make sure that an align buffer is returned.
Then we can drop the ugly workarounds and avoid the unnecessary memory
waste.
====================

Link: https://lore.kernel.org/r/20210204105638.1584-1-haokexin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dpaa2: Use napi_alloc_frag_align() to avoid the memory waste

The napi_alloc_frag_align() will guarantee that a correctly align
buffer address is returned. So use this function to simplify the buffer
alloc and avoid the unnecessary memory waste.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Reviewed-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: octeontx2: Use napi_alloc_frag_align() to avoid the memory waste

The napi_alloc_frag_align() will guarantee that a correctly align
buffer address is returned. So use this function to simplify the buffer
alloc and avoid the unnecessary memory waste.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Tested-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: Introduce {netdev,napi}_alloc_frag_align()

In the current implementation of {netdev,napi}_alloc_frag(), it doesn't
have any align guarantee for the returned buffer address, But for some
hardwares they do require the DMA buffer to be aligned correctly,
so we would have to use some workarounds like below if the buffers
allocated by the {netdev,napi}_alloc_frag() are used by these hardwares
for DMA.
    buf = napi_alloc_frag(really_needed_size + align);
    buf = PTR_ALIGN(buf, align);

These codes seems ugly and would waste a lot of memories if the buffers
are used in a network driver for the TX/RX. We have added the align
support for the page_frag functions, so add the corresponding
{netdev,napi}_frag functions.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mm: page_frag: Introduce page_frag_alloc_align()

In the current implementation of page_frag_alloc(), it doesn't have
any align guarantee for the returned buffer address. But for some
hardwares they do require the DMA buffer to be aligned correctly,
so we would have to use some workarounds like below if the buffers
allocated by the page_frag_alloc() are used by these hardwares for
DMA.
    buf = page_frag_alloc(really_needed_size + align);
    buf = PTR_ALIGN(buf, align);

These codes seems ugly and would waste a lot of memories if the buffers
are used in a network driver for the TX/RX. So introduce
page_frag_alloc_align() to make sure that an aligned buffer address is
returned.

Signed-off-by: Kevin Hao <haokexin@gmail.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dwc-xlgmac: Fix spelling mistake in function name

There is a spelling mistake in the function name alloc_channles_and_rings.
Fix this by renaming it to alloc_channels_and_rings.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Link: https://lore.kernel.org/r/20210204094944.51460-1-colin.king@canonical.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: mhi-net: Add re-aggregation of fragmented packets

When device side MTU is larger than host side MTU, the packets
(typically rmnet packets) are split over multiple MHI transfers.
In that case, fragments must be re-aggregated to recover the packet
before forwarding to upper layer.

A fragmented packet result in -EOVERFLOW MHI transaction status for
each of its fragments, except the final one. Such transfer was
previously considered as error and fragments were simply dropped.

This change adds re-aggregation mechanism using skb chaining, via
skb frag_list.

A warning (once) is printed since this behavior usually comes from
a misconfiguration of the device (e.g. modem MTU).

Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
Acked-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Link: https://lore.kernel.org/r/1612428002-12333-1-git-send-email-loic.poulain@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: qualcomm: rmnet: Fix rx_handler for non-linear skbs

There is no guarantee that rmnet rx_handler is only fed with linear
skbs, but current rmnet implementation does not check that, leading
to crash in case of non linear skbs processed as linear ones.

Fix that by ensuring skb linearization before processing.

Signed-off-by: Loic Poulain <loic.poulain@linaro.org>
Acked-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Link: https://lore.kernel.org/r/1612428002-12333-2-git-send-email-loic.poulain@linaro.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: sched: Return the correct errno code

When kalloc or kmemdup failed, should return ENOMEM rather than ENOBUF.

Signed-off-by: Zheng Yongjun <zhengyongjun3@huawei.com>
Link: https://lore.kernel.org/r/20210204073950.18372-1-zhengyongjun3@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

dccp: Return the correct errno code

When kalloc or kmemdup failed, should return ENOMEM rather than ENOBUF.

Signed-off-by: Zheng Yongjun <zhengyongjun3@huawei.com>
Link: https://lore.kernel.org/r/20210204072820.17723-1-zhengyongjun3@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: bridge: mcast: Use ERR_CAST instead of ERR_PTR(PTR_ERR())

Use ERR_CAST inlined function instead of ERR_PTR(PTR_ERR(...)).

net/bridge/br_multicast.c:1246:9-16: WARNING: ERR_CAST can be used with mp
Generated by: scripts/coccinelle/api/err_cast.cocci

Signed-off-by: Xu Wang <vulab@iscas.ac.cn>
Link: https://lore.kernel.org/r/20210204070549.83636-1-vulab@iscas.ac.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: ethernet: ti: fix netdevice stats for XDP

Align netdevice statistics when the device is running in XDP mode
to other upstream drivers. In particular report to user-space rx
packets even if they are not forwarded to the networking stack
(XDP_PASS) but if they are redirected (XDP_REDIRECT), dropped (XDP_DROP)
or sent back using the same interface (XDP_TX). This patch allows the
system administrator to verify the device is receiving data correctly.

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Link: https://lore.kernel.org/r/a457cb17dd9c58c116d64ee34c354b2e89c0ff8f.1612375372.git.lorenzo@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

dpaa2-eth: Simplify the calculation of variables

Fix the following coccicheck warnings:

./drivers/net/ethernet/freescale/dpaa2/dpaa2-eth.c:1651:36-38: WARNING
!A || A && B is equivalent to !A || B.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Acked-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Link: https://lore.kernel.org/r/1612260157-128026-1-git-send-email-jiapeng.chong@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'wireless-drivers-next-2021-02-05' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers-next patches for v5.12

First set of patches for v5.12. A smaller pull request this time,
biggest feature being a better key handling for ath9k. And of course
the usual fixes and cleanups all over.

Major changes:

ath9k
 * more robust encryption key cache management

brcmfmac
 * support BCM4365E with 43666 ChipCommon chip ID

* tag 'wireless-drivers-next-2021-02-05' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next: (35 commits)
  iwl4965: do not process non-QOS frames on txq->sched_retry path
  mt7601u: process tx URBs with status EPROTO properly
  wlcore: Fix command execute failure 19 for wl12xx
  mt7601u: use ieee80211_rx_list to pass frames to the network stack as a batch
  rtw88: 8723de: adjust the LTR setting
  rtlwifi: rtl8821ae: fix bool comparison in expressions
  rtlwifi: rtl8192se: fix bool comparison in expressions
  rtlwifi: rtl8188ee: fix bool comparison in expressions
  rtlwifi: rtl8192c-common: fix bool comparison in expressions
  rtlwifi: rtl_pci: fix bool comparison in expressions
  wlcore: Downgrade exceeded max RX BA sessions to debug
  wilc1000: use flexible-array member instead of zero-length array
  brcmfmac: clear EAP/association status bits on linkdown events
  brcmfmac: Delete useless kfree code
  qtnfmac_pcie: Use module_pci_driver
  mt7601u: check the status of device in calibration
  mt7601u: process URBs in status EPROTO properly
  brcmfmac: support BCM4365E with 43666 ChipCommon chip ID
  wilc1000: fix spelling mistake in Kconfig "devision" -> "division"
  mwifiex: pcie: Drop bogus __refdata annotation
  ...
====================

Link: https://lore.kernel.org/r/20210205161901.C7F83C433ED@smtp.codeaurora.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch '1GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
1GbE Intel Wired LAN Driver Updates 2021-02-03

This series contains updates to igc, igb, e1000e, and e1000 drivers.

Sasha adds counting of good transmit packets and reporting of NVM version
and gPHY version in ethtool firmware version. Replaces the use of strlcpy
to the preferred strscpy. Fixes a typo that caused the wrong register to be
output. He also removes an unused function pointer, some unneeded defines,
and a non-applicable comment. All changes for igc.

Gal Hammer fixes a typo which caused the RDBAL register values to be
shown instead of TDBAL for igb.

Nick Lowe enables RSS support for i211 devices for igb.

Tom Rix fixes checkpatch warning by removing h from printk format
specifier for igb.

Kaixu Xia removes setting of a variable that is overwritten before next
use for e1000e.

Sudip Mukherjee removes an unneeded assignment for e1000.

* '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  e1000: drop unneeded assignment in e1000_set_itr()
  e1000e: remove the redundant value assignment in e1000_update_nvm_checksum_spt
  igb: remove h from printk format specifier
  igb: Enable RSS for Intel I211 Ethernet Controller
  igb: fix TDBAL register show incorrect value
  igc: Fix TDBAL register show incorrect value
  igc: Remove unused FUNC_1 mask
  igc: Remove unused local receiver mask
  igc: Prefer strscpy over strlcpy
  igc: Expose the gPHY firmware version
  igc: Expose the NVM version
  igc: Add Host Good Packets Transmitted Count
  igc: Remove MULR mask define
  igc: Remove igc_set_fw_version comment
  igc: Clean up nvm_operations structure
====================

Link: https://lore.kernel.org/r/20210204004259.3662059-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'amend-hv_netvsc-copy-packets-sent-by-hyper-v-out-of-the-receive-buffer'

Andrea Parri says:

====================
Amend "hv_netvsc: Copy packets sent by Hyper-V out of the receive buffer"

Patch #2 also addresses the Smatch complaint reported here:

   https://lkml.kernel.org/r/YBp2oVIdMe+G%2FliJ@mwanda/
====================

Link: https://lore.kernel.org/r/20210203113513.558864-1-parri.andrea@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

hv_netvsc: Load and store the proper (NBL_HASH_INFO) per-packet info

Fix the typo.

Signed-off-by: Andrea Parri (Microsoft) <parri.andrea@gmail.com>
Fixes: 0ba35fe91ce34f ("hv_netvsc: Copy packets sent by Hyper-V out of the receive buffer")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

hv_netvsc: Allocate the recv_buf buffers after NVSP_MSG1_TYPE_SEND_RECV_BUF

The recv_buf buffers are allocated in netvsc_device_add().  Later in
netvsc_init_buf() the response to NVSP_MSG1_TYPE_SEND_RECV_BUF allows
the host to set up a recv_section_size that could be bigger than the
(default) value used for that allocation.  The host-controlled value
could be used by a malicious host to bypass the check on the packet's
length in netvsc_receive() and hence to overflow the recv_buf buffer.

Move the allocation of the recv_buf buffers into netvsc_init_but().

Reported-by: Juan Vazquez <juvazq@microsoft.com>
Signed-off-by: Andrea Parri (Microsoft) <parri.andrea@gmail.com>
Fixes: 0ba35fe91ce34f ("hv_netvsc: Copy packets sent by Hyper-V out of the receive buffer")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'r8152-adjust-flow-for-power-cut'

Hayes Wang says:

====================
r8152: adjust flow for power cut

The two patches are used to adjust the flow about resuming from
the state of power cut. For the purpose, some functions have to
be updated first.
====================

Link: https://lore.kernel.org/r/1394712342-15778-398-Taiwan-albertk@realtek.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

r8152: adjust the flow of power cut for RTL8153B

For runtime resuming, the RTL8153B may be resumed from the state
of power cut, when enabling the feature of UPS. Then, the PHY
would be reset, so it is necessary to be initailized again.

Besides, the USB_U1U2_TIMER also has to be set again, so I move
it from r8153b_init() to r8153b_hw_phy_cfg().

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

r8152: replace several functions about phy patch request

Replace r8153_patch_request() with rtl_phy_patch_request().
Replace r8153_pre_ram_code() with rtl_pre_ram_code().
Replace r8153_post_ram_code() with rtl_post_ram_code().
Add rtl_patch_key_set().

The new functions have an additional parameter. It is used to wait
the patch request command finished. When the PHY is resumed from
the state of power cut, the PHY is at a safe mode and the
OCP_PHY_PATCH_STAT wouldn't be updated. For this situation, it is
safe to set patch request command without waiting OCP_PHY_PATCH_STAT.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: xrs700x: Correctly address device over I2C

On read, master should send 31 MSB of the register (only even values
are ever used), followed by a 1 to indicate read. Then, reading two
bytes, the device will output the register's value.

On write, master sends 31 MSB of the register, followed by a 0 to
indicate write, followed by two bytes containing the register value.

Flexibilis' documentation (version 1.3) specifies the opposite
polarity (#read/write), but the scope indicates that it is, in fact,
read/#write.

Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Reviewed-by: George McCollister <george.mccollister@gmail.com>
Link: https://lore.kernel.org/r/20210202191645.439-1-tobias@waldekranz.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: bcm_sf2: Check egress tagging of CFP rule with proper accessor

The flow steering struct ethtool_flow_ext::data field is __be32, so when
the CFP code needs to check the VLAN egress tagging attribute in bit 0,
it does this in CPU native endianness. So logically, the endianness
conversion is set up the other way around, although in practice the same
result is produced.

Gets rid of build warning:

warning: cast from restricted __be32
warning: incorrect type in argument 1 (different base types)
   expected unsigned int [usertype] val
   got restricted __be32
warning: cast from restricted __be32
warning: cast from restricted __be32
warning: cast from restricted __be32
warning: cast from restricted __be32
warning: restricted __be32 degrades to integer

Signed-off-by: Vladimir Oltean <olteanv@gmail.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Link: https://lore.kernel.org/r/20210203193918.2236994-1-olteanv@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

drivers: net: ethernet: i825xx: Fix couple of spellings in the file ether1.c

s/initialsation/initialisation/
s/specifiing/specifying/

Signed-off-by: Bhaskar Chowdhury <unixbhaskar@gmail.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Link: https://lore.kernel.org/r/20210204031648.27300-1-unixbhaskar@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'fix-w-1-compilation-warnings-in-net-folder'

Leon Romanovsky says:

====================
Fix W=1 compilation warnings in net/* folder

Changelog:
v2:
 * Patch 3: Added missing include file.
v1: https://lore.kernel.org/lkml/20210203101612.4004322-1-leon@kernel.org
 * Removed Fixes lines.
 * Changed target from net to be net-next.
 * Patch 1: Moved function declaration to be outside config instead
   games with if/endif.
 * Patch 3: Moved declarations to new header file.
v0: https://lore.kernel.org/lkml/20210202135544.3262383-1-leon@kernel.org
====================

Link: https://lore.kernel.org/r/20210203135112.4083711-1-leon@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

netfilter: move handlers to net/ip_vs.h

Fix the following compilation warnings:
net/netfilter/ipvs/ip_vs_proto_tcp.c:147:1: warning: no previous prototype for 'tcp_snat_handler' [-Wmissing-prototypes]
  147 | tcp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
      | ^~~~~~~~~~~~~~~~
net/netfilter/ipvs/ip_vs_proto_udp.c:136:1: warning: no previous prototype for 'udp_snat_handler' [-Wmissing-prototypes]
  136 | udp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
      | ^~~~~~~~~~~~~~~~

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/core: move gro function declarations to separate header

Fir the following compilation warnings:
 1031 | INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb)

net/ipv6/ip6_offload.c:182:41: warning: no previous prototype for ‘ipv6_gro_receive’ [-Wmissing-prototypes]
  182 | INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
      |                                         ^~~~~~~~~~~~~~~~
net/ipv6/ip6_offload.c:320:29: warning: no previous prototype for ‘ipv6_gro_complete’ [-Wmissing-prototypes]
  320 | INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
      |                             ^~~~~~~~~~~~~~~~~
net/ipv6/ip6_offload.c:182:41: warning: no previous prototype for ‘ipv6_gro_receive’ [-Wmissing-prototypes]
  182 | INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
      |                                         ^~~~~~~~~~~~~~~~
net/ipv6/ip6_offload.c:320:29: warning: no previous prototype for ‘ipv6_gro_complete’ [-Wmissing-prototypes]
  320 | INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: move udp declarations to net/udp.h

Fix the following compilation warning:

net/ipv6/udp.c:1031:30: warning: no previous prototype for 'udp_v6_early_demux' [-Wmissing-prototypes]
 1031 | INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb)
      |                              ^~~~~~~~~~~~~~~~~~
net/ipv6/udp.c:1072:29: warning: no previous prototype for 'udpv6_rcv' [-Wmissing-prototypes]
 1072 | INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
      |                             ^~~~~~~~~

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ipv6: silence compilation warning for non-IPV6 builds

The W=1 compilation of allmodconfig generates the following warning:

net/ipv6/icmp.c:448:6: warning: no previous prototype for 'icmp6_send' [-Wmissing-prototypes]
  448 | void icmp6_send(struct sk_buff *skb, u8 type, u8 code, __u32 info,
      |      ^~~~~~~~~~

Fix it by providing function declaration for builds with ipv6 as a module.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: hns3: remove redundant null check of an array

The null check of filp->f_path.dentry->d_iname is redundant because
it is an array of DNAME_INLINE_LEN chars and cannot be a null. Fix
this by removing the null check.

Addresses-Coverity: ("Array compared against 0")
Fixes: 04987ca1b9b6 ("net: hns3: add debugfs support for tm nodes, priority and qset info")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Link: https://lore.kernel.org/r/20210203131040.21656-1-colin.king@canonical.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

nfc: pn533: Fix typo issue

change 'piority' to 'priority'
change 'succesfult' to 'successful'

Signed-off-by: wengjianfeng <wengjianfeng@yulong.com>
Link: https://lore.kernel.org/r/20210203093842.11180-1-samirweng1979@163.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-enable-udp-v6-sockets-receiving-v4-packets-with-udp'

Xin Long says:

====================
net: enable udp v6 sockets receiving v4 packets with UDP

Currently, udp v6 socket can not process v4 packets with UDP GRO, as
udp_encap_needed_key is not increased when udp_tunnel_encap_enable()
is called for v6 socket.

This patchset is to increase it and remove the unnecessary code in
bareudp in Patch 1/2, and improve rxrpc encap_enable by calling
udp_tunnel_encap_enable().
====================

Link: https://lore.kernel.org/r/cover.1612342376.git.lucien.xin@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

rxrpc: call udp_tunnel_encap_enable in rxrpc_open_socket

When doing encap_enable/increasing encap_needed_key, up->encap_enabled
is not set in rxrpc_open_socket(), and it will cause encap_needed_key
not being decreased in udpv6_destroy_sock().

This patch is to improve it by just calling udp_tunnel_encap_enable()
where it increases both UDP and UDPv6 encap_needed_key and sets
up->encap_enabled.

v4->v5:
  - add the missing '#include <net/udp_tunnel.h>', as David Howells
    noticed.

Acked-and-tested-by: David Howells <dhowells@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

udp: call udp_encap_enable for v6 sockets when enabling encap

When enabling encap for a ipv6 socket without udp_encap_needed_key
increased, UDP GRO won't work for v4 mapped v6 address packets as
sk will be NULL in udp4_gro_receive().

This patch is to enable it by increasing udp_encap_needed_key for
v6 sockets in udp_tunnel_encap_enable(), and correspondingly
decrease udp_encap_needed_key in udpv6_destroy_sock().

v1->v2:
  - add udp_encap_disable() and export it.
v2->v3:
  - add the change for rxrpc and bareudp into one patch, as Alex
    suggested.
v3->v4:
  - move rxrpc part to another patch.

Acked-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net-loopback: set lo dev initial state to UP

Traditionally loopback devices come up with initial state as DOWN for
any new network-namespace. This would mean that anyone needing this
device would have to bring this UP by issuing something like 'ip link
set lo up'. This can be avoided if the initial state is set as UP.

Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: Jian Yang <jianyang@google.com>
Link: https://lore.kernel.org/r/20210201233445.2044327-1-jianyang.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'net-consolidate-page_is_pfmemalloc-usage'

Alexander Lobakin says:

====================
net: consolidate page_is_pfmemalloc() usage

page_is_pfmemalloc() is used mostly by networking drivers to test
if a page can be considered for reusing/recycling.
It doesn't write anything to the struct page itself, so its sole
argument can be constified, as well as the first argument of
skb_propagate_pfmemalloc().
In Page Pool core code, it can be simply inlined instead.
Most of the callers from NIC drivers were just doppelgangers of
the same condition tests. Derive them into a new common function
do deduplicate the code.

Resend of v3 [2]:
 - it missed Patchwork and Netdev archives, probably due to server-side
   issues.

Since v2 [1]:
 - use more intuitive name for the new inline function since there's
   nothing "reserved" in remote pages (Jakub Kicinski, John Hubbard);
 - fold likely() inside the helper itself to make driver code a bit
   fancier (Jakub Kicinski);
 - split function introduction and using into two separate commits;
 - collect some more tags (Jesse Brandeburg, David Rientjes).

Since v1 [0]:
 - new: reduce code duplication by introducing a new common function
   to test if a page can be reused/recycled (David Rientjes);
 - collect autographs for Page Pool bits (Jesper Dangaard Brouer,
   Ilias Apalodimas).

[0] https://lore.kernel.org/netdev/20210125164612.243838-1-alobakin@pm.me
[1] https://lore.kernel.org/netdev/20210127201031.98544-1-alobakin@pm.me
[2] https://lore.kernel.org/lkml/20210131120844.7529-1-alobakin@pm.me
====================

Link: https://lore.kernel.org/r/20210202133030.5760-1-alobakin@pm.me
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: page_pool: simplify page recycling condition tests

pool_page_reusable() is a leftover from pre-NUMA-aware times. For now,
this function is just a redundant wrapper over page_is_pfmemalloc(),
so inline it into its sole call site.

Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: use the new dev_page_is_reusable() instead of private versions

Now we can remove a bunch of identical functions from the drivers and
make them use common dev_page_is_reusable(). All {,un}likely() checks
are omitted since it's already present in this helper.
Also update some comments near the call sites.

Suggested-by: David Rientjes <rientjes@google.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: introduce common dev_page_is_reusable()

A bunch of drivers test the page before reusing/recycling for two
common conditions:
 - if a page was allocated under memory pressure (pfmemalloc page);
 - if a page was allocated at a distant memory node (to exclude
   slowdowns).

Introduce a new common inline for doing this, with likely() already
folded inside to make driver code a bit simpler.

Suggested-by: David Rientjes <rientjes@google.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

skbuff: constify skb_propagate_pfmemalloc() "page" argument

The function doesn't write anything to the page struct itself,
so this argument can be const.

Misc: align second argument to the brace while at it.

Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mm: constify page_is_pfmemalloc() argument

The function only tests for page->index, so its argument should be
const.

Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: fix building errors on powerpc when CONFIG_RETPOLINE is not set

This commit fixes the errores reported when building for powerpc:

 ERROR: modpost: "ip6_dst_check" [vmlinux] is a static EXPORT_SYMBOL
 ERROR: modpost: "ipv4_dst_check" [vmlinux] is a static EXPORT_SYMBOL
 ERROR: modpost: "ipv4_mtu" [vmlinux] is a static EXPORT_SYMBOL
 ERROR: modpost: "ip6_mtu" [vmlinux] is a static EXPORT_SYMBOL

Fixes: f67fbeaebdc0 ("net: use indirect call helpers for dst_mtu")
Fixes: bbd807dfbf20 ("net: indirect call helpers for ipv4/ipv6 dst_check functions")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Link: https://lore.kernel.org/r/20210204181839.558951-2-brianvv@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: add EXPORT_INDIRECT_CALLABLE wrapper

When a static function is annotated with INDIRECT_CALLABLE_SCOPE and
CONFIG_RETPOLINE is set, the static keyword is removed. Sometimes the
function needs to be exported but EXPORT_SYMBOL can't be used because if
CONFIG_RETPOLINE is not set, we will attempt to export a static symbol.

This patch introduces a new indirect call wrapper:
EXPORT_INDIRECT_CALLABLE. This basically does EXPORT_SYMBOL when
CONFIG_RETPOLINE is set, but does nothing when it's not.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Brian Vazquez <brianvv@google.com>
Link: https://lore.kernel.org/r/20210204181839.558951-1-brianvv@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

netlink: add tracepoint at NL_SET_ERR_MSG

Often userspace won't request the extack information, or they don't log it
because of log level or so, and even when they do, sometimes it's not
enough to know exactly what caused the error.

Netlink extack is the standard way of reporting erros with descriptive
error messages. With a trace point on it, we then can know exactly where
the error happened, regardless of userspace app. Also, we can even see if
the err msg was overwritten.

The wrapper do_trace_netlink_extack() is because trace points shouldn't be
called from .h files, as trace points are not that small, and the function
call to do_trace_netlink_extack() on the macros is not protected by
tracepoint_enabled() because the macros are called from modules, and this
would require exporting some trace structs. As this is error path, it's
better to export just the wrapper instead.

v2: removed leftover tracepoint declaration

Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/4546b63e67b2989789d146498b13cc09e1fdc543.1612403190.git.marcelo.leitner@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

drivers: net: xen-netfront: Simplify the calculation of variables

Fix the following coccicheck warnings:

./drivers/net/xen-netfront.c:1816:52-54: WARNING !A || A && B is
equivalent to !A || B.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Link: https://lore.kernel.org/r/1612261069-13315-1-git-send-email-jiapeng.chong@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'gtp'

Jonas Bonn says:

====================
There's ongoing work in this driver to provide support for IPv6, GRO,
GSO, and "collect metadata" mode operation.  In order to facilitate this
work going forward, this short series accumulates already ACK:ed patches
that are ready for the next merge window.

All of these patches should be uncontroversial at this point, including
the first one in the series that reverts a recently added change to
introduce "collect metadata" mode.  As that patch produces 'broken'
packets when common GTP headers are in place, it seems better to revert
it and rethink things a bit before inclusion.
====================

Link: https://lore.kernel.org/r/20210203070805.281321-1-jonas@norrbonn.se
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

gtp: update rx_length_errors for abnormally short packets

Based on work by Pravin Shelar.

Update appropriate stats when packet transmission isn't possible.

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Harald Welte <laforge@gnumonks.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

gtp: set device type

Set the devtype to 'gtp' when setting up the link.

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Harald Welte <laforge@gnumonks.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

gtp: drop unnecessary call to skb_dst_drop

The call to skb_dst_drop() is already done as part of udp_tunnel_xmit().

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Harald Welte <laforge@gnumonks.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

gtp: really check namespaces before xmit

Blindly assuming that packet transmission crosses namespaces results in
skb marks being lost in the single namespace case.

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Harald Welte <laforge@gnumonks.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

gtp: include role in link info

Querying link info for the GTP interface doesn't reveal in which "role" the
device is set to operate.  Include this information in the info query
result.

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Harald Welte <laforge@gnumonks.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

gtp: set initial MTU

The GTP link is brought up with a default MTU of zero.  This can lead to
some rather unexpected behaviour for users who are more accustomed to
interfaces coming online with reasonable defaults.

This patch sets an initial MTU for the GTP link of 1500 less worst-case
tunnel overhead.

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Harald Welte <laforge@gnumonks.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Revert "GTP: add support for flow based tunneling API"

This reverts commit 9ab7e76aefc97a9aa664accb59d6e8dc5e52514a.

This patch was committed without maintainer approval and despite a number
of unaddressed concerns from review.  There are several issues that
impede the acceptance of this patch and that make a reversion of this
particular instance of these changes the best way forward:

i)  the patch contains several logically separate changes that would be
better served as smaller patches (for review purposes)
ii) functionality like the handling of end markers has been introduced
without further explanation
iii) symmetry between the handling of GTPv0 and GTPv1 has been
unnecessarily broken
iv) the patchset produces 'broken' packets when extension headers are
included
v) there are no available userspace tools to allow for testing this
functionality
vi) there is an unaddressed Coverity report against the patch concering
memory leakage
vii) most importantly, the patch contains a large amount of superfluous
churn that impedes other ongoing work with this driver

This patch will be reworked into a series that aligns with other
ongoing work and facilitates review.

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Harald Welte <laforge@gnumonks.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: tracepoint: exposing sk_family in all tcp:tracepoints

Similar to sock:inet_sock_set_state tracepoint, expose sk_family to
distinguish AF_INET and AF_INET6 families.

The following tcp tracepoints are updated:
tcp:tcp_destroy_sock
tcp:tcp_rcv_space_adjust
tcp:tcp_retransmit_skb
tcp:tcp_send_reset
tcp:tcp_receive_reset
tcp:tcp_retransmit_synack
tcp:tcp_probe

Signed-off-by: Hariharan Ananthakrishnan <hari@netflix.com>
Signed-off-by: Brendan Gregg <bgregg@netflix.com>
Link: https://lore.kernel.org/r/20210129001210.344438-1-hari@netflix.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tcp: use a smaller percpu_counter batch size for sk_alloc

Currently, a percpu_counter with the default batch size (2*nr_cpus) is
used to record the total # of active sockets per protocol. This means
sk_sockets_allocated_read_positive() could be off by +/-2*(nr_cpus^2).
This under/over-estimation could lead to wrong memory suppression
conditions in __sk_raise_mem_allocated().
Fix this by using a more reasonable fixed batch size of 16.

See related commit cf86a086a180 ("net/dst: use a smaller percpu_counter
batch for dst entries accounting") that addresses a similar issue.

Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20210202193408.1171634-1-weiwan@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'support-setting-lanes-via-ethtool'

Danielle Ratson says:

====================
Support setting lanes via ethtool

Some speeds can be achieved with different number of lanes. For example,
100Gbps can be achieved using two lanes of 50Gbps or four lanes of
25Gbps. This patchset adds a new selector that allows ethtool to
advertise link modes according to their number of lanes and also force a
specific number of lanes when autonegotiation is off.

Advertising all link modes with a speed of 100Gbps that use two lanes:

$ ethtool -s swp1 speed 100000 lanes 2 autoneg on

Forcing a speed of 100Gbps using four lanes:

$ ethtool -s swp1 speed 100000 lanes 4 autoneg off
====================

Link: https://lore.kernel.org/r/20210202180612.325099-1-danieller@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: selftests: Add lanes setting test

Test that setting lanes parameter is working.

Set max speed and max lanes in the list of advertised link modes,
and then try to set max speed with the lanes below max lanes if exists
in the list.

And then, test that setting number of lanes larger than max lanes fails.

Do the above for both autoneg on and off.

$ ./ethtool_lanes.sh

TEST: 4 lanes is autonegotiated                                     [ OK ]
TEST: Lanes number larger than max width is not set                 [ OK ]
TEST: Autoneg off, 4 lanes detected during force mode               [ OK ]
TEST: Lanes number larger than max width is not set                 [ OK ]

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: ethtool: Pass link mode in use to ethtool

Currently, when user space queries the link's parameters, as speed and
duplex, each parameter is passed from the driver to ethtool.

Instead, pass the link mode bit in use.
In Spectrum-1, simply pass the bit that is set to '1' from PTYS register.
In Spectrum-2, pass the first link mode bit in the mask of the used
link mode.

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: ethtool: Add support for setting lanes when autoneg is off

Currently, when auto negotiation is set to off, the user can force a
specific speed or both speed and duplex. The user cannot influence the
number of lanes that will be forced.

Add support for setting speed along with lanes so one would be able
to choose how many lanes will be forced.

When lanes parameter is passed from user space, choose the link mode
that its actual width equals to it.
Otherwise, the default link mode will be the one that supports the width
of the port.

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

mlxsw: ethtool: Remove max lanes filtering

Currently, when a speed can be supported by different number of lanes,
the supported link modes bitmask contains only link modes with a single
number of lanes.

This was done in order to prevent auto negotiation on number of
lanes after 50G-1-lane and 100G-2-lanes link modes were introduced.

For example, if a port's max width is 4, only link modes with 4 lanes
will be presented as supported by that port, so 100G is always achieved by
4 lanes of 25G.

After the previous patches that allow selection of the number of lanes,
auto negotiation on number of lanes becomes practical.

Remove that filtering of the maximum number of lanes supported link modes,
so indeed all the supported and advertised link modes will be shown.

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ethtool: Expose the number of lanes in use

Currently, ethtool does not expose how many lanes are used when the
link is up.

After adding a possibility to advertise or force a specific number of
lanes, the lanes in use value can be either the maximum width of the port
or below.

Extend ethtool to expose the number of lanes currently in use for
drivers that support it.

For example:

$ ethtool -s swp1 speed 100000 lanes 4
$ ethtool -s swp2 speed 100000 lanes 4
$ ip link set swp1 up
$ ip link set swp2 up
$ ethtool swp1
Settings for swp1:
        Supported ports: [ FIBRE         Backplane ]
        Supported link modes:   1000baseT/Full
                                10000baseT/Full
                                1000baseKX/Full
                                10000baseKR/Full
                                10000baseR_FEC
                                40000baseKR4/Full
                                40000baseCR4/Full
                                40000baseSR4/Full
                                40000baseLR4/Full
                                25000baseCR/Full
                                25000baseKR/Full
                                25000baseSR/Full
                                50000baseCR2/Full
                                50000baseKR2/Full
                                100000baseKR4/Full
                                100000baseSR4/Full
                                100000baseCR4/Full
                                100000baseLR4_ER4/Full
                                50000baseSR2/Full
                                10000baseCR/Full
                                10000baseSR/Full
                                10000baseLR/Full
                                10000baseER/Full
                                50000baseKR/Full
                                50000baseSR/Full
                                50000baseCR/Full
                                50000baseLR_ER_FR/Full
                                50000baseDR/Full
                                100000baseKR2/Full
                                100000baseSR2/Full
                                100000baseCR2/Full
                                100000baseLR2_ER2_FR2/Full
                                100000baseDR2/Full
                                200000baseKR4/Full
                                200000baseSR4/Full
                                200000baseLR4_ER4_FR4/Full
                                200000baseDR4/Full
                                200000baseCR4/Full
        Supported pause frame use: Symmetric Receive-only
        Supports auto-negotiation: Yes
        Supported FEC modes: Not reported
        Advertised link modes:  1000baseT/Full
                                10000baseT/Full
                                1000baseKX/Full
                                1000baseKX/Full
                                10000baseKR/Full
                                10000baseR_FEC
                                40000baseKR4/Full
                                40000baseCR4/Full
                                40000baseSR4/Full
                                40000baseLR4/Full
                                25000baseCR/Full
                                25000baseKR/Full
                                25000baseSR/Full
                                50000baseCR2/Full
                                50000baseKR2/Full
                                100000baseKR4/Full
                                100000baseSR4/Full
                                100000baseCR4/Full
                                100000baseLR4_ER4/Full
                                50000baseSR2/Full
                                10000baseCR/Full
                                10000baseSR/Full
                                10000baseLR/Full
                                10000baseER/Full
                                200000baseKR4/Full
                                200000baseSR4/Full
                                200000baseLR4_ER4_FR4/Full
                                200000baseDR4/Full
                                200000baseCR4/Full
        Advertised pause frame use: No
        Advertised auto-negotiation: Yes
        Advertised FEC modes: Not reported
        Advertised link modes:  100000baseKR4/Full
                                100000baseSR4/Full
                                100000baseCR4/Full
                                100000baseLR4_ER4/Full
Advertised pause frame use: No
Advertised auto-negotiation: Yes
Advertised FEC modes: Not reported
Speed: 100000Mb/s
Lanes: 4
Duplex: Full
Auto-negotiation: on
Port: Direct Attach Copper
PHYAD: 0
Transceiver: internal
Link detected: yes

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ethtool: Get link mode in use instead of speed and duplex parameters

Currently, when user space queries the link's parameters, as speed and
duplex, each parameter is passed from the driver to ethtool.

Instead, get the link mode bit in use, and derive each of the parameters
from it in ethtool.

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ethtool: Extend link modes settings uAPI with lanes

Currently, when auto negotiation is on, the user can advertise all the
linkmodes which correspond to a specific speed, but does not have a
similar selector for the number of lanes. This is significant when a
specific speed can be achieved using different number of lanes.  For
example, 2x50 or 4x25.

Add 'ETHTOOL_A_LINKMODES_LANES' attribute and expand 'struct
ethtool_link_settings' with lanes field in order to implement a new
lanes-selector that will enable the user to advertise a specific number
of lanes as well.

When auto negotiation is off, lanes parameter can be forced only if the
driver supports it. Add a capability bit in 'struct ethtool_ops' that
allows ethtool know if the driver can handle the lanes parameter when
auto negotiation is off, so if it does not, an error message will be
returned when trying to set lanes.

Example:

$ ethtool -s swp1 lanes 4
$ ethtool swp1
  Settings for swp1:
Supported ports: [ FIBRE ]
        Supported link modes:   1000baseKX/Full
                                10000baseKR/Full
                                40000baseCR4/Full
40000baseSR4/Full
40000baseLR4/Full
                                25000baseCR/Full
                                25000baseSR/Full
50000baseCR2/Full
                                100000baseSR4/Full
100000baseCR4/Full
        Supported pause frame use: Symmetric Receive-only
        Supports auto-negotiation: Yes
        Supported FEC modes: Not reported
        Advertised link modes:  40000baseCR4/Full
40000baseSR4/Full
40000baseLR4/Full
                                100000baseSR4/Full
100000baseCR4/Full
        Advertised pause frame use: No
        Advertised auto-negotiation: Yes
        Advertised FEC modes: Not reported
        Speed: Unknown!
        Duplex: Unknown! (255)
        Auto-negotiation: on
        Port: Direct Attach Copper
        PHYAD: 0
        Transceiver: internal
        Link detected: no

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

ethtool: Validate master slave configuration before rtnl_lock()

Create a new function for input validations to be called before
rtnl_lock() and move the master slave validation to that function.

This would be a cleanup for next patch that would add another validation
to the new function.

Signed-off-by: Danielle Ratson <danieller@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: dsa: fix SWITCHDEV_ATTR_ID_BRIDGE_VLAN_FILTERING getting ignored

The bridge emits VLAN filtering events and quite a few others via
switchdev with orig_dev = br->dev. After the blamed commit, these events
started getting ignored.

The point of the patch was to not offload switchdev objects for ports
that didn't go through dsa_port_bridge_join, because the configuration
is unsupported:
- ports that offload a bonding/team interface go through
  dsa_port_bridge_join when that bonding/team interface is later bridged
  with another switch port or LAG
- ports that don't offload LAG don't get notified of the bridge that is
  on top of that LAG.

Sadly, a check is missing, which is that the orig_dev is equal to the
bridge device. This check is compatible with the original intention,
because ports that don't offload bridging because they use a software
LAG don't have dp->bridge_dev set.

On a semi-related note, we should not offload switchdev objects or
populate dp->bridge_dev if the driver doesn't implement .port_bridge_join
either. However there is no regression associated with that, so it can
be done separately.

Fixes: 5696c8aedfcc ("net: dsa: Don't offload port attributes on standalone ports")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Tobias Waldekranz <tobias@waldekranz.com>
Tested-by: Tobias Waldekranz <tobias@waldekranz.com>
Link: https://lore.kernel.org/r/20210202233109.1591466-1-olteanv@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'chelsio-cxgb-use-threaded-interrupts-for-deferred-work'

Sebastian Andrzej Siewior says:

====================
chelsio: cxgb: Use threaded interrupts for deferred work

Patch #2 fixes an issue in which del_timer_sync() and tasklet_kill() is
invoked from the interrupt handler. This is probably a rare error case
since it disables interrupts / the card in that case.
Patch #1 converts a worker to use a threaded interrupt which is then
also used in patch #2 instead adding another worker for this task (and
flush_work() to synchronise vs rmmod).
====================

Link: https://lore.kernel.org/r/20210202170104.1909200-1-bigeasy@linutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

chelsio: cxgb: Disable the card on error in threaded interrupt

t1_fatal_err() is invoked from the interrupt handler. The bad part is
that it invokes (via t1_sge_stop()) del_timer_sync() and tasklet_kill().
Both functions must not be called from an interrupt because it is
possible that it will wait for the completion of the timer/tasklet it
just interrupted.

In case of a fatal error, use t1_interrupts_disable() to disable all
interrupt sources and then wake the interrupt thread with
F_PL_INTR_SGE_ERR as pending flag. The threaded-interrupt will stop the
card via t1_sge_stop() and not re-enable the interrupts again.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

chelsio: cxgb: Replace the workqueue with threaded interrupt

The external interrupt (F_PL_INTR_EXT) needs to be handled in a process
context and this is accomplished by utilizing a workqueue.

The process context can also be provided by a threaded interrupt instead
of a workqueue. The threaded interrupt can be used later for other
interrupt related processing which require non-atomic context without
using yet another workqueue. free_irq() also ensures that the thread is
done which is currently missing (the worker could continue after the
module has been removed).

Save pending flags in pending_thread_intr. Use the same mechanism
to disable F_PL_INTR_EXT as interrupt source like it is used before the
worker is scheduled. Enable the interrupt again once
t1_elmer0_ext_intr_handler() is done.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'support-for-octeontx2-98xx-cpt-block'

Srujana Challa says:

====================
Support for OcteonTX2 98xx CPT block.

OcteonTX2 series of silicons have multiple variants, the
98xx variant has two crypto (CPT) blocks to double the crypto
performance. This patchset adds support for new CPT block(CPT1).
====================

Link: https://lore.kernel.org/r/20210202152709.20450-1-schalla@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>