The driver can never use more queues of any type than the number of
possible CPUs, so a higher value causes the driver to allocate more
memory for IO queues than it could ever use. Limit the parameter at
module load time to the number of possible cpus.
Reviewed-by: NJens Axboe <axboe@kernel.dk>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

The number of poll or write queues should never be negative. Use unsigned
types so that it's not possible to break have the driver not allocate
any queues.
Reviewed-by: NJens Axboe <axboe@kernel.dk>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

We had been saving the last_cq_head seen from an interrupt so that a
polled queue wouldn't mistakenly trigger spruious interrupt detection. We
don't poll interrupt driven queues any more, so saving this value is
pointless.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

Despite NVM Express specification 1.3 requires a controller claiming to
be 1.3 or higher implement Identify CNS 03h (Namespace Identification
Descriptor list), the driver doesn't really need this identification in
order to use a namespace. The code had already documented in comments
that we're not to consider an error to this command.

Return success if the controller provided any response to an
namespace identification descriptors command.

Fixes: 538af88e ("nvme: make nvme_report_ns_ids propagate error back")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=205679Reported-by: NIngo Brunberg <ingo_brunberg@web.de>
Cc: Sagi Grimberg <sagi@grimberg.me>
Cc: stable@vger.kernel.org # 5.4+
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

If an error occurs on one of the ios used for creating an
association, the creating routine has error paths that are
invoked by the command failure and the error paths will free
up the controller resources created to that point.

But... the io was ultimately determined by an asynchronous
completion routine that detected the error and which
unconditionally invokes the error_recovery path which calls
delete_association. Delete association deletes all outstanding
io then tears down the controller resources. So the
create_association thread can be running in parallel with
the error_recovery thread. What was seen was the LLDD received
a call to delete a queue, causing the LLDD to do a free of a
resource, then the transport called the delete queue again
causing the driver to repeat the free call. The second free
routine corrupted the allocator. The transport shouldn't be
making the duplicate call, and the delete queue is just one
of the resources being freed.

To fix, it is realized that the create_association path is
completely serialized with one command at a time. So the
failed io completion will always be seen by the create_association
path and as of the failure, there are no ios to terminate and there
is no reason to be manipulating queue freeze states, etc.
The serialized condition stays true until the controller is
transitioned to the LIVE state. Thus the fix is to change the
error recovery path to check the controller state and only
invoke the teardown path if not already in the CONNECTING state.
Reviewed-by: NHimanshu Madhani <hmadhani@marvell.com>
Reviewed-by: NEwan D. Milne <emilne@redhat.com>
Signed-off-by: NJames Smart <jsmart2021@gmail.com>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

Remove unnecessary keyword in nvme_create_queue().
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NEdmund Nadolski <edmund.nadolski@intel.com>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

We've seen a few devices that return different controller id's to
the Fabric Connect command vs the Identify(controller) command. It's
currently hard to identify this failure by existing error messages. It
comes across as a (re)connect attempt in the transport that fails with
a -22 (-EINVAL) status. The issue is compounded by older kernels not
having the controller id check or had the identify command overwrite the
fabrics controller id value before it checked. Both resulted in cases
where the devices appeared fine until more recent kernels.

Clarify the reject by adding an error message on controller id mismatches.
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NHannes Reinecke <hare@suse.de>
Reviewed-by: NEwan D. Milne <emilne@redhat.com>
Signed-off-by: NJames Smart <jsmart2021@gmail.com>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

In nvme-fc: it's possible to have connected active controllers
and as no references are taken on the LLDD, the LLDD can be
unloaded.  The controller would enter a reconnect state and as
long as the LLDD resumed within the reconnect timeout, the
controller would resume.  But if a namespace on the controller
is the root device, allowing the driver to unload can be problematic.
To reload the driver, it may require new io to the boot device,
and as it's no longer connected we get into a catch-22 that
eventually fails, and the system locks up.

Fix this issue by taking a module reference for every connected
controller (which is what the core layer did to the transport
module). Reference is cleared when the controller is removed.
Acked-by: NHimanshu Madhani <hmadhani@marvell.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJames Smart <jsmart2021@gmail.com>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

nvme_loop_create_io_queues() preallocates a big buffer for the IO SGL based
on SG_CHUNK_SIZE.

Modern DMA engines are often capable of dealing with very big segments so
the SG_CHUNK_SIZE is often too big. SG_CHUNK_SIZE results in a static 4KB
SGL allocation per command.

If a controller has lots of deep queues, preallocation for the sg list can
consume substantial amounts of memory. For nvmet-loop, nr_hw_queues can be
128 and each queue's depth 128. This means the resulting preallocation
for the data SGL is 128*128*4K = 64MB per controller.

Switch to runtime allocation for SGL for lists longer than 2 entries. This
is the approach used by NVMe PCI so it should be reasonable for NVMeOF as
well. Runtime SGL allocation has always been the case for the legacy I/O
path so this is nothing new.
Tested-by: NChaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NChaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: NMax Gurtovoy <maxg@mellanox.com>
Signed-off-by: NIsrael Rukshin <israelr@mellanox.com>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

nvme_fc_create_io_queues() preallocates a big buffer for the IO SGL based
on SG_CHUNK_SIZE.

Modern DMA engines are often capable of dealing with very big segments so
the SG_CHUNK_SIZE is often too big. SG_CHUNK_SIZE results in a static 4KB
SGL allocation per command.

If a controller has lots of deep queues, preallocation for the sg list can
consume substantial amounts of memory. For nvme-fc, nr_hw_queues can be
128 and each queue's depth 128. This means the resulting preallocation
for the data SGL is 128*128*4K = 64MB per controller.

Switch to runtime allocation for SGL for lists longer than 2 entries. This
is the approach used by NVMe PCI so it should be reasonable for NVMeOF as
well. Runtime SGL allocation has always been the case for the legacy I/O
path so this is nothing new.
Reviewed-by: NMax Gurtovoy <maxg@mellanox.com>
Reviewed-by: NJames Smart <james.smart@broadcom.com>
Signed-off-by: NIsrael Rukshin <israelr@mellanox.com>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

nvme_rdma_alloc_tagset() preallocates a big buffer for the IO SGL based
on SG_CHUNK_SIZE.

Modern DMA engines are often capable of dealing with very big segments so
the SG_CHUNK_SIZE is often too big. SG_CHUNK_SIZE results in a static 4KB
SGL allocation per command.

If a controller has lots of deep queues, preallocation for the sg list can
consume substantial amounts of memory. For nvme-rdma, nr_hw_queues can be
128 and each queue's depth 128. This means the resulting preallocation
for the data SGL is 128*128*4K = 64MB per controller.

Switch to runtime allocation for SGL for lists longer than 2 entries. This
is the approach used by NVMe PCI so it should be reasonable for NVMeOF as
well. Runtime SGL allocation has always been the case for the legacy I/O
path so this is nothing new.

The preallocated small SGL depends on SG_CHAIN so if the ARCH doesn't
support SG_CHAIN, use only runtime allocation for the SGL.

We didn't notice of a performance degradation, since for small IOs we'll
use the inline SG and for the bigger IOs the allocation of a bigger SGL
from slab is fast enough.
Suggested-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NMax Gurtovoy <maxg@mellanox.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NIsrael Rukshin <israelr@mellanox.com>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

Slip_open doesn't clean-up device which registration failed from the
slip_devs device list. On next open after failure this list is iterated
and freed device is accessed. Fix this by calling sl_free_netdev in error
path.

Here is the trace from the Syzbot:

__dump_stack lib/dump_stack.c:77 [inline]
dump_stack+0x197/0x210 lib/dump_stack.c:118
print_address_description.constprop.0.cold+0xd4/0x30b mm/kasan/report.c:374
__kasan_report.cold+0x1b/0x41 mm/kasan/report.c:506
kasan_report+0x12/0x20 mm/kasan/common.c:634
__asan_report_load8_noabort+0x14/0x20 mm/kasan/generic_report.c:132
sl_sync drivers/net/slip/slip.c:725 [inline]
slip_open+0xecd/0x11b7 drivers/net/slip/slip.c:801
tty_ldisc_open.isra.0+0xa3/0x110 drivers/tty/tty_ldisc.c:469
tty_set_ldisc+0x30e/0x6b0 drivers/tty/tty_ldisc.c:596
tiocsetd drivers/tty/tty_io.c:2334 [inline]
tty_ioctl+0xe8d/0x14f0 drivers/tty/tty_io.c:2594
vfs_ioctl fs/ioctl.c:46 [inline]
file_ioctl fs/ioctl.c:509 [inline]
do_vfs_ioctl+0xdb6/0x13e0 fs/ioctl.c:696
ksys_ioctl+0xab/0xd0 fs/ioctl.c:713
__do_sys_ioctl fs/ioctl.c:720 [inline]
__se_sys_ioctl fs/ioctl.c:718 [inline]
__x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:718
do_syscall_64+0xfa/0x760 arch/x86/entry/common.c:290
entry_SYSCALL_64_after_hwframe+0x49/0xbe

Fixes: 3b5a3997 ("slip: Fix memory leak in slip_open error path")
Reported-by: syzbot+4d5170758f3762109542@syzkaller.appspotmail.com
Cc: David Miller <davem@davemloft.net>
Cc: Oliver Hartkopp <socketcan@hartkopp.net>
Cc: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: NJouni Hogander <jouni.hogander@unikie.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This function was using configuration of port 0 in devicetree for all ports.
In case CPU port was not 0, the delay settings was ignored. This resulted not
working communication between CPU and the switch.

Fixes: f5b8631c ("net: dsa: sja1105: Error out if RGMII delays are requested in DT")
Signed-off-by: NOleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: NVladimir Oltean <olteanv@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While enqueueing a broadcast skb to port->bc_queue, schedule_work()
is called to add port->bc_work, which processes the skbs in
bc_queue, to "events" work queue. If port->bc_queue is full, the
skb will be discarded and schedule_work(&port->bc_work) won't be
called. However, if port->bc_queue is full and port->bc_work is not
running or pending, port->bc_queue will keep full and schedule_work()
won't be called any more, and all broadcast skbs to macvlan will be
discarded. This case can happen:

macvlan_process_broadcast() is the pending function of port->bc_work,
it moves all the skbs in port->bc_queue to the queue "list", and
processes the skbs in "list". During this, new skbs will keep being
added to port->bc_queue in macvlan_broadcast_enqueue(), and
port->bc_queue may already full when macvlan_process_broadcast()
return. This may happen, especially when there are a lot of real-time
threads and the process is preempted.

Fix this by calling schedule_work(&port->bc_work) even if
port->bc_work is full in macvlan_broadcast_enqueue().

Fixes: 412ca155 ("macvlan: Move broadcasts into a work queue")
Signed-off-by: NMenglong Dong <dong.menglong@zte.com.cn>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The ENETC hardware support the Credit Based Shaper(CBS) which part
of the IEEE-802.1Qav. The CBS driver was loaded by the sch_cbs
interface when set in the QOS in the kernel.

Here is an example command to set 20Mbits bandwidth in 1Gbits port
for taffic class 7:

tc qdisc add dev eth0 root handle 1: mqprio \
	   num_tc 8 map 0 1 2 3 4 5 6 7 hw 1

tc qdisc replace dev eth0 parent 1:8 cbs \
	   locredit -1470 hicredit 30 \
	   sendslope -980000 idleslope 20000 offload 1
Signed-off-by: NPo Liu <Po.Liu@nxp.com>
Reviewed-by: NClaudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: NVladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add helpers to make locking/unlocking the MDIO bus easier.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Geert Uytterhoeven reported that using devm_reset_controller_get leads
to a WARNING when probing a reset-controlled PHY. This is because the
device devm_reset_controller_get gets supplied is not actually the
one being probed.

Acquire an unmanaged reset-control as well as free the reset_control on
unregister to fix this.
Reported-by: NGeert Uytterhoeven <geert@linux-m68k.org>
CC: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: NDavid Bauer <mail@david-bauer.net>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This enables the use of SW timestamping.

ax88179_178a uses the usbnet transmit function usbnet_start_xmit() which
implements software timestamping. ax88179_178a overrides ethtool_ops but
missed to set .get_ts_info. This caused SOF_TIMESTAMPING_TX_SOFTWARE
capability to be not available.
Signed-off-by: NAndreas K. Besslein <besslein.andreas@gmail.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

When mlxsw_sp_adj_discard_write() is called for the first time, the
value stored in 'mlxsw_sp->router->adj_discard_index' is invalid, as
indicated by 'mlxsw_sp->router->adj_discard_index_valid' being set to
'false'.

In this case, we should not use the value initially stored in
'mlxsw_sp->router->adj_discard_index' (0) and instead use the value
allocated later in the function.

Fixes: 983db619 ("mlxsw: spectrum_router: Allocate discard adjacency entry when needed")
Signed-off-by: NAmit Cohen <amitc@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

When a GRE tunnel is bound to an underlay netdevice and that netdevice is
moved to a different VRF, that could cause two tunnels to have the same
underlay local address in the same VRF. Linux in this situation dispatches
the traffic according to the tunnel key (or lack thereof), but that cannot
be offloaded to Spectrum devices.

Detect this situation and unoffload the two impacted tunnels when it
happens.
Signed-off-by: NPetr Machata <petrm@mellanox.com>
Signed-off-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Use the same bnxt_flash_package_from_file() function to support
devlink flash operation.

Cc: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: NVasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Currently, the driver does not allow PHY settings on a multi-function or
NPAR NIC whose port is shared by more than one function.  Newer
firmware now allows PHY settings on some of these NICs.  Check for
this new firmware setting and allow the user to set the PHY settings
accordingly.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

If the link settings have been changed by another function sharing the
port, firmware will send us an async. message.  In response, we will
call the new bnxt_init_ethtool_link_settings() function to update
the current settings.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Refactor this logic in bnxt_probe_phy() into a separate function
bnxt_init_ethtool_link_settings().  It used to be that the settable
link settings will never be changed without going through ethtool.
So we only needed to do this once in bnxt_probe_phy().  Now, another
function sharing the port may change it and we may need to re-initialize
the ethtool settings again in run-time.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

New firmware allows PHY loopback to be set without disabling autoneg
first.  Check this capability and skip disabling autoneg when
it is supported by firmware.  Using this scheme, loopback will
always work even if the PHY only supports autoneg.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

The driver currently only assignes 1 RSS context to each VF.  This works
for the Linux VF driver.  But other drivers, such as DPDK, can make use
of additional RSS contexts.  Modify the code to divide up and assign
RSS contexts to VFs just like other resources.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Some chips that need host context memory as a backing store requires
the memory to be initialized to a non-zero value.  Query the
value from firmware and initialize the context memory accordingly.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Driver calls HWRM_FUNC_RESET firmware call while resuming the device
which clears the context memory backing store. Because of which
allocating firmware resources would eventually fail. Fix it by freeing
all context memory during suspend and reallocate the memory during resume.

Call bnxt_hwrm_queue_qportcfg() in resume path.  This firmware call
is needed on the 57500 chips so that firmware will set up the proper
queue mapping in relation to the context memory.
Signed-off-by: NVasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

After driver unregister, firmware is erasing the information that
driver supports new resource management. Send FUNC_RESOURCE_QCAPS
command to inform the firmware that driver supports new resource
management while resuming from hibernation.  Otherwise, we fallback
to the older resource allocation scheme.

Also, move driver register after sending FUNC_RESOURCE_QCAPS command
to be consistent with the normal initialization sequence.
Signed-off-by: NVasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Everytime driver registers with firmware, driver is required to
register for async event notifications as well. These 2 calls
are done using the same firmware command and can be combined.

We are also missing the 2nd step to register for async events
in the suspend/resume path and this will fix it.  Prior to this,
we were getting only default notifications.

ULP can register for additional async events for the RDMA driver,
so we add a parameter to the new function to only do step 2 when
it is called from ULP.
Signed-off-by: NVasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

In the bnxt_init_one() failure path, if the driver has already called
firmware to register the driver, it is not undoing the driver
registration.  Add this missing step to unregister for correctness,
so that the firmware knows that the driver has unloaded.
Signed-off-by: NVasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Disable Bus master during suspend to prevent DMAs after the device
goes into D3hot state.  The new 57500 devices may continue to DMA
from context memory after the system goes into D3hot state.  This
may cause some PCIe errors on some system.  Re-enable it during resume.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Fix BNXT_CHIP_NUM_5645X() to include 57452 and 56454 chip IDs, so
that these chips will be properly classified as P4 chips to take
advantage of the P4 fixes and features.
Signed-off-by: NMichael Chan <michael.chan@broadcom.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

The rfs members of struct efx_channel are under CONFIG_RFS_ACCEL.
Ethtool stats which access those need to be as well.
Reported-by: Nkbuild test robot <lkp@intel.com>
Fixes: ca70bd42 ("sfc: add statistics for ARFS")
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Adjust indentation from spaces to tab (+optional two spaces) as in
coding style with command like:
	$ sed -e 's/^        /\t/' -i */Kconfig
Signed-off-by: NKrzysztof Kozlowski <krzk@kernel.org>
Reviewed-by: NJuergen Gross <jgross@suse.com>
Signed-off-by: NJuergen Gross <jgross@suse.com>

The recording of RSS hash should be controlled by NETIF_F_RXHASH.

Fixes: 1fac7ca4 ("hv_netvsc: record hardware hash in skb")
Suggested-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NStephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: NHaiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: NMichael Kelley <mikelley@microsoft.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

Rename the mac_link_state() method to mac_pcs_get_state() to make it
clear that it should be returning the MACs PCS current state, which
is used for inband negotiation rather than just reading back what the
MAC has been configured for. Update the documentation to explicitly
mention that this is for inband.

We drop the return value as well; most of phylink doesn't check the
return value and it is not clear what it should do on error - instead
arrange for state->link to be false.
Signed-off-by: NRussell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

rhashtable_lookup_fast() internally calls rcu_read_lock() then,
calls rhashtable_lookup(). So if rcu_read_lock() is already held,
rhashtable_lookup() is enough.
Signed-off-by: NTaehee Yoo <ap420073@gmail.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

When a phydev is created, the speed and duplex are set to zero and
-1 respectively, rather than using the predefined SPEED_UNKNOWN and
DUPLEX_UNKNOWN constants.

There is a window at initialisation time where we may report link
down using the 0/-1 values.  Tidy this up and use the predefined
constants, so debug doesn't complain with:

"Unsupported (update phy-core.c)/Unsupported (update phy-core.c)"

when the speed and duplex settings are printed.
Signed-off-by: NRussell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>

There are no users of phy_ethtool_sset() in the kernel anymore, and
as of commit 3c1bcc86 ("net: ethernet: Convert phydev advertize
and supported from u32 to link mode"), the implementation is slightly
buggy - it doesn't correctly check the masked advertising mask as it
used to.

Remove it, and update the phy documentation to refer to its replacement
function.
Signed-off-by: NRussell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>