NIX hardware context structure got changed to accommodate new
features like bandwidth steering, L3/L4 outer/inner checksum
enable/disable etc., on CN10K platform.
This patch defines new mbox message NIX_CN10K_AQ_INST for new
NIX context initialization.

This patch also updates the NPA context structures to accommodate
bit field changes made for CN10K platform.

This patch also removes Big endian bit fields from existing
structures as its support got deprecated in current and upcoming silicons.
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Firmware allocates memory regions for PFs and VFs in DRAM.
The PFs memory region is used for AF-PF and PF-VF mailbox.
This mbox facilitate communication between AF-PF and PF-VF.

On CN10K platform:
The DRAM region allocated to PF is enumerated as PF BAR4 memory.
PF BAR4 contains AF-PF mbox region followed by its VFs mbox region.
AF-PF mbox region base address is configured at RVU_AF_PFX_BAR4_ADDR
PF-VF mailbox base address is configured at
RVU_PF(x)_VF_MBOX_ADDR = RVU_AF_PF()_BAR4_ADDR+64KB. PF access its
mbox region via BAR4, whereas VF accesses PF-VF DRAM mailboxes via
BAR2 indirect access.

On CN9XX platform:
Mailbox region in DRAM is divided into two parts AF-PF mbox region and
PF-VF mbox region i.e all PFs mbox region is contiguous similarly all
VFs.
The base address of the AF-PF mbox region is configured at
RVU_AF_PF_BAR4_ADDR.
AF-PF1 mbox address can be calculated as RVU_AF_PF_BAR4_ADDR * mbox
size.
The base address of PF-VF mbox region for each PF is configure at
RVU_AF_PF(0..15)_VF_BAR4_ADDR.PF access its mbox region via BAR4 and its
VF mbox regions from RVU_PF_VF_BAR4_ADDR register, whereas VF access its
mbox region via BAR4.

This patch changes mbox initialization to support both CN9XX and CN10K
platform.
The patch also adds new hw_cap flag to setting hw features like TSO etc
and removes platform specific name from the PF/VF driver name to make it
appropriate for all supported platforms

This patch also removes platform specific name from the PF/VF driver name
to make it appropriate for all supported platforms
Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Firmware allocates memory regions for PFs and VFs in DRAM.
The PFs memory region is used for AF-PF and PF-VF mailbox.
This mbox facilitates communication between AF-PF and PF-VF.

On CN10K platform:
The DRAM region allocated to PF is enumerated as PF BAR4 memory.
PF BAR4 contains AF-PF mbox region followed by its VFs mbox region.
AF-PF mbox region base address is configured at RVU_AF_PFX_BAR4_ADDR
PF-VF mailbox base address is configured at
RVU_PF(x)_VF_MBOX_ADDR = RVU_AF_PF()_BAR4_ADDR+64KB. PF access its
mbox region via BAR4, whereas VF accesses PF-VF DRAM mailboxes via
BAR2 indirect access.

On CN9XX platform:
Mailbox region in DRAM is divided into two parts AF-PF mbox region and
PF-VF mbox region i.e all PFs mbox region is contiguous similarly all
VFs.
The base address of the AF-PF mbox region is configured at
RVU_AF_PF_BAR4_ADDR.
AF-PF1 mbox address can be calculated as RVU_AF_PF_BAR4_ADDR * mbox
size.
The base address of PF-VF mbox region for each PF is configure at
RVU_AF_PF(0..15)_VF_BAR4_ADDR.PF access its mbox region via BAR4 and its
VF mbox regions from RVU_PF_VF_BAR4_ADDR register, whereas VF access its
mbox region via BAR4.

This patch changes mbox initialization to support both CN9XX and CN10K
platform.

This patch also adds CN10K PTP subsystem and device IDs to ptp
driver id table.
Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com>
Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Stefan Chulski says:

====================
net: mvpp2: Add TX Flow Control support

Armada hardware has a pause generation mechanism in GOP (MAC).
The GOP generate flow control frames based on an indication programmed in Ports Control 0 Register. There is a bit per port.
However assertion of the PortX Pause bits in the ports control 0 register only sends a one time pause.
To complement the function the GOP has a mechanism to periodically send pause control messages based on periodic counters.
This mechanism ensures that the pause is effective as long as the Appropriate PortX Pause is asserted.

Problem is that Packet Processor that actually can drop packets due to lack of resources not connected to the GOP flow control generation mechanism.
To solve this issue Armada has firmware running on CM3 CPU dedicated for Flow Control support.
Firmware monitors Packet Processor resources and asserts XON/XOFF by writing to Ports Control 0 Register.

MSS shared SRAM memory used to communicate between CM3 firmware and PP2 driver.
During init PP2 driver informs firmware about used BM pools, RXQs, congestion and depletion thresholds.

The pause frames are generated whenever congestion or depletion in resources is detected.
The back pressure is stopped when the resource reaches a sufficient level.
So the congestion/depletion and sufficient level implement a hysteresis that reduces the XON/XOFF toggle frequency.

Packet Processor v23 hardware introduces support for RX FIFO fill level monitor.
Patch "add PPv23 version definition" to differ between v23 and v22 hardware.
Patch "add TX FC firmware check" verifies that CM3 firmware supports Flow Control monitoring.

v12 --> v13
- Remove bm_underrun_protect module_param

v11 --> v12
- Improve warning message in "net: mvpp2: add TX FC firmware check" patch

v10 --> v11
- Improve "net: mvpp2: add CM3 SRAM memory map" comment
- Move condition check to 'net: mvpp2: always compare hw-version vs MVPP21' patch

v9 --> v10
- Add CM3 SRAM description to PPv2 documentation

v8 --> v9
- Replace generic pool allocation with devm_ioremap_resource

v7 --> v8
- Reorder "always compare hw-version vs MVPP21" and "add PPv23 version definition" commits
- Typo fixes
- Remove condition fix from "add RXQ flow control configurations"

v6 --> v7
- Reduce patch set from 18 to 15 patches
 - Documentation change combined into a single patch
 - RXQ and BM size change combined into a single patch
 - Ring size change check moved into "add RXQ flow control configurations" commit

v5 --> v6
- No change

v4 --> v5
- Add missed Signed-off
- Fix warnings in patches 3 and 12
- Add revision requirement to warning message
- Move mss_spinlock into RXQ flow control configurations patch
- Improve FCA RXQ non occupied descriptor threshold commit message

v3 --> v4
- Remove RFC tag

v2 --> v3
- Remove inline functions
- Add PPv2.3 description into marvell-pp2.txt
- Improve mvpp2_interrupts_mask/unmask procedure
- Improve FC enable/disable procedure
- Add priv->sram_pool check
- Remove gen_pool_destroy call
- Reduce Flow Control timer to x100 faster

v1 --> v2
- Add memory requirements information
- Add EPROBE_DEFER if of_gen_pool_get return NULL
- Move Flow control configuration to mvpp2_mac_link_up callback
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Patch check that TX FC firmware is running in CM3.
If not, global TX FC would be disabled.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch fix GMAC TX flow control autoneg.
Flow control autoneg wrongly were disabled with enabled TX
flow control.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

New FIFO flow control feature was added in PPv23.
PPv2 FIFO polled by HW and trigger pause frame if FIFO
fill level is below threshold.
FIFO HW flow control enabled with CM3 RXQ&BM flow
control with ethtool.
Current  FIFO thresholds is:
9KB for port with maximum speed 10Gb/s port
4KB for port with maximum speed 5Gb/s port
2KB for port with maximum speed 1Gb/s port
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The PP2v23 hardware supports a feature allowing to double the
size of BPPI by decreasing number of pools from 16 to 8.
Increasing of BPPI size protect BM drop from BPPI underrun.
Underrun could occurred due to stress on DDR and as result slow buffer
transition from BPPE to BPPI.
New BPPI threshold recommended by spec is:
BPPI low threshold - 640 buffers
BPPI high threshold - 832 buffers
Supported only in PPv23.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch add ethtool flow control configuration support.

Tx flow control retrieved correctly by ethtool get function.
FW per port ethtool configuration capability added.

Patch also takes care about mtu change procedure, if PPv2 switch
BM pools during mtu change.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds RXQ flow control configurations.
Flow control disabled by default.
Minimum ring size limited to 1024 descriptors.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch enables global flow control in FW and in the phylink validate mask.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The firmware needs to monitor the RX Non-occupied descriptor
bits for flow control to move to XOFF mode.
These bits need to be unmasked to be functional, but they will
not raise interrupts as we leave the RX exception summary
bit in MVPP2_ISR_RX_TX_MASK_REG clear.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Flow Control periodic timer would be used if port in
XOFF to transmit periodic XOFF frames.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

BM pool and RXQ size increased to support Firmware Flow Control.
Minimum depletion thresholds to support FC are 1024 buffers.
BM pool size increased to 2048 to have some 1024 buffers
space between depletion thresholds and BM pool size.

Jumbo frames require a 9888B buffer, so memory requirements
for data buffers increased from 7MB to 24MB.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Reviewed-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch add PPv23 version definition.
PPv23 is new packet processor in CP115.
Everything that supported by PPv22, also supported by PPv23.
No functional changes in this stage.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Reviewed-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently we have PP2v1 and PP2v2 hw-versions, with some different
handlers depending upon condition hw_version = MVPP21/MVPP22.
In a future there will be also PP2v3. Let's use now the generic
"if equal/notEqual MVPP21" for all cases instead of "if MVPP22".

This patch does not change any functionality.
It is not intended to introduce PP2v3.
It just modifies MVPP21/MVPP22 check-condition
bringing it to generic and unified form correct for new-code
introducing and PP2v3 net-next generation.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds CM3 memory map.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

CM3 SRAM address space will be used for Flow Control configuration.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Signed-off-by: Konstantin Porotchkin <kostap@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Patch adds CM3 address space and PPv2.3 description.
Signed-off-by: Stefan Chulski <stefanc@marvell.com>
Acked-by: Marcin Wojtas <mw@semihalf.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In order to support the possibility of per-device event channel
settings (e.g. lateeoi spurious event thresholds) add a xenbus device
pointer to struct irq_info() and modify the related event channel
binding interfaces to take the pointer to the xenbus device as a
parameter instead of the domain id of the other side.

While at it remove the stale prototype of bind_evtchn_to_irq_lateeoi().
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: Wei Liu <wei.liu@kernel.org>
Reviewed-by: Paul Durrant <paul@xen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

In case of a common event for rx and tx queue the event should be
regarded to be spurious if no rx and no tx requests are pending.

Unfortunately the condition for testing that is wrong causing to
decide a event being spurious if no rx OR no tx requests are
pending.

Fix that plus using local variables for rx/tx pending indicators in
order to split function calls and if condition.

Fixes: 23025393 ("xen/netback: use lateeoi irq binding")
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Reviewed-by: Wei Liu <wl@xen.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The function fib6_walk_continue() cannot return a positive value when
called from register_fib_notifier(), but ignoring causes static analyzer to
generate warnings in users of register_fib_notifier() that try to convert
returned error code to pointer with ERR_PTR(). Handle such case by
explicitly checking for positive error values and converting them to
-EINVAL in fib6_tables_dump().
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Suggested-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'mlx5-for-upstream-2021-02-10' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-for-upstream-2021-02-10

Misc cleanups and trivial fixes for net-next

1) spelling mistakes
2) error path checks fixes
3) unused includes and struct fields cleanup
4) build error when MLX5_ESWITCH=no
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Due to the fact that ic_dev->dev is kept open in ic_close_dev, I had
thought that ic_dev will not be freed either. But that is not the case,
but instead "everybody dies" when ipconfig cleans up, and just the
net_device behind ic_dev->dev remains allocated but not ic_dev itself.

This is a problem because in ic_close_devs, for every net device that
we're about to close, we compare it against the list of lower interfaces
of ic_dev, to figure out whether we should close it or not. But since
ic_dev itself is subject to freeing, this means that at some point in
the middle of the list of ipconfig interfaces, ic_dev will have been
freed, and we would be still attempting to iterate through its list of
lower interfaces while checking whether to bring down the remaining
ipconfig interfaces.

There are multiple ways to avoid the use-after-free: we could delay
freeing ic_dev until the very end (outside the while loop). Or an even
simpler one: we can observe that we don't need ic_dev when iterating
through its lowers, only ic_dev->dev, structure which isn't ever freed.
So, by keeping ic_dev->dev in a variable assigned prior to freeing
ic_dev, we can avoid all use-after-free issues.

Fixes: 46acf7bd ("Revert "net: ipv4: handle DSA enabled master network devices"")
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Several years ago these two entries have been added, but it's not clear
why. There's no trace that there has ever been such a chip version, and
not even the r8101 vendor driver knows these id's. So let's disable
detection, and if nobody complains remove them completely later.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Nikolay Aleksandrov says:

====================
bonding: 3ad: support for 200G/400G ports and more verbose warning
xk
We'd like to have proper 200G and 400G support with 3ad bond mode, so we
need to add new definitions for them in order to have separate oper keys,
aggregated bandwidth and proper operation (patches 01 and 02). In
patch 03 Ido changes the code to use pr_err_once instead of
pr_warn_once which would help future detection of unsupported speeds.

v2: patch 03: use pr_err_once instead of WARN_ONCE
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The bond driver needs to be patched to support new ethtool speeds.
Currently it emits a single warning [1] when it encounters an unknown
speed. As evident by the two previous patches, this is not explicit
enough. Instead, promote it to an error.

[1]
bond10: (slave swp1): unknown ethtool speed (200000) for port 1 (set it to 0)

v2:
* Use pr_err_once() instead of WARN_ONCE()
Signed-off-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In order to be able to use 3ad mode with 400G devices we need to extend
the supported speeds.
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In order to be able to use 3ad mode with 200G devices we need to extend
the supported speeds.
Signed-off-by: Nikolay Aleksandrov <nikolay@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Bhaskar Upadhaya says:

====================
qede: add netpoll and per-queue coalesce support

This is a followup implementation after series

https://patchwork.kernel.org/project/netdevbpf/cover/1610701570-29496-1-git-send-email-bupadhaya@marvell.com/

Patch 1: Add net poll controller support to transmit kernel printks
         over UDP
Patch 2: QLogic card support multiple queues and each queue can be
         configured with respective coalescing parameters, this patch
         add per queue rx-usecs, tx-usecs coalescing parameters
Patch 3: set default per queue rx-usecs, tx-usecs coalescing parameters and
         preserve coalesce parameters across interface up and down

v3: fixed warnings reported by Dan Carpenter
v2: comments from jakub
 - p1: remove poll_controller ndo and add budget 0 support in qede_poll
 - p3: preserve coalesce parameters across interface up and down
===================
Signed-off-by: David S. Miller <davem@davemloft.net>

Here we do the initialization of coalescing values on load.
per queue coalesce values are also restored across up/down of
ethernet interface.
Signed-off-by: Bhaskar Upadhaya <bupadhaya@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

per queue coalescing allows better and more finegrained control
over interrupt rates.
Signed-off-by: Bhaskar Upadhaya <bupadhaya@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

handle netpoll case when qede_poll is called by
netpoll layer with budget 0
Signed-off-by: Bhaskar Upadhaya <bupadhaya@marvell.com>
Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Currently, a random stack value is being returned because variable
_ret_ is not properly initialized. This variable is actually not
used anymore and it should be removed.

Fix this by removing all instances of variable ret and return 0.

Fixes: 64749c9c ("net: hns3: remove redundant return value of hns3_uninit_all_ring()")
Addresses-Coverity-ID: 1501700 ("Uninitialized scalar variable")
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

plat_dat is initialized by stmmac_probe_config_dt().
So, initialization is not required by priv->plat.
This removes unnecessary initialization and variables.
Signed-off-by: Nobuhiro Iwamatsu <nobuhiro1.iwamatsu@toshiba.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>

It is simpler to make net->net_cookie a plain u64
written once in setup_net() instead of looping
and using atomic64 helpers.

Lorenz Bauer wants to add SO_NETNS_COOKIE socket option
and this patch would makes his patch series simpler.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Lorenz Bauer <lmb@cloudflare.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

So far we don't re-configure WOL-related register bits when waking up
from hibernation. I'm not aware of any problem reports, but better
play safe and call __rtl8169_set_wol() in the resume() path too.
To achieve this move calling __rtl8169_set_wol() to
rtl8169_net_resume() and rename the function to rtl8169_runtime_resume().
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Michael Walle says:

====================
net: phy: icplus: cleanups and new features

Cleanup the PHY drivers for IPplus devices and add PHY counters and MDIX
support for the IP101A/G.

Patch 5 adds a model detection based on the behavior of the PHY.
Unfortunately, the IP101A shares the PHY ID with the IP101G. But the latter
provides more features. Try to detect the newer model by accessing the page
selection register. If it is writeable, it is assumed, that it is a IP101G.

With this detection in place, we can now access registers >= 16 in a
correct way on the IP101G; that is by first selecting the correct page.
This might previouly worked, because no one ever set another active page
before booting linux.

The last two patches add the new features.
===================
Signed-off-by: David S. Miller <davem@davemloft.net>

Implement the operations to set desired mode and retrieve the current
mode.

This feature was tested with an IP101G.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

The IP101G provides three counters: RX packets, CRC errors and symbol
errors. The error counters can be configured to clear automatically on
read. Unfortunately, this isn't true for the RX packet counter. Because
of this and because the RX packet counter is more likely to overflow,
than the error counters implement only support for the error counters.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>