The __dev* removal patches for the network drivers ended up messing up
the function prototypes for a bunch of drivers.  This patch fixes all of
them back up to be properly aligned.

Bonus is that this almost removes 100 lines of code, always a nice
surprise.
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

CONFIG_HOTPLUG is going away as an option.  As result the __dev*
markings will be going away.

Remove use of __devinit, __devexit_p, __devinitdata, __devinitconst,
and __devexit.
Signed-off-by: NBill Pemberton <wfp5p@virginia.edu>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

Fixed the resource cleanup to act correctly and prevent a kernel oops when
mlx4_QUERY_ADAPTER() fails.
Signed-off-by: NDotan Barak <dotanb@dev.mellanox.co.il>
Reviewed-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

In the current driver, the SENSE_PORT firmware command is issued as a
"wrapped" command, but the command handling code doesn't have a
wrapper, so it will never do anything other than log an error message.
The latest ConnectX-3 2.11.500 firmware reports the SENSE_PORT
capability even in multi-function (SR-IOV) mode, so the driver will
try to issue the command.

At least until the driver has a proper wrapper for SENSE_PORT, make
sure we disable the command for multi-function devices.
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Instead of having a hard-coded "PCI device ID != 0x1003" (which
obviously breaks as newer devices with ID != 0x1003 become available),
instead let's set a flag in our PCI device table for the older devices
where we're supposed to force using SENSE_PORT.  This also avoids
enabling SENSE_PORT for virtual functions by mistake.
Signed-off-by: NRoland Dreier <roland@purestorage.com>

That way we can check flags later on, when we've finished with the
pci_device_id structure.  Also convert the "is VF" flag to an enum:
"Never do in the preprocessor what can be done in C."
Signed-off-by: NRoland Dreier <roland@purestorage.com>

On an SR-IOV master device, __mlx4_init_one() calls mlx4_init_hca()
before mlx4_multi_func_init().  However, for unlucky configurations,
mlx4_init_hca() might call mlx4_SENSE_PORT() (via mlx4_dev_cap()), and
that calls mlx4_cmd_imm() with MLX4_CMD_WRAPPED set.

However, on a multifunction device with MLX4_CMD_WRAPPED, __mlx4_cmd()
calls into mlx4_slave_cmd(), and that immediately tries to do

	down(&priv->cmd.slave_sem);

but priv->cmd.slave_sem isn't initialized until mlx4_multi_func_init()
(which we haven't called yet).  The next thing it tries to do is access
priv->mfunc.vhcr, but that hasn't been allocated yet.

Fix this by moving the initialization of slave_sem and vhcr up into
mlx4_cmd_init(). Also, since slave_sem is really just being used as a
mutex, convert it into a slave_cmd_mutex.
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Also put format string onto one line.
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Previously, the structure of a guest's proxy QPs followed the
structure of the PPF special qps (qp0 port 1, qp0 port 2, qp1 port 1,
qp1 port 2, ...).  The guest then did offset calculations on the
sqp_base qp number that the PPF passed to it in QUERY_FUNC_CAP().

This is now changed so that the guest does no offset calculations
regarding proxy or tunnel QPs to use.  This change frees the PPF from
needing to adhere to a specific order in allocating proxy and tunnel
QPs.

Now QUERY_FUNC_CAP provides each port individually with its proxy
qp0, proxy qp1, tunnel qp0, and tunnel qp1 QP numbers, and these are
used directly where required (with no offset calculations).

To accomplish this change, several fields were added to the phys_caps
structure for use by the PPF and by non-SR-IOV mode:

    base_sqpn -- in non-sriov mode, this was formerly sqp_start.
    base_proxy_sqpn -- the first physical proxy qp number -- used by PPF
    base_tunnel_sqpn -- the first physical tunnel qp number -- used by PPF.

The current code in the PPF still adheres to the previous layout of
sqps, proxy-sqps and tunnel-sqps.  However, the PPF can change this
layout without affecting VF or (paravirtualized) PF code.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

This is necessary in order to support > 1 VF/PF in a VM for software
that uses the node guid as a discriminator, such as librdmacm.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

This requires:

1. Replacing the paravirtualized P_Key index (inserted by the guest)
   with the real P_Key index.

2. For UD QPs, placing the guest's true source GID index in the
   address path structure mgid field, and setting the ud_force_mgid
   bit so that the mgid is taken from the QP context and not from the
   WQE when posting sends.

3. For UC and RC QPs, placing the guest's true source GID index in the
   address path structure mgid field.

4. For tunnel and proxy QPs, setting the Q_Key value reserved for that
   proxy/tunnel pair.

Since not all the above adjustments occur in all the QP transitions,
the QP transitions require separate wrapper functions.

Secondly, initialize the P_Key virtualization table to its default
values: Master virtualized table is 1-1 with the real P_Key table,
guest virtualized table has P_Key index 0 mapped to the real P_Key
index 0, and all the other P_Key indices mapped to the reserved
(invalid) P_Key at index 127.

Finally, add logic in smp_snoop for maintaining the phys_P_Key_cache.
and generating events on the master only if a P_Key actually changed.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

In addition, pass the proxy and tunnel QP numbers to slaves so the
driver can perform special QP paravirtualization.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

If mlx4_cmd_init() failed, the init_one function returned
success, although no resources were opened.
Signed-off-by: NEugenia Emantayev <eugenia@mellanox.com>
Signed-off-by: NYevgeny Petrilin <yevgenyp@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The order of operations was wrong on the teardown flow.
Signed-off-by: NAviad Yehezkel <aviadye@mellanox.com>
Signed-off-by: NYevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Port1=Eth, Port2=IB restriction is no longer required.
Having RoCE, there will always rdma port initialized over ConnectX
physical port, no matter whether the link layer is IB or Ethernet.
So we always have dual port IB device.
Signed-off-by: NYevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently the mlx4 drivers don't have the necessary callbacks to
implement EEH errors detection and recovery, so the PCI layer uses the
probe and remove callbacks to try to recover the device after an error on
the bus. However, these callbacks have race conditions with the internal
catastrophic error recovery functions, which will also detect the error
and this can cause the system to crash if both EEH and catas functions
try to reset the device.

This patch adds the necessary error recovery callbacks and makes sure
that the internal catastrophic error functions will not try to reset the
device in such scenarios. It also adds some calls to
pci_channel_offline() to suppress reads/writes on the bus when the slot
cannot accept I/O operations so we prevent unnecessary accesses to the
bus and speed up the device removal.
Signed-off-by: NKleber Sacilotto de Souza <klebers@linux.vnet.ibm.com>
Acked-by: NShlomo Pongratz <shlomop@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

To allow easy paravirtualization of P_Key and GID table sizes, keep
paravirtualized sizes in mlx4_dev->caps, but save the actual physical
sizes from FW in struct: mlx4_dev->phys_cap.

In addition, in SR-IOV mode, do the following:

1. Reduce reported P_Key table size by 1.
   This is done to reserve the highest P_Key index for internal use,
   for declaring an invalid P_Key in P_Key paravirtualization.
   We require a P_Key index which always contain an invalid P_Key
   value for this purpose (i.e., one which cannot be modified by
   the subnet manager).  The way to do this is to reduce the
   P_Key table size reported to the subnet manager by 1, so that
   it will not attempt to access the P_Key at index #127.

2. Paravirtualize the GID table size to 1. Thus, each guest sees
   only a single GID (at its paravirtualized index 0).

In addition, since we are paravirtualizing the GID table size to 1, we
add paravirtualization of the master GID event here (i.e., we do not
do ib_dispatch_event() for the GUID change event on the master, since
its (only) GUID never changes).
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Modify mlx4_dev_cap to allow IB support when SR-IOV is active.  Modify
mlx4_slave_cap to set the "rdma-supported" bit in its flags area, and
pass that to the guests (this is done in QUERY_FUNC_CAP and its
wrapper).

However, we don't activate IB support quite yet -- we leave the error
return at the start of mlx4_ib_add in the mlx4_ib driver.

In addition, set "protected fmr supported" bit to zero in the
QUERY_FUNC_CAP wrapper.

Finally, in the QUERY_FUNC_CAP wrapper, we needed to add code which
checks for the port type (IB or Ethernet).  Previously, this was not
an issue, since only Ethernet ports were supported.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

The SR-IOV special QP tunneling mechanism uses proxy special QPs
(instead of the real special QPs) for MADs on guests.  These proxy QPs
send their packets to a "tunnel" QP owned by the master.  The master
then forwards the MAD (after any required paravirtualization) to the
real special QP, which sends out the MAD.

For security reasons (i.e., to prevent guests from sending MADs to
tunnel QPs belonging to other guests), each proxy-tunnel QP pair is
assigned a unique, reserved, Q_Key.  These Q_Keys are available only
for proxy and tunnel QPs -- if the guest tries to use these Q_Keys
with other QPs, it will fail.

This patch introduces a mechanism for reserving a block of 64K Q_Keys
for proxy/tunneling use.

The patch introduces also two new fields into mlx4_dev: base_sqpn and
base_tunnel_sqpn.

In SR-IOV mode, the QP numbers for the "real," proxy, and tunnel sqps
are added to the reserved QPN area (so that they will not change).
There are 8 special QPs per port in the HCA, and each of them is
assigned both a proxy and a tunnel QP, for each VF and for the PF as
well in SR-IOV mode.

The QPNs for these QPs are arranged as follows:
 1. The real SQP numbers (8)
 2. The proxy SQPs (8 * (max number of VFs + max number of PFs)
 3. The tunnel SQPs (8 * (max number of VFs + max number of PFs)

To support these QPs, two new fields are added to struct mlx4_dev:

  base_sqp:  this is the QP number of the first of the real SQPs
  base_tunnel_sqp: this is the qp number of the first qp in the tunnel
                   sqp region. (On guests, this is the first tunnel
                   sqp of the 8 which are assigned to that guest).

In addition, in SR-IOV mode, sqp_start is the number of the first
proxy SQP in the proxy SQP region.  (In guests, this is the first
proxy SQP of the 8 which are assigned to that guest)

Note that in non-SR-IOV mode, there are no proxies and no tunnels.
In this case, sqp_start is set to sqp_base -- which minimizes code
changes.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

With IB SR-IOV, each slave has its own separate copy of the port
capabilities flags.  For example, the master can run a subnet manager
(which causes the IsSM bit to be set in the master's port
capabilities) without affecting the port capabilities seen by the
slaves (the IsSM bit will be seen as cleared in the slaves).

Also add a static inline mlx4_master_func_num() to enhance readability
of the code.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

The driver is modified to support three operation modes.

If supported by firmware use the device managed flow steering
API, that which we call device managed steering mode. Else, if
the firmware supports the B0 steering mode use it, and finally,
if none of the above, use the A0 steering mode.

When the steering mode is device managed, the code is modified
such that L2 based rules set by the mlx4_en driver for Ethernet
unicast and multicast, and the IB stack multicast attach calls
done through the mlx4_ib driver are all routed to use the device
managed API.

When attaching rule using device managed flow steering API,
the firmware returns a 64 bit registration id, which is to be
provided during detach.

Currently the firmware is always programmed during HCA initialization
to use standard L2 hashing. Future work should be done to allow
configuring the flow-steering hash function with common, non
proprietary means.
Signed-off-by: NHadar Hen Zion <hadarh@mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Instead of checking the firmware supported steering mode in various
places in the code, add a dedicated field in the mlx4 device capabilities
structure which is written once during the initialization flow and read
across the code.

This also set the grounds for add new steering modes. Currently two modes
are supported, and are named after the ConnectX HW versions A0 and B0.

A0 steering uses mac_index, vlan_index and priority to steer traffic
into pre-defined range of QPs.

B0 steering uses Ethernet L2 hashing rules and is enabled only
if the firmware supports both unicast and multicast B0 steering,

The current steering modes are relevant for Ethernet traffic only,
such that Infiniband steering remains untouched.
Signed-off-by: NHadar Hen Zion <hadarh@mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NYuval Mintz <yuvalmin@broadcom.com>
Signed-off-by: NEilon Greenstein <eilong@broadcom.com>

Cc: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix a crash at the error flow of NOP command which caused the driver to try and use
a completion vector which wasn't allocated.
Signed-off-by: NYevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The range check was performed after using the port number.

Reverse this to prevent a potential array overflow.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Reviewed-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

- pass the following parameters:
  - firmware version (added QUERY_FW paravirtualization for that)

  - disable Blueflame on slaves. KVM disables write combining on guests,
    and we get better performance without BF in this case. (This requires
    QUERY_DEV_CAP paravirtualization, also in this commit)

  - max qp rdma as destination

- get rid of a chunk of "if (0)" dead code
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Reviewed-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In SRIOV mode, the number of EQs used when computing the total ICM size
was incorrect.

To fix this, we do the following:
1. We add a new structure to mlx4_dev, mlx4_phys_caps, to contain physical HCA
   capabilities.  The PPF uses the phys capabilities when it computes things
   like ICM size.

   The dev_caps structure will then contain the paravirtualized values, making
   bookkeeping much easier in SRIOV mode. We add a structure rather than a
   single parameter because there will be other fields in the phys_caps.

   The first field we add to the mlx4_phys_caps structure is num_phys_eqs.

2. In INIT_HCA, when running in SRIOV mode, the "log_num_eqs" parameter
   passed to the FW is the number of EQs per VF/PF; each function (PF or VF)
   has this number of EQs available.

   However, the total number of EQs which must be allowed for in the ICM is
   (1 << log_num_eqs) * (#VFs + #PFs).  Rather than compute this quantity,
   we allocate ICM space for 1024 EQs (which is the device maximum
   number of EQs, and which is the value we place in the mlx4_phys_caps structure).

   For INIT_HCA, however, we use the per-function number of EQs as described
   above.
Signed-off-by: NMarcel Apfelbaum <marcela@dev.mellanox.co.il>
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Reviewed-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add missing resource tracking for XRC domains and complete the tracking for HCA
network flow counters.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently the slave and master resources are deleted after master freed
all bitmaps. If any resources were not properly cleaned up during the
shutdown process, an Oops would result.

Fix so that delete slave (only) resources during cleanup. Master resources
are cleaned up during unload process, and need not separately be cleaned.

Note that during cleanup, we need to split the resource-tracker freeing
functionality.

Before removing all the bitmaps, we free any leftover slave resources.
However, we can only remove the resource tracker linked list after
all bitmap frees, since some of the freeing functions (e.g.,
mlx4_cleanup_eq_table) use paravirtualized FW commands which expect
the resource tracker linked list to be present.
Found-by: NAviad Yehezkel <aviadye@mellanox.com>
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Consider the following scenario: 2 HCAs, where only one of which can run SRIOV.

If we reset the module parameter, all the VFs of the SRIOV HCA will be
claimed by the PPF host (-- the code relies on num_vfs being non-zero
to avoid this claiming, and num_vfs was reset when pci_enable_sriov failed
for the non-SRIOV HCA).

The solution is not to touch the num_vfs parameter.

Also, eliminate the unneeded check of num_vfs when disabling sriov
(the dev flag bit is sufficient).
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds a 64-bit flags2 features member to struct mlx4_dev to
export further features of the hardware.  The original flags field
tracks features whose support bits are advertised by the firmware in
offsets 0x40 and 0x44 of the query device capabilities command.
flags2 will track features whose support bits are scattered at various
offsets.

RSS support is the first feature to be exported through flags2.  RSS
capabilities are located at offset 0x2e.  The size of the RSS
indirection table is also given in this offset.
Signed-off-by: NShlomo Pongratz <shlomop@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Set the MTU for IB ports in the driver instead of using the firmware
default of 2KB (the driver defaults to 4KB).  Allow for dynamic mtu
configuration through a new, per-port sysfs entry.

Since there's a dependency between the port MTU and the max number of
HW VLs the port can support, apply a mim/max approach, using a loop
that goes down from the highest possible number of VLs to the lowest,
using the firmware return status to know whether the requested number
of VLs is possible with a given MTU.

For now, as with the dynamic link type change / VPI support, the sysfs
entry to change the mtu is exposed only when NOT running in SR-IOV
mode.  To allow changing the MTU for the master in SR-IOV mode,
primary-function-initiated FLR (Function Level Reset) needs to be
implemented.
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Commit 22c8bff6 ("mlx4_core: Exported functions can't be static")
fixed most of this up, but forgot about mlx4_is_slave_active().  Fix
this one too.

Cc: <stable@vger.kernel.org>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

While doing the work for commit a6f7feae ("IB/mlx4: pass SMP
vendor-specific attribute MADs to firmware") we realized that the
firmware would respond on all sorts of vendor-specific MADs.
Therefore commit 97285b78 ("mlx4_core: Add extended port
capabilities support") adds redundant code into the driver, since
there's no real reaon to maintain the extended capabilities of the
port, as they can be queried on demand (e.g the FDR10 capability).

This patch reverts commit 97285b78 and removes the check for
extended caps from the mlx4_ib driver port query flow.
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

ConnectX devices have a limit on the number of mappings that can be
done on an FMR before having to call sync_tpt.  The current
mlx4_ib driver reports the limit correctly in max_map_per_fmr in
.query_device(), but mlx4_core doesn't check it when actually
allocating FMRs.

Add a max_fmr_maps field to struct mlx4_caps and enforce this maximum
value on FMR allocations.
Signed-off-by: NEli Cohen <eli@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Signed-off-by: NYevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In port type change flow, need to set the new port types only after
all interfaces have finished the unregister process.
Otherwise, during unregister, one of the interfaces might issue a SET_PORT
command with wrong port types, it can cause bad FW behavior.
Signed-off-by: NYevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Under the spinlock we call request_irq(), which allocates memory with GFP_KERNEL,
This causes the following trace when DEBUG_SPINLOCK is enabled, it can cause
the following trace:

 BUG: spinlock wrong CPU on CPU#2, ethtool/2595
 lock: ffff8801f9cbc2b0, .magic: dead4ead, .owner: ethtool/2595, .owner_cpu: 0
 Pid: 2595, comm: ethtool Not tainted 3.0.18 #2
 Call Trace:
 spin_bug+0xa2/0xf0
 do_raw_spin_unlock+0x71/0xa0
 _raw_spin_unlock+0xe/0x10
 mlx4_assign_eq+0x12b/0x190 [mlx4_core]
 mlx4_en_activate_cq+0x252/0x2d0 [mlx4_en]
 ? mlx4_en_activate_rx_rings+0x227/0x370 [mlx4_en]
 mlx4_en_start_port+0x189/0xb90 [mlx4_en]
 mlx4_en_set_ringparam+0x29a/0x340 [mlx4_en]
 dev_ethtool+0x816/0xb10
 ? dev_get_by_name_rcu+0xa4/0xe0
 dev_ioctl+0x2b5/0x470
 handle_mm_fault+0x1cd/0x2d0
 sock_do_ioctl+0x5d/0x70
 sock_ioctl+0x79/0x2f0
 do_vfs_ioctl+0x8c/0x340
 sys_ioctl+0xa1/0xb0
 system_call_fastpath+0x16/0x1b

Replacing with mutex, which is enough in this case.
Signed-off-by: NYevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

BF can be disabled in some cases, the capability field, bf_reg_size is set
to zero in this case. Don't map the BF area in this case, it would cause
failures.  In addition, leaving the BF area unmapped
also alerts the ETH driver to not use BF.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>