Move fib6_nh to the end of fib6_info and make it an array of
size 0. Pass a flag to fib6_info_alloc indicating if the
allocation needs to add space for a fib6_nh.

The current code path always has a fib6_nh allocated with a
fib6_info; with nexthop objects they will be separate.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Similar to the pcpu routes exceptions are really per nexthop, so move
rt6i_exception_bucket from fib6_info to fib6_nh.

To avoid additional increases to the size of fib6_nh for a 1-bit flag,
use the lowest bit in the allocated memory pointer for the flushed flag.
Add helpers for retrieving the bucket pointer to mask off the flag.

The cleanup of the exception bucket is moved to fib6_nh_release.

fib6_nh_flush_exceptions can now be called from 2 contexts:
1. deleting a fib entry
2. deleting a fib6_nh

For 1., fib6_nh_flush_exceptions is called for a specific fib6_info that
is getting deleted. All exceptions in the cache using the entry are
deleted. For 2, the fib6_nh itself is getting destroyed so
fib6_nh_flush_exceptions is called for a NULL fib6_info which means
flush all entries.

The pmtu.sh selftest exercises the affected code paths - from creating
exceptions to cleaning them up on device delete. All tests pass without
any rcu locking or memleak warnings.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Before moving exception bucket from fib6_info to fib6_nh, refactor
rt6_flush_exceptions, rt6_remove_exception_rt, rt6_mtu_change_route,
and rt6_update_exception_stamp_rt. In all 3 cases, move the primary
logic into a new helper that starts with fib6_nh_. The latter 3
functions still take a fib6_info; this will be changed to fib6_nh
in the next patch.

In the case of rt6_mtu_change_route, move the fib6_metric_locked
out as a standalone check - no need to call the new function if
the fib entry has the mtu locked. Also, add fib6_info to
rt6_mtu_change_arg as a way of passing the fib entry to the new
helper.

No functional change intended. The goal here is to make the next
patch easier to review by moving existing lookup logic for each to
new helpers.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Move the existing pcpu walk in fib6_drop_pcpu_from to a new
helper, __fib6_drop_pcpu_from, that can be invoked per fib6_nh with a
reference to the from entries that need to be evicted. If the passed
in 'from' is non-NULL then only entries associated with that fib6_info
are removed (e.g., case where fib entry is deleted); if the 'from' is
NULL are entries are flushed (e.g., fib6_nh is deleted).

For fib6_info entries with builtin fib6_nh (ie., current code) there
is no change in behavior.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

rt6_info are specific instances of a fib entry and are tied to a
device and gateway - ie., a nexthop. Before nexthop objects, IPv6 fib
entries have separate fib6_info for each nexthop in a multipath route,
so the location of the pcpu cache in the fib6_info struct worked.
However, with nexthop objects a fib6_info can point to a set of nexthops
(yet another alignment of ipv6 with ipv4). Accordingly, the pcpu
cache needs to be moved to the fib6_nh struct so the cached entries
are local to the nexthop specification used to create the rt6_info.

Initialization and free of the pcpu entries moved to fib6_nh_init and
fib6_nh_release.

Change in location only, from fib6_info down to fib6_nh; no other
functional change intended.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Y.b. Lu says:

====================
ENETC: support hardware timestamping

This patch-set is to support hardware timestamping for ENETC
and also to add ENETC 1588 timer device tree node for ls1028a.

Because the ENETC RX BD ring dynamic allocation has not been
supported and it is too expensive to use extended RX BDs
if timestamping is not used, a Kconfig option is used to
enable extended RX BDs in order to support hardware
timestamping. This option will be removed once RX BD
ring dynamic allocation is implemented.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add ENETC 1588 timer node which is ENETC PF 4 (Physiscal Function 4).
Signed-off-by: NYangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add a new compatible for ENETC PTP.
Signed-off-by: NYangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch is to add get_ts_info interface for ethtool
to support getting timestamping capability.
Signed-off-by: NYangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch is to add hardware timestamping support
for ENETC. On Rx, timestamping is enabled for all
frames. On Tx, we only instruct the hardware to
timestamp the frames marked accordingly by the stack.

Because the RX BD ring dynamic allocation has not been
supported and it is too expensive to use extended RX BDs
if timestamping is not used, a Kconfig option is used to
enable extended RX BDs in order to support hardware
timestamping. This option will be removed once RX BD
ring dynamic allocation is implemented.
Signed-off-by: NYangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: NClaudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixes: 1b3fa5cf ("net: ll_temac: Cleanup multicast filter on change")
Signed-off-by: NEsben Haabendal <esben@geanix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Jeff Kirsher says:

====================
100GbE Intel Wired LAN Driver Updates 2019-05-23

This series contains updates to ice driver only.

Anirudh cleans up white space issues and other code formatting issues in the
driver.  Also implemented LLDP persistence across reboots and start/stop of the
LLDP agent.  Updated print statements for driver capabilities to include
if it is a device or function capability.

Bruce cleaned up variable declarations by removing unneeded assignment.

Dave fixes a potential hang due to a couple of flows that recursively
acquire the RTNL lock which results in a deadlock.

Tony updates the driver to advertise what link modes we are capable of
when the user does not request a specific link mode.

Usha fixes up the LLDP MIB change event handling by cleaning up
workarounds and print the DCB configuration changes detected.

Brett fixes the driver to handle failures in the VF reset path, which
was failing to free resources upon an error.

Richard fixed the reported of stats via ethtool to align with our other
Intel drivers.

Jesse optimizes the transmit buffer and ring structures to have more
efficient ordering to get hot cache lines to have packed data.  Also
optimized the VF structure to use less memory, since it is used hundreds
of times throughout the driver.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Recent versions of sparse warn about casting pointers to/from restricted
endian types in the Linux driver.  Silence those with the compiler
attribute __force macro from the Linux kernel to force casts to/from
restricted endian types.
Signed-off-by: NBruce Allan <bruce.w.allan@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Currently the driver is calling ice_napi_del() and then
unregister_netdev(). The call to unregister_netdev() will result in a
call to ice_stop() and then ice_vsi_close(). This is where we call
napi_disable() for all the MSI-X vectors. This flow is reversed so make
the changes to ensure napi_disable() happens prior to napi_del().

Before calling napi_del() and free_netdev() make sure
unregister_netdev() was called. This is done by making sure the
__ICE_DOWN bit is set in the vsi->state for the interested VSI.
Signed-off-by: NBrett Creeley <brett.creeley@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

The ice_vf struct can be used hundreds of times in our
driver so it pays to use less memory per struct.

ice_vf prior to this commit:
  /* size: 112, cachelines: 2, members: 25 */
  /* sum members: 101, holes: 4, sum holes: 8 */
  /* bit holes: 2, sum bit holes: 11 bits */
  /* padding: 3 */
  /* last cacheline: 48 bytes */

ice_vf after this commit:
  /* size: 104, cachelines: 2, members: 25 */
  /* sum members: 100, holes: 3, sum holes: 4 */
  /* bit holes: 1, sum bit holes: 3 bits */
  /* last cacheline: 40 bytes */
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

We can use bit fields to store boolean values and when the
bit fields are next to each other, the compiler will combine them
(as long as the size holds enough).
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Use more efficient structure ordering by using the pahole tool
and a lot of code inspection to get hot cache lines to have
packed data (no holes if possible) and adjacent warm data.

ice_ring prior to this change:
  /* size: 192, cachelines: 3, members: 23 */
  /* sum members: 158, holes: 4, sum holes: 12 */
  /* padding: 22 */

ice_ring after this change:
  /* size: 192, cachelines: 3, members: 25 */
  /* sum members: 162, holes: 1, sum holes: 1 */
  /* padding: 29 */

ice_tx_buf prior to this change:
  /* size: 48, cachelines: 1, members: 7 */
  /* sum members: 38, holes: 2, sum holes: 6 */
  /* padding: 4 */
  /* last cacheline: 48 bytes */

ice_tx_buf after this change:
  /* size: 40, cachelines: 1, members: 7 */
  /* sum members: 38, holes: 1, sum holes: 2 */
  /* last cacheline: 40 bytes */
Signed-off-by: NJesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Fixes ethtool -S reported stats in ice driver to match
format and nomenclature of the ixgbe driver.
Signed-off-by: NRichard Rodriguez <richard.rodriguez@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Currently if ice_reset_all_vfs() fails in ice_alloc_vfs() we fail to
free some resources, reset variables, and return an error value.
Fix this by adding another unroll case to free the pf->vf array, set
the pf->num_alloc_vfs to 0, and return an error code.

Without this, if ice_reset_all_vfs() fails in ice_alloc_vfs() we will
not be able to do SRIOV without hard rebooting the system because
rmmod'ing the driver does not work.
Signed-off-by: NBrett Creeley <brett.creeley@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

This patch fixes the LLDP MIB change event handling code by removing
the workarounds in the current code. Added ice_dcb_need_recfg() to
print the DCB configuration changes detected via MIB change event.
Signed-off-by: NUsha Ketineni <usha.k.ketineni@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

User requested link modes affect what is returned as an advertised
link mode.  If no modes have been requested, we are not advertising
any link modes.  Advertise what we are capable of supporting if no
link modes have been requested.
Signed-off-by: NTony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

When disabling and enabling VSIs, there are a couple of flows
that recursively acquire the RTNL lock which causes a deadlock.
Fix that.
Signed-off-by: NDave Ertman <david.m.ertman@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

ice_parse_caps is used to parse both device and function capabilities.
Currently, capabilities are printed with a cryptic "HW caps" prefix,
which makes it difficult to distinguish whether the capabilities being
printed are device or function capabilities.

This patch makes a change to add a "func cap" prefix when printing
function capabilities, and a "dev cap" prefix when printing device
capabilities.

This patch also changes some of the capability print strings for
consistency.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Fix checkpatch warning "WARNING:BRACES: braces {} are not necessary
for single statement blocks"
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Commit 3463688e6ced ("ice: Add more validation in ice_vc_cfg_irq_map_msg")
added an assignment of vsi making the assignment during declaration
unnecessary.

Also, cleanup the declaration and assignment of irqmap_info to not use two
lines in the variable declaration section.
Signed-off-by: NBruce Allan <bruce.w.allan@intel.com>
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Implement LLDP persistence across reboots, start and stop of LLDP agent.
Add additional parameter to ice_aq_start_lldp and ice_aq_stop_lldp.

Also change the ethtool private flag from "disable-fw-lldp" to
"enable-fw-lldp". This change will flip the boolean logic of the
functionality of the flag (on = enable, off = disable). The change
in name and functionality is to differentiate between the
pre-persistence and post-persistence states.
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Fix double spacing in ice_napi_disable_all
Signed-off-by: NAnirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: NAndrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>

Create if_rmnet.h and move the rmnet MAP packet structs to this
common include file. To account for portablity, add little and
big endian bitfield definitions similar to the ip & tcp headers.

The definitions in the headers can now be re-used by the
upcoming ipa driver series as well as qmi_wwan.
Signed-off-by: NSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This reverts commit f8b99585.

The reverted change instructed the QMan hardware block to fetch
RX frame annotation and beginning of frame data to cache before
the core would read them.

It turns out that in rare cases, it's possible that a QMan
stashing transaction is delayed long enough such that, by the time
it gets executed, the frame in question had already been dequeued
by the core and software processing began on it. If the core
manages to unmap the frame buffer _before_ the stashing transaction
is executed, an SMMU exception will be raised.

Unfortunately there is no easy way to work around this while keeping
the performance advantages brought by QMan stashing, so disable
it altogether.
Signed-off-by: NIoana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Adds support for validating hardware filter spec configured in firmware
before offloading exact match flows.

Use the new fw api FW_PARAM_DEV_FILTER_MODE_MASK to read the filter mode
and mask from firmware. If the api isn't supported, then fall-back to
older way of reading just the mode from indirect register.
Signed-off-by: NRaju Rangoju <rajur@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Esben Haabendal says:

====================
net: ll_temac: Fix and enable multicast support

This patch series makes the necessary fixes to ll_temac driver to make
multicast work, and enables support for it.so that multicast support can

The main change is the change from mutex to spinlock of the lock used to
synchronize access to the shared indirect register access.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Multicast support have been tested and is working now.
Signed-off-by: NEsben Haabendal <esben@geanix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Avoid leaving old address table entries when using multicast.  If more than
one multicast address were removed, only the first removed address would
actually be cleared.
Signed-off-by: NEsben Haabendal <esben@geanix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With .ndo_set_rx_mode/temac_set_multicast_list() being called in atomic
context (holding addr_list_lock), and temac_set_multicast_list() needing
to access temac indirect registers, the mutex used to synchronize indirect
register is a no-no.

Replace it with a spinlock, and avoid sleeping in
temac_indirect_busywait().

To avoid excessive holding of the lock, which is now a spinlock, the
temac_device_reset() function is changed to only hold the lock for short
periods.  With timeouts, it could be holding the spinlock for more than
2 seconds.
Signed-off-by: NEsben Haabendal <esben@geanix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When user has requested IFF_ALLMULTI or have set more than 4 multicast
addresses, we should just use promiscuous mode, but not set it in flags,
as it causes the interface to stay in promiscuous mode even when the
non-IFF_PROMISC condition that caused promiscuous mode to be enabled
has gone away.
Signed-off-by: NEsben Haabendal <esben@geanix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

All LXT PHYs implement the standard "power down" bit 11 of
BMCR, so this patch adds support using the generic
genphy_{suspend,resume} functions added by
commit 0f0ca340 ("phy: power management support").

LXT970 is left aside because all registers get cleared upon
"power down" exit.
Reviewed-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: NFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Prevent misbehavior of drivers who would not set port type for longer
period of time. Drivers should always set port type. Do WARN if that
happens.

Note that it is perfectly fine to temporarily not have the type set,
during initialization and port type change.
Signed-off-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently, the hv_sock send() iterates once over the buffer, puts data into
the VMBUS channel and returns. It doesn't maximize on the case when there
is a simultaneous reader draining data from the channel. In such a case,
the send() can maximize the bandwidth (and consequently minimize the cpu
cycles) by iterating until the channel is found to be full.

Perf data:
Total Data Transfer: 10GB/iteration
Single threaded reader/writer, Linux hvsocket writer with Windows hvsocket
reader
Packet size: 64KB
CPU sys time was captured using the 'time' command for the writer to send
10GB of data.
'Send Buffer Loop' is with the patch applied.
The values below are over 10 iterations.

|--------------------------------------------------------|
|        |        Current        |   Send Buffer Loop    |
|--------------------------------------------------------|
|        | Throughput | CPU sys  | Throughput | CPU sys  |
|        | (MB/s)     | time (s) | (MB/s)     | time (s) |
|--------------------------------------------------------|
| Min    |     407    |   7.048  |    401     |  5.958   |
|--------------------------------------------------------|
| Max    |     455    |   7.563  |    542     |  6.993   |
|--------------------------------------------------------|
| Avg    |     440    |   7.411  |    451     |  6.639   |
|--------------------------------------------------------|
| Median |     446    |   7.417  |    447     |  6.761   |
|--------------------------------------------------------|

Observation:
1. The avg throughput doesn't really change much with this change for this
scenario. This is most probably because the bottleneck on throughput is
somewhere else.
2. The average system (or kernel) cpu time goes down by 10%+ with this
change, for the same amount of data transfer.
Signed-off-by: NSunil Muthuswamy <sunilmut@microsoft.com>
Reviewed-by: NDexuan Cui <decui@microsoft.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently, the hv_sock buffer size is static and can't scale to the
bandwidth requirements of the application. This change allows the
applications to influence the socket buffer sizes using the SO_SNDBUF and
the SO_RCVBUF socket options.

Few interesting points to note:
1. Since the VMBUS does not allow a resize operation of the ring size, the
socket buffer size option should be set prior to establishing the
connection for it to take effect.
2. Setting the socket option comes with the cost of that much memory being
reserved/allocated by the kernel, for the lifetime of the connection.

Perf data:
Total Data Transfer: 1GB
Single threaded reader/writer
Results below are summarized over 10 iterations.

Linux hvsocket writer + Windows hvsocket reader:
|---------------------------------------------------------------------------------------------|
|Packet size ->   |      128B       |       1KB       |       4KB       |        64KB         |
|---------------------------------------------------------------------------------------------|
|SO_SNDBUF size | |                 Throughput in MB/s (min/max/avg/median):                  |
|               v |                                                                           |
|---------------------------------------------------------------------------------------------|
|      Default    | 109/118/114/116 | 636/774/701/700 | 435/507/480/476 |   410/491/462/470   |
|      16KB       | 110/116/112/111 | 575/705/662/671 | 749/900/854/869 |   592/824/692/676   |
|      32KB       | 108/120/115/115 | 703/823/767/772 | 718/878/850/866 | 1593/2124/2000/2085 |
|      64KB       | 108/119/114/114 | 592/732/683/688 | 805/934/903/911 | 1784/1943/1862/1843 |
|---------------------------------------------------------------------------------------------|

Windows hvsocket writer + Linux hvsocket reader:
|---------------------------------------------------------------------------------------------|
|Packet size ->   |     128B    |      1KB        |          4KB        |        64KB         |
|---------------------------------------------------------------------------------------------|
|SO_RCVBUF size | |               Throughput in MB/s (min/max/avg/median):                    |
|               v |                                                                           |
|---------------------------------------------------------------------------------------------|
|      Default    | 69/82/75/73 | 313/343/333/336 |   418/477/446/445   |   659/701/676/678   |
|      16KB       | 69/83/76/77 | 350/401/375/382 |   506/548/517/516   |   602/624/615/615   |
|      32KB       | 62/83/73/73 | 471/529/496/494 |   830/1046/935/939  | 944/1180/1070/1100  |
|      64KB       | 64/70/68/69 | 467/533/501/497 | 1260/1590/1430/1431 | 1605/1819/1670/1660 |
|---------------------------------------------------------------------------------------------|
Signed-off-by: NSunil Muthuswamy <sunilmut@microsoft.com>
Reviewed-by: NDexuan Cui <decui@microsoft.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Removing two 4 bytes holes allows to use kmalloc-32
kmem cache instead of kmalloc-64 on 64bit kernels.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>