ath9k: Cleanup get_rate() interface

The interface to calculate the TX rate for a data frame
was convoluted with lots of redundant arguments being
passed around. Remove all of that and make it simple.

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath5k: properly free rx dma descriptors

When freeing rx dma descriptors, use the right buffer size.
Fixes kernel oopses on module unload on ixp4xx and most likely
other platforms as well.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Acked-by: Nick Kossifidis <mickflemm@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

wireless: replace uses of __constant_{endian}

The base versions handle constant folding now.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: fix led naming

Fixed led device naming for the iwl driver. Due to the
documentation of the led subsystem/class the naming should be
"devicename:colour:function" while not applying sections
should be left blank.

This should lead to e.g. "iwl-phy0::RX" instead of "iwl-phy0:RX".

Signed-off-by: Danny Kukawka <dkukawka@suse.de>
Acked-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ipw2x00: correct Kconfig to prevent following entries from not indenting

not defining dependencies for LIBIPW caused the following entries to not be
indented. changing this entry to depend on PCI && WLAN_80211 corrects this
issue

Signed-off-by: Jay Sternberg <jay.e.sternberg@linux.intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: remove chain noise calibration functions from 6000 family

redefine structures that contain function pointer for chain noise reset
and chain noise gain for the 6000 family since these are not needed.

Signed-off-by: Jay Sternberg <jay.e.sternberg@linux.intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: define structures and functions externally for customization

defined the structures and functions as extern to alter behavior used by
5000 series for other products including 100 and 6000 series

Signed-off-by: Jay Sternberg <jay.e.sternberg@linux.intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: correct API command overlap

Correct the API commands where same command id used for two different
commands. Update max api versions for affected devices.

TX_ANT_CONFIGURATION_CMD was already using id 0x98, so
REPLY_TX_POWER_DBM_CMD moved to 0x95

Older API interfaces may used original value so V1 defines provided.

Signed-off-by: Jay Sternberg <jay.e.sternberg@linux.intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: parametrize configuration of the PLL for exclusion on 6000

added a config parameter to enable setting PLL_CFG.  older hardware has
this parameter set true.  the 6000 family does not support this setting,
so this parameter set false.

Signed-off-by: Jay Sternberg <jay.e.sternberg@linux.intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: simplify parameter setting to allow support for 6000 series

by parametrizing the set hw function, in addition to allowing for
supporting the 6000 family significantly simplify the addition of new
hardware.

Signed-off-by: Jay Sternberg <jay.e.sternberg@linux.intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: add new HW_REV_TYPEs for Intel WiFi Link 100, 6000 and 6050 Series

simply add definitions for the HW_REV_TYPEs for the new devices.

Signed-off-by: Jay Sternberg <jay.e.sternberg@linux.intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: IBSS join rework

I hold back this patch for around a week to avoid
confusion. This is the second step of
"mac80211: Fixed BSSID handling revisited".

With it, in the situation of a strange merge to the
same BSSID (e.g. caused by a TSF overflow) only
reset_tsf() is called.

And sta_info_flush_delayed() is only called if you
change the network manually, not on an automatic
BSSID merge.

Signed-off-by: Alina Friedrichsen <x-alina@gmx.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Fix LED blink pattern

Signed-off-by: Vasanthakumar Thiagarajan <vasanth@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: Creating new IBSS with fixed BSSID

This fixes a bug when creating a new IBSS network with a
fixed BSSID. The fixed BSSID situation is now with one of
my last patches handled in ieee80211_sta_find_ibss()
function.

It's more robust to test against
(ifsta->flags & IEEE80211_STA_PREV_BSSID_SET), because
ifsta->state is not seted right in every situation and so
the creating of the new IBSS network sometimes hangs after
the first try to scan for a network to merge.

Signed-off-by: Alina Friedrichsen <x-alina@gmx.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Remove a bunch of unused macros

RX filter masks are already defined in enum ath9k_rx_filter
in ath9k.h

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Fix typo in checking for chip revision

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: Reset assoc_scan_tries after an unsuccessful scan run

Trying to associate with a non-existent SSID stops the
state machine after the first run. Subsequent association
requests fail to start the scan engine. Fix this by resetting
assoc_scan_tries to zero after completing a scan run.

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: remove passive scan on 5 GHz if country IE knows better

If we have new found information about our location and the
current country regulatory domain does not have passive scan
flag requirements we should be able to actively scan now on those
channels.

Since AP functionality is not allowed where passive scan flags are
set this means if you have a world regulatory domain and you get a
country IE that allows that channel (with active scan) then we lift
the passive-scan requirement so you can then use AP mode.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: move check for radar freqs into a helper

This will be used later.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: replace usage of internal wireless_modes for conf

No need to use our internal wireless mode variable when
cfg80211 already has its own.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Enable MIB and TIM interrupts for station mode.

Enable operating mode specific interrupts in ath9k_add_interface instead
of ath9k_start.

Signed-off-by: Vivek Natarajan <vnatarajan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath9k: Fix lockdep warning

This patch fixes the lockdep warning shown below, and also
initializes the starting sequence number when starting a TX
aggregation session.

 =============================================
 [ INFO: possible recursive locking detected ]
 2.6.29-rc2-wl #21
 ---------------------------------------------
 swapper/0 is trying to acquire lock:
  (_xmit_IEEE80211#2){-+..}, at: [<ffffffff80456d71>] __qdisc_run+0x221/0x290

 but task is already holding lock:
  (_xmit_IEEE80211#2){-+..}, at: [<ffffffff80456d71>] __qdisc_run+0x221/0x290

 other info that might help us debug this:
 7 locks held by swapper/0:
  #0:  (rcu_read_lock){..--}, at: [<ffffffff80442a63>] dev_queue_xmit+0x53/0x620
  #1:  (_xmit_ETHER#2){-+..}, at: [<ffffffff80456d71>] __qdisc_run+0x221/0x290
  #2:  (rcu_read_lock){..--}, at: [<ffffffff80442a63>] dev_queue_xmit+0x53/0x620
  #3:  (_xmit_IEEE80211#2){-+..}, at: [<ffffffff80456d71>] __qdisc_run+0x221/0x290
  #4:  (rcu_read_lock){..--}, at: [<ffffffffa0154919>] ieee80211_master_start_xmit+0x219/0x6c0 [mac80211]
  #5:  (rcu_read_lock){..--}, at: [<ffffffffa01427c6>] ieee80211_start_tx_ba_session+0x66/0x4e0 [mac80211]
  #6:  (rcu_read_lock){..--}, at: [<ffffffff80442a63>] dev_queue_xmit+0x53/0x620

 stack backtrace:
 Pid: 0, comm: swapper Not tainted 2.6.29-rc2-wl #21
 Call Trace:
  <IRQ>  [<ffffffff8026c329>] __lock_acquire+0x1be9/0x1c40
  [<ffffffff80442af1>] dev_queue_xmit+0xe1/0x620
  [<ffffffff8026a8cc>] __lock_acquire+0x18c/0x1c40
  [<ffffffff8026c3d5>] lock_acquire+0x55/0x70
  [<ffffffff80456d71>] __qdisc_run+0x221/0x290
  [<ffffffff804dbeb9>] _spin_lock+0x39/0x50
  [<ffffffff80456d71>] __qdisc_run+0x221/0x290
  [<ffffffff804dbd2f>] _spin_unlock+0x1f/0x50
  [<ffffffff80456d71>] __qdisc_run+0x221/0x290
  [<ffffffff80442d18>] dev_queue_xmit+0x308/0x620
  [<ffffffff80442a63>] dev_queue_xmit+0x53/0x620
  [<ffffffffa0142a63>] ieee80211_start_tx_ba_session+0x303/0x4e0 [mac80211]
  [<ffffffffa01427c6>] ieee80211_start_tx_ba_session+0x66/0x4e0 [mac80211]
  [<ffffffffa0149dae>] rate_control_get_rate+0xae/0xc0 [mac80211]
  [<ffffffffa01526b5>] invoke_tx_handlers+0x655/0x1000 [mac80211]
  [<ffffffff802699fd>] mark_held_locks+0x4d/0x90
  [<ffffffff804dbcf5>] _spin_unlock_irqrestore+0x65/0x80
  [<ffffffffa0151aaa>] __ieee80211_tx_prepare+0x16a/0x310 [mac80211]
  [<ffffffffa0151adc>] __ieee80211_tx_prepare+0x19c/0x310 [mac80211]
  [<ffffffff80439cc2>] pskb_expand_head+0x112/0x190
  [<ffffffffa0154986>] ieee80211_master_start_xmit+0x286/0x6c0 [mac80211]
  [<ffffffffa0154919>] ieee80211_master_start_xmit+0x219/0x6c0 [mac80211]
  [<ffffffff8026a8cc>] __lock_acquire+0x18c/0x1c40
  [<ffffffff80456d8e>] __qdisc_run+0x23e/0x290
  [<ffffffff80442d18>] dev_queue_xmit+0x308/0x620
  [<ffffffff80442a63>] dev_queue_xmit+0x53/0x620
  [<ffffffffa0154221>] ieee80211_subif_start_xmit+0x4a1/0x980 [mac80211]
  [<ffffffffa0153f18>] ieee80211_subif_start_xmit+0x198/0x980 [mac80211]
  [<ffffffff80456d8e>] __qdisc_run+0x23e/0x290
  [<ffffffff80442d18>] dev_queue_xmit+0x308/0x620
  [<ffffffff80442a63>] dev_queue_xmit+0x53/0x620
  [<ffffffffa028ecfd>] ip6_output+0x62d/0x1230 [ipv6]
  [<ffffffff8024ca00>] __mod_timer+0xb0/0xd0
  [<ffffffffa02ad25a>] mld_sendpack+0x3fa/0x4a0 [ipv6]
  [<ffffffffa02ace60>] mld_sendpack+0x0/0x4a0 [ipv6]
  [<ffffffffa02adf90>] mld_ifc_timer_expire+0x0/0x340 [ipv6]
  [<ffffffffa02ae219>] mld_ifc_timer_expire+0x289/0x340 [ipv6]
  [<ffffffffa02adf90>] mld_ifc_timer_expire+0x0/0x340 [ipv6]
  [<ffffffff8024c097>] run_timer_softirq+0x147/0x220
  [<ffffffff802473fb>] __do_softirq+0x9b/0x180
  [<ffffffff80265516>] tick_dev_program_event+0x36/0xb0
  [<ffffffff8020d77c>] call_softirq+0x1c/0x30
  [<ffffffff8020f2c5>] do_softirq+0x65/0xb0
  [<ffffffff80246ebd>] irq_exit+0x9d/0xc0
  [<ffffffff80221db6>] smp_apic_timer_interrupt+0x86/0xd0
  [<ffffffff8020d1b3>] apic_timer_interrupt+0x13/0x20

Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

rt2x00: Validate firmware in driver

The get_firmware_crc() callback function isn't flexible
enough when dealing with multiple firmware versions.
It might in some cases be possible that the firmware
file contains multiple CRC checksums.

Create the check_firmware() callback function where the driver
has complete freedom in how to validate the firmware.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

rt2x00: Add kill_tx_queue callback function

provide rt2x00lib the possibility to kill a particular TX queue.
This can be useful when disabling the radio, but more importantly
will allow beaconing to be disabled when mac80211 requests this
(during scanning for example)

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

rt2x00: Move intf_work to mac82011 workqueue

ieee80211_iterate_active_interfaces() no longer acquires the
RTNL lock which means the intf_work handler can be safely
used from the mac80211 workqueue again.

Signed-off-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: unify iwlagn and 3945 power save management

This patch unifies 3945 and iwlagn power save management
This patch also better separates system state from user setting.
System state shall be removed later as this shall be shifted to user space

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Acked-by: Mohamed Abbas <mohamed.abbas@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: don't use implicit priv in IWL_DEBUG

Call IWL_DEBUG macro with explicit priv argument.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Acked-by: Samuel Ortiz <samuel.ortiz@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: check return value of pci_enable_device

pci_enable_device is tagged with __must_check therefore
don't ignore the return value in pci_resume handlers

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: iwl3945_send_tx_power must be static

iwl3945_send_tx_power must be static

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwlwifi: suppress unused variable warning when compiling w/o IWLWIFI_DEBUG

This patch adds __maybe_unused attribute to priv variables used in
functions that used it solely for debug printouts

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Cc: Helmut Schaa <helmut.schaa@googlemail.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwl3945: Getting rid of the *39_rxon iwl_priv fields

The iwl_rxon_cmd is really just a iwl3945_rxon_cmd structure extension.
So, we can use the *_rxon fields from iwl_priv instead of the 3945 specific
ones (*39_rxon). We have to then be careful when submitting REPLY_RXON host
commands, since the command length as to be set according to the HW. As
another precaution the reserved4 and reserved5 fields are cleared before being
sent to the 3945.

With the *39_rxon removal, a lot of duplicated code can be removed from the
3945 code base.

Signed-off-by: Samuel Ortiz <samuel.ortiz@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

orinoco: checkpatch cleanup

Fix errors and obvious warnings reported by checkpatch in all files
except orinoco.c. Orinoco.c is part of different patch series of Dave.

Signed-off-by: Andrey Borzenkov <arvidjaar@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

gro: Optimise TCP packet reception

gro: Optimise TCP packet reception

As this function can be called more than half a million times for
10GbE, it's important to optimise it as much as we can.

This patch uses bit ops to logical ops, as well as open coding
memcmp to exploit alignment properties.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

gro: Optimise IPv4 packet reception

As this function can be called more than half a million times for
10GbE, it's important to optimise it as much as we can.

This patch does some obvious changes to use 2-byte and 4-byte
operations instead of byte-oriented ones where possible.  Bit
ops are also used to replace logical ops to reduce branching.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

gro: Optimise Ethernet header comparison

This patch optimises the Ethernet header comparison to use 2-byte
and 4-byte xors instead of memcmp.  In order to facilitate this,
the actual comparison is now carried out by the callers of the
shared dev_gro_receive function.

This has a significant impact when receiving 1500B packets through
10GbE.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

gro: Remember number of held packets instead of counting every time

This patch prepares for the move of the same_flow checks out of
dev_gro_receive.  As such we need to remember the number of held
packets since doing a loop just to count them every time is silly.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Increase default NET_SKB_PAD to 32.

Several devices need to insert some "pre headers" in front of the
main packet data when they transmit a packet.

Currently we allocate only 16 bytes of pad room and this ends up not
being enough for some types of hardware (NIU, usb-net, s390 qeth,
etc.)

So increase this to 32.

Note that drivers still need to check in their transmit routine
whether enough headroom exists, and if not use skb_realloc_headroom().
Tunneling, IPSEC, and other encapsulation methods can cause the
padding area to be used up.

Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of /linux/kernel/git/davem/net-2.6

Conflicts:
drivers/net/wireless/iwlwifi/iwl-agn.c
drivers/net/wireless/iwlwifi/iwl3945-base.c

igb: remove dead code in transmit routine

Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: update version number and copyright dates

Update the version number to 1.3.16 and update copyright dates for 2009.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: fix two minor items found during code review

This patch addresses two minor items I found while cleaning up the igb
driver for our sourceforge version.

The first clears the context index if we don't flag that we need it.

The second item is that eims_other should be used instead of bit defines
when setting all of the EICS bits prior to reset.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: update stats before doing reset in igb_down

It was seen with repeated interface up/down testing that there was a large
stray between the stats reported by the queues and the stats reported by the
HW.  It was found to be an issue in that hw stats were being reset without
first being recorded.  This change records the stats before wiping them from
the system via the reset.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: remove redundant count set and err_hw_init

Remove the setting of ring->count variables from igb_probe as they are
duplicating the same configuration that is done igb_alloc_queues.

Remove the err_hw_init tag as it can be replaced by err_sw_init.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: remove disable_av variable from mac_info struct

The disable_av variable is never used by the driver and provides no value as
it is likely a leftover debugging variable.  I have removed it and replaced
the one spot that checked for it with a check for a valid address.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: change pba size determination from if to switch statement

As additional hardware is added to the igb driver it is easier to support
the expansion via switch statements instead of using nested ifs.  For
this reason I am changing this to a switch statement.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: move get_hw_control within igb_resume.

Move igb_get_hw_control up so that it is called just after the reset in
igb_resume.  This notifies the HW sooner that the driver is reassuming
control of the device.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: don't read eicr when responding to legacy interrupts

The interrupt handler was reading eicr and then doing nothing with the
result.  I have removed the variable and the register read since they
provide no value to the legacy interrupt handler.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: remove unnecessary adapter->hw calls when just hw-> will do.

There were several spots in the code making calls to adapter->hw when they
could have just been accessing hw-> directly.  I cleaned up the spots where
this was visibly apparent.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: rename igb_update_mc_addr_list_82575 to not include the 82575

There isn't much point in having the _82575 hanging off the end of this
function since there aren't any other version of this function running
around within this driver.  This also allows for a bit of whitespace
cleanup due to a shorter function name.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: remove redundant timer updates and cleanup watchdog_task

The igb watchdog task is modifying the watchdog timer twice duing a single
run.  It only needs to be called once to reschedule itself for 2 seconds from
the last time it ran.

In addition I removed the allocation of the mac_info structure since it is
only called twice and is easier to access via the e1000_hw struct.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: cleanup igb_netpoll to be more friendly with napi & GRO

This patch cleans up igb_netpoll so that it is more friendly with both the
current napi and newly introduced GRO features.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: add counter for dma out of sync errors

Add a counter for dma out of sync errors reported via interrupt.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: update testing done by ethtool

Most of the code for the testing has pretty much become stale at this point
and is need of update.  This update just streamlines most of the code,
widens the range of interrupt testing.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: update feature flags supported in ethtool

This driver is currently using HW_CSUM which is not correct.  Update this
to use the IP_CSUM and IPV6_CSUM flags.  In addition consolidate the TSO
flag setting.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: remove unused rx_hdr_split statistic

This statistic is not used and so it is safe to remove

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: rename nvm ops

All of the nvm ops have the tag _nvm added to the end which is redundant
since all of the calls to the ops have to go through the nvm ops struct
anyway.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: rename phy ops

This patch renames write_phy_reg to write_reg and read_phy_reg to read_reg.
It seems redundant to call out phy in an operation that is part of the
phy_ops struct.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: read address from RAH/RAL instead of from EEPROM

Instead of pulling the mac address from EEPROM it is easier to pull it from
the RAL/RAH registers and then just copy it into the address structures.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: make dev_spec a union and remove dynamic allocation

This patch makes dev_spec a union and simplifies it so that it does not
require dynamic allocation and freeing in the driver.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: add link check function

Add a link check function to contain all activities related to verifying
that the link is present.  The current approach is a bit cludgy and needs
to be cleaned up.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: remove check for needing an io port

Since igb supports only pci-e nics and there is no plan to support any
legacy pci parts in the driver there isn't really much need for checking to
see if an io port is needed.

In the unlikely event that we do begin supporting legacy pci parts then we
can see about adding this code back to the driver.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: move initialization of number of queues into set_interrupt_capability

This patch moves the initialization of the number of queues into
set_interrupt_capability.  This allows the number of queues to increase in
the unlikely event that the system initially fails to allocate enough msi-x
interrupts, does a suspend/resume, and then can allocate enough interrupts
on resume.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: move setting of buffsz out of repeated path in alloc_rx_buffers

buffsz is being repeatedly set when allocaing buffers.  Since this value
should only need to be set once in the function I am moving it out of the
looped portion of the path.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: optimize/refactor receive path

While cleaning up the skb_over panic with small frames I found there was
room for improvement in the ordering of operations within the rx receive
flow.  These changes will place the prefetch for the next descriptor to a
point earlier in the rx path.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sunhme: Don't match PCI devices in SBUS probe.

Unfortunately, the OF device tree nodes for SBUS and PCI
hme devices have the same device node name on some systems.

So if the name of the parent node isn't 'sbus', skip it.

Based upon an excellent report and detective work by
Meelis Roos and Eric Brower.

Signed-off-by: David S. Miller <davem@davemloft.net>
Tested-by: Meelis Roos <mroos@linux.ee>

ixgbe: Fix a set_num_queues() bug that can result in num_(r|t)x_queues = 0

Now that our set_num_queues() routines for each feature are re-entrant, and
can be called at any point, they shouldn't zero out the feature's indices
or mask bits.  Subsequent calls into those routines for those features can
result in zero Rx and Tx queues being assigned, causing a panic later in
driver reinitialization.

Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: bump version to 63

This patch bumps the version up to 63

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: recover error support

This patch adds another type of recoverable error to the driver. It also
modifies the sequence for recovery to include a mac reset and clearing
of interrupts.

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: ethtool tx csum fix

This patch fixes the ethtool tx csum "set" command. A recent patch was
submitted to remove HW_CSUM and use IP_CSUM instead. Therefore, the
corresponding ethtool command should also be modified.

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: msi interrupt fix

This patch fixes an issue with the suspend/resume cycle with msi
interrupts. See bugzilla number 10487 for more details. The fix is to
re-setup a private msi pci config offset field.

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: mgmt unit interface

This patch updates the logic used to communicate with the mgmt unit. It
also adds a version check for a newer mgmt unit firmware.

* Fixed udelay to schedule_timeout_uninterruptible

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/sunrpc/xprtsock.c: some common code found

$ diff-funcs xs_udp_write_space net/sunrpc/xprtsock.c
net/sunrpc/xprtsock.c xs_tcp_write_space
 --- net/sunrpc/xprtsock.c:xs_udp_write_space()
 +++ net/sunrpc/xprtsock.c:xs_tcp_write_space()
@@ -1,4 +1,4 @@
- * xs_udp_write_space - callback invoked when socket buffer space
+ * xs_tcp_write_space - callback invoked when socket buffer space
  *                             becomes available
  * @sk: socket whose state has changed
  *
@@ -7,12 +7,12 @@
  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
  * with a bunch of small requests.
  */
-static void xs_udp_write_space(struct sock *sk)
+static void xs_tcp_write_space(struct sock *sk)
 {
  read_lock(&sk->sk_callback_lock);

- /* from net/core/sock.c:sock_def_write_space */
- if (sock_writeable(sk)) {
+ /* from net/core/stream.c:sk_stream_write_space */
+ if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
  struct socket *sock;
  struct rpc_xprt *xprt;

$ codiff net/sunrpc/xprtsock.o net/sunrpc/xprtsock.o.new
net/sunrpc/xprtsock.c:
  xs_tcp_write_space | -163
  xs_udp_write_space | -163
 2 functions changed, 326 bytes removed

net/sunrpc/xprtsock.c:
  xs_write_space | +179
 1 function changed, 179 bytes added

net/sunrpc/xprtsock.o.new:
 3 functions changed, 179 bytes added, 326 bytes removed, diff: -147

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6/addrconf: common code located

$ codiff net/ipv6/addrconf.o net/ipv6/addrconf.o.new
net/ipv6/addrconf.c:
 addrconf_notify | -267
1 function changed, 267 bytes removed

net/ipv6/addrconf.c:
 add_addr |  +86
1 function changed, 86 bytes added

net/ipv6/addrconf.o.new:
2 functions changed, 86 bytes added, 267 bytes removed, diff: -181

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6/ndisc: join error paths

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

ax25: more common return path joining

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipmr: use goto to common label instead of opencoding

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

9p: fix endian issues [attempt 3]

When the changes were done to the protocol last release, some endian
bugs crept in.  This patch fixes those endian problems and has been
verified to run on 32/64 bit and x86/ppc architectures.

This version of the patch incorporates the correct annotations
for endian variables.

Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_dma: call dmaengine_get only if NET_DMA enabled

Based upon a patch from Atsushi Nemoto <anemo@mba.ocn.ne.jp>

--------------------
The commit 649274d993212e7c23c0cb734572c2311c200872 ("net_dma:
acquire/release dma channels on ifup/ifdown") added unconditional call
of dmaengine_get() to net_dma.  The API should be called only if
NET_DMA was enabled.
--------------------

Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Dan Williams <dan.j.williams@intel.com>

3c509: Fix resume from hibernation for PnP mode.

From: Ondrej Zary <linux@rainbow-software.org>

last year, I posted a patch which fixed hibernation on 3c509
cards. That was back in 2.6.24. It worked fine in 2.6.25. But then I
stopped using hibernation (as it did not work with my new IT8212 RAID
controller).

Now I fixed it and noticed that 3c509 does not wake up properly
anymore (in 2.6.28) - neither in PnP nor in ISA modes. ifconfig
down/up makes the card work again in PnP mode. However, in ISA mode,
ifconfig up ends with "No such device" error.

Comparing the 3c509 driver between 2.6.25 and 2.6.28, there's only
some statistics-related change. So the cause of the problem must be
somewhere else.

This patch makes the resume work in PnP mode, but it's still not
enough for ISA mode.

Signed-off-by: David S. Miller <davem@davemloft.net>

sungem: Soft lockup in sungem on Netra AC200 when switching interface up

From: Ilkka Virta <itvirta@iki.fi>

In the lockup situation the driver seems to go off in an eternal storm
of interrupts right after calling request_irq(). It doesn't actually
do anything interesting in the interrupt handler. Since connecting the link
afterwards works, something later in initialization must fix this.

Looking at gem_do_start() and gem_open(), it seems that the only thing
done while opening the device after the request_irq(), is a call to
napi_enable().

I don't know what the ordering requirements are for the
initialization, but I boldly tried to move the napi_enable() call
inside gem_do_start() before the link state is checked and interrupts
subsequently enabled, and it seems to work for me. Doesn't even break
anything too obvious...

Signed-off-by: David S. Miller <davem@davemloft.net>

RxRPC: Fix a potential NULL dereference

Fix a potential NULL dereference bug during error handling in
rxrpc_kernel_begin_call(), whereby rxrpc_put_transport() may be handed a NULL
pointer.

This was found with a code checker (http://repo.or.cz/w/smatch.git/).

Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

r8169: Don't update statistics counters when interface is down

Some Realtek chips (RTL8169sb/8110sb in my case) are unable to retrieve
ethtool statistics when the interface is down. The process stays in
endless loop in rtl8169_get_ethtool_stats. This is because these chips
need to have receiver enabled (CmdRxEnb bit in ChipCmd register) that is
cleared when the interface is going down. It's better to update statistics
only when the interface is up and otherwise return copy of statistics
grabbed when the interface was up (in rtl8169_close).

It is interesting that PCI-E NICs (like 8168b/8111b...) are not affected.

Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: Defeature Tx Head writeback

Tx Head writeback is causing multi-microsecond stalls on PCIe chipsets, due
to partial cacheline writebacks.  Removing this feature removes these
issues.

Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: Update flow control state machine in link setup

The flow control handling is overly complicated and difficult to maintain.
This patch cleans up the flow control handling and makes it much more
explicit.  It also adds 1G flow control autonegotiation, for 1G copper
links, 1G KX links, and 1G fiber links.

Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

udp: Fix potential wrong ip_hdr(skb) pointers

Like the UDP header fix, pskb_may_pull() can potentially
alter the SKB buffer.  Thus the saddr and daddr, pointers
may point to the old skb->data buffer.

I haven't seen corruptions, as its only seen if the old
skb->data buffer were reallocated by another user and
written into very quickly (or poison'd by SLAB debugging).

Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: enable msix to default

Impact: change default

msix and napic can work again

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: ck804 and mcp55 doesn't need timerirq

Impact: cleanup

so get less irq.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: disable irq at first before schedule rx

Impact: clean up

schedule it later after disable it.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: don't clear nic_poll_irq too early

Impact: fix bug

for msix, we still need that flag to enable irq respectively

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

forcedeth: make msi-x different name for rx-tx

Impact: make /proc/interrupts could show more info which irq is rx or other for msi-x

add three name fields for rx, tx, other

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

neigh: some entries can be skipped during dumping

neightbl_dump_info and neigh_dump_table  can skip entries if the
*fill*info functions return an error. This results in an incomplete
dump ((invoked by netlink requests for RTM_GETNEIGHTBL or
RTM_GETNEIGH)

nidx and idx should not be incremented if the current entry was not
placed in the output buffer

Signed-off-by: Gautam Kachroo <gk@aristanetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Disallow rediculious flowlabel option sizes.

Just like PKTINFO, limit the options area to 64K.

Based upon report by Eric Sesterhenn and analysis by
Roland Dreier.

Signed-off-by: David S. Miller <davem@davemloft.net>

netlink: change return-value logic of netlink_broadcast()

Currently, netlink_broadcast() reports errors to the caller if no
messages at all were delivered:

1) If, at least, one message has been delivered correctly, returns 0.
2) Otherwise, if no messages at all were delivered due to skb_clone()
   failure, return -ENOBUFS.
3) Otherwise, if there are no listeners, return -ESRCH.

With this patch, the caller knows if the delivery of any of the
messages to the listeners have failed:

1) If it fails to deliver any message (for whatever reason), return
   -ENOBUFS.
2) Otherwise, if all messages were delivered OK, returns 0.
3) Otherwise, if no listeners, return -ESRCH.

In the current ctnetlink code and in Netfilter in general, we can add
reliable logging and connection tracking event delivery by dropping the
packets whose events were not successfully delivered over Netlink. Of
course, this option would be settable via /proc as this approach reduces
performance (in terms of filtered connections per seconds by a stateful
firewall) but providing reliable logging and event delivery (for
conntrackd) in return.

This patch also changes some clients of netlink_broadcast() that
may report ENOBUFS errors via printk. This error handling is not
of any help. Instead, the userspace daemons that are listening to
those netlink messages should resync themselves with the kernel-side
if they hit ENOBUFS.

BTW, netlink_broadcast() clients include those that call
cn_netlink_send(), nlmsg_multicast() and genlmsg_multicast() since they
internally call netlink_broadcast() and return its error value.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: normalize usage of serdes_has_link

Cosmetic change to use struct e1000_mac_info.serdes_has_link
consistently as the 'bool' that it's declared as.

No functional change.

Signed-off-by: Alex Chiang <achiang@hp.com>
Acked-by: Jeff Kirsher <Jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: Display EEPROM version in ethtool -i queries

Currently ixgbe does not display the EEPROM version in ethtool -i, where
other drivers do.  The EEPROM version is located at offset 0x29.  This
patch adds support to display it.

Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: Update link setup code to better support autonegotiation of speed

The current code has some flaws in it when performing autonegotiation,
especially on KX/KX4 links.  This patch updates the code to better handle
the autonegotiation states on link setup.  The patch also removes a redundant
link configuration call on driver load, and moves link configuration to
the ->open() path.

Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: Refactor set_num_queues() and cache_ring_register()

The current code to determine the number of queues the device will want
on driver initialization is ugly and difficult to maintain.  It also
doesn't allow for easy expansion for future features or future hardware.
This patch refactors these routines, and make them easier to deal with.

Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gro: Fix frag_list merging on imprecisely split packets

The previous fix ad0f9904444de1309dedd2b9e365cae8af77d9b1 (gro:
Fix handling of imprecisely split packets) only fixed the case
of frags merging, frag_list merging in the same circumstances
were still broken.

In particular, the packet headers end up in the data stream.

This patch fixes this plus another issue where an imprecisely
split packet header may be read incorrectly (this is mostly
harmless since it'll simply cause the packet to not match and
be rejected for GRO).

Thanks to Emil Tantilov and Jeff Kirsher for helping to track
this down.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: fix some trailing whitespaces

Signed-off-by: Graf Yang <graf.yang@analog.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

tun: Limit amount of queued packets per device

Unlike a normal socket path, the tuntap device send path does
not have any accounting.  This means that the user-space sender
may be able to pin down arbitrary amounts of kernel memory by
continuing to send data to an end-point that is congested.

Even when this isn't an issue because of limited queueing at
most end points, this can also be a problem because its only
response to congestion is packet loss.  That is, when those
local queues at the end-point fills up, the tuntap device will
start wasting system time because it will continue to send
data there which simply gets dropped straight away.

Of course one could argue that everybody should do congestion
control end-to-end, unfortunately there are people in this world
still hooked on UDP, and they don't appear to be going away
anywhere fast.  In fact, we've always helped them by performing
accounting in our UDP code, the sole purpose of which is to
provide congestion feedback other than through packet loss.

This patch attempts to apply the same bandaid to the tuntap device.
It creates a pseudo-socket object which is used to account our
packets just as a normal socket does for UDP.  Of course things
are a little complex because we're actually reinjecting traffic
back into the stack rather than out of the stack.

The stack complexities however should have been resolved by preceding
patches.  So this one can simply start using skb_set_owner_w.

For now the accounting is essentially disabled by default for
backwards compatibility.  In particular, we set the cap to INT_MAX.
This is so that existing applications don't get confused by the
sudden arrival EAGAIN errors.

In future we may wish (or be forced to) do this by default.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>