The MIB counters are disabled when doing a chip reset.
Since ANI depends on the MIB registers for its operation, relying
on the contents of said registers during HW reset results in sub-optimal
performance.
Cc: stable@kernel.org
Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When TX DMA termination has failed, the HW has to be reset
completely. Doing a fast channel change in this case is insufficient.
Also, change the debug level of a couple of messages to FATAL.
Cc: stable@kernel.org
Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The internal, driver-specific maintenance of sequence
numbers is applicable only for HT frames.
Also, remove comments that are not relevant anymore.
Signed-off-by: Sujith <Sujith.Manoharan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Fix typo. The index should be multiplied by the entry size, not 'and'-ed.
Found via code-inspection.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Some devices have 40MHz operation disabled entirely. Ensure that driver do
not enable 40MHz operation if a channel does not allow this.
This fixes http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=2135
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
CC: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Recent powersaving work resulted in power management ops being called
during EEPROM initialization. The lock used by these functions is not
initialized at this time. Ensure lock is initialized before it is used.
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
we see from http://bugzilla.intellinuxwireless.org/show_bug.cgi?id=2125
that power saving does not work well on 3945. Since then power saving has
also been connected with association problems where an AP deathenticates a
3945 after it is unable to transmit data to it - this happens when 3945
enters power savings mode.
Disable power save support until issues are resolved.
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
CC: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
I've also for a long time had a problem with the
temperature calculation code, which I had fixed
by byte-swapping the values, and now it turns out
that was the correct fix after all.
Also, any use of iwl_eeprom_query_addr() that is
for more than a u8 must be cast to little endian,
and some structs as well.
Fix all this. Again, no real impact on platforms
that already are little endian.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: stable@kernel.org
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The construct "le16_to_cpu((__force __le16)(r >> 16))" has
always bothered me when looking through the iwlwifi code,
it shouldn't be necessary to __force anything, and before
this code, "r" was obtained with an ioread32, which swaps
each of the two u16 values in it properly when swapping the
entire u32 value. I've had arguments about this code with
people before, but always conceded they were right because
removing it only made things not work at all on big endian
platforms.
However, analysing a failure of the OTP reading code, I now
finally figured out what is going on, and why my intuition
about that code being wrong was right all along.
It turns out that the 'priv->eeprom' u8 array really wants
to have the data in it in little endian. So the force code
above and all really converts *to* little endian, not from
it. Cf., for instance, the function iwl_eeprom_query16() --
it reads two u8 values and combines them into a u16, in a
little-endian way. And considering it more, it makes sense
to have the eeprom array as on the device, after all not
all values really are 16-bit values, the MAC address for
instance is not.
Now, what this really means is that all the annotations are
completely wrong. The eeprom reading code should fill the
priv->eeprom array as a __le16 array, with __le16 values.
This also means that iwl_read_otp_word() should really have
a __le16 pointer as the data argument, since it should be
filling that in a format suitable for priv->eeprom.
Propagating these changes throughout, iwl_find_otp_image()
is found to be, now obviously visible, defective -- it uses
the data returned by iwl_read_otp_word() directly as if it
was CPU endianness. Fixing that, which is this hunk of the
patch:
- next_link_addr = link_value * sizeof(u16);
+ next_link_addr = le16_to_cpu(link_value) * sizeof(u16);
is the only real change of this patch. Everything else is
just fixing the sparse annotations.
Also, the bug only shows up on big endian platforms with a
1000 series card. 5000 and previous series do not use OTP,
and 6000 series has shadow RAM support which means we don't
ever use the defective code on any cards but 1000.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: stable@kernel.org
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
We've had many reports of rt61pci failures with powersaving enabled.
Therefore, as a stop-gap measure, disable powersaving of the rt61pci
until we have found a proper solution.
Also disable powersaving on rt2800pci as it most probably will show
the same problem.
Cc: stable@kernel.org
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
sizeof(iv16) and sizeof(iv32) are the sizes of pointers. Change them to
the size of the copied data.
Furthermore, iveiv_entry is a local structure that has just been
initialized and is not visible outside this function. Thus, there would
seem to be no point to copy data into it. The order of the arguments is
thus changed to copy the data into the parameters, which are provided as
pointers, suggesting in this case that they should be used to return values.
A simplified version of the semantic patch that finds the first problem is as
follows: (http://coccinelle.lip6.fr/)
// <smpl>
@@
expression *x;
expression f;
type T;
@@
*f(...,(T)x,...)
// </smpl>
Signed-off-by: Julia Lawall <julia@diku.dk>
Acked-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
First, we copy/paste the padding stuff from ath9k_tx to ath_tx_cabq since it
needs to same kind of padding, but for internally generated beacons.
Next, software padding done on TX needs to be removed before calling
ieee80211_tx_status. The code was already there in ath_tx_complete but it
was wrong. Fix it by using ath9k_cmn_padpos. This later code has been
tested by sending packets to a monitor interface and reading packets from the
same interface.
Signed-off-by: Benoit PAPILLAULT <benoit.papillault@free.fr>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When trigger event log dumping from debugfs, the entire event log
should be dumped and the size should match the number of events being
dump.
Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Recent commits "iwlwifi: remove power-wasting calls to apm_ops.init()" and
"iwlagn: power up device before initializing EEPROM" had the goal of
reducing device power consumption from the time the module is loaded until
the interface is brought up and the device's power saving mechanisms kick
in. The idea is that once the module is loaded there is no need for the
device to consume power until the interface is brought up.
With the current solution the device is only powered up during EEPROM read,
and then so also only if the EEPROM type is OTP. We have found that on
certain platforms even non-OTP devices require power to be up during EEPROM
read. On these platforms the driver never loads and the system log contains
the following:
iwlagn 0000:03:00.0: MAC is in deep sleep!. CSR_GP_CNTRL = 0x080403D8
We thus now power up all devices during EEPROM read.
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In iwlwifi, priv->alloc_rxb_page is used to keep track of the Rx
pages allocated by the driver. This cleans up the page free routines
by introducing __iwl_free_pages/iwl_free_pages so that the accounting
is more accurate and less error prone. This also fixes two instances where
the counter was not updated.
Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Add a definition of the amount of TX headroom reserved by mac80211 itself
for its own purposes. Also add BUILD_BUG_ON to validate the value.
This define can then be used by drivers to request additional TX headroom
in the most efficient manner.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
rt2800lib incorrectly detected whether RT2800USB was enabled because
it didn't account for a modularized RT2800USB driver.
Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Update "US" and "JP" for current rules, and replace "EU" rules with the
world roaming domain (since it was only a pseudo-domain anyway).
Signed-off-by: John W. Linville <linville@tuxdriver.com>
The patch ("mac80211: Use correct sign for mesh active path
refresh.") was actually a bug. Reverted it and improved the
explanation of how mesh path refresh works.
Signed-off-by: Javier Cardona <javier@cozybit.com>
Signed-off-by: Andrey Yurovsky <andrey@cozybit.com>
Cc: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Paths to mesh portals were being timed out immediately after each use in
intermediate forwarding nodes. mppath->exp_time is set to the expiration time
so assigning it to jiffies was marking the path as expired.
Signed-off-by: Javier Cardona <javier@cozybit.com>
Signed-off-by: Andrey Yurovsky <andrey@cozybit.com>
Cc: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Memset should be given the size of the structure, not the size of the pointer.
The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)
// <smpl>
@@
type T;
T *x;
expression E;
@@
memset(x, E, sizeof(
+ *
x))
// </smpl>
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
As shown in Kernel Bugzilla #14761, doing a controller restart after a
fatal DMA error does not accomplish anything other than consume the CPU
on an affected system. Accordingly, substitute a meaningful message for
the restart.
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Cc: Stable <stable@vger.kernel.org> [2.6.32]
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Michael Buesch reports that his rtl8187 gives warnings on suspend
("queueing ieee80211 work while going to suspend" warnings), as rtl8187
can call ieee80211_queue_delayed_work after mac80211 is suspended.
This change enhances rtl8187 led code so we can avoid queuing work after
mac80211 is suspended: now we register a radio led and make additional
checks to ensure led is off/on properly as mac80211 wants.
Signed-off-by: Herton Ronaldo Krzesinski <herton@mandriva.com.br>
Tested-by: Larry Finger <Larry.Finger@lwfinger.net>
Cc: Stable <stable@vger.kernel.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Without this we have no gaurantee of the integrity of the
EEPROM and are likely to encounter a lot of bogus bug reports
due to actual issues on the EEPROM. With the EEPROM checksum
check in place we can easily rule those issues out.
If you run patch during a revert *you* have a card with a busted
EEPROM and only older kernel will support that concoction. This
patch is a trade off between not accepitng bogus EEPROMs and
avoiding bogus bug reports allowing developers to focus instead
on real concrete issues.
If stable keeps bogus bug reports because of a possibly busted EEPROM
feel free to apply this there too.
Tested on an AR5414
Cc: stable@kernel.org
Cc: jirislaby@gmail.com
Cc: akpm@linux-foundation.org
Cc: rjw@sisk.pl
Cc: me@bobcopeland.com
Cc: david.quan@atheros.com
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
This part was missed in "cfg80211: implement get_wireless_stats",
probably because sta_set_sinfo already existed and was only handling
dBm signals.
Cc: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
When the current_bss is not set, 'iwconfig <iface> key off' does not
clear the private flag. Hence after we connect with WEP to an AP and
then try to connect with another non-WEP AP, it does not work.
This issue will not be seen if supplicant is used.
Signed-off-by: Vivek Natarajan <vnatarajan@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
commit 46ceb60ca8 ("gianfar: Add
Multiple group Support") introduced the following build error
with CONFIG_NET_POLL_CONTROLLER=y:
CC ggianfar.o
ggianfar.c: In function 'gfar_netpoll':
ggianfar.c:2653: error: invalid storage class for function 'gfar_interrupt'
ggianfar.c:2652: warning: ISO C90 forbids mixed declarations and code
ggianfar.c:2681: error: invalid storage class for function 'adjust_link'
ggianfar.c:2764: error: invalid storage class for function 'gfar_set_multi'
ggianfar.c:2855: error: invalid storage class for function 'gfar_clear_exact_match'
ggianfar.c:2877: error: invalid storage class for function 'gfar_set_hash_for_addr'
ggianfar.c:2898: error: invalid storage class for function 'gfar_set_mac_for_addr'
ggianfar.c:2922: error: invalid storage class for function 'gfar_error'
ggianfar.c:3020: warning: ISO C90 forbids mixed declarations and code
ggianfar.c:3032: error: invalid storage class for function 'gfar_init'
ggianfar.c:3037: error: invalid storage class for function 'gfar_exit'
ggianfar.c:3041: error: initializer element is not constant
ggianfar.c:3042: error: initializer element is not constant
ggianfar.c:3042: warning: ISO C90 forbids mixed declarations and code
ggianfar.c:3042: error: expected declaration or statement at end of input
make[1]: *** [ggianfar.o] Error 1
This patch fixes the issue.
Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
About 50% of shutdowns of b44 Ethernet adapter ends by kernel panic
with kernels compiled with stack-protector.
Checking b44_magic_pattern() return values, one call of
b44_magic_pattern() returns 127. It means, that set_bit(128, pmask)
was called on line 1509. It means that bit 0 of 17th byte of pmask was
overwritten. But pmask has only 16 bytes. Stack corruption happens.
It seems that set_bit() on line 1509 always writes one bit off.
The fix does not only solve the stack corruption, but also makes Wake
On LAN working on my onboard B44 on Asus A7V-333X mainboard.
It seems that this problem affects all kernel versions since commit
725ad800 ([PATCH] b44: add wol for old nic) on 2006-06-20.
Signed-off-by: Stanislav Brabec <sbrabec@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
It can happen, that tcp_retransmit_skb fails due to some error.
In such cases we might end up into a state where tp->retrans_out
is zero but that's only because we removed the TCPCB_SACKED_RETRANS
bit from a segment but couldn't retransmit it because of the error
that happened. Therefore some assumptions that retrans_out checks
are based do not necessarily hold, as there still can be an old
retransmission but that is only visible in TCPCB_EVER_RETRANS bit.
As retransmission happen in sequential order (except for some very
rare corner cases), it's enough to check the head skb for that bit.
Main reason for all this complexity is the fact that connection dying
time now depends on the validity of the retrans_stamp, in particular,
that successive retransmissions of a segment must not advance
retrans_stamp under any conditions. It seems after quick thinking that
this has relatively low impact as eventually TCP will go into CA_Loss
and either use the existing check for !retrans_stamp case or send a
retransmission successfully, setting a new base time for the dying
timer (can happen only once). At worst, the dying time will be
approximately the double of the intented time. In addition,
tcp_packet_delayed() will return wrong result (has some cc aspects
but due to rarity of these errors, it's hardly an issue).
One of retrans_stamp clearing happens indirectly through first going
into CA_Open state and then a later ACK lets the clearing to happen.
Thus tcp_try_keep_open has to be modified too.
Thanks to Damian Lukowski <damian@tvk.rwth-aachen.de> for hinting
that this possibility exists (though the particular case discussed
didn't after all have it happening but was just a debug patch
artifact).
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch moves retransmits_timed_out() from include/net/tcp.h
to tcp_timer.c, where it is used.
Reported-by: Frederic Leroy <fredo@starox.org>
Signed-off-by: Damian Lukowski <damian@tvk.rwth-aachen.de>
Acked-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes a problem in the TCP connection timeout calculation.
Currently, timeout decisions are made on the basis of the current
tcp_time_stamp and retrans_stamp, which is usually set at the first
retransmission.
However, if the retransmission fails in tcp_retransmit_skb(),
retrans_stamp is not updated and remains zero. This leads to wrong
decisions in retransmits_timed_out() if tcp_time_stamp is larger than
the specified timeout, which is very likely.
In this case, the TCP connection dies after the first attempted
(and unsuccessful) retransmission.
With this patch, tcp_skb_cb->when is used instead, when retrans_stamp
is not available.
This bug has been introduced together with retransmits_timed_out() in
2.6.32, as the number of retransmissions has been used for timeout
decisions before. The corresponding commit was
6fa12c8503 (Revert Backoff [v3]:
Calculate TCP's connection close threshold as a time value.).
Thanks to Ilpo Järvinen for code suggestions and Frederic Leroy for
testing.
Reported-by: Frederic Leroy <fredo@starox.org>
Signed-off-by: Damian Lukowski <damian@tvk.rwth-aachen.de>
Acked-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
use common_task instead of reset_task and link_chg_task, so it fix "call cancel_work_sync
from the work itself".
Signed-off-by: Jie Yang <jie.yang@atheros.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
add pci map direction in atl1c_buffer flags, it is used when call pci_unmap
apis.
Signed-off-by: Jie Yang <jie.yang@atheros.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Unified firmware image may not contain MN type of firmware.
Driver should fall back to NOMN firmware type instead
of going to flash.
Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o netif_running() check for enabling interrupt at end of napi poll is
not enough to cover firmwar recovery. Instead test __NX_DEV_UP bit.
o Avoid re-entry into to netxen_nic_down() with __NX_DEV_UP bit check.
Acked-by: Dhananjay Phadke <dhananjay.phadke@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
o To prevent race conditions with other reset events.
During suspend/resume and firmware recovery, acquire rtnl_lock,
while changing interface state.
Acked-by: Dhananjay Phadke <dhananjay.phadke@qlogic.com>
Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Various additions and improvements to the Gigaset driver's README
file, and added comments to its userspace visible include file.
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>
When built with debugging support, the Gigaset driver enabled some
debugging messages by default. Change the default to "all off".
Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the locking scheme on the fec_mpc52xx driver. This device can
receive IRQs from three sources; the FEC itself, the tx DMA, and the
rx DMA. Mutual exclusion was handled by taking a spin_lock() in the
critical regions, but because the handlers are run with IRQs enabled,
spin_lock() is insufficient and the driver can end up interrupting
a critical region anyway from another IRQ.
Asier Llano discovered that this occurs when an error IRQ is raised
in the middle of handling rx irqs which resulted in an sk_buff memory
leak.
In addition, locking is spotty at best in the driver and inspection
revealed quite a few places with insufficient locking.
This patch is based on Asier's initial work, but reworks a number of
things so that locks are held for as short a time as possible, so
that spin_lock_irqsave() is used everywhere, and so the locks are
dropped when calling into the network stack (because the lock only
protects the hardware interface; not the network stack).
Boot tested on a lite5200 with an NFS root. Has not been performance
tested.
Signed-off-by: Asier Llano <a.llano@ziv.es>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use a more effective rss hash by default (src + dst, rather than just
src).
Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
in routed mode, we don't have a hardware address so netdev_ops doesnt
need to validate our hardware address via .ndo_validate_addr
Reported-by: Manuel Fuentes <mfuentes@agenciaefe.com>
Signed-off-by: Chas Williams - CONTRACTOR <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>
due to reference counting sk_wmem_alloc now has a value of 1 when all
the outstanding data has been sent.
Signed-off-by: Chas Williams - CONTRACTOR <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>
fix oops when initializing lane interfaces. lec should probably be
changed to use alloc_netdev() instead.
Signed-off-by: Chas Williams - CONTRACTOR <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds kerneldoc for inet_twsk_unhash() & inet_twsk_bind_unhash().
With help from Randy Dunlap.
Suggested-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When we find a timewait connection in __inet_hash_connect() and reuse
it for a new connection request, we have a race window, releasing bind
list lock and reacquiring it in __inet_twsk_kill() to remove timewait
socket from list.
Another thread might find the timewait socket we already chose, leading to
list corruption and crashes.
Fix is to remove timewait socket from bind list before releasing the bind lock.
Note: This problem happens if sysctl_tcp_tw_reuse is set.
Reported-by: kapil dakhane <kdakhane@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
First patch changes __inet_hash_nolisten() and __inet6_hash()
to get a timewait parameter to be able to unhash it from ehash
at same time the new socket is inserted in hash.
This makes sure timewait socket wont be found by a concurrent
writer in __inet_check_established()
Reported-by: kapil dakhane <kdakhane@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Acked-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>
