net: introduce DST_NOCACHE flag

While doing stress tests with IP route cache disabled, and multi queue
devices, I noticed a very high contention on one rwlock used in
neighbour code.

When many cpus are trying to send frames (possibly using a high
performance multiqueue device) to the same neighbour, they fight for the
neigh->lock rwlock in order to call neigh_hh_init(), and fight on
hh->hh_refcnt (a pair of atomic_inc/atomic_dec_and_test())

But we dont need to call neigh_hh_init() for dst that are used only
once. It costs four atomic operations at least, on two contended cache
lines, plus the high contention on neigh->lock rwlock.

Introduce a new dst flag, DST_NOCACHE, that is set when dst was not
inserted in route cache.

With the stress test bench, sending 160000000 frames on one neighbour,
results are :

Before patch:

real 2m28.406s
user 0m11.781s
sys 36m17.964s

After patch:

real 1m26.532s
user 0m12.185s
sys 20m3.903s

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sctp: Fix break indentation in sctp_ioctl().

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

be2net: add multiple RX queue support

This patch adds multiple RX queue support to be2net. There are
upto 4 extra rx-queues per port into which TCP/UDP traffic can be hashed into.
Some of the ethtool stats are now displayed on a per queue basis.

Signed-off-by: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'for-davem' of git://git./linux/kernel/git/linville/wireless-next-2.6

qeth: tagging with VLAN-ID 0

This patch adapts qeth to handle tagged frames with VLAN-ID 0 and
with or without priority information in the tag. It enables qeth to
receive priority-tagged frames on a base interface, for example from
z/OS, without configuring an additional VLAN interface.

Signed-off-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: remove a bogus PCI function number check

Remove a bogus PCI function number check from the driver's .remove
method that causes pci_release_regions not to be called for function 0
if additional functions are attached and one of them is used as primary.

Signed-off-by: Dimitris Michailidis <dm@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/atm/idt77252.c: Remove unnecessary error check

This code does not call deinit_card(card); in an error case, as done in
other error-handling code in the same function.  But actually, the called
function init_sram can only return 0, so there is no need for the error
check at all.

init_sram is also given a void return type, and its single return statement
at the end of the function is dropped.

A simplified version of the sematic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@r exists@
@r@
statement S1,S2,S3;
constant C1,C2,C3;
@@

*if (...)
 {... S1 return -C1;}
...
*if (...)
 {... when != S1
    return -C2;}
...
*if (...)
 {... S1 return -C3;}
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers-net-tulip-de4x5c-fix-copy-length-in-de4x5_ioctl-checkpatch-fixes

ERROR: trailing statements should be on next line
#23: FILE: drivers/net/tulip/de4x5.c:5477:
+ if (copy_to_user(ioc->data, tmp.lval, ioc->len)) return -EFAULT;

total: 1 errors, 0 warnings, 8 lines checked

./patches/drivers-net-tulip-de4x5c-fix-copy-length-in-de4x5_ioctl.patch has style problems, please review.  If any of these errors
are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.

Please run checkpatch prior to sending patches

Cc: Dan Rosenberg <dan.j.rosenberg@gmail.com>
Cc: Grant Grundler <grundler@parisc-linux.org>
Cc: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipmr: cleanups

Various code style cleanups

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipmr: RCU protection for mfc_cache_array

Use RCU & RTNL protection for mfc_cache_array[]

ipmr_cache_find() is called under rcu_read_lock();

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipmr: RCU conversion of mroute_sk

Use RCU and RTNL to protect (struct mr_table)->mroute_sk

Readers use RCU, writers use RTNL.

ip_ra_control() already use an RCU grace period before
ip_ra_destroy_rcu(), so we dont need synchronize_rcu() in
mrtsock_destruct()

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipmr: __pim_rcv() is called under rcu_read_lock

No need to get a reference on reg_dev and release it, we are in a
rcu_read_lock() protected section.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gre: protocol table can be static

This table is only used in gre.c

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-next-2.6 into for-davem

enic: Update MAINTAINERS

Update MAINTAINERS list

Signed-off-by: Vasanthy Kolluri <vkolluri@cisco.com>
Signed-off-by: Roopa Prabhu <roprabhu@cisco.com>
Signed-off-by: David Wang <dwang2@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

enic: Make local functions static

Make functions used locally in a file as static

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Vasanthy Kolluri <vkolluri@cisco.com>
Signed-off-by: Roopa Prabhu <roprabhu@cisco.com>
Signed-off-by: David Wang <dwang2@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

enic: Remove dead code

Removed code that is unused

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Vasanthy Kolluri <vkolluri@cisco.com>
Signed-off-by: Roopa Prabhu <roprabhu@cisco.com>
Signed-off-by: David Wang <dwang2@cisco.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

neigh: reorder fields in struct neighbour

On 64bit arches, there are two 32bit holes that we can remove.

sizeof(struct neighbour) shrinks from 0xf8 to 0xf0 bytes

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: improve bas_gigaset USB error reporting

Rephrase some USB error messages to make them clearer and more consistent.
Downgrade some warning messages that may occur during normal operation to
debug messages.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: fix bas_gigaset interrupt read error handling

Rework the handling of USB errors in interrupt input reads
to clear halts correctly, delay URB resubmission after errors,
limit retries, and improve error recovery.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: unclog bas_gigaset AT response pipe

Recover from a lost HD_RECEIVEATDATA_ACK message by sending a
zero-length HD_READ_ATMESSAGE command when ev_layer sends "+++".

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: try USB reset for bas_gigaset error recovery

In error_reset(), if sending HD_RESET_INTERRUPT_PIPE to the device
fails, try performing an USB reset.
Also correct an error in the leading comment.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: bas_gigaset timer cleanup

Use setup_timer() and mod_timer() instead of direct assignment to
timer structure members, simplify the argument of one timer routine,
and make extra sure all timers are stopped during suspend.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: drop obsolete debug option

Remove the debug flag DEBUG_DRIVER and associated code.
It doesn't serve any useful purpose anymore.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: correct bas_gigaset rx buffer handling

In transparent data reception, avoid a NULL pointer dereference
in case an skbuff cannot be allocated, remove an inappropriate
call to the HDLC flush routine, and correct the accounting of
received bytes for continued buffers.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
CC: stable <stable@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: fix bas_gigaset AT read error handling

Rework the handling of USB errors in AT response reads
to fix a possible infinite retry loop and a memory leak,
and silence a few overly verbose kernel messages.

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
CC: stable <stable@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

isdn/gigaset: bas_gigaset locking fix

Unlock cs->lock before calling error_hangup() which is marked
"cs->lock must not be held".

Signed-off-by: Tilman Schmidt <tilman@imap.cc>
CC: stable <stable@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Update version to 3.114

This patch updates the tg3 version to 3.114.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Add extend rx ring sizes for 5717 and 5719

This patch increases the rx ring sizes for those asic revs that support
them.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Prepare for larger rx ring sizes

This patch adds two new variables to track the size of the standard and
jumbo rx producer ring sizes.  The code is then pivoted to these
variables from preprocessor constants.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Futureproof the loopback test

There are other multiqueue modes 5717 and 5719 devices can assume.  This
patch makes sure that the loopback test is safe, should those other
modes be enabled in the future.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Cleanup missing VPD partno section

This patch cleans up the default VPD partno section.  New entries for
5717 asic rev devices were also added.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Remove 5724 device ID

This product was never released to the public.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: 5719: Prevent tx data corruption

This patch enables a bit that prevents read DMA overflows and adjusts
the txmbuf margin from the hardware default.  The combination of these
modifications prevents a tx data corruption issue we were seeing on the
5719.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Fix potential netpoll crash

Up until now the tg3 driver would call netif_napi_add() for the maximum
number of NAPI instances the driver could use.  The problem is that
netpoll could call tg3_poll() on instances that are not active.  The net
effect is that the driver will crash attempting to dereference
uninitialized pointers.

The fix is to only allocate as many NAPI instances as the driver would
use in tg3_open() and deleted them in tg3_close().

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Reviewed-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: rcu conversion in ip_route_output_slow

ip_route_output_slow() is enclosed in an rcu_read_lock() protected
section, so that no references are taken/released on device, thanks to
__ip_dev_find() & dev_get_by_index_rcu()

Tested with ip route cache disabled, and a stress test :

Before patch:

elapsed time :

real 1m38.347s
user 0m11.909s
sys 23m51.501s

Profile:

13788.00 22.7% ip_route_output_slow [kernel]
 7875.00 13.0% dst_destroy          [kernel]
 3925.00  6.5% fib_semantic_match   [kernel]
 3144.00  5.2% fib_rules_lookup     [kernel]
 3061.00  5.0% dst_alloc            [kernel]
 2276.00  3.7% rt_set_nexthop       [kernel]
 1762.00  2.9% fib_table_lookup     [kernel]
 1538.00  2.5% _raw_read_lock       [kernel]
 1358.00  2.2% ip_output            [kernel]

After patch:

real 1m28.808s
user 0m13.245s
sys 20m37.293s

10950.00 17.2% ip_route_output_slow [kernel]
10726.00 16.9% dst_destroy          [kernel]
 5170.00  8.1% fib_semantic_match   [kernel]
 3937.00  6.2% dst_alloc            [kernel]
 3635.00  5.7% rt_set_nexthop       [kernel]
 2900.00  4.6% fib_rules_lookup     [kernel]
 2240.00  3.5% fib_table_lookup     [kernel]
 1427.00  2.2% _raw_read_lock       [kernel]
 1157.00  1.8% kmem_cache_alloc     [kernel]

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: introduce __ip_dev_find()

ip_dev_find(net, addr) finds a device given an IPv4 source address and
takes a reference on it.

Introduce __ip_dev_find(), taking a third argument, to optionally take
the device reference. Callers not asking the reference to be taken
should be in an rcu_read_lock() protected section.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: 82579 performance improvements

The initial support for 82579 was tuned poorly for performance.  Adjust the
packet buffer allocation appropriately for both standard and jumbo frames;
and for jumbo frames increase the receive descriptor pre-fetch, disable
adaptive interrupt moderation and set the DMA latency tolerance.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

e1000e: use hardware writeback batching

Most e1000e parts support batching writebacks.  The problem with this is
that when some of the TADV or TIDV timers are not set, Tx can sit forever.

This is solved in this patch with write flushes using the Flush Partial
Descriptors (FPD) bit in TIDV and RDTR.

This improves bus utilization and removes partial writes on e1000e,
particularly from 82571 parts in S5500 chipset based machines.

Only ES2LAN and 82571/2 parts are included in this optimization, to reduce
testing load.

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

ixgbe: fix link issues and panic with shared interrupts for 82598

Fix possible panic/hang with shared Legacy interrupts by not enabling
interrupts when interface is down.

Also fixes an intermittent link by enabling LSC upon exit from ixgbe_intr()

This patch adds flags to ixgbe_irq_enable() to allow for some flexibility
when enabling interrupts.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Stephen Ko <stephen.s.ko@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: __mkroute_output() speedup

While doing stress tests with a disabled IP route cache, I found
__mkroute_output() was touching three times in_device atomic refcount.

Use RCU to touch it once to reduce cache line ping pongs.

Before patch

time to perform the test
real 1m42.009s
user 0m12.545s
sys 25m0.726s

Profile :

16109.00 26.4% ip_route_output_slow   vmlinux
 7434.00 12.2% dst_destroy            vmlinux
 3280.00  5.4% fib_rules_lookup       vmlinux
 3252.00  5.3% fib_semantic_match     vmlinux
 2622.00  4.3% fib_table_lookup       vmlinux
 2535.00  4.1% dst_alloc              vmlinux
 1750.00  2.9% _raw_read_lock         vmlinux
 1532.00  2.5% rt_set_nexthop         vmlinux

After patch

real 1m36.503s
user 0m12.977s
sys 23m25.608s

14234.00 22.4% ip_route_output_slow   vmlinux
 8717.00 13.7% dst_destroy            vmlinux
 4052.00  6.4% fib_rules_lookup       vmlinux
 3951.00  6.2% fib_semantic_match     vmlinux
 3191.00  5.0% dst_alloc              vmlinux
 1764.00  2.8% fib_table_lookup       vmlinux
 1692.00  2.7% _raw_read_lock         vmlinux
 1605.00  2.5% rt_set_nexthop         vmlinux

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Phonet: restore flow control credits when sending fails

This patch restores the below flow control patch submitted by Rémi
Denis-Courmont, which accidentaly got lost due to Pipe controller patch
on Phonet.

commit 1a98214feef2221cd7c24b17cd688a5a9d85b2ea
Author: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Date:   Mon Aug 30 12:57:03 2010 +0000

Phonet: restore flow control credits when sending fails

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Kumar Sanghvi <kumar.sanghvi@stericsson.com>
Acked-by: Linus Walleij <linus.walleij@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: pxa168_etc.c recognize additional contributors

Signed-off-by: Philip Rakity <prakity@marvell.com>
Signed-off-by: Sachin Sanap <ssanap@marvell.com>
Signed-off-by: Mark Brown <markb@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip_gre: comments change

HARD_TX_LOCK no longer protects tunnels from dead loops,
but xmit_recursion percpu counter.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

de2104x: remove experimental status

It should be ready after 8 years...remove the experimental dependency.

Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
Acked-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

de2104x: disable media debug messages by default

Print media debug messages only when HW debug is enabled.

Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
Acked-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

myri10ge: DCA update (resubmit)

This patch contains the following DCA improvements to myri10ge:

1) Finally move myri10ge to use dca3 API

2) Disable PCIe relaxed ordering when enabling DCA on
     myri10ge.  This provides a performance boost on Nehalem
     based Xeons

3) Make sure to properly initialize NIC's DCA state when it is enabled,
     rather than giving the NIC a bogus tag (0) and waiting for
     the first received packet to trigger an update.  Not using a
     real tag can cause hardware exceptions on some motherboards
     when a CPU socket is empty.

3) Always update the cached CPU when our interrupt affinity changes
     so as to avoid excessive calls to dca3_get_tag()

Signed-off-by: Andrew Gallatin <gallatin@myri.com>
Signed-off-by: Loic Prylli <loic@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Moved enabling of MSI to the bnx2x_set_num_queues()

Moved enabling of MSI to the bnx2x_set_num_queues() - the same functions that
 handles the initialization of the MSI-X.

From: Vladislav Zolotarov <vladz@broadcom.com>
Signed-off-by: Dmitry Kravkov <dmitry@broadcom.com>
Signed-off-by: Vladislav Zolotarov <vladz@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp: tcp_enter_quickack_mode can be static

Function only used in tcp_input.c

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

arp: remove unnecessary export of arp_broken_ops

arp_broken_ops is only used in arp.c

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: rename netdev rx_queue to ingress_queue

There is some confusion with rx_queue name after RPS, and net drivers
private rx_queue fields.

I suggest to rename "struct net_device"->rx_queue to ingress_queue.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip6tnl: percpu stats accounting

Maintain per_cpu tx_bytes, tx_packets, rx_bytes, rx_packets.

Other seldom used fields are kept in netdev->stats structure, possibly
unsafe.

This is a preliminary work to support lockless transmit path, and
correct RX stats, that are already unsafe.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipip: enable lockless xmits

IPIP tunnels can benefit from lockless xmits, using NETIF_F_LLTX

Bench on a 16 cpus machine (dual E5540 cpus), 16 threads sending
10000000 UDP frames via one ipip tunnel (size:200 bytes per frame)

Before patch :
real 2m53.321s
user 0m10.277s
sys 46m0.597s

After patch:
real 0m32.063s
user 0m9.237s
sys 8m16.255s

Last problem to solve is the contention on dst :

16118.00 28.3% __ip_route_output_key         vmlinux
 6135.00 10.8% dst_release                   vmlinux
 3220.00  5.6% ip_finish_output              vmlinux
 2149.00  3.8% ip_route_output_flow          vmlinux
 1575.00  2.8% ip_append_data                vmlinux
 1481.00  2.6% ip_push_pending_frames        vmlinux
 1349.00  2.4% __xfrm_lookup                 vmlinux
 1216.00  2.1% csum_partial_copy_generic     vmlinux
 1208.00  2.1% udp_sendmsg                   vmlinux

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip_gre: lockless xmit

GRE tunnels can benefit from lockless xmits, using NETIF_F_LLTX

Note: If tunnels are created with the "oseq" option, LLTX is not
enabled :

Even using an atomic_t o_seq, we would increase chance for packets being
out of order at receiver.

Bench on a 16 cpus machine (dual E5540 cpus), 16 threads sending
10000000 UDP frames via one gre tunnel (size:200 bytes per frame)

Before patch :
real 3m0.094s
user 0m9.365s
sys 47m50.103s

After patch:
real 0m29.756s
user 0m11.097s
sys 7m33.012s

Last problem to solve is the contention on dst :

38660.00 21.4% __ip_route_output_key          vmlinux
20786.00 11.5% dst_release                    vmlinux
14191.00  7.8% __xfrm_lookup                  vmlinux
12410.00  6.9% ip_finish_output               vmlinux
 4540.00  2.5% ip_push_pending_frames         vmlinux
 4427.00  2.4% ip_append_data                 vmlinux
 4265.00  2.4% __alloc_skb                    vmlinux
 4140.00  2.3% __ip_local_out                 vmlinux
 3991.00  2.2% dev_queue_xmit                 vmlinux

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sit: enable lockless xmits

SIT tunnels can benefit from lockless xmits, using NETIF_F_LLTX

Bench on a 16 cpus machine (dual E5540 cpus), 16 threads sending
10000000 UDP frames via one sit tunnel (size:220 bytes per frame)

Before patch :

real 3m15.399s
user 0m9.185s
sys 51m55.403s

75029.00 87.5% _raw_spin_lock            vmlinux
 1090.00  1.3% dst_release               vmlinux
  902.00  1.1% dev_queue_xmit            vmlinux
  627.00  0.7% sock_wfree                vmlinux
  613.00  0.7% ip6_push_pending_frames   ipv6.ko
  505.00  0.6% __ip_route_output_key     vmlinux

After patch:

real 1m1.387s
user 0m12.489s
sys 15m58.868s

28239.00 23.3% dst_release               vmlinux
13570.00 11.2% ip6_push_pending_frames   ipv6.ko
13118.00 10.8% ip6_append_data           ipv6.ko
 7995.00  6.6% __ip_route_output_key     vmlinux
 7924.00  6.5% sk_dst_check              vmlinux
 5015.00  4.1% udpv6_sendmsg             ipv6.ko
 3594.00  3.0% sock_alloc_send_pskb      vmlinux
 3135.00  2.6% sock_wfree                vmlinux
 3055.00  2.5% ip6_sk_dst_lookup         ipv6.ko
 2473.00  2.0% ip_finish_output          vmlinux

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sit: fix percpu stats accounting

commit 15fc1f7056ebd (sit: percpu stats accounting) forgot the fallback
tunnel case (sit0), and can crash pretty fast.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipip: fix percpu stats accounting

commit 3c97af99a5aa1 (ipip: percpu stats accounting) forgot the fallback
tunnel case (tunl0), and can crash pretty fast.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

dummy: percpu stats and lockless xmit

Converts dummy network device driver to :

- percpu stats

- 64bit stats

- lockless xmit (NETIF_F_LLTX)

- performance features added (NETIF_F_SG | NETIF_F_FRAGLIST |
NETIF_F_TSO | NETIF_F_NO_CSUM | NETIF_F_HIGHDMA)

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: add a recursion limit in xmit path

As tunnel devices are going to be lockless, we need to make sure a
misconfigured machine wont enter an infinite loop.

Add a percpu variable, and limit to three the number of stacked xmits.

Reported-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6: Implement Any-IP support for IPv6.

AnyIP is the capability to receive packets and establish incoming
connections on IPs we have not explicitly configured on the machine.

An example use case is to configure a machine to accept all incoming
traffic on eth0, and leave the policy of whether traffic for a given IP
should be delivered to the machine up to the load balancer.

Can be setup as follows:
  ip -6 rule from all iif eth0 lookup 200
  ip -6 route add local default dev lo table 200
(in this case for all IPv6 addresses)

Signed-off-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: Allow configuring subnets as local addresses

This patch allows a host to be configured to respond to any address in
a specified range as if it were local, without actually needing to
configure the address on an interface.  This is done through routing
table configuration.  For instance, to configure a host to respond
to any address in 10.1/16 received on eth0 as a local address we can do:

ip rule add from all iif eth0 lookup 200
ip route add local 10.1/16 dev lo proto kernel scope host src 127.0.0.1 table 200

This host is now reachable by any 10.1/16 address (route lookup on
input for packets received on eth0 can find the route).  On output, the
rule will not be matched so that this host can still send packets to
10.1/16 (not sent on loopback).  Presumably, external routing can be
configured to make sense out of this.

To make this work, we needed to modify the logic in finding the
interface which is assigned a given source address for output
(dev_ip_find).  We perform a normal fib_lookup instead of just a
lookup on the local table, and in the lookup we ignore the input
interface for matching.

This patch is useful to implement IP-anycast for subnets of virtual
addresses.

Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

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

mac80211: Fix WMM driver queue configuration

The WMM parameter configuration function (ieee80211_sta_wmm_params) only
configures the WMM parameters to the driver is the wmm_last_param_set
counter value is changed by the AP.

The wmm_last_param_set is initialized to -1 on association in order to ensure
the configuration is made to the driver at least once on association, but
currently this initialization is done *after* the WMM parameter configuration
function was called.

This leads to unreliability in the driver getting properly configured on first
association (depending on what counter value the AP happens to use.) When
disassociating (the wmm default parameters are configured to the driver) and
then reassociating, due to the above the WMM configuration is not set to the
driver at all.

On drivers without beacon filtering the problem is corrected by later beacons,
but on drivers with beacon filtering the WMM will remain permanently incorrectly
configured.

Fix this by moving the initialization of wmm_last_param_set to -1 before
ieee80211_sta_wmm_params is called on association.

Signed-off-by: Juuso Oikarinen <juuso.oikarinen@nokia.com>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

cfg80211: always set IBSS basic rates

IBSS started from wireless extensions is currently
missing basic rate configuration, fix this by moving
the code to generate the default to the common code
that gets invoked for both nl80211 and wext.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

carl9170: interrupt urbs must not set URB_ZERO_PACKET

This patch fixes a bug in the driver which was
exposed by CONFIG_USB_DEBUG:
"usb 1-1.6.3: BOGUS urb flags, 40 --> 0"

The transfer flag "URB_ZERO_PACKET" is only valid
for bulk urbs.

Reported-by: André Erdmann
Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath5k: Fix bitmasks and typos for PCU Diagnostic register

As reported by Ryan Niemi, some bitmasks in the register definition for the PCU
Diagnostic register (DIAG_SW) were missing a zero at the end. While at it fix
some typos and add more comments.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath5k: Disable interrupts in ath5k_hw_get_tsf64

The code in ath5k_hw_get_tsf64() is time critical and will return wrong results
if we get interrupted, so disable local interrupts.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath5k: Increase "fudge" for beacon timers

We use FUDGE to make sure the next TBTT is ahead of the current TU.
Since we later substract AR5K_TUNE_SW_BEACON_RESP (10) in the timer
configuration we need to make sure it is bigger than that.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath5k: Check and fix ATIM window

This patch adds sanity-checks for the beacon timers and especially the ATIM
window to ath5k. It is basically the same what i did for madwifi two years ago
and fixes a problem in IBSS mode which has been described as "ramping" pings.

See the code comments for a more detailed description and these links:

http://madwifi-project.org/ticket/1154
http://madwifi-project.org/changeset/3867
http://thread.gmane.org/gmane.linux.drivers.madwifi.devel/6066

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

ath5k: Remove unused variable for atim window

It's not used and it's unlikely we will ever implement ATIM.

Signed-off-by: Bruno Randolf <br1@einfach.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

carl9170: revamp carl9170_tx_prepare

David Miller complained about the driver's excessive use
of variables in __packed structs. While I did not fully
agree with his sole "performance" argument on all accounts.
I do see some room for improvement in hot-paths on
architectures without an efficient access to unaligned
elements.

This first patch (dare I say?) optimizes an important tx
hot-path in the driver: carl9170_tx_prepare.

Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

carl9170: fix WARN_ON triggered by Broadcom HT STAs

Broadcom's Windows driver for the 4313 advertises
an ampdu density of 7 => 16 us. The AR9170 MAC on
the other hand only supports densities up to 8 us.

This patch removes the noisy WARN_ON, because
there is nothing we can do about it.

Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

carl9170: fix tx_ampdu_upload counter

tx_ampdu_upload was not decreased when an a-MPDU
frame had to be kicked out from the tx_pending
queues.

This broke ampdu aggregation, because the scheduler
waits until tx_ampdu_upload drops to zero, before
making the next aggregate.

Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

carl9170: remove lost-frame workaround

This patch removes some cruft, which survived
the RFC review phase.

Originally, carl9170_tx_ampdu_queue erroneously
dropped a lot of frames. As a result the ampdu
scheduler bogged down quite frequently and the
affected BA session timed out.

However this bug has been fixed and the WA and
its debugfs counter is no longer useful.

Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

carl9170: angle brackets for wiki link

"The convention seems to be angle brackets around
URLS in Kconfig."
 -- Finn Thain (to update web addresses in the kernel)

Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

carl9170: remove non-standard ba session teardown

802.11n-2009 demands in 11.2.1: "
When a STA enters normal (non-APSD) PS mode, any downlink
Block ACK agreement without an associated schedule is
suspended for the duration of this PS mode."

The operative word is "suspended" and not terminated.

Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

mac80211: fix offchannel assumption upon association

Association is dealt with as an atomic offchannel operation,
we do this because we don't know we are associated until we
get the associatin response from the AP. When we do get the
associatin response though we were never clearing the offchannel
state. This has a few implications, we told drivers we were
still offchannel, and the first configured TX power for the
channel does not take into account any power constraints.

For ath9k this meant ANI calibration would not start upon
association, and we'd have to wait until the first bgscan
to be triggered. There may be other issues this resolves
but I'm too lazy to comb the code to check.

Cc: stable@kernel.org
Cc: Amod Bodas <amod.bodas@atheros.com>
Cc: Vasanth Thiagarajan <vasanth.thiagarajan@atheros.com>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

iwl3945: queue the right work if the scan needs to be aborted

iwl3945's scan_completed calls into the mac80211 stack which triggers a
warn on if there is no scan outstanding.

This can be avoided by not calling scan_completed but abort_scan in
iwl3945_request_scan  in the done: branch of the function which is used
as an error out.

The done: branch seems to be an error-out branch, as, for example, if
iwl_is_ready(priv) returns false  the done: branch is executed.

NOTE:
I'm not familiar with the driver at all.
I just quickly scanned as a reaction to

https://bugzilla.kernel.org/show_bug.cgi?id=17722

the users of scan_completed in the  iwl3945 driver and noted the odd
discrepancy between the comment above this instance and the comment in
mac80211 scan_completed function.

Signed-off-by: Florian Mickler <florian@mickler.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

pch_gbe: add header files

Fix build errors, more header files needed.

drivers/net/pch_gbe/pch_gbe_main.c:965: error: implicit declaration of function 'tcp_hdr'
drivers/net/pch_gbe/pch_gbe_main.c:965: error: invalid type argument of '->' (have 'int')
drivers/net/pch_gbe/pch_gbe_main.c:968: error: invalid type argument of '->' (have 'int')
drivers/net/pch_gbe/pch_gbe_main.c:976: error: implicit declaration of function 'udp_hdr'
drivers/net/pch_gbe/pch_gbe_main.c:976: error: invalid type argument of '->' (have 'int')
drivers/net/pch_gbe/pch_gbe_main.c:980: error: invalid type argument of '->' (have 'int')

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

wl12xx: fix separate-object-folder builds

Make this go away (happens when building with a separate object
directory):

Assembler messages:
Fatal error: can't create drivers/net/wireless/wl12xx/.tmp_wl12xx_platform_data.o: No such file or directory
drivers/net/wireless/wl12xx/wl12xx_platform_data.c: In function 'wl12xx_get_platform_data':
drivers/net/wireless/wl12xx/wl12xx_platform_data.c:28: error: cannot open drivers/net/wireless/wl12xx/.tmp_wl12xx_platform_data.gcno
drivers/net/wireless/wl12xx/wl12xx_platform_data.c:28: confused by earlier errors, bailing out

Signed-off-by: John W. Linville <linville@tuxdriver.com>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>

Documentation: Update Phonet doc for Pipe Controller implementation

Updates the Phonet document with description related to Pipe controller
implementation

Signed-off-by: Kumar Sanghvi <kumar.sanghvi@stericsson.com>
Acked-by: Linus Walleij <linus.walleij@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tg3: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

8021q: Use netif_copy_real_num_queues() to set queue counts

This covers RX if necessary, as well as TX.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

niu: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

myri10ge: Use netif_set_real_num_{rx, tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mv643xx_eth: Use netif_set_real_num_{rx, tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Acked-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

mlx4_en: Use netif_set_real_num_{rx, tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ixgbe: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

igb: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

gianfar: Use netif_set_real_num_rx_queues()

Do not set num_tx_queues or real_num_tx_queues, since
alloc_etherdev_mq() does that.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4vf: Use netif_set_real_num_{rx, tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb4: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cxgb3: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2x: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

bnx2: Use netif_set_real_num_{rx,tx}_queues()

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Add netif_copy_real_num_queues() for use by virtual net drivers

This sets the active numbers of queues on a net device to match another.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Allow changing number of RX queues after device allocation

For RPS, we create a kobject for each RX queue based on the number of
queues passed to alloc_netdev_mq().  However, drivers generally do not
determine the numbers of hardware queues to use until much later, so
this usually represents the maximum number the driver may use and not
the actual number in use.

For TX queues, drivers can update the actual number using
netif_set_real_num_tx_queues().  Add a corresponding function for RX
queues, netif_set_real_num_rx_queues().

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: sk_{detach|attach}_filter() rcu fixes

sk_attach_filter() and sk_detach_filter() are run with socket locked.

Use the appropriate rcu_dereference_protected() instead of blocking BH,
and rcu_dereference_bh().
There is no point adding BH prevention and memory barrier.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

fib: use atomic_inc_not_zero() in fib_rules_lookup

It seems we dont use appropriate refcount increment in an
rcu_read_lock() protected section.

fib_rule_get() might increment a null refcount and bad things could
happen.

While fib_nl_delrule() respects an rcu grace period before calling
fib_rule_put(), fib_rules_cleanup_ops() calls fib_rule_put() without a
grace period.

Note : after this patch, we might avoid the synchronize_rcu() call done
in fib_nl_delrule()

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>