GitHub/exynos8895/android_kernel_samsung_universal8895.git
15 years agocnic: Fix ISCSI_KEVENT_IF_DOWN message handling.
Michael Chan [Mon, 27 Jul 2009 18:25:58 +0000 (11:25 -0700)]
cnic: Fix ISCSI_KEVENT_IF_DOWN message handling.

When a net device goes down or when the bnx2i driver is unloaded,
the code was not generating the ISCSI_KEVENT_IF_DOWN message
properly and this could cause the userspace driver to crash.

This is fixed by sending the message properly in the shutdown path.
cnic_uio_stop() is also added to send the message when bnx2i is
unregistering.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet: irda: init spinlock after memcpy
Deepak Saxena [Mon, 27 Jul 2009 17:49:44 +0000 (10:49 -0700)]
net: irda: init spinlock after memcpy

irttp_dup() copies a tsap_cb struct, but does not initialize the
spinlock in the new structure, which confuses lockdep.

Signed-off-by: Deepak Saxena <dsaxena@mvista.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoixgbe: fix for 82599 errata marking UDP checksum errors
Don Skidmore [Thu, 23 Jul 2009 18:00:39 +0000 (18:00 +0000)]
ixgbe: fix for 82599 errata marking UDP checksum errors

There is an 82599 errata that UDP frames with a zero checksum are
incorrectly marked as checksum invalid by the hardware.  This was
leading to misleading hw_csum_rx_error counts. This patch adds a
test around this counter increase for this condition.

Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agor8169: WakeOnLan fix for the 8168
françois romieu [Fri, 24 Jul 2009 12:34:19 +0000 (12:34 +0000)]
r8169: WakeOnLan fix for the 8168

More stuff for http://bugzilla.kernel.org/show_bug.cgi?id=9512

Some 8168 are unable to WoL when receiving is not enabled (plain
old 8169 do not seem to care).

It is not exactly pretty to leave the receiver enabled but we
should now enable DMA late enough for it to be safe. Some late
stage boot failure due to pxe and friends may benefit from the
delayed enabling of bus-mastering as well.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Tested-by: Jaromír Cápík <tavvva@volny.cz>
Cc: Edward Hsu <edward_hsu@realtek.com.tw>
15 years agonetxen: reset ring consumer during cleanup
Dhananjay Phadke [Mon, 27 Jul 2009 03:17:21 +0000 (20:17 -0700)]
netxen: reset ring consumer during cleanup

Reset consumer of status rings to 0 when cleaning
up sw resources. Status rings are not deleted
during suspend since they have napi objects.

This ensures correct rx processing across suspen-resume.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet/bridge: use kobject_put to release kobject in br_add_if error path
Xiaotian Feng [Thu, 23 Jul 2009 23:06:32 +0000 (23:06 +0000)]
net/bridge: use kobject_put to release kobject in br_add_if error path

kobject_init_and_add will alloc memory for kobj->name, so in br_add_if
error path, simply use kobject_del will not free memory for kobj->name.
Fix by using kobject_put instead, kobject_put will internally calls
kobject_del and frees memory for kobj->name.

Signed-off-by: Xiaotian Feng <dfeng@redhat.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agosmc91x.h: add config for Nomadik evaluation kit
Alessandro Rubini [Wed, 22 Jul 2009 12:49:08 +0000 (12:49 +0000)]
smc91x.h: add config for Nomadik evaluation kit

Signed-off-by: Alessandro Rubini <rubini@unipv.it>
Acked-by: Andrea Gallo <andrea.gallo@stericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoNET: ROSE: Don't use static buffer.
Ralf Baechle [Mon, 27 Jul 2009 02:11:14 +0000 (19:11 -0700)]
NET: ROSE: Don't use static buffer.

The use of a static buffer in rose2asc() to return its result is not
threadproof and can result in corruption if multiple threads are trying
to use one of the procfs files based on rose2asc().

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoeepro: Read buffer overflow
Roel Kluin [Mon, 27 Jul 2009 01:53:17 +0000 (18:53 -0700)]
eepro: Read buffer overflow

io[i] is read before the bounds check on i, order should be reversed

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agotokenring: Read buffer overflow
roel kluin [Sat, 25 Jul 2009 12:38:33 +0000 (12:38 +0000)]
tokenring: Read buffer overflow

io[i] is read before the bounds check on i, order should be reversed

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoat1700: Read buffer overflow
roel kluin [Sat, 25 Jul 2009 12:01:50 +0000 (12:01 +0000)]
at1700: Read buffer overflow

loop bound looks to be wrong, for an array of length 8

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agofealnx: Write outside array bounds
roel kluin [Sat, 25 Jul 2009 07:41:12 +0000 (07:41 +0000)]
fealnx: Write outside array bounds

phy_idx is checked to be < 4, but np->phys[] is 2 elements long

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoixgbe: remove unnecessary call to device_init_wakeup
Andy Gospodarek [Wed, 22 Jul 2009 09:34:00 +0000 (09:34 +0000)]
ixgbe: remove unnecessary call to device_init_wakeup

Calls to device_init_wakeup should not be necessary in drivers that use
device_set_wakeup_enable since pci_pm_init will set the can_wakeup flag
for the device when initialized.

Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoixgbe: Don't priority tag control frames in DCB mode
Lucy Liu [Wed, 22 Jul 2009 14:07:33 +0000 (14:07 +0000)]
ixgbe: Don't priority tag control frames in DCB mode

Certain types of control packets (LLDP, LACP, etc.) are not supposed to have a
priority tag or vlan tag inserted.  Ixgbe driver is currently priority
tagging everything (if packet is not on a VLAN interface).

This patch modifies DCB mode, so that packets marked with skb priority
TC_PRIO_CONTROL are not priority tagged.  It also transmits these packets on
the highest priority traffic class.
Programs (like dcbd) can set the skb priority using a socket option.  Or, a tc
filter can be configured to set the priority value. Using the value
TC_PRIO_CONTROL (7) has the benefit that it is already defined in the kernel,
and the bonding LACP code already sets the skb->priority field to this value.

Signed-off-by: Lucy Liu <lucy.liu@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>
15 years agoixgbe: Enable FCoE offload when DCB is enabled for 82599
Yi Zou [Wed, 22 Jul 2009 14:07:12 +0000 (14:07 +0000)]
ixgbe: Enable FCoE offload when DCB is enabled for 82599

Currently, FCoE offload feature is turned on when the kernel config has
CONFIG_FCOE or CONFIG_FCOE_MODULE set. However, we really want to turn
FCoE offload on when there is FCoE traffic passing and turn it off when
it's just LAN traffic. Since FCoE depends on a lossless network provided
by DCB, this allows us to have FCoE turned on/off when user turns on DCB
using dcbtool.

Signed-off-by: Yi Zou <yi.zou@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>
15 years agonet: Rework mdio-ofgpio driver to use of_mdio infrastructure
Mark Ware [Thu, 23 Jul 2009 17:56:48 +0000 (10:56 -0700)]
net: Rework mdio-ofgpio driver to use of_mdio infrastructure

Changes to the fs_enet driver aa73832c5a80d6c52c69b18af858d88fa595dd3c
("net: Rework fs_enet driver to use of_mdio infrastructure")
cause kernel crashes when using the mdio-ofgpio driver.

This patch replicates similar changes made to the fs_enet mii-bitbang
drivers.  It has been tested on a custom mpc8280 based board using an
NFS mounted root.

Signed-off-by: Mark Ware <mware@elphinstone.net>
Acked-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoregister at91_ether using platform_driver_probe
Uwe Kleine-König [Tue, 21 Jul 2009 10:11:39 +0000 (10:11 +0000)]
register at91_ether using platform_driver_probe

at91ether_probe lives in .init.text, so using platform_driver_register
to register it is wrong because binding a device after the init memory
is discarded (e.g. via sysfs) results in an oops.

As requested by David Brownell platform_driver_probe is used instead of
moving the probe function to .devinit.text as proposed initially.
This saves some memory, but devices registered after the driver is
probed are not bound (probably there are none) and binding via sysfs
isn't possible.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Andrew Victor <linux@maxim.org.za>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoskge: Enable WoL by default if supported
Rafael J. Wysocki [Wed, 22 Jul 2009 02:58:55 +0000 (02:58 +0000)]
skge: Enable WoL by default if supported

If skge hardware is capable of waking up the system from sleep,
enable magic packet WoL during driver initialisation.

This makes WoL work without calling 'ethtool -s ethX wol g'
for each adapter.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Tested-by: Michael Guntsche <mike@it-loops.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
David S. Miller [Wed, 22 Jul 2009 18:18:40 +0000 (11:18 -0700)]
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

15 years agonet: KS8851 needs to depend on MII
Alexander Beregalov [Wed, 22 Jul 2009 01:16:51 +0000 (01:16 +0000)]
net: KS8851 needs to depend on MII

fix this build error when CONFIG_MII is not set
drivers/net/ks8851.c:999: undefined reference to `generic_mii_ioctl'
drivers/net/ks8851.c:1050: undefined reference to `mii_link_ok'
drivers/net/ks8851.c:1056: undefined reference to `mii_nway_restart'
drivers/net/ks8851.c:1044: undefined reference to `mii_ethtool_sset'
drivers/net/ks8851.c:1038: undefined reference to `mii_ethtool_gset'

Signed-off-by: Alexander Beregalov <a.beregalov@gmail.com>
Acked-by: Ben Dooks <ben@simtec.co.uk>
15 years agobe2net: Bug fix in the non-lro path. Size of received packet was not updated in stati...
Ajit Khaparde [Wed, 22 Jul 2009 16:28:55 +0000 (09:28 -0700)]
be2net: Bug fix in the non-lro path. Size of received packet was not updated in statistics properly.

This patch fixes a bug in the non-lro path.  Wrong size of received
packet was being passed for updating receive statistics.  This patch
is against the net-2.6 git.

Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoucc_geth: Revive fixed link support
Anton Vorontsov [Thu, 16 Jul 2009 21:31:47 +0000 (21:31 +0000)]
ucc_geth: Revive fixed link support

Since commit 0b9da337dca972e7a4144e298ec3adb8f244d4a4 ("Rework
ucc_geth driver to use of_mdio infrastructure") the fixed-link
support is broken.

This patch fixes the support by removing !ug_info->phy_node check,
and adds a call to of_phy_connect_fixed_link() if a phy is not attached
to the MAC.

Also, remove an old fixed-link code that we don't use any longer.

Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agogianfar: Revive fixed link support
Anton Vorontsov [Thu, 16 Jul 2009 21:31:42 +0000 (21:31 +0000)]
gianfar: Revive fixed link support

Since commit fe192a49118f5b1272317d60c7930ece4e13ae49 ("Rework gianfar
driver to use of_mdio infrastructure") the fixed-link support is
broken, the driver oopses at init_phy():

  Unable to handle kernel paging request for data at address 0x000000e4
  Faulting instruction address: 0xc01cf298
  Oops: Kernel access of bad area, sig: 11 [#1]
  [...]
  NIP [c01cf298] init_phy+0x80/0xdc
  LR [c01cf250] init_phy+0x38/0xdc
  Call Trace:
  [cf81fe80] [c01d1cf8] gfar_enet_open+0x6c/0x19c
  [cf81fea0] [c024494c] dev_open+0xfc/0x134
  [cf81fec0] [c0242edc] dev_change_flags+0x84/0x1ac
  [cf81fee0] [c0399ee0] ic_open_devs+0x168/0x2d8
  [cf81ff20] [c039b2e8] ip_auto_config+0x90/0x2a4
  [cf81ff60] [c0003884] do_one_initcall+0x34/0x1a8

This patch fixes the oops, and removes phy_node checks, and adds a call
to of_phy_connect_fixed_link() if a phy isn't attached..

Also, remove an old fixed-link code that we don't use any longer.

Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agofs_enet: Revive fixed link support
Anton Vorontsov [Thu, 16 Jul 2009 21:31:36 +0000 (21:31 +0000)]
fs_enet: Revive fixed link support

Since commit aa73832c5a80d6c52c69b18af858d88fa595dd3c ("Rework
fs_enet driver to use of_mdio infrastructure") the fixed-link support
is broken in the fs_enet driver.

This patch fixes the support by removing a check for phy_node, and adding
a call to of_phy_connect_fixed_link().

Also set netdev parent device via SET_NETDEV_DEV() call, this is needed
so that OF MDIO core could find a node pointer for a device.

Plus, fix "if (IS_ERR(phydev))" check, in case of errors,
of_phy_connect() returns NULL, not ERR_PTR as phy_connect().

Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoof/mdio: Add support function for Ethernet fixed-link property
Anton Vorontsov [Thu, 16 Jul 2009 21:31:31 +0000 (21:31 +0000)]
of/mdio: Add support function for Ethernet fixed-link property

Fixed-link support is broken for the ucc_eth, gianfar, and fs_enet
device drivers.  The "OF MDIO rework" patches removed most of the
support. Instead of re-adding fixed-link stuff to the drivers, this
patch adds a support function for parsing the fixed-link property
and obtaining a dummy phy to match.

Note: the dummy phy handling in arch/powerpc is a bit of a hack and
needs to be reworked.  This function is being added now to solve the
regression in the Ethernet drivers, but it should be considered a
temporary measure until the fixed link handling can be reworked.

Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agosky2: Avoid races in sky2_down
Mike McCormack [Wed, 22 Jul 2009 03:57:56 +0000 (20:57 -0700)]
sky2: Avoid races in sky2_down

Reset rx chain before trying to drain it.
Shut interrupts off last, incase there's something to report.

Signed-off-by: Mike McCormack <mikem@ring3k.org>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agodrivers/net/mlx4: Adjust constant
Julia Lawall [Sun, 19 Jul 2009 06:09:25 +0000 (06:09 +0000)]
drivers/net/mlx4: Adjust constant

The values in the advertising field are typically ADVERTISED_xxx, not
SUPPORTED_xxx.  Both SUPPORTED_10000baseT_Full and
ADVERTISED_1000baseT_Full have the same value.

The semantic match that finds this problem is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
struct ethtool_cmd E;
@@
*E.advertising = SUPPORTED_10000baseT_Full
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agodrivers/net: Move a dereference below a NULL test
Julia Lawall [Sun, 19 Jul 2009 05:26:35 +0000 (05:26 +0000)]
drivers/net: Move a dereference below a NULL test

If the NULL test is necessary, then the dereference should be moved below
the NULL test.

The semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
type T;
expression E,E1;
identifier i,fld;
statement S;
@@

- T i = E->fld;
+ T i;
  ... when != E=E1
      when != i
  if (E == NULL||...) S
+ i = E->fld;
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agodrivers/net: Move a dereference below a NULL test
Julia Lawall [Sun, 19 Jul 2009 06:13:04 +0000 (06:13 +0000)]
drivers/net: Move a dereference below a NULL test

If the NULL test is necessary, then the dereferences should be moved below
the NULL test.

The semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
type T;
expression E,E1;
identifier i,fld;
statement S;
@@

- T i = E->fld;
+ T i;
  ... when != E=E1
      when != i
  BUG_ON (E == NULL||...);
+ i = E->fld;
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agolibertas: Fix problem with broken V4 firmware on CF8381
Marek Vasut [Thu, 16 Jul 2009 17:19:53 +0000 (19:19 +0200)]
libertas: Fix problem with broken V4 firmware on CF8381

Firmware V4 on CF8381 reports region code shifted by 1 byte to left.
The following patch checks for this and handles it properly.

Signed-off-by: Marek Vasut <marek.vasut@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoath: add support for special 0x8000 regulatory domain
Luis R. Rodriguez [Mon, 20 Jul 2009 15:32:47 +0000 (08:32 -0700)]
ath: add support for special 0x8000 regulatory domain

Two users of ar9170 devices have now reported their cards
have been programmed with a regulatory domain of 0x8000.
This is not a valid regulatory domain as such these users were
unable to use these devices. Since this doesn't seem to be
a device EEPROM corruption we must treat it specially. It
may have been possible the manufacturer intended to use 0x0
as the regulatory domain and that would ultimately yield
to US but since we cannot get confirmationf or this we
default this special case to one of our world regulatory
domains, specifically 0x64.

Reported-by: DavidFreeman on #linux-wireless
Reported-by: Joerg Albert <jal2@gmx.de>
Cc: Christian Lamparter <chunkeey@web.de>,
Cc: Stephen Chen <stephen.chen@atheros.com>
Cc: David Quan <david.quan@atheros.com>
Cc: Tony Yang <tony.yang@atheros.com>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agop54spi: fix potential null deref in p54spi.c
Dan Carpenter [Sun, 19 Jul 2009 19:26:13 +0000 (21:26 +0200)]
p54spi: fix potential null deref in p54spi.c

Fix a potential NULL dereference bug during
error handling in p54spi_probe.

This bug was discovered by smatch:
(http://repo.or.cz/w/smatch.git).

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Christian Lamparter <chunkeey@web.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoacer-wmi: fix rfkill conversion
Alan Jenkins [Sun, 19 Jul 2009 08:48:28 +0000 (09:48 +0100)]
acer-wmi: fix rfkill conversion

Fix another polarity error introduced by the rfkill rewrite,
this time in acer_rfkill_set().

Signed-off-by: Alan Jenkins <alan-jenkins@tuffmail.co.uk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoconnector: maintainer/mail update.
Evgeniy Polyakov [Tue, 21 Jul 2009 19:43:51 +0000 (12:43 -0700)]
connector: maintainer/mail update.

Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoUSB host CDC Phonet network interface driver
Rémi Denis-Courmont [Tue, 21 Jul 2009 01:58:35 +0000 (01:58 +0000)]
USB host CDC Phonet network interface driver

Many Nokia handsets support a Phonet interface to the cellular modem
via a vendor-specific USB interface. CDC Phonet follows the
Communications Device Class model, with one control interface, and
and a pair of inactive and active data alternative interface. The later
has two bulk endpoint, one per direction.

This was tested against Nokia E61, Nokia N95, and the existing Phonet
gadget function for the Linux composite USB gadget framework.

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agomacsonic, jazzsonic: fix oops on module unload
Finn Thain [Tue, 21 Jul 2009 19:21:49 +0000 (12:21 -0700)]
macsonic, jazzsonic: fix oops on module unload

Set the driver data before using it. Fixes an oops when doing rmmod.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agomacsonic: move probe function to .devinit.text
Uwe Kleine-König [Sat, 11 Jul 2009 20:52:48 +0000 (20:52 +0000)]
macsonic: move probe function to .devinit.text

A pointer to mac_sonic_probe is passed to the core via
platform_driver_register and so the function must not disappear when the
.init sections are discarded.  Otherwise (if also having HOTPLUG=y)
unbinding and binding a device to the driver via sysfs will result in an
oops as does a device being registered late.

Various other functions that are called by mac_sonic_probe need to move
to .devinit.text, too.

An alternative to this patch is using platform_driver_probe instead of
platform_driver_register plus removing the pointer to the probe function
from the struct platform_driver.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Tested-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agocfg80211: double free in __cfg80211_scan_done
Christian Lamparter [Sun, 19 Jul 2009 03:05:37 +0000 (05:05 +0200)]
cfg80211: double free in __cfg80211_scan_done

This patch fixes a double free corruption in __cfg80211_scan_done:

 ================================================
 BUG kmalloc-512: Object already free
 ------------------------------------------------

 INFO: Allocated in load_elf_binary+0x18b/0x19af age=6
 INFO: Freed in load_elf_binary+0x104e/0x19af age=5
 INFO: Slab 0xffffea0001bae4c0 objects=14 used=7
 INFO: Object 0xffff88007e8a9918 @offset=6424 fp=0xffff88007e8a9488

 Bytes b4 0xffff88007e8a9908:  00 00 00 00 00 00 00 00 5a 5a
 [...]
 Pid: 28705, comm: rmmod Tainted: P         C 2.6.31-rc2-wl #1
 Call Trace:
  [<ffffffff810da9f4>] print_trailer+0x14e/0x16e
  [<ffffffff810daa56>] object_err+0x42/0x61
  [<ffffffff810dbcd9>] __slab_free+0x2af/0x396
  [<ffffffffa0ec9694>] ? wiphy_unregister+0x92/0x142 [cfg80211]
  [<ffffffff810dd5e3>] kfree+0x13c/0x17a
  [<ffffffffa0ec9694>] ? wiphy_unregister+0x92/0x142 [cfg80211]
  [<ffffffffa0ec9694>] wiphy_unregister+0x92/0x142 [cfg80211]
  [<ffffffffa0eed163>] ieee80211_unregister_hw+0xc8/0xff [mac80211]
  [<ffffffffa0f3fbc8>] p54_unregister_common+0x31/0x66 [p54common]
  [...]
 FIX kmalloc-512: Object at 0xffff88007e8a9918 not freed

The code path which leads to the *funny* double free:

       request = rdev->scan_req;
       dev = dev_get_by_index(&init_net, request->ifidx);
/*
 * the driver was unloaded recently and
 * therefore dev_get_by_index will return NULL!
 */
        if (!dev)
                goto out;
[...]
rdev->scan_req = NULL; /* not executed... */
[...]
 out:
        kfree(request);

Signed-off-by: Christian Lamparter <chunkeey@web.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agorfkill: remove too-strict __must_check
Alan Jenkins [Sat, 18 Jul 2009 18:20:20 +0000 (19:20 +0100)]
rfkill: remove too-strict __must_check

Some drivers don't need the return value of rfkill_set_hw_state(),
so it should not be marked as __must_check.

Signed-off-by: Alan Jenkins <alan-jenkins@tuffmail.co.uk>
Acked-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoath9k: Tune ANI function processing on AP mode during ANI reset
Luis R. Rodriguez [Thu, 16 Jul 2009 17:17:35 +0000 (10:17 -0700)]
ath9k: Tune ANI function processing on AP mode during ANI reset

For AP mode we must tune ANI specially for 2 GHz and
for 5 GHz. We mask in only the flags we want to toggle
on ath9k_hw_ani_control() through the ah->ani_function
bitmask, this will take care of ignoring changes during
ANI reset which we were disabling before.

Testedy-by: Steven Luo <steven@steven676.net>
Cc: Bennyam Malavazi <bennyam.malavazi@atheros.com>
Cc: Jouni Malinen <jouni.malinen@Atheros.com>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agort2x00: Fix chipset detection for rt2500usb
Andy Whitcroft [Thu, 16 Jul 2009 15:28:11 +0000 (16:28 +0100)]
rt2x00: Fix chipset detection for rt2500usb

The commit below changed the semantics of rt2x00_check_rev so that it no
longer checked the bottom 4 bits of the rev were non-zero.  During that
conversion this part of the check was not propogated to the rt2500usb
initialisation.

    commit 358623c22c9fd837b3b1b444377037f72553dc9f
    Author: Ivo van Doorn <ivdoorn@gmail.com>
    Date:   Tue May 5 19:46:08 2009 +0200

        rt2x00: Simplify rt2x00_check_rev

Without this check rt73 devices are miss recognised as rt2500 devices and
two drivers are loaded.  Preventing the device being used.  Reinstate this
check.

Signed-off-by: Andy Whitcroft <apw@canonical.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agonl80211: Memory leak fixed
Niko Jokinen [Wed, 15 Jul 2009 08:00:53 +0000 (11:00 +0300)]
nl80211: Memory leak fixed

Potential memory leak via msg pointer in nl80211_get_key() function.

Signed-off-by: Niko Jokinen <ext-niko.k.jokinen@nokia.com>
Signed-off-by: Luciano Coelho <luciano.coelho@nokia.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agortl8187: Fix for kernel oops when unloading with LEDs enabled
Larry Finger [Tue, 14 Jul 2009 20:55:16 +0000 (15:55 -0500)]
rtl8187: Fix for kernel oops when unloading with LEDs enabled

When rtl8187 is unloaded and CONFIG_RTL8187_LEDS is set, the kernel
may oops when the module is unloaded as the workqueue for led_on was
not being cancelled.

This patch fixes the problem reported in
http://marc.info/?l=linux-wireless&m=124742957615781&w=2.

Reported-by: Gábor Stefanik <netrolller.3d@gmail.com>
Signed-off-by: Larry Finger <Larry.Finger@lwfinger>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoath5k: temporarily disable crypto for AP mode
Bob Copeland [Tue, 14 Jul 2009 01:57:39 +0000 (21:57 -0400)]
ath5k: temporarily disable crypto for AP mode

Pavel Roskin reported some issues with using AP mode without
nohwcrypt=1.  Most likely this is similar to the problem fixed
some time ago in ath9k by 3f53dd64f192450cb331c0fecfc26ca952fb242f,
"ath9k: Fix hw crypto configuration for TKIP in AP mode."

That only affects TKIP but it's easiest to just disable that and
WEP too until we get a proper fix in.

Signed-off-by: Bob Copeland <me@bobcopeland.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agomac80211: use correct address for mesh Path Error
Javier Cardona [Tue, 14 Jul 2009 00:00:10 +0000 (17:00 -0700)]
mac80211: use correct address for mesh Path Error

For forwarded frames, we save the precursor address in addr1 in case it
needs to be used to send a Path Error.  mesh_path_discard_frame,
however, was using addr2 instead of addr1 to send Path Error frames, so
correct that and also make the comment regarding this more clear.

Signed-off-by: Andrey Yurovsky <andrey@cozybit.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agomac80211_hwsim: fix use after free
Johannes Berg [Mon, 13 Jul 2009 11:25:58 +0000 (13:25 +0200)]
mac80211_hwsim: fix use after free

Once the "data" pointer is freed, we can't be iterating
to the next item in the list any more so we need to use
list_for_each_entry_safe with a temporary variable.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agomac80211_hwsim: fix unregistration
Johannes Berg [Mon, 13 Jul 2009 11:04:30 +0000 (13:04 +0200)]
mac80211_hwsim: fix unregistration

If you rmmod the module while associated, frames might
be transmitted during unregistration -- which will crash
if the hwsim%d interface is unregistered first, so only
do that after all the virtual wiphys are gone.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agorfkill: fix rfkill_set_states() to set the hw state
Alan Jenkins [Sun, 12 Jul 2009 16:03:13 +0000 (17:03 +0100)]
rfkill: fix rfkill_set_states() to set the hw state

The point of this function is to set the software and hardware state at
the same time.  When I tried to use it, I found it was only setting the
software state.

Signed-off-by: Alan Jenkins <alan-jenkins@tuffmail.co.uk>
Reviewed-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agomac80211: fix injection in monitor mode
Pavel Roskin [Fri, 10 Jul 2009 20:42:29 +0000 (16:42 -0400)]
mac80211: fix injection in monitor mode

The location of the 802.11 header is calculated incorrectly due to a
wrong placement of parentheses.  Found by kmemcheck.

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agorfkill: allow toggling soft state in sysfs again
Johannes Berg [Fri, 10 Jul 2009 19:41:39 +0000 (21:41 +0200)]
rfkill: allow toggling soft state in sysfs again

Apparently there actually _are_ tools that try to set
this in sysfs even though it wasn't supposed to be used
this way without claiming first. Guess what: now that
I've cleaned it all up it doesn't matter and we can
simply allow setting the soft-block state in sysfs.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Tested-By: Darren Salt <linux@youmustbejoking.demon.co.uk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agomac80211: disable mesh
Johannes Berg [Fri, 10 Jul 2009 09:38:14 +0000 (11:38 +0200)]
mac80211: disable mesh

My kvm instance was complaining a lot about sleeping
in atomic contexts in the mesh code, and it turns out
that both mesh_path_add() and mpp_path_add() need to
be able to sleep (they even use synchronize_rcu()!).
I put in a might_sleep() to annotate that, but I see
no way, at least right now, of actually making sure
those functions are only called from process context
since they are both called during TX and RX and the
mesh code itself even calls them with rcu_read_lock()
"held".

Therefore, let's disable it completely for now.

It's possible that I'm only seeing this because the
hwsim's beaconing is broken and thus the peers aren't
discovered right away, but it is possible that this
happens even if beaconing is working, for a peer that
doesn't exist or so.

It should be possible to solve this by deferring the
freeing of the tables to call_rcu() instead of using
synchronize_rcu(), and also using atomic allocations,
but maybe it makes more sense to rework the code to
not call these from atomic contexts and defer more of
the work to the workqueue. Right now, I can't work on
either of those solutions though.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoiwlwifi: only update byte count table during aggregation
Reinette Chatre [Thu, 9 Jul 2009 17:33:39 +0000 (10:33 -0700)]
iwlwifi: only update byte count table during aggregation

The byte count table is only used for aggregation. Updating it
in other cases caused fragmented frames to be dropped.

This fixes http://www.intellinuxwireless.org/bugzilla/show_bug.cgi?id=2004

Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoiwlwifi: only show active power level via sysfs
Reinette Chatre [Thu, 9 Jul 2009 17:33:37 +0000 (10:33 -0700)]
iwlwifi: only show active power level via sysfs

This changes the power_level file to adhere to the "one value
per file" sysfs rule. The user will know which power level was
requested as it will be the number just written to this file. It
is thus not necessary to create a new sysfs file for this value.

In addition it fixes a problem where powertop's parsing expects
this value to be the first value in this file without any descriptions.

Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
cc: stable@kernel.org
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agoiwmc3200wifi: fix NULL pointer dereference in iwm_if_free
Zhu Yi [Thu, 9 Jul 2009 09:24:15 +0000 (17:24 +0800)]
iwmc3200wifi: fix NULL pointer dereference in iwm_if_free

The driver private data is now based on wiphy. So we should not
touch the private data after wiphy_free() is called. The patch
fixes the potential NULL pointer dereference by making the
iwm_wdev_free() the last one on the interface removal path.

Signed-off-by: Zhu Yi <yi.zhu@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
15 years agocan: switch carrier on if device was stopped while in bus-off state
Wolfgang Grandegger [Mon, 20 Jul 2009 04:06:41 +0000 (04:06 +0000)]
can: switch carrier on if device was stopped while in bus-off state

This patch fixes a problem when a device is stopped while in the
bus-off state. Then the carrier remains off forever.

Signed-off-by: Kurt Van Dijck <kurt.van.dijck@skynet.be>
Signed-off-by: Wolfgang Grandegger <wg@grandegger.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agocan: restart device even if dev_alloc_skb() fails
Wolfgang Grandegger [Mon, 20 Jul 2009 04:06:40 +0000 (04:06 +0000)]
can: restart device even if dev_alloc_skb() fails

If dev_alloc_skb() failed in can_restart(), the device was left behind
in the bus-off state. This patch restarts the device nevertheless.

Signed-off-by: Kurt Van Dijck <kurt.van.dijck@eia.be>
Signed-off-by: Wolfgang Grandegger <wg@grandegger.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agocan: sja1000: remove duplicated includes
Wolfgang Grandegger [Mon, 20 Jul 2009 04:06:39 +0000 (04:06 +0000)]
can: sja1000: remove duplicated includes

Remove duplicated #include('s) in
  drivers/net/can/sja1000/sja1000.c

Signed-off-by: Huang Weiyi <weiyi.huang@gmail.com>
Signed-off-by: Wolfgang Grandegger <wg@grandegger.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoNew device ID for sc92031 [1088:2031]
Cesar Eduardo Barros [Sun, 19 Jul 2009 08:03:32 +0000 (08:03 +0000)]
New device ID for sc92031 [1088:2031]

rain_maker@root-forum.org wrote:
> Hello cesar,
>
> In a recent thread in a german linux forum, a user reported his PIC
> NIC not being recognized by the kernel.
>
> Fortunately he provided enough information and I was able to help him
> and get the device working with the sc92031 driver.
>
> The device ID is [1088:2031] (Vendor is called "Microcomputer Systems
> (M) Son"), here is the respective thread in "ubuntuusers.de"
>
> http://forum.ubuntuusers.de/topic/lankarte-unter-xubuntu-wird-nicht-erkannt/
>
> (Although you might not speak german, the code provided will show
> you, that the device is actually working with your driver).
>
> It would be nice, if you include this new device ID to the
> sc92031-driver.
>
> Regards,
>
> Axel Köllhofer (aka Rain_Maker)

Cc: rain_maker@root-forum.org
Signed-off-by: Cesar Eduardo Barros <cesarb@cesarb.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years ago3c589_cs: re-initialize the multicast in the tc589_reset
Ken Kawasaki [Sun, 19 Jul 2009 13:08:12 +0000 (13:08 +0000)]
3c589_cs: re-initialize the multicast in the tc589_reset

3c589_cs:
re-initialize the multicast in the tc589_reset,
and spin_lock the set_multicast_list function.

Signed-off-by: Ken Kawasaki <ken_kawasaki@spring.nifty.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoFix error return for setsockopt(SO_TIMESTAMPING)
Rémi Denis-Courmont [Mon, 20 Jul 2009 00:47:04 +0000 (00:47 +0000)]
Fix error return for setsockopt(SO_TIMESTAMPING)

I guess it should be -EINVAL rather than EINVAL. I have not checked
when the bug came in. Perhaps a candidate for -stable?

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: fix thermal check and shutdown
Dhananjay Phadke [Fri, 17 Jul 2009 15:27:08 +0000 (15:27 +0000)]
netxen: fix thermal check and shutdown

Check temperature for all PCI functions, that can allow
graceful shutdown of all interfaces on the overheated card.

Old code was only monitoring temperature for function 0 only.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: fix deadlock on dev close
Dhananjay Phadke [Fri, 17 Jul 2009 15:27:07 +0000 (15:27 +0000)]
netxen: fix deadlock on dev close

netxen: fix deadlock on dev close

The tx ring accounting fix in commit cb2107be43d2fc5eadec58b92b
("netxen: fix tx ring accounting") introduced intermittent
deadlock when inteface is going down.

This was possibly combined effect of speculative tx pause,
calling netif_tx_lock instead of queue lock and unclean
synchronization with napi which could end up unmasking
interrupt.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetxen: fix context deletion sequence
Dhananjay Phadke [Fri, 17 Jul 2009 15:27:06 +0000 (15:27 +0000)]
netxen: fix context deletion sequence

o Use D3 reset context deletion for NX2031, it cleans up
  more resources in the firmware.
o Release rx buffers after hardware context has been reset.
o Delete tx context after rx context, some firmware control
  commands are sent on tx context, so it should be the last
  to go.

Signed-off-by: Dhananjay Phadke <dhananjay@netxen.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet: Micrel KS8851 SPI network driver
Ben Dooks [Thu, 16 Jul 2009 05:24:08 +0000 (05:24 +0000)]
net: Micrel KS8851 SPI network driver

Network driver for the SPI version of the Micrel KS8851
network chip.

Signed-off-by: Ben Dooks <ben@simtec.co.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agotcp: Use correct peer adr when copying MD5 keys
John Dykstra [Fri, 17 Jul 2009 09:23:22 +0000 (09:23 +0000)]
tcp: Use correct peer adr when copying MD5 keys

When the TCP connection handshake completes on the passive
side, a variety of state must be set up in the "child" sock,
including the key if MD5 authentication is being used.  Fix TCP
for both address families to label the key with the peer's
destination address, rather than the address from the listening
sock, which is usually the wildcard.

Reported-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: John Dykstra <john.dykstra1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agotcp: Fix MD5 signature checking on IPv4 mapped sockets
John Dykstra [Thu, 16 Jul 2009 05:04:51 +0000 (05:04 +0000)]
tcp: Fix MD5 signature checking on IPv4 mapped sockets

Fix MD5 signature checking so that an IPv4 active open
to an IPv6 socket can succeed.  In particular, use the
correct address family's signature generation function
for the SYN/ACK.

Reported-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: John Dykstra <john.dykstra1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoAdd mac driver for w90p910
Wan ZongShun [Thu, 16 Jul 2009 02:55:05 +0000 (02:55 +0000)]
Add mac driver for w90p910

Add mac driver support for evaluation board based on w90p910.

Signed-off-by: Wan ZongShun <mcuos.com@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoUpdate Andreas Koensgen's email address
Ralf Baechle [Fri, 17 Jul 2009 04:47:19 +0000 (04:47 +0000)]
Update Andreas Koensgen's email address

The kernel has used a stale email address of Andreas for a few years.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoMAINTAINERS entry for STRIP driver
Ralf Baechle [Fri, 17 Jul 2009 04:42:28 +0000 (04:42 +0000)]
MAINTAINERS entry for STRIP driver

The web server does no longer exist, it's not on archive.org nor does there
seem to be any mirror.

 MAINTAINERS |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agocdc-eem: bad crc checking
Vincent CUISSARD [Thu, 16 Jul 2009 06:08:58 +0000 (06:08 +0000)]
cdc-eem: bad crc checking

When the driver received an EEM packet with CRC option enabled, driver must
compute and check the CRC of the Ethernet data. Previous version computes CRC
on Ethernet data plus the original CRC value. Skbuff is correctly trimed but
the old length is used when CRC is computed.

Signed-off-by: Vincent CUISSARD <vincent.cuissard@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoixgbe: Remove DPRINTK messages in DCB mode
Lucy Liu [Thu, 16 Jul 2009 13:43:31 +0000 (13:43 +0000)]
ixgbe: Remove DPRINTK messages in DCB mode

Remove debug DPRINTK in DCB mode netlink interface.

Signed-off-by: Lucy Liu <lucy.liu@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>
15 years agoixgbe: clear mac address data block in DCB mode
Lucy Liu [Thu, 16 Jul 2009 13:43:10 +0000 (13:43 +0000)]
ixgbe: clear mac address data block in DCB mode

This change clears the address data block memory space, which is needed for
the 82598 which does not have a SAN MAC.

Signed-off-by: Lucy Liu <lucy.liu@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>
15 years agosky2: revert shutdown changes
Stephen Hemminger [Thu, 16 Jul 2009 13:20:57 +0000 (13:20 +0000)]
sky2: revert shutdown changes

The commit changes to shutdown path broke startup on some systems.

revert commit c0bad0f2e4366d5bbfe0c4a7a80bca8f4b05272b

Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet: sock_copy() fixes
Eric Dumazet [Wed, 15 Jul 2009 23:13:10 +0000 (23:13 +0000)]
net: sock_copy() fixes

Commit e912b1142be8f1e2c71c71001dc992c6e5eb2ec1
(net: sk_prot_alloc() should not blindly overwrite memory)
took care of not zeroing whole new socket at allocation time.

sock_copy() is another spot where we should be very careful.
We should not set refcnt to a non null value, until
we are sure other fields are correctly setup, or
a lockless reader could catch this socket by mistake,
while not fully (re)initialized.

This patch puts sk_node & sk_refcnt to the very beginning
of struct sock to ease sock_copy() & sk_prot_alloc() job.

We add appropriate smp_wmb() before sk_refcnt initializations
to match our RCU requirements (changes to sock keys should
be committed to memory before sk_refcnt setting)

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoE100: work around the driver using streaming DMA mapping for RX descriptors.
Krzysztof Halasa [Tue, 14 Jul 2009 11:01:54 +0000 (11:01 +0000)]
E100: work around the driver using streaming DMA mapping for RX descriptors.

E100 places it's RX packet descriptors inside skb->data and uses them
with bidirectional streaming DMA mapping. Unfortunately it fails to
transfer skb->data ownership to the device after it reads the
descriptor's status, breaking on non-coherent (e.g., ARM) platforms.

This have to be converted to use coherent memory for the descriptors.

Signed-off-by: Krzysztof Halasa <khc@pm.waw.pl>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agobonding: clean muticast addresses when device changes type
Moni Shoua [Wed, 15 Jul 2009 04:56:31 +0000 (04:56 +0000)]
bonding: clean muticast addresses when device changes type

Bonding device forbids slave device of different types under the same
master.

However, it is possible for a bonding master to change type during its
lifetime.  This can be either from ARPHRD_ETHER to ARPHRD_INFINIBAND
or the other way arround.  The change of type requires device level
multicast address cleanup because device level multicast addresses
depend on the device type.

The patch adds a call to dev_close() before the bonding master changes
type and dev_open() just after that.

In the example below I enslaved an IPoIB device (ib0) under
bond0. Since each bonding master starts as device of type ARPHRD_ETHER
by default, a change of type occurs when ib0 is enslaved.

This is how /proc/net/dev_mcast looks like without the patch

5    bond0           1     0     00ffffffff12601bffff000000000001ff96ca05
5    bond0           1     0     01005e000116
5    bond0           1     0     01005e7ffffd
5    bond0           1     0     01005e000001
5    bond0           1     0     333300000001
6    ib0             1     0     00ffffffff12601bffff000000000001ff96ca05
6    ib0             1     0     333300000001
6    ib0             1     0     01005e000001
6    ib0             1     0     01005e7ffffd
6    ib0             1     0     01005e000116
6    ib0             1     0     00ffffffff12401bffff00000000000000000001
6    ib0             1     0     00ffffffff12601bffff00000000000000000001

and this is how it looks like after the patch.

5    bond0           1     0     00ffffffff12601bffff000000000001ff96ca05
5    bond0           1     0     00ffffffff12601bffff00000000000000000001
5    bond0           1     0     00ffffffff12401bffff0000000000000ffffffd
5    bond0           1     0     00ffffffff12401bffff00000000000000000116
5    bond0           1     0     00ffffffff12401bffff00000000000000000001
6    ib0             1     0     00ffffffff12601bffff000000000001ff96ca05
6    ib0             1     0     00ffffffff12401bffff00000000000000000116
6    ib0             1     0     00ffffffff12401bffff0000000000000ffffffd
6    ib0             2     0     00ffffffff12401bffff00000000000000000001
6    ib0             2     0     00ffffffff12601bffff00000000000000000001

Signed-off-by: Moni Shoua <monis@voltaire.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoatl1c: misplaced parenthesis
roel kluin [Sun, 12 Jul 2009 12:57:38 +0000 (12:57 +0000)]
atl1c: misplaced parenthesis

Fix misplaced parenthesis

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoatl1c: add missing parentheses
roel kluin [Sun, 12 Jul 2009 11:40:34 +0000 (11:40 +0000)]
atl1c: add missing parentheses

Parentheses are required or the comparison occurs before the bitand.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-2.6
David S. Miller [Fri, 17 Jul 2009 00:34:50 +0000 (17:34 -0700)]
Merge branch 'master' of git://git./linux/kernel/git/kaber/nf-2.6

15 years agonetfilter: nf_conntrack: nf_conntrack_alloc() fixes
Eric Dumazet [Thu, 16 Jul 2009 12:03:40 +0000 (14:03 +0200)]
netfilter: nf_conntrack: nf_conntrack_alloc() fixes

When a slab cache uses SLAB_DESTROY_BY_RCU, we must be careful when allocating
objects, since slab allocator could give a freed object still used by lockless
readers.

In particular, nf_conntrack RCU lookups rely on ct->tuplehash[xxx].hnnode.next
being always valid (ie containing a valid 'nulls' value, or a valid pointer to next
object in hash chain.)

kmem_cache_zalloc() setups object with NULL values, but a NULL value is not valid
for ct->tuplehash[xxx].hnnode.next.

Fix is to call kmem_cache_alloc() and do the zeroing ourself.

As spotted by Patrick, we also need to make sure lookup keys are committed to
memory before setting refcount to 1, or a lockless reader could get a reference
on the old version of the object. Its key re-check could then pass the barrier.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
15 years agonetfilter: xt_osf: fix nf_log_packet() arguments
Patrick McHardy [Thu, 16 Jul 2009 12:01:54 +0000 (14:01 +0200)]
netfilter: xt_osf: fix nf_log_packet() arguments

The first argument is the address family, the second one the hook
number.

Signed-off-by: Patrick McHardy <kaber@trash.net>
15 years agonet/can: add module alias to can protocol drivers
Lothar Waßmann [Tue, 14 Jul 2009 23:12:25 +0000 (23:12 +0000)]
net/can: add module alias to can protocol drivers

Add appropriate MODULE_ALIAS() to facilitate autoloading of can protocol drivers

Signed-off-by: Lothar Wassmann <LW@KARO-electronics.de>
Acked-by: Oliver Hartkopp <oliver@hartkopp.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet/can bugfix: use after free bug in can protocol drivers
Lothar Waßmann [Tue, 14 Jul 2009 23:10:21 +0000 (23:10 +0000)]
net/can bugfix: use after free bug in can protocol drivers

Fix a use after free bug in can protocol drivers

The release functions of the can protocol drivers lack a call to
sock_orphan() which leads to referencing freed memory under certain
circumstances.

This patch fixes a bug reported here:
https://lists.berlios.de/pipermail/socketcan-users/2009-July/000985.html

Signed-off-by: Lothar Wassmann <LW@KARO-electronics.de>
Acked-by: Oliver Hartkopp <oliver@hartkopp.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoRevert "NET: Fix locking issues in PPP, 6pack, mkiss and strip line disciplines."
David S. Miller [Tue, 14 Jul 2009 20:13:41 +0000 (13:13 -0700)]
Revert "NET: Fix locking issues in PPP, 6pack, mkiss and strip line disciplines."

This reverts commit adeab1afb7de89555c69aab5ca21300c14af6369.

As Alan Cox explained, the TTY layer changes that went recently
to get rid of the tty->low_latency stuff fixes this already,
and even for -stable it's the ->low_latency changes that should
go in to fix this, rather than this patch.

Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoskbuff.h: Fix comment for NET_IP_ALIGN
Tobias Klauser [Mon, 13 Jul 2009 22:48:16 +0000 (22:48 +0000)]
skbuff.h: Fix comment for NET_IP_ALIGN

Use the correct function call for skb_reserve in the comment for
NET_IP_ALIGN.

Signed-off-by: Tobias Klauser <klto@zhaw.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agodrivers/net: using spin_lock_irqsave() in net_send_packet()
Dongdong Deng [Sun, 12 Jul 2009 20:27:06 +0000 (20:27 +0000)]
drivers/net: using spin_lock_irqsave() in net_send_packet()

spin_unlock_irq() will enable interrupt in net_send_packet(),
this patch changes it to spin_lock_irqsave/spin_lock_irqrestore,
so that it doesn't enable interrupts when already disabled,
and netconsole would work properly over cs89x0/isa-skeleton.

Call trace:
netconsole write_msg()
{
 ...
 -> spin_lock_irqsave();
        -> netpoll_send_udp()
          -> netpoll_send_skb()
            -> net_send_packet()
              ->...

 -> spin_unlock_irqrestore();
 ...
}

Signed-off-by: Dongdong Deng <dongdong.deng@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoNET: phy_device, fix lock imbalance
Jiri Slaby [Mon, 13 Jul 2009 11:23:39 +0000 (11:23 +0000)]
NET: phy_device, fix lock imbalance

Don't forget to unlock a mutex in phy_scan_fixups on a fail path.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agogre: fix ToS/DiffServ inherit bug
Andreas Jaggi [Tue, 14 Jul 2009 16:35:59 +0000 (09:35 -0700)]
gre: fix ToS/DiffServ inherit bug

Fixes two bugs:
- ToS/DiffServ inheritance was unintentionally activated when using impair fixed ToS values
- ECN bit was lost during ToS/DiffServ inheritance

Signed-off-by: Andreas Jaggi <aj@open.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoigb: gcc-3.4.6 fix
Eric Dumazet [Mon, 13 Jul 2009 18:11:41 +0000 (11:11 -0700)]
igb: gcc-3.4.6 fix

forward declaration of inline function should be avoided, or
old gcc cannot compile.

Reported-by: Teck Choon Giam <giamteckchoon@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoatlx: duplicate testing of MCAST flag
roel kluin [Sun, 12 Jul 2009 13:12:37 +0000 (13:12 +0000)]
atlx: duplicate testing of MCAST flag

Fix duplicate testing of MCAST flag

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Acked-by: Jay Cliburn <jcliburn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoNET: Fix locking issues in PPP, 6pack, mkiss and strip line disciplines.
Ralf Baechle [Mon, 13 Jul 2009 04:09:20 +0000 (21:09 -0700)]
NET: Fix locking issues in PPP, 6pack, mkiss and strip line disciplines.

Guido Trentalancia reports:

I am trying to use the kiss driver in the Linux kernel that is being
shipped with Fedora 10 but unfortunately I get the following oops:

mkiss: AX.25 Multikiss, Hans Albas PE1AYX
mkiss: ax0: crc mode is auto.
ADDRCONF(NETDEV_CHANGE): ax0: link becomes ready
------------[ cut here ]------------
WARNING: at kernel/softirq.c:77 __local_bh_disable+0x2f/0x83() (Not
tainted)
[...]
unloaded: microcode]
Pid: 0, comm: swapper Not tainted 2.6.27.25-170.2.72.fc10.i686 #1
 [<c042ddfb>] warn_on_slowpath+0x65/0x8b
 [<c06ab62b>] ? _spin_unlock_irqrestore+0x22/0x38
 [<c04228b4>] ? __enqueue_entity+0xe3/0xeb
 [<c042431e>] ? enqueue_entity+0x203/0x20b
 [<c0424361>] ? enqueue_task_fair+0x3b/0x3f
 [<c041f88c>] ? resched_task+0x3a/0x6e
 [<c06ab62b>] ? _spin_unlock_irqrestore+0x22/0x38
 [<c06ab4e2>] ? _spin_lock_bh+0xb/0x16
 [<c043255b>] __local_bh_disable+0x2f/0x83
 [<c04325ba>] local_bh_disable+0xb/0xd
 [<c06ab4e2>] _spin_lock_bh+0xb/0x16
 [<f8b6f600>] mkiss_receive_buf+0x2fb/0x3a6 [mkiss]
 [<c0572a30>] flush_to_ldisc+0xf7/0x198
 [<c0572b12>] tty_flip_buffer_push+0x41/0x51
 [<f89477f2>] ftdi_process_read+0x375/0x4ad [ftdi_sio]
 [<f8947a5a>] ftdi_read_bulk_callback+0x130/0x138 [ftdi_sio]
 [<c05d4bec>] usb_hcd_giveback_urb+0x63/0x93
 [<c05ea290>] uhci_giveback_urb+0xe5/0x15f
 [<c05eaabf>] uhci_scan_schedule+0x52e/0x767
 [<c05f6288>] ? psmouse_handle_byte+0xc/0xe5
 [<c054df78>] ? acpi_ev_gpe_detect+0xd6/0xe1
 [<c05ec5b0>] uhci_irq+0x110/0x125
 [<c05d4834>] usb_hcd_irq+0x40/0xa3
 [<c0465313>] handle_IRQ_event+0x2f/0x64
 [<c046642b>] handle_level_irq+0x74/0xbe
 [<c04663b7>] ? handle_level_irq+0x0/0xbe
 [<c0406e6e>] do_IRQ+0xc7/0xfe
 [<c0405668>] common_interrupt+0x28/0x30
 [<c056821a>] ? acpi_idle_enter_simple+0x162/0x19d
 [<c0617f52>] cpuidle_idle_call+0x60/0x92
 [<c0403c61>] cpu_idle+0x101/0x134
 [<c069b1ba>] rest_init+0x4e/0x50
 =======================
---[ end trace b7cc8076093467ad ]---
------------[ cut here ]------------
WARNING: at kernel/softirq.c:136 _local_bh_enable_ip+0x3d/0xc4()
[...]
Pid: 0, comm: swapper Tainted: G        W 2.6.27.25-170.2.72.fc10.i686
 [<c042ddfb>] warn_on_slowpath+0x65/0x8b
 [<c06ab62b>] ? _spin_unlock_irqrestore+0x22/0x38
 [<c04228b4>] ? __enqueue_entity+0xe3/0xeb
 [<c042431e>] ? enqueue_entity+0x203/0x20b
 [<c0424361>] ? enqueue_task_fair+0x3b/0x3f
 [<c041f88c>] ? resched_task+0x3a/0x6e
 [<c06ab62b>] ? _spin_unlock_irqrestore+0x22/0x38
 [<c06ab4e2>] ? _spin_lock_bh+0xb/0x16
 [<f8b6f642>] ? mkiss_receive_buf+0x33d/0x3a6 [mkiss]
 [<c04325f9>] _local_bh_enable_ip+0x3d/0xc4
 [<c0432688>] local_bh_enable_ip+0x8/0xa
 [<c06ab54d>] _spin_unlock_bh+0x11/0x13
 [<f8b6f642>] mkiss_receive_buf+0x33d/0x3a6 [mkiss]
 [<c0572a30>] flush_to_ldisc+0xf7/0x198
 [<c0572b12>] tty_flip_buffer_push+0x41/0x51
 [<f89477f2>] ftdi_process_read+0x375/0x4ad [ftdi_sio]
 [<f8947a5a>] ftdi_read_bulk_callback+0x130/0x138 [ftdi_sio]
 [<c05d4bec>] usb_hcd_giveback_urb+0x63/0x93
 [<c05ea290>] uhci_giveback_urb+0xe5/0x15f
 [<c05eaabf>] uhci_scan_schedule+0x52e/0x767
 [<c05f6288>] ? psmouse_handle_byte+0xc/0xe5
 [<c054df78>] ? acpi_ev_gpe_detect+0xd6/0xe1
 [<c05ec5b0>] uhci_irq+0x110/0x125
 [<c05d4834>] usb_hcd_irq+0x40/0xa3
 [<c0465313>] handle_IRQ_event+0x2f/0x64
 [<c046642b>] handle_level_irq+0x74/0xbe
 [<c04663b7>] ? handle_level_irq+0x0/0xbe
 [<c0406e6e>] do_IRQ+0xc7/0xfe
 [<c0405668>] common_interrupt+0x28/0x30
 [<c056821a>] ? acpi_idle_enter_simple+0x162/0x19d
 [<c0617f52>] cpuidle_idle_call+0x60/0x92
 [<c0403c61>] cpu_idle+0x101/0x134
 [<c069b1ba>] rest_init+0x4e/0x50
 =======================
---[ end trace b7cc8076093467ad ]---
mkiss: ax0: Trying crc-smack
mkiss: ax0: Trying crc-flexnet

The issue was, that the locking code in mkiss was assuming it was only
ever being called in process or bh context.  Fixed by converting the
involved locking code to use irq-safe locks.

Review of other networking line disciplines shows that 6pack, both sync
and async PPP and STRIP have similar issues.  The ppp_async one is the
most interesting one as it sorts out half of the issue as far back as
2004 in commit http://git.kernel.org/?p=linux/kernel/git/tglx/history.git;a=commitdiff;h=2996d8deaeddd01820691a872550dc0cfba0c37d

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Reported-by: Guido Trentalancia <guido@trentalancia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetdev: restore MTU change operation
Ben Hutchings [Thu, 9 Jul 2009 17:59:01 +0000 (17:59 +0000)]
netdev: restore MTU change operation

alloc_etherdev() used to install a default implementation of this
operation, but it must now be explicitly installed in struct
net_device_ops.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonetdev: restore MAC address set and validate operations
Ben Hutchings [Thu, 9 Jul 2009 17:54:35 +0000 (17:54 +0000)]
netdev: restore MAC address set and validate operations

alloc_etherdev() used to install default implementations of these
operations, but they must now be explicitly installed in struct
net_device_ops.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agosit: fix regression: do not release skb->dst before xmit
Sascha Hlusiak [Sun, 12 Jul 2009 03:30:52 +0000 (20:30 -0700)]
sit: fix regression: do not release skb->dst before xmit

The sit module makes use of skb->dst in it's xmit function, so since
93f154b594fe47 ("net: release dst entry in dev_hard_start_xmit()") sit
tunnels are broken, because the flag IFF_XMIT_DST_RELEASE is not
unset.

This patch unsets that flag for sit devices to fix this
regression.

Signed-off-by: Sascha Hlusiak <contact@saschahlusiak.de>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet: ip_push_pending_frames() fix
Eric Dumazet [Wed, 8 Jul 2009 14:20:42 +0000 (14:20 +0000)]
net: ip_push_pending_frames() fix

After commit 2b85a34e911bf483c27cfdd124aeb1605145dc80
(net: No more expensive sock_hold()/sock_put() on each tx)
we do not take any more references on sk->sk_refcnt on outgoing packets.

I forgot to delete two __sock_put() from ip_push_pending_frames()
and ip6_push_pending_frames().

Reported-by: Emil S Tantilov <emils.tantilov@gmail.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Tested-by: Emil S Tantilov <emils.tantilov@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet: sk_prot_alloc() should not blindly overwrite memory
Eric Dumazet [Wed, 8 Jul 2009 19:36:05 +0000 (19:36 +0000)]
net: sk_prot_alloc() should not blindly overwrite memory

Some sockets use SLAB_DESTROY_BY_RCU, and our RCU code correctness
depends on sk->sk_nulls_node.next being always valid. A NULL
value is not allowed as it might fault a lockless reader.

Current sk_prot_alloc() implementation doesnt respect this hypothesis,
calling kmem_cache_alloc() with __GFP_ZERO. Just call memset() around
the forbidden field.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agocxgb3: Fix crash caused by stashing wrong netdev_queue
Roland Dreier [Thu, 9 Jul 2009 09:30:25 +0000 (09:30 +0000)]
cxgb3: Fix crash caused by stashing wrong netdev_queue

Commit c3a8c5b6 ("cxgb3: move away from LLTX") exposed a bug in how
cxgb3 looks up the netdev_queue it stashes away in a qset during
initialization.  For multiport devices, the TX queue index it uses is
offset by the first_qset index of each port.  This leads to a crash
once LLTX is removed, since hard_start_xmit is called with one TX
queue lock held, while the TX reclaim timer task grabs a different
(wrong) TX queue lock when it frees skbs.

Fix this by removing the first_qset offset used to look up the TX
queue passed into t3_sge_alloc_qset() from setup_sge_qsets().

Signed-off-by: Roland Dreier <rolandd@cisco.com>
Acked-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agoixgbe: Fix coexistence of FCoE and Flow Director in 82599
Yi Zou [Thu, 9 Jul 2009 02:29:50 +0000 (02:29 +0000)]
ixgbe: Fix coexistence of FCoE and Flow Director in 82599

Fix coexistence of Fiber Channel over Ethernet (FCoE) and Flow Director (FDIR)
in 82599 and remove the disabling of FDIR when FCoE is enabled.

Currently, FDIR is turned off when FCoE is enabled under the assumption that
FCoE is always enabled with DCB being turned on. However, FDIR does not have
to be turned off all the time when FCoE is enabled since FCoE can be enabled
without DCB being turned on, e.g., use link pause only. This patch makes sure
that when DCB is turned on or off, FDIR is turned on or off correspondingly;
and when FCoE is enabled, it does not disable FDIR, rather, it will have FDIR
set up properly so FCoE and FDIR can coexist regardless of DCB being on or off.

Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agomemory barrier: adding smp_mb__after_lock
Jiri Olsa [Wed, 8 Jul 2009 12:10:31 +0000 (12:10 +0000)]
memory barrier: adding smp_mb__after_lock

Adding smp_mb__after_lock define to be used as a smp_mb call after
a lock.

Making it nop for x86, since {read|write|spin}_lock() on x86 are
full memory barriers.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
15 years agonet: adding memory barrier to the poll and receive callbacks
Jiri Olsa [Wed, 8 Jul 2009 12:09:13 +0000 (12:09 +0000)]
net: adding memory barrier to the poll and receive callbacks

Adding memory barrier after the poll_wait function, paired with
receive callbacks. Adding fuctions sock_poll_wait and sk_has_sleeper
to wrap the memory barrier.

Without the memory barrier, following race can happen.
The race fires, when following code paths meet, and the tp->rcv_nxt
and __add_wait_queue updates stay in CPU caches.

CPU1                         CPU2

sys_select                   receive packet
  ...                        ...
  __add_wait_queue           update tp->rcv_nxt
  ...                        ...
  tp->rcv_nxt check          sock_def_readable
  ...                        {
  schedule                      ...
                                if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
                                        wake_up_interruptible(sk->sk_sleep)
                                ...
                             }

If there was no cache the code would work ok, since the wait_queue and
rcv_nxt are opposit to each other.

Meaning that once tp->rcv_nxt is updated by CPU2, the CPU1 either already
passed the tp->rcv_nxt check and sleeps, or will get the new value for
tp->rcv_nxt and will return with new data mask.
In both cases the process (CPU1) is being added to the wait queue, so the
waitqueue_active (CPU2) call cannot miss and will wake up CPU1.

The bad case is when the __add_wait_queue changes done by CPU1 stay in its
cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1 will then
endup calling schedule and sleep forever if there are no more data on the
socket.

Calls to poll_wait in following modules were ommited:
net/bluetooth/af_bluetooth.c
net/irda/af_irda.c
net/irda/irnet/irnet_ppp.c
net/mac80211/rc80211_pid_debugfs.c
net/phonet/socket.c
net/rds/af_rds.c
net/rfkill/core.c
net/sunrpc/cache.c
net/sunrpc/rpc_pipe.c
net/tipc/socket.c

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